Investigating the relationship between interpretability and performance for optimal rule-based control

Rule-based controls need to have a small number of nodes so that designers and engineers can understand, validate, and maintain them. This paper aims to investigate the relationship between the number of nodes and control performance. Using a case study, we found that models with a large number of nodes do not always have higher control performance. Specifically, optimal rules with a small number of nodes (n = 112) can achieve a 6.2% energy savings and an 8.8% increase in thermal satisfaction rate. The performance achieved is close to that with optimization using exhaustive search while maintaining high interpretability. Therefore, when designing optimal rule-based controls, it is important to investigate the relationship and select a suitable model based on the specific conditions or requirements of projects.