Modeling and Parameter Estimation of Electric Thermal Storage utilizing Residual Components for Residential Consumer

Electric Thermal Storage (ETS) systems are conventionally programmed for participation in the typical Demand Response programs. Particularly, in the context of Dynamic Energy Markets (DEMs), the ETS enables residential customers to actively participate in lowering their energy costs. It is imperative to build a model of the ETS-based heating of the residential thermal zone to achieve precise indoor temperature predictions. This model can also assist in estimating energy demands, providing advantages during their integration into DEMs. Accordingly, this work introduces a grey box modeling technique for predicting indoor temperatures. It leverages residual components to capture the differences between the trained model and the experimentally recorded data, thereby highlighting prediction discrepancies. Subsequently, the Least Square (LS)-based parameter estimation technique for predicting the thermal zone and ETS brick temperatures utilizing Quantile Regression (QR) and Huber Loss (HL) functions is proposed. Comparative prediction results for indoor and ETS brick temperatures over a 24-hour period with the experimental data utilizing the proposed method and ensemble learning techniques such as Extra Trees Regressor (ETR) and Random Forest Regressor (RFR) are presented. The proposed method performs a more accurate forecast than a conventional model without residuals and ensemble learning techniques. The reduction in Mean Absolute Error (MAE) to under 0.4°C for indoor temperature prediction compared to real data demonstrates better performance than conventional method and ensemble learning techniques.