An ANN-GA Semantic Rule-Based System to Reduce the Gap Between Predicted and Actual Energy Consumption in Buildings

This paper addresses the endemic problem of the gap between predicted and actual energy performance in public buildings. A system engineering approach is used to characterize energy performance factoring in building intrinsic properties, occupancy patterns, environmental conditions, as well as available control variables and their respective ranges. Due to the lack of historical data, a theoretical simulation model is considered. A semantic mapping process is proposed using principle component analysis (PCA) and multi regression analysis (MRA) to determine the governing (i.e., most sensitive) variables to reduce the energy gap with a (near) real-time capability. Further, an artificial neural network (ANN) is developed to learn the patterns of this semantic mapping, and is used as the cost function of a genetic algorithm (GA)-based optimization tool to generate optimized energy saving rules factoring in multiple objectives and constraints. Finally, a novel rule evaluation process is developed to evaluate the generated energy saving rules, their boundaries, and underpinning variables. The proposed solution has been tested on both a simulation platform and a pilot building - a care home in the Netherlands. Validation results suggest an average 25% energy reduction while meeting occupants' comfort conditions.