计及气象累积效应的特征解耦峰荷预测模型  

Feature Decoupling Peak-load Forecasting Model Considering Meteorological Cumulative Effect

作　　者：秦川[1] 丁鹏飞 刘波 鞠平[1] QIN Chuan;DING Pengfei;LIU Bo;JU Ping(College of Energy and Electrical Engineering,Hohai University,Nanjing 211100,China;Huaian Power Supply Company,State Grid Jiangsu Electric Power Co.,Ltd.,Huaian 223002,China)

机构地区：[1]河海大学能源与电气学院,江苏省南京市211100 [2]国网江苏省电力有限公司淮安供电分公司,江苏省淮安市223002

出　　处：《电力系统自动化》2022年第6期66-72,共7页Automation of Electric Power Systems

基　　金：国家自然科学基金资助项目(51837004);国家自然科学基金委员会-国家电网公司智能电网联合基金资助项目(U2066601);111引智计划“新能源发电与智能电网学科创新引智基地”资助项目(B14022)。

摘　　要：夏季峰荷是电力部门的关注焦点,温度累积效应对于夏季峰荷预测具有重要影响。为此,提出一种计及气象累积效应的特征解耦峰荷预测模型。建立具有3个输入分支的深度神经网络模型,从结构上对输入特征实现解耦,称为特征解耦模型。3个分支分别以时间特征、负荷特征、气象特征为输入,其中负荷分支、气象分支应用了长短期记忆(LSTM)网络隐藏层并基于LSTM网络的时序处理能力能对负荷及气象序列进行处理来反映累积效应,进而应用于峰荷预测。最后,通过实例分析,与温度修正等常规方法进行对比,验证了特征解耦模型更适合于计及气象累积效应的峰荷预测。The summer peak-load is the focus of the power industry, and temperature cumulative effect has an important impact on the summer peak-load forecasting. Therefore, a feature decoupling peak-load forecasting model considering meteorological cumulative effect is proposed. The deep neural network model with three input branches is established to realize the decoupling of input features from the structure, which is called feature decoupling model. The three branches take the time feature, load feature and meteorological feature as the input, respectively. The branches for the load and meteorology apply the hidden layer of the long short-term memory(LSTM) network. The load and meteorological sequences are processed based on the time sequence processing ability of LSTM network to the reflect cumulative effect, and then apply it to the peak-load forecasting. Finally, through case analysis and comparison with conventional methods such as temperature correction method, it is verified that the feature decoupling model is more suitable for peak-load forecasting considering cumulative effect.

关 键 词：峰荷预测 累积效应 长短期记忆网络 特征解耦模型 

分 类 号：TM715[电气工程—电力系统及自动化] TP183[自动化与计算机技术—控制理论与控制工程]