Portfolio for PhD Application

Portfolio

LIU ZIDONG Selected Works 2018 - 2021

Shenzhen Architectural Design Institute Graduate Student in Master degree at the Bartlett, UCL

Portfolio

LIU ZIDONG Selected Works 2018 - 2021

Shenzhen Architectural Design Institute Graduate Student in Master degree at the Bartlett, UCL

zidong.liu.19@ucl.ac.uk Copyright 2021 by LIU Zidong All Rights Reserved

LIU ZIDONG Shenzhen, China | 1 Jun 1996 zidong.liu.19@ucl.ac.uk | liuzidong2019@163.com GitHub: https://github.com/wodeyyf

EDUCATION Bartlet, University College London, UK Master of Urban Design

GPA: Distinction

09.2019 - 09.2020

DASP, Politecnico di Torino, Italy Architecture Design(Exchange)

MARK: 29 (full mark is 30, Top 10%)

02.2019 - 07.2019

Southeast University, China(Top 2 in China Architecture School, Project 985&211) Bachelor of Architecture Design GPA: 3.78 (Top 12%)

09.2014 - 07.2019

EXPERIENCES MENG Architects, Shenzhen, China A team of 12 architects led by MENG Jianmin (Academician of the Chinese Academy of Engineering, co-founder with Carlo Ratti of the Bi-City Biennale of Urbanism\Architecture) Architect China Computational Design Annual Conference, Beijing, China (CDAC 2021) Presenter

09.2020 - Now

12.2021

UCL Creative Design and Interactive Architecture Research Workshop Teaching assistant (Prof. Ruairi Glynn)

09.2021 - 10.2021

UCL Creative Design and Interactive Architecture Research Workshop Student (Prof. Ruairi Glynn)

05.2021 - 06.2021

Tsinghua University Online Summer School Epidemics, Health and Urban Architecture in a Global Perspective Teaching assistant (Prof. LIU Yishi)

09.2021 - 10.2021

Tongji University DigialFUTURES 2021 Digital Future Global Workshop Student

06.2021 - 07.2021

Politecnico di Torino Transitional Morphologies Research Unit, Turin, Italy Student

07.2018

PUBLICATION Predicting the Vitality of Stores along the Street based on Business Type Sequence via Recurrent Neural Network (First author).

CAADRIA 2022

Topological Networks Using a Sequential Method for Data Distribution in an Interactive Community Design (Sole author).

ACADIA 2021

HousingPrime App, a Bottom-up Platform for Housing Customization and Participatory Community Design (Sole author).

ICAMC 2021

The Development Modes of Postwar Middle-Class Housing in Turin, Italy (Second author).

City & House

SKILLS Code: Python, Mathematica, Java, Processing, CoLab, TensorFlow, VSCode Algorithm: cycleGAN, GAN, RNN, Genetic Algorithm, Simulated Annealing, K-means Modelling: Revit, Grasshopper, Kangaroo, Rhino, Speckle, SketchUp, Speckle

Contents

Part 1 ACADEMIC WRITING 01

Using GCN for Data-driven Urban Design PhD Research Proposal

02

Using RNN for Street Stores Vitality Prediction CAARIDA 2022

03

Topological Networks Using a Sequential Method ACADIA 2021

Part 2 RESEARCH PROJECTS 01

Housing Prime, A Digital Platform for Housing Customization Generative Design Algorithm Programming

02

App User Interface Coding based on Java Diploma Project of Master of Urban Design

Part 3 SKILLS EXPLORATION 01

Facade Styletransfer based on CycleGAN Paris to Barcelona

02

Game Design based on Processing an Environmental Awareness Game about CO2 Reduction

03

Mathematica Programming Systematic Training on Functional Programming

04

Speckle An Online Grasshoper Collaboration Platform

Part 4 DESIGN PROJECTS 01

HFC Central Tower (358m)

02

Taiping Waterway Bridge

03

Foshan Nanhai Art Center

04

Guangzhou Substation

05

Shenzhen Baoan Archives Center

06

Shenzhen Reform and Opening-Up Exhibition Hall

ACADEMIC WRITING

01

Zidong Liu Bartlett School of Architecture University College London

PhD Research Proposal Quantitative Analysis of Spatial Networks Using Graph Neural Networks (GCN) for Data-driven Urban Design

ABSTR ACT The research aims to develop a data-based spatial network assessment model through machine learning for data-driven design. The traditional top-down urban design method has been criticized for its over-reliance on the subjectivity of architects. As revealed by Space Syntax, spatial network is strongly associated with social logic. In the machine learning model, the input is the spatial network of the built environment, and the output is the weighted graph, in which the node connection represents the vitality degree of the corresponding street. The trained AI models will be evaluated for correlation of their predictions with actual city features and compared with the predictions of existing evaluation models, such as the NACH model. In this comparison, it is supposed to discover urban spatial network features that are overlooked by existing evaluation models, thus improving existing models and further contributing to data-driven urban design. The research is based on Graph Convolutional Network (GCN), an emerging neural network model tailored for harnessing topological networks. GCN has already been playing an influential role in many areas of society. However, its application in architecture is still rare, making this study of great academic value and practical significance.

RESE ARCH BACKGROUND There is a strong demand for objective quantitative assessment methods in urban design practice. At present, the urban design process still relies heavily on architects’ personal experience and subjective judgment, which bring great uncertainty to the project's results. Therefore, the quantitative assessment method of urban features for data-driven design and plan has a wide range of practical implications.

Urban features are strongly associated with spatial networks. As revealed by the Space Syntax team, spatial networks are an objective reflection of social structures.

(Hiller and Julienne 1984; Hiller 1996). The network centrality has been proved to have a strong correlation with urban features such as people movement. As early as 1938, architects began to use bubble diagrams to represent the spatial organization of architectural layout designs (Emmons and Paul 2017). The article a City Is Not a Tree opened the research horizon of analysing cities from the perspective of graph theory (Alexander 1965).

Nowadays, with the rapid development of network science (Barabasi 2016), the paradigm of city science has shifted

1

from Location to Network (Batty 2013). However, traditional analysis models for spatial networks are mathematical descriptions of centrality like integration and choice and lack the use of descriptive data.

The rapid development of AI and Big Data in recent years has opened up new possibilities for quantitatively assessing urban features. There have been many studies using AI to evaluate and predict urban performance. However, most of the current research is based on images, and applications of network-based machine learning are still rare. It is only in recent years that graph neural networks have matured with the establishment of GCN (Kipf and Welling 2017).

Currently, graph machine learning has been practiced in various scientific fields, such as recommender systems, new drug synthesis, particle-based physics simulation, etc. (Leskovec 2021), but the exploration based on GCN has hardly appeared in architecture and city at present.

Therefore, the use of machine learning to access and predict urban features from a spatial network perspective has huge research potential in the coming years.

RESE ARCH A IM This research aims to explore a novel spatial analysis model based on Graph Neural Networks (GCN), reveal the network attributes that might be neglected by existing computational analysis models (topological models and angular models) and promote their algorithm improvement, thus contributing to the data-driven design. It takes urban spatial networks as the research object and attempts to explore an assessment model that has a strong correlation with urban features such as community vitality and crime rates. Unlike existing analytical models of Space Syntax such as NACH, which is a purely mathematical calculation of network centrality, this machine learning model is driven by descriptive data and will evolve continuously with the improvement of neural networks and databases. In combination with optimization algorithms such as genetic algorithms, this evaluation model can be used for datadriven design and planning decisions.

LITER ATURE RE V IEW Space Syntax and Big Data

Space Syntax is a mixed method that includes both spatial computational analysis and data investigation. By exploring the correlation between the analytical model and the descriptive data, the data gains academic value and the effectiveness of the analytical model is verified. Through gate count, tracing and other observation methods, the Space Syntax team has accumulated a rich database for decades of research, such as the "Ten Minute Map" by Bill Hillier's team, which movement data for every road in the Barnsbury area of London (Hillier, 1996). In design projects, observation data is a good complement to network analysis, such as the impact of site attraction points on the crowd, thus making the design more relevant.

However, in the methodology of Space Syntax, spatial analysis and data observation are two relatively independent steps. The spatial analysis does not introduce spatial performance but takes space itself as the research object.

The existing spatial analysis models are divided into three categories: topological model, angular model, and distance model, which are all mathematical descriptions of network centrality. After decades of research, the most popular model is currently NACH (Normalised Angular Choice) (Hillier, Yang and Turner 2012). The research data only verifies the correlation of spatial analysis but does not directly become part of the computational model. In contrast, in machine learning models, the addition of new data automatically optimizes the existing analysis model.

Machine Learning in Urban Analysis and Feature Prediction

In recent five years, the application of AI in the architectural field can be roughly divided into three general directions: classification issues, generation issues and optimization issues. Detailed information on the latest use of neural network models can be found in the Statistical Table in the Appendix.

Most of the current prediction models are imagebased, among which GAN models are predominant. The researchers transformed the citizens’ cycling route data into an urban heat map to represent community vitality and explored its relationship with urban fabric (Sun, Jiang and Zheng 2020). Similar approaches can be used to predict crime rate (He and Zheng 2021) and commercial value (Shou, Chen and Zheng 2021). Besides, the GAN approach can also be applied to urban plans (Shen et al 2020).

However, image-based machine learning has an unavoidable drawback: due to the limitation of computing power their images are only about 100,000 pixels, and the generated results always have ambiguous areas. This problem is particularly severe in studies that require high information accuracy like housing layout generation. (Zheng and Ren 2020) This is the reason why some studies have attempted to vectorize images before performing machine learning. (Li et al. 2020; Xia and Tong 2020).

GCN in Urban Analysis and Feature Prediction

The graph-based neural network was proposed in 2008 (Scarselli et al. 2008), but it is only in recent years with the introduction of GCN (Kipf and Welling 2017) that graph machine learning has matured to be practiced in various fields. Different from the previous neural network model that takes regular (Euclidean structure) data as input such as two-dimensional image (CNN, GAN) and one-dimensional sequence (RNN), GCN focuses on the irregular multidimensional graph structure, which is closer to the real spatial network of the urban fabric.

