Physics-Informed Neural Network Model for Flexibility Modeling of Electric Water Heaters

Abstract

Forecasting water temperature in Electric Water Heaters (EWH) with limited information is essential for devising a control strategy for aggregated flexibility activation considering user comfort and rebound effects. In this paper, we present a novel Physics-Informed Unsupervised Neural Network (PINN) architecture for modeling the thermal dynamics of EWHs with only power consumption data. Physics from the single-zone thermal grey-box EWH model is encoded into the PINN loss function to incorporate domain knowledge. Additionally, a custom Recurrent Neural Network (RNN) cell is developed to capture the exponential evolution of water temperature profile. Two models for EWH ON- and OFF-dynamics are trained with power consumption as the sole input data. Temperature prediction results indicate that the proposed model has a similar performance as the traditional single-zone grey-box EWH model, thereby demonstrating the ability of proposed model to learn the underlying physics behind the EWH operation without water temperature data.