Comparing different parameter identification techniques for optimal control of building energy systems

Buildings account for a significant portion of global energy consumption and contribute to greenhouse gas emissions. Optimizing building energy systems during their operational phase is crucial for achieving carbon neutrality. This study aims to develop a technique for obtaining the parameters in RC (Resistor-Capacitor) models, which is a key step to develop Model Predictive Control (MPC) for real buildings. Two convex and nonconvex parameter identification methods are implemented to identify RC model parameters considering various quality and quantity of input data. The analyzed input data comprises weather conditions, heat flux from the heating system, and internal heat gain from occupants. The evaluation criteria include temperature prediction accuracy, the spread of identified RC parameters, and MPC performance (considering both the energy costs and thermal discomfort). The results demonstrate that RC parameters derived from the non-convex parameter identification (PI) model yield a lower Root Mean Square Error (RMSE) in predicting the operative temperature. Furthermore, increasing the quantity of data for PI improves the MPC performance. The extension of the PI optimization horizon has a negative impact on temperature prediction accuracy for both convex and non-convex models. However, it leads to a reduction in the standard deviation of RC parameters and an improvement in MPC performance under peak and typical heating conditions when their average is utilized. These findings underscore the significance of the precise identification of parameters and highlight the potential advantages of employing diverse optimization methods and appropriate input data for enhancing the operational efficiency of building energy systems.