Among the sparse applications of graph machine learning in architecture and planning, a notable one is the GCN-based household layout generation study published recently by Harvard University (Lu et al. 2021), in which they used the existing floor plan dataset CubiCasa5K as input and generated the graph structure dataset CubiGraph5K. They automated the process of transforming vectorized building plans into graphs, which provides a solid technical foundation for building graph training sets in the future. Another remarkable study is the paper Traffic Prediction based on GCN-LSTM Model (Wu et al. 2021), where they proposed a traffic prediction model using graph neural networks. Their research methods and training results are highly informative for this research proposal.

2

LITER ATURE RE V IEW The research process involves training an evaluation model in which the input is a topological network of the built environment, and the output is a weighted graph where the node connections represent spatial performance like movement or crime rates. Just as the existing analysis models of Space Syntax are used in different situations (among which NACH is the most popular analysis model at present), this neural network model will continuously evolve with the increasing enrichment of database and the continuous modification of algorithm.

Mastering Technical Skills

I have some research experience with using cycleGAN, RNN. For the future in-depth research on graph neural networks.

I need to study linear algebra, computer graphics, data science. My programming skills also need to be strengthened. Here are some specific ways to study: attending relevant courses at university, reading literature and attending open online courses. I am currently attending Stanford University's open course CS224W: Machine Learning with Graphs by Jure Leskovec.

Mastering Technical Skills

Building a database is a key step in machine learning. On the one hand, traditional data observations such as gate count for people movement data are still very important today due to their high resolution. On the other hand, the rapid development of Big Data has reduced the cost of data acquisition and increased the number of types of data, making real-time data acquisition possible. Specifically, data can be obtained from POI data from online maps, social media data, mobile phone signalling data, consumption data from O2O platforms, travel platforms, public data from government websites, etc.

Training Model

When building the dataset, the input data is urban spatial networks and the output data is the weighted graphs. The input data and output data are matched one by one. Data are organized into the different folders (80% in the training set and 20% in the test set). At the beginning of my research, I will call on open-source code available on the web and perform machine learning with only minor parameter adjustments. When I am skilled enough, I will try to significantly change the GCN structure, for example by combining it with LSTM model for more accurate data analysis. The process will be full of literature reading and collaboration.

Model Evaluation and Comparison Study

The trained GCN models will be put into a built environment to check their correlation with urban features and to

12

compare with the correlation of existing analytical models such as NACH. For example, an AI model based on pedestrian movement data from a street network will be tested for its ability to predict pedestrian traffic on an unknown street. The GCN model may perform better, or the NACH model may be better, but most importantly their focus must be different. The GCN model may reveal some potentially unique laws that can be used to improve the existing NACH algorithm. This process is very similar to how professional Go players develop moves by playing against AlphaGo.

E XPECTED RESULTS This study is not just about obtaining an AI model that predicts urban features from spatial networks. It can be seen as a new tool for urban analysis alongside topological models, angular models, etc. At the same time, the study of this neural network model may reveal some spatial network properties that are overlooked by existing spatial syntactic theories. and thus support the improvement of existing spatial computational analysis models. The research is highly extensible. It can be combined with optimization algorithms to achieve data-driven urban design. This method is also effective for problems of architectural scale such as house floor plans automation.

KE Y REFERENCES Book

Barabási, A.L. & Pósfai, M. (2016). Network Science.

Cambridge: Cambridge University Press.

Batty, M. (2005). Cities and Complexity. Understanding Cities with Cellular Automata, Agent-Based Models, and Fractals, MIT Press, Boston.

Batty, M. (2018). Inventing future cities. Boston: MIT Press.

Hillier, B., & Hanson, J. (1984). The Social Logic of Space.

Cambridge: Cambridge University Press.

Hillier, B. (2007). Space is the machine: a configurational theory of architecture. London: Space Syntax.

Journal Article

Emmons, P. (2006). Embodying networks: bubble diagrams and the image of modern organicism. The Journal of Architecture, 11(4), 441-461.

He, J., & Zheng, H. (2021). Prediction of crime rate in urban neighborhoods based on machine learning. Engineering Applications of Artificial Intelligence, 106, 104460.

Hillier, B., W. R. G., Yang, T. and Turner, A. (2012) Normalising Least Angle Choice in Depthmap and How It Opens up New Perspectives on the Global and Local Analysis of City Space.

Journal of Space Syntax, 3, 155-193.

Roth, J., & Hashimshony, R. (1988). Algorithms in graph theory and their use for solving problems in architectural design. Computer-aided design, 20(7), 373-381.

Structured Architectural Floor Plan Dataset. Proceedings of the 26th International Conference on Computer-Aided Architectural Design Research in Asia (CAADRIA), 81-90.

Hong Kong: CAADRIA. Shekhawat, K. (2021). A Graph Theoretic Approach for the Automated Generation of Dimensioned Floor plans.

Proceedings of the 26th International Conference on Computer-Aided Architectural Design Research in Asia (CAADRIA), 141-150. Hong Kong, CAADRIA. Shen, J., Liu, C., Ren, Y., & Zheng, H. (2020). Machine Learning Assisted Urban Filling. Proceedings of the 25th International Conference on Computer-Aided Architectural Design Research in Asia (CAADRIA), 679-688. Bangkok: CAADRIA. Shou, X., Chen, P., & Zheng, H. (2021). Predicting the Heat Map of Street Vendors from Pedestrian Flow through Machine Learning. Proceedings of the 26th International Conference on Computer-Aided Architectural Design Research in Asia (CAADRIA), 569-578. Hong Kong, CAADRIA. Sun, Y.J., Jiang, L., & Zheng, H. (2020). A Machine Learning Method of Predicting Behavior Vitality via Urban Forms.

Proceedings of the 40th Annual Conference of the Association for Computer Aided Design in Architecture (ACADIA), 160-168. Philadelphia, ACADIA. Xia, X., & Tong, Z. (2020). A Machine Learning-Based Method for Predicting Urban Land Use. Proceedings of the 25th International Conference on Computer-Aided Architectural Design Research in Asia (CAADRIA), 21-30. Bangkok: CAADRIA.

Scarselli, F., Gori, M., Tsoi, A. C., Hagenbuchner, M., & Monfardini, G. (2008). The graph neural network model.

IEEE transactions on neural networks, 20(1), 61-80.

Wu, Z., Huang, M., & Zhao, A. (2021, July). Traffic prediction based on GCN-LSTM model. In Journal of Physics: Conference Series (Vol. 1972, No. 1, p. 012107). IOP Publishing.

Xing, Z., Chen, Z.L., Gu, Y.Y., Bai, J. N. & Yao, Y. (2021).

Quantified Analysis on Affecting Factors of Urban Street Vigor Based on Big-Data. Journal of Human Settlements in West China, 36(3): 98-105.

Conference Paper

Lu, Y., Tian, R., Li, A., Wang, X., & Jose Luis, G. D. C. L. (2021).

CubiGraph5k: Organizational Graph Generation for

13

Appendix

Statistics of the Latest Use of AI at Top CAD Conferences

Conference

Title

Author

CAADRIA2021

Automatically Generating Layouts Of Large-Scale Office Park Using Position-Based Dynamics

Shuqi Cao, Guohua Ji

CAADRIA2021

Generative Design Of Urban Fabrics Using Deep Learning

CAADRIA2021

Cubigraph5K Organizational Graph Generation For Structured Architectural Floor Plan Dataset

CAADRIA2021

Genscan:A Generative Method For Populating Parametric 3D Scan Datasets

CAADRIA2021

Intuitive Behavior The Operation Of Reinforcement Learning In Generative Design Processes

CAADRIA2021

From Exploration To Interpretation Adopting Deep Representation Learning Models To Latent Space Interpretation Of Architectural Design Alternatives

CAADRIA2021

Deep- Performance Incorporating Deep Learning For Automating Building Performance Simlation In Generative Systems

CAADRIA2021

Hierarchical Multi-Label Architectural Image Recognition And Classification

CAADRIA2021

Reinforcement Learning For Architectural Design-Build Opportunity Of Machine Learning In A Material-Informed Circular Design Strategy

CAADRIA2021 CAADRIA2021 CAADRIA2021

Architecture Language And Ai Language, Attentional Generative Adversarial Networks（Attn Gan） And Architecture Design

CAADRIA2021 CAADRIA2021 ACADIA2020

Multi-Objective Optimisation Of A Free-Form Building Shape To Mprove The Solar Energy Utilisation Potential Using Artificial Neural Networks Spatial Findings On Chilean Architecture Stylegan Ai Graphics A Tool For Searching Active Bending Bamboo Strips In Construction Via Deep Learning The Experiment Of Neural Network On The Cognition Of Style Facilitating Archiitect-Client Communication In The Pre-Design Phase 21E8：Coupling Generative Adversarial Neural Networks (Gans) With Blockchain Applications In Building Information Modelling (Bim) Systems Can A Generative Adversarial Network Remove Thin Clouds In Aerial Photographs？Toward Improving The Accuracy Of Generating Horizontal Building Mask Images Or Ee L Earning In Urban Planning And Design Global Urban Cityscape-Unsupervised Clustering Exploration Of Human Activity And Mobility Infrastructure Predicting The Heat Map Of Street Vendors From Pedestrian Flow Through Machine Learning A Machine Learning method of Predicting Behavior Vitality Using Open Source Data

ACADIA2020

Clustering and Morphological Analysis of Campus Context Based on a Convolutional Autoencoder Model

ACADIA2020

Machine Learning for Comparative Urban Planning at Scale: An Aviation Case Study

CAADRIA2021 CAADRIA2021 CAADRIA2021 CAADRIA2021 CAADRIA2021 CAADRIA2021 CAADRIA2021

ACADIA2020

Generating and Optimizing a Funicular Arch Floor Structure

ACADIA2020 ACADIA2020

Encoded Images Representational Protocols for Integrating cGANs in Iterative Computational Design Processes BIM Hyperreality Data Synthesis Using BIM and Hyperrealistic Rendering for Deep Learning Text-to-Form 3D Prediction by Linguistic Descriptions How Machines learn to plan A Critical Interrogation of Machine Vision Techniques in architecture Spatial Assembly with Self-Play Reinforcement Learning An AI Lens on Historic Cairo A Deep Learning Application for Minaret Classification Making a new City Image DeepGreen Coupling Biological and Artificial Intelligence in Urban Design 3D Graph Convolutional Neural Networks in Architecture Design Steering into the Skid Arbitraging Human and artificial Intelligences to Augment the Design Process Reprogramming Urban Block by Machine Creativity How to use neural networks as generative tools to design space Deep Learning Methods for Urban Analysis and Health Estimation of Obesity Automatic Generation of the Schematic Mechanical System Drawing by Generative Adversarial Network

ACADIA2020 ACADIA2020 ACADIA2020 ACADIA2020 ACADIA2020 ACADIA2020 ACADIA2020 ACADIA2020 eCAADe2020 eCAADe2020 eCAADe2020 eCAADe2020

A machine-learning model driven by geometry, material and structural performance data in architectural design process

eCAADe2020

On AI Adoption Issues in Architectural Design Identifying the issues based on an extensive literature review.

