Comparison of metaheuristics and dynamic programming for district energy optimization

Metaheuristic optimization methods, as model-free methods, are expected to be applicable to practical issues (e.g., engineering problems). Although optimization methods have been proposed or improved through many theoretical studies, they should be tested using not only some benchmark functions, but also other models representing practical situations, such as those involving discrete control variables and equality or inequality constraints. Hence, differential evolution (DE)-based constrained optimization methods were applied to district energy optimization in this study. Several different types of DE-based methods and dynamic programming which was utilized to obtain theoretical results, were compared. The proposed DE-based method, ϵ-constrained DE with random jumping II (ϵDE-RJ-II), proved capable of producing results differing by only 2.1% from the theoretical results in a computation time 1/457 of that required by dynamic programming. Therefore, ϵDE-RJ-II has high potential to provide comprehensive district energy optimization within a realistic computation time.