Revolutionizing parameter identification of PEM fuel cell using adaptive differential evolution algorithm based on deeply-informed mutation strategy and restart mechanism optimization

Proper modeling of proton exchange membrane fuel cells (PEMFCs) depends on the accurate determination of the unknown parameters that are essential in predicting performance, but usually not given by manufacturers. The Adaptive Differential Evolution Algorithm Based on Deeply-Informed Mutation Strategy and Restart Mechanism (ADEMDR) that is proposed herein proposes a new method of optimization of the seven parameters (ξ1, ξ2, ξ3, ξ4, λ, R_c, and β) that control the activation, ohmic, and concentration overpotentials. In contrast to the traditional approaches, ADEMDR incorporates a highly informed mutation strategy that dynamically modifies evolutionary direction based on elite individuals, suboptimal solutions, and search history. Also, there is a restart mechanism that avoids premature convergence by reintroducing good solutions, which greatly improves the balance between exploration and exploitation. The results of validation of ADEMDR on 12 commercial PEMFC stacks under different operating conditions prove its superiority to the state-of-the-art algorithms, such as L-SHADE, PSO, DE, PLO, HARD-DE, and NDE. The algorithm has the minimum sum of squared error (SSE), absolute error (AE), and mean bias error (MBE), and the improvement of computational efficiency is more than 98% in terms of run time. The optimized parameters allow very precise prediction of I-V and P–V curves, which are very close to experimental data. The strength and scalability of ADEMDR render it applicable in real-time PEMFC modeling, and it has a great potential of being implemented in embedded systems in monitoring and control systems.