Research On Intelligent Control Algorithm For Automatic Train Operation

Railway transportation is a significant part in China 's social and economic development,and the railway energy consumption accounts for a large proportion of total energy consumption for transportation.Different operating ways have a different influence on the energy consumption of trains.Therefore,the study of ATO is of great significance to improve the efficiency of train operations,to liberate human resources,and to achieve run goal with train safety,stability,punctuality,and energy saving.The problem of ATO has the characteristics of high dimension,nonlinearity,complex constraints and time-varying characteristic.Thus,it has a huge space for optimization.In current solutions,analytical solutions can hardly handle situations with complex constraints.Numerical searching techniques can not be applied to real-time system because of the large amount of calculation.Heuristic solution relies too heavily on artificial design.And PID control is difficult to adjust parameter.In view of all analysis above,this paper proposes a scheme for ATO based on machine learning.The pre-optimized speed curve can be obtained by learning the excellent driver's driving pattern and PID parameter tuning can be realized by adopting a self-learning method.Thus,ATO can be achieved.The main work are as follows:1?Proposed an automatic train control scheme based on machine learning.It analyzes the route model,the train model,and the train operation constraints and objectives.And it proposes a scheme for automatic train control based on machine learning,That is the pre-optimized speed curve satisfying the operation constraints and energy-saving targets is obtained in advance,and then control the train to follow the pre-optimized speed curve.2?Proposed an driving patterns learning method based on high-order correlation learning.By analyzing the influence factors of train energy consumption,feature sets can be designed and it uses operating data after preprocessed combining with hypergraph to learn the excellent driving patterns,and finally,pre-optimized speed curves can be generated.3?Proposed an automatic train control method based on predictive PID.Designing a predictive PID controller and symbolic regression is used to learn the train state model and a PID parameter tuning is achieved through a reinforcement learning.4 ? The scheme of ATO based on machine learning has been validated by experiments.The experimental results show that the pre-optimized speed generated by this scheme can achieve about 9% energy saving effect on the constraints condition of safety,punctuality,and stability.The designed controller can control the train to follow the pre-optimized speed curve to verify the feasibility of the solution.