Renewable energy grid integration and stability

A 2 MW solar farm on a rural 11 kV feeder can meet its annual yield forecast and still lose output on every clear summer day. At noon its export raises the feeder voltage towards the upper limit, and the plant controller curtails to stay inside its connection offer. On 9 August 2019 a similar boundary failed at system scale in Great Britain. About 500 MW of small embedded generation disconnected on its own loss-of-mains protection during a transmission fault, and helped drive frequency down to 48.8 Hz.

In both cases each plant did what its settings told it to do. The problem appeared at a boundary: a feeder voltage limit in the first case, a protection setting repeated across many sites in the second. This guide covers the physics behind those boundaries, the measures that address each one, and what a renewable site should measure at its connection. For a real project, the connection agreement, the grid code and any service contract set the requirements. Read those first.

What renewable energy grid integration includes

The International Energy Agency sorts power systems into six phases of wind and solar integration. In phase 1, wind and solar have no significant effect at system level. By phase 6, supply comes almost entirely from them. The IEA's 2024 stocktake expects nearly two-thirds of the wind and solar added to 2030 to connect in systems in phases 1 to 3. There, better forecasting and dispatch, grid reinforcement and industrial demand response go a long way. Systems that already run at high instantaneous shares meet the stability problems first. Ireland is one: EirGrid and SONI operate the all-island system with System Non-Synchronous Penetration (SNSP) up to 75%, the share of power that comes through converters at any instant.

The rows interact, but a fix for one row rarely fixes another. A battery can relieve an export constraint and leave a protection-grading problem untouched. A monitor that records a voltage excursion at 20 ms resolution does not clear it.

Voltage rise and network capacity

Distribution networks were designed for power to flow one way, from the primary substation to the load. Export reverses the voltage drop along the feeder. A first estimate of the rise at the point of connection is:

ΔV / V ≈ (P × R + Q × X) / V²

P and Q are the plant's active and reactive export (Q is negative when the plant absorbs reactive power). R and X are the network resistance and reactance back to the source, and V is the line-to-line voltage.

Take a 2 MW plant at unity power factor on an 11 kV feeder with R = 1.5 Ω and X = 1.2 Ω back to the substation. The rise is (2 × 10⁶ × 1.5) ÷ (11 × 10³)² = 2.5%. Now run the inverters at 0.95 power factor, absorbing reactive power. Q = −0.66 Mvar, so the reactive term is −0.66 × 10⁶ × 1.2 ÷ (11 × 10³)² = −0.65%. The net rise falls to about 1.8%.

Reactive absorption recovers only a quarter of the rise here, because R is larger than X on this feeder. On a transmission circuit, where X is many times R, the same power factor cancels much more. The absorption also uses inverter capacity. At 0.95 power factor the plant needs 2.1 MVA, so a 2 MVA inverter fleet must reduce active power to 1.9 MW. The kW, kVA, kvar and power factor calculator relates P, Q, S and power factor for an operating point.

A connection study adds the conditions that set the real limit: minimum demand at maximum generation, tap-changer behaviour under reverse power flow, outage configurations, and every other project on the same feeder. The same capacity arithmetic applies in the import direction. The EV charging capacity calculator checks how much charging a site can run within its import limit, interval by interval.

Frequency, inertia and loss of mains

Frequency is common to a whole synchronous area and moves with the balance between generation and demand. In Great Britain, the system operator must keep it between 49.5 and 50.5 Hz and normally holds it within 49.8 to 50.2 Hz.

In the first moments after a sudden loss of generation, only the kinetic energy of rotating machines resists the change. The initial rate of change of frequency (RoCoF) is approximately:

RoCoF = ΔP × f₀ / (2 × Eₖ)

ΔP is the lost infeed in MW, f₀ is 50 Hz and Eₖ is the kinetic energy stored in synchronised machines, in MW·s. A synchronous generator stores roughly 2 to 10 seconds of its own rating (its inertia constant, H). A PV inverter stores none. For a system holding 150,000 MW·s, a 1,000 MW loss gives 1,000 × 50 ÷ (2 × 150,000) = 0.17 Hz/s. With half the synchronous plant online, the same loss gives 0.33 Hz/s.

