A common question for the data science team is “why is it so hard to detect light bulbs?” The simple answer is that light bulbs don’t draw a lot of power, making their electrical signatures often indistinguishable from noise and difficult to identify. We also touched on some of the challenges of identifying light bulbs in a previous article.
The three most common types of light bulbs are compact fluorescent lights (CFLs), light emitting diodes (LEDs), and incandescents. These consume 13-15 Watts, 6-8 Watts, and around 60 Watts, respectively. Figures 1, 2, and 3 illustrate the electrical signature for a CFL, an LED, and an incandescent. Of the three types, LEDs are the most environmentally friendly and energy efficient, leading more and more consumers to switch over to LEDs to save on their electrical bills. Although this makes the task of detecting light bulbs harder, the data science team does have a couple of tricks up its sleeve to improve performance.
a. Light clusters: Although single light bulbs, particularly CFLs and LEDs, do not consume a lot of power, devices such as vanities lights and chandeliers (yes people still own chandeliers!) tend to use more than one bulb. Since these are turned on simultaneously, the increase in power consumption leads to a stronger electrical signal that is easier to identify.
b. Light context: Lighting devices tend to be associated with other more significant appliances. For example, a lot of households activate the bathroom light and fan at or around the same time. Another example illustrated in Figure 4 is the light of the microwave which is turned on shortly before the microwave is actually used. The Sense energy disaggregation algorithm can make use of these associations to improve detection performance.
c. Light annotation: As we work on improving our disaggregation algorithm, light bulbs might be initially detected as unknown devices. In a previous article, we talked about how you can participate in device detection by renaming devices identified by the Sense monitor. As the data science team grows its database of labeled light bulbs, with your help, we can fine tune our feature extraction and machine learning algorithms using the annotated data. The more data, the better!
We end with a mystery wattage signal inspired by this blog (Figure 5). Can you tell what is going on? Leave us a comment on Facebook if you have figured it out. Good luck sensing!