Machine Learning based Load Forecasting using Efficient Feature Extraction Scheme in Smart Grid

The smart grid requires the integration of emerging technologies for the digitalization of the demand side. This study deals with short-term load forecasting using candidate machine learning models to utilize demand-side technologies. The dataset of Panama’s national grid has been implemented with new features through correlation mapping and tested for the five machine-learning models. The results illustrate that Random Forest (RF) and Extreme Gradient Boost (XGBoost), two ensemble-based machine learning models, perform better considering evaluation metrics, Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). After that, machine learning models are also tested with sequential historical data and used as extracted features for hourly and weekly prediction. It shows that model performances have improved if more sequential historical data is used. The study’s findings will be beneficial in using these outperformed models to anticipate an accurate load for current intelligent energy transitions.