Solar Photovoltaic Output Power Prediction with a Wavelet Transform-Based Long Short-Term Memory Network

The Sun is the most powerful source of energy present in the world, and sunlight is the most abundant energy source received by the Earth. The overall amount of solar energy which the earth receives exceeds the world’s present as well as future requirements by a large margin. If harnessed properly, solar energy has the ability to meet all future demands of energy. Solar-based energy is going towards becoming one of the most promising sources for producing power for commercial, residential and industrial applications. A wavelet transform-based deep learning long short-term memory (WTLSTM) network is used in this study to predict solar photovoltaic (SPV) output power using three dependent data variables, including temperature, direct horizontal radiation and diffuse horizontal radiation. The paper looks into the problem of selecting appropriate error metrics and features. The WTLSTM model was created and tested using actual SPV output power that was obtained from Greece, Southeast Europe. Two key metrics are used to assess how well-developed models predict the future solar power: nominal mean absolute error (nMAE) and nominal root mean square error (nRMSE). The simulation results show that the suggested approach yields respectable short-term forecasting outcomes.