Deviation detection in distributed control systems by means of statistical methods

Multi-agent systems have done long journey on their way to become a mature paradigm for solving dynamical tasks with distributed merit. There are many successful applications proving benefits of this methodology. However, future expansion of multi-agent technology requires development of appropriate assistive tools. One of the most important is a diagnostic framework that is capable to detect deviations from intended behavior. The proposed concept of model-based diagnostic framework is able to build stochastic model of a diagnosed system from observed events and interactions among agents within a community. Such a model is then used to evaluate likelihood of observed event sequences in runtime.