Due to its simplicity and high reliability, Oscillating Water Column (OWC) wave energy conversion devices hold great promise for practical applications. The optimization of OWC parameters is currently a hot research topic. In this study, we address the challenges of rapid response prediction and parameter optimization for OWC systems by developing a physics-informed neural networks (PINN) model. The influence of different physical constraints on the performance of the PINN model is investigated. Our research demonstrates that incorporating the key parameter of wave elevation in the intermediate layer enhances the interpretability, prediction accuracy, and scalability of the model but may affect its convergence. By including the prior physical knowledge of the correlation between water volume change rate and power generation efficiency into the loss function, the convergence of the neural network model can be improved, albeit with minimal impact on prediction accuracy and scalability. Additionally, introducing the critical laws discovered in subsequent analyses regarding the relationship between wave elevation distribution and power generation efficiency into the loss function enhances the convergence of the PINN model while significantly improving its scalability. In summary, by integrating various physical constraints, including key parameters, prior knowledge, and critical laws, into the neural network model, significant improvements can be achieved in the interpretability, accuracy, convergence, and scalability of the model. The findings of this study highlight the avenues through which neural network model performance can be improved from these three perspectives, aiming to provide guidance for related modeling endeavors.