Combined Prediction Model Of Wind Speed And Its Application In High-speed Railway Gale Warning System

With the expansion of the coverage area of China’s high-speed railway network,the mileage of high-speed rail in the windy areas is also getting longer.The strong crosswind changes the air pressure on the train surface,causing the train yaw exceeding the limit,which seriously affects the safe operation of high-speed trains.Derailment and overturning accidents caused by strong wind occur frequently at home and abroad.Therefore,constructing a perfect high-speed railway gale monitoring and early warning system is an important measure to ensure the safe operation of high-speed trains.Based on the wind speed monitoring data along the Lanzhou-Wulumuqi High-speed Railway,this paper establishes a wavelet-time series wind speed prediction model and an EEMD-GA-BPNN wind speed prediction model,and uses each single prediction result to establish a combined prediction model of wind speed.Combined with high-speed train operation alarm rules in gale environment,the design of high-speed railway gale early warning system embedded in wind speed combined prediction model is studied.The main research contents and conclusions are as follows:（1）Based on the wind speed monitoring data of Hami station of Lanzhou-Wulumuqi High-speed Railway,this paper used the db3 wavelet to perform 3-layer decomposition to obtain the trend signal and the detail signal.And then,the stationarity test was performed on each sub-signal,and the tailing and censoring of the ACF and PACF of each sub-signal were calculated.Combined with the AIC information criterion,the optimal order of the model was determined.Using the method of least square estimation,the unknown parameters in the model were calculated.By constructing the statistical quantity Q_LB,the adaptability of the model was tested,and the ARIMA model suitable for data characteristics was established to predict each sub-signal.Finally,the predicted values of each sub-signal were superimposed to obtain the final wind speed prediction result.In order to improve the accuracy of wind speed multi-step prediction,a rolling prediction method was proposed,and three-step and five-step prediction were performed.The result shows that:when the wind speed is predicted in one step,the prediction effect is good,and the overall trend of the wind speed change can be accurately predicted,but there is a large deviation in the prediction results when the wind speed changes rapidly.When the wind speed is predicted in multiple steps,the prediction result of rolling multiple steps is better than that of the direct prediction,indicating that the proposed rolling prediction algorithm can effectively improve the accuracy of multi-steps prediction.While,with the increase of prediction steps,the prediction error also increases.（2）Based on the EEMD theory,the original wind speed data was decomposed to obtain several IMF components and a residual information.The BPNN prediction model was established for each sub-information,and the genetic algorithm was used to optimize the initial weights and thresholds of the neural network.The network structure and related parameters were determined by several trial calculations.Based on the wind speed monitoring data along the Lanzhou-Wulumuqi High-speed Railway,the BPNN wind speed prediction model,the EEMD-BPNN wind speed prediction model and the EEMD-GA-BPNN wind speed prediction model were established,and the wind speed was predicted in one-step,three-step and five-step.The result shows that:when the original wind speed data without decompose is directly predicted by BPNN,the prediction delay occurs,but after decomposing by EEMD,the problem of prediction delay is solved.Using genetic algorithm to optimize the neural network can effectively improve the prediction accuracy.（3）Based on the results of each single prediction model,and talking the minimum sum of squared errors as the objective function,a dynamic variable weight combination prediction model and a neural network combination prediction model were established.Derive the process of calculating the weights of the dynamic variable weight combination prediction model;determine the input,output and structure of the neural network combination prediction model.Using the predict data to analyze the combined model.The result shows that:the prediction accuracy of single prediction model cannot be kept in a stable state;in some sections,the prediction error is large.However,the error fluctuation of the combined prediction model is small.Compared with the results of single prediction model,the combined prediction models can effectively improve the prediction accuracy,and the results of neural network combined prediction model are better.When the multi-step prediction of wind speed is carried out,the combined model still shows better prediction effect.Compared with the three-step combination prediction,the optimization effect of the five step combination prediction is better.（4）Relying on the wind speed monitoring station arranged in the disaster prevention network of the Lanzhou-Wulumuqi High-speed Railway,this paper described the location determination method of wind speed monitoring base station along the high-speed railway and expounded the collection and transmission of wind speed data,and established a big data platform that could store and process data efficiently..Based on the wind-vehicle-road（bridge）coupling mechanism,the relationship between the critical speed of different types of trains and the ambient wind speed was determined,and this was used as the warning rule of the early warning system.The established wind speed combined prediction model was used to predict the wind speed,and combined with the system alarm rules,the design of high-speed railway gale monitoring and early warning system embedded in wind speed combined prediction model was studied.