On September 10, I Skyped with a researcher at the Institute for Future Energy Consumer Needs and Behavior (FCN) at RWTH Aachen University. His work applies agent-based simulations, or models of interacting autonomous actors (“agents”) with encoded needs and/or preferences, to energy-related problems. He’s co-authored a number of papers in this area, including one on German biomass plant installation and one on the spread of solar PV systems in Italy. He is also working on projects involving techno-economic analyses of large energy systems and pricing/regulation in power markets. We talked primarily about the merits and limitations of agent-based modeling for Smart Grids and renewable energy.
(Note: The following interview is not a direct transcript. It is written by me, Priya Donti, and is based on notes I took during the meeting. The interviewee reviewed this writeup before its publication.)
Priya: In your perception, what is a Smart Grid?
Interviewee: Smart Grid definitions can vary depending on the focus of the research. For instance, as an economist within FCN, I tend to focus on consumer behavior, how technologies will spread across a region, and the role of prosumers. My institute’s Smart Grid definitions are framed by this background. However, in general, a Smart Grid has two major aspects. The first is that a Smart Grid encompasses both electricity generation units and electricity consumption nodes. The second is that the electricity generation units involved tend to be small and operate within a relatively small area. Smart Grids pertain not to a whole country’s grid, but instead to a closed system or subsystem within a country. They also tend to involve low- and medium-voltage distribution grids rather than higher level grids. A Smart Grid then tries to integrate the different aspects of generation and consumption in the most efficient way possible.
Priya: How are agent-based simulations useful in the context of energy and Smart Grids?
Interviewee: In an agent-based simulation, researchers create a micro-structure for each agent’s behavior. Each agent acts and reacts according to specific rules and algorithms. You can then examine how different types of individuals may behave or interact, as well as the aggregate-level outcomes of these micro-level interactions. Such models become important as, due to Smart Grids and decentralized energy generation, consumers play a larger role in the energy system. Specifically, there are two major ways agent-based models can help examine the effects of energy system policies. The first is to understand aggregate outcomes based on knowledge about agent behaviors. For instance, one of my papers examined the spatial distribution of biogas plant investments, by having regional agents make investment decisions based on biomass substrate availability. The second use is to understand the behavior of specific agents, given knowledge about the aggregate outcomes. Along this vein, my other paper examined what factors drive households who invest in photovoltaics. We encoded social and demographic information into the household agent models and saw if the aggregate outcomes matched the reality of photovoltaic investment. Similarly, there’s a whole literature on Smart Metering that tries to see whether and which types of customers alter their behavior given access to certain amounts of energy consumption information.
Priya: What are some limitations of agent-based models?
Interviewee: It’s difficult to gather reliable underlying information that properly represents your agent population. Even though there’s some prior research on technology diffusion and adoption decisions, the data often has shortcomings or is difficult to use. For instance, if your agent-based model looks at the United States but the study you’re looking at involves Germany, it’s hard to assume the same social factors are at play even if your model uses a similar parameterization. Ideally, you would conduct your own social experiments and build your agent-based simulation models on top of these, but that can be difficult in terms of both time and money. You end up having to take some shortcuts and make assumptions about the factors that drive representative agents, and it becomes complicated to justify why you constructed your model in a certain way. As an example, during our photovoltaic study, we had to first ask what factors motivate a household to buy solar PV, how different households differ from each other, and so on before encoding this information into the model. Gathering accurate real-world information about these topics can be difficult.
Priya: How do you validate the results of agent-based models?
Interviewee: There are multiple ways. One way is to validate your agent model beforehand with surveys, choice experiments, etc. so you have some idea of how individuals make decisions in certain contexts. Another way is to add shocks and vary the value of certain variables to see how outcomes change, which is what we did in the biomass paper. In the PV paper, which looked at individual behaviors in the context of past PV diffusion, we made sure the aggregate results of our models matched how photovoltaic diffusion in Italy actually took place. However, agents adapt and people change, so as soon as you make a simulation of the future, your validation is always partial. Such complications are typical of the social sciences.
Priya: How are agent-based model results used to affect policies and behaviors?
Interviewee: It’s a difficult question, because policies are often drafted not only based on academic results, but based on many general, external influences. Generally speaking, I think academics and policy are not well-connected. Politicians aren’t interested in articles that are too technical, but less technical articles aren’t taken seriously by the academic community. There are some general communication problems, but there is always an attempt to communicate between the academic and policy worlds.
Priya: What are some of the major changes necessary for Germany and the EU to reach 100% renewable energy?
Interviewee: First, Europe has to solve some major infrastructural problems in the electricity market, which are related to fundamental shifts in electricity production/consumption/transfer across different areas, individuals’ behavior, the availability of facilities and services, and the ICT layer. These changes involve massive long-term investments, which need to be solved in an economical way. For instance, if you could just store all excess electricity in batteries, our problems would be solved. However, batteries are costly, cause CO2 emissions when produced, and produce waste when disposed. Next, Europe needs to collaborate with the rest of the world. Europe doesn’t necessarily need to be fully 100% renewable if other countries also reduce their fossil fuel use. To accomplish this goal, technologies need to be cheap enough to be implemented in the rest of the world. We also need to raise carbon prices to prevent carbon leakage, or the phenomenon when companies relocate elsewhere due to high electricity prices. Such relocation can sometimes cause more pollution, since companies often shift to places with energy systems that are more polluting than ours. For instance, it wouldn’t be smart to have policies that drive industries outside of Europe to Latin America or India, whose emissions per unit of energy produced are far greater than Europe’s. These are all significant problems that may not be solved in the short term, but will be solved in the next 20-30 years.