NM SMART Grid Center Webinar Series: Student Research Spotlight

Anne Jakle: Hi, everyone. We’ll give folks just a couple of minutes to arrive to the webinar. So hang tight. Hi, everybody. Welcome to the new mexico Smart Grid center webinar series. My name is Anne Jakle. I’m the Associate Director of New Mexico EPSCoR. And today I’m pleased to bring you in Mexico Smart Grid center student research spotlight where graduate students from our project will be presenting on our various research goals. Just as a reminder, all of our webinars are recorded and archived on our website.

So you’ll be able to 00:45 - access this afterwards if you want to revisit any of the information. And you can use the question and answer box in your zoom interface to type in your questions. We will have a q&a session at the end of all of the presentations after they’re all completed and so on. If you have any questions at any time, though, you can type them in there and we will address them later. Our next webinar will be on April 22, also at noon to 1pm.

And Kevin Tomsovic who’s the Director of CURENT which is 01:15 - the Center for ultra wide area resilient electric energy transmission networks will be presenting. This is a complimentary center to the New Mexico Smart Grid center where they do research on some similar questions. And in his presentation, he’ll be emphasizing the need for developing new test beds to better understand grid modernization. So we hope you can make that too and you can access the link and download a calendar, a calendar feature on our website. Before we start in today, I wanted to just give a little overview if there’s new students or those of you who are who are coming to our project maybe for the first time, just the the general context.

So the New Mexico Smart Grid center is 02:00 - a five year project funded in 2018 through the National Science Foundation EPSCoR program that establishes an interdisciplinary research center to address the design operational data and security challenges of next generation electric power management. So the approach of this project is to look at how we can transform existing electricity distribution feeders and turn them into a set of interconnected distribution feeder micro grids. I know some of you have a background in power systems and some of you don’t. So we’ll just take a second to define these two terms. The distribution system is the final stage of the electricity delivery system where electricity is stepped down in voltage from the transmission system and delivered to residential, industrial and commercial users.

02:42 - And a micro grid is also a discrete entity, and that it’s a group of interconnected loads and distributed energy resources that acts as a single controllable entity with respect to the electricity grid. That is it can connect and disconnect from the grid and operate independently. So we’re looking at taking Existing distribution feeders and overlaying micro grids.We are organized into four research teams and research goals. The first is the architecture group or RG1 and they’re looking at optimizing this distribution feeder micro grid or DFM design, incorporating human preferences. RG2 is networking.

And they’re looking at creating new DFM 03:24 - networking and communication systems that are scalable, secure, and protect user privacy. The third group is integrating machine learning data mining and knowledge based techniques to make computer aided and data driven decisions in the smart grid. And RG4 the deployment group is validating the proposed models and technologies and simulations, diverse test beds and New Mexico and representative of scenarios in the nation. And this is all encompassed by our cyber infrastructure which we have high performance computing resources at nmsu that are available to all project team members and also data archive and sharing resources. If you’re interested in those, I encourage you to check out our last webinar in the archive where there’s more information provided on how you can access the high performance computing and data management resources.

So this is just to say, we have 04:18 - a wide ranging project. It encompasses the three main research universities Santa Fe Community College, Explora, we have many faculty and students who are a part of this and you will hear from five different graduate students today. So first, I am going to introduce Jee Won Choi. She is a PhD student at UNM in the mechanical engineering department and currently advised by Dr. Ali Bidram. Her research interests are in power systems, especially the smart micro grid area and she’ll be presenting on areas of research related to research goal one - Architecture.

04:54 - Go ahead Jee Won Jee Won: Hi everyone. My name is Jee Won. I’m a PhD student at UNM and today, I’m going to present about the secondary frequency restoration of islanded micro grids using a decentralized event triggered and finite time control strategies. Next slide, please. Okay, so first, I’m going to introduce the concept of the purpose control strategy. Then I’ll go through the preliminaries of the current research and how the newly proposed technique is developed from the preliminary work. And then I’ll present the case studies of the proposed technique simulate using the four DG micro grid system model. Next slide please.

