A decoupling control method using multi-model based on a new type of generalized inverse is proposed. The neural networks generalized inverse model is identified by partitioning the induction motor input space and the relevant controller is designed. The zeros and poles of the close-loop transfer function can be assigned arbitrarily. The switching method is carried out according to the dissimilarity of reference signal to the clusters of the training data sampled in the motor input space. The simulation results show the flux and speed of rotor can be decoupled effectively with this method,
and the whole control system has strong robustness.