 Great Yeah, thanks for the introduction. My name is being one. I'm a research scientist from Florence broken national lab for survey Yeah, hope the orders of our has been treating everybody. Well, I've heard it's at the very end of the semester Hopefully guys are back with a lot of things One of them I heard is that it is strong at most fair Entrepreneurship, so I hope the progress is off also, you know going well to to to be a millionaire Yeah, that's a joke, but let's take a millionaire more seriously the near term Yeah, the topic I would say I went to the lab getting to this business of electric vehicles Can you give us just a date? When your lab got into electric vehicles Well, I think the the date for my lab gets into this business was they did like decades ago Even before I joined the lab They used to be the electric vehicle and great integration study in terms of v2g Techno-economic analysis as well as the hardware components Yeah, the topic Mentioned by she was about the heavy duty I was actually planning to include a little bit more stuff. So I just made this topic as integration of EVs in the electric grid systems You know to include the work for my Study like the one in UCLA how the EVs can be Controlled and managed with different kind of great conditions and in this Profit in particular, I'm also gonna include another work when I was a postdoc LBL like how to aggregate the EVs together as a virtual resources or virtual battery as people You know following it right now How can use that as a proxy to integrate and interact with the electric rail and finally is the heavy load part Which also emerging topic in transport and electrification area The key question is how we should and how we can integrate the medium and heavy duty into the grid So those are the topics that I'm gonna Talk about hopefully some of the stuff is old, but I hope it's So the big motivation for EV is that you know You can see it is a big thing now and the adoption is celebrating the past one or two years As you might have heard that you know for companies in the state of California has achieved more than 30% market Intracellation right look maybe the nearest one is the county of San Martel At the national scale there are roughly more than 100 counties with more than 10% So EV is everywhere is the reality that's happening right now the national transportation decarbonization blueprint Published by US DOE also suggests that more than 50% new light duty and more than 30% new medium heavy duty zero machine vehicles Will be sold starting 2030 to meet the decarbonization goal of the entire transportation sector if we look at this From the entire sector it is actually huge envelope of applications Not only the on-road light duty vehicle medium heavy duty vehicle But also the off-road vehicles number of vehicles such as maritime vessels aviation as well as those rail applications In California particularly for the medium heavy duty vehicles We have the advanced clean truck advanced clean fleet and the upcoming proposed rule of EPA Will also promote the adoption of medium heavy duty EVs in this area However, you know as we are on if we target one billion as the final ultimate goal and Usually the first mile of this journey is very challenging. That's why one million dollar You know is a cannibal and we have to make it a tenable Things are in good shape right now because as you can see the adoption of the EVs at different levels Actually making much progress in the past few years and months From the research perspective, I want this topic to help students and audience to have some insights from those following angles But first is that How can we make the EV charging sessions more predictable? unmanageable, right if things are happening all over the place and at the different times Things can be chaotic because EVs in my words live two lives one with the grid one with the transportation A lot of times they are characterized as mobile Mobile storage right grab grabbing energy from the grid and this Consume this energy with very complex temporal and geospatial complex cities And the second question is How we can operate those chargers in terms of cost effectiveness and reliability Reliability I mean we have to deliver the energy requested by each EV driver, you know before the departure time and finally Which is a topic that have been working a lot recently which is to You know project the charging infrastructure needs for medium and heavy duty vehicles a different You know geospatial scales in terms of county, state and the entire nation and the key, you know parameters in this equations are the type of chargers, quantity of chargers and critical locations before those chargers So those are the high level and motivation of the talk Then we'll dive into the very first piece of work that have been doing in this business when I joined DCLA As a graduate student, we actually build a system as shown on the slide It is a system with a lot of components where you can see EV drivers interact with the system with the mobile apps They can submit their Schedule preferences energy demand energy requests and this information will be taken by the we call it the EV Aggregator in this model, but I don't know if people still call it. Maybe they have a new name for Virtual power plant or virtual battery for EVs But the notion is that for this aggregator we have algorithms developed to compute for optimized the schedules and distribute the schedule to individual vehicles and the EV aggregator is also able to interact with other components in our synthetic micro grid system like a PV generations energy storages as well as the underlying building load But how the EVs can be controlled to achieve the system-wide benefits is one of the objectives of the study In particular If we look at the user side, right, those are the data We collected from each hundreds of real-world EV drivers We characterize each charging session with a number of parameters like when the session starts Basically when the EV gets plugged in when the EV leaves probably, you know after school you know around 5 p.m. for faculty members and And What we try to determine is the actual time the charging session starts and the actual time the energy You know got delivered to the vehicle, right? Different users have quite different behaviors. Some of them have very stable start time Right, some of them, you know start late now will