Agent-Based Situational Modeling and Identification Technological Systems in Conditions of Uncertainty

Abstract

A structural-parametric situational model for identifying the state of a complex technological system under conditions of uncertainty, implemented by a trained intelligent agent, is proposed. The algorithm of neural network learning of an intelligent agent for recognizing difficult situations with finding the dividing surface between them upon presentation of a training sample of the parametric vector of the current state of the system is formalized. The methodology of developing a self-educating intelligent agent capable of identifying the current situation with incomplete and fuzzy information and making adequate decisions on their normalization in real time in the management of the technological system of the processing enterprise is described. The main stages of the software implementation of a trained intelligent agent in identifying and predicting the anomalous state of technological systems are formulated. For software implementation of a self-learning intelligent agent, a universal simulation system Simplex 3 with a specialized object-oriented language Simplex 3-MDL (Model Description Language) is proposed for describing system-dynamic, discrete-event and multi-agent models. The procedure for training an intelligent agent in the dynamics of its behavior is based on a multilayered neural network with pairs of interconnected input and output vectors and recurrent tuning of synaptic links by similarity measures.