Development of a Self-Learning and Adaptation Algorithm for Microclimate Control System Using Sensor Networks

Abstract

The paper presents an algorithm for self-learning and adaptation of an intelligent microclimate control system for industrial premises, operating based on a cyber-physical architecture with the use of a distributed sensor network. The mechanism of incremental retraining of an NNARX neural network model during operation is described, ensuring gradual refinement of the model’s predictive properties considering real production conditions. The architecture of the sensor subsystem is presented, including sensors for temperature, humidity, gas concentration, and differential pressure, integrated via industrial interfaces RS-485 and Modbus RTU/TCP protocols. The results of experimental studies are provided, confirming the effectiveness of the developed algorithm: periodic retraining of the model based on accumulated operational data ensures a reduction in the mean absolute prediction error by 12–15%, while the system maintains temperature with a deviation of no more than 0.5 °C from the setpoint under external temperature fluctuations of 5–7 °C. It is shown that adaptive tuning of the fuzzy logic controller parameters, performed in parallel with neural network retraining, provides an additional improvement in transient response quality and a reduction in stabilization time.