Prediction method of TBM excavation axis deviation for small turning tunnels based on LSTM neural network

During the construction of full face tunnel boring machine (TBM), due to the influence of geological environment, personnel operation, equipment itself and other factors, it is easy to cause large deviation between TBM boring route and design axis, especially in small turning tunnels, TBM heading orientation and attitude control are more difficult. To solve this problem, a prediction method of TBM tunnelling axis deviation based on Long Short-Term Memory (LSTM) neural network is proposed. The model is established by using MATLAB software, and takes the historical tunneling parameters generated during TBM tunneling as input data to predict the tunneling axis deviation of TBM in the future. Firstly, the data is preprocessed, including binary discriminant function and other methods, and the 24 dimensional tunneling parameters are selected by Pearson analysis method as the input of prediction model. Taking the horizontal deviation as an example, the TBM tunneling axis deviation is predicted. Then, the different structures of the model are analyzed. Different LSTM layers and different number of neurons are selected to form a new model structure, and the optimal model structure and optimal input time periods of different prediction time are determined. Finally, the tunneling data of R100 m turning section is taken as the test set, and the model after training is input for model verification. The results show that the prediction performance of the model structure is optimal when LSTM layer is 2 and the number of neurons per layer is 128. When the prediction time is 1 min, the goodness of fit of the model is 0.969, the average absolute error is 1.506 mm, and the root mean square error is 2.412 mm, achieving good prediction results. Through this method, the operator can know the deviation of TBM tunnelling axis in advance, adjust TBM posture in advance, avoid large construction deviation of TBM tunnel axis and improve the accuracy and safety of tunnel construction.