Data-driven indoor thermal comfort monitoring system based on soft sensors in the tropics

Monitoring indoor thermal parameters in buildings can improve energy efficiency while enhancing occupants’ comfort. Conventional monitoring requires sensor systems that are complicated, unreliable, and cost-intensive. This research aims to develop a real-time indoor thermal comfort monitoring system using soft sensors, considering outdoor environmental data and indoor equipment operations of a university classroom in Yogyakarta, Indonesia. We evaluated five machine-learning models against data from 192 CFD simulation scenarios. The models used eight environmental and operational input variables (including outdoor temperature, humidity, wind speed, and solar radiation) to predict three indoor thermal comfort parameters: air temperature, relative humidity, and air velocity. The Artificial Neural Network (ANN) is the best-performing model in predicting the three indoor thermal comfort parameters, with R2, NMSE, and NMAE values of 0.86, 0.006, and 0.041, respectively. We deployed this model in an actual indoor thermal comfort monitoring system and validated it. The monitoring platform reliably and accurately reports the real-time thermal environmental parameters of the classroom. This research provides an economically effective and reliable real-time indoor thermal comfort prediction framework, offering a practical alternative to conventional sensor-based monitoring systems.