A novel optimization method combining metaheuristics and machine learning for daily optimal operations in building energy and storage systems
Shintaro Ikeda, Tatsuo Nagai
Abstract
In recent years, research on operational optimization of buildings and regional energy systems has been actively conducted. There are several groups that utilized linear approximations, considered nonlinearity, conducted scenario-based research, and used an optimization algorithm to find an optimum solution. In terms of real-world implementation in buildings, the nonlinearity of machine characteristics should be considered within practical computation time because linearization incurs modeling costs, and computational resources are limited. Hence, the authors propose a hybrid algorithm that consists of metaheuristics and machine learning for optimizing daily operating schedules in building energy systems. The deep neural network machine learning technique was used to predict optimal operations of integrated cooling tower systems, and metaheuristics were used to optimize the operation of the other components. The proposed method may reduce daily operating costs by more than 13.4%. In addition, the integrated cooling tower system evaluated in this study reduced cost and energy requirements compared to an individual cooling tower system.