On calculating carpark charging loads

A classic problem demanding increasing attention concerns the amount of electrical reticulation and distribution capacity needed to support EV charging. AS/NZS 3000 is pretty clear on what it wants in your home (e.g., if it's permanently wired, demand calculation require 100% of device capacity to be considered) - in a workplace or apartment setting that might have tens, hundreds (or in future, more) charging stations that's simply not practical: energy management, long present in charging infrastructure management frameworks, specifically exists for capacity provisioning to be reasonable.

But what is actually reasonable, and how to calculate as much? You'll have heard it all - from lojacking complete fleets with aftermarket telematics solutions to digital twins and more, however in the spirit of not letting perfection/resolution be the enemy of good enough/accuracy I'm a fan of 80/20 (or greater) approaches that provide contextual understanding informing better decisions. And hey, if your application has this knowledge intrinsically there are applications that can compute directly (e.g., if you're computing EV load for public transport fleets and then some, ChargeSim is criminally underrated).

Some time ago I wrote some elementary code with a good student of mine to these ends which I've since cleaned up / debugged / added some more stuff to. What's presented here is a robust if elementary but scaleable version of things; Added Realism Caveats are made where appropriate. At Greenergenic we retain some much-evolved versions of this basic theme; if you want Jedi-level answers we're available to consult.

Let's dig in.

Defining when cars arrive and depart

Key to this problem is understanding when EVs turn up to the carpark and when they leave. With a wee bit of contextual understanding we can put together probability distributions (PD's) of arrivals and departures, e.g.,:

• Arrival and departure PD's will differ
• Weekends are going to be different to weekdays
• Public holidays are going be different again
• Etc

How many ways things need to be carved up to capture adequate realism is up to what captures salience in your given situation. I've split this one six ways and across hour time intervals, and allowed for the possibility that sometimes cars will not leave or return on a given day ('NP = Not Present').

So imaginatively on weekdays:

People in this parking location are most likely to depart between 5AM and 9AM, and no one's departing after 4PM. They're coming back in two distinct waves, either just after a morning or afternoon rush. Some of them (<3%) don't come back that day at all.

On weekends:

They're departing in the morning and coming back in the afternoon (the theoretical people here are responsible nightowls and take PT to get around at night). Whilst not shown here, I've assumed that on public holidays they'll tear off in the morning and come back from it but with a material proportion of trips coming back quite late or not at all that day. (Realistically one could be a bit more astute about how this breaks down and could derive the PDs from various sources for greater accuracy, however for demonstration purposes this is fine).

From here things are pretty simple:

• For each vehicle simulated we create a time vector, differentiating between weekdays, weekends and public holidays,
• We loop through the journey of a given car, using the PDs to generate probabilistic arrival and departure times, and
• We loop this through the number of cars in a parking location.

And from that we've modelled movements in and out of a carpark. Using the PDs above, here's one for 50 cars (just for fun I randomised results and processed results to 5-minute resolution):

Nothing unexpected. If we zoom in on a week we can see it's reasonably detailed, and pretty much as expected with the (deliberately inserted) weekend and weekday disparities visible:

It really doesn't matter how you come to this conclusion; you could use PDs or traffic data or whether else. It's only important that you come to a place where you have this data and that it comes from defensible perspectives. A carpark for an apartment block is going to look different to a workplace or a shopping centre, and so tune the PDs to suit.

What happened when they were out and about

Easy - they drove their cars. Which involves another PD for daily driven distances; we can pull that from the ABS or similar. Here's one we found earlier:

(Yes, it's be more realistic if arrival/departure times weren't modelled independently of arrival/departure times, though it's relatively easy to capture upper/lower bounds for skew in easier ways.) So at this point we have when they were parked, how long, and distance travelled when not in the car park.

Realising consumption

Obviously not all EVs consume the same amount of energy when they're out and about - there are other considerations too, e.g.,:

• EVs differ in homologated efficiency
• EVs differ in real-world efficiency in different ways
• EVs differ in battery size and charging power capacities
• Other external variances exist, e.g., prevailing weather

A simple approach is proposed:

• Vehicle specification data on EVs sold locally within the last five years is acquired and tabulated, which is in part used to develop vehicle segmentation per FCAI (or whatever market matters) guidelines. This data includes sales data (where available), vehicle dimensions, efficiency and key power system specifications,
• A simulated vehicle fleet is generated against a distribution function of known per-segment sales (i.e., vehicles are distributed based on how consumers are known to purchase among segments),
• A PD of real-world percentage deltas between homologated and actual EV range estimate is created and applied to all vehicles in the simulated fleet, and
• This efficiency is further weighted with known seasonal effects; we used a paper detailing seasonal variation across an EV fleet in a particular region and used that (along with contemporary weather data for that region) to estimate seasonal efficiency loadings for the time and place simulated (a few minutes on Google Scholar will find a several relevant examples).

