Abstract: Thermal batteries based on phase change materials (PCMs) are a key technology for energy savings and carbon reduction in building heating since latent thermal energy storage by PCMs with high energy storage density indicates great potential to utilize renewable energies. However, the complicated structure of finned-tube heat exchangers and the nonlinear melting-solidification process of PCMs require a great deal of computation and time resources, highly restricting the engineering design and application of PCM-based thermal batteries. To address these issues, this study proposed a universal machine learning-enabled performance prediction framework for finned-tube PCM-based thermal batteries designed for residential domestic hot-water supply, in which thermal energy is stored in PCM during the charging process and released to cold water during the discharging process to provide usable hot water for end users. First, a simplified simulation method coupling a 1D tube model with a 3D CFD model is established for the rapid performance computation of thermal batteries. Second, the deep operator network framework is introduced to directly map static parameters to the time-based temperature response of outlet hot water validated by the simulation and previous experimental results. Based on the machine learning-enabled performance prediction framework, the effects of critical parameters such as tube diameter and thickness combination (Or/Th), fin distance (Fd), and flow rate (Fr) on the heat storage capacity and heat release power are systematically analyzed, providing guidelines for the proposed flowrate feedback regulation strategy of thermal batteries to satisfy different usage temperatures Tuse and time constraints. The results show that the proposed framework can accurately predict the outlet temperature and available total volume of hot water output by thermal batteries under various conditions