FACTS Devices

FACTS (Flexible AC Transmission Systems) devices use power electronics to control quantities that a classical grid can only influence indirectly: bus voltage, branch flow, and the sharing of power between parallel paths. In a steady-state power flow a FACTS device reduces to a controllable network parameter, a reactive injection, a shunt susceptance, a series reactance, or a converter power, driven toward a target by the generic outer control loop around runpf!.

This page collects the device family, the limit characteristics that distinguish the devices from each other, and the modeling decisions behind Sparlectra's implementations. Hands-on: chapter 4 of the advanced workshop tour and the example examples/others/exp_facts_limit_modes.jl.

Device family and where each one lives

All controllers report through the same surfaces: ControlRunResult, controllableElements (element, device, actuator with live range, target, status/converged/at_limit), and the per-controller summary printers.

The limit characteristic is the device

In range, every shunt compensator does the same thing: it holds a voltage target by injecting reactive power. The devices differ at their LIMIT, and the limit is precisely where the distinction matters, because a compensator reaches it during the depressed-voltage conditions it was installed for.

Synchronous machine (constant-Q box). The classical limit is a fixed reactive capability, $Q \in [Q_{min}, Q_{max}]$, independent of the terminal voltage. At the limit the machine behaves like a fixed injection; this is the MachineVoltageControl default (limit_mode = :constant_q) and the exact outer-loop analogue of PV to PQ switching.

SVC (constant-B limit). An SVC regulates a continuous susceptance $B$ within $[B_{min}, B_{max}]$. At the limit the susceptance clamps and the delivered reactive power follows the voltage QUADRATICALLY through the Y-bus stamp:

\[Q_{SVC} = V^2 \, B_{lim}\]

This is the well-known weakness of the SVC: as the voltage sags, the support collapses with $V^2$. At $V = 0.9$ pu a fully switched-in SVC delivers only 81 percent of its nominal rating. Sparlectra's ShuntVoltageControl produces this behavior with no extra modeling: the clamped susceptance stays stamped in the Y-bus and the $V^2$ law falls out of the power flow itself.

STATCOM (constant-current limit). A STATCOM is a voltage-source converter; its bound is the converter CURRENT. The deliverable reactive power therefore scales LINEARLY with the terminal voltage:

\[Q_{STATCOM} = V \, I_{max} \quad\text{, in Sparlectra: } Q_{lim} = V \cdot S_{max}\]

with $S_{max}$ the rating at 1.0 pu. At $V = 0.9$ pu the STATCOM still delivers 90 percent of its rating, the decisive advantage over the SVC under exactly the sag conditions that matter. addMachineVoltageControl! with s_max_mva (or i_max_ka, converted via $\sqrt{3}\,U_n I_{max}$) switches the machine controller into this mode: the symmetric bound $\pm V \cdot S_{max}$ is re-evaluated from the solved terminal voltage before every outer step, so an at-limit STATCOM TRACKS the sagging or recovering voltage instead of freezing at a stale bound.

Summarized on one axis (delivered reactive power at the capacitive limit, relative to the 1.0-pu rating):

The machine column is idealized (a real machine derates too, through its capability curve); the STATCOM/SVC columns are the model behavior and match the device physics to first order.

Discrete banks: MSC/MSR

Strictly speaking a switched capacitor/reactor bank is not power electronics, but it is the working end of most voltage-control schemes and shares the shunt physics above, so it lives in the same controller (issue #324). With step_mvar the susceptance moves in whole switched blocks (e.g. four times 10 MVAr):

• the secant proposal is TRUNCATED toward the target to whole steps, so the bank approaches the voltage target from one side and never overshoots; this is the anti-hunting guarantee, a bank that never crosses its target cannot oscillate between two adjacent blocks;
• when no whole block improves the voltage further, the controller PARKS on the reached step (status = :parked, deliberately the last step BEFORE crossing: conservative under-compensation instead of a possible overvoltage) and releases itself when another controller moves the operating point far enough that a block helps again;
• at the outermost admissible block the constant-B limit region applies unchanged, the delivered Q follows $V^2$ with the last block connected.

The classical coordination case, a switched bank plus an OLTC on one bus, remains open follow-up work.

Series side: fixed window versus voltage-bounded window

TCSC (fixed reactance window). The thyristor-controlled series capacitor changes the branch reactance within a hardware-defined window $[x_{min}, x_{max}]$, independent of loading. At a window end the branch is a fixed compensated line. The resonance region between capacitive and inductive operation is excluded by the impedance-magnitude guard eps_z (see Series Compensation).

