Generator Load Management: Compare Dispatch Options Without Assuming One Ideal Load

A factory can use less fuel with one generator at a moderate load than with two sharing that load. Yet stopping a set may also remove the reserve needed for the next motor start or an unexpected failure. The useful question is which combination can supply the required service at the least justified cost.

Generator load management therefore needs more than a preferred load percentage. It needs a demand profile, fuel evidence and a clear account of what happens when demand changes. The worked case below separates those questions. The case uses invented machine ratings, fuel figures, prices and operating windows to teach the method.

Define the service before comparing fuel

Start with the loads that the generator system must supply. Some processes can wait briefly for another set to become ready. For others, an interruption may ruin a batch or erase a machine's reference position. It may also stop a required control function. List the required service and the permitted response for each group.

Keep that service requirement separate from the current operating habit. Running two sets all day may have a sound reliability reason, or it may simply be the way the system was first used. A fuel comparison should make the reason visible before proposing a change.

Next, identify the rating that applies to the actual duty. A standby rating is not automatically the rating for sustained factory supply. Caterpillar's C18 specification sheet shows why the label matters. It distinguishes duty ratings and states the conditions behind its performance data. Use the installed set's own approved rating and site conditions. Include any limits that apply to its operation.

Do not replace that review with a rule that every generator should run at one ideal percentage. Caterpillar's underloading discussion explains the concerns with running at light load for long periods. It points to operating guidance for the equipment concerned. It also distinguishes engine types and applications. It is not a universal load target for every diesel or gas set.

The practical output is a permitted operating envelope for each installed unit. Record the real-power and apparent-power limits that apply. State the permitted load range for the intended duration. Note any other restrictions on the proposed schedule. An economic model can compare options within that envelope. It cannot establish which operating states are permitted.

Describe demand on the timescale that matters

A daily electricity total cannot show whether two large loads start together. Assemble time-stamped power records and a separate list of material load changes. The sampling interval should reveal the events that the proposed control must handle.

For each large load, record its normal running demand and starting behaviour. Can its connection be delayed? A known compressor start can be requested in advance. An unannounced load change requires a different response. Do not hide the difference inside one average kilowatt figure.

Use the same electrical boundary for the load profile and the fuel curve. One record may include generator auxiliaries while another excludes them. Resolve that difference before comparing the options. Preserve the useful service delivered to the factory as a separate check.

A production plan can help explain the profile, but it is not a substitute for measurement. A scheduled machine may remain off while another runs longer than expected. Compare the proposed schedule with actual starts, stops and changes of operating state.

Also identify the information that is missing. Unknown load-step behaviour should remain an open constraint. Calling it a small percentage of nameplate capacity does not resolve it. A dispatch recommendation is only as strong as the conditions that it has checked.

Screen stable operating windows with a fuel model

Suppose a factory has two identical generator sets. Each has an applicable limit of 500 kW and 625 kVA under the assumed site conditions. For this example, operation from 100 to 500 kW per set has been accepted for the duty being studied. That range belongs only to this fictional case.

Assume measurements of these units support this simple fuel relation within that range:

F(P) = 12 + 0.22P

Here, F is litres per hour for one running set and P is its electrical output in kW at the stated boundary. An off set consumes no fuel in these stable windows. Starts, transitions and cooldown are considered separately.

This is a teaching model, not a published engine curve. HOMER's fuel-curve explanation describes a common approach: fit a straight line to load and fuel data. The slope represents the increase in fuel rate with output. Such a fit should only be used where the underlying data support it. A curved or poorly fitted response needs a different calculation.

The three stable demand windows are:

These windows provide a screening comparison. They do not describe an instant change from one running set to two. The real sequence between windows still needs to be designed and assessed.

Option A: keep both units running

With equal real-power sharing, the sets each supply 150 kW in the lower-demand window. One set then uses 12 + 0.22 × 150 = 45 L/h. Both together use 90 L/h, or 270 litres over three hours.

