Forecasting rooftop photovoltaic solar power using machine learning techniques

Solar power plants offer a healthy substitute for traditional thermal power plants. However, the management and quality of power in the current energy grids are threatened by the environmental effects of relying too much on solar power. Accurate solar power prediction is essential for designing and managing solar power plants. The distribution grid runs more smoothly due to improved solar power forecasting, which assures accurate solar power generation forecasts. Artificial intelligence (AI) based algorithms are becoming increasingly well-liked and successful at estimating solar power forecasts. Utilizing a machine learning (ML) model to explore solar power in KSA (Kingdom of Saudi Arabia) is limited, despite significant advances being made elsewhere. This study uses machine learning ensemble models to predict solar power at Commercial building of Saudi Arabia, KSA. The meteorological data gathered from the research location includes the module's temperature, ambient temperature, irradiance, and real power. The following significant elements briefly highlight the main contributions of this research. The data gathered from the solar photovoltaic system is initially visualized using a data analysis tool. Second, by employing multiple statistical indices to predict values from a time-series dataset on solar power, we assess the forecasting power of different machine-learning approaches. The comparative analysis reveals that the xGBOOST model performance is superior than that of the other ML models with a Mean-Square Error (MSE) value of 0.0036, Root Mean-Square Error (RMSE) of 0.0191, Mean Absolute Percentage Error (MAPE) of 0.6898, Mean Absolute Error (MAE) 0.0130 and Coefficient of determination (R 2 ) 0.975. Hence, the suggested prediction methodologies can minimize forecast intricacy with tiny reconstruction errors. Regarding forecasting accuracy, the xGBoost (Extreme Gradient Boost Regression) model outperforms several other popular models. • A novel machine learning technique is used for the Evaluation of Rooftop Photovoltaic Solar Power Potential. • Small datasets are used to train a machine-learning model for PV power potential assessment. • The results show that the Extreme Gradient Boost Regression, or xGBoost, performs better than the other models. • The efficacy of the developed models is evaluated using a variety of performance assessment metrics. • A comparison of different machine learning models is performed.