Abstract
This thesis deals with the post-processing lossy compression of non-equidistant univariate time series data in the industrial environment. Measurement data play an essentialrole in production facilities. They are used for monitoring, analyzing and optimizing
production processes. These time series data are typically stored only for a limited period and are also isolated within individual production facilities. In order for the data
to be used across processes and over the long term, they must be stored in long-term
archives. However, due to the continuously growing volume of data, this poses a significant challenge. Efficient data compression is therefore of great importance. The focus
is on compression methods that enable a reduction in storage space while preserving
important information. A central aspect of this thesis is the comprehensive evaluation of
existing compression algorithms and their testing with real measurement data from the
industrial environment. For this investigation, compression rate, algorithm speed, memory requirements during the compression process and the accuracy of the compressed
time series data are used as evaluation metrics. Using these metrics, the different compression results are compared based on various time series characteristics, highlighting
the advantages and disadvantages of each algorithm. To efficiently analyze these results,
they must be presented in a clear and concise form. Finally, various methods for visualizing the compression results both quantitatively and qualitatively are presented. These
visualizations aim to assist in selecting the optimal compression algorithm for different
time series characteristics. The individual tests show that each compression algorithm
used has advantages and disadvantages when applied to different structures in the time
series data. A certain knowledge of the data to be compressed is therefore an important
factor.
Date of Award | 2024 |
---|---|
Original language | German (Austria) |
Supervisor | Gabriel Kronberger (Supervisor) |