基于PNN/PCA/SS-SVR的光伏发电功率短期预测方法

光伏发电功率预测对太阳能开发利用、电网稳定安全运行具有重要意义。提出一种融合了概率神经网络(PNN)、主成分分析法(PCA)、分散搜索(SS)和支持向量机回归(SVR)的光伏输出功率预测模型。首先结合天气信息通过PNN将天气划分为晴、多云、阴、雨4种类型,然后在每种天气类型下,利用PCA对影响光伏出力的多个气象因素,如太阳辐射强度、温度和相对湿度等进行降维、转换成少数几个主成分作为输入向量,最后建立SS算法优化SVR的光伏发电功率短期预测模型。结果表明,该模型实现了对不同天气类型下的光伏出力较为精准的预测,具有一定的可行性及指导意义。

Short-term Forecasting Method of Photovoltaic Output Power Based on PNN/ PCA/SS-SVR

Forecasting of photovoltaic power generation plays a significant role in solar production and the stable and secure operation in grids. A new photovoltaic forecasting model based on the converging algorithm including probabilistic neural network(PNN), principal component analysis(PCA), scatter search(SS)and support vector regression(SVR)is proposed. First of all, the weather type is divided into sunny/cloudy/overcast/rainy by using PNN combined with meteorological information. Then under each weather type, a reducing dimension calculation using PCA is applied to various meteorological factors which affect the photovoltaic output. Finally, a short-term forecast model of photovoltaic output based on SVR optimized with SS is established, using the extracted fewer principal components as input vectors. The results show that the forecasting of photovoltaic power is realized with the proposed combination forecasting model and the preferable results are achieved. This work is supported by the State Key Program of National Natural Science Foundation of China(No. 61533012)and Shanghai Natural Science Foundation(No. 14ZR1421800).