Scalable Optimization for Mobility-Aware Coordinated Electric Vehicle Charging in Power Distribution Networks

arXiv:2604.11999 · eess.SY, cs.SY · Submitted 2026-04-13 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: We're looking at the paper: "Scalable Optimization for Mobility-Aware Coordinated Electric Vehicle Charging in Power Distribution Networks".

Dev: Here's the gist of the paper. Rapid growth in electric-vehicle (EV) charging demand is placing increasing stress on power distribution networks (PDNs), whose hosting capacity is often limited and spatially uneven <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

Rosa: Beyond demonstrating that coordination can help, this paper answers an open question that is central for planners: what is the maximal achievable benefit of EV charging demand flexibility from spatial and temporal shifting in reducing overload-driven distribution upgrades at a regional scale <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

Dev: We introduce MAC (Mobility-Aware Coordinated EV charging) to establish a credible upper bound, which entails rethinking charging flexibility around individual mobility, fusing travel itineraries with feeder-level hosting-capacity data, and solving population-scale optimization with spatio-temporal coupling to certified near-optimality <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

Rosa: Pruning redundant reasoning tokens The paper introduces MAC (Mobility-Aware Coordinated EV charging) to establish a credible upper bound on the benefit of EV charging demand flexibility from spatial and temporal shifting in reducing overload-driven distribution upgrades at a regional scale <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

Paper summary: Rosa: So, we're looking at this paper, "Scalable Optimization for Mobility-Aware Coordinated Electric Vehicle Charging in Power Distribution Networks." Basically, they're trying to figure out the biggest benefit you can get from EV charging flexibility—shifting when and where cars charge—to reduce those big power distribution upgrades that happen when the grid gets overloaded.

Dev: Right, so it tackles this central question for planners: what's the maximum gain we can achieve regionally by shifting charging demand spatially and temporally? They introduce this model called MAC, which stands for Mobility-Aware Coordinated EV charging, to set a realistic upper limit on that benefit.

Taro: What that means in practice is they are rethinking how we think about flexibility; instead of just focusing on recovering energy during a single charging session, it couples the entire travel itinerary with the actual capacity data at the feeder level.

Rosa: Exactly, and this coupling allows them to solve a population-scale optimization problem using spatio-temporal coupling to get a certified near-optimal result. It’s about making decisions that consider where you are going and how much power is available right there on that trip.

Dev: And they establish this upper bound by expanding the feasible scheduling by coupling charging decisions over the full mobility horizon instead of just enforcing per-session energy recovery, which is a key part of their approach to MAC (<ref:2604.11999#pg1>).

Taro: That expansion lets vehicles have more options, like allowing a lower departure state-of-charge from an intermediate stop if there's charging available downstream, which unlocks more flexibility when the grid is tight (<ref:2604.11999#pg3>).

Rosa: It’s computationally scalable because they use an iterated price response scheme where each round posts a locational-temporal price and gathers the fleet’s best responses in parallel (<ref:2604.11999#pg2>). They show that mobility-aware flexibility alone removes most of the overload, even better than fully coordinated session-based charging (<ref:2604.11999#pg2>).

Dev: I like that comparison, because it shows the power of considering the whole trip versus just optimizing a single charge; it suggests that mobility-aware flexibility is a much bigger lever for reducing overload than perfectly timed session-based charging (<ref:2604.11999#pg3>).

Taro: So, when you look at the optimization problem itself, they are minimizing grid-side cost and EV disutility while respecting those mobility constraints (<ref:2604.11999#pg2>). The social optimum is defined as the solution that minimizes that total cost subject to all those feasibility constraints (<ref:2604.11999#pg2>).

Paper summary: Rosa: That formulation is equivalent to a competitive equilibrium, meaning each vehicle acts like it’s minimizing its own disutility plus the charging payment under a posted locational price (<ref:2604.11999#pg2>). It frames the whole thing as a market interaction.

Dev: The core mechanism they use to find that optimal solution is an iterated price response method, which solves this coupled resource allocation problem by iterating a price against the agents’ best responses (<ref:2604.11999#pg2>). They perform this ascent on the dual function Q over the set Lambda using an entropic mirror scheme (<ref:2604.11999#pg2>).

Taro: The convergence analysis shows that under certain conditions, this ascent converges to the social optimum at a rate of Op one over kappa squared (<ref:2604.11999#pg2>). That tells us how fast we can trust the result when we're trying to find that absolute best solution.

