Generator paralleling: how it works and what two sets need to run together


Paralleling means running two or more generator sets onto one bus so that they share the load instead of taking turns. It is worth doing when the site needs N+1 redundancy, when the load grows in stages, or when one machine of the required size cannot be transported or is not made. It only works when four quantities match at the moment the breaker closes – voltage, frequency, phase angle and phase sequence – and when the governors, the excitation and the protection are designed to share rather than just to run.

What paralleling actually is

In a single-set standby installation the set owns the bus. It starts, closes onto a dead bus, and whatever it produces is what the load takes. Paralleling changes that in one specific way: the first set to run establishes a live bus, and every set after it has to be brought up to match that bus and then closed onto it. From that moment the sets are locked together – they run at one common frequency, and the bus voltage is common too. What is no longer common is how much of the load each one carries, and that becomes a control problem rather than a wiring problem.

The mechanical arrangement is simple: each set has its own circuit breaker, and all the breakers feed one busbar that feeds the load. What makes the installation a parallel system rather than a row of separate sets is the control layer between them – synchronising, load sharing, and protection that only makes sense when more than one source is connected.

A single set with a transfer switch and a parallel bus are two different engineering jobs, even when the sets themselves are identical.
Question One set with an ATS Two or more sets on a parallel bus
How the load is fed One source at a time; the transfer switch chooses All connected sets feed the bus at once and share the load
What decides the frequency The governor of the one set that is running The governors together; no single set sets the frequency alone
What decides the voltage The AVR of the one set that is running The AVRs together, through reactive load sharing
Redundancy None while the set is running, unless a second set is installed N+1 is possible: one set can be out of service and the load still carried
Extra equipment Transfer switch and its interlocks Paralleling switchgear, load sharing, synchronising and reverse power protection
What has to be commissioned Transfer and retransfer delays Governor settings, load share stability, protection settings and a load test on the bus

Four things have to match before the breaker closes

Closing a generator breaker onto a live bus is the only moment in the sequence that can damage something in milliseconds. The set has to arrive at the bus with the same voltage, the same frequency, the same phase angle and the same phase sequence. Two of those are fixed before the set is ever delivered: waveform is set by the alternator, and phase sequence is set by the alternator connections and the control wiring. The remaining three are what the synchronising equipment controls on every start.

The synchronising window that a paralleling controller is normally commissioned to. These are typical adjustable settings, not standard values.
Quantity Typical window What happens if it is wrong
Phase sequence Must match by construction; it is not a setting A reversed sequence puts system voltage against generator voltage and produces a heavy, damaging current on the first close
Voltage Within about 1 per cent of the bus A large voltage difference drives reactive current between the machines as soon as they are connected
Frequency Within about 0.1 to 0.2 Hz A set that is too fast or too slow takes or gives load the moment it closes, and the bus frequency swings
Phase angle Within about 5 to 10 degrees Closing far from the in-phase point transfers real power in one jolt and can trip the set on reverse power
Waveform Fixed by the alternator Dissimilar waveforms circulate harmonic current between sets on the same bus

The practical detail that catches people out is breaker closing time. The controller issues the close command, the breaker mechanism takes tens of milliseconds to actually make contact, and the phase angle drifts through that interval. A synchroniser that does not advance the close command to allow for it will close consistently late, which shows up as a small reverse-power kick on every start. It is a commissioning setting, and it is worth asking for it to be demonstrated rather than assumed.

The independent safety net is a check-sync relay: a device wired in series with the close circuit that simply will not pass the command unless the window is satisfied. An automatic synchroniser can be wrong; a check-sync relay set correctly is the thing that stops a wrong close from being a fault. For an operator closing by hand, the equivalent instrument is a synchroscope, whose pointer shows the set as fast or slow and is read as it approaches the in-phase position.

Two governors and one bus: how the load is shared

Once two sets are connected, two separate control loops decide what each one does, and they are easy to confuse. Real power in kilowatts is set by the engine governor: the set that is asked for more fuel produces more kW. Reactive power in kVAr is set by the alternator excitation: the set whose voltage is set slightly higher produces more kVAr. A parallel system that shares kW badly has an engine control problem; one that shares kVAr badly has an excitation problem. Both show up as current between the machines.

The classic mechanism for sharing kW is speed droop. A governor with 5 per cent droop is set so that the set runs at full-load speed at rated load and at 105 per cent of that speed at no load; the droop is the change in speed reference that produces full governor travel. Two sets of the same rating with the same droop curve share the load in proportion, because the common bus frequency forces them to settle at the same point on their curves. Change the droop on one machine and it takes a different share – which is exactly how a load-sharing controller trims the balance.

