Day-ahead Prediction of Building District Heat Demand for Smart Energy Management and Automation in Decentralized Energy Systems

Abstract

This paper proposes an Artificial Intelligence (AI) based data-driven approach to forecast heat demand for various customer types in a District Heating System (DHS). The proposed day-ahead forecasting approach is based on a hybrid model consisting of Imperialistic Competitive Algorithm (ICA) and Support Vector Machine (SVM). The model is built using two years (2015 - 2016) of hourly data from various buildings in the Otaniemi area of Espoo, Finland. Day-ahead forecast models are also developed using Persistence and four other AI based techniques. Comparative forecasting performance analysis among these techniques was performed. The proposed ICA-SVM heat demand forecasting model is tested and validated using an out-of-sample one-year (2017) hourly data of the buildings’ district heat consumption. The prediction results are presented for the out-of-sample testing days in a one-hour time interval. The validation results demonstrate that the devised model is able to predict the buildings’ heat demand with an improved accuracy and short computation time. Moreover, the proposed model demonstrates outperformed prediction accuracy improvement, compared to the other five evaluated models.