Abstract Since the traditional methods for predicting surrounding rock deformation in soft rock tunnels are lacking
in suitability and generalization, a multivariate algorithm fusion model based on the improved D-S evidence theory has been proposed, and the process framework for prediction of surrounding rock deformation in soft rock tunnel has been established by using the Tianqiaoshan Tunnel Project as an engineering example. First, the fruit fly optimization algorithm (FOA) is used to optimize the random forest (RF) model and the whale optimization algorithm (WOA) is used to optimize the long short-term memory neural network (LSTM) model. The partial least squares regression(PLSR) prediction model and the improved D-S evidence theory are used to calculate the fusion weight of the 3 models and realize decision-level fusion, which generates the surrounding rock crown settlement and horizontal convergence prediction model based on the multivariate algorithm fusion. Then the data from the measuring points G, S1,S2 and S3 of the DK110+600 cross section of Tianqiaoshan Tunnel are used as training and test samples, to compare the prediction results of PLSR model, RF model, LSTM model, FOA-RF model, WOA-LSTM model and multivariate algorithm fusion model. At last, the multivariate algorithm fusion model is put to engineering application at the DK110+605 cross section. According to comparison of the models, the multivariate algorithm fusion model has the highest prediction accuracy, its relative error is within the range of [-1.5 mm, 1.5 mm], its average R2 value is 0.998 5 and average MAPE value 3.09%, and it allows accurate prediction of the surrounding rock deformation pat?
tern of soft rock tunnel. Accounting to its engineering application, the overall prediction error of the model is within the range of [-3 mm, 2 mm], its average R2 value is 0.995 6 and average MAPE value 5.65%, and it has high enough prediction accuracy to guide construction activities.
Abstract：
Since the traditional methods for predicting surrounding rock deformation in soft rock tunnels are lacking
in suitability and generalization, a multivariate algorithm fusion model based on the improved D-S evidence theory has been proposed, and the process framework for prediction of surrounding rock deformation in soft rock tunnel has been established by using the Tianqiaoshan Tunnel Project as an engineering example. First, the fruit fly optimization algorithm (FOA) is used to optimize the random forest (RF) model and the whale optimization algorithm (WOA) is used to optimize the long short-term memory neural network (LSTM) model. The partial least squares regression(PLSR) prediction model and the improved D-S evidence theory are used to calculate the fusion weight of the 3 models and realize decision-level fusion, which generates the surrounding rock crown settlement and horizontal convergence prediction model based on the multivariate algorithm fusion. Then the data from the measuring points G, S1,S2 and S3 of the DK110+600 cross section of Tianqiaoshan Tunnel are used as training and test samples, to compare the prediction results of PLSR model, RF model, LSTM model, FOA-RF model, WOA-LSTM model and multivariate algorithm fusion model. At last, the multivariate algorithm fusion model is put to engineering application at the DK110+605 cross section. According to comparison of the models, the multivariate algorithm fusion model has the highest prediction accuracy, its relative error is within the range of [-1.5 mm, 1.5 mm], its average R2 value is 0.998 5 and average MAPE value 3.09%, and it allows accurate prediction of the surrounding rock deformation pat?
tern of soft rock tunnel. Accounting to its engineering application, the overall prediction error of the model is within the range of [-3 mm, 2 mm], its average R2 value is 0.995 6 and average MAPE value 5.65%, and it has high enough prediction accuracy to guide construction activities.
HOU Shoujiang
.Study on the Prediction Model of Surrounding Rock Deformation in Soft Rock
Tunnel Based on Multivariate Algorithm Fusion and Its Application[J] MODERN TUNNELLING TECHNOLOGY, 2023,V60(6): 151-164
2023, Vol. 60 