基于改进广义神经网络的光伏阵列短期功率预测  

Short Term Power Prediction of Photovoltaic Array Based on Improved Generalized Neural Network

作　　者：杨德州 尹立夫 王洲 刘永成 王慧娟 YANG Dezhou;YIN Lifu;WANG zhou;LIU Yongcheng;WANG Huijuan(State Grid Gansu Economic Research Institute,Lanzhou Gansu 730050,China;State Grid Gansu Electric Power Company,Lanzhou Gansu 730030,China;State Grid TianShui Power Supply Company,Tianshui Gansu 741000,China)

机构地区：[1]国网甘肃省电力公司经济技术研究院,甘肃兰州730050 [2]国网甘肃省电力公司,甘肃兰州730030 [3]国网天水供电公司,甘肃天水741000

出　　处：《电子器件》2022年第3期722-726,共5页Chinese Journal of Electron Devices

基　　金：甘肃省电力公司经济技术研究院项目(SGGSJY00PSWT2000053)。

摘　　要：光伏系统短期预测对调度电力资源、较少弃光用电、提升电站效能及维持光伏系统平稳运行具有重要意义。考虑到光伏系统输出参数如功率受到天气因素影响,是一个非平稳随机信号。对非平稳随机信号的短期预测,传统数学模型方法具有局限性。基于广义神经网络方法并通过前期数据预处理提取趋势项信号,进行分类别模型训练,获得较为精准不同天气类型的训练模型。实验表明所提出的方法在各类天气类型下,对于短期功率预测具有较高的准确性,可为光伏电站平稳运行、区域电力调度提供参考依据。The short-term prediction of photovoltaic system is of great significance for dispatching power resources,reducing the rate of abandonment of photovoltaic power,improving the efficiency of power station and mainteining the smooth operation of photovoltaic system.The output parameters of photovoltaic system such as power are affected by weather factors,so they are non-stationary random signals.For the short-term prediction of non-stationary random signal,the traditional mathematical model method has limited application.Based on the generalized neural network method,the trend signal is extracted by pre-processing data,and the classification models are trained to obtain more accurate training models of different weather types.The experimental results show that the proposed method has high accuracy for short-term power prediction under various weather types,and can provide reference for the smooth operation of photovoltaic power plants and regional power dispatching.

关 键 词：光伏阵列 广义回归神经网络 FCM聚类 功率预测 

分 类 号：TM615[电气工程—电力系统及自动化]