The drive to help consumers cut their fuel bills has been going on for decades. Governments regularly run adverts to advise on insulating houses and why consumers should change to more energy efficient appliances. One of the goals of smart meters is to show the consumer exactly how much energy they are using at any given point in time. With all the recent high profile attention on global warming, this has taken on a new urgency.
The problem is that effecting change is difficult. Getting accurate data and building models that can understand how energy is used is the first step. At the Spark+AI conference in Amsterdam, Enterprise Times talked with Ellissa Verseput, Machine Learning Engineer at Quby.
Verseput explained that Quby is in the smart energy domain. Its goal is to help households save as much energy as they can. How does it do this? Verseput says: “We inform them about their energy usage, which appliances are using more energy and which are doing fine. We tell them how their heating can be optimised to reduce gas.”
It is not all about energy saving as Verseput explained. “We also send predictive alerts to prevent boiler failures. We look at what heat pumps can provide. For customers who don’t have solar energy we are exploring how we can give them a good estimate of what solar panels can do. For those with them, we help them understand what they are producing.”
How do you build a model to deliver insights?
One of the challenges here is how to build a model of usage. A meter reading, even from a smart meter, is a single number. Changing behaviour requires far more granularity. Verseput is part of the team that set out to solve that model using machine learning.
Quby collects terabytes of raw data every day from customers. Customers installed smart plugs in their houses to provide detailed information on what individual appliances were consuming and at what times of day. Verseput said that this data allowed the machine learning team to create a training set that it then used to build the machine learning model.
Not everyone will want to install smart plugs. An alternative way of gathering the data would be for housebuilders to add smart data devices to fuse boxes. This would provide granularity based on what each fuse controlled. In houses, different circuits have their own fuse. For example, lighting, plugs upstairs and plugs downstairs. Large appliance such as cookers and washing machines have their own fuses and circuits. This is a safety issue and taking the data from these circuits would provide highly accurate energy usage data.
This detailed data makes it easy to identify usage patterns and classes of devices. This is important. If you don’t know what is consuming the most power, you can’t change how it is used. It allows the homeowner to decide:
- What devices should only be run on low cost overnight tariffs?
- What devices can you do without completely?
Unseen usage is often a surprise
Two examples of the challenge of identifying what is consuming energy in a household is consumer electronics and standby mode. Laptops, tablets, phones and other portable devices are regularly on and off charge in a household. The devices that charge these devices are often left connected and powered on. Those units draw energy, albeit just a small amount. People rarely share chargers so everyone in a household is frittering away money.
This is also true of standby mode. When users get an app or device that shows them their real-time usage, they go around turning devices off and unplugging them. It is a limited hit. Over time, they stop doing it for various reasons. With standby mode it is often as simple as the time it takes for a satellite or cable TV box to reboot or synchronise.
Changing user behaviour by evidencing cost savings
Changing behaviour requires accurate insights. Governments have had limited success in getting users to change old energy guzzling appliances for energy saving appliances. The reason is that they were unable to show that the cost of the new appliance could be funded from energy savings.
What Verseput and her team are able to show is much more accurate data around the cost of using specific appliances. This means that they can tell the customer what a given device is consuming and provide a reasonable estimate of savings by buying a more energy efficient one. This is something that government campaigns have failed to do. It will be interesting to see if governments begin to engage with the likes of Quby.
Another way that Verseput says Quby is helping customers understand their usage is by comparing them with other properties. This allows them to benchmark their consumption to similar types and sizes of households.
Such a move will always throw up outliers. Verseput gave the example of someone with an aquarium who would have a higher than expected energy consumption. Despite that, Verseput believes that the model can provide accurate information for 90% of customers. If that led to just a small reduction in consumption, it will have a significant impact on energy usage.
Not just about smart meters and engaged households
The rollout of smart meters around the world is taking far longer and is far more costly than previously thought. This has slowed the benefits and has already caused problems as different generations of technology are now in use.
For Verseput and her team, this is not a problem. They are focused on training their model using high resolution data from customers with smart plugs and smart thermostats. These customers provide data every 10 seconds. The model can then be used to interpret data from any generation of smart meter which generally provide data every 15 minutes. By being able to extend the model in this way, Quby can ensure it reaches the maximum number of households to get the maximum reduction in energy consumption.
Data security and privacy not forgotten
It was also refreshing to hear Verseput talk about how aware Quby was when it comes to customer data. All data it gathers from customers is encrypted and sent back through a VPN over the customers Internet connection. This should prevent criminals getting access to data and using it to know when a household is unoccupied.
Where the data comes from smart meters, it comes from registered third-parties who gather the data. Protection of that data is down to those organisations. However, there are well founded concerns over the weakness of smart meter security. While not a problem for Quby, it is a problem for the industry and one that needs addressing.
Verseput also pointed out that the nature of this data means that once it can be connected to a household, it has to be treated as personal data. Building in the security controls to the data management from the start is a smart move.
Enterprise Times: What does this mean
The use cases for machine learning systems are often focused on helping organisations improve their revenue and profits. It is uncommon to come across an organisation who is delivering a solution that can have an immediate impact on helping households save money. In the long-term this is about more than just saving on energy bills. Reducing energy consumption allows the more polluting forms of generation, such as coal and gas, to be swapped out for renewable energy.
At present, Quby is focused solely on the consumer household market. However, its model could be applied to businesses. There is nothing to stop enterprises using smart plugs to get a more granular understanding of where they are consuming energy. Given that this is often a significant cost to businesses, it should be an easy sell. It also plays well to the green agenda that large enterprise in particular, like to talk about.
Quby has a deal with a major utility company in the Netherlands that will more than double its customer base from 400,000 to 900,000 by the end of 2019. There is nothing to stop it using its machine learning model, adapting it to different countries and expanding rapidly. It could even do so through partners. This is the type of project that fits several EU targets. It will be interesting to see where Quby goes next.
