A Multi-source Data And Machine Learning Based Method For Generating Urban Spatial Layout

Since the founding of the People’s Republic of China,urban development has undergone several transformations.Today’s urban construction has to consider more complex and diverse factors,and the urban spatial layout tends to be mixed.The traditional experience-based urban planning and design methods have certain limitations,requiring new technologies.Since the emergence of computer technology,it has been widely used in the urban field.At first,the computer acted as a drawing assistant.Then,various technologies like geographic information system were invented,offered big help in urban research and analysis.And now,at the current information age,machine learning and deep learning technologies become the new hot,opening up a new stage for urban planning and design.Based on Tobler’s First Law of Geography,this paper pays attention to the relationship between the target plot and the surrounding environment,using the Artificial Neural Network(ANN)to learn the hidden rules among the city’s multi-source data,and explore the city’s logic of construction and development.The computer can then generate a spatial layout plan for the target plot based on this logic,aiming to assist the existing manual urban planning and design method.The method for generating urban spatial layout has three parts:data preparation,single-grid basic algorithm development and multi-grid expansion algorithm development.In the first part,the basic research unit is defined as the data grid obtained by dividing the research area.The research data is collected from three dimensions:land use,urban form,and traffic.After processing,the dataset input into the ANN is generated.Train and optimize the neural network and obtain a basic single-grid model.Then,based on two different algorithms of Monte-Carlo Tree Search(MCTS)algorithm and adjacent grid filling(AGF)algorithm,the result of multiple unknown grids is generated,which means the scope of application of the method is expanded.After the method is constructed,a case study of Nanjing is carried out.Select the study area,collect and process data,train the ANN model,and choose three different areas for single-grid and multi-grid testing.The tests’results verify the feasibility and effectiveness of the whole method.Finally,an algorithm analysis and a result analysis are carried out to evaluate this method for generating urban spatial layout.The algorithm analysis proves that the overall performance of the ANN algorithm is satisfactory,and the performance of the AGF algorithm is better than that of the MCTS algorithm.The result analysis proves that this method can generate an urban spatial layout plan with reference value,and by discussing the relationship between the data of the target and the surrounding,several reasonable rules of urban spatial layout are summarized.The whole thesis contains about 42 000 words,95 pictures and charts.