Autonomous Energy Systems: Reimagining Optimization and Control of Future Energy Systems (Text Version)

This is the text version of the video Autonomous Energy Systems: Reimagining Optimization and Control of Future Energy Systems.

Hello. My name is Ben Kroposki, and I'm the director of the Power Systems Engineering Center at the National Renewable Energy Laboratory. Today, I'm going to talk about autonomous energy systems and our thoughts around reimagining optimization and control of future energy systems. First off, I'd like to acknowledge the NREL team, including over 60 staff members from NREL's Computational Science, Power Systems Engineering, National Wind Technology Center, Integrated Mobility Science, and Building and Thermal Systems centers. Finally, I'd also like to acknowledge our external collaborators in this project.

If we think about the transformation that is happening with the power system, we are transitioning from our current power system, largely based on large-scale power plants, like hydro, nuclear, coal, and natural gas. At the heart of these power plants, there are synchronous generators, and, typically, power is flowing in one direction from these large, central-station plants out to the customers. The system is centrally controlled, and the generation follows the demand. As we look to future power systems, we're starting to see a lot of new technologies being integrated into this system.

First off are increasing levels of wind and solar. These are variable generators with their output based on weather conditions. Also, these technologies are a little different than the conventional generation in that they typically have power electronic inverter interfaces to connect to the grid. There's also more use of communications, controls, and data and information, so-called smart grids. This is enabling a lot more efficiency from the grid itself, but it also introduces interoperability and cybersecurity issues.

There's also a wide range of new distributed technologies, like electric vehicles, distributed storage, and flexible loads. In addition, there are increasing interdependencies between the electricity grids and other infrastructures, such as the natural gas grid and communications infrastructure. Finally, all of this is leading to a much more highly distributed and much more complex to control and operate energy system.

So, let's look at some of the solutions that handle this complexity. If we look at the current grid, it really operates from the top down. Large, synchronous AC interconnects, like the Eastern Interconnect, are made up of regional transmission operators. Those operate markets and coordinate reliability across large areas, then down to the local utilities, both at the transmission and distribution level, and, finally, to the customers. In general, the power system operates very reliably and can operate with around 10,000 of these large and bulk power system generators, providing power to their end use.

But we start to look at the future system, in addition to these 10,000 bulk generators and storage, we're looking at adding in large amounts of distributed resources at the customer sites, and potentially up to hundreds of millions of these devices. Right now, we don't think that this can be a centrally controlled solution, so we've been examining a number of ways to actively control hundreds of millions of generators, storage, and load at one-second optimizations.

This is a distributed hierarchical control approach, where we start at the bottom, and each of these areas are aggregated into small cells. The small cells are then aggregated into larger cells until you complete the entire system. So, this way of handling autonomous energy systems will be important as we move into the future.

As we started to look at how to develop advanced controls for the power grid, we noticed similarities with several other applications, including transportation, building systems, and renewable technologies. We noticed that there was a set of common problems, including nonlinear controls, optimization, advanced analytics, and complex systems. We needed to understand how we could control operations of these types of systems in real time, and these systems are always varying with time. In addition, we needed to be able to handle hundreds of millions of control points. Also, things like artificial intelligence and machine-learning techniques are needed to help with the massive amounts of data and information. And, finally, these complex systems also added in a variety of issues, including the randomness or stochasticity of renewable technologies and consumer and occupant behavior.

At the end of the day, we were looking at solutions where we could tie all of these applications together using a common framework. If we think about how we were able to formulate the problems to address the challenges, first, let's take a look at the individual challenges. We had to come up with a solution that was fast enough to operate in real time, and one technique for doing this is the use of distributed optimization techniques. You can see from the diagram in the middle a large distribution feeder in the black line at the bottom. On top of that, there are little green dots that represent distributed generators. And then the red hexagons represent the ability to break that system into smaller subsystems. We're able to then do distributed optimization to help speed the convergence of algorithms.

In addition to that, we wanted to make the system scalable, so we wanted to make sure that we, in order to control millions of devices, developed a hierarchical strategy that linked home and community scale, to the distribution scale, to transmission-scale systems. Finally, we needed to use data analytics to make [the] best use of all the data and information that is around these systems, although they're not all time-synchronized.

So, at the end of the day, we needed to create an integrated system that allowed integration of all energy sources. This includes the electric power grids themselves, grid-interactive efficient buildings, vehicles and mobility, advanced wind plants, and advanced solar plants. We needed to make sure we came up with solutions that were scalable, to be able to allow distributed control of millions of devices, and perform real-time optimizations. Finally, we wanted to integrate all the data and information that was available to help optimize these types of systems.

So, as we took a look at all these different challenges, we decided to look at each of them individually and start to apply a common framework for doing optimization and control. First, let's take a look at advanced wind plants and what we were able to do with distributed controls. So, the pictures that you see here represent a large wind farm, with each of these arrows representing an individual wind turbine, and the arrow direction is the direction the wind turbine is pointing into the wind to generate as much power as possible. Now, typically, wind farm controls are based on the local measurements, as you can see from the little picture there of the wind turbine and the arrow pointing to where it's making a measurement of the wind speed and direction.

So, in typical controls, you can see on the left-hand side, as the wind starts to change direction, the wind turbines start to search for which way the wind is blowing. If you look at the figure on the right, you can see that the wind turbines are acting in a much more synchronized manner and are seemingly finding the wind direction as a whole. So, how are we able to do this? First, we take the entire wind farm and break it into smaller cells. Then interior to each cell, we have the wind turbines communicate with each other and come to consensus on which way the wind is blowing. Once they do that, they pass that information to neighboring cells. In this way, the entire wind farm can come to consensus on which way the wind is blowing.

