Research Of Data Stream Prediction Algorithm Based On Directed Graph Construction

With the extensive application of computer technology in different industries, a lot of business data is produced.The data exceeds traditional lasting relationship and existes by the instantaneous data. People want to analyse it by using the characteristics of data, and excavate the change regularity in data stream, then to predict it. However, such data is usually very large and updates quickly, and we cannot preserve it in limited space, which brings about great difficulties in data stream process, analysis and prediction. Therefore, the research of related data stream prediction technology and algorithm is of great theory value and realistic significance.According to the characteristics of data stream, the thesis mainly researches the data stream prediction algorithm, and designs data storage model based on directed graph, to solve the problem of large data, complex calculation and high timely demands. Through the research of data stream prediction procession and data storage requirements, the thesis divides it into three processes which include data stream preprocessing based on directed graph storage, directed graph construction and maintenance, and prediction calculation. The algorithm proposed by this thesis includes the following advantages:The first, it uses the directed graph as data structures which stores the data stream state, and designs directed gragh construction and maintenance algorithm, then maps the unlimited data to the limited data state. Thereby, the difficulty of data storage and process can be reduced, and the storage space can be saved. The second, in the data stream preprocessing, the thesis is in the base of sliding window model in data stream clustering processing and refers to the ideas of density clustering. The result of clustering provides data input for directed graph construction and maintenance, which increases the precision of describing data stream by directed graph, and decreases the noisy point. The last, according to the migration statistics of data state at different paces descripted by directed graph, the probability is got and processed by superposition. The concept of mid-value is used to implement the prediction of point value based on the superimposed markov models, then the value of data item that may coming at the next moment is got. The result is changed from interval to point value, and the accuracy of prediction is increased.The experimental results show that, through the directed graph construction and data stream prediction model can efficiently record and describe the data stream changing tendency, and the data coming at the next moment can be predicted effectively.