Isochronous governing is the opposite setting: no droop at all, so the governor holds one frequency regardless of load. A single set on an islanded bus wants that. Two sets in isochronous mode will fight over the frequency unless the controllers are tied together by a load-sharing line, which is what happens on the control panel of a modern multi-set plant: the governors are isochronous, and the kW balance is imposed by the load-sharing signal rather than by the droop slope. Droop is still the right answer where the sets run in parallel with the utility, because a set with no droop has nothing to tell it how much of the grid’s load it is carrying.

The three governing arrangements that turn up in quotations, and what each one is for.
Mode Where it is used What it does to load sharing
Droop (typically 4 to 5 per cent) Parallel operation, and any scheme where a set runs alongside the utility Sets with the same droop curve share in proportion to rating; the bus frequency falls slightly as the load rises
Isochronous (no droop) A single set on its own bus, where the frequency has to be held Frequency is held regardless of load, so two isochronous sets cannot share without help
Isochronous plus load sharing Islanded multi-set plants: data centres, hospitals, industrial base load The controllers hold the frequency and impose the kW split between the sets

Reactive sharing is the half that gets forgotten. Two alternators paralleled with the same voltage setpoint will not agree on kVAr; the one whose excitation is marginally higher pushes reactive current into the other. The fix is either voltage droop in the AVR or cross-current compensation, where a current transformer on each alternator feeds the reactive difference back into the AVR so that the two settle at a share. Without one of those two, the bus runs with a permanent circulating current that heats both machines and wastes fuel. It is also the reason a set is not simply “parallel-ready” because it has a synchronising relay: the excitation side has to be built for it as well.

One more number belongs in this section because it decides how many sets to buy: a diesel set that runs for long periods below roughly a third of its rating runs cool, and unburnt fuel and carbon build up in the exhaust – wet stacking. A parallel plant is therefore sized from both ends. The largest set must be able to carry the smallest credible load on its own, and the number of sets running has to be stepped up and down as the load moves, which is done by the load demand logic rather than by an operator.

Protection that only exists because there are two sets

Everything a single set needs – overcurrent, earth fault, over and under voltage, over and under frequency – is still needed on each set of a parallel plant. What is added is the set of protections that exist because the sets can drive each other. Each one of them is cheap next to the equipment it protects.

Protection that a parallel bus needs and a single-set installation does not.
Protection What it protects against How it is set
Reverse power A set whose engine has shut down, but whose breaker is still closed, being driven as a motor by the other sets A few per cent of rated kW, with a short time delay so that a synchronising transient does not trip it
Circulating current and reactive imbalance One alternator carrying the reactive load of the others because the excitation is not sharing Alarm on the imbalance, trip at a level the machines can survive
Loss of mains or busbar supervision Feeding a bus that is no longer healthy, and closing onto a fault Voltage and frequency supervision on the bus, permissive for every close
Short-circuit withstand on the bus A busbar fault fed by every connected set at once The fault contribution of the sets adds up, so the switchgear, cable and settings are rated for the total
Load shedding and load demand Sets running below their minimum load, or the whole load landing on the first set to close Load is added in blocks as sets come on line, and non-essential boards are shed on the way down

The fault level point deserves to be said again in plain terms: a bus fed by four sets can deliver roughly the sum of their short-circuit contributions, and that is the number the switchgear and the protection coordination have to survive. It is the reason a parallel installation is not simply a single-set installation with more sets attached – the electrical design changes, and so does what the protection has to be able to do. The same is true of earthing: in a parallel system only one neutral-to-earth bond may exist, or part of the return current takes the earth path instead of the neutral.

What has to change on the set itself

A set that is going to be paralleled is ordered differently from a standby set of the same rating. The engine and alternator may be identical; the control and protection around them are not.

  • A paralleling-capable controller. Synchroniser, kW and kVAr load sharing, reverse power, and the bus supervision all live in or around the controller. A standard AMF panel does not have them and cannot be programmed to acquire them.
  • Current transformers on each alternator. Load sharing and reactive sharing both need to know what each machine is producing, which means CTs and the wiring to use them.
  • A breaker that can be operated and vetoed. Motor-operated or solenoid-closed, with a shunt trip, and with the close circuit routed through the check-sync relay so that a faulty controller cannot close a set out of step.
  • A paralleling panel and busbar. Sized for the total continuous current of the plant and for the total fault contribution of every set connected to it, not for one set.
  • An excitation arrangement that can share reactive load. Voltage droop or cross-current compensation in each AVR, otherwise the alternators circulate current between themselves.
  • A commissioning plan with a load bank. Load sharing stability, governor settings, protection settings and the synchronising window are all demonstrated under load; a parallel plant that has never been load tested is an untested design.

