MICROSERVICE-BASED ADAPTIVE INDOOR MICROCLIMATE CONTROL SYSTEM USING LSTM FORECASTING

Abstract

Maintaining stable indoor microclimate conditions is a critical challenge in smart buildings, precision agriculture, and industrial environments, where traditional centralized control systems often fail to provide sufficient flexibility, scalability, and resilience under dynamically changing conditions. This study proposes a microservice-based adaptive microclimate control system that integrates real-time IoT sensing, distributed service orchestration, and machine- learning-driven forecasting to enable proactive environmental management. The research methodology combines system-level domain analysis, architectural decomposition into autonomous microservices, implementation of lightweight communication mechanisms based on REST and MQTT protocols, containerized deployment using Docker technologies, and empirical evaluation of a functional prototype. The proposed system architecture separates data acquisition, storage, analytics, decision-making, actuator control, and monitoring into independent services, ensuring loose coupling, fault isolation, and independent horizontal scaling. Such decomposition enables continuous system evolution and flexible adaptation to varying operational scenarios without service interruption. A central contribution of the work is the integration of an analytical microservice utilizing Long Short-Term Memory (LSTM) neural networks for forecasting temperature and humidity parameters. Sensor streams undergo preprocessing, normalization, and aggregation before being used for predictive modeling, allowing the control subsystem to anticipate future environmental changes and execute preventive actions rather than reactive responses. Experimental evaluation demonstrated stable system performance with end-to-end response latency below two seconds under realistic workloads. The forecasting module achieved strong predictive accuracy, with determination coefficients of approximately R² = 0.90 for temperature and R² = 0.94 for humidity, while maintaining low prediction errors (MAE of 0.21 °C and 0.56 %, respectively). These results confirm that integrating predictive analytics into a microservice architecture significantly improves environmental stability and energy efficiency. The proposed approach demonstrates the feasibility and practical effectiveness of combining microservice- based software engineering principles with intelligent data-driven control strategies. The developed solution is suitable for deployment in smart buildings, indoor farms, and industrial facilities requiring scalable, resilient, and adaptive microclimate management systems.