October 26, 2026
Almost every failed power factor project fails at the selection step rather than the installation step. A fixed capacitor installed on a load that empties at night, a contactor-switched bank on a welding line, or an SVG bought for a hotel — each is a correct product in the wrong place, and each costs more to fix than it cost to buy.
This article separates the three topologies that actually exist in commercial and light industrial compensation, gives the conditions under which each is the right answer, and provides a decision table by load type. It then works the sizing arithmetic, because kvar is not kW and the two are constantly confused.
Fixed capacitor. A capacitor permanently connected, sized for the average load: the cheapest possible correct answer, with no moving parts, that over-compensates whenever the load drops.
Automatic bank. Several capacitor steps switched in and out by contactors or thyristors under a controller that follows the measured power factor: the industrial default, and the right answer for most commercial sites.
SVG (static var generator) or STATCOM. A power-electronic converter that injects or absorbs reactive current continuously, within milliseconds, in both directions: the most capable option and the most expensive by a factor of three to five.
Everything below is an elaboration of these three sentences. 
The fixed capacitor has a bad reputation because it is so often misapplied, not because it is obsolete. It is the correct choice when four conditions hold:
Two failure modes need to be designed out. The first is self-excitation: a fixed capacitor left connected to a motor that has been switched off will excite the still-spinning motor, which then generates voltage. On a single large motor this is a genuine hazard, and the rule is to keep the capacitor below the motor's no-load reactive demand or to interlock it with the contactor.
The second is night-time over-compensation. On a site that drops to near zero load, a fixed bank exports reactive power back to the utility, the power factor goes leading, voltage rises, and on a kVA-billed tariff the customer pays for the privilege. This single behaviour is why the fixed capacitor has such a poor reputation in restaurants and hotels, where it should never have been installed in the first place.
An automatic bank measures power factor — or, better, reactive current — at the incomer and switches steps in and out to hold a target, usually 0.95 to 0.98 lagging, with a deadband wide enough to stop the controller hunting between two adjacent steps.
The implementation choice that matters most is the switching element. Contactor-switched steps are cheap and robust, but each step takes twenty to sixty seconds to respond and the contacts wear; this suits slow, predictable variation such as restaurants, hotels, malls and cold stores. Thyristor-switched steps respond within a cycle, have no contact wear, but cost more and dissipate their own losses and heat; they are required where the load changes within seconds, as on welding equipment, lifts, cranes and some process lines.
Four design details separate a good bank from a bad one, and all four are routinely omitted from cheap quotations:
An SVG is a voltage-source converter with a coupling reactor. It produces whatever reactive current the controller asks for, continuously, and it can do six things a capacitor bank cannot:
The costs are equally concrete: roughly three to five times the price of a contactor-switched bank of the same rating, continuous losses of about 1% to 2% of its own rating, and a shorter service life for the power electronics than for a capacitor. An SVG earns that price where the load swings within seconds, where the network is weak (low short-circuit capacity, long feeder), where harmonic distortion is already high, or where the utility specifies a response time. It is not the economical choice for a hotel with air conditioning, and buying one there is simply overpaying.
The table below assumes a three-phase low-voltage site with a tariff that charges for kVA demand or reactive energy. If the tariff does neither, read the last section first, because the answer may be that no compensation is justified at all.
| Load type | Power factor variation | Harmonic risk | Recommended | Why |
|---|---|---|---|---|
| Restaurant, hotel, spa (HVAC, hot water, lighting) | wide, 0.70–0.95 across the day | low unless many drives | Automatic bank, 6–8 steps | the load swings several times a day; a fixed bank would over-compensate at night |
| Supermarket, shopping mall (central AC, refrigeration) | moderate | moderate — many small drives | Automatic bank with detuned reactor | continuous operation, but compressors cycle in and out |
| Internet café, gaming centre (many switch-mode supplies) | moderate | high — 3rd and 5th | Automatic bank with detuning; SVG if THD is very high | switch-mode supplies produce triplen and fifth harmonics |
| Factory, motor-dominated, steady shifts | narrow | low | Fixed capacitor, or a small automatic bank | the load is predictable; a fixed bank is the cheapest correct answer |
| Factory with many six-pulse VFDs | moderate | high — 5th and 7th | Detuned automatic bank, or active filter plus bank | a standard bank risks parallel resonance |
| Welding shop, press line, crane | very fast, within seconds | moderate | Thyristor-switched bank or SVG | contactor steps cannot follow a load that changes this fast |
| Water pumping, one large motor | narrow | low | Fixed, interlocked with the motor | constant load, but the capacitor must not self-excite the motor |
| Site already at 0.95 or above | — | — | No compensation | there is no reactive power left to remove |
| 480 V or 200 V service | — | — | Not the 380–415 V model | voltage mismatch; a different product is required |
THD = total harmonic distortion. Detuned reactor percentages assume a 50 Hz system; on 60 Hz the same percentage moves the series resonance to a proportionally higher frequency.
