How we run Tasmanian batteries on live wholesale prices

Managing a home battery against a fixed tariff is straightforward: charge from solar, discharge in the evening, repeat. Managing one against a price that changes every five minutes is a different problem entirely, and it needs real infrastructure behind it.

We’ve been building that infrastructure for our own home and for the Tasmanian sites we manage. This article describes how it works, including the parts that went wrong along the way, because the failures are more instructive than the successes.

The shape of the problem

A battery has to make one decision, repeatedly: right now, should it charge, discharge, or sit still?

On a spot plan, answering that well requires knowing a lot of things at once:

the current wholesale price, and how stale that number is
the network tariff at this site, and whether this moment is a peak period
how much solar the roof is producing now, and how much it will produce for the rest of the day
what the house is actually consuming
the battery’s state of charge, and its real charging limits
what the evening peak is going to demand
whether anything else — an EV, hot water — is competing for the same capacity

Get the combination wrong and you do something genuinely expensive: charging a battery from the grid during the most expensive hour of the day, or arriving at the evening peak with an empty battery because you sold everything that afternoon.

The system runs today across our own home and eight other Tasmanian sites, each with its own controller making these decisions on a five-minute cycle.

One price, everywhere

The most important architectural decision we made was also the least glamorous: every site prices from a single shared module.

There’s a principle underneath this. Wholesale pricing is a pass-through — the market sets one price for the whole Tasmanian region every five minutes. Every customer on a wholesale plan in Tasmania is paying that same energy price at that same instant. The only legitimate differences between two homes are the network tariff and a small location loss factor.

So if two sites we manage ever display different energy prices at the same moment, that is a bug, not a market. Making that structurally impossible — one module, one formula, one source — means a site cannot drift without the discrepancy being visible immediately. We run an automated check that compares every site’s price fields at the same interval and fails loudly if any of them disagree.

We learned to do this the hard way. Earlier versions had each site derive its own price. They diverged, and for a while sites showed export prices that were wildly different from each other at the same instant — some positive, some negative, during the same five minutes.

Forecasts are not prices

The cause of that divergence is worth calling out on its own, because it’s a trap anyone building on a wholesale data feed will hit.

Price feeds often publish a forecast for the upcoming interval alongside the settled price for the current one. They look alike. They are not alike. We were reading a forecast field and displaying it as the current price, and forecasts differ per connection point and routinely fail to eventuate — we watched one feed predict a price of nearly 30 c/kWh for an interval that actually settled slightly below zero.

The fix was to price exclusively from the settled market price, and to explicitly label any forecast data as a forecast wherever it appears. A number that moves when nothing happened is usually a forecast wearing a price’s clothing.

Fail closed, never freeze

The second rule that shapes a lot of the design: when the system doesn’t know the price, it must say so rather than guess.

The worst failure mode in a system like this isn’t an error message. It’s a stale number that still looks live — a price frozen twenty minutes ago that the battery is still confidently acting on. So a stale or missing price produces no price at all, and decisions that depend on it don’t run. The same applies to the customer-facing displays: they show an honest “unavailable” rather than the last number they happened to see.

This shows up in small places too. Prices in Tasmania sit exactly on the zero boundary for hours at a time, flickering between marginally negative and marginally positive. Any rule written as a bare “is the price below zero?” will flap dozens of times an hour, and every flap is a command sent to a battery. Every one of those decisions now runs through a latch that requires several consecutive readings to change state.

The rules that don’t bend

Some behaviours are enforced as hard constraints rather than left to the optimiser, because the cost of getting them wrong is high and the benefit of flexibility is low:

Never charge a battery from the grid during a network peak. This sounds obvious and is surprisingly easy to do by accident — a battery told to hold a small fixed output during a peak will quietly bank its own solar and let the house buy the shortfall back at peak rates. That is grid-charging at peak in everything but name. Holds are now sized against measured solar production, not fixed numbers.
Solar first, always. While the sun is up, a grid purchase only happens for what the rest of the day’s forecast sunshine genuinely cannot deliver before the evening.
Arrive at the evening peak with a full battery. The reserve needed is calculated from each site’s own measured load history against the solar forecast, not a fixed percentage.
Never exceed the site’s electrical limits. Battery charging is never stacked on top of EV charging in a way that would exceed the main breaker.
Every site’s tariff is pinned explicitly. A site with an unrecognised tariff refuses to start rather than assuming a default.

That last one is deliberate. An unknown value producing an error is far safer than an unknown value producing a plausible guess.

Learning from each site’s own data

Several decisions are driven by what a site has actually demonstrated rather than what its specifications claim.

The clearest example: a battery’s real charging rate often falls well short of its rated figure — cold weather alone can more than halve it. If the plan assumes a rate the hardware can’t deliver, it schedules a charge window that doesn’t finish, and the battery arrives at the peak underfilled. So each controller now derives a charging cap from its own telemetry: if the battery has repeatedly accepted less than its rating while surplus energy was still available, the lower figure is treated as the truth.

The same approach applies to load forecasting. Each site’s evening reserve comes from its own measured weekday consumption profile, not an average of other houses.

What it deliberately does not do

Being clear about the boundaries matters as much as describing the capabilities:

It does not trade on any market. Our LocalVolts partnership is a read-only data partnership. The API we hold has no trading functions, and we place no orders on anyone’s behalf.
It is not a virtual power plant. No vendor or retail VPP operates in Tasmania. Nobody outside the home is dispatching these batteries for someone else’s benefit.
It does not read a customer’s data without authorisation. Access comes from an authorisation the customer gives inside their own account.
Financial and physical changes are not auto-applied. The system can propose changes to its own logic, but anything touching money or hardware is reviewed by a person before it runs.

Is it actually worth it?

An honest answer has two parts.

The part that reliably pays is unglamorous: not buying power during expensive hours. Self-consumption and peak avoidance are where the bulk of the value is, and they work on any tariff, with or without a market plan. A battery that simply covers the evening peak every day is doing the heavy lifting.

The part that’s easy to overstate is the clever market behaviour. We measured one deep negative-price event where being paid to import was worth about 2.6 c/kWh for roughly an hour. That’s real, and it’s small. The much larger prize during the same window was storing surplus solar that would otherwise have been exported at a negative price, and using it that evening against rates of 25–36 c/kWh.

We try to be careful about this distinction, including internally. It’s tempting to present every optimisation as a money-saver. Most of the value in this system comes from a handful of boring behaviours executed reliably, every single day — not from occasional clever moments.

The honest summary

The infrastructure exists because the problem is genuinely harder than it looks, and because the failure modes are expensive and quiet. A battery making a bad decision doesn’t alarm. It just costs you money at 6pm, every day, until someone notices.

Most of what we’ve built is defensive: one price source instead of many, settled prices instead of forecasts, honest gaps instead of stale numbers, hard limits that can’t be optimised away, and figures derived from what each site has actually proven it can do.

None of that is exciting. All of it is the difference between a battery that responds to prices and a battery that responds to prices correctly.

This guide was written by the Tasmanian solar team at Solar Generation Tasmania.

#solargentas #solargen #solargenerationtasmania

Published by Solar Generation Tasmania. General information current as at publication — rebates, tariffs and prices change, so confirm current figures before deciding.