Prediction of Fine-grained Soils Resilient Modulus using Hybrid ANN-PSO, SVM-PSO and ANFIS-PSO Methods

Resilient modulus of subgrade soil is one of the most important parameters in terms of pavement analysis and design. This parameter is used for design of pavement structure based on both empirical (e.g. AASHTO 1993) and mechanistic-empirical methods (e.g. MEPDG). In order to determine resilient modulus, dynamic triaxial loading test should be conducted at different confining and deviator stresses on the soil samples and conducting such a test is very time consuming and costly. This paper aims to evaluate three hybrid neuro-computing methods including Artificial Neural Network-Particle Swarm Optimization (ANN-PSO), Support Vector Machine-Particle Swarm Optimization (SVM-PSO) and Adaptive neuro-Fuzzy Inference System-Particle Swarm Optimization (ANFIS-PSO) for predicting resilient modulus of fine-grained soils. Input parameters in all of these models were considered as particles passing #200 sieve, liquid limit, plastic index, moisture content, optimum moisture content, degree of saturation, unconfined compression strength, confining stress, and deviator stress and output was assumed as resilient modulus of soil. Results show that ANN-PSO method has the highest accuracy in comparison with other methods. Coefficient of determination (R2) for ANN-PSO method was determined as 0.992 in case of overall dataset and in most cases the prediction error of resilient modulus using this method was less than 20%. Coefficient of determination for SVM-PSO method and ANFIS-PSO method were determined as 0.989 and 0.951, respectively. Results of this study also showed that the input parameter of particles passing #200 sieve has maximum influence on the resilient modulus of fine grained soil materials while the deviator stress has minimum impact on the resilient modulus of this type of materials.