A combined approach to Measurement-System-Design and system identification in building energy retrofits

• Domain knowledge enhances measurement selection and system identification methods. • The optimal times for performing measurements can be effectively identified. • Parameter value uncertainty was reduced with half of the available measurements. • Maximum joint entropy is a promising metric for identifying key measurements. Effective building-energy retrofits are needed to enhance the energy efficiency of existing buildings. However, discrepancies between predicted and actual energy demand, known as the performance gap, undermine retrofit effectiveness. This work addresses the performance gap by enhancing measurement effectiveness through measurement-system design and system identification for assessing building performance. Measurement-system design involves selecting the type of measurement, timing, and locations. System identification consists of identifying models that align with observed data. Gaps include the lack of measurement-system-design methodologies that avoid preliminary measurements and consider shared information, as well as the need for practical system-identification approaches that incorporate uncertainty and modelling assumptions. This paper addresses these gaps by adapting two methodologies from other fields to building-energy retrofits, leveraging domain-specific knowledge alongside building-energy simulations and, as an initial exploratory step, synthetic measurements. A hierarchical algorithm that maximises joint entropy is implemented for measurement-system design, while Error Domain Model Falsification (EDMF), a Bayesian inference variant, is implemented for system identification. Their strengths are assessed, and a combined approach to building energy retrofits is proposed, focusing on measurement timing. EDMF significantly reduces uncertainty in model parameter values and energy demand predictions. The hierarchical algorithm yielded similar system identification results using only half of the available measurements. The successful adaptation of these methods is attributed to domain-specific knowledge, which differs significantly from previous applications of the techniques. EDMF effectively manages uncertainty and provides feedback on modelling assumptions, while the hierarchical algorithm optimises measurement selection, together showing potential for enhancing the effectiveness of energy retrofits.