Abstract: The forecasting techniques of data stream provide such an important role in the applications of financial portfolio management and network monitoring.The forecasting techniques of data stream in recent years are analyzed and made conclusion.A new forecasting method of data stream is proposed,which constructs the input data stream to digraphs structure.When new data stream comes,the digraphs is maintained.Then the probability matrix that describes the data stream approximately can be received,and the possible value in the next time of a data stream can be forecasted by the use of Markov models.Finally,the feasibility is verified through experiment.The conclusion indicates that the ratio of the algorithm's precision and success forecasting is high.