Rosa: To handle the massive scale, they developed custom primal dual routines and batched solvers specifically for the dominant EV subproblems to get around the per-iteration bottleneck at population scale (<ref:2604.11999#pg3>). They even show tractability by solving a MAC that coordinates one point eight four million EVs across more than thirteen hundred feeders over a week using high-resolution mobility and grid data (<ref:2604.11999#pg3>).

Dev: That's impressive for the scale, but what about the real world? The empirical results show that MAC reduces feeder overloading across the Bay Area by orders of magnitude compared to unmanaged charging (<ref:2604.11999#pg2>). Under MAC, the total maximum violation across the system is eight point one megawatts, which is way better than twelve thousand four hundred forty megawatts under the ASAP plus scenario (<ref:2604.11999#pg2>).

Taro: So that gap of eight point one versus twelve thousand four hundred forty megawatts shows that leveraging mobility-aware flexibility through full charging coordination can almost eliminate the need for overload-driven upgrade needs at a regional scale (<ref:2604.11999#pg2>).

Rosa: They compared MAC against session-based flexibility, which only optimizes the timing within a fixed session energy, and they found that mobility-aware flexibility is actually the larger lever because it removes most of the overload that even perfectly timed session-based charging cannot address (<ref:2604.11999#pg3>).

Paper summary: Dev: And finally, they showed that coordination completes the remaining step by pricing where and when feeders are actually short, which brings those violations down to eight point one megawatts, another tenfold reduction compared to just session-based flexibility (<ref:2604.11999#pg2>).

Taro: The paper does acknowledge some scalability challenges though; they note that long-horizon planning for millions of different people over many alternatives is hard for many methods that work well at small scales (<ref:2604.11999#pg3>).

Rosa: They also point out the data perspective challenge, saying both infrastructure and capacity are scarce resources where drivers compete with each other, creating system-level congestion effects that don't happen in single-agent settings (<ref:2604.11999#pg3>).

Dev: Looking at what this means for us on the ground, the framework offers a locational price signal that can inform how you design managed charging programs for EV drivers or aggregators (<ref:2604.11999#pg2>). It validates that coordination can turn what looks like a systemwide capacity crisis under unmanaged charging into something much more manageable operationally (<ref:2604.11999#pg2>).

Taro: They also provide a certified estimate of the upgrade investment that coordination can defer for distribution utilities, which sets up a baseline for future studies to include more operational constraints or model driver behavior beyond cooperative drivers (<ref:2604.11999#pg3>).

Rosa: So, this paper, "Scalable Optimization for Mobility-Aware Coordinated Electric Vehicle Charging in Power Distribution Networks," presents a framework that moves beyond simple session-based charging to use full mobility data to coordinate EV charging across the grid. It shows that this coordination can defer a huge amount of necessary distribution upgrades, and it gives utilities a concrete estimate of how much they can save by implementing such flexibility.

Dev: Ultimately, the implication is that we might be able to convert what seems like a massive systemwide capacity crisis under unmanaged charging into something far more manageable operationally through smart pricing and coordination (<ref:2604.11999#pg2>).

Taro: It sets up a baseline for future work, but the paper itself flags that individual drivers might charge across a very wide set of locations, and the charging cycles often span weeks or longer, which adds complexity to the modeling (<ref:2604.11999#pg3>).

Rosa: That's what you take away: coordination converts a capacity crisis into an operational problem, and it gives us a tool—a locational price signal—that can guide how we design charging programs for drivers or aggregators (<ref:2604.11999#pg2>).

Conclusion: Rosa: So, we're wrapping up our look at this paper, "Scalable Optimization for Mobility-Aware Coordinated Electric Vehicle Charging in Power Distribution Networks." Essentially, they’ve figured out how to use a system that knows where you are and where you’re going to make charging decisions.

Dev: Yeah, it really boils down to taking those individual car decisions and weaving them into a big picture optimization across the whole power distribution network.

Taro: They did this by creating something called MAC, Mobility-Aware Coordinated EV charging, which sets an upper limit on how much benefit you can get from that flexibility when reducing those massive power upgrades we talk about.

Rosa: What I found most interesting is how they built that model; it couples the travel itinerary with the actual capacity data at the feeder level. It's not just about a single charging stop anymore, it’s about your whole trip.

Dev: And from a controls engineering standpoint, they tackle this population-scale optimization problem by using an iterated price response method. It iterates a price and sees how the entire fleet reacts in parallel to find that best possible outcome.

Taro: I like that they use an entropic mirror scheme for their ascent on the dual function; it's a specific mathematical way to search through all those possible charging schedules efficiently.

