A Recognition Method of Reduced Evolutionary Neural Network and Its Application

In complex pattern recognition, it is difficult to evaluate by traditional method or single intelligent method. So a recognition method of reduced evolutionary neural network is presented, which includes, an algorithm for continuous attribute discretization based on attribute similarity, an algorithm for sample attribute reduction based on rough set and granularity computation, a stable speedy algorithm for neural network study-train based on particle swarm optimization, and an optimization algorithm for neural network hidden layer nodes based on golden section principle. The actual application shows the recognition method not only achieves the perfect precision in complex gas layer recognition, but also saves cost, improves processing speed, and so on. The applied effect is better than that of BP algorithm, improved BP algorithm and Levenberg-Marquardt algorithm