In theory, a world powered entirely by renewables is possible. But given the natural fluctuations of the wind, waves and sun, how can we ensure that the green energy market of the future is safe, fair and truly sustainable? An open-source project Power TAC is using machine learning simulations to help us find out.
Today, arounda quarter of the world’s electricity comes from renewable energy sources, and although the implementation of renewables will have to accelerate even more if we’re to meet global climate targets, supplies are nevertheless growing faster than expected. This can be attributed to multiple factors: from the falling cost of renewables technology (especially solar) to the rising number of people becoming ‘prosumers’ (consumers who produce their own energy with self-installed rooftop- and balcony-based solar systems, or by joining local decentralised neighbourhood microgrids). Such technologies promise a wealth of financial and social benefits for locals wishing to take their energy supply into their own hands. But they also emphasise an issue that is growing larger as access to renewable energy becomes more widespread: what will a complex, global renewable energy market actually look like in practice?
Even in 2020, it’s easy to think the only thing stopping our world converting to 100% renewable energy is a lack of renewable energy itself. Or perhaps you could say demand comes into it, too. Either we need more wind farms, or we need to persuade more consumers to buy wind energy. In fact, it’s what happens in between these two factors that can prove most problematic.
The green energy revolution has to work with existing policies and infrastructure – systems that have been optimised for fossil fuels since the industrial revolution. Traditionally, fossil fuels have been easy to value and trade because their production is controlled. Renewables, on the other hand, are more unpredictable – you can build a wind farm on a windy shoreline, but there’s no way to schedule the wind itself. This can produce problems for the people selling and buying that energy because of the resulting fluctuations in supply – and value.
A project called PowerTAC (the "TAC" part stands for Trading Agent Competition) seeks to resolve those issues using artificial intelligence. The open source platform uses digital “agents” to construct hypothetical scenarios, imagining every way in which traders might interact with one another in the context of an volatile energy market. The end game? Policy-makers and industry players can obtain the understanding they need to support the implementation of green energy systems around the world. As PowerTAC co-founder Wolfgang Ketter explained in his Ted Talkon the subject, “we have to introduce sustainable energy in a sustainable manner.”
Ketter founded the project with John Collins in 2009, after giving a keynote speech on the potential for future energy systems to be enabled by AI software and machine learning. His talk piqued the interest of federal government representatives who encouraged him to develop his ideas further. Fast forward a decade and PowerTAC currently holds the title for the largest open source smart grid project in the world. “Quite astounding considering in 2009, I was sitting in a biergarten with my friend writing ideas out on a napkin!” Ketter laughs.
The software has since been downloaded over 10,000 times by a range of users – from startups to PhD students – and the project also has ties with numerous federal and educational bodies, providing “a rich teaching tool for groups studying the economics of sustainable energy.” Since 2012, PowerTAC has hosted 8 competitions and workshops on topics like ‘Policy Awareness Sustainability and Systems,’ which invites the participation of policy-makers, tech innovators, behavioral researchers, and business actors, to name but a few – all working together to “harness AI for social good.”
How does the platform actually work?
The software itself is a competitive simulation that mimics the interactions of energy suppliers and consumers, providing potential avenues for balance and regulation. Exploring these market behaviours then highlights where more regulation is needed to protect the system from exploitation. Being both flexible and modular (outside the simulation core, that is), the platform is well-suited to welcome new collaborators, and new developers are free to contribute regardless of previous experience.
Ketter tells RESET that PowerTAC functions in two modes: tournament mode, where specific rules and regulations apply; and research mode, which is more customisable and can therefore be used to create hypothetical environments – such as a snowstorm, for example – more quickly. “There’s no other tool in the world which does this,” he notes. The platform’s tournament mode is in keeping with PowerTAC’s solid track record of hosting tournaments globally since 2012; events that are set up to develop the PowerTAC community as much as the software itself. Ketter is enthusiastic about this aspect of the project, stressing that all are welcome to these tournaments, not just software developers.
How does PowerTAC use artificial intelligence to project hypothetical situations?
Put simply, PowerTAC uses a range of algorithms to map the possibilities of supply and demand within the sustainable energy industry: as an example, the same genetic algorithms used in evolutionary computing can be used to simulate the way market players trade, partner and succeed (or fail). Evaluating each algorithm’s ease-of-use against its accuracy is an important part in making these processes efficient: depending on what needs to be achieved, PowerTAC might use learning techniques taken from Q-learning, neural networks, or deep reinforcement learning.
PowerTAC explores and tests options that are theoretically desirable, without compromising energy markets in practice. Ketter cites an article entitled “Agents of Change” written by The Economist in 2010, which suggested that crises like the California energy crisis of 2001 – perhaps even the financial crash of 2008 – might have been avoided had agent-based simulation like PowerTAC been used. These kinds of systemic failures had occurred as a result of loopholes that were then exploited to the detriment of the system overall.
There are still challenges ahead for the project, however, such as how to make it more accessible to users who lack a software developer’s know-how. In these cases, Ketter advises collaborating with developers to get the best out of PowerTAC, and while the project currently employs a handful of people internally to tackle the issue of accessibility, they still need further capacity to share the benefits of the platform effectively. Ketter is looking to find more students taking a specific interest in this over the coming years, hopefully aided by future university partnerships across the Erasmus scheme.
In fact, the project has wider implications for education in the energy sector, too. For instance, students have used the platform to explore new business models and to build renewable energy cooperatives in Africa. The platform also invests in partnerships with a number of universities, whose students contribute to the software by modelling it in various ways: wind energy parks and residential scenarios are just two of 180 different customer models that now exist. Because of the sheer extent of its use cases and its inherent flexibility, it’s a tool that lends itself to experimentation and exploration, and the discovery of inventive and scalable solutions. As Wolf himself observes: “PowerTAC is a tool for inspiration as well as innovation.”
PowerTAC is always developing as a project and as a community. Later this year they look forward to introducing new research tools to better facilitate studies that use the platform; as well as planning a third iteration of their Policy, Awareness, Sustainability and Systems (PASS)workshops.
This article is part of the RESET Special Feature "Artificial Intelligence - Can Computing Power Save Our Planet?"