A New Long-Term Photovoltaic Power Forecasting Model Based on Stacking Generalization Methodology

In recent times, solar energy has become a highly promising source of energy and one of the most regular types of sustainable energy. Forecasting the availability of solar energy has become a concern of many studies because of the intermittent characteristics of solar power. This study proposes a new stacked generalization methodology for predicting long-term photovoltaic power. In the proposed methodology, the base learners used consisted of group method of data handling (GMDH), least squares support vector machine (LSSVM), emotional neural network (ENN), and radial basis function neural network (RBFNN). The backpropagation neural network (BPNN) served as the meta-learner in the stacked approach. The proposed stacked generalization method showed superiority over the four standalone state-of-the-art methods (GMDH, LSSVM, ENN, and RBFNN) when tested with real data using performance indicators such as Bayesian information criteria (BIC), percent mean average relative error (PMARE), Legates and McCabe index ( LM ), mean absolute error, and root mean square error. The stacked model had the lowest BIC and PMARE values of 10,417.54 and 0.3617% for testing results. It also had the highest LM score of 0.996711 as compared with the benchmark standalone models, confirming its strength in forecasting photovoltaic power.