For time series prediction, using autoregressive moving average (ARIMA model, this paper presents a new) based on dynamic fuzzy neural network (DFNN) model, this model is considered the main factors is output error of ARIMA model, using DFNN model with nonlinear characteristics. Producing the result is more accurate than single DFNN and ARIMA model. Therefore, it can be used as an appropriate alternative model to predict the task, especially when the need of higher precision. Last time sequence Mackey - whipping validated the effectiveness of the model.