Space Filling Curves for Optimising Single Point Incremental Sheet Forming using Supervised Learning Algorithms Machine Learning Methods in Energy Simulations for Architects and Designers: The implementation of supervised machine learning in the context of the computational design process Smart Structures - A Generative Design Framework for Aesthetic Guidance in Structural Node Design: Application of Typogenetic Design for Custom-Optimisation of Structural Nodes Machine Learning Methods for Clustering Architectural Precedents: Classifying the relationship between building and ground Exploration & Validation Making sense of generated data in large option sets Generative and synthetic biological design imaginations for the Miami bay area

eCAADe2020

Applying Deep Learning and Databases for Energyefficient Architectural Design

eCAADe2020 eCAADe2020 eCAADe2020 eCAADe2020 eCAADe2020 eCAADe2020

eCAADe2020 eCAADe2020

An Academy of Spatial Agents Generating spatial configurations with deep reinforcement learning The Emoting City Designing feeling and artificial empathy in mediated environments

eCAADe2020

Automatic Generation of Horizontal Building Mask Images by Using a 3D Model with Aerial Photographs for Deep Learning

eCAADe2020

Drawing Recognition Integrating Machine Learning Systems into Architectural Design Workflows

eCAADe2020

Architectural Visualisation with Conditional Generative Adversarial Networks (cGAN). What machines read in architectural sketches.

eCAADe2020

User-driven Configurable Architectural Assemblies Towards artificial intelligence-embedded responsive environments

eCAADe2020

From the Cognitive to the Sentient Building: Machine Learning for the preservation of museum collections in historical architecture

eCAADe2020

Methods for the Prediction and Specification of Functionally Graded Multi-Grain Responsive Timber Composites Designing with a Robot Interactive methods for brick wall design using computer vision

eCAADe2020

14

Exploring Optimal Ways To Represent Topological And Spatial Features Of Building Designs In Deep Learning Methods And Applications For Architecture Sketch With Artificial Intelligence (Ai) A Multimodal Al Approach For Conceptual Design

Jinmo Rhee, Pedro Velos

Yueheng Lu, Runjia Tian, Ao Li, Xiaoshi Wang, Garcia

Mohammad Keshavarzil, Oladapo Afolabi, Luisa Ca Zakhor

Dasong Wang, Roland Sno

Jielin Chen, Rudi Stouffs

Shermeen Yousif, Daniel Bol Jielin Chen, Rudi Stouffs, Filip Chien-Hua Huang

Viktor Eisenstadt, Hardik Aror, Christoph Ziegle, Langenhan, Klaus-Dieter Althoff, An Yifan Zhoui, Hyoung-June P Matias Del Campo

Xin Zhao, Yunsong Han, Linha

Tomas Vivanc, Larrain Antoni, Valencia Xuyou Yang, Weishun Xu Wei Hu Cheng-Lin Chuang, Sheng-Fen Provides Ng

Kazunosuke Ikeno, Tomohiro Fukuda, No

Tania Papasotiriou, Stephan C Xinyue Shou, Pinyang Chen, Hao Yunjuan Sun, Lei Jiang, Hao Z Peiwen Li, Wenbo Zhu Ahmed Meeran, Sam Conrad

Hao Zheng, Xinyu Wang, Zehua Qi, Shixuan Su

Gabriella Rossi, Paul Nicho Mohammad Alawadhi, Wei Hang Zhang

Matias del Campo, Alexandrea Carlson, S Tyson Hosmer, Panagiotis Tigas, David R

Islam Zohier, Ahmed El Antably, Ahm Brian Ho

Claudia Pasquero, Marco Po

Matias del Campo, Alexandrea Carlson, S Geoff Kimm, Mark Burry De Yu

David Newton, Dan Piatkowski, Wesley Mars

Gen Sato, Tsukasa Ishizawa, Hajime Iseda Sevil Yazici Mateusz Zwierzycki

Kunaljit Chadha, Alexandre Dubor, La

Adam Sebestyen, Jakub T

Manuel Muehlbauer, Andy Song, J

Abdulrahman Alymani, Wassim Jabi, Pa Rachel Tan, Trevor Patt, Seow Jin Koh,

Thomas Spiegelhalter, Alfredo Andia, Juhasz Lev Manav Mahan Singh, Patricia Schneider-Marin, Ha Philipp Geyer

Pedro Veloso, Ramesh Krishna Sayjel Vijay Patel, Raffi Tchakerian, Renata Lemos Cropper

Kazunosuke Ikeno, Tomohiro Fukuda, No

Lachlan Brown, Michael Yip, Nicole Gardner, M. Han Yannis Zavoleas, Cristina Ra

Yick Hin Edwin Chan, A. Benjami Peter Buš Federico Mario La Russa, Cettina

Vasiliki Fragkia, Isak Worre Fo

Avishek Das, Isak Worre Foged, Mads

Research type

Research topic

Space generation

Generation of office park planning based on image with site information

so

Space generation

Generation of urban morphology diagrams

a Del Castillo Lopez Jose Luis

Space generation

Generation of room relation graphs

GCN

Space generation

Generation of the building layout variations based on 3D scan

DNN+VGG

ook

Space generation

Multi-agent generative urban design

DRL Deep Reinforcement Learning

s

Space generation

Generation of architectural image

StyleGAN

Performance optimization

Performance prediction of daylight simulation

Deep Performance

Space classification

Image recognition and classification

Space generation

A material-informed design-build workflow

CNN DRL Deep Reinforcement Learning

aldas, Allen Y. Yang, Avideh

lojan Biljeck

Jessica Bielski, Christoph ndreas Dengel Park

Tool

PBD (Position-Based Dynamic) GAN

Space generation

Generation of relation map

Space generation

Image predication based on sketches and textual information

CNN GAN

Space generation

Visual output based on lauguage

AtteGAN

ai Shen

Performance optimization

Improvement of solar enegy utilization

ANN

a Philip F. Yuan u

Space classification Performance optimization Space classification Space generation

Image recognition of Chilean buildings Prediction of the number and locations of nodes on bamboo strips Image style classification Generation of photo images based on sketches

StyleGAN ANN CNN LA+SA+GAN

Space generation and analysis

Utilization of E8 network to combine GAN with BIM

GAN

n Chien

Nobuyoshi Yabuki

Space generation

Generation of mask images and photographs without thin clouds

GAN

Space classification Space classification Space classification

Identification of similar characteristics and patterns among urban areas Prediction of vendor's locations Prediction of cyclists' behaviours based on built environment planes Clustering and quantitative analysis of campus morphology based on compressed feature vectors Recgonition of satellite images of airports Generation of funicular supporting structures based on the floor plan with the positions of columns Iterative computational design and digital fabrication processes Recgonition of synthetic photos of BIM and photorealistic renderings

Unsupervised clustering Pyramid scene parsing network (PSPNet)+GAN GAN

Space generation

3D form generation based on text

Stanford Scene Graph Parser+GCN

Space generation Space generation

Generation of plans based on existing planning solutions Generation of intelligent spatial assemblies from encoded spatial parts

GAN RL

Space classification

Classification of minaret styles

CNN

Space generation

Generation of city images

CNN

oletto

Space generation

Generation of biomorphological maps based on satellite images

GAN

Sandra Manninger y

Space generation Literature Review

Prediction of style, functionality value and asthetic value

GCN

Chalup o Zheng Zheng

Space classification Joyce

Space classification

un, Masoud akbarzadeh

Space generation

olas Yan

Space generation Space generation

Sandra Manninger Reeves, Ziming He

med S. Madani

shall, Atharva Tendle

a, Hideo Kitahara

Clustering CNN CNN cGAN GAN

Space generation

Generation of spatial layouts based on existing cases

GAN

Space classification

Prediction of ovesity rates based on satellite images

CNN

Performance optimization

Evaluation of Piping Coverage Rate

GAN

Performance optimization

Optimization of time efficiency and design results

ANN

Literature Review

aura Puigpinos

Performance optimization

Optimization of long-drawn-out ISF operation

K-NEAREST NEIGHBOUR MACHINE LEARNING

Tyc

Performance optimization

Evaluation of the complex façade performance

CNN

Space generation and performance optimization

Stucture generation and optimization

Genetic Algorithm

adraig Corcoran

Space classification

Identification of the typological and topological characteristics of architectural solutions

K-means Clustering

Edmund Chen

Space generation

Slection of generated design options

K-means Clustering

Space generation

Design generation for Miami Bay

Performance optimization

Prediction and optimization of buildings' energy

AI Cloud Algorithms Component-based machine learning model (CBML)

Jane Burry

vente, Srikanth Namuduri annes Harter, Werner Lang,

amurti Morais, Jie Zhang, Simon

Nobuyoshi Yabuki

nk Haeusler, Nariddh Khean, amos

in Spaeth

Santagati

Space generation

Interactive generative design

MADRL+DDQN+DCNN

Space classification

Implementation of artificial empathy in the built environment.

