The control of dish-like underwater robot motion is complex.It involves many influencing factors and it’s also a nonlinear control process.The model of attitude motion control is very important for the accurate control and self-adaptive predictive control.SVM algorithm is proposed to apply in indentifing the model of dish-like underwater robot and a set of MIMO identifier based on SVM is designed because the SVM has excellent performance on nonlinear system by strict mathematical derivation.This identifier can predict the attitude change according to control parameters.The identification and prediction performance of the SVM identifier is proved by the experimental data of pool test.The result shows that the MSE is less than 0.004.It verifies that this algorithm has good effect on the identification and prediction of dish-like underwater robot attitude motion control system.