基于网络集成的太阳全辐射超短期混合预测方法

A Hybrid Model for Very Short-term Forecast of Global Horizontal Irradiance Based on Network Ensemble

针对引起光伏发电量波动和突变的太阳全辐射,提出了一种基于网络集成的太阳全辐射超短期混合预测模型。首先,根据太阳全辐射在不同天气情况下的变化特性,构建混合模型的初级子模型,分别为针对变化平缓的多元线性回归模型和针对剧烈变化的非线性支持向量回归模型和神经网络模型;然后,设计一个含有两个隐层的四层感知器模型,融合三个初级子模型的预测结果,实现对未来15分钟太阳全辐射量的预测。采用网上公开数据库数据测试提出的混合预测模型性能,实验结果为所提出的混合模型预测值和真实测量值的相关系数高达0.945、预测平均误差约为0.9Wm-2,表明该混合模型的有效性。

A hybrid model has been proposed for very short-term forecast of global horizontal irradiance(GHI)which causes the volatility and change of photovoltaic power generation.Firstly,primary sub-models were constructed according to the change characteristic of GHI,including a multiple linear regression model for gentle change and a support vector regression model and neural network model for strong volatility.Secondly,a four-layer perceptron model with two hidden layers is developed to weight the predictions of three sub-models above in order to achieve the 15-min ahead GHI.Measured data in an open database online were used to evaluate the performance of the proposed hybrid model.The results show that the correlation coefficient of the predictions by the hybrid model and the measured values reaches 0.945 and the mean error is about 0.9Wm-2,which verifies the effectiveness of the presented hybrid model.