Why the PCS decides whether a storage project earns or burns

The battery gets the headlines; the PCS decides the settlement statement. A two-hour system cycling twice a day moves roughly 26 GWh of energy per year for a 40 MWh plant — a single point of round-trip efficiency is therefore worth about 1% of gross arbitrage revenue, and a badly matched DC/AC ratio can clip far more than that. According to BloombergNEF, global battery storage turnkey prices fell to roughly USD 165/kWh in 2024, which means margins are now thin enough that conversion losses dominate the business case rather than capital cost.
There is a second, harder cost. Grid-code non-compliance discovered at commissioning means re-testing, firmware rework and delayed commercial operation — typically 4 to 12 weeks of zero revenue, plus penalties under the grid connection agreement. In several provincial markets, a plant that fails the primary frequency regulation (一次调频) test simply does not receive its dispatch priority, and therefore never captures the peak spread it was financed against.
The practical rule we give clients: select the PCS before finalising the battery DC layout, not after. The PCS datasheet sets the DC voltage window, the maximum DC current per input, the reactive capability window and the protection settings — five constraints that flow backwards into module string design, cable sizing and even the container footprint.

Sizing the PCS: overload margins and the DC/AC ratio trap

Storage is not PV, and copying PV sizing habits is the single most common error we see. In PV, a DC/AC ratio of 1.2–1.3 is normal because output is clipped only near noon. In storage, a DC/AC ratio above roughly 1.1 starts to throttle discharge power at the exact moment the price is highest — and it throttles charge power too, extending the cycle and cutting daily throughput.
A reasonable starting range for a 2-hour lithium iron phosphate (LFP) system is 1.00 to 1.08 DC/AC, with the final number driven by the dispatch profile, not by habit. If the plant sells frequency regulation (调频) or runs a high-rate primary frequency response service, size up: those services require 1.5–3C bursts and a PCS with at least 1.2× continuous overload for 10–60 seconds.
Be equally careful with headline efficiency. Manufacturers quote peak number, usually 98.5–99.0% for a central PCS and 97.5–98.5% for a string PCS, measured at nominal DC voltage and 50–75% load. Real annual efficiency sits lower because the plant spends hours at 10–20% load during the top-up phase. Ask for the weighted efficiency curve across 5–100% load, and verify it against the Chinese standard GB/T 36548 test method.
| Configuration | Typical peak efficiency | Best fit | Main drawback |
|---|---|---|---|
| Central PCS, 2.5–3.5 MW | 98.7–99.0% | Large plants, low cost/W | Single point of failure, long DC runs |
| String PCS, 100–250 kW | 97.5–98.5% | Distributed, containerised, rooftop | More devices, higher BOS cost |
| Modular/stacked, 50–125 kW | 98.0–98.6% | Flexible expansion, redundancy | Firmware complexity, spare-part mix |

Grid-code compliance: the GB/T stack that trips up importers

China's grid requirements are published, mandatory and strict. The core documents are GB/T 36547 (technical requirements for connecting electrochemical storage to the grid) and GB/T 36548 (test specification), supplemented by provincial dispatch rules and by GB/T 19964 for the converter interface itself. A PCS certified only to IEC 62933 or UL 9540 will not automatically pass a provincial acceptance test. This is the most expensive mistake in the category.
The specific gaps usually appear in four places: the voltage and frequency ride-through envelope, the reactive power response time, the active power ramp rate after a dispatch instruction, and the communication protocol. Chinese dispatch typically requires the plant to follow an automatic generation control (AGC) instruction within 1 second and to reach 90% of the target within 10 seconds. Many imported PCS units specify a 30-second ramp by default.
Communication is the quiet killer. Provincial dispatch centres commonly require IEC 61850 plus a dedicated telemetry channel, with specific point tables. If the PCS only exposes Modbus TCP, the energy management system (EMS) integrator has to build a protocol gateway — practical, but it adds latency and a failure point that shows up during the frequency response test.

