Abstract: Proton exchange membrane electrolyzer cell （PEMEC） has emerged as one of the most promising devices for hydrogen production. However, the two-phase flow dynamics within the PEMEC severely affect its hydrogen production efficiency. To address the challenge of unobservability and immeasurability of two-phase flow state distributions, this paper proposes a collaborative states estimation strategy based on physics-Informed neural network (PINN). The method extracts spatiotemporal dependencies of state variables, such as volume fraction and flow velocity, through a deep neural network architecture. It also incorporates initial condition loss, boundary condition loss, and partial differential equation (PDE) loss terms to decouple the complex two-phase flow dynamics. The PDE terms are based on a mixture model combined with the Schiller-Naumann slip model, and are further refined by automatic differentiation and second-order central moment difference algorithms, which improve the model’s accuracy and interpretability. Additionally, to accelerate model convergence, Bayesian optimization algorithm is adopted for fine-tuning the weights of loss terms and difference algorithms. Finally, the proposed method's effectiveness is validated based on finite element simulations. The results show that the method’s predictions for volume fraction and flow velocity achieve mean absolute errors of 1.09×10⁻² and 4.37×10⁻⁴, respectively, demonstrating improvements of 39.80% and 43.25% compared to traditional deep neural network models.