Autonomous AI-driven home energy management: Optimizing co-generation efficiency and thermal comfort in hybrid solar infrastructure

Hybrid photovoltaic-thermal water heating systems present a highly efficient path toward decentralized decarbonization by simultaneously co-generating electricity and capturing waste heat. However, the commercial deployment of this infrastructure remains limited by the volatile, stochastic nature of residential hot water extraction profiles. Traditional rule-based hysteresis frameworks lack active adaptability, frequently resulting in severe consumer thermal discomfort during early morning load spikes or structural tank overheating during peak midday solar irradiance intervals. This study tries to address these critical limitations by developing a physics-consistent double deep Q-network (DDQN) reinforcement learning agent tailored for active home energy management systems (HEMS). Leveraging high-resolution, multi-variable field experimental datasets; a rigorous thermodynamic environment is constructed incorporating active mass flow rate regulation, Polyak target smoothing, and gradient clipping architectures. A dynamic 3.0 kW auxiliary electric heating element is integrated into the system to handle low-irradiance conditions. Performance validation benchmarks indicate that the safety-constrained DDQN agent balances thermal comfort against structural constraints. The autonomous controller restricts the peak storage temperatures to well below the atmospheric boiling point as it elevates the mean tank temperature. This dynamic optimization enhances the consumer satisfaction which rises from a baseline of 19.8% to 81.5%. Furthermore, the system successfully maximizes zero-carbon solar thermal utilization, harvesting 72.40 kWh of renewable heat while restricting grid-tied auxiliary consumption to 18.04 kWh. This adaptive framework mitigates intense evening peak-load strains on decentralized grids which can be beneficial in many regions of the globe. This proposition transforms passive solar collectors into intelligent elements of the built environment. Therefore, this methodology can provide scalable socio-economic protection against rising fossil-fuel tariffs while significantly lowering municipal greenhouse gas emissions across high-humidity tropical microclimates.