Obviously there's some further realism for those wanting or needing as much; if your're modelling a workplace that only runs electrified Landcruisers with drivers carrying significant loads at all times it's possible to create a more representative distribution with ease. It's also possible to add complexity and relations between e.g., efficiency impact and vehicle classes - to shortcut that I've added a bit more penalty against homologated numbers than is statistically average (i.e., a conservative approach).

From this and considering all previous we have when the EVs arrive at a carpark, how long they stay there and (per occasion) how much energy they need when parked based on their last simulated trip.

Modelling charging

So now we've the stage set for a simple spill 'n' fill problem: when constrained by vehicle, trip, garage movement and EVSE characteristics, what power would a variable number of EVs tend to use when parked at a given location.

I'm going to simplify this one too:

• Minimum charging current is 6A per IEC 61851/SAE J1772 (if you want to model mixed fleets on ISO 15118 capable of dropping to lower power levels too, so be it) at 240VAC (1.44kW). Below this, zero. Maximum charge power 7kW. Charging efficiency is constant at 95% (reality will vary, particularly at low power levels).
• No V2G (easy enough to add)
• Charging is instant on parking (easy enough to randomise delay, or to model cars that park but don't charge)
• Everyone's trying to get to 100% SoC (easy enough to change or randomise)
• Site capacity limits are constant throughout the simulation, and no additional is modelled to skew charging load at one time or another

Outputs are to be power required over time from which any statistical measures can be realised (e.g., average, mean, etc).

Now because we want to understand and contrast the impact of managed charging, we're also going to model a reasonable lower bound in an easy way:

• Cars arrive with 'perfect visibility' of next departure (V2G charge sessions under ISO 15118-20 can actually capture what's anticipated in practice; for now some ML + user exception management can work fine), and
• Charging current is modelled as the lesser of capacity required to attain full SoC over any parking session divided by the time of that parking session, or maximum power permitted in a given simulation (being simpler than coding for time-slicing/'Energy Jenga' but giving effectively the same answer - but for simplicity we still respect the 6A minimum, which in future could be lesser).

We get the same output structure this way. In either case we're also able to capture the number of charging sessions that did not completely fill an EV's battery - call that a proxy for needing charging at other locations (e.g., public, workplace relative to home, home relative to workplace, etc).

Results

For the above hypothetical scenario

So:

• On average, cars are parked in our hypothetical garage on average 61.5% of the time and each making an average 357 trips in the year modelled.
• The maximum coincident energy demand was 85.6kW, which is no where near the ~350kW (plus line losses) needed to charge all 50 parking spaces at the maximum capacity of the provisioned charging infrastructure in one go.
• The maximum fully-managed energy demand was 51.3kW (40% or so less) - in other words there's certainly value in smart charging.

The comparison of power demand looks as follows:

Zoomed in per the same time period explored earlier with respect to parking movements:

So what if the carpark size changed?

At this point many will observe that even with coincident charging, the odds of all charging at the same time is very, very low. It stands to reason; vehicles don't return to plug in at the same time, and most trips in our simulation are under 30km in duration - even if an EV were relatively inefficient at 250W/km, that's 7.5kWh total - a bit over an hour at ~7kW charging. So to work out how 'what you need' might scale with 'how many cars have you got' in the present scenario, let's do the following:

• Run the simulation a number of times across a range of car park sizes (assuming there's an EV in each),
• Average runs per car park size, and
• Compare results, expressed as the power capacity required per vehicle for a theoretical carpark of a given size.

We should reasonably expect that a single carpark with a 7kW charging station and a car able to utilise that fully will top out at 7kW. We should reasonably expect that two cars are pretty close to twice that (assuming the EVs are good for it), but as more and more cars come into it that the probability of needing maximum power for every charging space all the time starts to diminish.

It indeed plays out as such (five runs at each carpark size are averaged here):

(Whilst the above needs to have all the practical things added - line losses, etc) this does scale down quite nicely; from 7kW per car for one of them to just under 1.5kW per car at 100 cars - dropping to 0.8kW if 'perfectly managed'. Whilst 'perfect' management isn't possible in practice, it does have value: the 250-car-space case modelled above requires 287kW for coincident charging and less than 175kW for perfectly managed - if you've ever dealt with building facilities or a DNSP, you'll know these are numbers that do make a difference to project cost and timing: you'd likely value (and pay for) management.

