基于多维时序信息融合的短期电力负荷预测方法  

A Short-term Power Load Forecasting Method Based on Multidimensional Temporal Information Fusion

作　　者：李丹[1] 孙光帆 缪书唯 章可 谭雅 张远航 贺帅 LI Dan;SUN Guangfan;MIAO Shuwei;ZHANG Ke;TAN Ya;ZHANG Yuanhang;HE Shuai(College of Electrical Engineering&New Energy,China Three Gorges University,Yichang 443000,Hubei Province,China;Hubei Provincial Key Laboratory for Operation and Control of Cascaded Hydropower Station(China Three Gorges University),Yichang 443002,Hubei Province,China;Hubei Provincial Collaborative Innovation Center for New Energy Microgrid(China Three Gorges University),Yichang 443002,Hubei Province,China)

机构地区：[1]三峡大学电气与新能源学院,湖北省宜昌市443000 [2]梯级水电站运行与控制湖北省重点实验室(三峡大学),湖北省宜昌市443002 [3]新能源微电网湖北省协同创新中心(三峡大学),湖北省宜昌市443002

出　　处：《中国电机工程学报》2023年第S1期94-106,共13页PROCEEDINGS OF THE CHINESE SOCIETY FOR ELECTRICAL ENGINEERING

基　　金：国家自然科学基金项目(青年科学基金)(51807109)

摘　　要：为综合考虑负荷的周期性、时序性和非线性,提高短期负荷预测精度,该文提出一种基于多维时序信息融合的短期电力负荷预测方法。它将负荷的时序性和日、周等周期性先验知识引入预测模型结构和处理单元的设计中。首先,通过一种新颖的多维门控循环单元融合时刻间、日间和周间等多个时序维度的负荷信息;然后,以此为基本单元构建序列到序列的多维时序结构预测模型,差异化处理历史与未来不同特征维数的输入信息,并融合多回溯周期下的负荷时变模式;最后,通过全连接网络输出未来多个时刻的负荷预测值。两个实际地区日前负荷预测的算例结果表明,相比常见预测方法,该文方法可实现历史和未来差异化输入条件下负荷多维时序规律的自适应学习,对组成性质、时间步长、日类型或季节不同的负荷具有更稳定的预测性能和更高的预测精度。This paper proposes a short-term power load forecasting method based on multidimensional temporal information fusion to improve the accuracy of short-term load forecasting by considering the load’s temporality,periodicity,and nonlinear characteristics.The prior knowledge about the load time series,including temporal and multi-periodic(daily,weekly,and so on)features,is introduced into the designs of forecasting model structure and processing units.First,a novel multidimensional gated recurrent unit is designed to fuse the load’s multidimensional temporal information,including hourly,daily,and weekly.Then,based on the basic unit,a sequence-to-sequence multidimensional temporal structured forecasting model is constructed.It processes the input information of the history and the future with different feature dimensions differently and fuses the load time-varying patterns under multiple lookback periods.Finally,the load prediction values at future multiple time points are outputted by the fully-connected network.The day-ahead load forecasting results from two areas show the proposed method’s advantages compared with several popular methods.It can adapt the load’s multidimensional temporal patterns under different historical and future input conditions.And it has a more stable prediction performance and higher accuracy for different load compositions,time steps,day types,or seasons.

关 键 词：短期负荷预测 序列到序列模型 多周期 多维门控循环单元 多维时序模式 

分 类 号：TM715[电气工程—电力系统及自动化]