We demonstrated that with this technique, we were able to solve the optimization routines much quicker. Using central control, it took about 13 minutes to solve, but using these distributed control techniques, it only took us around 2 seconds to solve, and this allowed about 2% more energy production on an annual basis.

We then started trying to apply these techniques to a variety of other domains, including buildings. So, if we think about buildings, for the most part, they're individually operated and run. But if we could look at how buildings—say, in a city block—are actually using their energy and if they could communicate with their neighbors, they potentially could reduce the overall energy use of the entire block. This is one of the ways we're thinking about how to connect building technologies with advanced distributed controls.

In order to do this, we needed to look at how buildings were being modeled. Traditionally, buildings are modeled in a way that focuses on energy reduction and occupant comfort, and this is fine for individual buildings. But as we looked at how they interact with the larger electric grid, we needed to find ways for these buildings to provide ancillary services while still keeping people happy. So, this allowed us to build models that would then integrate with larger-scale grid models to coordinate large number buildings and energy systems.

We also took a look at how vehicles and mobility fit into this picture. If we think about autonomous vehicles, we usually think of individual vehicles driving around a city. But what if these could communicate with each other and make decisions on what is best for an entire fleet of vehicles? Our thoughts around distributed optimization and control can lead to this type of application.

So, in order to do this, we needed to think about all the various parameters that go into modeling electric vehicle fleets, including the battery size, the fleet size, the occupants in the vehicles themselves. Also connected to that are things like the price of electricity and where power charging stations are located in the city. Also, things like pick-up and drop-off times play a critical role in this. We developed a highly integrated vehicle ecosystem simulator that allowed us to take all of this information and then utilize how the vehicles actually perform in terms of overall fleet and energy performance.

If we take a look at this particular figure, you'll see the little green dots in the City of Austin. These represent an, actually, a model fleet of electric vehicles. The green dots represent passengers that are in the vehicles, the red dots are empty, the black dots are idling, and the orange dots are charging. As we can see as we run through a particular period in time here, the vehicles drive around the city and go through all of these various stages, but a higher-level control algorithm is actually optimizing the total energy use across the entire fleet.

Next, we'll take a look at how these start to integrate together through autonomous energy grids. One of the things we really wanted to do was understand a really large-scale deployment of distributed resources, so we took as an example the San Francisco Bay area. This is a very complex distribution-level grid with over 10 million electrical nodes. If you think about how different distributed resources could be integrated into this system, you have electric vehicles, charges, building loads, local PV. If each customer had all of these devices, you could be talking 10 to 20 million controllable devices on the grid just inside the San Francisco Bay area itself.

As we looked at how to take the power system and do some of this distributed optimization, we took the natural strategy of looking at large, complicated distribution circuits, as you can see on the left-hand side, and breaking that into smaller areas, then doing optimization and control on that, and then bringing those solutions back together. On the right-hand side, you can see that if we even went to smaller-scale cells, it actually provided a much higher-speed improvement in terms of doing the optimization while maintaining the accuracy.

So, let's look at how we pulled all this information together across the San Francisco Bay area. We developed a large-scale simulation platform that allowed us to evaluate hundreds of thousands of devices. In this case, around 400,000 PV systems, shown by the yellow dots; around 400,000 building loads, shown by the blue dots; and then 10,000 electric vehicles; and around 200 charging stations.

This particular figure is showing the simulation running through time, so around 2 hours compressed into several seconds. But we were able to develop a complex multi-domain energy simulation of the San Francisco Bay area with millions of controllable assets running distributed hierarchical controls at around 1-second time steps. This demonstrated, at least in simulation, that we could deploy these large-scale distributed control algorithms on extremely complex systems.

In addition to the large-scale simulation, we wanted to demonstrate that these types of control algorithms would work on actual operating devices, so we set up one of the largest hardware-in-the-loop experiments ever at our Energy Systems Integration Facility, where we tied together 100 different operating devices—from electric vehicles, to energy storage, to PV inverters and loads. We then deployed these distributed-control algorithms and looked at how they operated across this large-scale simulation and experimental platform.

In addition to our laboratory experimentation, we wanted to demonstrate that these control algorithms could be used in the real world. Working with one of our local utilities, Holy Cross Energy, we were able to set up these control algorithms on a new subdevelopment of smart homes that included photovoltaic arrays, electric vehicles, local energy storage, and smart home controllers. The use of these distributed-control algorithms show that these can improve overall energy efficiency and optimization in the real world.

So, let's tie this all together with the concept of autonomous energy grids. As you can see from the figure, we use centralized control, represented by these blue rings. We get a voltage profile that isn't optimized. If we look at ways that we can take the entire grid and operate it as coordinated cells, we can have these cells communicate with each other and better optimize a voltage profile. This is important when we look at integrating large bubbles of distributed resources.

We are continuing to do research on autonomous energy systems. This includes ideas like resilient science, where we're looking at how to improve the ability of energy systems to respond to disruptive events and recover more quickly using distributed controls. We're also looking at how to take a bottoms-up approach to integrating buildings and transportation infrastructure through autonomous urbanization. Both of these ideas are continuing improvements in autonomous energy systems.

I'll wrap up with a set of references that you can look at for further reading. And with that, I'll say, thank you, and if you'd like more information, please visit this website. Thanks.