The Surprising Math That Makes EV Charging 30% Cheaper — and Privacy-Safe
A network of three charging hubs in the Netherlands, wired together by direct-current cables and kitted out with rooftop solar and parked batteries, can cut the cost of charging electric vehicles by 30 percent and the emissions associated with that electricity by 25 percent — and it can do it without any one station handing over its private data. The trick, according to a team of engineers at Eindhoven University of Technology, is not to build bigger batteries or wait for a greener grid. It is to think of every car plugged in as a tiny, movable lever, and to pull all those levers in coordination rather than letting each hub fend for itself.
That is the core result of "Optimal Control Strategies for a Network of Electric Vehicle Charging Energy Hubs with Smart Scheduling via Distributed Optimization" by Diego Fernandez-Zapico, Finn Vehlhaber, Maedeh Izadi, Theo Hofman, and Mauro Salazar. On the surface, the paper is a piece of mathematical engineering: a convex optimization problem, a decomposition, an algorithm called ADMM. But beneath the notation sits a practical question that will shape the next two decades of the energy transition: when millions of electric vehicles plug in, who decides when each one draws power, and can that decision be made intelligently without anyone's private data leaking out?
The answer, as the researchers show, is yes — and the savings are not marginal.
The Science
The object of study is what the authors call a Charging Energy Hub (CEH): a charging station that does not just take power from the grid but produces some itself. Each hub in this network has rooftop solar panels, a stationary battery energy storage system (BESS — a fixed battery that banks energy for later), and a connection to the main distribution grid. Some hubs lack certain components: in the Dutch case study, hub 2 has no grid connection at all and hub 3 has neither batteries nor solar, relying entirely on its DC links to its neighbors.
The clever part is the network itself. The hubs are physically interconnected by direct-current (DC) power lines, so energy can flow not just vertically between grid and car, but horizontally between one hub and another. A hub with sun-drenched afternoon solar and no cars can push power to a hub whose cars all arrive at 5 p.m. An inter-hub cable, in other words, acts like a small private power line that bypasses the grid entirely.
The optimization problem the researchers set up is elegant in its scope. They track, at fifteen-minute intervals over a two-day horizon, the charging power delivered to each individual vehicle, the charge and discharge of each stationary battery, and the power flowing along each DC cable. The objective is to minimize the total cost of electricity bought from the grid — multiplied by the real Dutch day-ahead price — plus a small regularization term that nudges the system toward smooth, non-erratic power profiles rather than jittery ones. The whole thing is formulated as a convex quadratic program, a type of problem with a single guaranteed optimum, which means the math will never get stuck in a bad local answer.
One detail worth pausing over: the system operates on what engineers call V1G. That is smart, schedulable charging — moving when a car charges within its window of availability — but not V2G, vehicle-to-grid, which would let a car discharge its battery back into the network. The authors deliberately exclude V2G, so every improvement they report comes purely from smarter scheduling, not from treating cars as mobile power plants. That is a conservative and more realistic baseline, because V2G brings its own costs — battery degradation, customer consent, complex metering — that this work deliberately sidesteps.
The method, though, is what gives the paper its title. A centralized optimizer that knew everything about every hub could solve this problem in one shot, but that requires each station to share its full internal state — how many cars are plugged in, how much energy each requests, when each will leave — with a single operator. That is a privacy problem and a scalability problem. The Eindhoven team instead decomposes the global problem so that each hub solves its own local optimization while sharing only one small piece of information: the power it wants to exchange along each DC cable. A central "collector" then nudges the shared values toward a consensus.
This is the Alternating Direction Method of Multipliers (ADMM), a workhorse distributed optimization algorithm. Its value proposition is that it converges to the same global optimum as a centralized solver while moving only a few numbers between agents. Each hub guards its own data; no hub needs to know how many vehicles its neighbor is serving or at what price that neighbor sells back to the grid. The algorithm iterates: each hub updates its private variables, reports its proposed cable flows, the collector averages them and gates them to stay within physical limits, and the dual variables — the mathematical "penalties" that pull everyone toward agreement — climb toward a consensus. When the primal residual (how far apart the hubs' proposals are) and the dual residual (how much the consensus is still shifting) both fall below tight tolerances, the network has reached an agreement that is provably optimal.
