Thermoelectric generation (TEG) systems suffer from severe power losses under heterogeneous temperature distributions (HTDs). This paper proposes a battery-based dynamic voltage compensation scheme optimized by an adaptive hierarchical actor–critic (AHAC) reinforcement learning algorithm. Unlike conventional methods, the AHAC controller is rigorously mapped to the physical TEG model, where the state vector explicitly incorporates temperature-dependent Seebeck coefficients, internal resistances, column voltages, currents, thermal profiles, and battery states. The action corresponds directly to battery voltage injection, and the reward function is strictly derived from the net power maximization objective defined by the system’s power balance equations. By complying with thermoelectric material characteristics and thermal–electrical coupling dynamics, the proposed method ensures physical interpretability and reproducibility. The simulation and hardware-in-the-loop (HIL) results confirm the real-time feasibility of online inference and control execution, with power enhancement rates from 3.14% to 13.91% (9 × 9) and 0.44% to 13.23% (9 × 6), outperforming Dyna-Q, GA, and PG methods. The revised framework guarantees methodological coherence between the control algorithm and the TEG’s physical and optimization models.
Huang et al. (Mon,) studied this question.