Abstract: For the cable-driven parallel robot, the main model uncertainties that effect the accuracy of end-effector position include geometric parameter error and non-geometric parameter error. The two kinds of errors have strong nonlinearity and are coupled with each other, so it is difficult to calibrate parameters by traditional calibration methods. To solve it, a compensation method for end-effector position error based on neural network is proposed. The above two types of parameter errors are regarded as pseudo errors equivalently, and the curve of the end-effector position error caused by pseudo errors is approached by neural network. The mapping relationship between the end-effector position error and the cable length is established, and the position error is compensated in joint space. In order to improve the fitting accuracy of the neural network, a neural network optimization method is designed by the multi-group co-evolutionary algorithm and the backpropagation algorithm. The optimization method can optimize the network weight, threshold and structure at the same time, and improve the generalization ability and fitting accuracy of the neural network. The position error compensation experiment is carried out on the 3-DOF (degree of freedom) cable-driven parallel robot. The experimental results indicate that the mean value of the position error after compensation is reduced from 6.64 mm to 1.08 mm, the mean value of the trajectory error is reduced from 7.5 mm to 1.6 mm, and the position accuracy of the end-effector is significantly improved.