Towards an automated detection of self-organizing behavior

Abstract

In this paper we present an entropy based method to analyze complex systems. Systems are treated as black boxes, which only expose information by some specified parameters. To decide whether a system shows self-organizing behavior in terms of the specified parameters, the calculated entropy values are preprocessed using low-pass filters. The time derivative of the smoothed entropy curve is used to analyze the short-term behavior of the system. A larger filter is used for the analysis of the long-term behavior. Different evaluations show that in principle self-organizing behavior can be recognized with our approach.