HVAC optimisation studies: Sizing by genetic algorithm

Abstract

Previous research into the optimum sizing of hvac systems has focused on the use of direct search optimisation methods. Although these methods can find a solution, it is difficult for them to move discrete variables along nonlinear constraint boundaries and they often fail as a result. This paper describes the performance of a simple genetic algorithm search method when applied to such a problem. The formulation of the problem is described together with the operation of the algorithm. It is concluded that the algorithm exhibits rapid initial progress but that final convergence is slow due to the highly constrained nature of the optimisation problem. It is suggested that a more effective use of the constraint functions could improve the convergence and robustness of the algorithm. The performance of the algorithm is also sensitive to the problem formulation.