Research on GAN based sample expansion method for dam earthquake damage assessment

Rapid and accurate assessment of dam damage after an earthquake is crucial for post-disaster emergency response and rapid reconstruction. Aiming at the lack of stability and accuracy of the assessment model due to the scarcity of dam-damage samples, this study proposes a Generative Adversarial Network (GAN)-based seismic data enhancement method, which learns the features of historical dam-damage samples and expands the data, and generates a high-quality enhancement dataset that conforms to the real distribution. To verify the reliability of the data, four types of machine learning models, namely decision trees, random forests, gradient boosting trees, and BP neural networks, were used for comparison experiments. The results show that the augmented dataset significantly improves the generalization ability of the models: the mean absolute error (MAE) of the predictions of each model was reduced by 70.3%, the root-mean-square error (RMSE) was reduced by 64.8% after augmentation, and the goodness-of-fit (R²) of the tree models was improved by more than 50%. Therefore, the method proposed in this study can effectively overcome the limitations of sample scarcity, which is of great significance for improving the accuracy of dam damage assessment, disaster prevention, and mitigation.