Optimizing the performance of ANFIS using the genetic algorithm to estimate the deformation modulus of rock mass

Abstract

Among the rock mass properties, deformation modulus of rock mass (Em) is important for implementation and successful execution of rock engineering projects. The direct field measurements of modulus determination is costive and sometimes difficult to execute; however indirect estimation of the modulus using regression based statistical methods, artificial neural networks (ANN) and fuzzy logic (FL) systems are recently employed. Despite the extensive application of ANN and FL in rock mass properties estimation, they are also associated with some disadvantages. In order to improve FL performance, it is possible to incorporate it to ANN. Therefore, adaptive neuro-fuzzy system (ANFIS) was presented. In this system, ANN is used to learn fuzzy rules. However, some parameters of ANN which are left should be optimized. As ANN is structured within the ANFIS, finding the optimum architecture of ANFIS will be very time-consuming via a trial-and-error approach. This study focuses on the efficiency of the genetic algorithm (GA) to find the optimum ANFIS structure and its application to predict the deformation modulus of rock mass. GA is utilized to find the optimal number of membership function, the learning rates and the momentum coefficients and to select the input variables. The results are then compared with those of trial-and-error procedure. A database including 188 data sets from six dam sites in Zagros Mountains in Iran was employed using the purpose method. It has been shown that the hybrid ANFIS-GA model has higher accuracy than the trial-and-error model for estimation of Em.