A novel integrated photovoltaic power forecasting model based on variational mode decomposition and CNN-BiGRU considering meteorological variables

Abstract

• An integrated photovoltaic power forecasting model considering meteorological variables is proposed. • Multivariate models considering meteorological variables outperform the univariate ones. • CNN is superior in data-mining than the traditional statistical feature selection methods. • BiGRU yields better forecasting performance than the other machine leaning models. • VMD can enhance the forecasting accuracy of the single machine learning methods. The present study aims to develop an integrated multivariate model based on variational mode decomposition (VMD), convolutional neural network (CNN) and Bi-directional gated recurrent unit (BiGRU) for photovoltaic (PV) power forecasting. The PV power series is firstly decomposed into several sub-modes using VMD. For each sub-mode, the antecedent values and several meteorological variables including solar radiation, temperature, relative humidity and wind speed are employed to construct an input feature matrix to forecast the value of the future moment. And then the CNN technique is adopted to mine the underlying input-output relationship between the feature matrix and the target variable. After that the BiGRU network is adopted as a predictor to forecast the future value of each sub-mode. The final PV power forecasting result is acquired by combining the forecasting values of the sub-modes. To demonstrate the performance of the proposed approach, the univariate models which consider antecedent PV power series as input features and the multivariate models which also consider meteorological variables are firstly compared. And then the proposed VMD-CNN-BiGRU method is compared with other eight contrast methods. Experimental results lend strong support to the superior performance of the VMD-CNN-BiGRU model considering meteorological variables.