AI model predicts thermal comfort in regions lacking climate data
Researchers have developed a machine learning approach that forecasts indoor thermal comfort using limited local data, by borrowing patterns from global datasets. The technique could help cities and building operators prepare for heat waves in developing regions where monitoring infrastructure is sparse—a growing concern as climate change intensifies extreme weather events.
Originaltitel: Transfer the thermal comfort prediction to data-scarce regions: a model for future climate
<p>Climate change is expected to increase the frequency and intensity of extreme weather events, such as heat waves, which pose significant challenges to human thermal comfort and public health. Recently, data-driven thermal comfort models have shown superior performance compared with traditional knowledge-based methods such as the Predicted Mean Vote (PMV) model, highlighting their potential in large-scale application in many local areas for future climate. However, in many regions, thermal comfort data are scarce, making it difficult to develop local thermal comfort prediction models. To address this, a transfer learning approach is proposed. Large-scale source domain data are taken from the ASHRAE RP-884, while small-scale target domain data are collected from local climate chamber experiments simulating real indoor environments. The proposed multilayer perceptron (MLP)-based model achieves an accuracy of 0.719 and a weighted F1-score of 0.616 on the target dataset, demonstrating the effectiveness of transfer learning for thermal comfort prediction in data-scarce regions under future climate conditions.</p>