1 Outliers in control engineering—they exist, like it or not

Authors:
  • PDPaweł D. Domański
    Paweł D. Domański
  • YCYangQuan Chen
    YangQuan Chen
  • MŁMaciej Ławryńczuk
    Maciej Ławryńczuk

Abstract
Major achievements in control engineering, like the least squares estimation, minimum variance, LQG (Linear Quadratic Gaussian), MPC (Model Predictive Control) or adaptive control strategies rely on the assumption that process disturbances exhibit Gaussian properties. Though this applies to the original algorithms’ formulations and modifications relaxing this assumptions exist, these conditions are rarely verified. On the other hand, anyone who has worked in industry is very often confronted with signals and processes that do not meet these conditions. Analysis of real industrial signals very often discloses their non-Gaussian properties, which manifests itself in more or less frequent anomalies. The easiest method to identify them is through statistical analysis as they are responsible for the tails in distributions. Therefore, statistical analysis should be considered as the appropriate analytical tool. Fat-tail properties are often closely related to the fractional properties of the process behind this data and fractional calculus complements statistical methods, giving further insight into the generating mechanism. Outliers do not have to be artificial exceptions. Their observation and analysis may enrich our knowledge and give new, and open up further, perspectives for our research curiosity.