Particle swarm optimization trained neural network for aquifer parameter estimation

Numerical simulation in aquifers require knowledge of parameters that govern flow through aquifers, however, at times, these parameters are not available. Estimation of such parameters has thus gained importance in the recent years and researchers have suggested various ways by which these parameters can be obtained. Amongst the various methods used for parameter estimation, Artificial Neural Networks (ANN) has shown promise in determining parameters for non leaky confined aquifers. Usually some gradient algorithms including the Back Propagation (BP) technique are used for training a network, however these procedures exhibit slow convergence. Besides this, the solution gets easily entrapped in a local minima. The ANN proposed in this study employs a Particle Swarm Optimization (PSO) technique to train the perceptrons to predict the storage coefficient and transmissivity of aquifers. PSO technique could be an effective alternate training algorithm for ANN's since it is found to much accurate when compared to the existing conventional algorithms. Besides this, since PSO is a heuristic optimization technique, a global optimal solution can be obtained. Further, a sensitivity analysis is later carried out in the study to evaluate the most suitable ANN characteristics which includes the learning rate, the momentum factor, and the number of neurons in the input, hidden and output layers. Also, the impact of maximum velocity and acceleration constants of PSO on ANN convergence is studied so as to obtain the best possible value of parameters to minimize error. Further the proposed Particle Swarm Optimization trained Neural Network is employed in aquifer Parameter Estimation for the specific cases and the results are compared with the other existing gradient algorithms.