短期电力负荷在小波包分解下的径向基神经网络预测方法

针对短期电力负荷预测问题,提出一种在小波包分解下的径向基神经网络预测方法。通过小波包分析,将电力载荷及其温度变量对称地分解为低频的近似系数和高频的细节系数。针对不同的小波系数,设计径向基神经网络作为预测器,并通过试错法确定网络合适的结构。网络的训练过程中,采用滑动窗口数据选择策略减少数据样本集,采用随机梯度法更新权值、中心位置和扩展参数。预测的小波系数用于重构出最终的电力载荷值。与前馈多层神经网络的对比数值,实验结果表明,新提出的方法具有较高的预测准确性。

Short term load forecasting method using the wavelet packet decomposition and RBF neural network

This paper provides a short term load forecasting method using the wavelet packet decomposition and RBF neural network. The power loads and temperature variables are decomposed by wavelet packet analysis. The decomposed wavelet coefficients are fed to the RBF neural network. The moving window selection strategy is applied for decreasing the input data sets. The RBF network is trained by the stochastic gradient learning algorithm. The forecasted wavelet coefficients are reconstructed to the final loads. The numerical experiments, compared with the method based on the multi layer feedforward neural network, demonstrate that the proposed method can offer a higher accuracy of the short term load forecasting.