Quantifying the Impact of Demand Response & Informing Investments: Exclusive Interview with Miha Grabner of the Milan Vidmar Electric Power Research Institute
image credit: Miha Grabner
- Apr 25, 2019 1:45 pm GMTApr 19, 2019 1:46 am GMT
- 295 views
As demand response (DR) becomes a more prevalently integrated energy management strategy in the utility industry, it’s becoming more and more crucial that companies are able to foresee and quantify the impact of potential demand side management strategies. Before investing resources into large rollouts that these types of programs, energy providers want to know that they are going to pay off in the end and provide the types of benefits to the grid that they are intended to—minimize the maximum peak load, prevent over congestion, and accurately match supply with demand.
The Milan Vidmar Electric Power Research Institute (EIMV), the leading Slovenian engineering and scientific research organization for the electric power industry, has been working in this area in a number of studies. As a part of the Slovenian-Japanese NEDO Project (a partnership between the Japanese agency NEDO and the Slovenian TSO), they studied DR potential for utilities and the response of customers, ultimately making direct recommendations to the nation’s Energy Agency—marking the first demand response project in Slovenia as Europe starts to catch up to the United States when it comes to these strategies.
Miha Grabner of EIMV is set to present at the upcoming the Grid Asset Management 2019 conference, where his talk is titled "Demand Response – accurately quantifying the future impact of demand side management to inform smarter investment planning.” Ahead of the conference next month, Miha agreed to share some of his insights in an interview for the community of Energy Central. And if you want to hear more, don’t miss his presentation at the conference in May.
Matt Chester: Your presentation at the conference will focus on how to assess the potential that DR has using data analysis techniques before actual implementation. Is this strategy not currently terribly common? Are utilities more likely to jump in blind? If so, what problems has that caused?
Miha Grabner: Very good question! It should be! But here we have to distinguish two types of analyses: a typical old way using Excel spreadsheets or data analysis which are performed by a trained data scientist. There is a big difference between these two. Some utilities have already started employing data scientists and some are still doing it the old way. As you know, this is a whole new scientific field and electrical engineers have not learned these new methods at a University. There are also two important things that add extra complexity, which is a high integration of PVs that affects daily load profiles and a new data sources such as smart metering data, which provide better insights to the consumer behavior.
Probabilistic baseline load prediction during the DR event.
MC: You'll also discuss how consumers respond to DR strategies. In your experience, what are the behavioral challenges on the consumer end? Are customers typically on board when told about the potential savings or do you find you have to use data to convince skeptics?
MG: I will discuss a data approach by analyzing smart metering data using machine learning after the DR events. But on the other hand, a social aspect is very important, which was also one of the major conclusions of this project. It is hard to get people to enroll in this kind of programs, but the ones that enrolled were really very satisfied with the outcome.
The result of the survey before the program showed that the major reason for people to enroll in this project were financial savings and most of them were satisfied with it after the program. But financial savings are still quite small, for 50 hours of interruptions during a program which lasted for almost a year, the mean financial saving per household for consumers without additional equipment installed were a little less than 10 EUR, and for the ones with equipment (direct load control) a little more than 20 EUR. In my opinion, this is not enough, thus other international studies suggest that it is better to promote such events by telling people that they contribute to saving an environment. I think that in the future, one of the biggest challenges in the scope of smart grids will be to convince people to change from passive to active users or we will have to wait that smart homes will optimize our energy without interrupting our daily habits. Another interesting fact is that a lot of participating consumers were not willing to be interrupted for more than an hour.
MC: On the other hand, do you find you also have to provide proof and convincing to the utility heads before they understand the importance of this type of work and why such rigorous data analysis methods should be integrated?
MG: The easiest way to convince people that data analytics are crucial is to show them the analysis that we have already performed. Usually people are surprised what can be done using modern data science approaches, but of course, the domain knowledge is crucial here. In my opinion, in 80 % of cases, you are able to answer people’s questions just by looking at the data from the right perspective and visualizing it properly. Fancy data visualizations have gained a lot of popularity in data science, but are still not widely used in the energy industry. We use all modern data science approaches and combine it with our power engineering knowledge to answer the right questions. I also recently wrote a short blog about energy data visualization that people can check out.
Analyzing substation peak daily demand throughout the years.
MC: How widespread are the data analysis techniques in analyzing and assessing DR strategies that you're studying? Are these groundbreaking developments that haven't been rolled out in large numbers or are there already ahead-of-the-curve utilities who are implementing them?
MG: It is hard to say because a lot of internal studies that are carried out worldwide are not publicly available. What I have seen, people still use simple and inaccurate baseline (the load that would occur in the absence of a DR activation) methods for DR assessments, which is a very important aspect of the DR evaluation. Inaccurate baseline predictions can lead to false conclusions about the effectiveness of the DR program. Whereas machine learning based techniques provide much better results, but the main problem here is, that simple methods can be used without prior knowledge, whereas for applying machine learning based methods you need to hire a data scientist.
In general, a lot of machine learning methods for demand response are already developed. Right now, when talking about AI for DR there are two interesting topics in my opinion where AI will play a crucial role: first is forecasting the demand flexibility and the second is assessing the response of individual consumers from smart metering data. Forecasting demand flexibility is very hard, due to lack of data from the past and since DR events are performed under different conditions. On the other side, smart meter roll-out is happening all over the world, whereas typical methods for assessing the response of consumers at the aggregated level are not applicable to individual consumers, due to high volatility of the data.
MC: What has been the most surprising result from your study into this topic that you weren't necessarily expecting? Anything that's made you change your way of thinking about DR and it's role in utilities?
MG: The most surprising fact from the data perspective was a high variance of the consumers’ responses. There were DR events that were really successful and the mean response of consumers was as high as 0.25 kilowatts per consumer without equipment and 2 kilowatts per consumer with equipment. Whereas in a few cases the demand was actually higher during the DR event than the predicted baseline. From this kind of perspective, DR can be effective but is not a reliable solution according to our data, but further research is still needed.
MC: Can the lessons learned from this study be applied to important utility topics other than just DR—for example, with energy efficiency or variable pricing techniques?
MG: Almost all of the methods can be applied to other problems utilities are facing. The main idea is to understand the demand and in case of critical peak pricing DR program, everything is about analyzing annual peak demand. Good understanding of annual peak demand is crucial since network reinforcements and other equipment are dimensioned according to annual peak demand in the power engineering industry.
Analyzing PV generation during substation daily peaks.
MC: Aside from your own presentation, are there any topics or talks you're excited to take in while attending the Grid Asset Management conference?
MG: I am really excited about the conference. I was on a similar conference, which was organized by Smart Grid Forums last year and I must say it was really well organized.
What I especially like is that it is industry focused, as opposed to academic conferences where a lot of methods that are usually presented were not applied in practice. We are an Electric Power Research Institute and we work close to the industry. Our goal is applying state of the art methods to actual problems industry is facing, therefore it is great to talk with other colleagues about the methods they have applied in practice.
For the end, I would like to invite everyone interested in this topic to join LinkedIn group AI in Smart Grids where I post about this topic.
Interviewer's Note: Miha will be discussing these issues and more during his presentation at Grid Asset Management 2019 Conference, taking place in London from May 14 to 16. As mentioned, this presentation is titled “Demand Response – accurately quantifying the future impact of demand side management to inform smarter investment planning.”