基于特征选择和组合预测模型的负荷短期预测方法  

Short-term Load Forecasting Method Based on Feature Selection and Combination Forecasting Model

作　　者：陆嘉华 梅飞[1] 杨赛 唐瑜 华昊辰 Lu Jiahua;Mei Fei;Yang Sai;Tang Yu;Hua Haochen(College of Energy and Electrical Engineering,Hohai University,Nanjing 211100,China)

机构地区：[1]河海大学能源与电气学院,江苏南京211100

出　　处：《南京师大学报（自然科学版）》2023年第4期114-124,共11页Journal of Nanjing Normal University(Natural Science Edition)

基　　金：国家重点研发计划项目(2022YFE0140600);江苏省重点研发计划项目(BE2020027).

摘　　要：提高配电网母线负荷短期预测准确性对配电网调度运行具有重要意义,随着配电网中气象敏感负荷数量的不断增加,气象因素带来的负荷波动性对预测精度的影响也随之增大.针对传统负荷预测建模中的气象特征选择问题,基于特征的最大相关性、最小冗余性和最大协同作用提出了一种最优气象特征选择流程.在此之上,基于不同特征的不同时间维度提出了回归模型与时间序列模型相结合的短期负荷组合预测模型.采用回归模型对未来的日期时间特征、气象特征进行拟合,以卷积神经网络-双向门控循环单元构建时间序列模型描述历史负荷、历史气象时间序列中的时序特征和气象因素的累积效应.算例分析证明了本文特征选择方法比传统特征选择方法有更低的预测误差,且组合预测模型相比于传统回归、时序模型的预测准确性提升显著.Improving the accuracy of short-term forecasting of bus load in a distribution network is crucial to its dispatch and operation.With the increasing number of weather-sensitive loads in the distribution network,the influence of load fluctuation caused by meteorological factors on forecasting accuracy also increases.Aiming at the problem of meteorological feature selection in traditional load forecasting modeling,an optimal meteorological feature selection process was proposed based on the maximum correlation,minimum redundancy,and maximum synergy of features.On this basis,based on different time dimensions of different characteristics,a short-term load combination forecasting model combining regression model and time series model is proposed.The model uses the regression model to fit the future date-time characteristics and meteorological characteristics and uses the convolutional neural network and bidirectional gated recurrent unit to construct a time series model to describe the time series characteristics of the historical load and meteorological time series and the cumulative effect of meteorological factors.Example analysis proves that the feature selection method has a lower forecast error than the traditional feature selection methods.Compared with the traditional regression and time-series model,the forecasting accuracy of combination forecasting model is improved significantly.

关 键 词：特征选择 回归模型 时间序列模型 组合预测模型 

分 类 号：TM715[电气工程—电力系统及自动化] TP183[自动化与计算机技术—控制理论与控制工程]