Research on Modeling and Forecasting Driven by Time Series Stream Data

Abstract

In view of the potentially infinite, fast-reaching, single-scan and noise-related characteristics of time series stream data, a method of data driven modeling and predicting is proposed in this paper. In order to enhance the real-time performance and the modeling accuracy of algorithm to the time series stream, we use a double sliding window to divide the time series stream data. One window is designed as a fixed length window to evaluate the fluctuation of the actual data, the other, as a variable length window, is used to establish a prediction model by GEP algorithm and generate prediction data. And then, the actual data and the prediction data calculated by the prediction model are fused to generate a fusion data. The colony climbing algorithm is applied to improve the population diversity of GEP to improve the prediction accuracy of modeling. The numerical simulation to the four test data sets shows that the proposed algorithm has better prediction accuracy than the Hierarchical Temporal Memory algorithm.