Abstract:Currently, in shield tunneling construction, the precision of geological exploration is relatively low, and it is difficult to accurately obtain the stratum of the palm surface and the bad geological condition during shield tunneling with some construction risks. Relying on a municipal rapid connector tunnel project, a method for real-time identification of geological conditions during shield tunneling based on machine learning is proposed. Firstly, by performing a correlation matrix analysis on geological data, several shield tunneling parameters are selected as model inputs. According to the situation of composite strata, the model is classified and sorted as output. Time series data set and ring representative data set are formed to compare the effect of data volume on classification results. Secondly, the synthetic minority oversampling technique (SMOTE) is used to enhance the data and to solve the problem of data imbalance, which forms SMOTE enhanced datasets for both ring representative data and time series data. Finally, various machine learning models are used for classification prediction, and accuracy (ACC) and confusion matrix are used to test the model effect. The classification prediction performance variations caused by different datasets and models are compared. A larger amount of data can be obtained by acquiring the time series data for data set collation to ensure the stratum identification effect of machine learning model. The use of SMOTE to process the stratum data can realize the balance of different sample sizes and the expansion of total sample sizes. The classification effects of KNN, SVM and ANN models have been further improved.
