基于灰色系统校正-小波神经网络的光伏功率预测

Photovoltaic Output Prediction Based on Grey System Correction-Wavelet Neural Network

为提高非理想天气条件下的光伏功率预测精度,提出基于灰色系统校正-小波神经网络(wavelet neural network,WNN)的预测方法。首先以基于相似日算法的WNN进行逐时功率预测,并进行累加获得日累加功率。根据光伏出力历史数据,确定各广义天气类型的平均偏差比,并以平均偏差比进行平滑处理后的相邻日功率建立离散灰色系统模型(discrete gray model,DGM),进行日总功率预测并获得及其判断区间。最后以日总功率值判断区间为标准对累加功率值进行校正,得到校正后的各时段的预测值。算例结果验证了所提方法的有效性。

To improve prediction accuracy of photovoltaic power under non-ideal weather conditions, a wavelet neural network（WNN） prediction model of PV output is proposed based on correction of gray system model. Firstly, hourly power is predicted with WNN based on similar day algorithm, daily accumulative power is obtained by summing the hourly powers. Then, average deviation ratio of each generalized type of weathers is determined according to historical photovoltaic output data. Adjacent daily power is smoothed with average deviation ratio used to build discrete gray model（DGM）, to obtain daily total power and its judgment interval. Finally, daily accumulative power is corrected with judgment interval as reference to obtain post-correction hourly power. Effectiveness of the method is validated with a practical example.