A comparison of building energy optimization problems and mathematical test functions using static fitness landscape analysis

Computational optimization is gaining popularity in energy-efficient building design. For choosing an algorithm or setting its parameters, often mathematical test functions are employed. This study, therefore, investigates differences and similarities between such test functions and building energy optimization (BEO) problems. A fitness landscape analysis (FLA) is conducted with existing and newly proposed metrics, shedding light on the characteristics of optimization problems. We use the COCO testbed and compare it to BEO problems from the literature. Results suggest that for most FLA metrics there is no statistical difference between the set of test functions and BEO problems. Also, characterizing an archetypical BEO problem appears infeasible due to the high heterogeneity we can observe in FLA metric scores. However, using hierarchical clustering we can identify similarities between test functions and groups of BEO problems. Such knowledge may be exploited for selecting or calibrating an algorithm, thus facilitating its effective use in practice.