铁路风速单步高精度混合预测性能对比研究

Study on Performance Comparison of Wind Speed Hybrid High-precision One-step Predicting Models along Railways

为实现风速高精度预测以保障强风下铁路沿线列车的运营安全,利用小波分析理论和神经网络理论所形成的两种不同混合预测算法对我国典型的强风线路青藏铁路沿线的实测大风序列开展超前单步预测研究。小波-神经网络法采用小波分解理论对原始非平稳风速进行分解,对各分解层建立神经网络模型以实现最终的加权预测输出。小波型神经网络法将小波母函数作为神经网络隐含层节点的传递函数训练网络,用训练好的神经网络模型对原始风速进行预测计算。通过对青藏铁路3个测风站实测风速的预测算例表明:两种混合算法的预测指标都优于单种神经网络,小波-神经网络法比小波型神经网络法拥有更加出色的预测性能。

Realization of the wind speed prediction is one of the most effective measures to protect the running safety of the trains along the railways under strong-wind environments. To get high-precision forecasting results ,two hybrid prediction methods named Wavelet-Neural Networks and Wavelet Embedded Neural Networks were provided by combing the theory of Wavelet Decomposition and the theory of Neural Networks, to carry out one-step-ahead prediction study on the measured wind series along the Qinghai-Tibet railway, a typical railway under strong wind environment in China. The modeling process of the Wavelet-Neural Networks is explained as follows： The Wavelet Decomposition theory was used to decompose the original non-stationary wind speed series into a group of sub wind speed layers. A neural network model was established for all the decomposed wind speed layers to sum the multi-step forecasting results of the decomposed layers to realize the final weighted predictions for the original wind speed signals. Different to the Wavelet-Neural Networks, in-stead of using the wavelet method to do the decomposition, the Wavelet Embedded Neural Networks adopted the wavelet functions as the transferring functions in the network hidden layers, using trained neural network model to predict the original wind speed. Three real forecasting cases from the Qinghai-Tibet railway showed that： （a）Both those two hybrid methods can attain high-precision one-step ahead forecasting results, （b）The Wavelet-Neural Networks showed better prediction performance than the Wavelet Embedded Neural Networks.