Optimizing an ann model with genetic algorithm (ga) predicting load-settlement behaviours of eco-friendly raft-pile foundation (erp) system

• Main Authors: Liu, L. Q., Moayedi, H., Rashid, A. S. A., Rahman, S. S. A., Nguyen, H.
• Format: Article
• Published: Springer 2020
• Subjects: TA Engineering (General). Civil engineering (General)
• Online Access: http://eprints.utm.my/id/eprint/87764/; http://www.dx.doi.org/10.1007/s00366-019-00767-4

Eco-friendly raft-pile foundation (ERP) system is one of the most recent developed types of pile foundations that the original materials can be provided from local Bakau. A precise prediction of its behaviour is of interest for many engineers. This paper presents three intelligent systems, namely, adaptive neuro-fuzzy inference system (ANFIS), conventional artificial neural network (ANN), and optimized ANN model with genetic algorithm (GA) for prediction of vertical settlement in ERP system. In this regard, a database compiled from 43 load-settlement results obtained from full-scale maintained load test (PLT). Note that, these floating raft-pile system piles were subjected to vertical axial loading. The ERP system was installed at the marine soft clay soil site at Rantau Panjang Kapar, Selangor, Malaysia. The values of subgrade modulus (K-s), Young's modulus (E-s), soil properties beneath the footing, and applied load were set as model input to predict vertical settlement (s). To evaluate the reliability of the network output, several well-known statistical indexes were used. The results show that the new proposed GA-ANN model could provide a better performance in estimating the maximum settlement of ERP system. In terms of statistical indexes (R-2, and RMSE), the values of (0.998, 0.0259, and 99.99) and (0.997, 0.0324, and 99.998) were obtained for both data sets of training and testing, respectively. Besides, comparing the training and testing data sets, R-2 values of (0.994, 0.9884, 0.995, and 0.9984) and (0.996, 0.985, 0.994, and 0.9973) were found for ANN-LMBP, ANFIS, GA, and GA-ANN models, respectively, which proves the superiority of the proposed GA-ANN model comparing to other methods.