Time-varying hurst parameter estimation method based on dynamic step size overlap window
Debin Wei, Xin Yan Li, Chengsheng Pan, Jingyun Liu · 7th International Symposium on Advances in Electrical, Electronics, and Computer Engineering · 2022
Accurately estimating the self-similar characteristics of time series has great application value for data analysis, prediction and system control. Because of the high mutability of time series, the existing Hurst parameter estimation methods fail to describe the local self-similarity of time series and consider its dynamic change, resulting in low estimation accuracy. Therefore, this paper proposes a time-varying Hurst parameter estimation method based on dynamic step size and overlapping windows (DSSOW-H). The method introduces to the dynamic step size based on the existing sliding window time-varying Hurst parameter estimation method and adjusts the sudden change of data through the change of dynamic step size, in order to effectively estimate the self-similar characteristics of local data when it suddenly changes. MATLAB is used to generate artificial FGN sequences with self-similar characteristics. At the same time, this method compare with the periodic diagram method, R/S method, variance-time diagram method, sliding window time-varying Hurst parameter estimation method (SWTV-H). According to the experimental results, the relative error of DSSOW-H is reduced by 53% compared with SWTV-H, and the relative error of DSSOW-H is 0.9875% when applied to satellite network traffic. Therefore, it is proved that the method in this paper is effective in improving the estimation accuracy and reducing the estimation error when the data has abrupt changes.