The walking office, harvests and threshing system of combine was analyzed, and the fundamental mathematical model of system was proposed. Since this model was high order, nonlinear system, without the accurate system parameter. The difference equation of model was decomposed into two parts. The first part had a bearing on the control signal, another part did not. By using the RBF neural network technology, the two parts was separately identified online. Then, the controller functionated according to the identified result. Under the dissimilar condition, simulation and experiment was carried out which proved that using the RBF network can fit the controlled plant effectively. When the controlled plant parameter changes, the system can track it, achieving the good control effect.