An integrated decision support system for multi-target forecasting: A case study of energy load prediction for a solar-powered residential house

Abstract

This paper proposes a data-driven framework for multi-target forecasting. We apply the proposed framework to forecast energy load in solar-powered residential houses using meteorological and temporal inputs. We adopt five predictive models of gradient boosting, least angle regression, multi-layer perceptron, random forest, and partial least square as well as three information fusion techniques of simple average, proportionate mean squared error, and learned weights, and compare their forecasting accuracy. We also perform a comprehensive sensitivity analysis and extensive tuning of the parameters of these models. The results of the energy load application reveal that the learned weights information fusion technique generates the most accurate predictive model. The results unfold that energy consumer can reliably plan for their forecasted energy loads considering expected meteorological conditions. Finally, we apply the best-worst multi-criteria decision-making method to find the weight of importance of target features using energy experts’ opinions and the percentage overall load of each output. The proposed framework can be applied to any multi-target forecasting problem.