This study addresses the inverse problem of identifying adiabatic temperature rise (ATR) characteristics for mass concrete using the Physics-Informed Neural Network (PINN). The characteristics are defined by parameters representing the maximum ATR and temperature increasing rate. The PINN-based identification of these parameters was conducted using virtual experimental data generated through numerical simulation with three different ATR models. To assess the robustness of the PINN in the identification process, noise was introduced into the data. The observation period and noise condition of the data were used as variables to evaluate the performance of PINN-based parameter identification. In addition, 10 independent PINN training sessions were conducted, and the results were statistically analyzed. The identification performance of the unknown parameters was influenced by the observation period. The PINN accurately identified the parameters used in the virtual experiments, even with short-term observation data, regardless of the noise. Statistical analysis indicates that the PINN demonstrates significant reliability and consistency in parameter identification.