Development of an AI advisor for conceptual land use planning

Abstract

To efficiently manage a time-consuming and expensive process in traditional urban planning, we aim to develop an artificial intelligence (AI) advisor that assists nonprofessionals participating in the urban planning process by using a generative adversarial network (GAN). This study presents the process of developing the AI model, which suggests appropriate land use plans for user-targeted sites and urban density scenarios. We first create an image dataset that embeds land use, floor area ratio (FAR), and building cover ratio (BCR) information in the RGB channel. Then, an algorithm is developed for establishing an optimized training set with 1000 images and methods for validating the obtained results from both quantitative and visual perspectives. We set up a pilot test to generate three urban density scenarios in Sewoon-Sangga district by constructing urban data-encoded image datasets of Seoul. The pilot test results reveal that our proposed model successfully suggests appropriate land use plans according to the three scenarios. Based on the pilot test, the AI advisor improves its output, training performance, and usability by reflecting block morphology with the Hamming distance and accepting user -designed road patterns. We expect that the novel approach developed in our study will contribute to research on AI-based urban planning.