基于深度信念网络的电力负荷短期预测方法  

Short Term Forecasting Method of Energy Demand Based on Deep Belief Network

作　　者：曹华珍 韦斌 高超 吴杰康[3] 雷振 隋宇 陈亚彬 CAO Huazhen;WEI Bin;GAO Chao;WU Jiekang;LEI Zhen;SUI Yu;CHEN Yabin(Guangdong Power Grid Corporation Planning Research Center,Guangzhou 510600,China;Guangdong Power Grid Co.,Ltd.,Guangzhou 510600,China;School of Automation,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广东电网有限责任公司电网规划研究中心,广州510600 [2]广东电网有限责任公司,广州510600 [3]广东工业大学自动化学院,广州510006

出　　处：《新型电力系统》2025年第1期111-124,共14页NEW TYPE POWER SYSTEMS

基　　金：广东电网有限责任公司电力规划专题研究项目(031000QQ00220019)。

摘　　要：随着电力市场深入改革,电力逐渐显现商品属性,实时电价走向将会影响短期用电需求,为提高考虑实时电价情况下电力负荷短期预测精度问题,有效映射短期负荷随机性,提出一种电力市场环境下基于实时电价的负荷短期预测方法。选取以实时电价为核心的多种关联指标,利用互信息方法进行指标选择并加权处理,优化预测模型的输入。针对日类型负荷问题,利用改进灰色关联分析原理,构建双层结构相似日集选取方法,分两步确定相似日集。针对深度信念网络权值过于随机化的问题,采用鲸鱼算法优化深度信念网络,以期实现电力负荷短期预测。以美国某地区实时电价与相应负荷数据为例,进行电力负荷短期预测,结果表明所提预测方法可以有效处理电价与负荷相关性,提高了预测精度。With deepening reform of the power market,electricity gradually shows commodity attributes,and the trend of real-time price will affect the short-term demand for electricity.In order to improve the accuracy of short-term energy and power demand forecasting considering the real-time price,and effectively map the randomness of short-term energy and power demand,a short-term forecasting method of energy and power demand based on real-time price in the electricity market environment is proposed.In order to optimize the input of the forecasting model,a variety of related indicators with real-time price as the core are selected and weighted by using mutual information method.Aiming at the problem of daily energy and power demand,the method of selecting similar day sets with double-layer structure is constructed by using the improved grey correlation analysis principle,and the similar day sets are determined in two steps.Aiming at the problem that the weights of the deep belief network are too random,the whale algorithm is used to optimize the deep belief network in order to realize the short-term energy and power demand prediction.Taking the real-time electricity price and corresponding energy and power demand data of a region in the United States as an example,the short-term energy and power demand forecast is carried out.The results show that the proposed forecasting method can effectively deal with the correlation between electricity price and energy and power demand,and improve the prediction accuracy.

关 键 词：电力负荷短期预测 深度信念网络 鲸鱼算法 互信息 相似日 

分 类 号：TM715[电气工程—电力系统及自动化]