Energy that stays on
Holding an island grid at frequency is now a question of how fast inverters agree among themselves, and of what must never wait for a radio.
Small island systems are commonly described as expensive to power, and the description is accurate. It is not the figure that governs daily life. That figure is availability: the hours in a year during which the power is actually present. Once a system carries a substantial quantity of solar and wind, what decides availability is no longer how much generation has been installed. It is whether the generation can be coordinated quickly enough.
The number the system watches
An alternating-current grid carries one continuous measure of its own health. Generation and demand are matched instant by instant or they are not matched at all, and frequency is the reading: when demand runs ahead of supply the machines slow and the frequency falls; when supply runs ahead the frequency rises. Fifty hertz, or sixty depending on the island, is a statement that the balance is being kept. A tenth of a hertz away from it is a statement that it is not, and the correction has to arrive in seconds.
Why an island feels a cloud
Large interconnected systems defend that balance with inertia. The rotating mass of synchronous machines — turbines, diesel sets — stores kinetic energy and gives it up automatically as the frequency falls, which buys the operator time before anything has to be decided. Inverter-coupled solar and wind carry no such mass. Every megawatt of it that displaces a running diesel set removes a little more of the system’s tolerance, and a small island grid did not have much to begin with.
The consequence is measurable. Cloud passing across a photovoltaic plant can remove 60 to 80 per cent of its output in two to five seconds. On a continental system that is noise. On an island where the same plant is a fifth of the load it is an event, and in microgrids dominated by inverter-based generation the rate of change of frequency can exceed 2 Hz per second — brisk enough to trip the anti-islanding protection on generation that is still working perfectly well, which is how a shortfall becomes a cascade.
Islands accordingly carry expensive insurance against their own weather: 20 to 30 per cent spinning reserve is not unusual where penetration is high and inertia low. That reserve is diesel, running below its efficient point, to cover a cloud.
The loop that cannot wait for a radio
The first response cannot be instructed, because there is no time to instruct it. Under IEEE 1547-2018 each inverter measures frequency at its own terminals and adjusts its output along a droop curve it already holds: a deadband of 0.036 Hz by default, a droop of four to five per cent, and full response inside five seconds. Battery inverters answer faster still, in under 100 milliseconds. Nothing is transmitted and nothing is awaited. Several hundred devices arrive at a coherent collective response without any of them being told to, because each is reading the same physical signal at the same moment.
This is the part worth being clear about, since it is where most descriptions of a “smart” island grid go astray. Fast frequency control is local, autonomous and communication-free by design, and it must remain so: a control loop that depends on a radio link fails when the radio link does.
What the radio layer is for
The slower work is a different matter, and it is the work a small utility is least equipped to do. LoRa suits it well. It runs in the sub-gigahertz bands, reaches ten kilometres and more in the open, costs little per node, draws almost nothing, and depends on neither cellular coverage nor a fibre run — which matters on an island where both are thin and where a storm may take out whichever exists.
Its limits are equally plain and should be stated. Duty-cycle rules leave a gateway on the order of 500 to 1,000 downlink messages in a day, shared across every device beneath it. Delivery is not guaranteed. Payloads are tens of bytes. A device that listens continuously for instructions must be mains powered, which at an inverter it is.
None of that suits a control loop, and all of it suits the layer above one:
- Enrolment and telemetry. Knowing what is connected, where, and what it is presently doing. Most island utilities cannot answer that question about their own distributed generation.
- Changing the rules the local loops follow. Droop curves, curtailment ceilings, ramp limits and export caps are parameters held in each inverter. They need revising seasonally or before a storm, not every second.
- Reporting state. Battery state of charge, availability, fault codes — the inputs to a dispatch decision made minutes or hours ahead.
- The record afterwards. What each asset did during an event, which is the evidence a regulator or a lender later asks for.
The cascade, by timescale
| What responds, and how it is carried | Acts within |
|---|---|
| Inertia, from whatever synchronous mass remains | instantaneous |
| Battery inverters, on local measurement | under 0.1 s |
| Inverter frequency-droop, on local measurement | under 5 s |
| Setpoints and curtailment, over the radio layer | minutes |
| Dispatch and unit commitment | minutes to hours |
Read downwards, the demand on the communications layer falls away to nothing at the top. That is the design: each tier does what its latency permits, and the tiers that must act in milliseconds are given no dependency they could lose.
What it does not settle
Radio does not replace protection, which stays hardwired. It does not confer inertia; that comes from storage configured to synthesise it, or from grid-forming inverters able to establish a voltage and frequency reference rather than follow one — and a system intending to run without any synchronous machine at all needs those, LoRa or no LoRa. A single gateway is a single point of failure and should not be one. Setpoints that can be sent to an inverter can be sent by somebody else, so authentication is not optional. And none of it substitutes for maintenance by people who live within reach of the plant.
Why it decides whether a project is financeable
What a lender underwrites in an island generation asset is contracted availability. Nameplate capacity is a statement about equipment; availability is a statement about how the equipment behaves together, and the coordination layer is what produces the second from the first. A plant able to show what it did through a year of cloud, from its own records, arrives at a credit committee in a different condition from one that can only say what it was rated at.
There is nothing novel in any of this. The modulation was lifted from radar and sonar and put to work reading utility meters; frequency droop is older still. Only the arrangement of them is recent, and it is on the arrangement that an island’s availability now depends.