Machine learning-driven performance optimization of perovskite solar cell-thermoelectric generator hybrid systems

Prevailing single-variable optimization underutilizes the performance of perovskite solar cell/thermoelectric generator hybrid system by neglecting strong inherent opto-electro-thermal couplings. Herein, we overcome this for the first time by introducing a machine learning-driven global optimization to the developed numerical model, which is rigorously validated. Six optimizable variables are systematically analyzed and identified: light side temperature, backlight side temperature, absorption layer thickness, semiconductor leg length, cross-section area ratio between different semiconductor leg, and thermoelectric element density. Meanwhile, their impact mechanisms are investigated. Under typical temperature conditions, global optimization of the rest variables achieves an energy efficiency limit of 20.009%, improving by 4.284% over non-optimized system and 6.817% over standalone perovskite solar cell. Correspondingly, the optimal variables are precisely quantified: absorption layer thickness of 367.3 nm, semiconductor leg length of 3.400 × 10−3 m, cross-section area ratio of semiconductor leg of 1.290, and thermoelectric element density of 800 m−2. Furthermore, energy efficiency limits under diverse operational temperatures are quantified, extending their practical reference. Current work underscores the necessity of synergistic contributions of multiple variables and thoroughly exerts it in achieving energy efficiency potential, offering low-cost, rapid, and flexible optimization design strategies for practical systems facing new material systems or operational conditions.