基于PCA-LMBP神经网络的北京地区直散分离预测

Prediction of Beam-Diffuse Radiation Separation in Beijing Based on PCA-LMBP Neural Network

为确定气象环境变化对散射比的影响程度,提出主成分分析(PCA)与LMBP神经网络相结合的光伏直散分离模型,利用北京地区5年逐日地面辐射资料,通过相关系数矩阵,选出清晰度指数、日照百分率、PM2.5、总云量和气温日较差作为突出气象影响因子,采用PCA法对多维影响因子作预处理,根据贡献率选出3个主成分,将其作为LMBP神经网络的输入参数,进而通过误差分析方法分别对MLR模型、PCA-MLR模型、PCA-BP模型和PCA-LMBP模型进行评估。结果表明,PCA-MLR模型和PCA-LMBP模型的散射比预测值与实测值更吻合,其中基于PCA-LMBP神经网络的直散分离模型预测精度最高、泛化性能最好,具有一定的实际应用价值。

In order to explore the influence of the change of meteorological environment on the diffuse irradiation ratio, this paper proposed a model of beam-diffuse radiation separation of the PV system based on principal component analysis and LMBP neural network. The daily irradiation and meteorological environmental data of five years in Beijing were collected. On the basis of correlation coefficient matrix, the clearness index, the percentage of sunshine, PM2.5, the total cloud cover and the diurnal temperature range were chosen as a highlight of the meteorological environmental factors. Principal component analysis （PCA） method was used to pre-process the multidimensional influencing factors, and three principal components were selected according to the contribution rate. The principal components were used as input parameters of LMBP neural network. Finally, the error analysis method was used to evaluate the effect of MLR model, PCA-MLR model, PCA-BP model and PCA-LMBP model. The results show that the predictions of PCA-MLR model and PCA-LMBP model for the diffuse irradiation ratio are closer to the measured values. Based on PCA-LMBP neural network, the prediction accuracy and generalization performance of the model are the best. So, it has a certain value for practical application.