(A) method of automated urban planning using the generative adversarial network model

Abstract

With the development of information and communication technology (ICT), big data, artificial intelligence (AI), and machine learning have made remarkable progress, and are actively penetrating urban research. In particular, research using generative adversarial networks (GANs) that show high performance in generating images is drawing attention in urban design and urban design research. In this study, I explore the possibility of an AI model that automatically generates land use planning in Korea, focusing on the ’overlay zoning’ policy in Korea and the schematic and visual characteristics of land use planning. To this end, data processing is performed to produce a land use planning tiles dataset. To train the GAN-based pix2pix model, a question-answer set is created with a doughnut-shaped hole in the center. After filtering and augmenting the data for better performance models, I have succeeded in confirming that the pix2fix model implements the context of land use and the features of the road network, similar to the surrounding plans. It is expected that AI can be used for actual land use planning with better performance if it is trained with more amount of data with the questions formed with the shape of real districts so that it can handle the questions from various regions.