SSSC (injected-voltage window). The static synchronous series compensator injects a voltage in quadrature with the line current. In steady state that is equivalent to a reactance DEVIATION from the natural line reactance $x_{base}$, bounded by the injectable voltage magnitude:

\[|V_{inj}| = |I| \cdot |x - x_{base}| \le V_{inj,max} \quad\Longleftrightarrow\quad |x - x_{base}| \le \frac{V_{inj,max}}{|I|}\]

The usable window is therefore CURRENT-dependent and shrinks with loading: at high transfer, exactly when a large flow correction would need a large reactance swing, the SSSC saturates, while a TCSC keeps its full window. Conversely, at light loading the SSSC window is wide. In addSeriesReactanceControl! the mode is selected with v_inj_max_pu; the window $x_{base} \pm V_{inj,max}/|I|$ is re-evaluated from the solved branch current before every outer step, with a floor on $|I|$ (a currentless branch is physically unconstrained) and the same eps_z resonance guard applied as a clamp.

Live bounds in the outer loop

Both converter-based modes (STATCOM, SSSC) share one mechanism, the LIVE BOUND: the actuator range is a function of the solved operating point and is refreshed at the start of every outer iteration, before the secant step is clamped against it. Two consequences:

• At-limit tracking. A parked controller whose bound still moves (the voltage keeps sagging, the current keeps rising) is released and keeps adjusting; it reports at_limit only once its bound has settled. The delivered quantity therefore follows the physical limit law across outer iterations instead of freezing at the first clamp.
• Honest element rows. controllableElements and the report rows show the bounds of the LAST evaluated operating point, so actuator_min/ actuator_max are the currently deliverable range, not the nameplate.

In range, both modes behave like their fixed-limit counterparts: the same secant iteration on the same scalar map, converging into the same deadband.

Usage

Programmatic (see the docstrings for the full keyword sets):

# SVC: continuous susceptance, quadratic limit collapse
addShuntVoltageControl!(net; bus = "B", target_vm_pu = 1.0,
                        bs_min_mvar = -60.0, bs_max_mvar = 60.0)

# MSC/MSR: the same controller as a switched bank, whole 10-MVAr blocks
addShuntVoltageControl!(net; bus = "B", target_vm_pu = 1.0,
                        bs_min_mvar = -40.0, bs_max_mvar = 40.0,
                        step_mvar = 10.0)

# STATCOM: current-based limit, linear in V
addMachineVoltageControl!(net; bus = "B", target_bus = "C",
                          target_vm_pu = 1.0, s_max_mva = 25.0)

# TCSC: fixed reactance window
addSeriesReactanceControl!(net; fromBus = "A", toBus = "B",
                           p_target_mw = 35.0, x_min_pu = 0.02, x_max_pu = 0.30)

# SSSC: injected-voltage window, shrinks with loading
addSeriesReactanceControl!(net; fromBus = "A", toBus = "B",
                           p_target_mw = 35.0, v_inj_max_pu = 0.05)

Declarative, under control.controllers (see Control Framework):

control:
  enabled: true
  controllers:
    statcom_c:
      type: machine_voltage
      bus: B
      target_bus: C
      target_vm_pu: 1.0
      s_max_mva: 25.0
    sssc_ab:
      type: series_reactance
      from_bus: A
      to_bus: B
      p_target_mw: 35.0
      v_inj_max_pu: 0.05

UPFC: the stationary quadrature composite

The unified power flow controller combines a STATCOM (shunt side, bus voltage) and an SSSC (series side, branch flow) behind one DC link; the link couples the two converters through an active-power balance. That makes the full device a TWO-actuator controller with one coupling constraint, which does not fit the single-actuator secant pattern the outer loop is built on: each Sparlectra controller owns one actuator and one target, and the loop coordinates controllers only through the shared power flow. That was the reason for deferring the UPFC in issue #297 Draft G, and it still holds for the full device.

The way in is the QUADRATURE argument (issue #325): restrict the injected series voltage to quadrature with the line current. Then the series converter exchanges (approximately) no active power with the line, the DC link carries about zero, the coupling constraint degenerates, and what remains is exactly an SSSC on the branch plus a STATCOM at the bus. Both controllers exist, so addUpfcControl! registers them together as one named device:

• one call, one composite name; the series controller steers the branch active power inside the injected-voltage limit v_inj_max_pu, the shunt controller holds a remote bus voltage inside the current-based rating s_max_mva (or i_max_ka);
• registration is all-or-nothing (a rejected call leaves the net untouched), and the composite behaves exactly like the manually registered pair, to machine precision;
• the result table keeps one row per actuator with at_limit per converter side; both rows carry the device string UPFC series/shunt (VSC pair, stationary quadrature model).

What the composite is NOT: it has no series ACTIVE-power injection. The phase-shifter degree of freedom stays unavailable, and independent P and Q steering of the line needs the full model below.

control:
  enabled: true
  controllers:
    upfc_main:
      type: upfc                 # model: quadrature is the default
      from_bus: A
      to_bus: B
      shunt_bus: B
      target_bus: LOAD
      target_vm_pu: 1.0
      p_target_mw: 35.0
      v_inj_max_pu: 0.05
      s_max_mva: 25.0

UPFC: the full DC-link-coupled model

The full model (issue #326, model = :full) delivers the phase-shifter degree of freedom: a series voltage V_se of ARBITRARY phase, so the line carries INDEPENDENT active and reactive targets at once. The active part of the series injection, P_se = Re(V_se·conj(I_s)), flows through the DC link and is balanced by the shunt converter (P_sh = -P_se). In quadrature the in-phase component is zero and the device collapses onto the composite above; the picture is the split of the injected voltage relative to the line current:

