A Study of the Design and Parameters Optimization of BP Neural Network Using Improved GEP

For the disadvantages of BP neural network(NN), which easily traps into a local optimum and is sensitive to the initial parameters of the network, an algorithm for the optimization of the architecture, the weights and the thresholds of neural networks using an improved gene expression programming(IGEP) was presented. First, the basic principles of BP neural network (BP-NN) and GEP was introduced and both their advantages and disadvantages was analyzed, and then it is shown how to optimize BP-NN using GEP. Therefore, the dynamic adjustment methods of the fitness function, genetic operators and evolutionary generations of GEP were presented, thus effectively combining the global search ability of GEP and the local search ability of BP-NN. Then, the algorithm of the design and the parameters optimization of BP neural network using improved GEP was proposed and its design was fully discussed. Finally, the comparisons between the algorithm and the basic GEP via testing their optimal performance show that this algorithm has obvious improvement by experience.