A new forecast method is proposed in order to improve the estimation accuracy of the ultimate bearing capacity (UBC) of single driven piles. The performance of the adaptive neuro-fuzzy inference system (ANFIS) and multilayer perceptron (MLP) models have been boosted using gray wolf optimization (GWO). To train the models, data were used from 100 driven piles in c-u soil in different locations. The input parameters were pile area (m 2 ), pile length (m), flap number, average cohesion (kN/m 2 ) and friction angle (°), average soil specific weight (kN/m 2 ) and average pile-soil friction angle (°), and the output was UBC. The results showed that both the MLP and ANFIS methods had good abilities to predict the UBC of the piles; however, the MLP-GWO model with a topology of 7 9 10 9 10 9 1 provided better results. The performance indices of this model were a RMSE of 1.86 kN and R 2 of 0.991 for the test data. The validity of the MLP-GWO model was tested using a new experimental dataset, and the difference between actual and estimated UBC was very low (about 2%) and confirmed the high accuracy of the model. It was observed that the theoretical methods underestimated the UBC significantly, while the proposed model estimated the UBC with high accuracy. Sensitivity analysis confirmed that the pile area and length were the most effective factors for estimation of UBC.Keywords Piles Á ANFIS Á Gray wolf optimization Á Ultimate bearing capacity Á Sandy soil Á Clay soil Communicated by V. Loia.

No abstract

No abstract