基于集合经验模态分解和Q学习策略的短期负荷预测模型  

Short-term Load Forecasting Model Based on Ensemble Empirical Mode Decomposition and Q Learning Strategy

作　　者：段秦尉 何祥针 潮铸 谢祥中 兰萱丽 DUAN Qinwei;HE Xiangzhen;CHAO Zhu;XIE Xiangzhong;LAN Xuanli(Power Dispatching Control Center of Guangdong Power Grid.,Ltd.,Guangzhou 510000,Guangdong Province,China;Beijing Qingneng Internet Technology Co.,Ltd.,Haidian District,Beijing 100084,China)

机构地区：[1]广东电网有限责任公司电力调度控制中心,广东省广州市510000 [2]北京清能互联科技有限公司,北京市海淀区100084

出　　处：《现代电力》2025年第2期360-368,共9页Modern Electric Power

基　　金：中国南方电网有限责任公司科技项目036000KK52210065(GDKJXM20210096)。

摘　　要：短期负荷预测对电力系统的安全稳定运行有着重要意义,为此,提出一种基于集合经验模态分解和Q学习策略优化的短期负荷预测模型。首先,采用集合经验模态分解对原始负荷序列进行分解,以降低预测难度。其次,在此基础上分别采用卷积神经网络、残差神经网络、长短时记忆神经网络和门控循环单元网络4个深度学习模型进行预测得到4个预测结果,再对其加权组合获得最终的负荷预测值。第三,利用Q学习策略对组合权重进行优化,进而最大化组合模型的预测性能。最后,通过某地区真实采集的负荷数据进行实验,结果表明文中所提出的组合预测模型优于其他预测模型,并验证了所提模型的有效性。Short-term load forecasting is of great significance to the safe and stable operation of power systems.For that reason,a short-term load forecasting model based on ensemble empirical mode decomposition(abbr.EEMD)and Q learning strategy optimization was proposed.Firstly,the original load series was decomposed by EEMD to reduce the difficulty of forecasting.Secondly,on this basis,four deep learning models,namely,convolution neural network(abbr.CNN),residual neural network(abbr.ResNet),long short-term memory(abbr.LSTM)neural network and gated recurrent unit(abbr.GRU)were respectively used for forecasting to obtain four forecasting results,of which weighted combination was used to obtain the final load forecasting value.Thirdly,the combination weight was optimized by Q learning algorithm to maximize the forecasting performance of the combination model.Finally,the experiment was conducted using real collected load data from a certain region,and the results showed that the proposed combined forecasting model is superior to other forecasting models,and the effectiveness of the proposed model was verified.

关 键 词：短期负荷预测 集合经验模态分解 深度学习模型 Q学习策略 

分 类 号：TM73[电气工程—电力系统及自动化]