Situation assessment in autonomous systems

Abstract

For autonomous systems, it is essential to be able to identify the current state i.e to perform situation assessment, in order to determine the action to execute to ensure correct functionalities. Specifically, an autonomous system requires a supervision component with the goal to detect and to diagnose the current situation, and then to determine the self-adaptive operations. We propose a self-adaptive architecture for autonomous communicating systems in which the situation assessment process feeds the reconfiguration system with an estimation of the traffic situation so that it can decide about the reactions appropriate to cope with the dynamic changes. The estimation of the traffic situation is implemented at the transport protocol level from the time-stamped quality of service (QoS) parameters by using a learning approach. A fuzzy clustering method is used to classify the system states in classes called primitive patterns and the transitions between classes are expressed in term of events. A set of simulated network traffic scenarios are used to illustrate the main principles of the approach.