基于主成分分析线性回归的光伏发电功率预测研究

针对经典光伏发电功率物理模型预测精度不高、可能遗漏关键气象因子的问题,基于数据驱动思想提出一种主成分分析、逐步线性回归以及多种检验相结合的光伏发电功率预测模型混合建模方法.采用相关性分析提取关键气象因子自变量,通过逐步线性回归对历史样本数据进行训练得到初始模型;对初始模型进行拟合程度、有效性以及多重共线性等多种检验,根据检验结果,采用主成分分析对初始模型自变量进行降维重构,得到修正模型;通过残差分析实现修正模型正确性的自校验;基于多种时间尺度对经典模型、未修正模型以及修正模型的预测结果进行详细对比分析.仿真结果表明修正模型提高了预测精度及所提混合建模方法的有效性与优越性.

Research on Photovoltaic Power PredictionBased on Principal Component Analysis and Linear Regression

The classical model for physical photovoltaic power prediction has problems of poor prediction accuracy and missing key meteorological factors. To address those problems, a hybrid modeling method combining principal component analysis and stepwise linear regression, and multiple verification for photovoltaic power prediction is proposed based on the idea of data-driven. Firstly, the independent variables of key meteorological factors are extracted by correlation analysis, and then the initial model is obtained by training the historical sample data by stepwise linear regression. Secondly, the fitting precision and validity and multi-collinearity of the initial model are verified. According to the verification results, the principal component analysis is used to reduce the dimension of the independent variables of the initial model and reconstruct the modified model. Then, the self-verification of the correctness of the modified model is realized by residual analysis. At last, the prediction results of the classical model, the unmodified model and the modified model are compared and analyzed in detail based on a variety of time scales. The simulation results show that the modified model improves the prediction accuracy, and the effectiveness and superiority of the proposed hybrid modeling method are verified.