TH97873A - Forecast of subsurface subsidence from tunneling with the clay-balanced borehole by the neural network. - Google Patents

Abstract

------15/11/2019------(OCR) Page 1 of 1 Summary of Invention The method for predicting surface settlement during tunnel boring with a compressed soil boring machine (CSO) in clay soil using an artificial neural network, as described in this invention, has been developed to be suitable for tunnel construction in clay soil conditions. It can establish relationships between complex settlement factors to predict surface settlement values. ------------ ------10/01/2018------(OCR) Page 1 of 1 Summary of Invention The method for predicting surface settlement during tunnel boring with a compressed soil boring machine (CSO) in clay soil using an artificial neural network, as described in this invention, has been developed to be suitable for tunnel construction in clay soil conditions. It can establish relationships between complex settlement factors to predict surface settlement values. ------------ Prediction of surface settlement during tunnel boring with a compressed soil boring machine (CSO) The neural network-based equilibrium in clay soil, as described in this invention, has been developed to be suitable for tunnel construction in clay soil conditions. It can establish relationships between complex settlement factors to predict surface settlement values.

Claims (2)

------15/11/2019------(OCR) Page 1 of 2 pages. Claims 1. A method for predicting surface settlement from tunnel boring with a balanced soil pressure tunnel boring machine in clay soil using an artificial neural network will consist of 11 sub-units in the data layer, which include: (1) distance from the tunnel boring starting point (m) (2) Stability Number 10 (N) at the top of the tunnel (3) Stability Number (N) at the middle of the tunnel (4) Stability Number (N) at the bottom of the tunnel (5) height of the groundwater from the bottom of the tunnel (m) (6) pressure at the tunnel boring face (kPa) (7) boring speed (mm/min) (8) tunnel boring head elevation angle (degree) (9) grout injection pressure to fill voids behind the tunnel walls (bar) (10) grout filling percentage (%) (11) type of tunnel boring machine. The number and sub-units of the latent layer, as well as the calculation cycles, The optimal neural network (ANN) can be found through trial and error by dividing the selected data into two groups: 1) the training set, which comprises 25% of the total data, and 2) the testing set, which contains all the data. The ANN is then tested to find the optimal neural network (Optimum ANN) based on the smallest Mean Squared Error (RMSE) obtained from the training and testing processes. In the hidden layer calculations, the number of factors from the input layer is passed through each hidden node for complete calculation using the Sigmoid function. The results from each hidden node are then recalculated using the Sigmoid function in the output layer. This calculation constitutes one iteration. The ANN then sends the results back to each hidden node for repetition of the calculation until 3,000 iterations are completed, with varying biases in each iteration. The resulting calculations are then compared to the input values used for learning the ANN. This is expressed as the mean square error (RMSE) on page 2 of 2. Therefore, the optimal neural network architecture will consist of 11 sub-units in the input layer, 1 latent layer with 15 sub-units, and 3,000 computation cycles. 2. Method for predicting surface settlement from tunnel boring with a stress-balanced soil boring machine in clay soil using a neural network, according to claim 1, which uses 11 factors in the calculation. ------------ ------10/01/2018------(OCR) Page 1 of 2. Claim 1. The method for predicting surface settlement from tunnel boring with a stress-balanced soil boring machine in clay soil using a neural network will consist of 11 sub-units in the input layer. This consists of (1) distance from the tunnel boring starting point (T), (2) Stability Number 10 (N) at the top of the tunnel, (3) Stability Number (N) at the middle of the tunnel, (4) Stability Number (N) at the bottom of the tunnel, (5) height of groundwater from the bottom of the tunnel (T), (6) pressure at the drilling head (kPa), (7) drilling speed (T/T), (8) elevation angle of the drilling head (degree), (9) pressure of grout injection to fill gaps behind the tunnel wall (bar), (10) percentage of grout filling (%), and (11) type of drilling head. The number and sub-units of latent steps, as well as the calculation cycles, can be found using trial and error by dividing the selected data into two groups: 1) Training set, which contains 25% of the total data, and 2) Testing set, which contains all the data, to test for network compatibility. An optimized neural network (ANN) minimizes the mean square error (RMSE) through training and testing processes. In the hidden layer, each hidden node uses a number of input factors passed through the input layer for computation. The results from each hidden node are then used in a computational process in the output layer using the Sigmoid function. This process is considered one iteration. The neural network then sends the results back to each hidden node for further calculations, repeating this process 3,000 times. Each iteration has a different bias, and the final results are compared to the input values used for learning. This is expressed as the mean square error (RMSE) on page 2 of 2. Therefore, the optimal neural network architecture will consist of 11 sub-layers, 1 latent layer, 15 latent layers, and 3,000 iterations. 2. A method for predicting surface settlement from tunnel boring with a balanced soil pressure tunnel boring machine in clay soil using a neural network according to claim 1, which uses 11 factors in the calculation. 1. Forecasting soil settlement from tunnel boring using a pressure-regulating soil equilibrium tunneling system in clay soil using an artificial neural network will involve 11 subunits in the receiving layer, one latent layer with 15 subunits, and 3,000 calculation cycles. This network will be considered suitable for forecasting soil settlement caused by tunnel boring with a pressure-regulating soil equilibrium tunneling system in clay soil. 2. Forecasting of surface settlement from tunnel excavation with a pressure-regulating boring machine in clay equilibrium using an artificial neural network as per item 1. The 11 sub-units in the receiving layer will consist of: (1) distance from the tunnel excavation starting point, (2) Stability Number (N) at the tunnel top, (3) Stability Number (N) at the tunnel middle, (4) Stability Number (N) at the tunnel bottom, (5) groundwater height from the tunnel bottom, (6) boring machine face pressure, (7) excavation speed, (8) boring machine elevation angle, (9) grout injection pressure to fill voids behind the tunnel walls, (10) grout filling percentage, and (11) boring machine type.