Chapter 2 - Prediction of lateral and oblique load for batter pile group using GRNN, NN, and ANFIS

Abstract

This chapter evaluates the potential of three modeling approaches, i.e., generalized regression neural network (GRNN), neural network (NN), and adaptive neuro-fuzzy inference system (ANFIS) with three membership functions viz. triangular, generalized bell-shaped, and Gaussian function to model lateral and oblique load resisting capacity of batter pile groups. A total of 64 and 147 laboratory experiments were used as dataset for modeling lateral and oblique load tests, respectively. A comparison of results suggests that NN is found to work best among all three modeling approaches with both lateral and oblique load tests. Among the three membership functions used with ANFIS, triangular membership gives better performance for lateral load test, whereas the Gaussian membership function gives better performance with oblique load test. Sensitivity analysis indicates a number of vertical pile in the pile group and batter angle were important parameters in resisting lateral load and negative batter piles group were more efficient than the positive batter pile group. Sensitivity analysis on oblique load test indicates that number of batter piles, pile length, and angle of oblique load are important parameters. Parametric analysis indicates that 25 degree batter angle is the most efficient batter angle for resisting lateral load.