Deep Learning Based Solar Power Forecasting with Improved Representation

Abstract

The growing number of solar power plants is making it a viable option for renewable energy sources to meet the energy needs of communities. Moreover, accurate prediction of output improves integration of these plants into the grid. Deep learning models, which can take advantage of high-performance processors, and the data, have the potential to improve solar power prediction. This study proposes an auto-encoder and Gated Recurrent Unit based method for improved forecasting. The effectiveness of this approach is compared to other state of art for predicting solar power over different time periods. The performance of these models is evaluated using different performance parameters. The outcome of the analysis validates the effectiveness of the proposed model.