No abstract

No abstract

Indoor thermal comfort immensely impacts the health and performance of occupants. Therefore, researchers and engineers have proposed numerous computational models to estimate thermal comfort (TC). Given the impetus toward energy efficiency, the current focus is on data-driven TC prediction solutions that leverage state-of-the-art machine learning (ML) algorithms. However, an occupant’s perception of indoor thermal comfort (TC) is subjective and multi-dimensional. Different aspects of TC are represented by various standard metrics/scales viz., thermal sensation (TSV), thermal comfort (TCV), and thermal preference (TPV). The current ML-based TC prediction solutions adopt the Single-task Learning approach, i.e., one prediction model per metric. Consequently, solutions often focus on only one TC metric. Moreover, when several metrics are considered, multiple ML models for a single indoor space lead to conflicting predictions, rendering real-world deployment infeasible. This work addresses these problems by leveraging Multi-task Learning for TC prediction in naturally ventilated buildings. First, a survey-and-measurement study is conducted in the composite climatic region of north India, in 14 naturally ventilated classrooms of 5 schools, involving 512 primary school students. Next, the dataset is analyzed for important environmental, physiological, and psycho-social factors that influence thermal comfort of children. Further, “DeepComfort”, a deep neural network based Multi-task Learning model is proposed. DeepComfort predicts multiple TC output metrics viz., TSV, TPV, and TCV, simultaneously through a single model. It is validated on ASHRAE-II database and the primary student dataset created in this study. It demonstrates high F1-scores, Accuracy (≈90%), and generalization capability, despite the challenges of illogical responses and data imbalance. DeepComfort is also shown to outperform 6 popular metric-specific single-task machine learning algorithms.

Recently, deep learning-based positioning systems have gained attention due to their higher performance relative to traditional methods. However, obtaining the expected performance of deep learning-based systems requires large amounts of data to train model. Obtaining this data is usually a tedious process which hinders the utilization of such deep learning approaches.In this paper, we introduce a number of techniques for addressing the data collection problem for deep learning-based cellular localization systems. The basic idea is to generate synthetic data that reflects the typical pattern of the wireless data as observed from a small collected dataset.Evaluation of the proposed data augmentation techniques using different Android phones in a cellular localization case study shows that we can enhance the performance of the localization systems in both indoor and outdoor scenarios by 157% and 50.5%, respectively. This highlights the promise of the proposed techniques for enabling deep learning-based localization systems.