Powerful machine learning algorithms have existed for decades, but only recently have they become widely usable: “We have a lot of data now. Decades ago, you would design and run an electricity system, and you wouldn’t get much data about how it operated. Nowadays, we can ‘sensor up’ almost anything, and machine learning can then provide you with insights that you might not otherwise get”. In recent years, Nigel Goddard and other computer scientists at the University of Edinburgh have been exploring innovative ways to use data from smart meters: “The tagline we had for our projects was ’putting the smarts in the smart meter’ because smart meters just give you a lot of data. They’re not actually smart”, jokes the expert. Nigel and his colleagues wanted to use data from smart meters to get insights into how people use their energy. “There's a specific thing that we worked on quite intensively; it's known as the disaggregation problem in machine learning.”
Disaggregation refers to the ability of machine learning algorithms to identify which appliances are being used. For neural networks to be effective, they need to recognise individual appliances even if several are used simultaneously, and only the “aggregate”, or combined, data is available: “For something like a kettle, you turn it on, the energy use goes up and stays at a constant level while the kettle is on. When the kettle turns itself off, it goes down. Very simple. If you look at something like a washing machine, when you turn it on, nothing happens at first because it's letting some water in. Then the water heater turns on, so that's a bit like a kettle. Then the water heater turns off, a motor turns on, and it moves your washing around”. By training neural networks on individual appliances, Nigel and his team were able to detect energy consumption patterns of each of these machines, even when their signals were mixed together: “We did some fairly groundbreaking work on that disaggregation problem, and it's had a number of applications”.
Nigel and his colleague Lynda Webb cooperated with Blackwood Homes and Care to test how their technology could support vulnerable people. “The people gave us rules like, ‘Here’s what I normally do. If I haven’t used an appliance by lunchtime, that’s probably a problem’”. Nigel and Lynda arranged for the participants to have responders who would be alerted if someone didn’t use an appliance within their usual timeframe and wasn’t responding. “We actually had one real-life incident where one of the participants had a medical event that caused her to be unable to respond, move or do her normal things. Our system kicked in, and it tried to contact her but couldn’t. It contacted her responder, who went around and found that she needed medical attention. She was fine in the end, but it was a matter of a few hours from when this thing happened to when the medical attention got to her - if that system hadn't been there, it could have been days”. Having consulted medical professionals, Nigel believes that the NHS or similar services could, in the future, offer systems like his, helping vulnerable people and their families feel more secure
Reducing energy demand
While the researchers' feedback from vulnerable people and their families was overwhelmingly positive, it remained unclear for some time if people were interested in receiving advice on their energy consumption. Providing such advice was the goal of Nigel’s original project, IDEAL, which involved over 250 UK households. “We were getting going with this kind of advice around 2015 and 2016. One of the things that became quite clear to me was that most people don't want to think about energy. They might think about what they do with energy, like: ‘Oh, I need to do the laundry’, or ‘I need to cook’. They don't want to think about energy usage”.
Yet, Nigel believes that with the recent increases in energy prices, people have become more aware of their energy usage and hopes that the insights from the original project may be even more relevant today than when it was first launched. “People think that because the light is bright, it must use a lot of energy. Whereas, as engineers know, energy really goes to heating and cooling. We gave every household a tablet and designed an interface for them to look at how their energy was being used. We would give them suggestions on, for example, what sort of wash cycles they were using in their washing machine and suggest to them that perhaps they didn't need to run at such a high temperature, or if they weren't spinning, that it's a lot cheaper to spin the water out of the clothes than it is to tumble dry it out”. As the computer scientist points out, the homes of the future will likely use similar technology to what he and his team developed at the University of Edinburgh, automating energy use to be as efficient as possible so that people won’t have to constantly think about how to save energy.
Automating energy systems is only part of the answer when it comes to reducing energy demand, a lesson that became evident in another project called Enhance, which focused on energy use in public buildings. “A lot of energy use is determined by people's behaviour, not by some system. The systems underneath are what people have to work with, but how they choose to use those systems is what determines the energy use. We developed methods for working with people to try and engage them in coming up with ideas about how to do things more efficiently”. To advance the energy transition, scientists can’t rely solely on engineering solutions – they need to collaborate with various groups of people, including companies, policymakers, city councils, homeowners and commuters. “If you don't get them all somehow involved in crafting net zero solutions, people will say: ‘That doesn't work for me’”.
Energy as the foundation of economic activity
The impact of informatics on the search for energy solutions extends beyond machine learning or enquiries into how AI algorithms interact with people. Teaching a university course on dynamic modelling of energy systems, Nigel Goddard hopes to provide researchers from all disciplines with an alternative to conventional economic models. “Energy is what we use to process all the materials, provide all the services, and run our computers. It’s all energy. We use system dynamics models to try to create a model of the economy, thinking of it not as a financial system but as a biophysical system where materials flow in, being transformed with energy into products and services. It’s very different from the economic way of modelling, which is to do everything in terms of prices and money. The system dynamics models allow us to model time processes. If you want to build a lot of wind turbines because you decide to use renewable energy, that takes a long time because it's a physical engineering problem”.
While the early system dynamics models, including the “Limits to Growth” model from 1972, did not always succeed in predicting future developments, Nigel Goddard stresses that these models have since significantly improved. “Most recently, there's an EU project called Locomotion that has built a global system dynamics model looking at the net zero transition”. Nigel’s own past projects studied the planned deployment of renewables in the United Kingdom: “One of the key things was that to build all these wind turbines and put them out in the ocean, we needed a lot of equipment. You don't just need the factories; you also need the ships to take it out, the various machines that will install it on the seabed, and so on. That wasn't really factored in at the time we did this”. The computer scientist believes that system dynamics modelling could be just as valuable when studying other emerging technologies, such as those relying on hydrogen. “I'm most excited about using these system dynamics models as tools, for policymakers to test their ideas”.
Multidisciplinarity and energy at Edinburgh
In the university context, knowledge of policies and technology often comes from scientists in various disciplines, each of which can contribute to Nigel’s system dynamics models: “If I want to build the kind of models I was talking about, I need to include some of the social and organisational aspects. However, I also need engineering expertise to understand what goes into building and using wind turbines. So, I need to bring those together. Having a network like Energy at Edinburgh, where people are already talking together, is great; it’s perfect”. Recognising the importance of different disciplines is part of a broader understanding that, in energy, solutions are rarely straightforward: “What I'm really trying to do is to bring an appreciation of the complexity of the systems that we live in and rely upon in the world, how they interact, and that there aren't simple answers. A solution that you produce in one place may have some negative effects somewhere else, and that's just the way these systems work”.
Interview audio recording
Read the full interview with Nigel Goddard, where he delves into how machine learning in drones could help monitor energy grids and offers advice on getting started with system dynamics modelling.
Author and interviewer: Jan Žižka
