DEVELOPMENT OF THE DIGITAL-TWIN FOR BUILDING FACILITIES (PART 3): A COMPARISON OF METAHEURISTICS AND REINFORCEMENT LEARNING FOR OPTIMAL CONTROLS

Abstract

In this study, to quantitatively evaluate a metaheuristic and a model-based reinforcement learning which are control methods using a predictive model, these methods were compared for energy costs and computational loads. As a result, it was revealed that a metaheuristic has more saving energy costs, whereas model-based reinforcement learning has lower computational loads. Therefore, it is necessary to select an appropriate method to a target system, for example, a metaheuristic is more suitable for a mode control, and a model-based reinforcement learning is more suitable for a water flow control of pumps.