Evaluation of the maximum horizontal displacement around the power station caverns using artificial neural network

Abstract

Prediction of the maximum displacement on the sidewalls of the powerhouse is a crucial task in the caverns excavation that depends on the site characteristics including geological and geomechanical parameters. Limitations of the available methods have caused utilizing of new predictive methods. In this research, maximum horizontal displacement around the caverns has been investigated using artificial neural network (ANN), numerical and empirical models for different conditions. The effective parameters including RMR (rock mass rating), overburden depth, coefficient of lateral pressure, pillar width and vertical difference of crown level between two adjacent caverns (powerhouse and transformer) are considered as the input parameters to predict the maximum horizontal displacement. Accordingly, the numerical modeling was utilized to introduce a sufficient database to construct the ANN model. The obtained results from the ANN model were compared with the results of the available numerical and empirical models based on the measured data gathered from different case studies in Iran and other countries. To compare the performance of utilized models, determination coefficient (R2), variant account for (VAF), mean absolute error (Ea) and mean relative error (Er) indices between predicted and measured values were calculated. Comparison results showed that based on the geomechanical parameters, the constructed optimum neural network can reliably predict the maximum displacement around the caverns. Finally, the sensitivity analysis of ANN model results shows that overburden depth is recognized as the most effective parameter, whereas tensile strength is the least effective parameter on the maximum displacement around the power station caverns in this study.