That matters because loss-of-mains protection on embedded generation also measures RoCoF. On 9 August 2019, a lightning strike on a transmission line started this sequence:

1. About 150 MW of embedded generation tripped on vector-shift protection when the fault occurred.
2. Hornsea One offshore wind farm lost 737 MW, and the Little Barford steam turbine tripped at 244 MW.
3. The falling frequency tripped a further 350 MW of embedded generation whose RoCoF protection was set at 0.125 Hz/s.
4. The total loss reached 1,481 MW. The system operator held reserve for 1,000 MW, the largest single infeed at the time.
5. A Little Barford gas turbine then tripped at 210 MW. Frequency fell to 48.8 Hz, and low-frequency demand disconnection removed 931 MW of demand, about 1.1 million customers.

The GB response was the Accelerated Loss of Mains Change Programme. Where RoCoF protection is used for loss of mains, the setting is now 1 Hz/s with a definite time delay of 500 ms, and vector-shift protection must be removed. The requirement became mandatory on 1 September 2022. Ireland uses the same figure from the system side: since May 2023, EirGrid and SONI have operated to an enduring RoCoF limit of ±1 Hz/s.

After inertia, frequency response services take over. Their timescales differ by an order of magnitude:

A battery inverter can change output within a fraction of a second, but only from headroom it already holds, and only after it has measured the frequency change. Test the whole chain from frequency input to the measured power change at the connection point. The dynamic response guide covers the GB services, and the FCAS guide covers the Australian measurement and verification rules.

Stability classes for converter-dominated systems

The 2004 IEEE/CIGRE classification had three classes: rotor-angle, voltage and frequency stability. The IEEE PES report TR-77 (2020) keeps them and adds two classes for systems with a large share of power electronics:

• Resonance stability covers oscillations from energy exchange between series-compensated lines and turbine-generator shafts (torsional resonance), or between series compensation and generator electrical controls, such as those of doubly fed wind turbines (electrical resonance).
• Converter-driven stability covers interactions between converter control loops, the network and other converters. The report separates fast interactions with the network's electromagnetic dynamics from slow interactions with its electromechanical dynamics.

Name the class, the disturbance and the timescale in the study scope. Slow voltage collapse can be studied with RMS phasor models. A control interaction between two converters usually needs electromagnetic transient (EMT) models of both.

System strength

Short-circuit ratio (SCR) is the usual first screen. SCR is the three-phase fault level at the connection point in MVA divided by the plant rating in MW. A 50 MW plant at a 33 kV node with a 150 MVA fault level has an SCR of 3. The lower the SCR, the more the connection-point voltage moves for a given change in injected current, and the harder it is for a grid-following inverter to track that voltage.

NERC's guideline on low short-circuit strength systems warns that simple SCR is optimistic when several plants connect close together, because they interact. It describes weighted SCR (WSCR) and composite SCR for that case. ERCOT used detailed EMT studies in PSCAD to set a WSCR threshold of 1.5 for its Panhandle region. Below that value, it requires more detailed models. The usual remedies are synchronous condensers, grid-forming converters and network reinforcement.

Grid-following and grid-forming inverters

A grid-following inverter measures the network voltage with a phase-locked loop and injects a controlled current. A grid-forming inverter sets its own voltage magnitude and phase, and current flows in response to the network, as it does from a synchronous machine. Both share the same semiconductor limit. A converter's controls hold its fault current close to its rating, while a synchronous machine supplies several times its rated current for the first cycles of a fault. When converters replace machines, check any protection that relies on fault current to operate. The NREL grid-forming inverter roadmap sets out the open questions, including behaviour at current limit and in systems with many grid-forming units.

Ride-through, protection and disturbance behaviour

Connection codes specify ride-through as a voltage-against-time envelope. The plant must stay connected while the voltage stays inside it. IEEE 1547-2018 defines three abnormal-performance categories (I, II and III) for distributed resources in North America. In Europe, the RfG network code (Regulation 2016/631) sets requirements by generator type, A to D. In Great Britain, G99 puts the type boundaries at 1, 10 and 50 MW for connections below 110 kV.

Two events show where ride-through settings fail. On 28 September 2016, storms brought down three transmission lines in South Australia. Five faults in 87 seconds caused six voltage dips. The AEMO final report found that eight of nine affected wind farms had turbine protection that allowed only a preset number of voltage dips in two minutes. The turbines rode through each individual dip, then reduced output when the count was exceeded. A sustained reduction of 456 MW occurred in less than seven seconds. About 700 ms later, import over the Heywood interconnector from Victoria rose high enough to trigger a special protection scheme, which tripped the interconnector. The state then went black.

