Abstract: To deal with the nonlinear and time-variant characteristic of the electrode regulator system in arc furnace, an IMC controller is designed. The controller composes by two RBF neural networks, which are used to identify the controlled object and its inverse, from this to eliminate steady-state error and let output track input. Center vectors and the shape parameters of the networks are adjusted online, which speeds up the convergence rate and improves anti-jamming capability. Simulation results verify the effectiveness of the method.