stay late. So the duration is more More stable But some of them will stick to the, you know, 5 p.m. Rule that, you know, it's time to go then we have to go Right, and there's also similar patterns for the energy demand for charging session versus the duration The question is how can we represent this, you know behaviors and this dynamics into some decision models and come up with the reasonable Schedules for personalize the energy management. That was the research goal in my study So we come up with the kernel based approach Giving each data point a Gaussian kernel to represent the behavior and we aid them up to have probability distribution it could be multiple dimensions and For some some type of users tuning the bandwidth of the Gaussian kernel and can get you best personalized estimation accuracy and the next thing is Taking the, you know numerical model to represent the use of behaviors into this decision process Which follows the rule of model predictive control, right? In each time step the predictor makes a prediction and fit that Estimated the value as input to the optimization model an optimization model computes the conditional Optimal solutions and only implement the first element in the series and this first element is broadcast to the individual Fathers that gots implement. That's how we Adaptively solve this puzzle when the new EV arrives at charging station the whole time table will be updated. So the So the optimization will be able to take the new information and we run it and redistribute the control signals So that's how the system was implemented a long time ago At that time reinforcement learning and deep learning techniques were not so popular So I'm actually not sure how powerful this is as of now compared with Reinforcement learning deep learning techniques, but from the result you can see, you know, if we set the cost as a minimization objective, you know in some low price periods You know some charging load got shifted to to them when there's a renewable generation the solar panel on The rooftop of a building the EV charging load of tries to follow us the you know solar generation So as a result the total system-wide cost can be minimized Also, we can set the total flatness of the that the Aggregated a net load as of jet beam. So the idea is to how how can we Let the load as much as we can It can also be achieved by this methodology Here's a you know preliminary comparison Of the cost-effectiveness in terms of dollars per kilowatt hour delivered to the EV battery and the this method effectively reduced the cost delivered per kilowatt hour and You know this estimate another estimate is based on the outdated information You can see the payback time, you know by this lower average operation cost Then How to take this approach to the next level, right? We are dealing with individual vehicles individual drivers But in terms of scalability when we have more EVs aggregated the same well, how can we Move this org EV into the power system How can we operate them in terms of the multiple time periods and in terms of individual vehicles dynamics? so that's the extension of the previous work and This work, you know, it's also leveraging the real-world EV Itinerary information that we collected that from Location called Al Park in a CEC funded a project So in that problem scenario, we have a bunch of real-cause electric vehicles Underneath the office buildings with solar panel and you know, you know battery storage, etc And the idea is that we first aggregate them together based on their schedule constraints and demand as a virtual battery and Then participate into a number of different markets, right? For instance, if this thing happens in the utility grid, right? the energy charge will be the first consideration of cost and Some areas the demand charge is also a big thing big factor But in California particularly in this study we considered the PG&E territory There are a number of other incentives that we can bring onto the table. For instance, the you know, something called Accelerate service market proxy demand resource market time-of-the-peak day pricing and demand bid programs each program has Quite specialized the rule for resources to be able to join the market in the first place And if they join the market, they have to comply with their rule For instance, state in the market for a period of time So all these constraints are very complex to model. We've spent a lot of efforts Just to make sure the comprehensive formulation covers all these cost factors and these constraints can be properly represented it turned out that this the this ended as a very giant mix integer programming problem Yeah, so you know the complexity We leverage some solvers. That's pretty efficient solving the problem on behalf of the aggregator So this is how the aggregator looks like based on some of the previous study Then for instance the the number of EVs, you know We want to use the Apple energy bonds versus lower energy bonds to characterize this EV battery The blue curve is the scenario where all the EVs charge as soon as they can as soon as they arrive at the location So this is a fasted ramp ramping up curve Right and the red curve is for each EV to start charging as late as they can As long as they can get the energy to fill the departure So the optimization objective is to find the optimal trajectory in between eventually to reach the, you know Total energy state for the whole EV boot So the the problem turns out to be solvable We can see, you know, pretty significant cost savings or running revenue generations by considering these market opportunities And we also tested some different sensitivity Some different knobs for instance the flashboat who participate in the market For instance, you have to have certain amount of resource that you can commit To join the market, right? If I have a phone battery with like five Watts, that's too small But I have a number of EVs with like a 500 kilowatt power that I can commit for certain periods Then I'm eligible to participate in those programs So those are modeled and the revenue are, you know, summarized here Another big factor we find is the flexibility And this flexibility is in terms of time flexibility that means If I can park there for a number of hours and, you know, the maximum or the fastest charging can only make like You know 20% of the parking duration That means I have a lot of flexibility