What They Found
The headline numbers come from a 30-day evaluation across 2025, using real charging sessions from a Dutch EV company called Maxem, real day-ahead electricity prices and emission factors from ENTSO-E (the European grid operators' association), and real Dutch weather history from Open-Meteo to drive the solar model. Three evaluation days were drawn from each of the ten months studied, at a 15-minute resolution.
Compared against the baseline — where each car charges at a flat, fixed power for the whole window it is plugged in, and only the batteries and cables are optimized — the V1G system delivered a 30 percent reduction in total electricity cost and a 25 percent reduction in total electricity emissions. Those are not incremental gains; they are the kind of numbers that change business cases.
Total electricity cost: V1G versus fixed-power Baseline
Total electricity cost comparison between the fixed-power Baseline (problem 2) and the optimized V1G (problem 1) over the 30-day evaluation. V1G achieves a 30% reduction in total electricity cost.
| Label | Value |
|---|---|
| Baseline | 100 |
| V1G | 70 |
The savings are not distributed evenly, and the asymmetry is itself instructive. Hub 1, which has the full complement of components including a grid connection it can sell power back to, captures the largest share of both cost and emission reductions. That is partly because a grid-connected hub with a big battery can buy cheap, buy renewable, and sell the surplus at the best moments — a flexibility that hubs 2 and 3, connected only to each other and the cables, cannot fully exploit.
Total electricity emissions: V1G versus fixed-power Baseline
Total electricity emissions comparison between the Baseline and V1G over the 30-day evaluation. V1G achieves a 25% reduction in total electricity emissions.
| Label | Value |
|---|---|
| Baseline | 100 |
| V1G | 75 |
The second major result is about trust: the distributed ADMM solution matches the centralized optimum. The researchers verified that their decomposed, privacy-preserving algorithm converges to the same answer a monolithic solver would produce, within tight numerical tolerances. In practical terms, that means a network operator does not have to choose between privacy and optimality. It can have both.
There is an important caveat hanging over the emissions number. The 25 percent figure is computed from the emissions factor of the grid electricity in the Netherlands on each evaluation day. Shifting charging to moments when the grid mix is cleaner — times when wind and solar are abundant, so the marginal electricity is low-carbon — is exactly what the optimizer exploits. The paper models solar production and price, and the emission factor, but the interplay is the point: because the Dutch grid's carbon intensity swings through the day, and because V1G lets the system chase the low-carbon hours, the emissions follow the cost down.
Components per hub in the Dutch case-study network
Schematic of the three-hub network studied: hub 1 has all components including a grid connection it can sell power back to, hub 2 lacks a grid connection, and hub 3 lacks both battery storage and solar panels. Hub 1 captures the largest share of cost and emission reductions.
| Label | Value |
|---|---|
| Hub 1 (grid, BESS, PV) | 1 |
| Hub 2 (no grid) | 2 |
| Hub 3 (no BESS, no PV) | 3 |
Why This Changes Things
To see why these numbers matter, consider what the baseline actually represents. In much of the world today, when an electric vehicle plugs in, it charges at whatever power the station delivers until the battery is full. The car is a passive load. The grid has no say in when that load lands, and the station operator has no reason to care whether the marginal electron came from a Dutch wind farm or a German coal plant. The baseline in this paper is not a straw man; it is closer to the status quo.
V1G changes the relationship between car and grid. Every vehicle that is plugged in for longer than its charging takes is a reservoir of deferred demand — an asset the optimizer can shuffle around its window of availability. A car that arrives at 9 a.m. and leaves at 5 p.m. does not need to charge at noon; it can charge at 2 p.m. when solar peaks, or at 4 p.m. if the price drops. The paper shows that simply exploiting these windows, on a network of only three hubs, cuts a third off the energy bill.