Human-AI Interfaces

Space generation

Generation of building mask images

SpaceNet

Space classification

Recognition of rooms within floor plan images

CNN

Space generation

Cyclic transformation of sketches and images

Cyclic-cGAN

Space generation

An assembly linked with human activator's participative inputs

Learning-to-Design-and-Assembly method Machine Learning algorithms

Space classification

Recognition of impact of actions

oged

Performance optimization

Generation of grain pattern based on gradient color map

cGAN

s Brath Jensen

Performance optimization

Recognition of brick positions

TensorFlow-Keras

15

Predicting the Vitality of Stores along the Street RNN Model with Business Type Sequence as Input Accepted by CAADRIA 2022

Zidong Liu Bartlett School of Architecture University College London Xiao Xiao Department of Architecture and Design Politecnico di Torino Yan Li Faculty of Engineering University of Sydney

1

ABSTR ACT The rational planning of store types and locations to maximize street vitality is essential in real estate planning. Traditional business planning relies heavily on the subjective experience of developers. Some studies have been conducted to predict urban vitality from the scale of urban texture through image-based neural networks. Currently, developers have access to low-resolution urban data to support their decision making, and researchers have done much image-based machine learning research from the scale of urban texture.

However, there is still a lack of research on the functional layout with shop-level accuracy. This paper uses a sequence-based neural network (RNN) to explore the relationship between the sequence of store types along a street and its commercial vitality. We use customer review data of 80streets from O2O platforms to represent the store vitality degree. In the machine learning model, the input is the sequence of store types on the street, and the output is the corresponding sequence of business vitality indexes. After training and evaluation, the model was shown to have acceptable accuracy. We further combined this evaluation model with a genetic algorithm to develop a business planning optimization tool: it automatically gives the best ranking order based on the input store types to maximize the overall street business value.

KE Y WORDS Machine Learning; Big Data Analysis; Recurrent Neural Network; Genetic Algorithm; SDG8 Decent Work and Economic Growth; SDG9 Industry, Innovation and Infrastructure

1

Genetic Algorithm Based Vitality Enhancement of Nanjing Gongyuan West Street.

INTRODUCTION

possibilities to develop evaluation models based on data.

Background

Business management experience shows that the order in which stores are located on the street has a significant effect on business. Stores at the top of the street tend to have higher popularity and therefore higher rents. People without specific shopping goals are more likely to shop at the first supermarket or fruit store they see. At the same time, however, some store functions may not be as sensitive to their street location. For example, a bookstore or a clinic may not need to be in the most exposed location to get a high number of customers with a specific purpose.

In addition to a store's location in the street, its neighbours may also have a complex impact on its operation, depending on the types of stores. For example, suppose two supermarkets are located close to each other. In that case, they may have a vicious competition, but for McDonald's and KFC, putting them together may help enhance their visibility on the street. A bank placed next to a luxury store may help to increase the sales of the luxury store.

Problem Statement

The current business planning model still relies heavily on the subjective experience of real estate developers, which leads to uncertainty in planning results and adversely affects the profitability of businesses and the vitality of cities. There has been increasingly detailed data analysis such as customer base analysis and regional vitality analysis to support low-resolution issues like the proportion of store types (Schlegel et al., 2021). However, the resolution of these data is still not sufficient to guide shop-level planning. Store business information used to be a trade secret until the rise of O2O platforms made available some public data. This is why real estate business consultants are heavily monopolized by a few companies, such as Savills, CBRE, Colliers International, Jones Lang LaSalle and DTZ. There is still a lack of scientific research based on public data for more precise planning such as the location sequence of commercial types along a street. Therefore, this paper has a strong practical research significance.

Literature Review Machine Learning in Urban Analysis and Feature Prediction

As mentioned above, there is a very complex relationship between the order of location of stores and their commercial viability. Some types of stores are sensitive to location; others are not sensitive to their location but are easily influenced by their neighbours. In the past, it has been difficult to describe this complex relationship accurately with a mathematical function. Machine learning gives us new

There have been many studies on region vitality through machine learning. However, most of them are currently image-based and they are not accurate to store-level precision. Among these studies, GAN models are predominant. A study transforms the citizens’ cycling route data into an urban heat map to represent community vitality and explores its relationship with urban fabric (Sun, Jiang and Zheng 2020). Similar approaches can be used to predict other urban metrics, such as urban crime rate (He and Zheng 2021) and commercial value (Shou, Chen and Zheng 2021). However, image-based machine learning has an unavoidable drawback: due to the limitation of computing power, their images are only about 100,000 pixels, and the generated results always have ambiguous areas. This is the reason why some studies have attempted to vectorize images before performing machine learning. (Xia and Tong, 2020).

Recurrent Neural Network

In this study, we choose RNN as the basic neural network model. RNN is based on sequential data, widely used in natural language processing, advertising recommendations and so on. The principle of RNN is to scan the data in sequence, and the parameters of the previous step will be involved in the next step of prediction. Compared with other neural network models, its outstanding features are: 1. RNN uses sequential data as input and output 2.

In RNN models, the relative position of data has a decisive influence on the prediction results. 3. The input and output of each item in the training set of the RNN model can be of different lengths. These features are highly compatible with our research object and goal.

Currently, the use of RNNs in the architectural and urban fields is very rare, because space is a complex network structure and is not suitable for linear sequential studies.

Among the sparse RNN-based studies, there is one very relevant to the topic of this paper on the optimization of business. (Karoji et al., 2019) Using the behaviour of a pedestrian inside a mall as data, the researchers trained a behavioural predictor that can infer the pedestrian's forward direction based on the information of his current location and orientation. This model in turn guides the design of the mall, leading to higher commercial value on the pedestrian's expected route. In addition, given the inspiration of RNNs in content recommendation, some researchers have tried to use RNNs from the perspective of software operation. Toulkeridou describes a method to train RNNs to assist in parametric design decisions.

(Toulkeridou, 2019) In this study, the user's parametric

2

2

programming process in the Dynamo editor is recorded as a sequence as input. After machine learning, the RNN can predict the list of possible paths for the next step by judging the existing state. This method was later further extended to assist 3D modelling operations. (Gao et al., 2021; Gao et al., 2022).

Project Goal

The paper aims to explore the relationship between the order of store businesses along the street and their commercial vitality by a sequence-based neural network (RNN). In the machine learning model, the input is the sequence of store businesses on the street, and the output is the corresponding sequence of business vitality indexes.

After training, this machine learning model can predict the vitality of each store, thus guiding real estate business planning at a high resolution.

After obtaining the prediction model, this study selects a street sample outside the training set to verify the effectiveness of the model experimentally. We will observe the change in the prediction results of the model by moving the location of stores or adding or deleting stores in the street to see if the model is consistent with existing theories and real-life experiences.

Furthermore, we can combine this prediction model with a genetic algorithm to develop a business planning

3

2

Research framework

optimization tool: it automatically gives the best ranking order based on the input store types to maximize the business value of the whole street.

METHODOLOGY Overview Workflow

The whole research process is divided into three parts: data collection, model training and model evaluation (Figure 2). We collected data of stores along the street from O2O platforms such as Gaode Map, Meituan, and Dianping, and processed these data into a sequence that can represent the types of stores and their sales status. After that, the store type sequences are input into the seq2seq attention model, trained in the attention and LSTM layers through dictionary encoding, one hot encoding, embedding and other steps. The the model outputs the sequence of letters that can represent the vitality level through SOFTMAX and decoding layers. Finally, we use the Cross Entropy Loss Function and the prediction accuracy function to evaluate the effectiveness of this prediction model.

Data Collection

We selected 80 streets, 1261 stores, and 29 store types from 8 representative cities in China from O2O platforms (Figure 3 and 4). As the mainstream O2O platforms vary from city to city and different merchants on the same street might choose different O2O platforms, it was necessary to collate data from multiple mainstream platforms. In

3

4

this research, the commercial data was comprehensively collected on Meituan, Dianping and Gaode Map. In this way, we collect as complete data as possible for every store on each of the 80 streets. In the face of a tiny number of shops with missing data, we take the average of the nearby shops of the same type as a replacement for the missing stores.

O2O platforms provide a variety of information: shop type, number of reviews, per capita spending. There is also information on sales volumes (some semi-annual, some monthly).

Data Processing

Quantitative assessment of business vitality is very complex as no platform publicly provides information on the sales of every shop in the street. Based on the assumption that all shops have the same review rate, we can use the number of reviews multiplied by the per capita spend to estimate the sales of each shop. However, after research, we found that the type of shop significantly impacts the number of reviews. Milk tea shops, bakeries and fast-food restaurants tend to have very high review rates. In contrast, some support facilities such as banks and bicycle repair points have meager review rates though their presence can have a significant impact on the surrounding stores.

In order to provide a more objective assessment of the commercial viability of shops, a relative value comparison approach is applied here. For these 1261 shops, we ignore the effect of city and compare the number of reviews multiplied by the value of per capita consumption within each type of shop, and then classify their relative vitality into five classes: ABCDE. For example, there are 75 pastry shops, so we rank their vitality, then the top 10 are ranked A, 11-25 are ranked B, and so on (Figure 5). For those supporting facilities with few reviews like banks, we unify their vitality value C. After calculating the vitality values of the stores in these 80

5 3

Data set range

4

POI data statistics

5

Translating business data into relative vitality values

4

be trained in reverse, so the dataset was expanded from 80 streets to 160.

To expand the sample size further, we obtained all subsequences of length greater than five from the beginning of these 160 sequences (Figure 7). This is reasonable because we may not go through the whole street in daily shopping but finish shopping after passing several stores.

By this method, we obtained a total of 1820 sequential data. This method of expanding the database is inspired by the research of Weixin Huang's team on the modelling operation process, in which they also applied a similar subsequence approach (Gao et al., 2022).

Machine Learning

We base our training on the Seq2Seq attention model (Figure 1). The input data is 1820 sequences representing the order of shop types, and the output data is 1820 sequences representing the relative vitality level of stores.

The two are matched one by one. We divided the training set, validation set and test set according to the ratio of 7:2:1.

6

We evaluate the effectiveness of this prediction model by two functions, one is the Cross Entropy Loss Function and the other is the Prediction Accuracy Function. The Cross Entropy Loss Function is widely used in machine learning to evaluate probabilistic losses (Formula 1). In this RNN model, the machine is also evolving based on this metric.

