Abstract: An exact inverse model controller is constructed,which in accordance with the characteristics of the electrode control system in arc furnace.The arc furnace model can be identified on-line.The nonlinear multiple-input-multiple-output (MIMO) plant is converted into isolated dynamic decoupling pseudo linear system based on radial basis function (RBF) neural network that applies nearest neighbor clustering algorithm.A dual-mode control strategy is presented,which is based on the proposed RBF neural network inverse controller and a proportion differential (PD) controller.Result shows that the controller has rapid compliant ability of process changing and good robust performance in practical applications.