特征增强驱动下基坑临近构筑物沉降预测OA

To address the issue of pseudo-variations in automated monitoring data during deep foundation pit excavation,this study proposes a data feature enhancement method integrating exponential smoothing and variational mode decomposition(VMD)to improve the deformation warning accuracy of machine learning models.This study focuses on settlement monitoring data from a deep foundation pit in Pujiang County,Chengdu,proposing a data enhancement method that integrates exponential smoothing and Variational Mode Decomposition(VMD).The impact on the predictive performance of four models was systematically evaluat-ed:Long Short-Term Memory(LSTM),Gated Recurrent Unit(GRU),Transformer and Convolutional Neu-ral Network(CNN).Results indicate that VMD effectively filters high-frequency noise,elevating the sig-nal-to-noise ratio from 22.1 dB to approximately 41.0 dB.Following data enhancement,all models exhibit substantially improved performance.Specifically,the CNN model achieves sub-millimeter accuracy,with its Mean Squared Error(MSE)reduced to 0.000 1 and Mean Absolute Error(MAE)reaching 0.001 mm.The Transformer model demonstrates the highest sensitivity to enhanced data,showing a 99.92%improvement in MSE.Exponential smoothing effectively boosts model stability,while VMD further suppresses prediction outliers,reducing the Mean Absolute Percentage Error(MAPE)to below 5%.