M: Number of categories yic: Sign function (0 or 1) pic: The predicted probability that ith item belongs to category c 7 6

Comparison of city store vitality

7

Sub-sequences generation for the training set expansion

streets, we can get some interesting statistical conclusions.

Shanghai, Nanjing, Wuhan and Suzhou have higher average store vitality than Kunming and Changsha, which is in line with daily experience: store vitality is positively correlated with the economic development of a city (Figure 6).

Training Set Expansion

The machine learning model simulates the behaviour of people walking down the street and passing through stores that is a one-way experience. However, since the starting point of pedestrians is not specified, the sequences can all

Formula 1. Cross-entropy loss function The Prediction Accuracy Function is formulated by the specific problem of this paper. We compare the prediction value with the actual value and then quantify this accuracy (Formula 2). In assessing the accuracy, the difference between the predicted value and the target value is divided by the floor, which varies depending on the predicted value (Figure 8). The formula can be explained by the following case: Suppose the current target sequence is A, B, C, D, E. And the predicted sequence is C, C, C, C, C. Then the prediction accuracy is (0.5+0.66+1+0.66+0.5)/5=66.4%.

The accuracy of random guess is the sum of all the values in Table 3 divided by 25 equals 46.56%.

8

R: Range of vitality level np: Predicted sequence length nt: Target sequence length rip: Predicted vitality rit: Target vitality level of the ith term level of the ith term

11

Formula 2. Prediction accuracy function

13

12

8

We initially train with 30 batches per epoch (Figure 9). After training about 100 epochs, the model quickly goes into an overfitting state. Then we adjusted the parameters to train with 15 batches per epoch (Figure 10). This time the training effect is very thick, the model never enters the overfitting state, the training loss curve and the validation loss curve remain the same, and the accuracy curve keeps increasing.

9

Training results for batch size = 30

Accuracy calculation table

11 Vitality of stores in Gongyuan West Street

12 POI data statistics

13 Accuracy of prediction results

situation of the site is shown in Figure 11 and 12.

Experiment Validation

The types of stores in West Street were input into the trained model, and the output vitality prediction was "b c b c b c b c b c b c b c b c b c b c", with an accuracy of 77% according to Formula 2.

Experiment 1 added a movie theatre at the beginning of the street, and we could see that the model had a higher expectation of street vitality, appearing more b (Figure 14).

Experiment 2 arranges the same kinds of stores together, and we can see that the model also has a higher expectation of the overall vitality of the street (Figure 15).

14 Impact of adding a movie theater

10 Training results for batch size = 15

RESULTS AND DISCUSSION Case Study

15 Impact of clustering similar stores

We chose Gungyuan West Street in Nanjing, outside the training set, to apply our trained evaluation model.

Gongyuan West Street is in the historical centre area of Nanjing, with a wide variety of businesses and high popularity. Gongyuan West Street can be divided into two sequences of stores in the east and west. The commercial

Further, we combined this evaluation model with a genetic algorithm to develop a reference tool that can provide suggestions for optimizing the location of stores. The vitality levels correspond to specific numbers: A scores 5, B scores

Optimization Tool Based on Genetic Algorithm

4, C scores 3, D scores 2, and E scores 1. The genetic algorithm uses the total score of the predicted vitality sequence as the optimization target. At each iteration, the genetic algorithm randomly swaps two store locations for vitality reassessment. Through continuous iterations, the genetic algorithm then gives the optimal solution of this prediction model. Figure 16 records an evolutionary process.

After hundreds of iterations, the system did find a solution with a high vitality index: "CS F STS CS AS JS IS STS B HC B DH B H F FAFR". The vitality prediction for this sequence is: "B C B A A A A A A A A A A A A A A A A" with a score of 76.

CONCLUSION This paper presents a method that uses machine learning to predict commercial vitality along streets and provide optimization advice. This study has important practical value for high-precision business planning. Although there have been many machine learning studies based on urban texture images, few studies are accurate to the prediction of individual vitality of stores. Compared with previous studies, this study creatively interpreted people's walking and shopping behaviour in the street as a linear sequence.

It converted POI data collected from the O2O platform into a sequence format to train the RNN model.

In the future, this study still has much room for improvement. From the training graph and genetic algorithm optimization graph, it can be seen that the accuracy of the currently trained model is still not high enough, and it is easy to overfit. In the data collection stage, a larger data set is needed in the future. Since the information accuracy of the sequence of stores along the street is very demanding, the traditional POI information is challenging to sort accurately by geographical coordinates alone, so the current method is manual statistics. In the future, however, automated data collection algorithms will have to be developed to replace the current manual methods to remarkably expand the scale of the training set. In the data processing stage, there are many noise points in the data set due to many factors affecting the vitality of the real-world stores.

In the future, homogenized data algorithms will be used to eliminate the effect of noise (Sheng et al., 2018). In the model training phase, we can use more RNN models such as Transformer, GRU, BiLSTM to compare who is more suitable for this research in the future.

16 optimization process based on genetic algorithm

REFERENCES

IMAGE CREDITS All drawings and images by the authors.

Gao, W., Wu, C., Huang, W., Lin, B., & Su, X. (2021). A data structure for studying 3D modeling design behavior based on event logs.

Automation in Construction, 132(103967), 103967.

Gao, W., Zhang, X., He, Q., Lin, B., & Huang, W. (2022). Command prediction based on early 3D modeling design logs by deep neural networks. Automation in Construction, 133(104026), 104026.

Schlegel, A., Birkel, H. S., & Hartmann, E. (2021). Enabling integrated business planning through big data analytics: a case study on sales and operations planning. International Journal of Physical Distribution & Logistics Management, 51(6), 607–633.

He, J., & Zheng, H. (2021). Prediction of crime rate in urban neighborhoods based on machine learning. Engineering Applications of Artificial Intelligence, 106, 104460.

Sheng, Q., Zhou, C., Karimi, K., Lu, A., & Shao, M. (2018). The application of space syntax modeling in data-based urban design — an example of Chaoyang square renewal in Jilin city. Landscape Architecture Frontiers, 6(2), 102.

Karoji, G., Hotta, K., Hotta, A., & Ikeda, Y. (2019). Pedestrian dynamic behaviour modeling. In 24th International Conference on Computer-Aided Architectural Design Research in Asia: Intelligent and Informed, CAADRIA 2019 (pp. 281-290). The Association for Computer-Aided Architectural Design Research in Asia (CAADRIA).

Shou, X., Chen, P., & Zheng, H. (2021). Predicting the Heat Map of Street Vendors from Pedestrian Flow through Machine Learning.

In 26th International Conference on Computer-Aided Architectural Design Research in Asia: Projections, CAADRIA 2021 (pp. 569-578).

The Association for Computer-Aided Architectural Design Research in Asia (CAADRIA).

Sun, Y.J., Jiang, L., & Zheng, H. (2020). A Machine Learning Method of Predicting Behavior Vitality via Urban Forms. In 40th International Conference on Computer Aided Design in Architecture: Distributed Proximities, ACADIA 2021 (pp. 160-168).

The Association for Computer Aided Design in Architecture (ACADIA).

Xia, X., & Tong, Z. (2020). A Machine Learning-Based Method for Predicting Urban Land Use. In 25th International Conference on Computer-Aided Architectural Design Research in Asia: Anthropocene, CAADRIA 2021 (pp. 21-30). The Association for Computer-Aided Architectural Design Research in Asia (CAADRIA).

8

Topological Networks Using a Sequential Method

Zidong Liu Bartlett School of Architecture University College London

Data Structure Simplification for Interactive Design Accepted by ACADIA 2021, Presented at ACADIA 2021 on Nov.6 2021.

1

ABSTR ACT The paper shares preliminary results of a novel sequential method to expand existing topology-based generative design. The approach is applied to building an interactive community design system based on a mobile interface. In the process of building an interactive design system, one of the core problems is to harness the complex topological network formed by user demands. After decades of graph theory research in architecture, a consensus on self-organized complex networks has emerged. However, how to convert input complex topological data into spatial layouts in generative designs is still a difficult problem worth exploring. The paper proposes a way to simplify the problem: in some cases, the spatial network of buildings can be approximated as a collection of sequences based on circulation analysis. In the process of network serialization, the personalized user demands are transformed into activity patterns and further into serial spaces. This network operation gives architects more room to play with their work. Rather than just designing an algorithm that directly translates users’ demands into shape, architects can be more actively involved in organizing spatial networks by setting up a catalogue of activity patterns of the residents, thus contributing to a certain balance of top-down order and bottom-up richness in the project. The research on data serialization lays a solid foundation for the future exploration of Recurrent Neural Network (RNN) applied to generative design.

2

ACADIA 2021

1

Design Result of Living Cluster and Community

INTRODUCTION In interactive generative design, a common problem is harnessing the topological network formed by user demand information. After data collection, user information is transformed into two parts: topological and geometrical (Damski and John 1997). If designers set too many parameters in collecting data, they face a complex network that is hard to be directly geometrized. Therefore, the method to analyze and reorganize the topological network for data distribution is significant.

2

The research aims to explore a sequential approach to organizing the topological network, thus reducing the complexity of data distribution and expanding the possibilities of generative design. The prototype system presented in this paper is based on a mobile interface, providing an interactive platform for people to design their houses and community. The traditional top-down model of community planning relies heavily on the subjectivity of architects, which leads to waste of land resources and lack of community vitality, exacerbating the housing crisis. (Wong, 2010).

This research takes the view that architects should not only be the designers of space but the developers of interactive platforms that facilitate inclusive design.