And these 05:44 - days micro grids are becoming the substantial components of the modern power grid systems. And as a result, there are a lot of ongoing researches on how to control this micro grids, and the hierarchical micro grids controls system is one of them. And it consists of the two major control layers, ones the primary and the secondary. And one of the roles of the primary control is to stabilize the system frequency and voltage when islanded and the other role is to share the power to the distributed energy resources or distributed generators of a micro grid based on their power ratings, so that the more loads can be assigned on the DG with the higher generation capacity. However, the primary control makes the micro grid frequency and voltage deviate from the nominal value. Next slide, please.

And this graph here shows how the DG frequency is 06:54 - determined as a consequence of the active power sharing So, this slope MP comes from the active power ratings of a DG. So, corresponding to the amount of active power p assigned to the DG, the frequency of that DG omega is determined. So in order to place this omega at the nominal value, the only thing we can do is to adjust the Omega star because P is fixed from the power sharing. So finding the proper omega star is the duty of the secondary control. And the similar process applies for the voltage restoration and reactive power sharing, but I’m going to only cover the frequency restoration and active power sharing.

And note from the second bullet item, that keeping 07:52 - MP times p equal in every DG is the goal of the power sharing Next, and decentralized control is applied to bring higher reliability and flexibility to the system. Next. So there are various decentralized secondary control techniques out in the literature. I won’t have enough time to go through the details, but notice that omega star dot is determined from the frequency measurements and the active power sharing. So, the conventional secondary control is advanced by applying auxiliary control techniques. For instance, the addition of the finite time control provides the shorter settling time or the shorter restoration time than the conventional control.

And the event triggered control 08:55 - reduces the communication burden by only triggering the communication between the distributed generators only if some condition is met. So finally, the goal of this study is to combine and apply those two existing techniques to the conventional secondary control in order to achieve both faster restoration time and the lower communication burden. And I’ll refer to them as the event finite time control. Next, so to test the proposed control technique, the four DG micro grid system is implemented in Simulink. So there are four DGs and there are two loads defined in the system and inside each DG the primary control is implemented. next.

in the first 09:58 - case, study We assume a fixed communication. So here each dotted line indicates the mutual communication between the two DGs. So each DG transfers its frequency and active power measurements to its neighbors through the communication link. And specifically for this communication graph, DG1 is set to be the only DG that has the access to the nominal frequency value. So that the rest of the DGs will be able to reach to that nominal value indirectly through DG1. next.

so the first 10:46 - and second graphs show the frequencies and active power ratios, which is mp times p of the DGs of the micro grid. So first, let’s look at the first graph. Then at zero second the micro grid is islanded from the main grid, so, the frequency becomes unstable, but soon we can see that the frequency becomes stable around 1.5 seconds. So, we know that the primary control is working, but the frequency is not at the nominal frequency which should be 60 hertz then the secondary control is applied at 2 second and the frequencies of the DGs are restored at 60 hertz around 2.7 seconds. And also in the second graph we can see that the active power ratios of all DGs are equal. So, we can verify that the active power is properly distributed to the DGs next And in the second case study, we test the response of the proposed control to the load variation.

And the same 12:09 - communication graph is used from the case one. And likewise, the micro grade is Island at zero seconds and the dg frequencies are restored after the secondary control is applied that two seconds and at three and four second the active power of load one and two changes and the system frequencies become unstable when the load changes, but shortly they are restored to the nominal frequency after by the secondary control. And we can also see that the active power ratios are kept equal after the loads changes Next, the third case study, we assume the time varying communication graph. So the graph changes from the left to the right every point 05 seconds. And in this case Did you want is the only digit that receives the nominal value.

