Research And Application Of Short-term Wind Speed Prediction Method Along High-speed Railway Based On Deep Learning

In recent years,China’s railway industry has developed rapidly,and the mileage of high-speed railway has reached 40 thousand kilometers.The high-speed rail network has the characteristics of many stations,long lines and wide areas,which leads to the more complicated running environment of trains.Among them,the ambient wind speed is an important factor affecting the safety of high-speed trains,and train overturning and derailment accidents caused by strong winds occur from time to time.In Germany,Japan and other developed countries,in order to ensure the safe operation of trains,corresponding railway gale early warning systems have been built,and short-term wind speed prediction in such systems is one of the key technologies.This thesis summarizes the existing research background of wind speed prediction along the high-speed railway and related technologies,puts forward two improved short-term wind speed prediction models along the high-speed railway,and designs a wind speed early warning system along the high-speed railway on this basis.The contents of this thesis are as follows:(1)Research on short-term wind speed prediction model along high-speed railway.In view of the strong non-stationarity and volatility of wind speed,the wind speed is decomposed by using the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)by referring to the idea of time series decomposition model.In order to solve the over-decomposition problem and ensure that the reconstruction is sensitive to the original data,an Adaptive Reconstruction method(AR)based on Permutation Entropy and Correlation Coefficient is proposed.Based on the Gated Recurrent Unit(GRU),in order to improve the accuracy of time series data correlation analysis,combined with the Attention Mechanism to enhance the information expression of the input features,the model CEEMDAN-AR-AGRU is constructed to improve the prediction accuracy.At the same time,the early warning performance of the model was evaluated through the precision rate and recall rate.The results showed that the improved model proposed in this thesis had better prediction accuracy and early warning performance than the comparison model in single-step and multi-step prediction.(2)Research on short-term wind speed interval forecasting combination model along high-speed railway.Based on the CEEMDAN-AR-AGRU model,combined with Quantile Regression(QR)and Kernel Density Estimation(KDE),the permutation entropy in the adaptive method is used to group each component,and different components are selected according to the components in different groups.Forecasting model,an interval forecasting combination model CEEMDAN-AR-QRAGRU-GRU was constructed.Experiments on the data of two monitoring points show that the combined interval prediction model proposed in this thesis has better interval prediction performance than the comparison model.In terms of early warning performance,the recall rate is higher than 90% and maintains a relatively high precision rate,which is significantly improved compared with point prediction,which verifies the effectiveness of interval prediction in high-speed rail wind speed early warning.(3)Design and implementation of wind speed early warning system along highspeed railway.As a part of the high-speed railway disaster prevention monitoring network,the early warning system designed in this thesis adopts modular design ideas to build an easy-to-access,easy-to-expand and cross-platform wind speed early warning system along the high-speed railway,and integrates the above prediction model in the algorithm layer of the system to provide support for wind speed trend prediction,and at the same time,the system realizes the visualization of wind speed trend,providing more intuitive decision-making information for relevant dispatchers of high-speed railway.