基于小波变换的混合神经网络短期负荷预测

Short-term load forecasting by using hybrid neural network and wavelet transform

准确的短期负荷预测是作出正确营销决策的依据。采用小波变换对负荷序列进行分解,对于每一分解序列,分别按照各自的特点选择出比较合适的影响因素,采用信息熵理论和主成份分析相结合的属性约简法对其进行约简,并利用动态聚类对各分解序列的样本归类,通过灰色关联分析找到与预测时刻负荷模式最接近的一些典型样本,训练各分解序列相应的神经网络预测模型,最后通过序列重构,得到完整的负荷预测结果。采用实际负荷数据进行测试,表明这一方法预测效果较好。

The accurate short-term load forecast is an accordance to make right marketing decision. The detailed process is as follows. Firstly, wavelet transform is employed to decompose the load sequence. To every decomposed sequence, according to each character the adequate effect factors are chosen. Secondly, information entropy and principal component analysis are combined for data reduction. By means of dynamic clustering each decomposed sequence is sample classified. Thirdly through gray relationship analysis it finds out some typical samples that are most close to forecasting-hour load model Then we can train the neural network forecast model correspond to every decomposed sequence. At last through reconstructing sequence, the complete load forecast result is gotten. The actual load data testing shows that the forecast result through using this method is good.