The retrofit question comes up often enough to answer directly: a set bought without a paralleling controller can be brought into a parallel system later, but it means replacing the controller and adding the load sharing, CTs and protection – so it is cheaper to specify parallel-ready at the quotation stage even if the second set is two years away. If the plant is going into a container package, the control panel is part of the package and the parallel option is settled with the box rather than later – see containerized diesel generators. The same calculation is set out for a specific rating on the 300 kVA page.

One large set, or two smaller ones

This is the decision that paralleling exists to serve, and it is not decided by price alone. Redundancy, transport, the shape of the load curve and the maintenance regime all push in the same direction: at the top of the range, sites buy in multiples rather than in one machine.

One machine against two smaller ones, for the same load and the same duty.
Question One large set Two or more smaller sets in parallel
Capital cost Lower for the machines and the switchgear Higher: more machines, a paralleling panel, more protection and more commissioning
Redundancy None. A fault or a service means an outage N+1 is available, and a set can be serviced while the load runs
Transport and handling One heavy lift; at high ratings the weight decides the route Smaller, lighter packages that reach sites a single large set cannot
Part-load running The set runs light whenever the load is light, and light running is where wet stacking starts The control system adds and removes sets, so the running sets stay in their efficient band
Footprint One large machine, one large radiator, one large exhaust More floor area, more exhausts, and a paralleling room or panel
Efficiency at the design point Usually the best fuel figure for a steady full load Slightly worse at full load, better across a varying load

The arithmetic that matters is the redundancy rule. With two sets, N+1 means each set can carry the whole load on its own, so both are rated at the full load. With three sets, N+1 means any two can carry the load, so each is rated at about half. A 1000 kVA load is therefore either two 1000 kVA machines or three 500 kVA machines, and the three-set arrangement usually wins on transport, on maintenance and on part-load running – which is why the larger ratings on this site are normally discussed as multi-set plants rather than as single machines: see two 500 kVA sets in parallel, the single-or-parallel decision at 400 kVA, parallel control at 1000 kVA and multi-megawatt halls built from 2000 kVA sets. The duty of each set in the group still has to be settled first, because a set that carries base load is a different machine from one that only runs when the mains fails – see prime and standby ratings. Data centre standby plants are quoted this way from the first enquiry, in N+1 or 2N groups rather than as one machine – see data centre power.

When paralleling is the wrong answer

Paralleling buys redundancy and flexibility, and it charges for both in switchgear, engineering and commissioning. It is the wrong answer when the site cannot use what it buys.

  • The load is small. If the site’s critical load is one 200 kVA board, two sets are two machines to maintain in exchange for redundancy that a single set plus a service contract can provide more cheaply.
  • Nobody will run it. A parallel plant has governor settings, load sharing, protection settings and a synchronising window. An installation with no maintenance regime will have all of those drift, and a parallel bus is less forgiving than a single set.
  • The load curve is flat and steady. The part-load argument disappears when one set runs at its rated point all day; the extra switchgear then buys only redundancy.
  • The site cannot take the minimum load. If the smallest credible load is below the minimum stable load of one set, adding sets makes the fuel and carbon problem worse rather than better.
  • The switchgear has nowhere to go. A paralleling panel, its protection and its cabling need space and a controlled environment; a plant room that has none should not be planned around them.

What to put in the specification

The items below are the ones that decide whether a parallel plant works. Each of them is cheaper to settle at quotation stage than to add later, and the last one is the one most often left out.

