Research on an Improved BP Neural Network Based on Fast Quantized Orthogonal Genetic Algorithm

This paper mainly proposes a new improved BP neural network training algorithm based on fast quantized orthogonal genetic algorithm (FQOGA), so as to overcome the disadvantage of neural networks that their structure and parameters were decided stochastically or by someone´s experience. In the algorithm, the global property and high-speed convergence of FQOGA and the parallelism of neural network were combined. FQOGA was used to evolve and design the structure, the initial weights and thresholds and the training ratio of neural network, and then the improved training samples were used to search for the optimal solution again by the evolved neural network. Test experiments run for the verification and validation of a logic operation, and the approach is proved to be effective and feasible especially in speeding up the convergence.