Maximum demand in volume is hardly ever 'maximum charging station power times the number of stations.'

"What you really need are 22kW charging stations"

In the examples above a portion of vehicles will need charging away from the hypothetical parking garage they're based at - part of it is simply a function of travelling a greater distance than their batteries have range for (some 1.8% of over 17,850 trips modelled). On some occasions, however, there's not enough time to charge vehicles to full SoC either - this could conceivably be improved by faster charging.

To these ends quite a number of fleet charging sales efforts tend to focus consumer needs on 22kW charging infrastructure; let's examine what happens if we (for simplification's sake):

• Model the aforementioned theoretical carpark of 50 spaces,
• Assume charging power isn't limited by the On-Board Charger but instead by the power capacity of the charging station involved (e.g., more akin to a DC scenario),
• Slew the charging power limits,
• Repeat 5 times each, average the results per power limit (yes statistical peeps, usually we'd do 30 but time), and
• Assess the effect of charging power on charge session completion to full SoC.

We get this:

1.7% of charging sessions in this theoretical scenario don't make full SoC at 7kW. That drops to 0.5% with 22kW charging infrastructure - assuming every car that turns up can use it, which is realistically a very small portion of the EV market.

Other things worth discussing

There are some topical adjacencies worth unpicking.

What data builds a workplace charging facility differs, right?

It certainly does. Let's quickly repeat the above with a mythical workplace:

• 20 vehicles,
• Vehicles are all utility vehicles with a 120kW battery, 7kW OBC, and 275 Wh/km efficiency,
• The facility is closed on weekends and public holidays, and
• All vehicles leave their depot between 7AM and 9AM, they all drive a minimum of 50km with an average of 180km per day, and all vehicles return between 2PM to 6PM - no one ever takes a vehicle home overnight (I did suggest it was a 'mythical workplace').

So the parking movements look as expected:

All good; the trip distribution is considerably skewed relative to the last simulation:

And the site power consumption reflects the above:

We can see that:

• Vehicle arrivals and departures are (expectedly) quite clustered,
• The carpark is periodically empty, and
• In the perfectly-managed charging scenario, charging times run close to (or are otherwise limited by) parking time limits.

Accordingly we can expect that at this power transfer limit (7kW) there will be little marginal difference in capacity requirements for coincident charging across a range of carpark sizes, and that whilst management is valuable the site reticulation requirements (as a function of vehicle numbers) should remain high relative to the previous example - both prove true:

Coincident charging needs the full 7kW per car from 1-50 parking spaces; by 50 parking spaces there's a 40% reduction from perfect management; still, compared to the previous case (per 20-space car park) with approximately 60kW needed to support coincident charging, this application needs nearly 140kW (with management able to reduce this peak by a maximum of ~35%).

As for what increasing charging capacity might do for fleet uptime:

In this instance the increase from 7kW to 11kW reduces incomplete charging sessions from nearly 7% to 0.25% - a 28-fold reduction. Beyond this power level there are no charging sessions that don't run to full.

x% of missed charging events isn't a problem... right?

This really depends on the nature of the mission a given vehicle undertakes. The marginal cost of imperfect capacity will differ significantly between e.g., a vehicle used in a residential setting that is unlikely to run out of energy if not charged to 100% SoC (and which would not materially suffer from using public charging infrasructure if it needed to) and a fleet of electrified ambulances with very strict criticality requirements.

It is thereore incumbent on fleet operators to be objective and realistic to these ends and for policymakers to understand, at least in a local context, the marginal value of incentivising e.g., fleet vs public charging infrastructure: the end goal is not to e.g., build more DC-fast charging infrastructure, it is to have vehicles charged affording appropriate and appropriately reliable utility.

V2G

V2G is certainly Flavour of the Month with consultants, policymakers and Internet People Attempting Relevance - does it make a difference? I'm a V2G solution developer for ~7 years now and would love to tell you/sell you that it does, though in all honestly the effects on a relevant simulation are straightforward and limited:

• Any export event that generates value will ultimately seek to maximise line capacity, and
• Any additional line capacity for the purposes of marginal value sought through V2G activities needs to be considered against the marginal cost of capacity augmentation; the costs and values of both are in turn defined by local network factors and market conditions.

