小波CWT-MLP网络预测负荷方法改进

Improved Wavelet CWT-MLP Network Forecasting Load Method

为提高对短期负荷的预测精度,提出基于小波变换,并加入电价因子构建MLP神经网络对负荷进行短期预测的方法。首先通过小波变换将原始负荷、电价序列进行分解,得到高、低频率的时间序列带;其次分别利用高频、低频电价序列对高频、低频负荷序列进行MLP神经网络训练与预测;最后,将预测的高频、低频负荷值通过小波变换,重构完整的负荷预测值。采用美国电力联盟实例对该方法进行验证,并与含电价因子的MLP网络预测法、经典MLP网络预测法,以及不含电价因子的CWT-MLP网络预测法预测效果进行比较。结果证明,含有电价因子,并结合小波与MLP神经网络构建的模型能够丰富数据信息,提高负荷预测精度。

To improve the accuracy of short term load forecasting,the paper presents a composition of modes based on continuous wavelet transformation,factor hybrid electric price and Multi-layer Perceptron.Firstly,original series of electric price and load are decomposed into high and low frequency time sequencies,respectively.Secondly,high and low frequency time series are learned and predicted by back propagation neural network.Thirdly,the full prediction of high and low frequency time series is reconstituted by continuous wavelet transform.We use the real electric tariff and load of the PJM company of the United States Power Alliance for 2016-2017 years to verify the method proposed in this paper.It is compared with classical MLP neural network,the single load of CWT-MLP neural network and the load with price factor CWT-MLP neural network.The case based on the improved wavelet CWT-MLP network forecasting load method proposed in this paper shows the feasibility of the method and nice application of the forecasted load model.Adding electric prices to the wavelet CWT-MLP network to establish a new model is good for improving the data’s quality and accuracy.