Settlement Prediction of Shallow Foundations on Cohesionless Soil Using Hybrid PSO-ANN Approach

Abstract

Settlement estimation of shallow foundations on cohesionless soil possesses a higher level of complexity, sole reason for which can be pointed towards the uncertainties involved in factors that affect the magnitude of settlement. For a safe and perfect shallow foundation design an unerring estimation of foundation settlement is rather essential. Unlike the conventional settlement prediction techniques AI techniques have shown greater accuracy, the potential which can be exploited for the settlement prediction of shallow foundations. This study approaches the settlement prediction problem using hybrid PSO—ANN technique (Particle swarm optimization—Artificial neural network). Dataset consisting of footing dimensions, net applied pressure, depth of embedment of footing, SPT N value and depth of water table are used as input data for developing the PSO—ANN model, whereas settlement is chosen as the output data. From about more than 300 runs an optimum network of 6-13-1 was developed. The developed model obtained coefficient of correlation, R = 0.953 and mean square error, MSE = 0.119 m. For assessing effectiveness of the model developed, different performance indices such as RMSE, VAF, MAE, PI, RSR, NS etc. were chosen. All these parameters gave values corresponding to a model with good predictive capacity.