Intelligent multi-pollutant prediction for indoor air quality management in industrial buildings using adaptive sensor fusion

Abstract

Maintaining healthy indoor air quality (IAQ) in industrial buildings is critical for occupant safety, regulatory compliance, and operational efficiency. This study presents a practical, sensor-driven framework for real-time prediction of indoor pollutants in industrial factory settings, aimed at supporting adaptive ventilation and building management. The system integrates multi-sensor data—including CO2, PM2.5, PM10, CH2O, O3, CO, TVOC, NO2, temperature, humidity, ventilation rates, and workforce density—collected across three factories (galvanizing, textile, and moulding) in Istanbul over four months, totaling more than 100,000 data samples. To handle sensor variability, environmental fluctuations, and data imbalance, a soft computing-based aggregation method was developed using a two-stage fuzzy inference system. This framework adaptively fuses local pollutant prediction models running on edge devices, enabling building-specific prediction accuracy and robustness. The system achieved up to 93 % local accuracy for key pollutants (CO2, PM2.5, PM10) and improved global prediction reliability under varying environmental stress conditions. Compared to conventional averaging methods, the proposed fusion method yielded higher consistency (90.4 %–92.57 % accuracy) and demonstrated statistical improvements in prediction fairness and adaptability (p < 0.05). By linking multi-pollutant forecasting with operational context, the framework supports intelligent HVAC control and contributes to improving environmental quality in industrial buildings.

keywords Air pollutant prediction, Fuzzy logic, Indoor air quality, Industrial buildings, Multi-sensor fusion, Air quality, Forecasting, Fuzzy inference, Fuzzy systems, Indoor air pollution, Industrial plants, Intelligent buildings,Sensor data fusion, Soft computing, Ventilation.