基于时序分解的自适应在线学习电力负荷预测算法研究  

An Adaptive Online Learning Load Forecasting Combination Algorithm Based On Time Series Decomposition

作　　者：谢小鹏 胡为明 何基龙 王理 向五济 罗湘 郑洲顺[2] Xie Xiaopeng;Hu Weiming;He Jilong;Wang Li;Xiang Wujing;Luo Xiang;Zheng Zhoushun(Hunan Datang Xianyi Technology Co.,Ltd,Changsha 410000,China;School of Mathematics and Statistics,Central South University,Changsha 410083,China)

机构地区：[1]湖南大唐先一科技有限公司,长沙410000 [2]中南大学数学与统计学院,长沙410083

出　　处：《数学理论与应用》2022年第4期93-104,共12页Mathematical Theory and Applications

基　　金：湖南国家应用数学中心建设项目(No.2020ZYT003)资助。

摘　　要：传统的机器学习方法无法捕捉到电力负荷需求的不确定性以及动态变化规律.本文将最新提出的隐马尔可夫模型在线学习算法应用于电力负荷预测研究,充分提取历史数据中的不确定性特征和动态变化规律,并结合分解算法,更精确利用数据中的动态变化特征,从而提高预测精度.算法基于隐马尔可夫概率预测模型,在获得新样本时对模型进行在线更新,适应最新数据;利用STL时序分解算法对负荷数据进行分解,使具有不同不确定性和动态变化规律的分量分离开,再分别使用在线学习算法对不同特征的分量进行预测,构造电力负荷预测组合算法.基于三个公开电力负荷数据集的测试结果表明,相比于单一的在线学习模型,本文提出的组合算法提高了预测精度,预测相对误差最高减少了27%.Since it is troublesome for conventional machine learning methods to extract the main features relevant to the uncertainties and variations of electrical load, in this paper, a recently proposed hidden Markov model based online learning algorithm is used to solve the load forecasting problems, extracting the uncertainties and variations from the load data. By combining with the decomposition algorithm, the variation features can be estimated more precisely and forecasting accuracy can be improved. Based on the hidden Markov model, the proposed algorithm is updated once new samples are received, thus adapting to real-time data;the STL algorithm is implemented to decompose the load data, leading to the separation of components with different trends. The online learning algorithm is then applied to each component of data, composing the hybrid load forecasting algorithm. Validated by three public datasets, it is shown that the proposed algorithm can improve the forecasting accuracy and reduce the relative error up to 27% when compared with the existing technique.

关 键 词：负荷预测 时间序列分解 在线学习 组合算法 隐马尔可夫模型 

分 类 号：TM715[电气工程—电力系统及自动化]