Study on a New Rapid Learning Algorithm of Feedforward-ANN with High Precision-Hybrid Genetic Algorithm

摘要

In order to overcome the serious drawbacks in learning of Feedforward-ANN(FANN), it presents a new learning algorithm-Hybrid Genetic Algorithm. First use genetic algorithm (GA) to train the initial weight vector of FANN with a rough error precision. Second, use conjugate gradient algorithm(CGA) to improve the trianing precision by its property of quadratic termination. In simulations and the identification to a nonlinear system of a weather satellite, the HGA shows ideal performance in both precision and speed.