Abstract: According to the equation of redundancy pose of parallel mechanism establishment of redundant actuation, selection problem is sensitive to the initial value of iteration and the calculation speed is slower and the use of the Newton-Raphson iterative method is proposed, which can satisfy the real-time requirements of the online calculation of Levenberg-Marquardt algorithm (L-M algorithm) improved BP neural network model and off-line training can meet the optimization BP neural network model of Genetic Algorithm based on improved high precision (GA-BP), to solve the new redundancy driven parallel mechanism of initial positive solutions of pose prediction problems. Compared with the commonly used quasi Newton algorithm (BFGS) and the quantitative conjugate gradient algorithm (SCG) based neural network model. The results show that the GA-BP model and the L-M algorithm model in error performance analysis is obviously superior to that of BFGS and SCG quasi Newton algorithm model; L-M algorithm in terms of accuracy slightly inferior to the GA-BP model, GA-BP model and the iteration time is longer, so it is suitable for the high precision posture is offline.

Key words: forward kinematics, GA-BP, hybrid strategy, Levenberg-Marquardt, Newton-Raphson, redundant actuation parallel mechanism