Structural Adaptation of Data Collection Processes in Autonomous Distributed Systems Using Reinforcement Learning Methods

A method of structural adaptation of data collection processes has been developed based on reinforcement learning of the decision block on the choice of actions at the structural and functional level subordinated to it, which provides a more efficient distribution of measuring and computing resources, higher reliability and survivability of information collection subsystems of an autonomous distributed system compared to methods of parametric adaptation. In particular, according to the results of experimental studies, the average amount of information collected in one step using the method of structural adaptation is 23.2 % more than in the case of using the methods of parametric adaptation. At the same time, the amount of computational costs for the work of the structural adaptation method is on average 42.3 % more than for the work of parametric adaptation methods. The reliability of the work of the method of structural adaptation was studied using the efficiency preservation coefficient for different values of the failure rate of data collection processes. Using the recovery rate coefficient for various values of relative simultaneous sudden failures, the survivability of a set of data collection processes organized by the method of structural adaptation has been investigated. In terms of reliability, the structural adaptation method exceeds the parametric adaptation methods by an average of 21.1 %. The average survivability rate for the method of structural adaptation is greater than for methods of parametric adaptation by 18.4 %.