基于数据增强技术的大坝变形GRU预测模型OA

The accuracy of dam deformation prediction models depends on high-quality data inputs;however,many projects face the issue of sparse or missing monitoring data.Water pressure,temperature,and time effects were selected as deformation influence factors to construct a dam deformation prediction model using a gated recurrent unit(GRU).To address the limited samples of measured deformation data,four generative data augmentation algorithms,i.e.,SMOTE,GAN,GMM,and Diffusion models,were employed to expand sample heterogeneity and broaden the model's learning scope.Results demonstrate that datasets enhanced by GAN,GMM,and Diffusion models improved the GRU model's accuracy,increasing the goodness of fitting from 0.925 to 0.940,0.932,and 0.959 respectively.Based on comprehensive analysis of evaluation metrics,the GRU deformation prediction model enhanced by the Diffusion model demonstrates the most outstanding performance and generalization ability,making it the preferred method for dam deformation data enhancement processing.SHapley Additive exPlanations(SHAP)analysis reveals that temperature is the most significant factor affecting dam deformation,with upstream water level and time effects also showing notable influence,while downstream water level has minimal impact.