考虑负荷耦合影响的农场配电网光伏功率短期修正预测  

Short-term corrected prediction of photovoltaic power in farm distribution network considering effect of load coupling

作　　者：车凯 朱进 金沫含 吴金辉 张晋铭 朱慧君 冯子瑜 梁睿[2] CHE Kai;ZHU Jin;JIN Mohan;WU Jinhui;ZHANG Jinming;ZHU Huijun;FENG Ziyu;LIANG Rui(Lianyungang Power Supply Company of State Grid Jiangsu Electric Power Co.,Ltd.,Lianyungang 222000,China;China University of Mining and Technology,Xuzhou 221116,China;Jiangsu Dongxin Farm,Lianyungang 222200,China)

机构地区：[1]国网江苏省电力有限公司连云港供电分公司,江苏连云港222000 [2]中国矿业大学,江苏徐州221116 [3]江苏省东辛农场有限公司,江苏连云港222200

出　　处：《供用电》2025年第3期76-85,共10页Distribution & Utilization

基　　金：江苏省碳达峰碳中和科技创新专项资金(BE2022609)。

摘　　要：在高比例光伏接入下的农场配电网中,光伏实际功率与负荷水平之间存在强耦合关系,影响光伏功率预测的准确性。针对该问题,提出一种考虑负荷耦合影响的农场配电网光伏功率短期修正预测方法。首先,分析负荷水平与光伏实际功率的耦合影响关系,通过设置影响因子指标衡量不同负荷点对光伏功率的影响程度,对耦合关系进行解耦;其次,在对农场配电网负荷类别划分和配电网拓扑简化的基础上,对农场配电网负荷进行区域性聚合,以实现负荷降维;最后,采用基于欧氏距离期望最大化聚类算法的相似日聚类和长短期记忆网络时序模型,获取光伏功率初始预测值,同时基于反向传播神经网络对负荷造成的预测误差进行估计,最终获取农场配电网光伏功率的修正预测值。实验结果表明,在聚类区分晴天、阴天、雨天数据集后,适当减小时序预测模型的历史步长和预测步长能够提高初始预测的准确性,在此基础上,通过考虑负荷耦合影响进行光伏预测误差估计,能够有效减小光伏预测误差。There is a strong coupling relationship between PV actual power and load level in farm distribution grids under high percentage of PV access,which affects the accuracy of PV output prediction.Aiming at this problem,a short-term correction prediction method of PV power in farm distribution network considering the influence of load coupling is proposed.Firstly,the coupling influence relationship between load level and PV output power is analyzed,and the coupling relationship is decoupled by setting the influence factor index to measure the degree of influence of different load points on PV output power;Secondly,based on the load category division of the farm distribution network and the simplification of the topology of the distribution network,loads of the farm distribution network are aggregated regionally for load dimensionality reduction;Finally,the Euclidean distance expectation maximization algorithm based on the similar day clustering and long short-term memory network(LSTM)time-series model to obtain the initial prediction value of PV output,and at the same time estimate the prediction error caused by the load based on BP neural network,and finally obtain the corrected prediction value of PV power of the farm distribution network.The experimental results show that after clustering to distinguish between sunny,cloudy,and rainy day data sets,appropriately reducing the history step and prediction step of the time-series prediction model can improve the accuracy of the initial prediction.On this basis,the PV prediction error estimation by considering the effect of load coupling can effectively reduce the PV prediction error.

关 键 词：农场配电网 光伏功率预测 源荷耦合 长短期记忆网络 反向传播神经网络 

分 类 号：TM74[电气工程—电力系统及自动化]