A Novel BP Algorithm Based on Self-adaptive Parameters and Performance Analysis

The standard back-propagation(BP) algorithm converges slowly and is easy to trap into local minimum, which is the main reason why it cannot be used widely in practical applications. Therefore, a new BP algorithm based on self-adaptive parameters, called the SAPBP algorithm for short, is put forward in this paper. The convergence stability is enhanced by adding the positive matrix in Hessian matrix. The SAPBP algorithm is used in approximating nonlinear function in order to test its performance. The data of simulation experiment have proved that the improved algorithm can converges very fast and avoid trapping into local minimum. The convergence stability, also, is very good. The results demonstrate the proposed scheme is efficient, and has good performance