基于Wasserstein生成对抗网络过采样的城市洪涝风险智能预测模型OA

Urban flooding is a critical issue affecting social public safety,and risk prediction serves as an important measure for flood disaster prevention and control.Aiming at the imbalance of inundation depth data in intelligent prediction of urban flood risk,this paper proposes a minority class oversampling method based on Wasserstein Generative Adversarial Network(W-GAN)to generate a balanced deep learning dataset with high,medium,and low-risk samples.Combined with Convolutional Neural Network(CNN),an intelligent prediction model for urban flood risk is established.Taking Haidian Island in Haikou City as the study area,the results show that the number of extremely very low risk samples in the study area is approximately 10 times that of very high risk samples,presenting a significant data imbalance characteristic.The samples generated by the W-GAN algorithm are highly similar to the original samples in features.Compared with the original imbalanced dataset,the optimized balanced dataset effectively improves the prediction accuracy of every risk level,among which the prediction performance of the very high risk level is the most significantly enhanced,with its F1-score increased from 0.712 0 to 0.944 0.This study can provide theoretical support and technical reference for the accurate identification and scientific prevention and control of urban flood risk.