So the second and the 13:25 - third graphs represent the instant but repeating failure of the nominal value acquisition in the system, because Digi one is not communicating with any DJs in those two graphs. Next, so, under the time varying communication graph, the proposed secondary control is able to restore the Digi frequences at the nominal value While successfully distributing the active powers to the teachers next And lastly, the proposed control is compared with the existing second grade thorough methods in order to show its throne points. So, these comparisons are made under the condition given in the case one, which is the fixed communication. The first event finite time control controls compared with the event triggered only control. So, the times taken until the frequency restoration in each case are found as shown in the slide and the new method provides almost three times faster response than the existing method.

The event finite time method is compared 15:04 - with the finite time only control and the average average communication time intervals in each case between two to 2.7 seconds, that is the transience duration of the secondary control are found as shown in the slide. And using the new methods, the communication burden has reduced yearly three times compared to the existing method. And let me conclude the presentation here. So, in this presentation, a decentralized event triggered and finite time sec country frequency restoration is proposed.

And it has been proved to be able to 15:52 - restore the frequency while successfully sharing the active powers among beaches and in addition It has been proved to have faster restoration speed as well as the lower communication burden than the event only or the finite time only controls. Anne Jakle: Thank you. Thank you, everyone. A reminder to Thank you. everyone on the webinar if you have questions for Jee Won. We’ll hold them till the end, but you can use the q&a box to type in your questions. For our next presenter will be Jacob Marks, who is a computer science master’s student at New Mexico Tech, working with Dr. Dongwon Shin. He’s studying how differential privacy can be applied to the smart grid and he’ll be presenting for research goal two, which is networking.

16:50 - Jacob Marks: Hello, I’m going to be talking about differential privacy in the smart grid. And I’m going to try and give a really basic overview not go into a ton of details but just try and get an understanding on what differential privacy is. So first, I’ll talk about what are some of the privacy issues that we faced with smart meters? What are some of domain privacy preserving solutions that we have? What is differential privacy? And finally, how can it actually be used in this market? So what are the privacy concerns? So, with the kind of fine grained data that we can get from smart meters, over older types of electrical meters, there’s a lot of concern about what kind of information people could get out of households with those meters. So studies have shown that it’s possible to identify the household occupancy of people with a smart meter, their economic status, and even their appliance usage. So some studies have been able to compare different electrical loads of appliances and identify which ones are being used at what time of day in a household with a smart meter.

Or even more amazing to me, in this one 18:05 - paper, multimedia content identifications through smart meter power usage profiles, they were able to identify what somebody was watching on their television, using just a small amount of data, using only five minute chunks of viewing was enough to identify a movie. Now, with this study, they were only able to identify it using smart meter data that was collected every point five seconds. It wasn’t real smart meter data. It was a simulation. And so that’s not something that can be done yet. But I think it illustrates the large privacy concerns that people have with smart meters. So how can we solve those privacy problems? So the two primary types of privacy preservation techniques that people are considering using for the smart grid.

One of them is cryptographic privacy that 19:01 - people are probably more familiar with. And another is statistical privacy. Differential privacy is a type of statistical privacy. And these are two different very different type ideas. So with cryptographic privacy, that’s the concept that if I want to send something to a friend of mine, both of us are allowed to see it. But we don’t want an adversary, some interceptor to read our data. So it would ruin our privacy if somebody was able to intercept the data. But with statistical privacy, legitimate receiver is the same as an adversary. So we want to keep ourselves private, even from a trusted data analyst. We still want our privacy preserved, which seems like a very difficult goal, and it can be difficult, but differential privacy seems like a pretty good solution to that. So differential privacy has a very specific goal a definition of privacy that I think is different from the way a lot of us think about it.

So there, the 20:07 - definition of differential privacy is that we want it to be very unlikely that an attacker can identify if we’re in a data set, we want to have plausible deniability. So if I’m in a data set, or if I’m not, it should be extremely unlikely that an attacker could detect that. Now, what this plot plausible deniability gives us is the chance that an attacker may suspect that somebody was in a data set, but it’s extremely unlikely that their results are actually true. So they can’t use it to damage somebody’s privacy. Now, that’s different from the actual definition, it’s more to the goal.