The most common sizing error is treating the capacitor rating as if it were a power rating in kW. It is not; it is reactive power, and it must be calculated from the active power and the two power factor angles:
Qc = P (tan φ₁ − tan φ₂)
Worked at four sizes, going from the present power factor to a target of 0.95:
Two corrections are then applied, and both are regularly forgotten.
Voltage. A capacitor's output scales with the square of the applied voltage: Qactual = Qrated × (Uactual ÷ Urated)². A unit rated 50 kvar at 400 V delivers 50 × (380 ÷ 400)² ≈ 45 kvar on a 380 V busbar. Every kvar figure in a quotation should state the voltage at which it applies.
Target. Raising the target from 0.95 to 0.98 costs roughly as much capacitance as going from 0.80 to 0.95, for a far smaller saving. 0.95 is the usual economic point; 0.98 only where the tariff explicitly demands it. And the cabinet rating is not the same as the kvar: the cabinet must be rated for the current, which depends on voltage and on harmonic content.
The decision table is a starting point, not an answer. Three situations override it.
1. Very low short-circuit capacity. On a long rural feeder or a small transformer, the same kvar produces a much larger voltage rise, and the resonant order moves down because Ssc is smaller. Recalculate h with the real Ssc before choosing a bank, and expect the switching steps to need to be smaller.
2. Predominantly single-phase load. Small shops, salons, small restaurants and internet cafés often have most of their load on one phase. Three-phase compensation leaves that imbalance untouched: the neutral current and the negative-sequence losses remain. What is needed is phase-to-phase or per-phase compensation, or a unit with active unbalance correction.
3. A tariff with no reactive or kVA charge. Then compensation saves only line losses, and the payback stretches from two years to something that may never arrive. Check the bill before the brochure — and note that a supplier who has not asked to see the bill has not done the calculation.
Our power loss regulating equipment sits in the automatic class. It connects in parallel to the distribution circuit without modifying existing wiring, adjusts compensation current continuously to the measured load rather than in fixed steps, corrects three-phase unbalance to within 2%, and is rated 15–100 kW at 380–415 V, 50/60 Hz, with over-voltage, over-current and over-temperature protection, an aluminium alloy enclosure with industrial cast-resin sealing, and an 8 to 10 year design life. The published saving range is 10% to 20%, which depends — as this article has argued throughout — on the starting power factor, the tariff and the load profile.
Where it fits: restaurants, hotels, internet cafés, small factories, malls and supermarkets with fluctuating inductive load, a power factor below about 0.85, and a tariff that charges for kVA demand or reactive energy. The continuous adjustment is the reason it suits sites whose load changes several times a day, which is most of the list.
Where it does not: 480 V or 200 V services, where the voltage does not match; sites already at 0.95, where there is nothing left to remove; sites with very high harmonic distortion, which need the measurement sequence described in the companion article before any product is chosen; and any site expecting a 20% reduction on a bill that has no reactive component at all.
The full specification and the selection route are on the power loss regulating equipment page.
Five items are enough to choose between the three topologies rather than guess:
With those we will say which of the three topologies applies, what rating is needed, and — if the honest answer is that none of them pays back on your tariff — that too.
Tell us your supply type, transformer capacity and main loads, and we will match a suitable model and send a quotation within 24 hours.
Get a Free QuoteParallel connection, 15–100 kW, 380–415 V / 50–60 Hz, continuous compensation with three-phase unbalance correction to within 2%, triple protection and an 8–10 year design life.
See specifications and pricing →