At 600 kW total demand, each set supplies 300 kW. Its fuel rate is 78 L/h, giving 156 L/h for the pair. Five hours use 780 litres.

At 800 kW demand, each supplies 400 kW. Each uses 100 L/h, so the pair uses 400 litres over two hours. Total fuel for the three stable windows is 1,450 litres.

The units are at 30%, 60% and 80% of their assumed 500 kW limit in these windows. Those percentages describe the case. They do not establish a universal ranking of acceptable generator loads.

Option B: use one unit during the lower-demand window

A single set supplying 300 kW uses 78 L/h under the same model. Over three hours it uses 234 litres. The middle and higher windows still require both sets and use the same 780 and 400 litres as Option A.

The potential stable-window reduction is 36 litres. Both options supply the same electricity at the same stated boundary. Divide by that common output to compare fuel per unit of electricity. The modelled windows give about 0.2636 L/kWh for Option A and 0.2571 L/kWh for Option B.

Explain why the model produces that difference

For two running units, adding their curves gives:

Combined fuel rate = 24 + 0.22 × total kW.

For one running unit, the fixed part is 12 rather than 24. The lower-demand difference is therefore 12 L/h, which explains 12 × 3 = 36 litres. The assumed fuel curves produce this result. No preferred load percentage enters the calculation.

Within this identical linear model, assigning 200 kW to one unit and 100 kW to the other would still use 56 + 34 = 90 L/h. Changing the split does not remove the second running unit's fixed fuel term. This arithmetic is not an instruction to override the installed load-sharing control.

With unequal units or curved fuel responses, the answer may change. Calculate each unit's fuel at its assigned load, then add the results. Keep the capacity and operating constraints attached to every proposed allocation.

Check the part of the curve that drives the choice

The important comparison here is one set at 300 kW against two sets at 150 kW each. Strong evidence at full load alone does not establish either side of that comparison. Collect evidence at these operating states before relying on the predicted difference.

Check the fit against points that were not used to establish it. If the errors are large near the proposed stopping region, improve the model or narrow its use. A smooth curve can look convincing on a report. Yet the evidence may be weak exactly where the dispatch decision changes.

The two units are identical in this teaching model. In a real comparison, units with equal nameplate ratings may have different fuel curves. Keep each unit's evidence separate until the data justify combining them. Also preserve the load basis and operating conditions when using a supplier curve.

Finally, report the fuel benefit at an appropriate precision. The calculated 36 litres is exact within the invented model. A measured forecast with uncertain inputs should carry that uncertainty into the decision. A small predicted benefit may need a better trial. It may not yet justify repeated starts or a reduced reserve.

Include the transitions before claiming a saving

A load that needs both units cannot be accepted until the second set is ready. Starting, synchronizing, taking load and later unloading take place through an approved sequence. Stopping a unit also has consequences beyond the stable fuel rate.

Cummins' paralleling application manual explains how load-dependent start/stop affects the capacity available online. Its discussion includes stabilization, load transfer and cooldown. Evaluate these operating stages for the site. This article cannot prescribe their delays for another installation.

The worked comparison calls the additional fuel penalty C. It covers Option B's complete transitions compared with Option A. It must be assessed for the same useful service. Include the net effect of overlap, ramping, starts and cooldown. Reconcile any change in useful energy delivered during those stages.

C is not simply the fuel used during one start. Some fuel would have been used to supply the load under Option A during that same period. Compare the two complete options, including any extra auxiliaries. Simply adding a gross event total can misstate the difference if the boundaries do not match.

Suppose that a valid whole-cycle comparison gives an additional penalty of 10 litres. This is another invented assumption. The net fuel reduction becomes:

Net reduction = 36 − 10 = 26 litres.

At an assumed fuel price of 120 currency units per litre, that is 3,120 currency units. It is a fuel-cost result only. It excludes changes in maintenance cost, spending on controls and the cost of an interruption.

If the additional penalty were 40 litres, the same proposal would increase consumption by 4 litres. The stable-window benefit is too small to cover that penalty. A claim based only on the three low-load hours would miss this reversal.

