In power system stability computation， accurate load models can ensure the reliability of computed results. Improving the aggregation precision of multi induction motors is one of the most important contents in load modeling with the component-based approach. In this paper， an aggregation method is presented， in which the multi induction motors are classified into different types using the self-organized neural network and each type of motor is aggregated with a new steady state model equivalence method. Finally， simulations are made on the typical residential and commercial induction motors data in IEEE 9 grid with the PSD-BPA program developed by CEPRI. The results have proved the validity of the method proposed.