The Application of the BP Neural Network in the Nonlinear Optimization

In this paper, a hybrid algorithm is proposed which combines the conjugate gradient theory with BP network algorithm. The algorithm regards the overall average error of neural network as the objective function, and the seeking weights and thresholds of the neural network as the design variables. The calculation process replaces the adjustment of original weights and thresholds of BP network with the conjugate gradient theory. It makes each iteration obtain the optimal step in the search direction, thus it improves the disadvantages of slow convergence of the original BP network. Numerical simulation is applied to the adjustment of weights and thresholds and the calculation process. By using specific examples, convergence rate and effective application are verified. Finally it fully proves the feasibility and superiority of the hybrid algorithm.