Feature Selection for Electricity Power Forecasting of Solar Power Plants

Forecasting electricity production has an essential role in overcoming the instability of electricity supply, especially in the context of Solar Power Plants (SPP), which are highly dependent on sunlight. Accurate forecasting of SPP can increase the efficiency of electricity generators and reduce high operational costs. One of the main challenges in forecasting Solar Electricity Production lies in selecting the appropriate input features in machine learning models. External factors, such as different weather conditions, play an essential role in influencing the availability of electricity at the SPP. This research focuses on determining the proper input features for machine learning-based active power production forecasting in SPP. We use two deep-learning models, the Long Short-Term Memory (LSTM) and Bidirectional Long Short-Term Memory (BiLSTM) models, to design active power production in SPP. Other than weather features, we also add a time feature, i.e., hourly time information, which can improve the accuracy of SPP forecasting. The results of this research show that selecting the proper weather features and adding time features can increase the model's accuracy in forecasting active power production. The LSTM produced the best model with an R-squared of 0.842 and an RMSE of 34.122 for a one-month prediction. The significance of this research is the potential to optimize the use of generators and reduce expensive operational costs in the context of SPP.