To shift my charging to other time periods or change my charging power to achieve different objectives Right, so we tested different flexibility threshold to see how You know, how the cost performance looks like So these are the results you can see that more flexibility more lucrative But it's not the, you know To some level it will stop increasing because there's a threshold of the, you know Total energy needs by that bill as a construct um This part is addition to this work because so far we have been talking about at the things that the aggregator But once the aggregator gets the energy commitment How can the aggregator disaggregate and distribute the energy to individual vehicles? And in reality, as we, you know Had a lot of experience In the real world settings, you know, for instance, the communication is not very robust We lost data packet and some of the eb drive some one of the eb is failed to communicate There's a data packet lost something like that How can we make the decentralized the optimizing or decision making more robust Uh in the context of the, you know, market Opportunities, right? So in this case, we developed a synchronized Decentralized the eb charging algorithm You know, each eb call communicate with this aggregator while solving its local optimization problem This is based on the adn methodology However, by tuning the step size and, you know, Search direction for each eb It is possible to come up with a robust solution if some of the eb are offline for further periods That's still the convergence to the global optimality is still achievable in this case Um, if you are interested, feel free to take a closer look and Um, the still, um, in my opinion has a lot of realistic meaning in the eb implementation If we are really seeking the optimality of the whole system Um, then fast switching from The individual eb control The aggregator control decentralized control to the next topic, which is the, you know, electrifying the medium and heavy duty vehicles So this the work that have been focusing on for the recent years um, the big motivation is that ACT, ACF and the EPA rules are projecting a lot of medium heavy duty ebs on the road in the future and LBL My team was tasked by, uh, California Energy Commission for their assembly of bill 2127 assessment So this work, um, was focusing on the projection of infrastructure needs in terms of the quality high ground location of those chargers for for the future electric trucks, including buses Um, here is the snapshot of the, you know, uh adoption Um, it looks like based on 83 scenario, which is, uh, um, you know, some remodeling of the ACP ACF rules on top of the cap cap data and We are focusing on the vehicles with more, um, with gross vehicle weight more than 10,000 pounds And these vehicles, um, have a wide range of applications. For instance, the drage is long hauled twice Right and also the regional, uh, vocational vehicles As well as the transit buses um That does not exist a such model that comprehensively, you know, model the universal medium heavy duty at the time where we are doing this work and I will show, uh, in the next slide that how this process is carried out Maybe you have been a little bit familiar with this process in some other locations, but I want to emphasize that this workflow um Really starts from some P inputs as the most important one That's a technology Yeah, as the most important one is the travel demand model It's a probable probabilistic model telling us how many trips are coming from one part On the map as for trip origins to other parts of the destination So it's a probabilistic model. Um, that we can synthesize The trip distributions based on how many vehicles we are considered in the system And the second thing is the charge gps location data, which includes the trip start time trip duration trip distance All these statistics that are coming from the real world The synthesize the travel demand and trip volume We have to make them share the some statistical pattern of the real world data. That's one of the motivations there Uh adoption piece is very important. Um, this is super interesting to the policy makers in different scenarios There are different composition Compositions of vehicle types in the scenario. There's also a different vehicle volume The vehicle stock a different year. So all these parameters should be synchronized and the synthesize should be harmonized with the travel demand model tool Um, then the next part is agent based the simulation Which is the core of the heavy load tool which takes the travel demand model at the trip level and resolve the integrated the activities like how the vehicle drives arcs Charges and routes itself if you know, if the vehicle finds the energy Very low in the battery that and it cannot support the tool trip right the vehicle is has to do its own decisions to find the Optimal or near rest or shortest distance shortest travel time location in its original route So we are trying to mimic how the you know truck drivers drivers do in the reality As a result, you can see the trap You know the trial the charging demand and infrastructure needs can be quantified at different geospatial scales in term For instance at the county level at the traffic analysis long long level or we can also quantify Um, the charging events at the location level by location I'm for instance, uh, this example shows the truck stops We found in the state of california Um, if the truck stop is really in a popular area, it can be selected by a lot of trips as a wake up So it's a natural Instinct that a lot of trucks can stop by here with some energy demand It's a also indication that you know infrastructure can be needed in this location to support the charging demand And the next thing we did you're through uh CDC's edge tool By adding the simulated the charging demand charging load on top of the circuit base load And compare the new load profile versus the circuit capacity that age data can cover The hope is that the comparison can be very informative to the utility companies Saying okay in this area. There's a huge transportation charging demand But your circuit capacity is not enough to support the demand It's a good indicator for you Action requests or transformer or feeder upgrade, etc um Here's a snapshot of those, uh, you know, uh, vocations and the gps data represent in in a model um, we really appreciate the data set from