Abstract To achieve precise prediction of vibrations induced by blasting construction, an optimized GRU deep learning model is proposed based on the t-SNE feature dimensionality reduction algorithm and the BWO algorithm.Using blasting vibration monitoring data from the shallow-buried land section of the Xiamen Haicang tunnel, six parameters—rock uniaxial compressive strength, rock mass integrity coefficient, distance from blasting source, explosive consumption, auxiliary hole spacing, and peripheral hole spacing—were selected as input variables. The key blasting vibration parameters, including blasting vibration velocity and blasting dominant frequency, were set as output variables to validate the predictive accuracy of the model. Comparative analysis was conducted with traditional machine learning models, including SVR and BPNN algorithms. Results show that the t-SNE-BWO-GRU deep learning model achieves an average R2 value of 0.976 0, an average MAPE value of 5.70%, a RMSE of 0.019 3 for blasting vibration velocity, and a RMSE of 2.214 0 for blasting dominant frequency, demonstrating high accuracy in predicting blasting vibration parameters.
Abstract：
To achieve precise prediction of vibrations induced by blasting construction, an optimized GRU deep learning model is proposed based on the t-SNE feature dimensionality reduction algorithm and the BWO algorithm.Using blasting vibration monitoring data from the shallow-buried land section of the Xiamen Haicang tunnel, six parameters—rock uniaxial compressive strength, rock mass integrity coefficient, distance from blasting source, explosive consumption, auxiliary hole spacing, and peripheral hole spacing—were selected as input variables. The key blasting vibration parameters, including blasting vibration velocity and blasting dominant frequency, were set as output variables to validate the predictive accuracy of the model. Comparative analysis was conducted with traditional machine learning models, including SVR and BPNN algorithms. Results show that the t-SNE-BWO-GRU deep learning model achieves an average R2 value of 0.976 0, an average MAPE value of 5.70%, a RMSE of 0.019 3 for blasting vibration velocity, and a RMSE of 2.214 0 for blasting dominant frequency, demonstrating high accuracy in predicting blasting vibration parameters.
GAO Fuzhong
.Prediction of Blasting Vibration Parameters in Urban Tunnels Based on Feature Dimensionality Reduction and Deep Learning[J] MODERN TUNNELLING TECHNOLOGY, 2024,V61(6): 100-110
2024, Vol. 61 