基于ARIMA时间序列的分布式光伏系统输出功率预测方法研究  

Photovoltaic power prediction in distribution network based on ARIMA model time series

作　　者：赵滨滨[1] 王莹[1] 王彬[1] 宣文博[1] 雷铮[1] 葛磊蛟 徐晓萌 Zhao Binbin;Wang Ying;Wang Bin;Xuan Wenbo;Lei Zheng;Ge Leijiao;Xu Xiaomeng(State Grid Tianjin Electric Power Company, Tianjin 300010, China;School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China)

机构地区：[1]国网天津市电力公司,天津300010 [2]天津大学电气自动化与信息工程学院,天津300072

出　　处：《可再生能源》2019年第6期820-823,共4页Renewable Energy Resources

基　　金：国家重点研发计划项目(2018YFB1500801);国家自然科学基金(51807134);国网天津市电力公司科技项目“基于灵活负荷特性及其优化策略的电力需求预测研究”

摘　　要：分布式光伏系统输出功率的预测是对配电网进行协调调度,进而有效消纳分布式光伏发电的关键。文章对天津某地区配电网中分布式光伏电站的功率特性进行研究,建立了基于ARIMA时间序列的分布式光伏系统输出功率预测模型。此外,还分别建立了基于ARIMA时间序列与神经网络的分布式光伏系统输出功率预测模型,以及基于ARIMA时间序列与支持向量机的分布式光伏系统输出功率预测模型,并比较了3种预测模型的预测误差。分析结果表明,与其他2种预测模型相比,基于ARIMA时间序列与支持向量机预测模型的预测误差较小,晴天、雾霾天、阴天和雨天条件下,该模型的预测误差分别为7.02%, 9.13%, 9.35%和9.48%,该模型的年预测误差为13.65%。Power prediction of distributed photovoltaic power stations in distribution network is an essential procession. In this paper, the power characteristics of distributed photovoltaic power station in Tianjin, the ARIMA time series model was established, combined with neural network,support vector machine(SVM) and the results were established respectively. The prediction errors of the three models are compared. The results show that the combined prediction effect of ARIMA model and SVM in sunny weather is good, the error can reach 7.02%, the prediction error in haze day is 9.13%, cloudy day is 9.35%, rainy day reaches 9.48%, and the annual error is 13.65%.

关 键 词：配电网 分布式光伏发电系统 时间序列 功率预测 

分 类 号：TK519[动力工程及工程热物理—热能工程] TM715[电气工程—电力系统及自动化]