So the actual 20:51 - definition is on my next slide. So the definition is a little confusing in order to To preserve differential privacy, we need some form of mechanism that will modify our data set. So that mechanism could be adding random noise, or it could be a random selection of data. And differential privacy only works with an aggregate of the data. So if we have an individual’s data set, we can’t protect their privacy individually.

But if we have 21:24 - some kind of aggregate of a number of people’s smart meter data, we can protect their privacy. So in the definition, we need the probability that the mechanism applied to one data set, D1 being an element is set as the probability is less than or equal to e to the power of epsilon times the probability of the mechanism applied to the data set two being an element of S. So the important part there is the epsilon value. So epsilon is the privacy parameter for differential privacy. In the paper, I have a dream differentially private smart metering, I found with a really good definition that I think helped the modification of any single users data in the data set changes the probability of any output only up to a multiplicative factor e to the epsilon. So our privacy parameter epsilon, is defining how much privacy we have.

And an 22:27 - extremely low value of epsilon will guarantee higher privacy, while a high value of epsilon will guarantee low privacy. Now, why would we want to have less privacy? Well, the problem is that differential privacy is usually using some form of noise. And if we have perfect privacy and epsilon value is zero, then the data will be unusable because it’ll be essentially random noise, we won’t be able to get any useful information out of it. So most differential privacy solutions Use random noise added to the data set. And most differential privacy solutions for the smart grid use the Laplacian mechanism.

So the Laplacian mechanism was, I believe, 23:12 - introduced by Cynthia Duework, one of the people primarily responsible for differential privacy, and it pulls noise from a Laplacian distribution in order to try and add noise to the data. So, the noise is added to the data set based on what the epsilon value is. So, a lower epsilon value will mean more noise added. And it’s also tuned depending on what the data set is, what your values are supposed to be. So, a very important consideration when choosing differential privacy solutions for the smart grid is finding ways to choose a good epsilon value.

And there’s a lot of research that’s been done on 23:56 - ways to add noise ways to confirm differential privacy. But there is not a tremendous amount of research done on what epsilon values are usable in the real world. So that’s something I’m hoping to see more research on. Now, one problem with differential privacy is that you cannot guarantee people’s privacy if the individual’s data is compromised. So in this image from the I Have a Dream paper, it shows an individual’s electrical data over today.

And in the second image, it shows 24:31 - with random noise added, and only some noise has been added. And it’s quite obvious. Some of the peaks are exactly the same in both of these images. I don’t remember exactly how much noise was added in this example. But you can see that an attacker could still get useful information from this and somebody’s privacy may be compromised. So a number of differential privacy solutions for the smart grid proposed combining it with cryptographic privacy in order to ensure that An individual’s data isn’t compromised.

Because if we add 25:03 - noise to this, it’s still not differentially private until we aggregate it with other people’s data. And once we have the secure aggregate, we’re able to confirm differential privacy, and customers can be assured that their data is safe. So the problems that are still rather open and research is being done on them, but it’s unclear exactly which solutions are good as what kind of accuracy will actually be enough. How much privacy Can we really confirm for people while still having useful data for researchers and utilities companies also who will be trusted with the original data? Many different solutions use different privacy models. Some will have a trusted utility company while others trust nobody and trying to aggregate the data outside of any human interaction.

Other concerns are the speed with which this can be 26:02 - done. Can we add this noise or confirm cryptographic privacy at a fast enough rate to accommodate the high fine grained data of future smart meters? Also, the privacy and accuracy are concerns. So part of my work with Dr. Shin has been trying to look at the modern solutions using differential privacy, and try and find out which ones are actually good enough to use. So, in conclusion, there’s a lot of privacy concerns associated with smart meters. But there are also a lot of privacy solutions, cryptographic ones statistical solutions, dozens! Differential privacy is especially promising.

