Quantum physics-informed neural network with multi-source data fusion for prediction in PV-driven water heating systems

Real-time prediction of stratified photovoltaic (PV)-driven water heating (PVWH) storage systems remains challenging because fluctuating PV power induces strongly nonlinear storage-state evolution through buoyancy-driven transport and thermocline migration. To address this problem, a quantum-inspired physics-informed neural network (QPINN) is developed for direct multi-step prediction of the transient temperature field in a stratified storage tank over a 60 min horizon. The framework integrates CFD-resolved fields, sparse thermocouple measurements, and PV electrical and meteorological variables within a unified learning architecture. PV-thermal coupling is established by representing electrical power supplied to the heater under both constant-power and PV-driven dynamic conditions as distributed heat generation in the heater region. In addition, a shallow quantum-circuit-inspired embedding is introduced to strengthen nonlinear feature representation and stabilize long-horizon prediction. Compared with a classical PINN, QPINN reduces RMSE by 64%–76% under constant-power heating and by 81%–87% under PV-driven dynamic heating. Under an unseen highly dynamic cloudy-day forcing case, it still achieves an RMSE reduction of 59%–64% while preserving the plume-thermocline structure. These results demonstrate that QPINN enables fast and physically grounded short-horizon forecasting of PVWH storage states, providing a practical basis for online monitoring and energy-efficient operation.