Optimizing indoor environmental prediction in smart buildings: A comparative analysis of deep learning models

This paper presents a comprehensive investigation into the application of deep learning models for predicting indoor environmental quality in smart buildings. Using data collected from a network of microclimate sensors deployed across a university campus in Sydney, we evaluated the performance of Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and hybrid CNN-LSTM models. Our study encompassed various aspects of model development, including data preparation, architecture design, hyperparameter optimization, and model interpretability. Contrary to common assumptions in time series forecasting, our results demonstrate that CNN models consistently outperformed LSTM and hybrid models in predicting indoor temperature. We found that multivariate input configurations enhanced prediction accuracy across all model types, highlighting the importance of capturing complex interactions between environmental parameters. Through SHapley Additive exPlanations (SHAP) analysis, we identified temperature, humidity, and Heating, Ventilation, and Air Conditioning (HVAC) status as the most influential features for predictions. Our experiments also revealed optimal configurations for historical input length and prediction horizon, providing practical guidelines for model implementation. This research contributes valuable insights for the development of more efficient and accurate smart building management systems, potentially leading to improved energy efficiency and occupant comfort in built environments.