基于嵌入式小波的CNN-LSTM电力系统负荷预测  

CNN-LSTM Power System Load Prediction Based on Embedded Wavelet

作　　者：王才倩 胡倩 苗文捷 陆沈雄 WANG Caiqian;HU Qian;MIAO Wenjie;LU Shenxiong(Hangzhou Electric Power Design Institute Company,Hangzhou 310014,China;Zhejiang Huayun Information Technology Co.,Ltd.,Hangzhou 310008,China)

机构地区：[1]杭州市电力设计院有限公司,浙江杭州310014 [2]浙江华云信息科技有限公司,浙江杭州310008

出　　处：《湖南电力》2025年第1期86-92,共7页Hunan Electric Power

基　　金：国家电网有限公司科技项目(5100-202119574A-0-5-SF);浙江大有集团有限公司科技项目(DY2022-05)。

摘　　要：针对电力系统中负载端耗电预测精准性的问题,提出一种结合卷积神经网络和长短期记忆网络的预测方法。利用小波变换原理,对卷积神经网络中的下采样方法进行改进,使时频特征信息在下采样过程中理论上得以保留和增强。以实际耗电量数据为实验内容,结果表明,所提方法与传统方法相比,各评估指标和短周期负载预测精度方面得到显著提升。To address the issue of the accuracy of power consumption prediction at the load end in power systems,a prediction method combining convolutional neural networks and long short-term memory networks is proposed.By utilizing the principle of wavelet transform,the downsampling method in the convolutional neural network is improved,allowing time-frequency characteristicsinformation to be theoretically preserved and enhanced during the downsampling process.Experiments are conducted using actual power consumption data,and the results show that compared with traditional methods,the proposed method significantly improves various evaluation index and the accuracy of short-term load prediction.

关 键 词：电力系统 时频特征 下采样 嵌入式小波 电量预测 

分 类 号：TM715.1[电气工程—电力系统及自动化]