New Intelligence on Future Energy Markets
The face of Europe’s energy supply is set to change. If latest plans by nine European countries go ahead, the North Sea could become home to an intricate network of undersea cables – part of Europe’s first supergrid for renewable energy. Thousands of kilometres of cable will re-route the power from various sources – be it Scotland’s wind turbines, Germany’s solar panels or Norway’s hydro-electric stations – to where it is most needed.
Initiatives of this type signal a marked departure from the traditional model of energy supply management, says Wolfgang Ketter of RSM’s Department of Decision and Information Sciences. Until now a few large companies have held centre stage in each country, operating in a largely top-down, command-and-control fashion. But with governments trying to reduce their dependence on non-sustainable fossil fuels, and attention turning towards greater use of alternative ‘cleaner’ energy resources, energy generation will become a far more dynamic and dispersed process – adding significantly to control issues and the complexities of balancing supply and demand. Add into the mix the predicted growth in electric vehicles, and the picture becomes more complex still.
To model what a future energy market might look like, Ketter is using multi-agent systems, an artificial intelligence technique developed in the mid-90s, to develop an ‘energy trading agent competition’. This will bring together research teams from some of the world’s leading institutions and from industry, and allow them to simulate what happens in the energy market under a variety of conditions.
‘We’re looking at how to ensure that energy market designs of the future are both efficient and robust,’ says Ketter, ‘so that we can avoid another disaster like Enron, which was an energy market design problem. Using the kind of tools we are developing you can simulate what could potentially happen before you risk money and people on a system which might in real life break down. We can examine, for example, the effects of policy changes such as taxes and incentives, the impact of external shocks on the system’s overall stability, or how rapid technological changes affect the infrastructure.’
‘Our second goal – which is vital for us as researchers – is to work out how to design software agents that will perform really effectively in these markets. The competition provides a perfect testbed for that.’
The ‘trading agents’ which will be tested in the competition are pieces of intelligent software that can make decisions autonomously. Because they have complex algorithms built in they can judge situations and reach decisions without necessarily involving a human user.
‘They can do everything you would expect to see in a market: submitting bids, requesting quotes, negotiating deals and accepting offers,’ Ketter explains. ‘They are very much faster than humans at processing information, and they also have reactivity, enabling them to adjust to conditions in the market. So they can constantly monitor prices, set prices, and react on the spot to market changes. They can be used as a kind of “personal assistant”, helping us to make better decisions, particularly in scenarios where information changes very rapidly.’
Ketter envisages many potential applications for such agents, both in commercial and domestic settings. In the energy market simulation the agents act as a type of broker – purchasing power from distributed sources and from regional energy exchanges, and selling power to consumers and exchanges. They have to solve a series of complex supply-chain problems, working with a product that is perishable and an environment notorious for being highly variable and uncertain – with shifts in weather, and equipment and networks subject to outages. They are operating in a dynamic network that involves multiple timescales – from negotiating long-term contracts with energy suppliers and tariffs for customers that will balance expected supply and demand, through to spot-market trading and real-time load balancing in the grid.
The agent software is designed up front, by the research teams who have signed up to compete. But once a game starts and the agents are put into play, the process is completely free from human intervention, as Ketter explains:
‘Across the many games that make up a round of the competition, the agents act entirely autonomously. The humans just observe what is going on in the market. Only between rounds can agents be tweaked, or new pieces of software added.’
‘The agents all have to do the same task: the issue is, how they will actually go about it. The algorithms on how they buy materials, how they produce and sell to customers will be completely different. Competitions of this type pit the agents against one another – hence the name multi-agent – so in a sense we are testing many different algorithms at once.’
The beauty of the system, he says, is its flexibility. ‘If a new type of energy resource comes along, and someone is building a model for it, we can just plug it in, providing it complies with the competition’s programming interface. The framework can cover all types of consumers or producers and play through many different scenarios.’
The energy market competition is the fifth such trading agent competition that Ketter and his colleagues have devised. The benefits to industry are already evident from an earlier supply chain competition, in which industry research teams competed alongside academics.
‘Companies like SAP, Ford and Daimler have adopted some of those algorithms in their own work,’ Ketter explains. ‘And many of the ideas developed will not necessarily be domain-dependent. The best algorithms to emerge from this latest competition can also apply outside energy management.’
What competitions of this kind also do is spark innovation, Ketter argues. ‘In a competitive setting, people generally become much more creative than if they just sit alone in the design process. They devise far more innovative ways of doing things. If you are competing against the top schools in the world, you have to be really sharp. But if your agent wins, then you have something really credible.'
The competition design is due to be completed this Spring, to be followed by rigorous testing and evaluation – including its usefulness as a tool for policy makers. After a demo competition at the ACM Conference on Electronic Commerce at Harvard University in June, the full competition should be open to participating teams next year.
‘It's a truly exciting area to be working in,’ says Ketter. ‘One thing is sure, trading agents are going to become increasingly essential, both for a sustainable energy market and for various kinds of supply chain management.’
Wolfgang Ketter is Assistant Professor in the Department of Decision and Information Sciences at Rotterdam School of Management and a member of the Erasmus Research Institute of Management. He is also director and founder of the Learning Agents Research Group at Erasmus (LARGE) and on the board of directors of the Association for Trading Agent Research.
His collaborators on this project include Carsten Block and Christof Weinhardt of Karlsruhe Institute of Technology, John Collins of the University of Minnesota, and Eric van Heck of RSM. Further details available in this working paper.