Automatic Train Operating Algorithm Of High-Speed Train Based On Attention Mechanism And State Awareness

As an important infrastructure in the national strategy,high-speed railway plays a vital role in promoting the sustainable development of national economy and improving people’s travel conditions.With the significant increase in the speed and traffic density of high-speed trains,the traditional manual driving has been unable to meet the development needs of high-speed railway.Therefore,Automatic Train Operation(ATO)has become an important direction for the future development of high-speed railway in China.However,due to the complex external operating environment and vehicle operating conditions,it is difficult for the existing ATO methods to take into account the goals of energy saving,punctuality and comfort of train operation control.The rapid development of computer,big data,artificial intelligence and other technologies has brought a profound revolution to the traditional control field.In this context,the in-depth integration of ATO and new generation information technology to build an intelligent ATO method has become a research hotspot in recent years.This paper takes ATO algorithm as the research object,proposes a train control model based on attention mechanism,and further designs a multi-objective optimal control method for high-speed trains based on state awareness.The specific research contents are summarized as follows.(1)High-speed trains have the characteristics of strong nonlinearity,large control time delay and uncertain parameters,so it is difficult to accurately describe the dynamic operation of trains using traditional mechanism-based modeling method.Data-driven models can effectively solve the environmental and parameter adaptive problems,but most of them are "black-box" mechanisms,which cannot explain the physical meaning of model parameters.Based on the above reasons,a data-driven train control model LAG-LSTM(Long Short-Term Memory with Lagged Information)with interpretability is investigated by combining train dynamics and deep learning framework.A deep learning module based on attention mechanism is constructed to obtain the variation pattern of control command delay time,and the model validity and parameter visualization are analyzed through experiments.After comparing with the standard LSTM and other benchmark models,it is verified that LAG-LSTM can significantly improve the prediction accuracy of train trajectory when the operating conditions are frequently switched,and effectively reduce the impact of control command time delay on the accuracy of the model.(2)In order to solve the problem that the existing ATO is difficult to take into account multiple control objectives,a multi-objective optimal control method for high-speed trains based on state awareness is further proposed.For the ATO station stopping process,a deep learning model is used to sense the stopping position in real time during the coast process after the train enters the station and selects an appropriate time to output a braking command to realize energy-saving and comfortable stopping control.Aiming at the problem of unbalanced parking data samples,an accurate stopping control method based on Few-Shot Convolutional Neural Network(FSCNN)is constructed.For the ATO section driving process,a deep reinforcement learning model based on Double-Agent Deep Deterministic Policy Gradient(DA-DDPG)is proposed with LAG-LSTM as the state transfer model and multiple control objectives such as train operation safety,energy efficiency,on-time performance and comfort as the reward function.The actual line and driving data are selected to compare with the control method based on manual driving,and the results show that the designed algorithm achieves the multi-objective cooperative and optimal control better.(3)On the basis of the above research,a set of automatic train operation simulation platform is developed based on the Flask framework.The platform has the function of ATO algorithm simulation experiment,and can also collect operation data and sends ATO control commands through the train interface to complete the field test function.Py Qt5 is used to design the user service interface,visualize the train operation data and simulation/test results,and this paper gives the functional implementation flow to support the research and development staff to complete the ATO algorithm design efficiently.Figures 63;Tables 25;References 99.