Regularization of the problem of monitoring the states of group objects of fight test

Abstract

The problem of monitoring the technical condition of prototypes of multi-agent systems is considered. The classical control procedure is analyzed in relation to a new type of group test objects - multi-agent systems. It has been established that the fulfillment of the necessary control condition - the observability of the states of the test object in relation to a multi-agent system is not obvious, since it is required to ensure a sufficiently high probability of correct detection of each element of the object. Otherwise, the control result becomes unreliable, since the elements of the vector of measured parameters are mixed. It is shown that for a group object, the states of its elements act as such parameters. In order to ensure the required observability and, consequently, the correctness of the control problem, it is proposed to use additional highly informative features as regularizes. The search for these signs is carried out in three directions: analysis of the hyperspectral image of elements with the search for unique forms of the spectrum corresponding to materials characteristic of a particular element; analysis of the location of these materials and the possibility of various combinations of these two features (the shape of the spectrum and the location of materials); infrared portrait of elements in the mid-IR range, in which characteristic bright areas can be distinguished, corresponding to the location of functional equipment. It is noted that the use of these features in terms of processing data from information-measuring systems requires some preparation and, preferably, automation. For automation, it is proposed to use single-pass neural network detectors. The results obtained will be useful in developing a system for collecting and analyzing information for testing prototypes of multi-agent systems.