26:44 - However, it’s a relatively new solution and it needs more data on which differential privacy solutions will actually work best. So, my best references were the I Have a Dream differentially private smart metering. paper, that paper presents a very good solution that takes into consideration some of the problems with differential privacy, and also some of the cryptographic solutions. And since the works, work on differential privacy is really the foundation of it, and provides most of the information I’ve gotten on how differential privacy functions and how it can be used. Thank you. Anne Jakle: Thank you, Jacob. Great presentation. So our next presentation will be for research group three, which is decision support, and Adnan Bashir, who’s a PhD student in Computer Science at the University of New Mexico, working with Dr.

Trilce Estrada will be presenting on database 27:45 - that he has helped develop his primary research areas include applied machine learning, and he will be grabbing control of the screen, I think at some point to also do a demonstration so go ahead Adnan. Adnan Bashir: Thanks Anne for the introduction. Hello, everyone. Today I’m going to be presenting my research work, which primarily consists of two different projects. One is smart grid data generation. And the other one is how we can host the data and make it accessible for our group, and also for our public under the EPSCoR project. So, first of all, we have to, why do we need to synthesize smart grid data? There are many, many reasons and one of them is this is still like an image or field right now. And it’s still under development.

So we don’t 28:40 - get to see a lot of data that we can use for decisions of support in our project. And also like the people who collect this data they don’t really share it with, you know, the public and the data which is accessible by public doesn’t have like a lot of parameters which we need So that’s why we need to look for our framework or a software which can synthesize smart grid data that that is very realistic, and can help us in training our neural net for using it in machine learning. And also, there are like a lot of very good mathematical models up there, which we can use to generate synthetic smart grid data. So here are like some options which we can use. First one is Ganz, generative adversarial networks. It’s basically a neural net. And the problem with Ganz is that we still need real data to generate some new data.

So first of all, 29:45 - we should have a lot of like real data, then we can generate some new ones. So currently, we don’t have enough data. So I’ll skip on that and the next one is Mosaik. Mosaik is basically a framework that can add different simulators. And each different simulator can be like a generator or like a bus or like a hub or like a branch in the smart grid, and we can analyze the power flow through it. And we can then collect the readings. And the third one is MATPOWER. It is the primarily one I’m working on right now, it is basically a MATLAB language based framework.

And I will show you what kind of configuration 30:35 - parameters we can use and how we can use it to generate a lot of data and then use it in our project. Fourth one is Pypower, Pypower is a Python based software, similar to MATPOWER. But and it’s also been used with the Mosaik framework as well. So now I’ll explain what I’m doing with MATPOWER. MATPOWER was first affiliate launched in 2010. And a lot of people have been using this. And you can basically run it on MATLAB. But the problem is MATPOWER is that it gives you the configurations or the simulations. But when you have to create a big chunk of data and run simulations over a longer period of time, then you have to use some kind of scheduling software. And the MATPOWER comes with a software called MOST, which is MATPOWER scheduling software. And right now I’m in the process of figuring out how we can add these to and use MATPOWER to generate data and then see whether it’s realistic enough or we can use it for our decision support or not. So here’s a summary of like, configuration format, power. These are as you can see, on the left column.

We 32:00 - can select how many buses are there, how many generators you want and how many branches are there, you want any loss in the transmission lines and for what is the total capacity. And also, we can select different voltage levels active power reactive power. sizes, there are like a lot of many other functions and a lot of other parameters which we can tweak to make our Smart Grid more realistic. And besides that the results are like saved in a CSV file or MATLAB data structure. So we can also like import that data from there. So right now, I’m working on scheduling part in MATPOWER. So as soon as I get some data, I will share it with the growth and patch it with the MATPOWER and now I will Give a demo about the data repository which we have created for actual project. So now I’m going to switch over to my screen. So this is the main page. I’m currently I’m hosting it locally. And at some point, we will be given a machine with more RAM and with an actual physical machine or maybe a high power PM, on which this will be installed. And for the for the demo, I have hosted two data sets.

