On the basis of matrix analytical theory, the identifiability of inertial moment parameters was studied. A linear regression model in term of the unknown parameters was derived from attitude dynamics equations; the condition number of regression matrix was defined as the quantitative index of identifiability degree; taking the maximum of identifiability degree as objective function, an optimal control model was built. The pseudospectral optimal control was employed to solve the command trajectory of control moment gyro and to improve the identification precision and convergence velocity. The method of dual-unscented Kalman filter identified parameters under simulated circumstances. Results demonstrate that the optimal excitation design can improve the performance of parameter's identification and has high robustness to inaccurate priori information.