Abstract
Time series segmentation is an important vehicle of data mining and extensively applied in the areas of machine learning and anomaly detection. In real world tasks, dynamics widely exist in time series but have been little concerned. This paper proposes an algorithm which can partition a multivariate time series into subsequences at points where the dynamics change. In process industries, the change of dynamics often relates to operation regime change, working condition shifting or faults. Therefore, to segment time series according to change of dynamics can be useful in data preprocessing and getting deep insight of the process in various industrial process monitoring tasks. The proposed algorithm recursively merges neighborhood subsequences through a heuristic bottom-up procedure. The cost of merging is defined as the mutual predictability of the subsequence models which are constructed using the dynamic-inner principal component analysis algorithm. Then, the goal becomes finding the segmentation scheme which minimizes the averaged dynamic prediction errors of each subsequence model. The proposed algorithm is evaluated on both simulated data and the time series data collected from an industrial processing plant. The results show that it outperforms the static principal component analysis based methods.
Highlights
There has been a growing recognition of the value of time series data because many applications, such as anomaly detection [1], [2], working condition perception [3], [4] and soft measurement [5]–[7], are being enabled by the rapid development of data driven algorithms
Industrial process time series are characterized by its inherent dynamic nature
This paper investigates the problem of partitioning multivariate industrial time series data into piecewise stationary subsequences
Summary
There has been a growing recognition of the value of time series data because many applications, such as anomaly detection [1], [2], working condition perception [3], [4] and soft measurement [5]–[7], are being enabled by the rapid development of data driven algorithms. For the problem dynamic time series data segmentation, Dobos and Abonyi [3] proposed an algorithm based on the dynamic PCA (DPCA) [26] to support the detection of operation regime changes. Assuming that the time series data are white noised and autocorrelated, the key idea of this work is to represent the time series data using a dynamic latent variable (DLV) model as inner model and a PCA mapping model as outer model. B. DYNAMIC LATENT VARIABLE MODEL Let T represent a multivariate time series composed of the sensory data collected from a dynamic industrial process. The key idea of this work is to synthesize a cost function based on the prediction accuracy of the dynamic model trained using the subsequence data.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