One of them is 33:36 - provided by Olga, I think she is at the meeting with us right now as well. This is a faulted data set from nmsu. And also there is another one, which is from Dr. Martinez Ramon, from ECU from at UNM. And so currently I have posted these two just to give a an idea so The main thing which we require from this data repository is a search field. So, for searching as you can see if I ECE both of these sets are related to ECE. So these two are coming up. And also if I type like this they are named for only one record shows up.

So this is based on see can which 34:24 - is pretty good in searching. And if we want to add a new one, we can also make an edit title from like a local local machine or also a URL. And then we can add some descriptions. We can add as many tags as we want. We can choose the license pointer, and we can have to the author emails and everything which is pretty common. And then we have three custom fields we can also increase the number of these custom t these these can be anything if someone wants to add a tag which is not already there. They can write the keyA d then then for the value f r that.

And even and this can al o be accessed publicly, if I if 35:10 - I just have a logout. And then f I go back to the website agai , I can see two data sets agai , because these two are availab e publicly. So, that was also o r goal to share that data with t e public. And I’ve also added a publication. This is a succe s publication. So, we can also a d like a sample data. This is a sample image of the image da a set, which is included in th s data set. And we also have li e a sample CSV file, we have lik , hundreds and thousands of C V files.

So, just if you can p 35:48 - t one as a sample and people c n view in the form of a graph as well like this, so, This also gives a nice idea before downloading, that’s waht this dataset actually looks like. So that’s it for the data repository thing. And once we deploy this on an actual DM, I will be sharing the tutorial with everybody in the group and with instructions on how they can download and share their and upload their own data sets to this website. Thanks. That’s it. Anne Jakle: Thank you, Okay, our our final presentation will be by Shuba Pati who is a PhD student in Electrical and Computer Engineering at New Mexico State University. And she’s working with doctors Satish Ranade and Olga Lavora.

She’s currently working on the 36:53 - resiliency considerations of the electricity grid in the event of natural disasters and she’ll also be joined by Rusty Nail who’s pursuing a master’s degree at nmsu and the same Research Group, a reminder to all if you have any questions for these presenters or any other ones, please use the q&a box as part of your zoom interface and go ahead shuba Shuba Pati: Thank you Anne. So, this is Shuba and with me there is Rusty and together we are going to talk about some of the research that we do at nmsu on power system resilience. So, what happens to the grid in case of a natural disasters like a tsunami, cyclone, tornado or hurricane or in case of a cyber physical threat. Part of the grid might get damaged, some of the lines may get disconnected, some of the generators might fail, or some of the command might get damaged, which ultimately leads to the shut down on the power system. For example, in 2012 In case of the Hurricane Sandy, there was a loss of $50 billion.

So, then that term resiliency comes to 38:08 - our mind. So, what is the resiliency, the resiliency is the ability of the power system to recover either completely or partially from the adversity is the definition of resilience. So, now, the resiliency is depend on the adaptability of the grid to unexpected failures. Here the term adaptability is similar to that used in biology, which is the ability of an organism to respond and survive in case of it and when under distress similar phenomenon is expected to happen in case of power system resiliency. So, now, what’s the basic difference between the reliability and resiliency? So, the goal of the resiliency is is to maintain the continuity of the service How do you maintain the continuity of the service, we make the infrastructure more redundant in order to make the system reliability.

Whereas, in case of the resiliency, the goal is to 39:13 - quickly recover either completely or partially to active management of the grid. One more key difference between the reliability and resiliency is the objective in reliability is to minimize the cost whereas, in resiliency The objective is to maximize the throughput. So, in order to do the resiliency study, we have taken IEEE 34 bus reliability system. As you can see in this picture, we have taken two cases and we have taken two critical loads at bus 19 and bus 20. So, in order to do the resiliency study, we need to define the objective function So, here the objective function is to maximize the throughput or minimize the mismatch between the generation and the total demand. Here.

