基于小波分解和相同尺度序列的神经网络短期电价预测  

Neural networks short-term forecasting of electricity price based on wavelet decomposition and homo-layer series combination

作　　者：肖霖[1] 张婧[2] 曾鸣[3] 董军[3] 

机构地区：[1]重庆市电力公司南岸供电局,重庆400060 [2]重庆市电力公司市区供电局,重庆400015 [3]华北电力大学能源与电力经济研究咨询中心,北京102206

出　　处：《电力需求侧管理》2011年第4期19-22,29,共5页Power Demand Side Management

基　　金：国家自然科学基金项目(70671041)

摘　　要：基于小波分解的短期电价预测主要是对电价样本分解后的各个子序列进行预测,并重构各个预测结果得到最终预测电价。在这一基础上,对电价和负荷样本进行多分辨率小波分解至2尺度,然后剔除噪声信号,再将其中相同制度的电价和负荷子序列相结合,并根据该尺度的时频特征设计建立神经网络模型进行预测,最后将各个子序列的预测结果重构得到预测电价。在算例分析中采用PJM市场2007年3月至2008年2月的数据,并通过绘制误差持续曲线,测试对比本文提出的预测方法和其他预测模型,证明了该方法的有效性和可行性。Short-term electricity price forecasting based on wavelet decomposition is mainly to forecast each subsequence after power price sample is decomposed and reconstruct each forecast result to get the final forecasting price.On this basis,tariff and load samples are decomposed by multi-resolution wavelet to 2 layers,and then remove the noise signals.Then homo-layer series of power price and load are combined as the input of neural networks.According to time frequency character,neural network model is designed and built to forecast.At last forecast results of each sub series are reconstructed to get forecast price.In the numerical example analysis,the data in PJM market from March 2007 to February 2008 is adopted.Error duration curves are drawn to verify the effectiveness and feasibility of the proposed forecasting model in contrasting with alternatives.

关 键 词：电价预测 小波分解 多分辨率 相同尺度序列 神经网络 

分 类 号：F407.61[经济管理—产业经济]