Voltage sag response characteristics reflect the time required for the physical parameters of an industrial process to cross the threshold when subjected to a voltage sag, which can be characterized by the process immunity time (PIT) curve. However, due to the lack of measured data in actual production processes, existing methods for analyzing voltage sag response characteristics suffer from issues such as insufficient consideration of influencing factors and difficulty in determining parameter values, making it impossible to accurately fit the PIT curve. Therefore, this paper proposes a method for analyzing voltage sag response characteristics based on the fusion of mechanism models and data-driven approaches. Firstly, the sensitive equipment in typical industrial processes and their con-nection relationships are analyzed to establish a mechanism model of the PIT curve and obtain mechanism data. Then, an adaptive weight allocation strategy is adopted to dynamically assign weights to the mechanism data and measured data, strengthening the learning of the measured data by the Wasserstein generative adversarial network with gradient penalty (WGAN-GP) and enabling the model to more accurately capture the characteristics of the measured data. Next, a bidirectional long-short-term memory network is used to extract temporal features and capture the time correlation between data, thereby improving the quality of the generated data. Finally, a feature-aware loss and a dynamic reconstruction loss are constructed to constrain the model training process through the deep features and dynamic characteristics of the data, thus enhancing the fitting accuracy of the PIT curve. The proposed method is applied to a simulation experiment of a high-power electrically driven centrifugal compressor system in a natural gas compression station in southwestern China. The results verify the effectiveness and accuracy of the method proposed in this paper.