Forecasting the Electricity Consumption for Energy Management Software Using an Ensemble Model

The paper investigates the problem of poor-quality of consumption forecasts in energy management software. It was found that the ensemble machine learning method can provide better prediction accuracy than each individual method by combining the strong features of different methods that have different structural characteristics. Based on this idea, a choice of basic models with linear, recursive, and tree structure was offered. This article presents an ensemble electricity consumption prediction model combining several base models: Autoregression, Random Forest, Support Vector Regression, Extreme Gradient Boosting, and Long Short-Term Memory. The individual models were built and trained. Considering the benefits from each other's structure advantages, each basic model produced a prediction weighting coefficient that was integrated into the combined model. Weighting coefficient, were adjusted using the Root Mean Squared Error loss function with a Sequential Least Squares Programming optimization algorithm. The proposed ensemble prediction model showed improved metrics, compared to individual base models. Therefore, energy management software can use our ensemble model architecture to forecast electricity consumption with higher accuracy.