基于开发智慧云平台的日前建筑用电负荷预测方法研究  

Research on Day-ahead Forecasting Method of Building Electricity Load Based on the Development of Smart Cloud Platform

作　　者：潘广旭 宫池玉 李兴玉 胡军 李英杰 王瑞琪 PAN Guangxu;GONG Chiyu;LI Xingyu;HU Jun;LI Yingjie;WANG Ruiqi(State Grid Shandong Electric Power Co.,Ltd.Rizhao Power Supply Company,Rizhao 276800,Shandong,China;State Grid Shandong Comprehensive Energy Service Co.,Ltd.,Jinan 250002,Shandong,China)

机构地区：[1]国网山东省电力公司日照供电公司,山东日照276800 [2]国网山东综合能源服务有限公司,山东济南250002

出　　处：《电气传动》2022年第12期47-53,共7页Electric Drive

基　　金：国网山东省电力公司科技项目(520617200001)。

摘　　要：作为电力系统设计规划、运行调度的重要一环,电力负荷预测受到强随机性、低精度的困扰,同时先进预测算法的落地实施关联数据管理系统,而传统数据管理系统底层数据资源传输与治理、预测信息应用十分不便。为了克服以上问题,基于云平台,在实现数据高效采集与治理的基础上,为电负荷预测提供天气预报以及历史电负荷信息;在日前负荷预测过程中,针对单一长短期记忆(LSTM)神经网络对时序数据挖掘能力不充分的情况,利用小波变换(WT)细化时序负荷高频分量,同时借助下一日温度、相对湿度预报信息,提升日前电负荷预测精度。结果表明,所提WT-LSTM方法表现了良好的预测效果,其两日均方根误差分别为185.56和179.56,比传统的LSTM网络预测精度分别提高了61.48%和12.51%。As an essential part of power system design and operation scheduling,electric load forecasting is troubled by strong randomness and low accuracy.Moreover,the application of advanced forecasting algorithms is associated with data management system,while the traditional data management system is very inconvenient for the transmission and management of the data resources and the application of prediction information.In order to overcome the above problems,based on the cloud platform,weather forecast and historical electric load information was provided for electric load prediction on the basis of achieving efficient data collection and management.In the process of day-ahead load forecasting,aiming at the insufficient ability of single long short-term memory(LSTM)neural network to mine time series data,the method of wavelet transform(WT)was adopted at the same time,which refines the high-frequency components,and improves the forecasting accuracy of the day-ahead power load with the help of the next day's temperature and relative humidity forecast information.The results show that the proposedWT-LSTM method has a good prediction effect,and its two-day root mean square errors are 185.56 and 179.56 respectively,prediction accuracies are 61.48%and 12.51%higher than the simple LSTMneural network.

关 键 词：电力负荷预测 云平台 长短期记忆神经网络 小波变换 

分 类 号：TM715[电气工程—电力系统及自动化]