Simulation Study of Genetic Algorithm Optimized Neural Network Controller

The selection of neuron number in hidden layer of RBF neural network is key point in RBF network training. Unreasonable selection of center point and long training time are also usual problems in RBF network. Since the time efficiency can be lowered to improve network performance for basic genetic algorithms, the schemes have subjectivity and lower convergence rate. Then this paper proposes a new RBF neural network learning method based on improved adaptive genetic algorithm. It introduces optimal retention mechanism, adaptive intersection probability and sequence comparison to overcome the precocity, which improve the convergence rate and accuracy of network. It can also get the optimal solution of RBF neural network controller for global searching weight, to acquire an ideal controlling effect.