the University of Los Angeles Virginia University that we leverage to characterize the uh applications and vocations in the model um It's a a little bit more explanation of the our charger selection algorithm that I just mentioned um Even the origin and destination of our treat There's an assumption that you know, there could be a depot chargers needed at both ends of the treat Because we assume, you know, it's either loading unloading or parking at the public local warehouse or distribution centers Or it returns to the depot at the end of the trip So the depot chargers are needed at those places However, when the vehicles are on the road each road segment We allow these vehicles to evaluate their situations by looking at the energy state of charge See, okay, if the energy state of charge is Below third and slash fold, so a threshold, right? It's reasonable for us to look for the, you know, best chargers along the way so that we can You know, still finish the trip, but avoid the range anxiety um This is a simplistic way to to do the charger selection for trip based the travel demand model but in the current practice The data we received from our partner has more granularities Not only one trip to represent one vehicle We can actually Represent one vehicle with multiple trips over multiple days. It's more closer. It's closer to the reality because Work was made multiple trips multiple stops during the day and the decision making about which stop to charge Assigned chargers is getting more challenging So a lot of criterias such as the most cost from location The duration of stop and the arrival time of those trips at these locations are considered to characterize these locations um, this is an ongoing work, um, but Um, this that those are our mindset to further refine the location based the energy analysis Uh, we got the opportunity to extend the california work for the Assembled For to the national scale Analysis, this is the work is funded by the u.s. Department of energy real-time technology office um, you can see the process is very similar, but the scalability is huge If you take a look at the travel demand model it covers the entire nation and we took national scale um scenario as an input to tell us how many vehicle stops for different vehicle vehicle types and we you know cross validate it with the uh, you know the truck od data from national renewable energy lab and also the vehicle registration data so that we can you know, um Calibrate the volume of trips starting from one particular location to other location and at the state level the statistics still match the those from the From from the adoption scenario In the simulation each one of this od is routed and the routes are computed in terms of shortest the travel time shortest the distance and if you can You know if you can see the green versus Red routes those are two different kinds of routes we consider one maybe the The this is an example for the new york. Maybe the red one is the depot based routes which starts directly start from the state of new york and And also or or they ended at a state of new york, but the green one is more complex um It doesn't start from from the state or it doesn't end, but it just passed through the state That doesn't You know That was not counted as the default charging trip, but it shows some opportunity for the public interrupt charging right if it It is a common scenario that um for the state Which is popular location for trips to stop by even though they are not originated from them But you have a lot of You know carried over emissions and the energy demand in your territory So that's the dynamics we want to uh, you know capture in this model As a result we can also project the load at the county level and also the infrastructure needs In terms of people chargers versus public chargers um I don't know if this is visible, but the public charger demand over the counties aligned pretty well with the freight corridors Which is always makes sense right? um And also we can quantify the truck volume over load segments as another by by product of the simulation Just trying to make the uh make it more fun by bringing some cartoons Those are the activities being simulated. Um, it's just the moving vehicles Where they are at different time of the day It was not perfect, but it shows you know different types of vehicles gone. How they got model um This one is the charging uh depot charging load Um at different time of the day as a result So you can see some concentrations in those hot areas But another layer actually, um is um the intensity of the load at different locations that was not not visualized here um similar scenarios, but slightly different assumptions made for the hydrogen fuel cell vehicles um One key distance distinction is that we don't assume the default refueling location Refueling infrastructure for hydrogen fuel cell vehicles So more energy was shifted towards the public infrastructure Um, this is the animation of the simulated the refueling demand um Of hydrogen fuel cell vehicles, um, aggregated the overall those uh truck stops Interestingly, you can see the magnitude relative magnitude of demand at different locations And also you can see how the demand slowly Um shifted from the east coast to to the west coast um Yeah Uh with the model, um we the the the the goal for us is to provide a tool To test drive different scenarios as you can, um, you know See a lot of parameters a lot of key assumptions um, you know, are quite different Depending on who the decision makers are right, so we want to make it as a convenient platform for users to test drive different scenarios and assumptions so, um, here's an example that if we um in this result if we compare the load, um, you know Low shape at different, uh, location scales For instance, the first one is at the entire National scale and then if we zoom into the state of california and zoom into the, you know, county of san Bernadino The load shapes are different And also the smoothness of those low curves are different And the the more granular that your special scales got, um The spikiness the more spikiness we will observe Right because in this scenario, we consider not only different applications and also The supply of different type of charges with a maximum charging power up to, uh, one magma Well, I go on, you know in in some occasions, uh, like energy load of, uh, you know Neighborhood load