Time series data is data derived from consecutive measurements over time. It appears in all domains of applied science and engineering. Algorithms for processing such data face the additional challenge of exploiting the linear structure of the data rather than treating the time component as a mere additional data value.
Storing time series data can quickly overload any available storage. The random noise introduced by physical measurements fundamentally limits the achievable compression rate. For an effective compression of high-frequency data, we are therefore forced to apply lossy compression schemes.
We want to evaluate existing algorithms and develop new algorithms for lossy compression schemes for high-frequency time series data from different domains. The key question is what compression can be achieved without losing too much information while still being able to analyze the data and using it for modeling and statistical predictions. The scientific method will be two-fold:
- an empirical study of different methods on real-world data;
- a thorough analysis of the theoretical limitations;
Our project partner Industrial Analytics IA GmbH is a young IoT company with a scientific background in physics and mechanical engineering. They supply the team with high-dimensional industrial data sets spanning a period of several years. The data has been obtained from physical measurements of large industrial machines. For understanding the time series data, the partner company provides details of the particular industrial machines and the underlying thermodynamics. During the project, they will support the team with the required background in mechanical engineering and industrial data analytics.