Evolution of approaches to intelligent microclimate control in industrial environments: A review of models for cyber-physical systems

Abstract

This article presents a systematic review of modern approaches to intelligent forecasting and control of microclimate parameters in industrial environments within the context of cyber-physical systems (CPS) development. The analysis covers classical control methods (PID), heuristic techniques (Fuzzy Logic), machine-learning-based approaches, as well as neural network architectures of varying complexity, including multilayer perceptrons (MLP), recurrent models (RNN, LSTM, GRU), and nonlinear autoregressive neural networks with exogenous inputs (NNARX). Special attention is given to the NNARX model as one of the most promising solutions for short-term forecasting of microclimate parameters in inertial industrial ventilation and thermal regulation systems. The article highlights the specific aspects of integrating forecasting models into CPS structures, where the combination of sensors, actuators, and the computational cyber layer forms an adaptive real-time control loop. A comparative analysis of the main neural network architectures is conducted based on accuracy, dynamic modeling capability, incorporation of external factors, and suitability for CPS implementation. The literature review reveals several research gaps, including an insufficient number of studies addressing complex microclimate interactions ( ), the lack of universal parametric models, limited availability of real industrial datasets, and the challenges of ensuring real-time operation of forecasting models. Promising future research directions are outlined, such as the development of microclimate digital twins, the adoption of Edge AI solutions, hybrid NNARX+FLC control strategies, and the adaptation of microclimate control systems to the principles of Industry 5.0.