The robustness of genetic algorithms in solving unconstrained building optimization problems

This paper investigates the robustness of a Genetic Algorithm (GA) search method in solving an unconstrained building optimization problem, when the number of building simulations used by the optimization is restricted. GA search methods can be classified as being probabilistic populations based optimizers. The probabilistic nature of the search suggests that GA’s may lack robustness in finding solutions. Further, it is a common perception that since GA’s iterate on a population (set) of solutions, they require many building simulations to converge. It is concluded here that a particular GA was robust in finding solutions with 1.4% mean difference in building energy use from that for the best solution found in all trial optimizations. This performance was achieved with only 300 building simulations (in any one trial optimization).