Predicting Hourly PV Production of FESB Solar Plant Using SKTIME Library

Accurate forecasting of solar power production is essential for the effective integration of renewable energy into power grids, reducing reliance on conventional energy sources and enhancing grid stability. This paper presents a day-ahead solar power forecasting model developed using Python’s sktime library, employing Gradient Boosting Regression to predict PV power output based on weather and historical power data. The dataset used for prediction includes detailed weather features such as temperature, humidity, rain, and cloud opacity from multiple sources, including Open Weather and Solcast APIs. The model’s performance improved significantly by performing hyperparameter tuning with ForecastingGridSearchCV, reducing the Mean Absolute Error (MAE) from 1.107 to 1.006. Results indicate that high-quality cloud data is crucial for enhancing forecast precision, particularly under cloudy conditions, yielding a 31.1% MAE reduction compared to models without this feature. The findings demonstrate sktime’s flexibility in time series forecasting for renewable energy applications and highlight the importance of detailed meteorological data in improving the accuracy of solar power prediction.