多尺度特征提取与非线性融合的综合能源系统多元负荷短期预测  

Short-term Multivariate Load Forecasting of Integrated Energy System Based on Multiscale Feature Extraction and Nonlinear Fusion

作　　者：付文龙 章轩瑞[1] 张海荣 刘嘉睿 缪书唯 李丹 FU Wenlong;ZHANG Xuanrui;ZHANG Hairong;LIU Jiarui;MIAO Shuwei;LI Dan(College of Electrical Engineering&New Energy,China Three Gorges University,Yichang 443002,China;Hubei Key Laboratory of Cascaded Hydropower Station Operation&Control,China Three Gorges University,Yichang 443002,China;China Yangtze Power Company Limited,Yichang 443133,China)

机构地区：[1]三峡大学电气与新能源学院,宜昌443002 [2]三峡大学梯级水电站运行与控制湖北省重点实验室,宜昌443002 [3]中国长江电力股份有限公司,宜昌443133

出　　处：《电力系统及其自动化学报》2023年第12期89-99,共11页Proceedings of the CSU-EPSA

基　　金：湖北省自然科学基金资助项目(2022CFD170)。

摘　　要：为提高综合能源系统多元负荷短期预测的精度,提出一种基于多尺度特征提取与非线性融合的综合能源系统多元负荷短期预测方法。首先,采用皮尔逊相关系数对气象数据进行关联因子优选;然后,通过嵌入式分解模块将输入的时间序列分解为周期分量和趋势分量,并将分解后得到的输入矩阵并行送入到具有不同尺度卷积核的时间卷积网络中,进行多尺度特征提取;接着,将多尺度时间卷积网络输出的特征向量输入到各自对应的注意力机制,以进行全局信息的学习与融合;最后,采用自适应非线性融合模块对各注意力机制的输出进行非线性融合,得到最终多元负荷预测结果。实验结果表明,所提方法具有较好的预测性能及泛化性。To improve the accuracy of short-term multivariate load forecasting of an integrated energy system(IES),a short-term multivariate load forecasting method for IES based on multiscale feature extraction and nonlinear fusion is proposed.First,the Pearson correlation coefficient is used to optimize the correlation factors for the meteorological data.Second,the input time series are decomposed into period components and trend components by an embedded decomposition module,and the input matrix obtained after decomposition is fed into the temporal convolutional network(TCN)with convolution kernels of different scales in parallel for multiscale feature extraction.Third,the output feature vectors of multiscale TCN are fed into the corresponding Attention mechanisms for global information learning and fusion.At last,an adaptive nonlinear fusion module is used to nonlinearly fuse the output of each Attention mechanism,thus obtaining the final multivariate load forecasting results.Experimental results show that the proposed method has a better prediction performance as well as a better generalizability.

关 键 词：综合能源系统 多元负荷预测 多尺度时间卷积网络 嵌入式分解 自适应非线性融合 

分 类 号：TM721[电气工程—电力系统及自动化]