Rosa: The results are pretty striking when you look at the numbers. They show that this mobility-aware flexibility, even without perfect coordination, removes most of the overload that we usually see in these situations.

Dev: That's what’s important; they compare it to session-based charging and show that full coordination brings those violations down significantly more than just timing a single charge.

Taro: They also gave us a certified estimate of how much upgrade investment distribution utilities can defer by using this kind of coordination framework.

Rosa: It really shifts the view on this problem from managing crises to designing something that is largely manageable operationally for the grid.

Dev: It sets up a baseline for future work, but we still have to think about how these models handle things when the world misbehaves or when drivers aren't perfectly cooperative.

Yi Ju, Lunlong Li, Jingchun Wang, Scott Moura

University of California, Berkeley · Hong Kong University of Science and Technology

eess.SY, cs.SY

Submitted: 2026-04-13

Updated: 2026-11-06

Comments: 20 pages, 8 figures, 3 tables

License: http://creativecommons.org/licenses/by-nc-nd/4.0/

Importance score: 95/100

The gist: Rapid growth in electric-vehicle (EV) charging demand is placing increasing stress on power distribution networks (PDNs)…

Key concepts

MAC (Mobility-Aware Coordinated EV charging)
This scheme optimizes EV charging by considering the full travel itinerary of a vehicle rather than just single charging sessions. It coordinates energy allocation across the entire journey based on known mobility patterns and feeder capacity data to find an optimal, system-wide solution.
Iterated Price Response (IPR) Scheme
This is the computational method used to solve the large optimization problem. Instead of solving everything at once, IPR iteratively posts a locational-temporal price signal and collects the best responses from all EVs in parallel until a stable, near-optimal solution is reached.
Social Optimum (SO) vs. Competitive Equilibrium (CE)
The optimization aims for the Social Optimum (SO), which minimizes total grid cost and EV disutility. This is equivalent to a Competitive Equilibrium (CE) where each individual EV acts selfishly, minimizing its own cost plus payment under a posted price signal.

Terminology

Summary

Rapid growth in electric-vehicle (EV) charging demand is placing increasing stress on power distribution networks (PDNs), whose hosting capacity is often limited and spatially uneven <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0

Where the paper says it

Scalable Optimization for Mobility-Aware Coordinated Electric Vehicle Charging in Power Distribution Networks Yi Ju1 , Lunlong Li2 , Jingchun Wang1 , and Scott Moura1 Abstract—Rapid growth in electric-vehicle (EV) charging demand is placing increasing stress on power distribution networks (PDNs), whose hosting capacity is often limited and spatially uneven. Beyond demonstrating that coordination can help, this paper answers an open question that is central for planners: what is the maximal achievable benefit of EV charging demand flexibility from spatial and temporal shifting in reducing overload-driven distribution upgrades at a regional scale?

Beyond demonstrating that coordination can help, this paper answers an open question that is central for planners: what is the maximal achievable benefit of EV charging demand flexibility from spatial and temporal shifting in reducing overload-driven distribution upgrades at a regional scale <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0
Where the paper says it

Scalable Optimization for Mobility-Aware Coordinated Electric Vehicle Charging in Power Distribution Networks Yi Ju1 , Lunlong Li2 , Jingchun Wang1 , and Scott Moura1 Abstract—Rapid growth in electric-vehicle (EV) charging demand is placing increasing stress on power distribution networks (PDNs), whose hosting capacity is often limited and spatially uneven. Beyond demonstrating that coordination can help, this paper answers an open question that is central for planners: what is the maximal achievable benefit of EV charging demand flexibility from spatial and temporal shifting in reducing overload-driven distribution upgrades at a regional scale?

We introduce MAC (Mobility-Aware Coordinated EV charging) to establish a credible upper bound, which entails rethinking charging flexibility around individual mobility, fusing travel itineraries with feeder-level hosting-capacity data, and solving population-scale optimization with spatio-temporal coupling to certified near-optimality <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0
Where the paper says it

Beyond demonstrating that coordination can help, this paper answers an open question that is central for planners: what is the maximal achievable benefit of EV charging demand flexibility from spatial and temporal shifting in reducing overload-driven distribution upgrades at a regional scale? We introduce MAC (Mobility-Aware Coordinated EV charging) to establish a credible upper bound, which entails rethinking charging flexibility around individual mobility, fusing travel itineraries with feeder-level hosting-capacity data, and solving population-scale optimization with spatio-temporal coupling to certified near-optimality.

