Temporal Dataset for Indoor and In-Vehicle Thermal Comfort Estimation

Abstract

Thermal comfort estimation is essential for enhancing user experience in static indoor environments and dynamic in-vehicle scenarios. While traditional datasets focus on buildings, their application to fast-changing conditions, such as in vehicles, remains unexplored. We address this gap by introducing two temporal datasets collected from (1) a self-built climatic chamber with 31 sensor signals and user-labeled ratings from 18 participants and (2) in-vehicle studies with 20 participants in a BMW 3 Series.

Our results show that leveraging time-series data significantly improves thermal comfort prediction, with recurrent neural network models outperforming single-vector baselines. We benchmark our datasets against publicly available ones, demonstrating superior predictive performance and insights into key signal importance.

Key Contributions

• Datasets: Temporal multimodal datasets for indoor and in-vehicle thermal comfort estimation.
• Machine Learning Models: Comparative studies using recurrent architectures for state recognition and prediction.
• Signal Importance: Identification of key factors like ambient temperature, relative humidity, and skin response.
• Benchmarking: Evaluation against existing thermal comfort datasets.

Keywords

Thermal comfort, temporal datasets, machine learning, recurrent neural networks, automotive research