is 100 households Because if you can imagine the turn on and off of this 100 households in your neighborhood, that's a huge impact to the grid Um, so through this, um, you know tool, we can see the composition of the load contribution from these chargers And how sensitive this load can be if we switch the low power charger to the high power charger As we what we want, um, what capability we want to enable the decision makers The next example Analysis is on the state of charge Um in the simulation, there's also a knob for us to turn like, um How much so see how much percentage of energy left in the battery when the vehicle starts their tree Right if we assume You know these vehicles start start the so c is randomly distributed or it's more Alligated to close to the level close to 100 right, which is you know by some assumption it is true Right the early adopters of the medium heavy duty EVs start their trade from the poll With overnight charging for the so c is could be relatively high close to 100 Right, but in some cases if it's not that case, the so this is a randomly distributed How the charging demand will change How you know how it will change in terms of the public The public charging demand versus the default charging demand So this actually gives us some kind of some uh insight For instance, um, if we assume there's a high so c compared with the base so c You know the curve the public charging curve got shifted a little bit Afterwards during the day But if we assume a randomly assumed you can see more aggressively people Because more vehicles, you know, um, they draw the energy So c levels below the threshold As the starting so c is relatively low than the previous right so you can see the the ratio between People entity in total versus the public engine total and also the shape Of the energy demand for different type of loads This will give um the utility companies. Hopefully a lot of insights when They should manage their distribution assets given the demand at the time There's just two examples. So a lot of work are still Um, you know in the process we're testing every possible now we can and we're still validating with our Um collaborators in this regard. So please stay tuned for and we definitely work on any future collaboration or insights into this Because this is really a new area for us to take baby steps Okay, it's pretty much Uh, to the end of my discussion, but this is the most exciting part usually for me to talk about the future directions The one that's most interesting to me is autonomous load management. There's a lot of Multiple dimensions in this question First is as we discussed, you know, we have been considering the temporal flexibility of EVs aggregators, but how about the geosubstantial flexibility? right If some locations have higher or more available city capacity for the EV charging Maybe it's opportunity to shift the load overall EV that way Right to take take advantage of the capacity so you can defer the billion dollar infrastructure upgrade Request, right So there's something we can't necessarily post temporarily and to especially and the other Angle as I was just chatting this with our colleagues the other day That if there is an option for every EV to do manage charging with one button as a do a smart charging Now the EVs in the future will be turned as energy trading robot Very similar to the You know stock market nowadays because the majority of the transactions are made by algorithms and robots Right. How do we trust those robots playing against each other playing against the system? So that's a really interesting area for us to look how can we develop market mechanism develop control algorithm that regulates the EV EV charging behaviors so that the overall system performance can be achieved But also the energy delivery to individual vehicles can also be satisfied Um, the next one is on the infrastructure side, um, you know, even though right right now we can be coming the, you know, initially the I wanted the allocation of chargers, but how can we do it? optimally Right. What are the best locations? Should we at least regret locations we should consider for the initial phase? Um, this makes me I think that's very similar to the real estate business Right starting from the key locations, but how can we identify them in the first place? Right. If some locations are identified, how can we assign the right number of chargers? And right type of charger in those locations so that optimally the system evolves in a way that's both cost effective to the public and the government but also It's reliable. It's It's they can meet the charging demand covers most of the trips charging demand, right? And there's also the battery electric. I'm 100 Yourself eb considerations that we have to consider this work is ongoing to Hopefully I'm gonna have more results to share in the near future um, the next one is also from my recent discussions colleagues that you know, as I mentioned there's so many different knobs in the scenario in one scenario, but policy makers and the public are more interested in more scenarios How things will look if I turn one of the long knobs, right? How the system performs differently How the results will look differently um, you know running one simulation using the high performance computer is quite But how can we leverage the existing runs existing simulations and all the existing simulator as a trainer or expert and extrapolate to the scenarios that have not been you know, comprehensively evaluated Can the generative AI be used for? large language model I'm just randomly thinking bendy directions and as you know, a natural extension of this capability and lastly You know, people have been talking a lot about the non-world and all for the electrification Each one of this application needs dedicated considerations in terms of duty cycle visibility cost-effectiveness cetera so in terms of research and you know Modeling there's still a long way to go I don't want to discourage people, but this we have you know Make a serious first step towards achieving the first one meaning go Um, I heard that first first part is very very challenging, but later on on the scale So, uh, that's pretty much what I have. Um, I think First thank thank you for having me here and there are really opportunity to communicate with You know faculty and kind of students. Yeah, let me know if there any questions. Yeah, how