Autonomous Train Speed Trajectory Control for Virtual Coupling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current train control systems lack the ability to efficiently and safely integrate train dispatching and control, particularly in high-speed or intercity railways, due to limitations in communication between trains, inflexible route locking, and suboptimal resource utilization, which hinders the implementation of virtual coupling and increases operational inefficiencies.
Innovation Solution
An integration-oriented intelligent speed trajectory optimization method and system for autonomous trains, utilizing a Markov decision process and deep reinforcement learning algorithm TD3 to train a neural network and agent, which optimizes speed trajectories under virtual coupling, considering constraints and objectives to ensure safe, efficient, and comfortable operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional train control systems are used with manual driving, then operation simplicity is maintained, but uniformity and accuracy of train operation are poor
Solution Approach 1:
The autonomous train equips itself with decision-making capabilities through onboard intelligence, enabling it to autonomously determine speed trajectories and make coupling/decoupling decisions without continuous human intervention or complex centralized control, thereby improving operation accuracy while avoiding proportional increases in control system complexity
Solution Approach 2:
The patent replaces traditional mechanical control systems with intelligence-based systems using deep reinforcement learning algorithms. The neural network agent processes sensor data and generates optimal control commands, substituting manual mechanical operations with automated intelligent decision-making to achieve higher precision and uniformity
2Adaptability or versatility
If interlocking system locks and clears routes for train control, then safety is ensured, but flexibility of virtual coupling is seriously reduced
Solution Approach 1:
The patent implements dynamic route control where the autonomous train can dynamically adjust its speed trajectory and make real-time coupling/decoupling decisions based on line resource utilization states. The system transitions from static interlocking to dynamic adaptive control, allowing flexibility while maintaining safety through intelligent prediction and constraint satisfaction
Solution Approach 2:
The autonomous train continuously monitors line resource utilization states and front train operation states, using this feedback to dynamically adjust speed trajectories and make informed coupling/decoupling decisions. This closed-loop feedback mechanism ensures safety while enabling flexible virtual coupling operations
3Productivity
If trains operate with larger headway to ensure safety, then safety margin is increased, but line capacity and operation efficiency are reduced
Solution Approach 1:
The patent changes the critical parameter from fixed safety margins to dynamic safety assessment based on real-time state information. By using deep reinforcement learning to optimize speed trajectories, the system achieves tighter headways while maintaining safety through continuous state monitoring and adaptive control adjustments
Solution Approach 2:
The autonomous train performs preliminary planning of speed trajectories and coupling/decoupling decisions based on predicted future states and line resource availability. This advance planning allows optimization of headways while pre-ensuring safety constraints are met, increasing line capacity without compromising reliability
4Extent of automation
If deep reinforcement learning algorithm TD3 is used to train neural network and agent, then decision intelligence is improved, but training time and computational resources increase
Solution Approach 1:
The neural network and agent are trained offline using the TD3 algorithm before deployment on the autonomous train. This preliminary training phase separates the computationally intensive learning process from real-time operation, allowing extensive training without impacting operational response time
Solution Approach 2:
The patent uses simulation environments to train the deep reinforcement learning agent, creating a virtual copy of the railway system for training purposes. This allows extensive training iterations without risking actual train operations or consuming physical resources, significantly reducing real-world training time and costs
Data Source
AI summary
The present invention relates to an integration-oriented intelligent speed trajectory optimization method and system for an autonomous train. The method includes: constructing an autonomous train speed trajectory optimization model under virtual coupling based on a discrete distance; converting the autonomous train speed trajectory optimization model into a Markov decision process; using a deep reinforcement learning algorithm TD3 to train a neural network and an agent in the Markov decision process, to obtain a trained neural network and agent; and deploying the trained neural network and agent to an autonomous train, to perform an autonomous train speed trajectory optimization decision, so that safe, efficient, and comfortable train autonomous operations can be implemented.