Parameter identification is a key task in the research of proton exchange membrane fuel cell (PEMFC), which provides a basis for establishing an accurate and reliable PEMFC model. However, the nonlinear characteristics of the PEMFC model and the inevitable insufficient measurement data often make traditional optimization techniques difficult to solve. In particular, inadequate measurement data can introduce bias or lead to data loss. To solve the above problems, a new hybrid optimization strategy is proposed. Firstly, the semi-mechanism PEMFC model is established by SimuNPS, and the feedforward neural network (FNN) is used to expand and predict the data. Then, the teaching-learning-based algorithm (TLBO) is used to identify the expanded PEMFC parameters. Finally, the effectiveness of the FNN-TLBO strategy is verified by comparing the two identification methods. For example, under high temperature and low pressure, the identification of FNN expansion processing shows that TLBO is 99.21% and 85.93% higher than particle swarm optimization algorithm and Harris hawk algorithm respectively. The parameter identification of PEMFC model based on FNN-TLBO shows that FNN-TLBO has higher robustness and optimization quality in the simulation results by comparing the original data and the preprocessed data.