Data-driven thermal comfort modeling: Comparing AI-based predictions with PMV-PPD model
Accurate thermal comfort modeling is essential for optimizing energy-efficient, occupant-centric indoor environments. While widely used, traditional models such as Predicted Mean Vote (PMV) and Predicted Percentage Dissatisfied (PPD) often fail to capture individual variability in thermal perception...
Saved in:
| Summary: | Accurate thermal comfort modeling is essential for optimizing energy-efficient, occupant-centric indoor environments. While widely used, traditional models such as Predicted Mean Vote (PMV) and Predicted Percentage Dissatisfied (PPD) often fail to capture individual variability in thermal perception and dynamic environmental changes. This study proposes a data-driven framework integrating objective environmental parameters (air temperature, mean radiant temperature, humidity, and air velocity) with subjective human responses (thermal sensation, comfort level, and satisfaction) to enhance thermal comfort prediction. Multiple Artificial Intelligence (AI) techniques–including Multiple Linear Regression (MLR), XGBoost, Random Forest, Multi-Layer Perceptron (MLP), and a Fuzzy Logic System (FLS)–were developed and systematically benchmarked against the PMV-PPD model. Results demonstrate that incorporating subjective data significantly improves prediction accuracy, reducing MLR’s RMSE from 0.989 (objective-only) to 0.688 (combined data). The FLS achieved competitive performance (RMSE = 0.704) while offering high interpretability through transparent rule-based modeling. In addition to RMSE, Mean Absolute Error (MAE) and Mean Bias Error (MBE) were used to evaluate consistency and bias, confirming that MLR and FLS delivered low-error, low-bias predictions suitable for practical use. The novelty of this work lies in (i) the integration of objective and subjective data streams within a unified framework, (ii) the statistical validation of AI models’ superiority over traditional PMV-PPD methods, and (iii) the introduction of an interpretable fuzzy logic model suitable for occupant-centered HVAC applications. These findings support the development of adaptive, explainable, and human-centric building systems for real-world deployment. |
|---|