It is of great significance how to model and mine historical data quickly and effectively. Based on the statistical characteristics of Markov model, this study designs and implements a variable order Markov model based on suffix array and suffix automata, in view of the limitations of the model in practical data mining applications. Based on the realization of suffix tree structure, the suffix chain is introduced to realize the quick jump of each state subsequence, and the requirement of different order length probability can be dynamically and adaptively calculated. The experimental results show that compared with the traditional Markov model, the model constructs the link between suffix sequence characteristics of probability and statistics of historical data and the state in linear time and space complexity, which can greatly reduce the storage space and time, and realize online learning and application of large data.