Optimization and operation of integrated air-water heat pump systems with energy storage and renewable energy based on deep learning

As renewable energy becomes more widely integrated into the power grid, building energy systems are evolving toward distributed, multi-source configurations. This study develops a deep learning (DL)-based control strategy for an integrated system comprising an air-to-water heat pump (AWHP), photovoltaic (PV) solar panels, and an energy storage system (ESS). Designed for residential use, the all-in-one system enables renewable energy generation and its storage in either a battery or a water tank. For efficient operation, accurate modeling of the flexible heating and cooling load, PV generation, and AWHP power demand is essential. These predictions are utilized by transactive control and model predictive control to schedule ESS charging and discharging, thereby managing peak demand and responding to time-of-use electricity rates and equipment temperature setpoints. DL models achieved CvRMSE values below 10% across all targets, with the Transformer model reducing training time by up to 20% while maintaining high accuracy. Simulations conducted using EnergyPlus validate the proposed approach, demonstrating up to 42% energy cost savings compared to conventional methods and a 68% cost reduction during peak periods. These results underscore the effectiveness of the strategy in optimizing operational efficiency, reducing energy costs, and maintaining thermal comfort, all while leveraging renewable energy and storage systems. By integrating the AWHP, PV, and ESS into a single package, the system enables users to efficiently produce and store renewable energy for later use, providing a comfortable and sustainable means of reducing electricity bills.