        Im (quadrature to I_s: reactance, NO DC power)
         ^
         |      V_se
         |     /
         |    /  in-phase part -> P_se -> DC link -> shunt   (the UPFC DOF)
         |   /
         +--------------->  Re, aligned with the line current I_s

The series source is realised as an equivalent series impedance z_add = V_se / I_s added to the branch (Re(z_add) < 0 when the converter injects active power), so the line stays an ordinary branch and no fictitious injection is created. With the terminal voltages frozen each outer iteration the from-end flow is affine in V_se, so the series step is an exact 2x2 solve; the coupled iteration is globalised with an adaptive damping line search.

• one call, one controller (not a pair): model = :full steers the from-end line flow to p_target_mw AND q_target_mvar;
• the shunt converter provides the DC-link balance plus a reactive SETPOINT q_shunt_mvar, inside the current-based rating whose reactive headroom is coupled to the active load, Q_max = sqrt((V·s_max)^2 - P_sh^2);
• the result row carries V_se magnitude and angle, P_se, P_sh, the shunt Q, and the DC-link residual |P_se + P_sh| (a genuine convergence quantity, since the balance holds by construction only at the frozen state);
• series_phase = :quadrature forces P_se = 0 and reproduces the composite.

Honest limitations of the first cut:

• Stationary model. No dynamics or transients; IPFC (a shared DC bus across several lines) is out of scope.
• Shunt reactive SETPOINT, not closed-loop voltage. Coupling a shunt-voltage secant with the line reactive-flow control does not converge in the sequential outer loop (a known behaviour of injection-model UPFCs); closed-loop shunt voltage regulation is a follow-up that needs the AC power-flow sensitivity framework (issue #217) or an augmented in-solver state.
• No explicit series current limit. Only the injected-voltage magnitude |V_se| <= v_inj_max_pu is clamped, not the series-converter current |I_s| <= i_max; adding it is one more clamp on the same step.
• The branch impedance is modified in place (like the SSSC/TCSC): the equivalent series impedance z_add stays on the branch after the control run, and for the full model its resistance part goes NEGATIVE. That is the correct steady-state power-flow construct, not the physical line. Sparlectra keeps the two apart (issue #329): every branch carries its physical base impedance (r_base_pu/x_base_pu) next to the live value, and runShortCircuit! and the CGMES/MATPOWER exports read the BASE. A fault calculation or an interchange export therefore matches the equipment network automatically, with no manual reset, while the power flow keeps solving with the live (compensated) impedance. restoreBaseImpedances!(net) returns the live field to the base, and clearUpfcFullControllers!(net) does the same while dropping the controller.
• Low line current. z_add = V_se / I_s is floored at a minimum current (|I_s| guard) so a lightly loaded or dead line stays finite and keeps its base impedance; a UPFC on an essentially currentless line has nothing to steer.
• Convergence regime. The full model converges reliably for feasible, moderate flow targets (the realistic operating envelope of a UPFC). Very aggressive targets near the injectable-voltage limit may not converge in the outer loop; a robust envelope needs the sensitivity/Jacobian work above.

control:
  enabled: true
  controllers:
    upfc_full:
      type: upfc
      model: full
      from_bus: I
      to_bus: J
      shunt_bus: I
      p_target_mw: 40.0
      q_target_mvar: 10.0
      q_shunt_mvar: 0.0
      v_inj_max_pu: 0.20
      s_max_mva: 120.0

Validation

• test/test_tap_controller.jl: STATCOM registration validation, limit tracking on a depressed-voltage corridor ($Q = V \cdot S_{max}$ across operating points), in-range equivalence with the constant-Q mode, and the SVC-versus-STATCOM contrast on one case ($V^2$ versus $V$ collapse, asserted numerically).
• test/test_series_reactance_control.jl: SSSC registration validation, converged operation inside the live window, pinned operation with the effective injected voltage at $V_{inj,max}$, and TCSC-mode regression.
• test/test_upfc_control.jl: the quadrature composite equals the manually registered SSSC+STATCOM pair to machine precision, both limit characteristics at their clamps, all-or-nothing registration, and the YAML type upfc with the double-apply no-op; the full model reaches independent P and Q on one line simultaneously with the DC-link balance closed, reduces to the SSSC when the series phase is forced to quadrature, round-trips through model: full in YAML, and (after a full-model control run) has short circuit and the CGMES/MATPOWER exports read the physical base impedance so the compensated net matches the equipment network (#329).
• examples/others/exp_facts_limit_modes.jl: the three limit characteristics side by side on one weak corridor plus the SSSC window on a loop network.
• examples/others/exp_facts_base_impedance.jl: a full UPFC on a meshed corridor, then short circuit and MATPOWER export reading the physical base impedance so the compensated net matches the equipment net (#329).
• Chapter 4 of the advanced workshop tour walks the same contrasts interactively.