Gravity anomaly interpolation based on Genetic Algorithm improved Back-Propagation Neural Network

The principal weakness of the traditional BP Neural Network (BP NN) is that it cannot avoid local minimum, while the Genetic Algorithm (GA) has the ability of globally optimum-searching, and therefore a new approach, GA-improved BP NN method, was presented for gravity anomaly interpolation. Firstly GA was used for optimizing the initial link weights as well as the threshold of the layers of the traditional BP NN, and then the training was completed using BP method. Numerical experiments were performed for gravity anomaly interpolation based on field measurements using BP NN and GA-improved BP NN respectively. Through comparison among the results, we found that not only the convergence rate and generalization ability of GA improved BP NN are higher than those of the traditional BP NN, but also the efficiency of the GA improved BP algorithm is more satisfactory.