Metamodel-Based Hyperparameter Optimization of Optimization Algorithms in Building Energy Optimization

Building energy optimization (BEO) is a promising technique to achieve efficient designs. The efficacy of algorithms imperative for the BEO and significantly dependent on algorithm hyperparameters. Currently, studies focusing hyperparameters are scarce, common agreement how set their values, especially problems, still lacking. This study proposes metamodel-based methodology hyperparameter applied in BEO. aim maximize algorithmic avoid failure because improper settings. method consists three consecutive steps: constructing specific problem, developing an ANN-trained metamodel optimizing with nondominated sorting genetic II (NSGA-II). To verify validity, 15 benchmark problems different properties, i.e., five building models design variable categories, were constructed numerical experiments. For each four commonly used algorithms, (GA), particle swarm (PSO) algorithm, simulated annealing (SA), multi-objective (MOGA), optimized. Results demonstrated that MOGA benefited most from terms quality obtained optimum, while PSO computing time.