Machine learning methods for prediction of hot water demands in integrated R744 system for hotels

Abstract

Load forecasting can help modern energy systems achieve more efficient operation by means of more accurate peak power shaving and more reliable control. This paper proposes a framework based on machine learning algorithms to forecast the hot water usage for a Norwegian hotel. The framework is tested on the real data from an integrated R744 HVAC and domestic hot water system with a 6 m3 thermal storage. Recorded operational data and ambient temperatures are utilized to build several forecasting models that can predict demands with high accuracy. The hot water usage accounts for 52 % of hotels’ heat load, where strategic accumulation of the hot water storage can improve the overall system performance. Charging the hot water storage according to three-hour-ahead demand predictions presents significant savings potential. This work can facilitate a demand management strategy and thus improve the energy efficiency of the integrated 744 system.