Coordinated Intelligent Optimization And Control Of Energy-saving Operation Of Urban Rail Trains

The urban rail transit system,with its advantages of safety,punctuality,green environmental protection and large transportation volume,has increasingly become the primary choice for people’s daily commuting,effectively alleviating the problem of road traffic congestion in large and medium-sized cities.In recent years,the scale of rail transit has been continuously expanding,and the operation of multiple trains on a railway line has shown a hugely networked trend.The coordination of traction and braking between different trains fully utilizes regenerative braking energy,which has the characteristics of energy-saving potential and complex tempo-spatial dynamic characteristics.In this case,the integrated research on train optimal control strategy and regional multi-train intelligent energy-saving regulation is of great significance for improving the industry competitiveness of urban rail transit and reducing operation costs.The continuous development of artificial intelligence,communication technology and cloud computing has been providing new ideas for intelligent cooperative optimization of multi-train operation from the perspective of integration combing train operation plan and control adjustment.How to apply these advanced technologies to make full use of regenerative braking energy is a hot issue in the energy-saving research of urban rail transit system.Through the applications of advanced information technology means such as deep reinforcement learning,cloud computing,and edge computing,this dissertation explores the optimization and control problems of energy-saving operation synergy of metro trains.Under the Communication Based Train Control(CBTC)system,for the automatic train operation(ATO)subsystem,automatic train supervision(ATS)subsystem,automatic train protection(ATP)subsystem,etc.this dissertation studies the adjustment schemes of integrated and coordinated energy-saving operation.The main research contents and innovations of this dissertation are described as follows:(1)In order to improve the intelligent control performance of ATO subsystem of urban rail trains,with regard to the complex characteristics of train operation processes constrained by speeds,times,positions and comfort requirements,an automatic train control algorithm based on deep reinforcement learning is studied.By accumulating a large number of controlling solution examples into the decision intelligence,the deep reinforcement learning model can directly output control decisions based on current states during train operation,so as to achieve real-time energy-saving control.In order to meet the speed protection requirements of train ATP subsystem,considering active constraint handling,a deep reinforcement learning algorithm with reference systems is proposed,which can meet the safety and reliability requirements and save energy during train running.Simulation examples show that the algorithm has faster and better convergence performance,and effectively reduces the train traction energy consumption.(2)In order to improve the intelligent cooperation level of multi-train energy-efficient operation,the optimization of inter-station operation curve is integrated to the multi-train timetable design in the ATS subsystem.The integrated research of these two aspects is a complex,large-scale,multi-variable and multi-constrained task,which involves the optimization processes of nesting,coupling and mutual inputs.An energy-saving inner-outer integrated model is proposed,in which the single train operation curve is optimized in the inner model and the multi train timetable is optimized in the outer model to promote the effective utilization of regenerative braking energy.A parallel particle swarm optimization algorithm based on Map-Reduce is established to solve the optimization of the synthetic model and to reduce the execution time through parallel computing.By introducing the concepts of tempo-spatial record and time shift,the time-space sequences of specific inter-station train operation on a subway line are derived.The experimental results show that the proposed integrated control model solved by parallel optimization method can achieve overall lower energy consumption at a faster execution speed.(3)In order to realize the intelligent adjustment of multi-train energy-saving operation synergy,an optimization model is established for systematically describing the traffic environment of multi-train operation in urban rail transit under the study about a synthetic case of multi-train synergistic operation timetable considering energy-saving driving strategies.Then a multi-agent cooperative actor-critic deep reinforcement learning method is designed to adjust the multi-train timetable to achieve energy saving.According to the high-dimensional characteristics of control actions in the timetable adjustment of train operation,action representation technology is applied for generalization.The multiple agents are deployed on the Spark cloud platform to effectively improve the training efficiency through parallel computing on multiple worker nodes.By employing deep neural networks to perceive the nonlinear relationships of multi-train operation,on the premise of satisfying various time constraints,the optimal energy-saving driving between stations and the optimal traction and braking coordination between multiple trains and multiple stations are realized at the same time,so as to save energy consumption to the greatest extent.(4)In order to realize dynamic real-time energy-saving dispatching and control intelligent adjustment of urban rail trains,make full use of regenerative braking energy,and reduce the total energy consumption of urban rail transit,a multi-agent actor-critic deep reinforcement learning architecture based on edge computing is proposed.The multi-agent actor-critic deep reinforcement learning model carries out parallel offline training deployed on the cloud servers.And according to demands,the edge servers download the multi-agent model well-trained in the cloud servers to the local place for online real-time adjustment of synergistic operation strategies of metro trains.Through the dynamic perceptions of train operation status and the tempo-spatial characteristics of multi-train real-time driving modes,the adjustment of departure intervals,inter-station operation time,dwell time and driving strategies can be implemented in real-time at the edge sides to improve the utilization of regenerative braking energy among multiple trains,and reduce the global energy consumption of trains in a regional railway line network.