جستجوی مقالات مرتبط با کلیدواژه « الگوریتم ژنتیک » در نشریات گروه « مهندسی معدن »

Surface settlement induced by tunneling is one of the most crucial problems in urban environments. Hence, accurate prediction of soil geotechnical properties is an important prerequisite in the minimization of it. In this research work, the amount of surface settlement is predicted using three-dimensional numerical simulation in the finite difference method and Artificial Neural Network (ANN). In order to determine the real geotechnical properties of soil layers around the tunnel; back-analysis is carried out using the optimization algorithm and monitoring data. Among the different optimization methods, genetic algorithm (GA) and particle swarm optimization (PSO) are selected, and their performance is compared. The results obtained show that the artificial neural network has a high ability with the amounts of R=0.99, RMSE=0.0117, and MSE= 0.000138 in predicting the surface settlement obtained from 150 simulations from randomly generated data. Comparing the results of back-analysis using the optimization algorithm, the genetic algorithm shows less error than the particle swarm algorithm in different initial populations. In all cases of analysis, the calculation time for both algorithms lasts about 5 minutes, which indicates the applicability of both algorithms in optimizing the parameters in mechanized tunneling in a short time.

The surface settlement is an essential parameter in the operation of mechanized tunneling that should be determined before excavation. The surface settlement analysis caused by mechanized tunneling is a geo-technical problem characterized by various sources of uncertainty. Unlike the deterministic methods, the reliability analysis can take into account the uncertainties for the surface settlement assessment. In this work, the reliability analysis methods (second-order reliability method (SORM), Monte Carlo simulation (MCS), and first-order reliability method (FORM)) based on the genetic algorithm (GA) are utilized to build models for the reliability analysis of the surface settlement. Specifically, for large-scale projects, the limit state function (LSF) is non-linear and hard to apply based on the reliability methods. In order to resolve this problem, the GMDH (group method of data handling) neural network can estimate LSF without the need for additional assumptions about the function form. In this work, the GMDH neural network is adapted to obtain LSF. In the GMDH neural network, the tail void grouting pressure, groundwater level from tunnel invert, depth, average penetrate rate, distance from shaft, pitching angle, average face pressure, and percent tail void grout filling are used as the input parameters. At the same time, the surface settlement is the output parameter. The field data from the Bangkok subway is used in order to illustrate the capabilities of the proposed reliability methods.

Bedrock unconfined compressive strength (UCS) is a key parameter in designing thegeosciences and building related projects comprising both the underground and surface rock structures. Determination of rock UCS using standard laboratory tests is a complicated, expensive, and time-consuming process, which requires fresh core specimens. However, preparing fresh cores is not always possible, especially during the drilling operation in cracked, fractured, and weak rocks. Therefore, some attempts have recently been made to develop the indirect methods, i.e. intelligent predictive models for rock UCS estimation, which require no core preparation and laboratory equipment. This work focuses on the application of new combinations of intelligent techniques including adoptive neuro-fuzzy inference system (ANFIS), genetic algorithm (GA), and particle swarm optimization (PSO) in order to predict rock UCS. These models were constructed based on the collected laboratory datasets upon 93 core specimens ranging from weak to very strong rock types. The proposed hybrid model results were compared with each other, and the real data and multiple regression (MR) results. These comparisons were made using coefficient of correlation, mean of square error, mean of absolute error, and variance account for indices. The comparison results proved that the ANFIS-GA combination had a relatively higher accuracy than the ANFIS-PSO combination, and both had a higher capability than the MR model. Furthermore, the ANFIS-GA and ANFIS-PSO model results were completely in accordance with the UCS laboratory test, and they were more accurate than the previous single/hybrid intelligent models. Lastly, a parametric study of the suggested models showed that the density and Schmidt hammer rebound had the highest influence, and porosity had the lowest influence on the output (UCS).