For a class of nonlinear discrete dynamic system, it is expressed as the sum of low-order model and high-order nonlinear term by the characteristics of the system running near the working point. On this basis, the intelligent modeling method of the whole nonlinear system is studied by using alternating identification strategy. Recursive least square method with forgetting factor is used to identify unknown parameters of low order model, and stochastic configuration networks is used to estimate the high order nonlinear part. Therefore, an improved intelligent modeling method for alternating identification of nonlinear system is proposed. The algorithm makes full use of the feature that the stochastic configuration networks can randomly assign the input weights and deviations of the hidden nodes according to the monitoring mechanism, automatically corrects the output weights, and gradually increases the hidden nodes until the pre-set estimation accuracy is reached. The combination of stochastic configuration networks and recursive least square algorithm can effectively improve the identification accuracy of nonlinear systems. Finally, the numerical simulation experiments are carried out to illustrate the effectiveness of the proposed algorithm.