An Ensemble Model for Prediction of Day Ahead Photovoltaic Power Using Learning Approaches

Forecasting the generation of photovoltaic power in a day to day basis is of prime importance for efficient operation of smart grid power system. The day ahead photovoltaic power can be predicted using machine learning models, but their accuracy is generally unstable due to changing atmospheric conditions. Stacking, an advanced ensemble learning approach implements high level meta-model which is trained on the base learner model outputs to make the predicted values more accurate in changing weather conditions. In this work, a Stacked Ensemble Model (SEM) is designed to increase the accuracy and precision of the forecasted photovoltaic power generated by the power grid. The proposed model implements two machine learners namely; Generalized Regression Neural Network and Elman Neural Network. The base learners are trained using different sub-training datasets and integrated with meta model to predict day ahead photovoltaic power. The stacked ensemble model is tested across three weather scenarios namely sunny, cloudy and rainy and it significantly outperforms the existing single models and traditional ensemble models with respect to Mean Square Error. The error is substantially reduced in Stacked Ensemble Model and accuracy of predicted power is 0.98 for the various sub-training datasets.