Pareto optimal design of photovoltaic thermal systems with grooved helical channels using GMDH-type neural networks and multi-objective grey wolf optimizer

Photovoltaic thermal (PVT) systems integrated with phase change materials (PCM) and nanofluids hold great promise for enhancing solar energy utilization. However, their practical design faces two major challenges: (i) capturing the complex nonlinear interactions among operating and design parameters, and (ii) balancing conflicting performance objectives such as electrical efficiency, thermal efficiency, and pressure drop. Existing studies often address these issues in isolation, either focusing on nanofluid-based thermal enhancement or on optimization using simplified models, leaving a gap for a comprehensive and intelligent design strategy. To address this problem, the present study proposes a novel hybrid AI-driven optimization framework that combines GMDH-type neural networks for highly accurate performance prediction with a multi-objective grey wolf optimizer to generate Pareto-optimal trade-offs. The framework is further coupled with the TOPSIS method to support decision-making across different performance priorities, offering a systematic and application-specific solution for PCM-enhanced PVT systems with plain, parallel, and grooved helical microchannel heat sinks. The GMDH models achieved excellent predictive accuracy (R 2 > 0.99), validating their robustness across plain, parallel, and staggered configurations. Optimization revealed that all Pareto-optimal solutions converged to a nanoparticle volume fraction near 2%, highlighting its decisive role in thermal enhancement. Mass flow rate distributions varied by design: in the plain unit, over 60% of solutions operated above 60 kg/hr, while the staggered unit favored medium ranges (45–60 kg/hr, 28%), indicating balanced operation. Thermal efficiency rose monotonically with flow rate, reaching up to ∼54%, whereas electrical efficiency followed a V-shaped trend with a maximum of 12.4%. The staggered configuration produced more diverse Pareto fronts, offering higher flexibility, while the parallel unit consistently minimized pressure drop (as low as ∼893 Pa). Multi-criteria decision-making confirmed that the staggered design is optimal for maximizing thermal output, while the parallel system excels when low hydraulic losses are prioritized. These findings demonstrate how nanoparticle concentration and flow strategies critically shape performance, and confirm the framework’s value as a decision-support tool for designing application-specific, high-efficiency PVT systems.