基于FCM和LSTM的光伏功率超短期预测  

Ultra-short-term prediction of PV power based on FCM and LSTM

作　　者：张磊 王小明 吴红斌 谢毓广 滕越 魏英杰 ZHANG Lei;WANG Xiaoming;WU Hongbin;XIE Yuguang;TENG Yue;WEI Yingjie(Anhui Province Key Laboratory of Renewable Energy Utilization and Energy Saving,Hefei 230009,China;State Grid Anhui Electric Power Research Institute,Hefei 230601,China;State Grid Shanghai Pudong Power Supply Company Dispatch Control Center,Shanghai 200122,China)

机构地区：[1]新能源利用与节能安徽省重点实验室(合肥工业大学),安徽合肥230009 [2]国网安徽省电力有限公司电力科学研究院,安徽合肥230601 [3]国网上海浦东供电公司调度控制中心,上海200122

出　　处：《供用电》2023年第1期10-17,共8页Distribution & Utilization

基　　金：安徽省自然科学基金项目(2108085UD05)。

摘　　要：光伏电站输出功率的不确定性对电网的运行调度产生了严重影响。基于模糊均值聚类(fuzzy C-means,FCM)和长短期记忆网络(long short-term memory networks,LSTM),提出了一种光伏功率超短期预测方法。首先,通过滑动时窗将历史数据划分为时序片段,将功率波动特性以时段为样本进行分析;然后,定义了3种特征指标提取波动规律,采用FCM方法进行时序片段聚类,在聚类后重构数据集,结合LSTM网络,建立光伏功率预测模型;最后,采用澳大利亚光伏电站实测数据集验证所提方案的聚类性能与预测性能。仿真结果表明,所提方法具有更高的聚类准确性和预测精确度。The uncertainty of the output power of photovoltaic power plants has a serious impact on the operational dispatch of the grid.This study proposes an ultra-short-term prediction method for photovoltaic power based on fuzzy C-means(FCM)clustering and long short-term memory networks(LSTM).Firstly,the historical data are divided into time series segments by sliding time windows,the power fluctuation characteristics are analyzed in time series segments as samples.Then,three feature indicators are defined to extract the fluctuation patterns and the FCM method is used to cluster the time series segments,the data set is reconstructed after clustering and combined with LSTM network to build a photovoltaic power prediction model.Finally,the clustering performance and prediction performance of the proposed method are verified using the measured dataset of Australian photovoltaic plants.The simulation results show that the proposed method has higher clustering accuracy and prediction precision.

关 键 词：光伏功率预测 模糊均值聚类 长短期记忆网络 特征提取 相关性系数 

分 类 号：TM74[电气工程—电力系统及自动化]