采用小波多尺度分解的方法,将需水量时间序列分解为多个较平稳的细节子序列和一个趋势序列,再利用BP神经网络对分解后的各序列进行预测,把预测后的序列聚合重构,得到预测结果。以新疆石河子地区的需水量为例对该方法作了验证。表明多尺度分析与神经网络耦合预测,比单一BP神经网络预测精度更高,可满足实际需要。 Using wavelet multi-scale decomposition method, the water demand time series is decomposed into a plurality of relatively stable detail sub-series and a trend series, and then the BP neural network is used to predict the decomposed series. The predicted sequence aggregation is reconstructed , Get the forecast result. Taking the water demand of Shihezi region in Xinjiang as an example, this method is validated. It shows that the coupling prediction between multi-scale analysis and neural network is more accurate than single BP neural network to meet the actual needs.