Application Of Interval Prediction In Small-amplitude Hunting Prediction Of High-speed Trains

The hunting instability of high-speed trains will seriously affect the safe operation of the train.However,the normal state and the hunting instability is mainly focused in the existed monitoring methods.These methods need a long calculation time in the feature extraction process,and the feature extraction method is fixed,and the generalization ability is not strong,so it is difficult to adapt to the complex running state of high-speed train.In addition,there are few studies on the evolution trend of small-amplitude hunting.In the process of high-speed train from normal to hunting instability,it needs to go through a small-amplitude hunting stage.There are two kinds of evolution in the transition stage of small-amplitude hunting: one is from small amplitude hunting to hunting instability,that is,small-amplitude hunting divergence.The other is from small-amplitude hunting to normal state,that is,small snake convergence.Therefore,it is of great significance for the prediction of hunting instability to study the variation trend of small-amplitude hunting.The structure of high-speed train is complex,it belongs to nonlinear system.The running environment is changeable,which leads to the lateral acceleration of bogie frame has a certain randomness.The traditional point prediction error is large,and it is difficult to give the prediction confidence.Therefore,in this paper,interval estimation method is applied to predict the evolution trend of small-amplitude hunting.Lower Upper Bond Estimation(LUBE)has been commonly used in recent years.However,due to its heuristic algorithm,it is difficult to optimize when there are many parameters.Therefore,improving its optimization algorithm is of great significance to improve the performance of LUBE.In order to solve the above problems,the high-speed trains have been taken as the research object to study the evolution trend of small-amplitude hunting.The main research contents are as follows;(1)The normal state and the hunting instability are mainly focused in the existed monitoring methods,but few studies on the evolution trend of small-amplitude hunting.In this paper,the evolution trend of small-amplitude hunting is studied to judge whether it will occur hunting instability when the train is in small-amplitude hunting.(2)The lateral acceleration of bogie frame has a certain randomness,and the prediction error of using point is large,so it is difficult to quantify the randomness of signal.Therefore,an improved interval estimation method is applied to the trend prediction of small-amplitude hunting.It can judge whether the train will be hunting instability through the prediction interval of small-amplitude hunting.Compared with the traditional feature recognition method,this method can judge whether the train will be hunting instability when it is in small-amplitude hunting,the confidence of prediction can be given,so it has higher efficiency and better stability.(3)The improved LUBE is applied to predict the trend of small-amplitude hunting movements.The method is based on the neural network structure,which has the characteristics of simple structure and strong generalization ability,so it is widely used in time series prediction.However,because of the heuristic algorithm is difficult to optimize the model parameters.A Two-level trained LUBE(TL-LUBE)method is proposed,which improves the optimization mode of the LUBE model.LUBE model is divided into primary model and secondary model training to improve the optimization efficiency of parameters.The results show that the prediction interval coverage probability(PICP)is 88.5% and the prediction interval normalized average width(PINAW)is 0.138 in public data.In high-speed trains data,the PICP is 100%and the PINAW is 0.187.It is concluded that the predicted interval generated by TL-LUBE can get higher coverage and smaller width.(4)In order to realize the practical application of TL-LUBE,a software is developed for hunting monitoring of high-speed trains.The test results show that the software realizes the monitoring function of high-speed train and verifies the practical significance of TL-LUBE in the application of high-speed train hunting monitoring.