OPTIMIZATION OF CONTROL ALGORITHMS FOR DISTRIBUTED TECHNICAL SYSTEMS THROUGH REINFORCEMENT LEARNING

Abstract

Optimization of control algorithms in distributed technical systems is one of the central problems of modern industrial automation and the Industry 4.0 paradigm. This article presents an adaptive control architecture developed on the basis of Reinforcement Learning (RL) for distributed industrial systems. The proposed system includes a Multi-Agent Proximal Policy Optimization (MAPPO) algorithm, a real-time sensor data processing module, and a distributed reward function. Preliminary experiments conducted in a synthetic industrial environment showed that the proposed method can increase system efficiency by 18.7% and reduce energy consumption by 22.3% compared to a baseline PID control system. The obtained results confirm the practical applicability of the reinforcement learning approach in distributed industrial systems