Summary
In this study, an improved variant of chicken swarm optimization (CSO), named I‐CSO, is proposed to find the unknown parameters of the proton exchange membrane fuel cell (PEMFC) models. Although the basic CSO has a well‐established population hierarchy mechanism that gives it an important advantage over its competitors, it suffers from premature convergence and can be easily trapped into the local optima because of inadequate use of population information in the update rule of the rooster's position. In the proposed I‐CSO, this shortcoming is addressed by introducing a new learning strategy for the roosters, which play leadership roles in the foraging behavior of the chicken swarm, to improve the algorithm convergence capability. Moreover, an adaptive inertia weight is introduced to make the algorithm more stable by striking a better balance between the exploration and exploitation phase. The sum of absolute error between the actual and estimated voltage outputs of the stack is suggested as the objective function to perform the optimization. Besides the suggested one, two other objective functions are also used to evaluate the impact of objective function choice on the optimization results. The test of the method is performed on two commercial PEMFCs, which are BCS 500‐W Stack and NedStack PS6, and the results of I‐CSO are compared with those of other competitive algorithms published in the literature. The final results show that the use of the proposed I‐CSO with the suggested objective function demonstrates excellent performance in estimating the PEMFC model parameters with fewer errors.