In addition to that, we have also taken the 40:13 - constraint on generation as well as on the line capacity. In addition to that, we have also taken the constraint on the on supplying the power to the critical load at all point of time without any disruption. So, in case one, so, we have taken out, as you can see in the picture (what we are seeing on the radar) So, once we took out those parts of the line from bus 23, to bus 12 from bus 13, to bus 11, from bus 13 to 12. So when we took out those parts of the line, and then when we saw the optimization function, you see that there are some of the lines that are in red that gets overloaded and the line which are in green that gets under loaded. Here overloaded means, the load in the line is increased by 30%.

Similarly, we 41:09 - have taken another case (case 2) when we disconnect the line that are shown in case two, from 15 to 21, from 15 to 16, from 15 to 24, from 15 to 21. And then when we saw the optimization function, you can see in this picture, the lines that are in red that get overloaded and the line that are in green that gets unloaded. So, the story of the multiple resiliency test cases on a simulated model of a power system can differ about the changes required to improve the resiliency of the grid. Like in our case, as you can see, we identify the lines for which the capacity needs to be increased. Over to Rusty. Rusty Nail: Hi, everybody, my name is Rusty.

I’m extending 42:00 - what we’re talking about with resiliency, I’m working on a current project right now in Cordova, Alaska, we coordinate with several national research labs as well as some other universities. In the image on the right, that’s a picture of the town of Cordova. What’s fascinating about this place is it’s kind of a natural, micro grid. We’re currently in the process of transforming into a smarter smart grid. It’s known as a natural micro grid, because it’s naturally enclosed in its geological location, self sustaining, and they provide their own distribution network.

And so this provides a really 42:30 - awesome testbed for our university, our group, as well as these other national labs to take a look at, you know, how we can mitigate some other contingency issues when it comes to now potential national natural disasters, cybersecurity threats, especially as we get more of those sensing and monitoring systems. And that’s kind of been a really cool new frontier. And that’s the work that we’ll be talking about here in a second. So as you can see in the last slide, I’m trying to use these to kind of illustrate to you that you know, these are we can we can envision a lot of these natural disasters that have happened in the past. While they aren’t very occurrant, they do happen.

And so some of these you can, you can see, it’s 43:07 - pertinent to know or relative, it’s important to know where the rivers what potential things could happen like tsunamis, the avalanches, as well as mudslides and things of that nature. So when we when we’re considering these different contingencies, it’s important to note not only the type of generation but also the location of that generation, because that’s depending on the potential type of natural disaster that will affect that each of them differently, a hydro plant, located closer to these water resources will probably be affected differently in the case of a tsunami, or some type of mudslide coming down, then a diesel generator located elsewhere. And so those are the types of things that we have to consider, in addition to those growing concerns of cybersecurity threats. So my current job is taking that distribution system that we have that is in Cordova. And shown to you on the left side is the one line diagram and that’s a distribution network.

And the 43:57 - top portion you can see there’s orchids substation, that’s one of the diesel generation locations. And then the other two Humpback Creek substation, and Power Creek on the right, those are the hydro-substations, these are the different areas where we might be able to need to plan for contingencies as well as if you look at the other components within the system. Those are each switch gears in which case, these are where potential contingencies or issues, grid down failures and areas may occur. On the right, I’ve translated that one line diagram to a MATLAB Simulink system visible visual visualization that you can take a look at. This is the actual thing that I’ve made in Simulink.

We simulate this on a 44:36 - target hardware that is known as Opal RT. It’s a real time simulator that that allows us to do real time simulations with higher computational power stronger than that of just your what you regularly run on Simulink just on your own computer. So here I’d like to illustrate, you know, the potential scenario where we’re looking at some type of natural disaster that comes in and I’m for simplicity sake, I’m going to say okay, we’re going to take out this particular type of switchgear. In the simulation, I’m making this I’m simulating the switchgear as a bus 530. As a, just a compilation of breakers. In this scenario, we’re going to say that one of those breakers goes down or is open.

