CNN-Based Prediction of Oxygen Supply Needs in under-Actuated HVAC Zones | Chapter 6 | Engineering Research: Perspectives on Recent Advances Vol. 11

 

Aims: This study aims to develop a predictive model for oxygen consumption in under-actuated HVAC zones by integrating occupant-centric parameters such as metabolic rate and activity behaviour with environmental and mechanical variables.

 

Study Design: This is an experimental and computational modelling study utilising a dataset of Variable Air Volume (VAV) damper measurements, fan energy estimation, and occupant metabolic rate calculations.

 

Place and Duration of Study: The dataset was collected and processed at Universitas Trilogi, Jakarta, Indonesia, during 2024–2025.

 

Methodology: A curated dataset comprising 11,696 multivariate records was used, including occupant activity counts, environmental parameters (temperature, humidity, air pressure), HVAC operational metrics (airflow, fan speed, brake horsepower, damper percentage), and physiological measures (metabolic rate, daily and minute-based kilocalorie expenditure). Oxygen consumption (L/min) was computed using metabolic energy equivalence. A Convolutional Neural Network (CNN) model was developed and benchmarked against a Multi-Layer Perceptron Artificial Neural Network (MLP-ANN) and a standard Multi-Layer Perceptron (MLP) Regressor. Model performance was assessed using RMSE, MAE, and R² metrics.

 

Results: Descriptive statistics indicated significant variability in metabolic rate and oxygen demand, ranging from near zero during sedentary activities to peaks above 0.9 L/min during high-intensity activities. The CNN model achieved the best performance (RMSE = 0.0074, MAE = 0.0038, R² = 0.9923), outperforming MLP-ANN (RMSE = 0.0314, MAE = 0.0151, R² = 0.8733) and other benchmarks. Learning curves demonstrated stable convergence without overfitting, while parity plots confirmed strong alignment between predicted and observed oxygen consumption.

 

Conclusion: The findings confirm that occupant behaviour, particularly metabolic rate, plays a critical role in predicting oxygen demand in under-actuated zones. CNN demonstrated superior predictive accuracy and generalisation by effectively capturing nonlinear and spatio-temporal dependencies. These results suggest that HVAC systems can be optimised through adaptive oxygen control strategies, enhancing both energy efficiency and indoor air quality.

 

Author(s) Details

Yaddarabullah
Department of Informatics, Universitas Trilogi, Indonesia.

 

Erneza Dewi Krishnasari
Department of Visual Communication Design, Universitas Trilogi, Indonesia.

 

Idea Alvira
Department of Tourism, Institute Seni Indonesia Padangpanjang, Indonesia.

 

Please see the book here :- https://doi.org/10.9734/bpi/erpra/v11/6530