The model saves 12 L/h during the stable window. With a fixed 10-litre transition penalty, the low-demand window must last 10 ÷ 12 = 0.8333 hours, or 50 minutes, to break even on fuel. This is a screening threshold, not a controller timer. Real transition costs may change with conditions. Reliability needs can rule out stopping even during a much longer window.

Check capacity after the event you need to withstand

Online spare capacity and protection against a unit failure are different measures. Two 500 kW units supplying 800 kW have 200 kW of unused steady-state capacity in total. If either unit fails, the remaining 500 kW unit cannot supply all 800 kW. The shortfall is 300 kW.

Suppose the factory identifies 450 kW of essential load within that 800 kW total. A single surviving unit has 50 kW of steady-state margin above that essential group. But keeping it supplied requires 350 kW of other load to be removed in time. An acceptable control response and transient performance must still be demonstrated.

At 600 kW total demand, loss of one unit leaves a 100 kW shortfall if all demand is retained. Two engines alone do not make this a redundant supply. That label would conceal the limit.

The lower-demand window has a different tradeoff. One 500 kW unit supplying 300 kW has 200 kW of unused steady-state capacity. Yet failure of that running unit removes the active source. A stopped second unit is available equipment, but it is not already supplying power.

With both units online at 300 kW, one remaining unit has enough steady-state rating to carry 300 kW. That does not prove it can accept the sudden change. Voltage and frequency must remain within the factory's requirements.

Caterpillar's transient-performance discussion distinguishes site load steps from general performance labels. The specific load sequence and acceptable recovery need agreement and suitable evidence. A nameplate sum cannot establish the performance of that sequence.

Keep the required contingency beside each dispatch option. An option may suit production that can pause. It may still fail to meet the essential group's needs. That is a service decision with a fuel consequence, rather than an error in the fuel arithmetic.

Check kVA as well as kW

A generator can be within its real-power limit while exceeding another electrical limit. Schneider Electric's apparent-power guidance relates real power, apparent power and power factor. With compatible measurements, kVA = kW ÷ power factor.

Return to the fictional 800 kW window. Assume balanced conditions. The identical units share both real and reactive demand equally. At a power factor of 0.8, the total is 1,000 kVA, or 500 kVA per unit. Each also supplies 400 kW. Both values fit the assumed 500 kW and 625 kVA limits.

At a power factor of 0.6, the same 800 kW requires about 1,333.33 kVA, or 666.67 kVA per unit. Real power remains 400 kW per unit. Apparent power now exceeds 625 kVA.

Under those assumptions, the 625 kVA limit at a power factor of 0.6 permits only 375 kW per unit, or 750 kW for the pair. That leaves 50 kW of the proposed 800 kW demand unsupported by this simple limit check.

This example does not size a complete electrical system. Engineers still need to review phase currents, imbalance and harmonics. Protection and the actual alternator capability also need review. It shows why a controller decision based on aggregate kW alone can miss a constraint.

The Cummins manual explains the two forms of sharing. Engine control governs real-power sharing; excitation control governs reactive-power sharing. For unequal units, equal percentage loading does not mean equal kW. Keep the installed system's approved sharing arrangement. Investigate abnormal sharing before changing dispatch.

Turn the comparison into an operating sequence

Can the plant still online carry the load after the proposed stop? That is the useful control question. A spare-capacity check made only before stopping can mislead. The stopped unit's capacity then disappears.

DEIF's power-management manual illustrates this distinction in a defined controller family. Its load-dependent start/stop functions account for available capacity and delays. They use separate start and stop limits. It also describes heavy-consumer requests that wait for capacity before permission to connect. These are examples of control functions, not ready-made settings for another controller.

In the worked case, a known change from 300 to 600 kW cannot be connected while only one 500 kW unit is ready. It exceeds that unit's steady-state limit by 100 kW even before considering the load step. The second unit must reach an accepted state before permission to connect is granted. Otherwise, the plan needs an approved alternative sequence.