As you can see, they’re 45:19 - externally controlled. This is something we can use to maybe plan for these types of contingencies, it can be timed. And that’s great thing about using the simulator and Opal RT platform. In the case of this occurring, I would then go with into the simulation, either to one of the substations, there also be other measurement devices and whatnot, we’re still building part of this model. That would then if you look in the top right of the slide, you can see that’s the generator setup.

That’s the simulation that I have for it, we would 45:48 - then open up that scope and then I can see some of the implications of that particular incident. Putting in some other control mechanisms this goes towards the actual planning side of the resiliency and collaborating this with some of the other, we’re cooroperating this with some of the other simulations we have with the other groups, we can then develop a better holistic contingency plan. So again, in closing for my portion this I wanted to point out again, with, this…it’s an evolving process, we’re going towards the evolutionary point of this, that’s something that I’ve not had pointed out, we’re looking towards adaptability, you know, we have to consider some of these larger issues. And the big things with resiliency is how can we maintain those critical loads before, during and after these big issues? And it’s, it’s that’s the difference between it being reliable and resilient.

It’s you know, how quickly can we recover and adapt 46:36 - to these different senarios. Back to Shuba. Shuba Pati: Thank you Rusty. So, previously, as I have shown in the two cases, we have taken the load as constant load, but in future we are also exploring like working on the considering statistical distribution of the load and also we are going to address the time sensitive loads. For example, suppose in case of a hurricane and there is a train in there in Metro in midway. So, our first priority would be to bring the metro or the train to the station and then once that train is at the station and then we can like divert the power to somewhere else. So it’s like, how how to like manage the load based on the time time like, which is more important.

And first we 47:25 - need to assign the load of the model, the power line over there and then to the next critical load. Also, it’s like Rusty talked before, we are also working on the switch placement. And also we are working on the battery placement like where how quickly we can place a battery in a system. So that in case of any natural disasters, we have some backup and also like, what would be the size of the battery we are exploring in those directions. Thank you, everybody. Let us know If you have any questions Anne Jakle: Thank you, Shuba, Thank you, Rusty for that great presentation that really brings it together for the deployment part of our project.

I’ll remind everybody, so you can use the 48:13 - q&a feature or you’re welcome to use the Zoom webinar chat, feature two type in any questions for any of our presenters here today. please type them in. I will give you a second and let us know if you’re having any problems with your, your technology to if there’s a way that you could communicate with us. Right now. I don’t see any questions though. I do see a comment from Satish which says excellent presentations with two exclamation points, which I agree with. Brittney Van Der Werff: We do have one question. Anne Jakle: Oh, what’s the question? Brittney Van Der Werff: It’s from Rusty Nail to Jee Won,“Was the need for time varying communications to lower operators costs? You had mentioned that it reduces the communication burden” Jee Won: Okay.

So the time varying communication wasn’t 49:12 - like designed for the lower operation costs, but just to represent the communication links failure. So sometimes like communication may fail between the DGs. So, yeah, for that case study, I just wanted to show you guys the response of the purpose control to those communication link failure. Did it answer your question? Rusty Nail: Yes. Thank you so much Jee Won - appreciate it. Jee Won: No problem. Anne Jakle: Okay, great. Thank you. Okay, well, we will see you next month, April 22 at noon, link on our website and stay well everybody.

Bye 55:54 - Brittney Van Der Werff: You all didn’t get to see because I could see them on my screen but it wasn’t translating through zoom but this The reminder that our next webinar will be April 22, noon to 1pm. Kevin Tom civic, the director of the current Center, the NSF do e funded center doing complimentary research to our So please, please do come to that webinar. No questions have arrived. So I think we’ll turn us loose a little early. But please know that we’re here at F square to help you with any of your remote working needs. And for our project participants. We have zoom, we have slack help that we can do, we can set up Google drives.

We’re here to 56:33 - support you in any way we can. And we hope that everybody stays well and safe and socially distance. So we’ll see you on the internet again, sometime soon. We do have one question. .