The use of artificial intelligence and artificial neural network analysis as new tools in evaluating soil properties can play an important role in engineering work, especially in predicting seismic geotechnical hazards. Liquefaction is one of the important issues that can occur in unsaturated saturated granular soil layers during an earthquake. Given that the occurrence of this phenomenon can cause serious damage to surface and subsurface structures, it is very important to evaluate its potential occurrence in soil layers. Various laboratory, field, and numerical methods are available to estimate the risk of liquefaction in soil layers. In this research, the risk of liquefaction along Line 2 of Tabriz Metro has been studied using neural network analysis. For this purpose, first, two experimental methods (based on the results of standard penetration resistance tests and shear wave velocity measurements) and then, the neural network analysis method has been used to evaluate the liquefaction potential of soil layers. In order to assess the liquefaction potential of soil layers, geotechnical data from 54 boreholes drilled along Line 2 of Tabriz Metro has been collected and used. To analyze boreholes and determine the potential for liquefaction risk in soil layers, the maximum ground surface acceleration must be determined. For this purpose, a maximum surface acceleration zoning map based on a finite fault stochastic model has been prepared and used. In the first step, using the simplified method of Idriss and Boulanger (Idriss and Boulanger, 2010), the liquefaction potential in soil layers was determined using the results of the Standard Penetration Test (SPT). Then, the liquefaction potential of soils is evaluated using the shear wave velocity (Vs), which is a new method and instead of the standard penetration resistance test (SPT) number, the shear wave velocity value at the studied depth is used. For this purpose, the method of Andrus et al. (2004) has been used. In estimating the liquefaction potential using a neural network, the fitnet neural network method is used, which uses a sigmoid function in the middle layer and a linear function in the output layer. In this study, five catalogs (models) are considered as input data for the neural network. Matlab software (Matlab-ver:7.11) is used to analyze the data of the neural network. From a total of 249 data series obtained, 149 series are selected for training, evaluation and checking the neural network and 100 series are selected for prediction of the neural network (89 series for training, 30 series for evaluation, 30 series for checking and 100 series for prediction). The number of hidden layers of the neural network is considered to be 10. The results of the data analysis can be seen below.
The results obtained show that the shear wave velocity method is somewhat consistent with the neural network analysis method, which takes into account factors such as the disproportion of the maximum shear wave velocity for the occurrence of liquefaction in the study area and the disproportion of the cementation coefficient of the soils. On the other hand, the results obtained show that the neural network can be used as a powerful and efficient tool in assessing liquefaction resistance, and the results obtained from the neural network analysis are somewhat consistent with the studies conducted by (Sahebkaram et al., 2023). In this study, due to the lack of access to real data on the occurrence of liquefaction, data obtained from the shear wave velocity method was used to train the neural network, while to achieve more realistic results, it is better to train the neural network based on real data on the occurrence of liquefaction.