Capacity augmentation is unlikely to get less expensive and, on a long enough timescale, Primary Frequency Response (PFR) service delivery (what we call fast/very fast FCAS) is likely to be the most lucrative of all opportunities - increasing in market size with power system volatility and decreasing in opportunity value with more players, the latter of which will grow in magnitude quicker at any rate. So unless talking to first-mover projects, investment efficiency is likely to be optimised in design for primary use (charging).

There's also the possibility of exporting from one or more vehicles to charge another within the same circuit (or otherwise distribution) as a way of dynamically increasing demand capacity though this is an extreme case (the marginal cost of simply deferring other charging in a fleet is likely lower and in most cases effects the same end).

Dynamic cost optimisation

This one's arguably more important - today we have ToU tariffs with (in some cases) demand charges, but increasingly we have consumer outcomes directly availed to merchant energy opportunities (e.g., wholesale spot pricing or financial derivatives thereof, DR, Amber Electric etc).

This is particularly important because we've actually got a strategy for this; AEMO and CSIRO have some robust and detailed predictions from the 2022 ISP (haven't had time to run 2024 yet) which, when rearranged into a nice spreadsheet allowing scenarios, years effective and vehicle classes to be flicked on and off: it looks a bit like this for the predicted national EV load per time of day, year averaged, end of decade, residential-classed vehicles only in the favoured 'step change' scenario:

Which is about what we expect - coincident charging dominating the shape with most consumers plugging in when they get home (since 2022 most would suggest the market's to get more optimistic than the above). What of the 2050 scenario then - strong electrification - otherwise same vehicles:

To these ends there's a broad expectation that charging should align with VRE (particularly solar PV), with the implication that consumers will do so as net price signals, charging infrastructure availability and charging session management should align accordingly. Note also the expectation that V2G should contribute strongly to the load sector's behaviours.

The point is that our market - like most others - is in the early stages of a long transition wherein what defines value in charging will change and evolve in time: if future-planning site capacity in detail, should consider value to these ends - though I'd argue that most local sales efforts presently trail as much. We've the makings of a government-endorsed strategic direction that the line capacity required to charge during the day will exceed that required to support networks and the power system at night. Yes, today's charging infrastructure is unlikely to be around in 30 years, but today's reticulation can last as much (and then some) without issue.

PFR service delivery (what we call FCAS)

Some regard this as an extreme, out-there, experimental thing; in truth Primary Frequency Response (PFR) service delivery is one of the first, oldest uses of modern V2G - demonstrations in modern V2G history are now more than 25 years old. Presently it is a particularly topical thing in Europe particularly, and later CCS standards (e.g., ISO 15118-20) incorporate functionality to better optimise service delivery from EVs (at least from communications perspectives).

That a group of 100 vehicles may have 11kW power interfaces does not mean that the available capacity from such a carpark is necessarily 11*100 = 1.1MW; asides from site capacity limits gating what's ultimately possible, maximum power requires all vehicles needing to be present/all parking spaces with charging infrastructure needing to be occupied and all participating vehicles needing to have sufficient SoC to effect the PFR response sought, not least because:

• Not all services are available at all battery SoC (e.g., a battery at 100% SoC cannot import further, and is therefore unavailable), and because
• Attempts to optimise SoC levels for e.g., FCAS delivery can have material opportunity costs with respect to other value opportunities, which need to be captured and duly considered.

Other considerations exist, however simulations of this nature can go some ways towards characterising available capacity from parking locations towards aggregate bids and value contribution from (prospective) participating assets.

Conclusions

Caveat emptor: design of reliable, competitive and strategically-relevant charging facilities requires competitive advisory and can be made more relevant with elementary characterisation of fleet characteristics and vehicle movements.

Simulations of this nature are expected to grow in number and relevance, whether able to simulate power and movements, or detailed OCPP behaviours to test backends or components thereof that can be integrated into bespoke software stacks. Whatever your choice, approaches do not need to be high-touch or expensive, they simply need to be relevant to contribute meaningfully towards operational and strategic considerations.

In many physical solution sales engagements charging infrastructure may be oversold from a capacity perspective which limits investment efficiency, which can in turn slow the proliferation of solutions essential for vehicle electrification.

In moderate-to-larger facilities, competitive load management solutions are demonstrably effective and can offer (given the cost of fixed capacity) exceptionally short payback.

Get in touch

Happy to talk V2G, EVs, DER and anything else relevant, and always happy to meet new people or to reconvene otherwise.

Reach out on LinkedIn for a chat - I develop solutions with Steve Oh at Greenergenic Inc. and occasionally collaborate with Jon Sibley at enX consulting .