On-line update of situation assessment: A generic approach

Situation is an abstraction that characterizes a complex system consisting of semi-autonomous objects striving to achieve certain particular goals (sequences of goals) and operating in a coordinated mode controlled by a meta-goal of the system on the whole. Situation and objects are discriminated by their "states" taking values from finite sets of classes' labels. Situation assessment, which is the topmost task in many practically important applications, is a classification task aimed at assessing the situation current state and assigning a class label to the situation. The paper covers certain key issues of the situation assessment problem. It analyses formal frameworks proposed for specification of the situation models, and highlights a number of challenging problems associated with situation assessment, the most crucial of which is caused by the peculiarities of input data used for situation assessment, which are a superposition of asynchronous discrete streams of heterogeneous data received from multiple sources and having finite "life time". The paper is focused on the approaches and algorithms intended for on-line updating of the situation assessment, on the situation assessment systems generic architecture and on the demonstration of the proposed techniques and architecture using a computer assurance system solving anomaly detection task.