Local Short-Term Prediction Method For Wind Speed And Direction On High-speed Railway

With the increasing running speed of high-speed railway train,the impact of strong winds on high-speed trains has begun to be valued.High-speed trains are easily affected by strong winds and tend to produce large pitching torque which can easily cause traffic accidents in process of driving.In the special railway section with complicated geographical environment,the wind flow field around the train changes fiercely,leading to significant changes in aerodynamic forces,resulting in a great increase in possibilities of derailment and overturning.Therefore,it is of great significance to predict the gale weather along high-speed railway lines.In this thesis,the method of short-term local advance prediction of gale weather along high-speed railway lines is studied,and the wind speed and direction prediction model is built.The proposed model was trained by data collected by WindLog wind speed and direction sensor,and the wind speed and direction lmin,5min and 10min ahead were predicted.In order to realize wind speed and direction prediction,the prediction model based on long short-term memory network(LSTM)is proposed.With data preprocessed,the reasonable learning step set up,the double LSTM network structure is established.Moreover,the features of data within the historical step are obtained.The performance of LSTM model is compared with different data input of single variable and double variable.A prediction model based on convolutional neural network(CNN)is built to optimize the wind speed prediction model.Multiple CNN convolutional modules were superimposed to extract the deep features of wind speed data.Combining LSTM network and CNN,the prediction model with double layer LSTM and double layer CNN convolution layer is constructed.LSTM studies features along time line.CNN studies deep partial features.The wind speed is predicted with the improved optimization model.For the wind direction prediction problem,the ARIMA time series model is constructed based on the seasonality and correlation of wind direction data and the model prediction performance is analyzed.The wind direction data is differentiated,and ACF and PACF tests are carried out to enhance its stationarity and continuity.On this basis,a wind direction prediction model combining of LSTM and CNN is constructed to optimize the wind direction prediction problem.The data of this thesis are detected under the laboratory peripheral environment.Data are divided into two sections for model training and verification.The experimental results show that the LSTM model has higher accuracy for wind speed prediction,and has certain following tendency for wind direction prediction.The optimized wind speed and wind direction prediction model has higher performance.Moreover,the prediction model has strong adaptability.