The second event is on the demand side. On 8 May 2025, a remote transient fault on the all-island system caused an immediate 387 MW reduction in data-centre demand, 52% of all data-centre load at the time. EirGrid and SONI report that generation recovers after such a fault while part of the demand stays off, so frequency rises. With HVDC export also lost, the imbalance could exceed 1,150 MW. The TSOs kept the inertia floor above its planned level and put the trial of 80% SNSP on hold. They are also preparing a Grid Code modification that gives large demand facilities a voltage-against-time ride-through profile.

Protection and ride-through pull in opposite directions. Early tripping deepens a system event, as both events show. Late tripping can damage equipment or keep an island energised. The protection study sets the balance, and commissioning must prove that the relay carries the approved settings.

Grid integration measures and what they solve

Battery storage for grid integration

A battery is two constraints at once: a power rating and an energy store. Consider a 10 MW / 20 MWh battery contracted for 10 MW of Dynamic Containment in both directions. NESO lists the service at up to 15 minutes of delivery. The battery must therefore hold 10 MW × 0.25 h = 2.5 MWh available to discharge and 2.5 MWh of room to charge. Its stored energy must stay between 2.5 and 17.5 MWh, before round-trip losses, the manufacturer's usable-energy window and any margin the provider adds. All 10 MW of power is committed, so nothing is left for wholesale trading in those hours. To stack services, the operator splits the megawatts between them, and each service then also claims its own share of the energy.

The battery management system owns state of charge, cell temperature and safety limits. An independent AC meter at the battery connection verifies power and energy in each direction. It cannot reconstruct cell state or prove that the battery is safe. The battery demand-response guide shows how to record availability, requested and achieved response, stale data and withdrawal at a site boundary.

Demand response and virtual power plants

Flexible demand can reduce consumption during scarcity or congestion and increase it during a renewable surplus. An aggregator or virtual power plant coordinates many such assets, and each asset needs these answers in writing:

1. What may be controlled, and which process or safety limit always has priority.
2. The increase or reduction available now, and how long it can be sustained.
3. The rebound after the event.
4. Who may issue a command, and how the command is authenticated.
5. What the asset does when communications or measurements go stale.
6. Which meter and baseline settle performance.

A request, an acknowledgement and the net response at the connection point are three different records. The demand response event calculator compares event and rebound power with their baselines interval by interval. The Demand Response & VPP page describes the wider architecture.

Site control and measurement layers

Only the first three layers act on the plant fast enough to matter during a disturbance. A gateway is not a protection device, and a cloud platform is not a plant controller. The SCADA architecture guide explains how telemetry, control authority, alarms and safety interlocks stay separate.

Data points and intervals for a renewable site

The connection agreement and the operating question decide the point schedule. A typical schedule at the connection point and at each generating or storage group includes:

• voltage and current on each phase;
• active, reactive and apparent power, with a written sign convention;
• import and export energy as separate cumulative registers;
• frequency;
• plant, inverter, breaker and controller states;
• active and reactive power setpoints, and the system that issued each one;
• curtailment flags, alarms and protection indications;
• battery state of charge and limits, read from the battery management system; and
• a source timestamp, a receipt timestamp and a stale flag on every value.

Write each delay in the chain down separately. Take a battery contracted for Dynamic Containment, which must reach full delivery within 1 s. Build the control loop through a cloud platform: a 1 s meter sample, a 1 s gateway poll, a 5 s publication interval and 2 s of platform processing. The loop has used 9 s before a command leaves the platform. The response must be calculated on site from local frequency, and telemetry reports afterwards what happened.

Reconciling the connection point

Use one sign convention everywhere: import positive, export negative. Then site demand = PV output + connection-point power.

At 12:30 on a clear day, the independent PV meter reads 820 kW and the connection-point meter reads −310 kW. Site demand is 820 − 310 = 510 kW. The inverters' own registers sum to 842 kW. The 22 kW (2.7%) difference comes from cable and transformer losses between the inverter terminals and the PV meter, plus the inverters' own measurement uncertainty. If the difference stays near 2.7% across the day, it is a property of the installation. If it steps, look for a failed meter channel, a changed CT or an inverter reporting incorrectly.

Now give the site a 250 kW export limit. On the next clear day, the connection-point meter sits at −250 kW for two hours while plane-of-array irradiance keeps rising. The plant controller is curtailing. Three signs separate this from a fault. Connection-point export is flat at the limit. All inverters reduce output together. The controller reports a curtailment state. In a fault, one inverter falls against its neighbours and export drops below the limit.