Topological Design

The research on architectural topological design has a decades-long history. In its early development period, the approach is based on simple topological rules, such as floorplan automation (Levin 1964), cellular automation (Wolfram 1983) and shape grammars (Stiny 2011). The booming development of network science provides an increasingly rich set of tools for spatial network analysis (Barabasi 2016; Turnbull et al. 2018). These studies strongly support the development of architecture layout automation and gradually extend its boundaries from rectangular layout to non-rectangular layout. (Marson and Musse 2010; Koenig et al. 2012, 2020; Shekhawat and Duarte 2018; Wang et al. 2018, 2020). However, most current topology-based generative designs directly transform topological networks into geometric forms, and lack secondary processing of spatial networks. When faced with overly complex networks, it is difficult for the architect to control the spatial form. Among the few studies on this issue, the computing efficiency of different topologies has been discussed (Koenig and Knecht 2014).

Sequential Method

The sequential method means interpreting the topological network from the perspective of activity sequence (Fig. 2). In some projects with distinct circulations, the spatial network can be approximated as the interweaving and

TOPIC (ACADIA team will fill in)

3 2

Sequential Interpretation of Complex Network.

3

Sequence Analysis of Topological Network of Soane Museum.

REALIGNMENTS

3

4

6

5

overlapping result of multiple activity sequences. The sequence acts as an intermediate hierarchy to cluster nodes and form the system, where the embedding vectors of nodes are close to each other mathematically.

The idea of activity sequence can be found in centuries of architectural practice, such as Vasari Corridor (1565), Soane Museum(1813) and Miller House(1953). The Soane Museum is a classic case of using sequential approach to organize floor layout. It serves multiple functions simultaneously within a narrow interior space: family life, party, teaching, exhibition and so on. Besides, due to the historical background, people with different identities had their own specialized paths, such as the specific logistics circulation for servants. Sir John Soane used a narrative approach to organize these various life patterns (Kyriafini, 2007). The topological analysis of space reveals that the structure of network is composed of interwoven sequences (Fig. 3).

The sequential approach has the potential to be applied to community design as well. Today’s community design is faced with the similar design problems as the Soane Museum: complex functions, interwoven circulations, customized living patterns, and adaptability for flexible use.

METHODOLOGY Overall Method

The sequential approach can be applied across scales, from urban to architectural scales. In the paper, the sequential approach is demonstrated with a community design experiment aiming at housing customization and

4

ACADIA 2021

7

Paper Title Author last names, separated by commas

4

Overall Logic Structure of the Generative Design System.

5

Data Translation for Community Design Part.

6

Grouping Residents into Living Cluster by K-means Algorithm.

7

The paper takes ten users as an example to illustrate the algorithm logic of sequential method.

8

Transforming 200 Hypothetical Users’ Demands into Sequential Network.

8

community self-organization (Fig. 4). An interactive platform based on a mobile interface is built on the front end for user data collection and design result preview. The process of transforming topological and geometric information into community design is carried out in the back end. The entire generative design system consists of three parts: site selection, community design and housing unit design. This paper focuses on sharing the results of community design part where the sequential approach is mainly used.

User Data Collection

User demands includes dwelling characteristics, neighborhood, surrounding environment and facilities, social identity and so on (Roy et al. 2018). To quantify the input variable, these influencing factors are reclassified into three categories: • • •

Housing unit design: room size, function types.

Community design: shared rooms, community facilities.

Site decision: housing prices, distance to work, urban facilities.

This experimental design is based on data from 200 simulated users. To maximize system complexity and verify the effectiveness of the sequential approach, each virtual user data is completely randomized. After data collection, these

TOPIC (ACADIA team will fill in)

mixed-up user demands will be translated and redirected to the different parts of the generative design system (Fig. 5).

Clustering Residents

The data about the community design is the most complicated one, because the type and number of shared rooms and community facilities for each user are uncertain. Due to the limitation of computing power, it is impossible to input 200 user data simultaneously for generative design, and structural hierarchies are necessary (Koenig and Schneider 2012). Therefore, the virtual 200 users are classified into 20 living clusters based on matching degree, which is evaluated by 3 parameters: house similarity, demands for sharing rooms and community facilities.

The pairing process is based on K-means Clustering algorithm (MacQueen, 1967) (Fig. 6). In this algorithm, each resident is regarded as a piece of three-dimensional data, mathematically equal to a point with 3 coordinates. The weighted approach is used to distinguish the importance of different attributes.

Sequencing Topological Network

People living in the same cluster share the similar living

REALIGNMENTS

5

a9 are considered as a living cluster, forming a sequence with offices and study rooms. The remaining five users and facilities form another sequence. It should be noted that this classification is only an approximation. For example, we ignore the need of a3 for offices. So we need to provide a common sequence connected to the public transport network, which facilitates the residents to access other places.

9

In order to strike a balance between diversity and simplicity, the following rules are set: a cluster has a maximum of 2 personalized activity sequences and a common sequence.

Based on these rules, 20 clusters generate 52 activity sequences of 5 categories in total, including 35 and 20 common sequences. These sequences merge and separate, generating a total of 106 supporting facilities (Fig. 8).

Physically Based Design Constraint

10

12 9

Physics Engine Simulation for Community Structure Organization.

10 Physics Engine Simulation for Units Locating inside Living Cluster.

12 Four Steps of Detail Automation in Living Cluster Generation Process.

patterns and infrastructure demands. For example, they might take a walk in the community park after work, and then exercise at the community gym before returning home.

We regard such a life pattern as a type of activity sequence.

The algorithmic logic of the sequential approach is demonstrated here using the example of ten users(Fig.7). Suppose there are ten users from a1 to a10 who have personalized needs for four facilities, which form a complex network that is difficult to geometrize. But after clustering, we find that a1, a2, a4, a6, a9 have similar needs for offices and study rooms, while the remaining five users have needs for shared kitchens and living rooms. Thus, a1, a2, a4, a6,

6

ACADIA 2021

We use the physics engine to set the design constraints translated from the topological network. This approach is widely used in generative design (Scott and Donald 1999, 2002; Koenig et al. 2012; Daniel et al. 2013). The advantage of this analogy is that the strength of connection can be adjusted freely and multiple forces can be applied to one object without restriction.

Elements are restricted to move within the community boundary by fb. They have repulsive forces fa against each other to maintain their own space. There is a strong attraction fc between adjacent elements in a sequence to ensure that they stick close to each other. In addition, adjacent elements in a sequence have rotational forces f_d to keep the angle between them at 0 or 90 degrees, thus keeping the layout structure orthogonal (Fig. 8).

In the process of generating the internal layout of a living cluster, highly matched living units will have an attraction in the middle, otherwise, they will be mutually exclusive. Paths automatically follow the change of cells’ location (Fig. 9).

Optimizing Design Result

Physics engine can only roughly simulate the relationship between elements because the results are very easy to be affected by the initial position of elements. In order to overcome this uncertainty, genetic algorithm is used to optimize the result based on specific performance criteria P. Here, we optimize the result by minimizing l_n and maximizing b_n (Fig. 11). The parameter f and g is used for weighting.

By this way, more compact floor plans will be screened out. This process is based on an interactive platform Biomorpher (Harding and Brandt-Olsen 2018).

Paper Title Author last names, separated by commas

11 Using Genetic Algorithm to Get a More Compact Floor Plan.

Geometrizing Details

After determining the location and boundary of elements, the system will further automatically complete the detailed design of the public space and living units. (Fig. 12)

RESULTS AND DISCUSSION Based on 200 hypothetical users’ data, the prototype system presented in this paper obtains a complete community design result with diverse architectural space emerging (Fig. 13). In this generative design process, the sequential approach, as a novel design methodology to organize the topological network, plays a key role in the community design step. This approach creatively translates the individual demands into different life patterns and corresponds them to specific sequences of activities.

Consequentially, the complex network is transformed

TOPIC (ACADIA team will fill in)

into interwoven sequences, successfully reducing the complexity of the problem.

At present, the system is still in the preliminary explorative development. The current process of serializing networks is highly subjective. RNN will be further explored in the future due to its huge potential for processing sequential information. In the process of simulating the sequential network, we have found that the traditional physics engine simulation requires considerable refinement due to the instability of the results.

CONCLUSION This paper verifies the effectiveness of a novel sequential approach in organizing topological networks in generative design process. The importance of sequential approach is

REALIGNMENTS

7

that it provides a novel and effective paradigm for designing complex networks. Architects and planners have developed various methods to quantitatively analyze existing complex networks, but how to create complexity remains a daunting task. Design is always faced with the challenge of neither being monotonous or chaotic. Sequence is a new perspective for people to control complexity. As a bridge, the sequence powerfully connects the mathematical topological network with the vitality of individual human patterns of life.

ACKNOWLEDGEMENTS The work in this article is part of Housingprime, a research project of RC11, B-Pro, the Bartlett, UCL 2020 led by Philippe Morel and Paul Poinet, with members including Zidong Liu, Shiyuan Huang, Yi Li and Huiyu Pan. I would like to thank our tutors Philippe and Paul who constantly gave us inspiration and guidance. I would also like to thank my team members for their company and hard work. Last but not least, I would like to thank Professor Ruairi Glynn for his many hours of close guidance. This paper would not have been possible without his help.

11 11 Using Genetic Algorithm to Get a More Compact Floor Plan.

8

ACADIA 2021

Paper Title Author last names, separated by commas

REFERENCES Arvin, Scott A., and Donald H. House. 1999. “Making Designs Come Alive: Using Physically Based Modeling Techniques in Space Layout Planning.” In Computers in Building, 245–62. Boston, MA: Springer US. Arvin, Scott A. and Donald H. House. 2002. “Modeling Architectural Design Objectives in Physically Based Space Planning.” Automation in Construction 11 (2): 213–25.

Barabasi, Albert-Laszlo. 2016. Network Science. Cambridge, England: Cambridge University Press.

Damski, José C., and John S. Gero. 1997. “An Evolutionary Approach to Generating Constraint-Based Space Layout Topologies.”

In CAAD Futures 1997, 855–64. Dordrecht: Springer Netherlands.

Gavrilov, Egor, Sven Schneider, Martin Dennemark, and Reinhard Koenig. 2020. “Computer-Aided Approach to Public Buildings Floor Plan Generation. Magnetizing Floor Plan Generator.” Procedia Manufacturing 44: 132–39.

Harding, John, and Cecilie Brandt-Olsen. 2018. “Biomorpher: Interactive Evolution for Parametric Design.” International Journal of Architectural Computing 16 (2): 144–63.

