可变带宽核估计与卷积神经网络结合的充电负荷预测  

Charging load prediction using variable bandwidth kernel estimation combined with convolutional neural networks

作　　者：王国君 王立业[1] 廖承林[1] 王丽芳[1] 袁晓冬 王明深 WANG Guojun;WANG Liye;LIAO Chenglin;WANG Lifang;YUAN Xiaodong;WANG Mingshen(Key Laboratory of Power Electronics and Electric Drive,Institute of Electrical Engineering,Chinese Academy of Sciences,Beijing 100190,China;University of Chinese Academy of Sciences,Beijing 100149,China;Electric Power Research Institute,State Grid Jiangsu Electric Power Co.,Ltd.,Nanjing 211103,China)

机构地区：[1]中国科学院电工研究所电力电子与电气驱动重点实验室,北京100190 [2]中国科学院大学,北京100149 [3]国网江苏省电力有限公司电力科学研究院,南京211103

出　　处：《北京交通大学学报》2024年第5期155-161,共7页JOURNAL OF BEIJING JIAOTONG UNIVERSITY

基　　金：国家重点研发计划(2021YFB2501602)。

摘　　要：针对电动汽车充电负荷预测研究中存在的充电负荷预测耗时长、效率低、结果不准确等问题,提出一种可变带宽核估计与卷积神经网络时间序列预测相结合的预测方法.首先,结合电动汽车的充电行为和行驶习惯,获得大规模电动汽车的充电行驶数据,基于大量的实时数据,深入分析大规模电动汽车充电负荷的多种影响因素,并基于影响因素和实际路况等构建单位里程耗电量模型.然后,为准确拟合数据,引入3种传统概率模型,分析并比较它们的优缺点和拟合的准确度.最后,基于拟合结果,采用拟合准确度最高的可变带宽核估计模型对电动汽车充电负荷进行拟合,基于拟合结果结合卷积神经网络对电动汽车充电负荷进行预测.研究结果表明:所提方法将电动汽车充电负荷预测的平均误差降至3.11%,最大误差降至6.42%,有效提高了预测准确度,可为电网系统的维护提供借鉴和参考.To address the challenges of time-consuming processes,low efficiency,and inaccurate results in Electric Vehicle(EV)charging load prediction,this study proposes a prediction method combining variable bandwidth kernel estimation with Convolutional Neural Network(CNN)-based time series prediction.First,the charging and driving data of large-scale EVs are collected by analyzing their charging behaviors and driving habits.Using extensive real-time data,the study conducts an in-depth analysis of multiple factors influencing large-scale EV charging load and constructs a unit mileage energy consumption model based on these influencing factors and actual road conditions.Next,to improve data fitting accuracy,three traditional probabilistic models are introduced,and their advantages,disadvantages as well as fitting accuracy are analyzed and compared.Finally,based on the fitting results,the variable bandwidth kernel estimation model with the highest fitting accuracy is used to fit the EV charging load.The fitted results are then combined with a CNN to predict EV charging load.The results demonstrate that the proposed method reduces the average prediction error of EV charging load to 3.11%and the maximum error to 6.42%,which significantly improves the prediction accuracy,providing reference and guidance for the maintenance of power grid systems.

关 键 词：电动汽车 可变带宽核估计 卷积神经网络 负荷预测 

分 类 号：TM73[电气工程—电力系统及自动化]