Monitoring solar PV performance

IEC 61724-1 defines the measurements and methods for PV performance monitoring. Its central metric is the performance ratio: PR = Yf / Yr. The final yield Yf is AC energy divided by the array's rated DC power (kWh/kWp). The reference yield Yr is plane-of-array irradiation divided by the 1 kW/m² reference irradiance.

For example, a 1,000 kWp array delivers 4,800 kWh on a day with 6.0 kWh/m² of plane-of-array irradiation. Yf = 4.8 h, Yr = 6.0 h, and PR = 0.80.

Module temperature moves PR with the seasons on a healthy plant. A module with a power temperature coefficient of −0.35 %/°C, running at 55 °C instead of the 25 °C rating temperature, produces 30 × 0.35 = 10.5% less than its rating. Summer PR therefore reads lower than winter PR. IEC 61724-1 also defines temperature-corrected performance ratios for comparisons across seasons.

Curtailed hours must be flagged before PR is calculated. Otherwise the export limit in the example above reads as underperformance. To calculate PR and locate losses, measure:

• plane-of-array irradiance, with a pyranometer or reference cell at the module angle;
• module temperature, with the coefficient from the module datasheet;
• AC output of each inverter, to find a trip or clipping;
• string or MPPT current and voltage from the inverter, to find a failed string, a blown fuse or shading; and
• import and export at the connection point, for self-consumption and the export limit.

Compare each inverter with its neighbours on the same day. A string failure or an inverter trip shows as a step. Soiling and degradation show as a slow fall against irradiance.

Grid integration design and commissioning

Design steps:

1. Mark the point of connection, the ownership boundary and every measurement location on the single-line diagram.
2. Obtain the current connection agreement, grid code, protection requirements and any service specification.
3. Study power flow, fault level, protection, power quality and the named stability phenomena, in normal and outage configurations.
4. Assign protection, inverter control, plant control, settlement metering, operational monitoring and onward communication to named equipment.
5. Record the active and reactive power envelope, ramp rates, headroom, energy duration and fail-safe state.
6. Define the data contract: point identities, units, sign convention, timestamps, stale flags, intervals and retention.
7. Authenticate instructions, restrict who may issue them, record acknowledgements and keep local safety or process vetoes.

Commissioning checks:

• Confirm that every meter and transducer sits on its documented boundary.
• Verify phase association, CT orientation, voltage references, ratios and import and export signs during a known import period and a known export period.
• Reconcile PV output, connection-point exchange and site demand in at least two operating states.
• Exercise active and reactive power setpoints only through an approved, bounded test procedure, and check which one wins when both reach the current limit.
• Measure ramp limits, deadbands and settling time at the connection point.
• Check the controller, gateway, historian and receiving-system clocks against the stated tolerance.
• Interrupt each communication path, and prove the stale indication, local fallback, buffering and recovery.
• Prove that an old or replayed command cannot pass as a current instruction.
• Compare the operational meter with the settlement meter, and record the expected difference.
• Record the model, firmware, settings and configuration revision of the accepted result, and repeat the relevant checks after any change.
• Keep protection and ride-through testing within the authorised engineering and safety procedure.

Monitoring a renewable site with EpiSensor

EpiSensor hardware covers the operational monitoring layer. A ZEM-65 meters the PV output and the grid connection independently of the inverters. It is Class 0.5S to IEC 62053-22 with its current sensors, and it measures frequency to ±0.01 Hz. A ZMB-31 reads inverter registers (string data, state and alarms) over Modbus RTU on RS-485. A ZIO-20 takes a 4 to 20 mA irradiance or module-temperature transmitter. The ZGW-20 runs EpiSensor Edge, which calculates site demand from the balance, keeps local history, applies rules and exports over MQTT or HTTP.

For frequency response, the ZDR controllers sample frequency at least every 100 ms at 0.01 Hz resolution. ZDR-22 sends a changing set point to a battery or UPS over Modbus, so the response is calculated on site. ZDR-21 and ZDR-22 record each event at 20 ms, with GPS timestamps. None of this equipment replaces the protection relays, the plant controller or the settlement meter.

The solar PV monitoring guide shows the meter placement. For a grid-connected project, agree these points before installation:

• the authoritative connection-point meter and its sign convention;
• whether each value is operational, compliance or settlement evidence;
• the measurement interval, timestamp source and freshness limit for each point;
• which system owns setpoints and control permission;
• what the site does locally when data, cloud or communications fail;
• how buffered and replayed data is marked for the recipient; and
• who owns studies, commissioning, service qualification and later changes.