Koenig, Reinhard, and Katja Knecht. 2014. “Comparing Two Evolutionary Algorithm Based Methods for Layout Generation: Dense Packing versus Subdivision.” Artificial Intelligence for Engineering Design, Analysis and Manufacturing: AI EDAM 28 (3): 285–99.

Koenig, Reinhard, and Sven Schneider. 2012. “Hierarchical Structuring of Layout Problems in an Interactive Evolutionary Layout System.” Artificial Intelligence for Engineering Design, Analysis and Manufacturing: AI EDAM 26 (2): 129–42.

Algorithm.” International Journal of Computer Games Technology 2010: 1–10.

Piker, Daniel. 2013. “Kangaroo: Form Finding with Computational Physics.” Architectural Design 83 (2): 136–37.

Roy, Noémie, Roxanne Dubé, Carole Després, Adriana Freitas, and France Légaré. 2018. “Choosing between Staying at Home or Moving: A Systematic Review of Factors Influencing Housing Decisions among Frail Older Adults.” PloS One 13 (1): e0189266.

Shekhawat, Krishnendra, and José P. Duarte. 2018. “Introduction to Generic Rectangular Floor Plans.” Artificial Intelligence for Engineering Design, Analysis and Manufacturing: AI EDAM 32 (3): 331–50.

Stiny, George. 2011. “What Rule(s) Should I Use?” Nexus Network Journal 13(1): 15-47.

Turnbull, Laura, Marc-Thorsten Hütt, Andreas A. Ioannides, Stuart Kininmonth, Ronald Poeppl, Klement Tockner, Louise J. Bracken, et al. 2018. “Connectivity and Complex Systems: Learning from a Multi-Disciplinary Perspective.” Applied Network Science 3 (1): 11.

Wang, Xiao-Yu, Yin Yang, and Kang Zhang. 2018. “Customization and Generation of Floor Plans Based on Graph Transformations.”

Automation in Construction 94: 405–16.

Wang, Xiao-Yu, and Kang Zhang. 2020. “Generating Layout Designs from High-Level Specifications.” Automation in Construction 119 (103288): 103288.

Wolfram, Stephen. 1983. “Statistical Mechanics of Cellular Automata.” Reviews of Modern Physics 55(3): 601-644.

Wong, Joseph Francis. 2010. “Factors Affecting Open Building Implementation in High Density Mass Housing Design in Hong Kong.”

Habitat International 34 (2): 174–82.

Kyriafini, Magdalini. 2007. “Narrative and Exploration in Small Museums: The Wallace Collection and the Soane Museum.” PhD

IMAGE CREDITS

diss., University College London.

All drawings and images by the authors.

Levin, Peter Hirsch. 1964. "Use of graphs to decide the optimum

Zidong Liu holds a Master of Architecture degree from the

layout of buildings." The Architects' Journal 7: 809-815.

University College London and a Bachelor of Architecture degree from the Southeast University. His research interests focus on

MacQueen, James. 1967. “Some Methods for Classification and

spatial network-based generative design, urban analysis, and

Analysis of Multivariate Observations.” Proceedings of the Fifth

machine learning-based urban feature prediction.

Berkeley Symposium on Mathematical Statistics and Probability 1(14):281-297.

Marson, Fernando, and Soraia Raupp Musse. 2010. “Automatic Real-Time Generation of Floor Plans Based on Squarified Treemaps

TOPIC (ACADIA team will fill in)

REALIGNMENTS

RESEARCH PROJECTS

02

01

Housing Prime

A Digital Platform for Housing Customization and Community Self-organization Diploma Project of RC11, Urban Design, B-Pro, Bartlett

Type: Urban Design/Group Work Site: London, UK Date: Sep.2019 to Sep.2020 Supervisor: Professor. Philippe Morel Dr. Paul Poinet

The housing crisis is a serious phenomenon worldwide that may lead to severe problems for society. Though, in the information age, the design approaches in housing seem still remains stagnant and unable to adapt to the dynamic changing human activities. The project aims to develop an application for both architects and users to create an agent-based platform to build a complex adaptive housing system. Through the application, customers are able to design their own customized housing with the help of architects in the back end, so as to combat continuously changing demands and expectations for more and better in homes.

1. Back-end Side

Social Background

Housing crisis is a serious and common phenomenon all over the world. Although the number of houses and homeownership rates are constantly increasing in recent years, the homeowner vacancy rates has remained high due to some reasons, which resulting in the occupancy rate of houses has not been improved. As a result, the problem of housing crisis has not been solved.

In spite of the fact that there is a slowing trend of population growth worldwide, the reduction of avertage household size in the past dacades has led to a continuous increase in the demand for housing, which as a result cause more serious housing crisis.

Every generation is shaped by its circumstances, and millennials are no exception. They’re no less ambitious than previous generations: More than half want to earn high salaries and be wealthy. But their priorities have evolved, or at least been delayed by financial or other constraints. Having children, buying homes, and other traditional signals of adulthood “success markers” do not top their list of ambitions.

Millennials, the Biggest Victim

Case Study Market Survey

Top Apps of Design

2D

2.5D

3D

There are Top 60 Apps of interior and home deisgn, and we categorized them based on 3 dimensions. as we can see, the 3D apps occupy 50% of these apps, so customer would prefer the 3D tyoe which could Help them get more intuitive feedback.

Its no doubt that apps for deisgn home have quite potential market, and inspaired us to design a platform to help people deisgn by themselves.

Project planning

Data from customer

Detailed Design Basic demand and preferences (size, function, style, community)

Site decision Housing storage

Construction

Media-platform

Tracking the process of manufactring

Prime factor

Combination Customer Feedback

Delivery

New house for customer

CIty Analysis Automatic collection of facility data

Operation Mechanism Firstly, we map these lands for sale, data of which we collect from website by python, and then we need to filter them. According to Clarence Perry’s neighborhood unit theory, we set the 400-meter circle as an evaluation area for each land.We try to map the information of different facilities Including the amenities, education, transportation and sports.Quantify them in each evaluation area of lands to get the lands with higher score, which means with much more convenience.

Site Decision Divide and filter

Quantify the quality of areas and make decision

Unit Layout

Housing C

Operation Mechanism Firstly, we map these lands for sale, data of which we collect from website by python, and then we need to filter them. According to Clarence Perry’s neighborhood unit theory, we set the 400-meter circle as an evaluation area for each land.We try to map the information of different facilities Including the amenities, education, transportation and sports.Quantify them in each evaluation area of lands to get the lands with higher score, which means with much more convenience.

Furniture Folding Furniture

Construction details

House unit generation For house unit, based on the customer’s house hold type, we give the proposal to satisfy their basic needs and the classified room size. we input more specific rules, like the wall side, the window side, the initial boundary for generation. Then we could generate the unit based on the strength of the linkage between different rooms.The customer could also choose their furniture on the platform. We provide folding furniture, you can choose different size, color and organize them as you need. In short, you could choose to live in a house with more flexibility.

ustomization

Options

Comments

Material

Lving Cluster Organization

Shared Functions in Living Cluster We have 2 types of clusters, one is public cluster, another is living cluster. In the living cluster we provide users variety of functions to choose and classify them into different groups based on their choices.

Step 1

Step 2

Step 3

Step 4

Layour Automation This is a dynamic algorithm to locate and circulate housing units according to their preferences.The residents who enjoy socializing will stay closer to public space.

Living Cluster Generation Shape Optimization Then, we optimize the inner space based on the circulation routes.

Step 1

Step 2

Step 3

Liv Structure of cluster

S

After that we got a global structure of living cluster and here are the circulation routes inside the cluster.

Shared Garden

Cafe

Automatically Circulating

ving Clusters

Show

Exhibition

Resataurant

Private Garden

Main Entrance

Public Cluster

Office Working space is aim for our target groups, who are dedicated to their works. It is easy for them to access to.

Living Cluster Garden The green space goes through the entire building, which enables people a place to relax and socialize with others.

Fitness Center Fitness centre are located in each clusters and common spaces.

Mall The shopping centre is located in the lower part of the building, providing residents daily supplies.

Besides living clusters, we create some types of public clusters. They are professional for one collective function.We design another algorithm. Each time, we input the shape of floor, it will generate circulation and then create different-sized rooms along the circulation.Getting the data of layout, the algorithm will automatically generate all details from stairs to furniture.

Step 1

Step 2

Step 3

Step 4

Cluster Customization Based on customers data, we would offer several cluster types as options, and people could do basic design on them.In addition, customers could share their design to other media-platform.

Community Generation

Force SImulation The next step of whole project is about combining different clusters in a community. We use the idea of activity sequence to organize the structure of clusters. We use physical engine to simulate their connections. After running, finally, we can get the position of each cluster.

Positions of clusters

Volume Generation

Genetic Algorithm The process of physical simulation has some uncertainty, so it is necessary for us to use genetic algorithm to optimize the result. Our standard is simple, if the total length of paths is shorter, the structure of whole building is more compact, the result is more reasonable.

Step 1

Step 2

Step 3

Step 4

“Green Hole”

From Point to box Then next step is to transform the points to mass.

Each point will occupy a number of closest boxes according to the cluster area. We design an algorithm to avoid overlapping when two point is too closer.

Volume of Cluster

rs

Paths of Community

2. Using Side -- Android as Front-end development This chapter will demontrate how the app(Housing-prime) could give customers better experiences, and how these serise of UI coming with Andriod technically.

13

2.1 Housing-Prime Using flow

Technical workflo

The using flow would be divided into four sections, First of all, customers would get some tips for using this app, and then is follwed by the site decision, house design, and shared platform.

14

Using Guide

This is the more logical UI of APP its site decision part

ow

Its housing decision part,you could choose some basic information, such as the household type

Then you could participate in some basic housing design choosing the function,material ect

Finally,you couldshare the design to some mediaplatform,you could inspair others to join your design community 15

2.2 Guide Pages LoopViewPager : Giving some guidelines for customers through Infinitely looping

Layout file

Java file (more details would be shown in chapter)

16

17

EditText : The user enters specific information through the keyboardbox

if else : Determine if it is a valid password

Radio Box : The user could choose certain information

18

19

2.3 Site decision Pop Window : A new dialog will pop up after the user clicks

20

Our project focus on 4 main cities, which are quite popular place among young people. and for each city, we have several sites with different theme.

