Research On The Generation Method Of Railway Line Scheme Based On Deep Reinforcement Learning

As a pivotal aspect of railway construction,route design plays a crucial role in determining the overall project cost,construction difficulty,and subsequent operational efficiency.The route design process is influenced by a combination of objective factors such as geographical environment and subjective factors including expert experience and local policies,making it a complex and time-consuming task.In the initial stages of a project,there is a pressing need to showcase route alternatives to facilitate subsequent work and establish project feasibility.To address the short-term requirement of designing railway route plans during project planning,this paper proposes an intelligent approach for generating railway route plans using deep reinforcement learning theory.The main contributions of this research are as follows:(1)A construction method for an intelligent railway route optimization model based on deep reinforcement learning is proposed.By analyzing the current research status of railway route design both domestically and internationally,deep reinforcement learning theory is selected.A neural network is established with the agent’s state as input and the agent’s exploration actions as output.The agent learns the route design task guided by reward signals,breaking the limitations of previous computer-aided design decisions and achieving intelligent integrated railway route planning.(2)Automatic generation of railway route alignments is implemented.By simplifying the terrain,a reinforcement learning environment model is established,optimizing the relationship between the agent’s state and actions based on the experience of route design work.Feedback in the form of rewards and penalties is provided to the agent during the railway route selection task.After multiple rounds of exploration and iteration,the model can output the optimal route alignment within the selected area.A performance comparison between the Deep Q-Network(DQN)and Proximal Policy Optimization(PPO)algorithms in this model reveals that PPO demonstrates superior overall performance.(3)A linear fitting method based on an adaptive line container is proposed.To address the issue of route alignments generated by the intelligent route selection model not conforming to specifications,an adaptive line container is constructed using the orthogonal least squares fitting method.This approach involves segmenting the path and then fitting the lines,transforming the initial route alignment into a railway route plan.By improving the fitting formula,the proposed method ensures that the route plan adheres to specifications while closely aligning with the original route alignment.(4)Development and validation of the intelligent route selection program.By combining deep reinforcement learning theory,geographic information systems,and computer programming techniques,a railway intelligent route selection system is developed using the Python language.The effectiveness of the proposed method is validated through a case study of railway route design in a mountainous region of the Yungui Plateau.The research results demonstrate that the railway intelligent route selection method based on deep reinforcement learning can efficiently explore route plans that meet requirements.By comparing it with an existing railway route plan for a mountainous region,this method significantly reduces the design time of the initial route plan and saves 20.55% in costs compared to the original route plan.This method provides a solid basis for manual selection of route plans in subsequent stages.