Pruning redundant reasoning tokens

The paper introduces MAC (Mobility-Aware Coordinated EV charging) to establish a credible upper bound on the benefit of EV charging demand flexibility from spatial and temporal shifting in reducing overload-driven distribution upgrades at a regional scale <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. This involves expanding feasible scheduling by coupling charging decisions over the full mobility horizon instead of enforcing per-session energy recovery <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0

Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

The MAC scheme is computationally scalable via an iterated price response (IPR) scheme where each iteration posts a locational-temporal price and collects the fleet’s best responses in parallel <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

The resulting best-case benchmarks show that mobility-aware flexibility alone removes most of the overload, outperforming even fully coordinated session-based charging <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

How the training signal is built

The optimization problem seeks to minimize grid-side cost (total feeder violations) and EV-side disutility while respecting mobility constraints <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. The social optimum (SO) is defined as the solution to a minimization problem subject to charging profile feasibility constraints <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0

Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

This formulation is equivalent to a competitive equilibrium (CE) where each EV minimizes its disutility plus charging payment under a posted locational price <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

Where the gains come from

The core mechanism for finding the optimal solution is the iterated price response (IPR) method which solves a coupled resource-allocation problem by iterating a price against agents’ best responses <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. This ascent is performed on the dual function Q over the set Λ via an entropic mirror scheme which is better matched to the geometry of the peak violation dual <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0

Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

The convergence analysis shows that under certain conditions, this ascent converges to the social optimum at a rate of Op1/kq <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

The paper develops custom primal–dual routines and batched solvers for the dominant EV subproblems to remove the per-iteration bottleneck at population scale <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. The custom solver for the best response (D1) exploits the structure of Xi to accelerate it while preserving certifiable primal–dual bounds <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0

Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

This approach demonstrates tractability by solving a MAC that coordinates 1. p. 2
Where the paper says it

To remove per-iteration bottlenecks at population scale, we derive custom primal–dual routines and batched solvers for the dominant EV subproblems. 3) We demonstrate tractability by solving a MAC that coordinates 1.

84 million EV charging across 1300+ feeders over a week using high-resolution mobility and grid data <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

The empirical results show that MAC reduces feeder overloading across the Bay Area by orders of magnitude relative to unmanaged charging <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. Under MAC, the realized system-wide total max-violation is 8. p. 1

Where the paper says it

The iterated price response (IPR) method solves it by alternating fleet-wide best responses (D1) with price updates (D+). Results: PDN upgrade needs (total max-violation) across the flexibility and coordination spectrum.

1 MW (the IPR deployed response) compared to 12,440 MW under ASAP+ <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

This gap indicates that leveraging mobility-aware flexibility through full charging coordination can nearly eliminate overload-driven upgrade needs at the regional scale <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

The paper compares MAC against several baselines organized along two orthogonal axes: how charging flexibility is modeled and how coordination is used <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. The session-based baseline FLEX optimizes only the intra-stay timing of a fixed session energy while MAC additionally reallocates energy across the itinerary <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0

Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

The mobility-aware flexibility is shown to be the larger lever as it removes most of the overload that even perfectly timed session-based charging cannot <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

The paper concludes that coordination completes the remaining step by pricing where and when feeders are actually short <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. By pricing where and when feeders are actually short, MAC brings the violation down to 8. p. 0

Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

1 MW another tenfold reduction compared to FLEX <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

The ordering of policies in Table I remains unchanged at every resolution when comparing hourly vs sub-hourly solutions <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

The work provides a certified estimate of the upgrade investment that coordination can defer for distribution utilities <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. The framework and results set up a baseline for future studies to incorporate more operational constraints or model driver behavior beyond cooperative drivers <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0

Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

The work also sets up a baseline that further studies can refine in three directions including incorporating more operational constraints or modeling driver behavior beyond the cooperative assumptions <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

The framework offers a locational price signal that can inform the design of managed-charging programs for EV drivers or aggregators <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0

Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

The paper validates that coordination converts what appears to be a systemwide capacity crisis under unmanaged charging into a largely manageable operational problem <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

The results show that 83% of feeders would require upgrades under ASAP+ but only 2. p. 1
Where the paper says it

Under the projected EV growth trajectory and unmanaged EV charging behaviors, Li et al. [8] estimate that approximately 50% of feeders in California by 2035 and 67% by 2045 would need upgrades, totaling about 25 GW in capacity expansion at an estimated cost of 6-20 billion US dollars.

