Forecasting capacitor banks for improving efficiency of grid-integrated PV plants: A machine learning approach

Abstract

Grid-connected rooftop PV systems are becoming more popular to promote renewable energy. The rooftop PV may diminish the system's energy efficiency by lowering the power factor (PF) on the grid side. The current work provides a machine learning approach that estimates the necessary capacitor banks to boost the PF to unity, enabling proactive remedial action for energy savings. Various machine learning models, such as linear regression, ridge regression, lasso regression, random forest, decision tree, XGBoost, Adaboost, and gradient boosting, are evaluated to improve the system's efficiency. The best model is Lasso Regression, which produces a high R² score of 0.89 with low MSE and MAPE values. The model is based on real-time data collected from a 100 kWp PV plant connected to an 11 kV grid supply and an institutional building load. The model undergoes validation by implementing the forecasted capacitor banks. According to the findings, a 10.60 kVAR-rated shunt capacitor is required to maintain the PF at unity and save an average of 1673.52 kWh of energy per month. This work highlights the necessity of implementing Lasso regression in energy management systems to improve PF, decrease electricity costs, and reduce environmental impacts.