A real number coded GA based wavelet neural network learning for oil well yield modeling

A real number coded genetic algorithm based wavelet neural network (WNN) learning for oil well yield modeling was proposed in this paper. Learning algorithm using stochastic gradient method usually gets local optimal solution, especially in higher dimension. We code parameter of WNN (mean and weight, dilation and translation of each wavelon) as a float array. To prevent premature convergence, we use aggregated fitness to evaluate each individual of population. A distance based fitness measure gives higher fitness to those individuals that are farther away from other individuals intended for maintaining population diversity; A MSE based fitness measure gives higher fitness to those individuals that are smaller MSE intend to achieve proximity, and gives an additional fitness to current best individual. Experimental results demonstrate our GA based WNN learning algorithm gets better solution.