1% under MAC <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

This demonstrates the great potential of eliminating overload-driven upgrade needs at the regional scale by coordinated schemes <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. p. 0
Where the paper says it

Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations. NOMENCLATURE Acronyms ADMM: alternating direction method of multipliers CE: competitive equilibrium DER: distributed energy resource DR: demand response EV: electric vehicle HC: hosting capacity IPR: iterated price response KKT: Karush–Kuhn–Tucker (optimality conditions) MAC: mobility-aware coordinated (EV charging) PDN: power distribution network SO: social optimum SOC: state of charge Sets & Indices Symbol Description i P rrIss Index of EVs; I = total number of EVs t P rrTss Index of time slots (hourly); T = horizon length s P rrSss Index of distribution feeders; S = total number of feeders Is,t Set of EVs within feeder s service region at time t 1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

The paper provides a certified estimate of the upgrade investment that coordination can defer for distribution utilities <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. The framework offers a locational price signal that can inform the design of managed-charging programs for EV drivers or aggregators <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. The paper validates that coordination converts what appears to be a systemwide capacity crisis under unmanaged charging into a largely manageable operational problem <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

The paper provides a certified estimate of the upgrade investment that coordination can defer for distribution utilities <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. The framework offers a locational price signal that can inform the design of managed-charging programs for EV drivers or aggregators <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley. The paper validates that coordination converts what appears to be a systemwide capacity crisis under unmanaged charging into a largely manageable operational problem <ref:1Yi Ju, Jingchun Wang, and Scott Moura are with the Department of Civil and Environmental Engineering, University of California, Berkeley.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Mobility-aware Coordinated EV charging (MAC) can eliminate overload-driven upgrade needs by coupling charging decisions over a week long mobility horizon, as it almost eliminates overload-driven upgrade needs relative to unmanaged charging in a future scenario for the San Francisco Bay Area.

  2. The system can implement an iterated price response (IPR) scheme, which is deployable by design and carries a certified anytime bound on global optimality, allowing planners to obtain a certified estimate of the upgrade investment that coordination can defer.

  3. The AI system can utilize custom primal dual routines and batched solvers for EV subproblems to remove the per-iteration bottleneck of solving millions of best responses, leading to a significant reduction in solve time compared to standard distributed optimization pipelines.

  4. The system can perform robust feasibility checks using Theorem 5, which allows it to verify if a charging profile is feasible by checking Bi equals Bi, providing a certifiable way to ensure scheduling constraints are met.

  5. The AI system can leverage the dual function Qpλq as an anytime optimality certificate, allowing for a practical stopping criterion based on the relative optimality gap, which is defined as Jpprq minus Qpλq.

  6. The system can use an adaptive optimizer like Projected Adam with active-set masking to solve the per-EV best response problem (D1) efficiently, achieving convergence substantially faster than vanilla projected gradient ascent.

  7. The system can accurately predict and manage feeder load by solving the grid-side subproblems (S2) exactly using a sortand-search procedure in time complexity OpT ln Tq, ensuring that the grid update step is negligible relative to the main EV optimization loop.

Abstract

Rapid growth in electric-vehicle (EV) charging demand is placing increasing stress on power distribution networks (PDNs), whose hosting capacity is often limited and spatially uneven. Beyond demonstrating that coordination can help, this paper answers an open question that is central for planners: What is the maximal achievable benefit of EV charging demand flexibility from spatial and temporal shifting in reducing overload-driven distribution upgrades at a regional scale? We introduce MAC (Mobility-Aware Coordinated EV charging) to establish a credible upper bound, which entails rethinking charging flexibility around individual mobility, fusing travel itineraries with feeder-level hosting-capacity data, and solving population-scale optimization with spatio-temporal coupling to certified near-optimality. (i) MAC expands feasible scheduling by coupling charging decisions over the full mobility horizon. Instead of enforcing per-session energy recovery, it only requires the EV state-of-charge (SOC) to remain sufficient for upcoming trips. (ii) MAC is computationally scalable via an iterated price response (IPR) scheme. Each iteration posts a locational-temporal price, collects the fleet's best responses in parallel, and updates the price from the observed capacity shortage. Custom batched subproblem solvers remove the per-iteration bottleneck of solving millions of best responses. In a future-oriented 30% EV adoption scenario for the San Francisco Bay Area, MAC almost eliminates overload-driven upgrade needs relative to unmanaged charging. Comparing across the baseline spectrum, mobility-aware flexibility alone removes most of the overload, outperforming even fully coordinated session-based charging, and coordination on top suppresses most of the remainder. Both levers are thus essential, and the resulting best-case benchmarks provide references for PDN planning and operations.

Sources

Related papers