Aiming at the problem of inverse kinematics in a single neural network, such as the low resolution and poor generalization ability, the working space and inverse kinematics algorithm of industrial robots are studied. Based on the analysis of Back Propagation (BP) and Radial Base Function (RBF) neural network, a 7-input 6-output neural network model with parallel BP network and RBF network was proposed. A cooperative industrial robot arm was taken as an example. Firstly, its kinematics model was established and the workspace was analyzed, then the positive kinematics was solved to obtain the data set, which was used to train, verify and test the network, and finally the network model was gotten that meets the requirements. The simulation results verify the correctness of the network, which shows that the parallel network method improves the solution accuracy of the single neural network, and the solution speed is faster than the inverse kinematics speed of the analytical method, which proves the practicality of the method.