A Pareto-optimal modeling procedure for thermostatically controlled loads

The integration of thermostatically controlled loads (TCL) into microgrids offers significant potential for demand-side management and flexibility provision. However, effective utilization requires models that are both accurate in representing thermal dynamics and interpretable in terms of physical parameters. Existing TCL models often prioritize data fitting, neglecting the physical meaning of parameters, or fail to guarantee a balanced trade-off between accuracy and interpretability. This paper presents POMP-TCL, a novel and generalizable Pareto-optimal modeling procedure for TCL. The method quantitatively defines interpretability as the deviation from theoretical thermal parameters and integrates it alongside accuracy into a multi-objective optimization framework. The approach is demonstrated through a progressive evolution of the authors’ prior works, moving from black-box to gray-box models and culminating in the full application of POMP-TCL to a wine cooler system equipped with a Peltier module. Results show that POMP-TCL consistently reduces root mean square error (RMSE) while significantly improving interpretability, enabling the construction of a Pareto front from which optimal trade-off solutions can be selected. Comparative analysis with related studies reveals that most achieve only a single point on the trade-off curve, lacking guaranteed balance. The proposed framework addresses this limitation and offers a systematic, transparent, and physically meaningful modeling procedure applicable to a wide range of TCL in microgrid applications.