Scale the thought experiment up and the resonance becomes obvious. The European Union and most major economies are betting that electrification of road transport will be one of the largest levers on both emissions and oil dependence. But electrification without smart scheduling risks slamming an enormous, inflexible new load onto grids that were never designed for it. Every analyst's worst-case scenario for EV charging — the "charging in the evening peak" problem that forces new peaker plants and transformer upgrades — is precisely the baseline this paper models. The 30 and 25 percent figures are therefore not just an efficiency curiosity; they are evidence that the grid does not have to be a victim of EV adoption. It can be a beneficiary, absorbing vehicle demand into the hours when renewables are cheap and clean.
The privacy result deserves its own moment in the sun. Centralized control of a large charging network is, in principle, the most efficient: one brain, all the information, one perfect answer. But that brain would need to know, in real time, the battery state and departure time of every single vehicle at every hub — commercially sensitive data that charging operators treat as trade secrets, and personally sensitive data that regulators in Europe are increasingly protective of. The paper's ADMM decomposition dissolves that tension. Each hub keeps its vehicle data private and shares only its proposed cable power flows, a single scalar time series per connection. The algorithm is provably convergent to global optimality, meaning decentralized operation does not cost efficiency. That is the property that makes the approach scalable to the real world: a national network of thousands of hubs, none of which is willing to bare its books to a central planner, can still be operated as if it were one organism.
The DC-interconnection idea is also quietly radical. Most EV charging today is a star topology: every station draws from the grid independently. Wiring hubs to each other with DC lines creates a mesh — a mini transmission grid — through which solar power can travel without ever touching the AC distribution network, avoiding its losses and constraints. Hub 2 in the case study has no grid connection at all, yet still operates because it is fed entirely by its neighbors through the DC links. The paper treats inter-hub power sharing and internal management as one single optimization, something the authors note no prior study has done — earlier work either optimized charging profiles without inter-hub exchange, or modeled charging demand as a fixed aggregate that could not be shaped.
What's Next
The results rest on a deliberate idealization the authors are frank about: perfect foresight. Every evaluation run assumes the optimizer knows the day-ahead price, the solar production, the energy requests, and the availability windows of every vehicle with certainty. Real life is messier. Cars arrive late, leave early, request more energy than announced, and solar panels underperform an overcast sky. The authors flag this explicitly, noting that previous work in their group investigated how prediction errors degrade online hub performance — and they position the current paper as the foundation on which a closed-loop, model-predictive version would build.
That said, the two-day-ahead, 15-minute-resolution deterministic case study is not an empty exercise. It establishes the ceiling: if you could predict perfectly, you could save 30 percent on cost and 25 percent on emissions. It also establishes the math and the decomposition that a stochastic or receding-horizon implementation would reuse nearly unchanged. ADMM is naturally suited to online operation, where each hub re-solves its local problem each time step with fresh measurements and forecasts, since the local problems never grow with the network size — they only grow with each hub's own complement of vehicles and components.
A few open questions loom. The first is V2G: allowing cars to discharge into the network would add another layer of flexibility and presumably larger savings, but it drags in battery-degradation modeling and customer-incentive design that this work deliberately avoids. A second is the renewables uncertainty treatment: the solar forecast here is perfect, and a robust or chance-constrained variant would speak more directly to a grid operator's worst-case planning. A third is market design — the model assumes hubs can buy and sell at the day-ahead price with a fixed selling ratio, but real tariffs are more tangled, with capacity charges and network fees buried in them.
None of these cut against the central finding. The paper's contribution is to show that the coordination problem of a physically connected charging network — privacy included — is soluble at all, and that the prize for solving it is large. As the world piles millions of electric vehicles onto grids already struggling to absorb cheap renewables, the most valuable infrastructure may not be new power plants or bigger batteries but the software that decides, fifteen minutes at a time, which car charges when, and which hub lends its surplus solar to its neighbor a few kilometers down the cable. The Netherlands case study is small — three hubs, thirty days. But the twenty-five and thirty percent it reports are the shape of what coordinated charging can do at any scale.