The specification checklist for a parallel installation.
Item What to state
Bus arrangement How many sets, whether the redundancy is N+1 or N, and whether the bus is single or split
Rating and duty of each set The prime or standby rating of each machine, and the load each one has to be able to carry alone
Governing mode Droop, isochronous, or isochronous with load sharing; and the droop percentage if droop is used
Reactive sharing Whether the AVRs use voltage droop or cross-current compensation, and the CT ratios that go with it
Synchronising Automatic synchroniser plus an independent check-sync relay, and the window the controller is set to
Protection Reverse power, circulating current or reactive imbalance, bus supervision, and load shedding
Breakers Motor-operated or solenoid-closed, with shunt trip, and the close circuit routed through the check-sync relay
Busbar and switchgear Continuous current and short-circuit withstand for the whole plant, not for one set
Earthing and neutral One neutral-to-earth bond in the system, and whether the neutral is switched
Commissioning A load bank test, a load sharing demonstration, injected protection tests, and the settings handed over in writing

Common mistakes

  • The sets were rated and the bus was forgotten. The fault contribution of every connected set adds up. Switchgear, cable and protection coordination have to be rated for the total, and that is a different exercise from sizing one set.
  • Two isochronous governors and no load sharing. Two sets in isochronous mode with nothing between them will hunt for the frequency between themselves. They need a load-sharing signal, or droop, or both.
  • No check-sync relay. The automatic synchroniser is allowed to be wrong; the check-sync relay is the device that makes being wrong harmless. It is one relay, and it is the difference between a trip and a fault.
  • “Parallel-ready” assumed from a synchronising relay. Reactive load sharing is a separate problem from synchronising, and it needs the excitation side – AVR droop or cross-current compensation – to be built for it.
  • Droop left at the factory default. If the curves do not match the load does not split evenly, and one set carries more than its share for the whole life of the plant. The settings belong in the commissioning record.
  • No one checked the minimum load. A parallel plant with a small night load runs its sets cold. The number of sets and the load demand logic have to be designed around the smallest load, not only the largest.
  • The retrofit was left until later. A set bought without a paralleling controller cannot be connected to another one by wiring alone; the controller and the protection have to be replaced. Specify it with the first set.

Questions

Can any two diesel generators be paralleled?

No. Both sets need a paralleling-capable controller, matching phase sequence, and the load sharing and protection that go with a parallel bus. Two ordinary standby sets with AMF panels can be installed side by side, but they cannot be connected to the same bus.

How close do the voltage and frequency have to be before the breaker closes?

The window is set in the controller and is typically a voltage difference of about 1 per cent, a frequency difference of about 0.1 to 0.2 Hz, and a phase angle of about 5 to 10 degrees, with the close command advanced to allow for the breaker’s closing time. The settings are adjustable project by project; the synchroniser and the check-sync relay together decide whether the close happens.

What is the difference between droop and isochronous governing?

With droop, the governor lets the speed fall slightly as load rises – typically 4 to 5 per cent from no load to full load – and that slope is what makes several sets share in proportion. With isochronous governing the speed is held constant regardless of load, which is what a single set on an islanded bus needs; several isochronous sets need a load-sharing controller between them to agree on the split.

Do I need a separate synchronising panel?

Not necessarily. On modern sets the synchroniser, load sharing and reverse power protection are built into the controller and the paralleling panel carries the breakers, busbar and protection. What cannot be left out is the independent check-sync relay on the close circuit, wherever the synchronising function physically lives.

Can two sets of different ratings run in parallel?

Yes. With equal droop the load splits in proportion to rating, so a 400 kVA and a 200 kVA set share roughly two to one. Load-sharing controllers can also be set to a deliberate ratio. What matters is that the smaller set is not asked to carry more than its share, and that the protection is coordinated for the pair.

Can a set be paralleled with the utility?

It can, but that is a different design from paralleling two sets. It needs droop governing, synchronising against the utility, and the utility’s agreement, and it is normally done for peak shaving or load curtailment rather than for standby. The transfer arrangements for a site that runs alongside the mains are covered in ATS vs AMF.

Give us the load schedule, not the set list

Send the load schedule, the redundancy you need, the duty of each set and the site conditions, and we will come back with the number of sets, the rating of each one, the bus arrangement and the controls that go with it.

Request a quote WhatsApp the sales desk

Written and maintained by the KVA Grid sales engineering desk. The synchronising conditions – equal voltage, frequency, phase angle and phase sequence, with waveform fixed by the alternator – and the droop definition (a governor with 5 per cent droop runs at full-load speed at rated load and 105 per cent at no load, and sets with equal droop share in proportion to rating) follow the published descriptions of synchronisation and droop speed control. Duty ratings follow ISO 8528-1, and transient performance classes follow ISO 8528-5. Synchronising windows, droop percentages, protection settings and minimum load figures are typical adjustable values set project by project and are not a guarantee; the datasheet, the protection study and the settings of the quoted assembly govern. Reviewed 2 October 2026.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top
WhatsApp