Structural optimization of the frame was usually implemented after the dimensions of mobile parallel robot have been determined. For the dimensional synthesis of the robot aimed at large load capacity and light weight, separation of structural optimization and dimensional synthesis might lead to unnecessary iterations. To solve this problem, a light weight design approach is proposed for the structural optimization by the surrogate model. Position of the connecting point of the parallel limbs are regarded as input parameters. Sample data were then obtained by the LHS method with additional rules for model fitting. Sensitivity analysis is introduced to screen the core parameters to achieve dimensionality reduction. Sample compensation is carried out to compensate for the regions with large prediction errors in the initial model, so as to improve the model accuracy through secondary reconstruction. The study shows that the constructed surrogate model can effectively balance the computational efficiency and accuracy. It solves the problem of multiple input parameters and time-consuming simulation in structural topology optimization. Meanwhile, this design strategy improves the ability of regulating the load distribution of the vehicle, providing a new solution for the lightweight design of parallel mobile robot.