| In recent years, the amount of data produced in sensor networks, system performance application field increases continuously, with the speed of millions a day, and even not has a ceiling rate, these data streams contain a large number of information which could be used as the basis for intelligent decision-making, how to mine useful information from these non-stationary continuous data streams is becoming new challenges we need to face, this problem can't be solved by traditional data mining methods.This paper focused on the research for time series data streams. It gave a new time series data stream prediction method based on sliding widow technique in stream data mining field and HHT method which is used for analyzing non-linear, non-stationary data, and proposed a common prediction module which is fit for time series data streams prediction in all application field. The continuity, large amount of data, and only once processing for one data characters are difficult problems for prediction. For the continuity of data stream, this paper adopted sliding window technique to model on the original data, made applications only concern with the data changes in sliding window in a recent time, sliding widow was realized by for loop in this paper; for the data in sliding window of any time point, used HHT method to analyze for prediction; In this paper, HHT method had been added the idea of parallel processing, when the amount of data in sliding window was too much, first divided the data into several same length data sections, then made EMD decomposition independently, enhanced the processing speed of this algorithm. Online-HHT method solved the problems of being unable to do on-line adaptive prediction and decomposing accurately of existed method(such as wavelet decomposition).Finally, Online-HHT method was used to analyze Guizhou Power Grid time series data stream, simulation experiment showed that this method could effectively do trend prediction on time series data stream. |
PDF Full Text Request