Highway Route Generation Method Based On Deep Reinforcement Learning

The design of highway line selection is a complex design work integrating many specialties.In view of the importance and complexity of this work,line selection and researchers continue to optimize and improve their working methods.Intelligent optimization,knowledge engineering and other artificial intelligence methods have been applied to this work,but the methods still need to be improved.Therefore,this paper proposes a method of route generation based on deep reinforcement learning.DQN(Deep Q Network)algorithm was used to generate the initial route path with consideration of elevation information,and then the route plane and longitudinal section satisfying the design specification were obtained by dynamic determination box and horizontal and vertical optimization.The main research contents and achievements are as follows:(1)a highway route design method system based on deep reinforcement learning is proposed.Based on the literature analysis at home and abroad,combined with the characteristics of highway route selection,the method of deep reinforcement learning is used to generate the initial route of highway route under certain geographical environment.Then,based on the dynamic decision box designed by myself and the horizontal and vertical optimization method,the initial path is fitted into the line plane and vertical section satisfying the design specification.The innovation of highway route design method is realized.(2)The optimal path generation based on deep reinforcement learning is realized.First of all,the problem of generating the initial direction of the route is described as the markov decision-making process,that is,each step is represented as the behavior taken by the agent in a certain state,and the reward is fed back from the environment.Then build the reinforcement learning model of line selection problem,extract the line selection area environment information with GIS platform,integrate it into the basic environment of agent exploration path,and design reasonable reward andpunishment function.Then,DQN algorithm is used to learn the route strategy.Under the existing environmental conditions,the optimal initial route direction is generated.(3)This paper presents a method to optimize the alignment of highway horizontal and vertical sections based on dynamic decision box.Aiming at the problem that the method of fitting could not be used to determine the route elements directly in the past,a method of dividing the route into reasonable sections by dynamic judgment box was designed.A series of rectangular boxes with variable range were used to divide the route into sections,and the straight line and curve elements were fitted,judged and adjusted to finally generate the highway route conforming to the design specification.(4)Program development and method validation.A highway intelligent route selection system based on deep reinforcement learning is developed based on Python language framework by integrating relevant technologies such as deep reinforcement learning,GIS and numerical analysis.The feasibility of the method is verified by a concrete engineering example.The research results of the paper show that the proposed method can effectively explore a better route scheme that satisfies the constraint conditions,the selected route can avoid the adverse environment and adapt to the terrain,the economy is reasonable,and the scheme is feasible.It can provide reliable route reference scheme for highway route designers in the early stage of engineering design.