Abstract: Authors take advantage of the genetic algorithm (GA) to automatically obtain structures, weights and bias of neural networks (NN). A classified coding scheme is presented to get modeling parameters of an NN. Then a practical fitness function along with a new method that can automatically adjust the number of hidden nodes and connection weights according to the individual fitness values is described in detail. The proposed method is applied to calculate the parameters of a manipulator with a freedom of degree 2. Simulation result is compared with data obtained from practical experience and the back propagation(BP) learning algorithm. Comparison study indicates that the proposed method has many advantages such as higher simulation accuracy, less resource utilization and higher computational efficiency.

Key words: Genetic algorithm, neural network, machinery example, back propagation algorithm, self-adaptive parameter adjustment