短期电力负荷的小波—混沌时序预测

Wavelet-Chaotic Time Series Prediction of Short-term Electric Power Load

提出了一种基于2层小波分解的混沌时序相空间重构预测模型。该模型利用小波分解原始负荷时间序列为周期项、趋势项和随机项,采用不同的混沌相空间重构高低频信号,再分别用相应的小波神经网络工具箱拟合混沌吸引子,将其输出进行信号重构得到最终预测结果。该方法兼有频率特征提取和相空间重构的优点,使短期电力负荷时序列的动力学系统得到更加细致的恢复。通过对欧洲电力负荷竞赛数据的实验证明了所提方法的有效性,仿真结果表明方法预测精度优于常规混沌时序预测方法。针对神经网络预测不稳定的问题给出了一种解决措施,并提出了一种小波神经网络工具箱,该方法比编程实现的小波网络可以大幅度提高训练速度,尤其适合于大批量数据的训练,对小波神经网络的推广应用和混沌时序预测具有重要意义。

A prediction model of chaotic phase space restructuring based on two layer wavelet decomposition was proposed. This model used wavelet decomposition original load time series to periodic item, tendency item and random item, and adopting different chaotic phase spaces to restructure low and high frequency signals. Then it used the correspondent wavelet neural network toolbox to fit chaotic attractors, and conduct signal reconstruction on the output to obtain the final result of the prediction. This method combined the merits of frequency characteristic extraction and phase space restructuring, giving a more meticulous recovery of the dynamical system of short-term power load time series. The effectiveness of the method presented was proved by experiments on the data of European power load competition. The simulation results show that the prediction method presented is better than the conventional chaotic time series prediction method. The study offers a solution for the instability of neural network prediction and puts forward a wavelet neural network toolbox prediction method. The method has much faster training speed, compared with the one used with programming. It is especially suitable for the training of mass data, and has a practical significance for promotion of application of wavelet neural networks and prediction of chaotic time series.