Enhancing Indoor Temperature Forecasting through Synthetic Data in Low-Data Environments

Forecasting indoor temperatures is crucial for efficient HVAC system control. Data scarcity is a major challenge, as acquiring data representing extreme scenarios and transient regimes is costly and energy-intensive. This paper investigates the use of synthetic data augmentation, specifically leveraging state-of-the-art AI methods like GANs and VAEs, to improve forecasting models in low-data environments. The study evaluates the performance of synthetic data generators independently and assesses their utility in improving forecasting accuracy and mitigating training variance. Results demonstrate significant improvements in forecasting accuracy by integrating synthetic data, paving the way for more robust models in low-data regimes.

Zachari Thiry, Massimiliano Ruocco, Alessandro Nocente, Michail Spitieris