中文摘要：

为了提高传统神经网络对非平稳风速的预测精度,提出将改进的量子粒子群算法(QPSO)和小波神经网络(WNN)相结合的滚动预测算法。将小波神经网络的初始连接权值及小波基函数参数组成一个多维向量,作为改进量子粒子群算法的粒子进行计算更新,将搜索得到的解空间范围内全局最优参数作为小波神经网络的初始参数。针对已经训练好的小波神经网络的预测误差会随着时间推移而增大的问题,采用每隔1h滚动式训练的方法训练小波神经网络。运用优化算法对我国海南东环铁路某测风站实测风速进行超前多步预测。实例结果表明,相对于传统小波神经网络,优化算法的风速平均相对误差和均方根误差都有所降低,其超前3min、9min、15min的风速预测平均相对误差为8.28%、9.93%、11.37%。

英文摘要：

This paper combined Improved Quantum Particle Swarm Optimization (QPSO)with Wavelet Neural Network (WNN)to improve the prediction accuracy of Traditional Neural Networks for non -steady wind.A multi-dimensional vector consisting of WNN initial connection weights and wavelet function parameters served as variables of Improved Quantum Particle Swarm Algorithm to update calculation.The global optimal parameters were obtained and used in initial parameters of Wavelet Neural Network.A problem that the forecast error of trained Wavelet Neural Network would increase by time was solved using rolling training method to Wavelet Neu-ral Network every one hour.This optimized algorithm was tested in one of the wind stations in Hainan East Cen-tral Railway to multi-step forecast wind speed.The results show that average relative error and root mean square error of average wind speed are reduced.Ahead of 3min,9min,15min of the wind speed,the relative error of average wind speed is 8.28%,9.93%,11.37% respectively.