Load Balancing and Dynamic Power Sharing for EV Charging Sites
A site with ten 120kW dispensers is rarely a 1.2MW site. Understanding how capacity is actually shared is what separates a workable electrical design from an expensive upgrade.
The most common mistake in charging site design is multiplying connector count by nameplate power and treating the result as the site load. A site with eight 120kW dispensers is not a 960kW site in any practical sense. It is a site with a supply limit, and the arithmetic of how that limit is shared between connectors decides whether the installation fits the existing service or triggers a transformer upgrade.
Load balancing is the discipline of deciding which connector gets power, at what level, and at what moment. It sits between the electrical design and the software, and it is one of the first things to settle before ordering hardware. A wider list of site design questions is in the EV charger site selection checklist.
The Site Limit Is Electrical, Not Mechanical
The constraint is almost never the chargers. It is the supply: the transformer rating, the service entrance, the cable feeding the site, and whatever load the building already draws. Adding charging to an existing site means sharing that supply with the load already on it, which is why the existing demand profile matters as much as the charger specification.
- Transformer and service capacity, in kVA, including spare capacity the utility is willing to allow
- The existing load curve: what the building or facility draws across a typical day and week
- Demand charges and the tariff structure, which often penalize peaks more than total energy
- Physical constraints such as cable routing, cabinet space, and where a new service could land
Those four items decide the design more than the chargers do. A site with a low overnight base load and a high daytime peak has very different headroom from one with flat industrial load, even at the same total capacity.
Static Limits, Dynamic Sharing, and Queues
There are three broad approaches, and they are not mutually exclusive. Most real sites combine at least two of them.
| Strategy | How it works | Best suited to | Trade-off |
|---|---|---|---|
| Static current limit | Each connector is capped at a fixed current, chosen so that all connectors together stay inside the site limit | Small AC sites, simple installations, overnight depots | Capacity is stranded whenever a connector is idle |
| Dynamic power sharing | A controller distributes available power across active sessions moment to moment | DC sites with varying session counts, retail and corridor locations | Needs a controller, a communication path, and a clear allocation policy |
| Priority queue | Sessions are ranked and power follows the ranking, up to a full-speed allocation for the top session | Fleet depots, bus terminals, sites with contractual priorities | Drivers on low-priority connectors wait longer, which must be communicated |
| Scheduled profile | Time-based limits, often aligned to tariff windows or a local generation curve | Sites with demand charges or on-site solar | Requires accurate input data and a backend that enforces the schedule |
The important practical difference between static and dynamic is utilization. Static limits are cheap and robust, and on a site where every connector is busy all day they perform as well as anything else. On a site where two of eight connectors are in use most of the time, a static limit holds those two sessions to an eighth of the site capacity for no reason.
Why Split Architecture Shares Power Naturally
Where the power conversion hardware physically lives decides how easy sharing is. In a combined dispenser, each unit carries its own rectifiers, so sharing means coordinating independent hardware over a network. In a split design, rectifiers sit in a shared power cabinet and feed dispensers over DC cabling, which means one pool of modules serves every connector attached to it.
That is the structural reason a split DC charger main cabinet is a natural fit for a site that expects uneven usage. The cabinet can be specified with fewer modules than the sum of its dispensers and still deliver full power to whichever connector needs it, because the modules are pooled. A quad-connector four-gun DC charger works on the same principle at a smaller scale, sharing one conversion stage among four outlets.
It also makes growth cheaper. Adding a dispenser to an existing cabinet is a cabling job, not a new service connection, as long as the cabinet has module slots free and the site supply has headroom.
What OCPP Smart Charging Profiles Actually Do
The protocol that carries these instructions is OCPP, published and maintained by the Open Charge Alliance. Both 1.6J and 2.0.1 define charging profiles: structured instructions that tell a charger or a group of chargers how much current they may draw, and when. A profile is not a suggestion. It is a limit the charger is expected to enforce locally even if the connection to the backend drops.
