Using wearable sensors to detect thermal preference in the field

Paper published in Building and Environment

Again, we are experiencing another hot summer this year. Temperature neared 110 °F in many cities of U.S. and Europe in July. Your building managers turned down the thermostat, city managers set up misters, and your friend handed you a cold beer from the fridge. All looked natural and reasonable because they “predicted” your thermal preference correctly.

However, such accurate prediction does not always happen. In fact, the high prediction power on thermal preference may not exist at extreme conditions, being hot or cold (Figure 1), because of the obscure detection regions (Figure 2).

So what is the prediction accuracy of thermal comfort in a real living environment using wearable sensors?

Would the prior results be too optimistic when putting people in a controlled environment?

To answer the question, 14 participants wore lab-grade wearable sensors (e.g., heart rate, skin temperature, accelerometry) for two weeks (Figure 3) and self-reported their thermal preference (cooler, no change, warmer) as many times as possible every day. They wore the sensors for at least 20 hours per day in order to capture their daily activity dynamics that may not be covered in a controlled lab environment.

We found the best and the median prediction power of thermal preference are approximately 43%/77%/0.78 ( Cohen’s kappa/accuracy/AUC) and 24%/78%/79%, respectively, much higher than the conventional PMV and adaptive models as expected.

However, the prediction power is highly dependent on thermal sensation. For example, the Cohen’s kappa is only 3% (0% means random guess) when the thermal sensation is from -0.5 to 0.5 as the general design practice aims (Figure 4).

In the study, we also calculated the prediction powers of different parameter combinations (Figure 5). Skin temperature at ankles or feet might be a useful feature for prediction.

For more details on the paper:

Liu, Shichao, Stefano Schiavon, Hari Prasanna Das, Ming Jin, and Costas J. Spanos. “Personal thermal comfort models with wearable sensors.” Building and Environment (2019): 106281. doi: 10.1016/j.buildenv.2019.106281

Complimentary copy: (link)