A Real Prediction Method for Underground Construction Based on Machine

Abstract: To address the problems of lack in training dimensions and length of prediction step widely abounding in civil engineering prediction, a new prediction method, multi-layer perception (MLP)-Regression, is proposed, which can be used in practical construction. Using machine learning, a large number of similar construction parameters (deformation, stress, construction stage, structure form, etc.) as training data, the prediction model is obtained. The algorithm can automatically evaluate the influence of each factor on the prediction target. The final deformation value of the target will be obtained by setting the environmental parameters (structural type, measuring point position, etc.) manually. In addition, regression and dynamic correction of the prediction results are carried out according to the in-situ monitoring. In the process of introducing construction measured data, anomaly detection will be conducted to realize hazard warning. Through some previous cases, the effectiveness of this method is proved, and it has the potential to be applied in underground construction and even the whole civil engineering.