Proton exchange membrane fuel cell (PEMFC) is regarded as the future energy capable of replacing existing fossil fuels. Performance degradation prediction is an effective method to improve the durability of PEMFC. However, traditional performance degradation prediction focuses solely on the accuracy of estimation results without focusing on how to achieve better estimation results with fewer resources. In this paper, we proposed an echo state network (ESN) to estimate the performance degradation. To improve the prediction accuracy and decrease the computing complexity, NSGA-II is used to optimize the hyperparameters of ESN. The simulation results show the RMSE is less than 0.0035. The NSGA-II-ESN can obtain Pareto solutions between accuracy and computing complexity, then provide corresponding optimal ESN models for devices with different performances.