A two-stage method for system identification based on asymptotic theory was proposed.Firstly

an unbiased estimation and its frequency variance were obtained by the high-order ARX model.Then each sub model was reduced by the OE structure and MDL criterion.It translated multivariable model structure selection into a simple SISO problem and realized model validation through frequency variance

which resolved the difficult problem of model order selection and model validation for the multivariable system.The use of multivariable test signal reduced time for plant test and disturbance to operation.The application results were given to demonstrate the effectiveness of the identification method.