Profiles are stackable. A site can run a baseline profile that caps total site draw, a tariff-driven profile that shifts energy into cheaper windows, and a local profile from a building management system that reacts to the building load. The charger or the local controller resolves the stack and applies the most restrictive combination at any given moment.
The gap between specifications and reality is worth knowing about. OCPP 1.6J profiles are implemented inconsistently enough that a multi-vendor site often falls back to static limits. OCPP 2.0.1 tightens the data model and adds explicit external constraints, which is why sites that expect real dynamic sharing increasingly specify 2.0.1 hardware from the start.
Local Controller or Backend Scheduling
Scheduling can run in the cloud, on a local controller, or in both places with the local device holding authority. The choice is about what happens when the network fails.
- Backend scheduling gives one view across many sites and is easier to change without a site visit
- A local controller keeps enforcing limits when the uplink drops, which is the common failure case
- A layered design lets the backend set policy while the local device holds the safety envelope
The safety envelope is the part that must be local. Whatever the backend intends, the site must not exceed its service capacity because a cellular link went down. Any design where the fallback behavior is uncontrolled full power on every connector is a design that will eventually trip a main breaker at the worst possible time.
What the Site Owner Has to Provide
Dynamic sharing is only as good as its inputs. Without real data, the controller either over-restricts the site or puts it at risk. The owner or facility manager should be asked for four things before the design is fixed.
- Transformer and service capacity in kVA, plus the maximum demand the utility will permit
- A load profile for the existing facility, ideally interval data from the meter rather than an estimate
- The tariff structure, including demand charges and any time-of-use windows
- Constraints on when charging demand may peak, such as a shift start that cannot be disturbed
Interval meter data is the item most often missing and most valuable. Fifteen-minute or hourly readings over a few weeks will show the real headroom far better than a nameplate rating will, and they usually reveal that the site has more usable capacity than the owner assumed.
Storage as a Peak Shaving Tool
Where the supply is genuinely tight or the demand charge is punitive, a battery can absorb the peaks instead of the grid. A 261kWh energy storage cabinet paired with a site controller can charge at low demand and discharge into the chargers during peak sessions, which reduces the maximum site draw without reducing peak charging power.
The economics depend entirely on the tariff. Storage is close to pointless on a flat energy tariff with ample supply. It becomes compelling where demand charges dominate, where a service upgrade would cost more than the battery, or where time-of-use windows are wide enough to arbitrage. European sites also have to account for the grid connection rules set out in the EU Alternative Fuels Infrastructure Regulation, which pushes sites toward managing their own peaks rather than drawing whatever the connection allows.
The combination that works best in practice is storage plus dynamic sharing plus a sensible priority policy. Storage absorbs the site peak, sharing keeps every connector productive, and priorities keep the vehicles that must leave soonest moving fastest. Running only one of the three leaves capacity on the table.
Can I avoid a service upgrade with load balancing?+
Often yes, but not always. If the existing supply has genuine headroom, dynamic sharing can fit a substantial charging site inside it. If the site is already near its maximum demand, no software policy will create capacity that the transformer does not have.
Does dynamic sharing slow down individual sessions?+
Only when the site is busy. On a quiet site a single session gets the full power its vehicle can accept. On a full site every session gets a share, which is usually better for drivers overall than a queue where some cars wait to start.
Is OCPP 2.0.1 required for dynamic power sharing?+
No, but it is more reliable. OCPP 1.6J supports charging profiles, though implementation quality varies between vendors. OCPP 2.0.1 defines the data model and external constraints more precisely, which matters on a multi-vendor site.
What happens if the network connection to the backend drops?+
The charger or local controller should hold the last known limits, or fall back to a conservative preconfigured site cap. Any design where the fallback is unrestricted full power is asking for a main breaker trip.
Do I need battery storage to run load balancing?+
No. Storage is a separate measure for sites where the supply is tight, the demand charge is high, or a service upgrade is expensive. Dynamic sharing alone handles the common case where connectors simply do not all run at full power at once.