The Improvement Of BP Algorithm Based On GA In The Mineral Prediction

As genetic algorithms have been widely used to optimize feedforward neural networks foryears, especially in the field of predicting, this paper will do some research on the topic:making full use of the global searching trait of genetic algorithms to optimize feedforwardneural networks, for the sake of predicting.In this paper, three-layered feedforward neural network which has wide applicability beingthe object of optimizing, a real-coded genetic algorithm with synthetic control strategy ispresented, synthetically considering the effects of coded pattern, fitness function,distributing of initial population and genetic operators, which forms high-efficient algorithmin optimizing neural networks.1. Choose real-coded pattern, shortening the length of coded individual, and codeneural network's topology structure and weights simultaneously;2. Design fitness function which can accurately indicate neural networks' performance:sample's approximation precision is the main factor, at the same time neural network'stopology structure should not be neglected;3. Guarantee individuals' variety and even distribution of initial population: someindividuals are normal random numbers whose mean is zero, the others are random numbersin choosing bound;4. Design and improve genetic operators which adapt to real-coded genetic algorithm,avoiding prematurity.All algorithms above are implemented in Visual Studio.Net, and approximation of nonlinearfunction is given to test its efficiency and rapidity in optimizing neural networks.Comparing with improved BP algorithm before and after, the simulation results indicate thatthese improving algorithms can acquire simultaneously topology and weights rapidly andeffectually, and the acquired network has perfect generalizing ability.