Forecasting of Solar Power by LSTM Derivatives: A Comparison Study

The utilization of solar energy diminishes the reliance on non-renewable fossil fuels and mitigates climate change consequences by lowering carbon emissions. Photovoltaic systems are among the most used technologies for transforming solar energy into useful electrical energy. However, power generation in solar systems varies due to numerous system-related factors. Therefore, the estimation of photovoltaic energy is important in terms of management, planning, integration and sustainability. Additionally, manual solar energy forecasting requires significant human effort and expertise to analyse and interpret the data. This process is time-consuming and open to errors. Thus, recently, the researchers around the world have been focusing on the application of artificial intelligence and machine learning approaches to forecast power production in the highly unpredictable renewable energy sector. In this study, the performances of the derivatives of long-short term memory (LSTM) neural networks are investigated in one hour ahead forecasting of the solar power output for a solar farm located in Türkiye. The aim of the study is to determine the best performing LSTM derivative for this problem and demonstrate the application of the algorithms in a specific case study, where the utilized data is publicly available. Thus, the methodology presented here can be utilized by anyone, including government agencies, to forecast solar power output of any solar farm, making this study a significant contribution to the existing literature.