Research On The Process Model Of Autonomous Decision Making For Traffic Network Path Planning

Path planning in traffic network is challenging due to the dynamic changes in traffic network status over time and space as vehicles traverse various decision points from the origin to the destination.These changes have characteristics of repeatability,randomness,and uncertainty.Consequently,relying solely on the traffic network status at the origin is insufficient to accurately capture this dynamic nature.However,if replicable traffic network features can be characterized at the origin and decision points along the path,an intelligent agent can autonomously make path planning decisions.The research content includes:(1)Research on autonomous decision making algorithm for traffic network path planning based on deep reinforcement learning.The algorithm adopts a deep Q-network as the fundamental framework,perceiving real-time traffic network status at the origin or decision points along the path,and enabling the intelligent agent to make optimal decisions.Ultimately,the algorithm aims to plan the shortest travel time path from the origin to the destination.The traffic network status is considered as the environment,and the research focuses on the reward and punishment mechanisms for the agent’s interaction with the environment at the origin or decision points,the action selection strategy of the agent,and the switching of environmental states after executing actions.Experience comprising the current state,action,reward,and next state will be accumulated to learn repeatable traffic network features,enabling the agent to achieve autonomous decision making ability at the origin and each decision point.(2)Research on the process model of autonomous decision making for traffic network path planning.To achieve automated and repeatable execution of the business process,the "autonomous decision making for traffic network path planning based on deep reinforcement learning" is decoupled into functional modules,and a process model is designed to combine and collaborate these modules.The resulting process model serves as a means to receive path planning training requests and execute autonomous decision making requests.Upon accepting a request,the training service process model is executed to train the intelligent agent and equip it with the ability to make autonomous decisions for traffic network path planning.Likewise,the execution decision service process model is executed to enable the intelligent agent to autonomously make decisions for traffic network path planning.The decoupled functional modules possess loose coupling,which imbues the process model with flexibility and scalability.As a result,when there are changes in training data or updates in deep learning algorithms,adjusting some functional modules or the collaborative relationships within the process model is sufficient to meet the new requirements.The research findings provide valuable insights for building service architectures with intelligent features.