Reactive power, harmonics and fault ride-through in practice

A PCS must deliver a continuous power factor range of −1.0 to +1.0 at full active power. In practice, many units can only reach ±0.95 at 100% load and hit ±1.0 only below 50% load, because the converter current limit is shared between active and reactive components. On a 100 MVA plant, that difference can mean the difference between passing and failing the voltage support requirement.
Harmonic performance matters more than the datasheet suggests. GB/T 14549 limits total harmonic distortion (THD) at the point of common coupling to around 3–5% depending on voltage level, and a PCS that quotes 3% THD at rated load can drift to 5% at 20% load. Where the grid is weak — short-circuit ratio below 3 — add active harmonic filtering or specify a PCS with a higher switching frequency and an LCL output filter.
Fault ride-through is now a hard requirement, not an option. Plants must stay connected through low-voltage and high-voltage disturbances and inject reactive current during the dip, following a defined Q-U curve. Verify that the PCS has the curve implemented in firmware and that the test report shows measured curves, not a compliance claim. Ask for the oscilloscope traces.

Case study: a 20 MW/40 MWh plant in Hebei

A developer commissioned a 20 MW/40 MWh LFP plant in Hebei in 2024, connected at 35 kV and operating on a two-cycle-per-day peak-valley arbitrage profile with a provincial capacity lease on top. The original design used a 1.15 DC/AC ratio, chosen by copying the team's PV experience, with central PCS units rated 3.15 MW.
The first month of operation showed discharge clipping of 6–9% during the evening peak window, and the measured round-trip efficiency came in at 85.1% versus the 88% modelled. Re-sizing to 1.05 DC/AC and adding one PCS unit raised the annual throughput by roughly 5.5%, worth about RMB 0.7 million per year on a project with a 45 million RMB EPC cost. Round-trip efficiency after re-tuning the thermal setpoints and switching to a wider-load-range efficiency curve reached 87.6%.
The lesson is not "buy bigger." The lesson is that the DC/AC ratio, the efficiency weighting and the thermal derating curve must be modelled together against the actual dispatch schedule before the purchase order is signed. For a broader view of how storage pairs with generation assets, see our /tech/battery-storage technical notes.

Cooling, enclosures and derating in real climates

Datasheet ratings are quoted at 25°C ambient. In a Hebei summer, container internal temperatures reach 45–50°C, and in southern China they exceed that. A PCS rated 3.15 MW at 25°C may deliver only 2.9 MW at 45°C — a 8% derating that silently erodes the revenue model. Always request the derating curve versus ambient temperature and versus altitude; above 2,000 m, air-cooled units lose roughly 1% per additional 100 m.
Liquid-cooled PCS units cost 5–10% more upfront but hold full power to 50°C ambient and reduce auxiliary cooling energy by 1–2% of throughput. On a plant cycling twice daily for 15 years, that auxiliary saving alone often repays the premium.
Also check ingress protection and corrosion class against the site. IP65 with C4 corrosion resistance is the sensible baseline for coastal provinces; IP54 is adequate inland but should be paired with positive-pressure ventilation and filter maintenance schedules. Dust is the most common cause of premature fan failure we encounter in the field.

Commercial checklist before signing a PCS purchase order

1. Demand the weighted efficiency curve, not the peak figure, measured per GB/T 36548, and model annual throughput at your actual dispatch profile. A 0.5-point difference in weighted efficiency is worth roughly 0.5% of revenue.
2. Fix the DC/AC ratio against the dispatch schedule, not a rule of thumb. Stay near 1.00–1.08 for arbitrage; go higher only for services proven to need short overload bursts.
3. Require the full GB/T 36547 and GB/T 36548 test reports, including measured ride-through oscilloscope traces, AGC ramp times under 1 second, and the IEC 61850 point table.
4. Verify reactive capability at full active power and harmonic performance at 20% load, both measured on a weak-grid emulator if your site's short-circuit ratio is below 3.
5. Price the thermal derating curve into the revenue model and confirm liquid cooling economics for sites above 40°C ambient.
If you are comparing PCS topologies for a specific site, our team can review your single-line diagram and dispatch profile — start with /tech/inverters or send the specification to /contact. Reference installations and measured performance data are available under /projects, and matching battery options are listed in /products/lithium-battery.