为提高传统神经网络对非平稳风速的预测精度,提出一种基于小波分析法与神经网络法混合建模的优化算法.该优化方法引入小波分析法对实测非平稳风速信号进行分解,将非平稳性原始风速序列转化为多层较平稳分解风速序列,再利用BP神经网络对各分解层风速序列建立预测模型,最终加权各层预测结果获得风速超前多步预测结果.仿真结果表明;该优化算法实现了风速的高精度短期多步预测,将传统神经网络法对应超前步数的平均绝对相对误差分别提高了55.56％,32.43％和34.58％,其超前1步、3步和5步预测的风速平均相对误差分别为0.48％,1.50％和2.97％.优化网络具备信号分解与自学习能力.%To promote the forecasting performance of traditional neural networks for non-stationary wind speed signal, an optimization algorithm was proposed based on wavelet analysis method and neural networks method. This optimization algorithm employed wavelet analysis method to make signal decomposition and reconstruction calculations for original wind speed series attain more steady sub-series. Then BP neural networks method was used to build unsteady prediction models for each layer to realize multi-step rolling forecast calculation. Simulation results show that the optimization algorithm can attain high-precision multi-step ahead forecast results, respectively improve forecast precision of traditional BP neural networks method by 55.56%, 32.43% and 34.58%, and the mean relative error of one-step, three-step and five-step ahead forecast are 0.48%, 1.50% and 2.97%. The optimization has signal decomposition and self-learning ability.
展开▼