21

2.4 Site decision Recycler View : Use tag flow to help customers mark their Location Preferences,

22

We would recommond different site based on the customer's preferences, for example, someone give priority to the transportation, their new house will be arranged to the one near subway.

23

2.5 Housing Decision Toast : According to the user input information to give the corresponding prompt box

24

Then, customer could choose how long they wo u l d l i ke t o s t ay i n our community and the household type. Based on these information ,we would offer various options, if a customer come with core family and want to stay a couple of months, they would have higner probability to have housing alternative option with bigger size or extra storage space etc···

25

OnClickListener : Record customer behavior by listening events

26

In addition, they could choose which function they want in their home, and if they want to share some of them with others, for instance, we would offer a collective kitchen in a cluster with many people who want to share kitchen with others.

27

2.6 Housing Design

28

Customer will recive 5 options to macth their options before. They can choose 1-2 options they are satisfied with and give comments, such as the size of certain rooms. Finally, The server will select the final plan for the client Besides, client also have chance to do some interior deisgn by themselves.

29

2.7 Sharing

Now, Customer's deisgn has done and be packged to delivery to the site.

Client could build their own living coummnity with new network, to share their design to mediaplatform, or find friends with similar interests in the community.

30

31

If you want to get more information,please click on :

https://youtu.be/Q2L0C1mNNyY

Overall, human could have extension through media-platform, the internet help people reshape their life and the relationship between them and the world.The app give customers a nwe way to consider their home and even help them particapate into into the deisgn process, rather than just passively accept the plan from real-state or so-called professionals. This means that public have a chance to get the real sence of belonging of home by the result of their own work towards their house.

215 HOUSING PRIME

SKILLS EXPLORATION

03

01

Facade Styletransfer based on CycleGAN

Paris to Barcelona For more information, please vist https://github.com/wodeyyf/cycleGAN-ParisToBarcelona

The object of this machine learning research is cycleGAN, currently the hottest in the field of architecture. 2019, Harvard Stanislas Chaillou's GAN-based housing layout research has had a very important impact on the field of architectural computing. He launched ArchiGAN, which led the way for the entire architectural AI research, and image-based generative design for architecture is still an absolute hot topic to this day.

However, since there has been a lot of research in the field of GAN-based architectural AI, I simply learned how to use GAN through the example of building façade style migration without further publishing it as a paper. I think new opportunities appear in GCN and RNN (LSTM).

Code Sample

Train Result

02

Game Design based on Processing

an Environmental Awareness Game about CO2 Reduction For more information, please vist https://github.com/wodeyyf/co2generatorGame1031

This is a game design that I completed with my students as a teaching assistant at the UCL Creative Design and Interactive Architecture Research Workshop held by Professor Ruairi Glynn.

While there has been a great deal of work done on data visualization to help people understand the environmental urgency. But people have never been able to truly appreciate the meaning behind the boring numbers. We developed this survival game that translates CO2 emissions per capita into anxiety values for players so they can empathize with environmental degradation. Players will choose a specific country as their place of origin and then experience the change in CO2 emissions per capita in that country from 1978 to 2018. We represent the value of CO2 emissions per capita in terms of the number of white blobs. Players need to dodge the white blobs to survive.

Although this is a very small game, it involves many aspects such as data reading, data output, game interface design, and game rule design. I learned a lot from this project.

Screenshot of the Game Interface

03

Mathematica Programming

Systematic Training on Functional Programming Part of Skill Course in Master of Urban Design, B-Pro, Bartlett For more information, please vist https://github.com/wodeyyf/MMA-Training

3.1 Data Collection and Reading In addition to direct data search, the reading of existing data and the conversion between different platforms are also important means of obtaining data.

Read the csv file to get the relevant information of the target area, and then get the map information from the internet.

Overlapping two kinds of information to get the data analysis we want.

Gismo in GH It is another method to import network map information.

More importantly, it can directly import the building geometry information in the block and directly connect with GRASSHOPPER. This provides us with an efficient platform for large-scale urban site analysis.

Site Data Extraction

3.2 Data Manipulation

MMA is a programming software widely used in scientific research.Compared with common languages such as python and Java, Wolfram is more concise and has more powerful mathematical functions.

We architects study this language mainly for developing our mathematical thinking.

Pure Function.

Conditional Function.

Replacing Function

Logic Gramma

Mapping Function

Foldlist Function

A complex function based on Foldlist Function.

Step 1 Create Initial List

Step 4 Evaluation

Step 2 Random Mutation

Step 5 Selection

Step 3 Crossover Function

Step 6 Create Loop

Genetic Algorithm

Step 7 Pack Final Function

In addition to manipulating data, we sometimes use algorithms to optimize the results.At this time, genetic algorithm is a powerful tool.

Geometry Calculation Sometimes we need to convert the geometric figure to algebra for calculation and solution to obtain useful geometric information.

Geometry Calculation Sometimes we need to convert the geometric figure to algebra for calculation and solution to obtain useful geometric information.

An algorithm study for finding intection between 2 lines.

An algorithm study for testing whether a point is in a circle.

Data Visualization MMA is mainly a data calculating software, but it also has a strong ability to draw graphics, very efficient when dealing with specific problems.

04

Speckle

An online Grasshoper collaboration platform Part of Skill Course in Master of Urban Design, B-Pro, Bartlett

Speckle is an advanced platform for gh file sharing. It makes Real-time data transmission in GH possible. I am sure that this platform will greatly improved teamwork efficiency in the future.

Step 1 Site Data Recieve

Step 3 Site Data Input

Step 5 Volumn Data Input

P1

P2

P3

Step 2 Site Data Output

Step 4 Volumn Data Output

Step 6 House Data Output

For more information, please vist https://vimeo.com/417049622

DESIGN PROJECTS

04

01

HFC Central Tower

A Natural High-tech Landmark(358m) Shenzhen General Institute of Architectural Design and Research Co. Ltd.

Situation: under Contruction Site: Shenzhen, CN Date: July.2019 to Now Leader: MENG Jianmin WU Chao

The housing crisis is a serious phenomenon worldwide that may lead to severe problems for society. Though, in the information age, the design approaches in housing seem still remains stagnant and unable to adapt to the dynamic changing human activities. The project aims to develop an application for both architects and users to create an agent-based platform to build a complex adaptive housing system. Through the application, customers are able to design their own customized housing with the help of architects in the back end, so as to combat continuously changing demands and expectations for more and better in homes.

02

Taiping Waterway Bridge

The Beauty of Clean Mechanics Shenzhen General Institute of Architectural Design and Research Co. Ltd.

Situation: Competition Site: Shenzhen, CN Date: June.2020 to July.2020 Leader: MENG Jianmin YI Yu

This proposal is based on the crane, and the power of the bridge is condensed in a few strokes on the bridge tower with a minimalist approach. The geometric angles are reinforced with origami-like folds, highlighting the strength of the bridge structure.

The scheme shows the cranes singing in the nine gorges, metaphorically showing that all the talents from all over the world will come here to create a splendid new area.

03

Foshan Nanhai Art Center

Sail, Wind, Sun and Sea Shenzhen General Institute of Architectural Design and Research Co. Ltd.

Situation: Competition Site: Shenzhen, CN Date: Feb.2020 to Apr.2020 Leader: WU Chao OT Architects

Ta k i n g t h e v a l u a b l e o p p o r t u n i t y o f t h e F o s h a n N a n h a i A r t C e n t e r I n t e r n a t i o n a l Competition, we hope to create a large public residential cultural center that integrates a grand theater, museum, exhibition hall and stadium. In this project, we have an ind e p t h d e s i g n s t u d y o n t h e f u n c t i o n a l f l o w o f t h e t h e a t e r, m u s e u m a n d e x h i b i t i o n hall, and there have been many discussions on the curtain wall construction.

04

Guangzhou Substation

An Attempt at a Futuristic Style Shenzhen General Institute of Architectural Design and Research Co. Ltd.

Situation: under Contruction Site: Shenzhen, CN Date: July.2020 to Now Leader: MENG Jianmin YI Yu

Substation is a very rare type of building. We were given full freedom to play with this p r o j e c t b e c a u s e t h e s u b s t at i o n i s n ot fo r h u m a n u s e , b ut fo r m a c h i n e u s e . D u r i n g t h e d e s i g n p r o c e s s , w e ex p e r i m e n t e d w i t h m a n y v a r i a t i o n s o f t h e f u t u r i s t i c s t y l e .

05

Shenzhen Baoan Archives Center

Geometry and Shadow Shenzhen General Institute of Architectural Design and Research Co. Ltd.

Situation: under Contruction Site: Shenzhen, CN Date: Octv.2019 to Now Leader: MENG Jianmin YI Yu

This design has three strategies: establish appropriate façade rhythm to render the building an elegant urban interface; integrate the programs in a compact way while creating a relaxing spatial atmosphere; create a Double-Foyer programmatic structure to provide more efficient circulation.

The form is developed based on the geometric vocabulary of rectangle and trapezoid. The different density of the shapes indicates the level of publicity and privacy. Meanwhile, the dynamics of trapezoids juxtapose solid and void volumes, and generate various gradients of light and shadow.

06

Shenzhen Reform and Opening-Up Exhibition Hall

Waves of the Times Shenzhen General Institute of Architectural Design and Research Co. Ltd.

Situation: under Contruction Site: Shenzhen, CN Date: Nov.2020 to Dec.2020 Leader: MENG Jianmin YI Yu

The project has an important political symbolism. The design is inspired by the "Lead the Wave of Reform". Shenzhen has created an unprecedented economic miracle in the history of mankind in just forty years, behind which there is a tremendous innovative energy and a spirit of struggle.

Portfolio LIU ZIDONG Selected Works 2018 - 2021

Shenzhen Architectural Design Institute Graduate Student in Master degree at the Bartlett, UCL Copyright 2021 by LIU Zidong All Rights Reserved.