Different start and stop conditions can also prevent repeated cycling near one threshold. However, a delay should not leave essential demand exposed while the controller waits. Document the event that starts the request and the signal that confirms readiness. Also state what happens if the requested unit fails to become available.

Review the reverse sequence too. A fall in demand may justify unloading one set, but a brief dip may end before the stop is useful. Preserve the unit's required shutdown process and the site's next expected load change. The 50-minute economic illustration does not replace these checks.

An operator should be able to explain the permitted states. Doing so should not require changed protection settings or improvised live switching. The approved sequence, alarms and fallback state belong in the controlled operating instructions. Changes to those functions need the responsible controls and electrical personnel.

Add solar or storage with a new balance

Solar output can reduce the generator's load, but it also changes the power that the running units must balance. Woodward's PV-with-generator guidance discusses reserve, synchronization and load management in an integrated arrangement. It does not make a fixed fuel-saving percentage valid for every factory.

If solar or a battery is added, replace the simple demand profile with a time-based balance at the agreed bus. Account for charging, discharging, conversion losses, state-of-charge limits and the required reserve. An empty battery cannot provide backup when the running set fails. Reserved energy is part of the service being compared.

Keep the comparison on equal useful service and stated starting and ending conditions. An option may otherwise seem cheaper because it ends with less stored energy. The two-generator case above includes no solar or storage contribution. It cannot serve as a hybrid-system forecast.

Release a measured operating change

Before a trial, retain the baseline fuel, electricity and event records with the same time boundaries. Agree on the permitted production states and the conditions that end the trial. The evidence should reveal both the expected fuel effect and any change in service.

During the trial, record which units run and their kW and kVA. Log starts, connection permissions, alarms and completed transitions. Compare the observed low-demand duration with the plan. Reconcile fuel across the complete comparison cycle. Include the periods omitted from the stable-window screen.

Judge the trial against its required load and contingency cases. A lower fuel total during a quieter shift does not establish a saving from dispatch. Nor does a successful ordinary transition prove the response to a failed start or loss of a running unit.

Maintenance condition is a separate influence on the result. ApparelDiary's preventive and predictive maintenance guide can help organize condition evidence without treating a dispatch change as a repair. Keep applicable service requirements attached to the revised operating plan.

The running-hour record should also remain distinct from elapsed factory time. In the ten stable hours of this example, Option A accumulates 20 engine-hours across the pair. Option B accumulates 17 engine-hours: one unit runs for three hours, then two run for seven hours. Both supply the same ten hours of modelled service.

That reduction of 3 engine-hours, or 15%, does not establish a 15% reduction in maintenance cost. The work due depends on the relevant service basis and the condition of each set. Some costs may remain unchanged during the comparison. This stable-window count excludes transition time. Add that time to the complete operating record.

Record starts and running hours by unit, including which set leads during the lower-demand period. An aggregate total can hide one unit doing most of the work. If the approved plan rotates the lead unit, evaluate that sequence. Equal final hour counts do not guarantee the same operating result.

If the proposal needs new controls or equipment, carry the verified operating cases into an investment comparison. ApparelDiary's capacity investment decision-tree guide explains how to compare uncertain scenarios. Count each reliability benefit once. Retained production and a separate avoided-loss allowance may describe the same benefit.

Release the change with its measured benefit, permitted states and fallback instructions. A useful result may be permission to stop a unit only during particular stable windows. That precise operating rule can be more useful than a broad fuel-saving promise. It accounts for the next load event.

References

• Caterpillar: C18 specification, LEHE1891-04.

• Caterpillar: The impact of generator-set underloading.

• HOMER: Fuel Curve.

• Cummins: Paralleling Application Manual, T-016.

• Caterpillar: Transient-performance specifications for diesel generator sets.

• Schneider Electric: Installed apparent power.

• DEIF: G4, G5 and G8 power management, document 4189340696H.

• Woodward: Photovoltaic systems with generator sets.