Intelligent Generation Of Urban Block Road Network Based On Deep Reinforcement Learning

Introducing artificial intelligence into urban design is an important trend in the process of Chinese urbanization,which growth from quantity to quality.At present,the research of intelligent urban design is still primitive.Some of the existing methods are completely in the way of black box,lacking of interpretability and interactivity;some are based on parametric semi-automatic generation,which is unable to improve the efficiency of urban designing.This paper focuses on the design of road network in urban block,exploring how to introduce behavioral intelligence into the generation of road network to assist planners to design urban road network more efficiently.As result,an automatic road network generation method with good automaticity,heuristic and interaction is proposed.Firstly,the urban road network generation problem is modeled as a reinforcement learning problem in this paper.By selecting the grid-based spatial structure,the geographic information of the environment is represented by the state space represented by a three-dimensional grid matrix,and the corresponding action space is also set for the agent.As a result,the road network generation problem is formulated as a Finite Markov Decision Process formed by the interaction between the agent and the environment.Secondly,this paper proposes a road network generation model based on deep reinforcement learning.The model enables agents to learn how to interact with the environment with appropriate policies,so as to generate the desired road network design scheme.In order to better controlling the generation process,this paper designs the rules and rewards of the agent.Rules restrict the behavior of the agent in case of the result do not conform to the road design specifications.Rewards determines the values that the agent follows,that is,the direction of policy optimization.The reward structure designed in this paper combines the evaluation system of traffic performance and morphological characteristics of road networks,as well as the internal incentive from the environment.Finally,experiments applying the above model to block scene is implemeted.The experimental results indicate that the method proposed in this paper realize the automatic generation of a large number of road network design schemes.Besides,users can adjust the weight of some reward items to control the morphological characteristics of the generated schemes,so as to realize interactive design.In addition,since the model completely retains the structural geographic information inside the block,the generated results can be directly applied to the mainstream urban design software through data format conversion.