Effect of Steady-State Wind Turbine Generator Models on Power Flow Convergence and Voltage Stability Limit Genevieve Coath Mechatronics Research Group Department of Mechanical Engineering The University of Melbourne, Australia Email: gcoath@ieee.org Majid Al-Dabbagh Electrical Energy and Control Systems School of Electrical and Computer Engineering RMIT University, Australia Email: majid@rmit.edu.au A BSTRACT Environmental concerns over the use of fossil fuels as energy sources for electric power generation have resulted in the rapid increase in the amount of wind-generated power connected to electrical power networks worldwide. As such, the inclusion of accurate wind turbine generator models in power system simulation software has become increasingly important. Steady-state and dynamic models of wind turbine generators are required for different types of simulation studies. Steady-state fixed-speed and variable-speed models are examined in this paper in terms of their effect on the power flow convergence, voltage stability limits, and their ability to be used to provide good initial values in dynamic models. 1. I NTRODUCTION In recent years, there has been a rapid increase in the amount of wind-generated power on electrical power grids worldwide. Subsequently, the inclusion of accurate models of wind turbine generators (WTGs) in power system analysis software is important. Steady-state WTG models are required for voltage stability analysis and to initialise dynamic models; dynamic simulation models require correct initialisation otherwise time will be wasted as the models try to find a steady-state operating point as the initial condition [1]. Thus, load flow data can provide dynamic models with an initial starting point [1]. The way in which the WTG is modelled in load flow studies can influence the load flow convergence and the system’s voltage stability limit (VSL). Variablespeed WTGs have seen a recent increase in popularity as the problems associated with fixed-speed WTGs become more prominent with the increase in penetration level of windgenerated power into electrical power grids. 2. W IND T URBINE G ENERATORS 2.1. S QUIRREL -C AGE I NDUCTION G ENERATORS Squirrel-cage induction generators (SCIGs) are fixed-speed machines that are directly grid connected, as shown in Figure 1. The slip and the rotor speed of a SCIG vary with the generated power; these variations in rotor speed are so small that SCIGs are described as fixed-speed [2]. They are widely employed in WTGs today, however their inability to generate reactive power is a limiting factor in their future use. Fig. 1. Squirrel cage induction generator [2] extraction from the wind [3]. This is achieved by feeding the rotor circuit with real and reactive power from the rotorside converter, as shown in Figure 2. The converter circuit allows the production or consumption of reactive power, making it different to the SCIG, which can only consume reactive power. Hence, DFIGs do not cause the same voltage instability problems as do SCIGs. Fig. 2. Doubly-fed induction generator [2] 2.3. D IRECT-D RIVE S YNCHRONOUS G ENERATORS Figure 3 shows a direct-drive synchronous generator (DDSG) which is studied in this paper, and has a wound rotor, and no gear box due to its operation at low speeds and its many poles. It is connected to the grid via a back-to-back voltage source converter [1]. 2.2. D OUBLY-F ED I NDUCTION G ENERATORS Doubly-fed induction generators (DFIGs) have seen a recent surge in popularity for wind turbine applications for several reasons [2]. The primary reason for this is their ability to vary their operating speed in order to gain optimum power Fig. 3. Direct drive synchronous generator [2] 3. W IND T URBINE G ENERATOR M ODELLING 3.1. S QUIRREL C AGE I NDUCTION G ENERATOR M ODEL The SCIG is usually modelled as a conventional PQ bus, with the real power generated and reactive power demand specified. However, the reactive power demand can be expressed as a funtion of the bus voltage if the SCIG is modelled as an improved PQ bus by using the representation introduced in [4] and later used in [5], as shown below: Q≈V2 Xc − X m X + 2P2 Xc Xm V (1) where Q is the generator’s reactive power consumption, and is calculated taking into account the capacitive reactance Xc , magnetising reactance Xm , sum of stator and rotor reactances X, terminal voltage V , and real power P of the generator. In this paper it was assumed that the capacitor was there primarily to compensate the no-load reactive power consumption, hence Xc = Xm . The improved PQ bus representation involves updating the specified reactive power at every load flow iteration [4]; a full derivation of the approximation can be found in [4]. 3.2. D OUBLY-F ED I NDUCTION G ENERATOR M ODELS DFIGs can be included in load flow studies as PQ or PV buses, as they can operate in either power factor controlled or voltage-controlled modes. When modelling a DFIG as a PQ bus, it is assumed that the DFIG is operating in power factor controlled mode, meaning that the specified reactive power is zero. In voltage controlled mode, the DFIG can be represented as a PV bus with Q limits applied. 3.3. D IRECT-D RIVE S YNCHRONOUS G ENERATOR M ODEL The DDSG can be modelled as a PV bus in the load flow study, with or without its reactive power limits enforced. When enforcing these limits, if the limit is reached the PV bus is converted to a PQ bus. attached to PQ bus 30 of the 57-bus network. The DFIG was then modelled in voltage controlled mode as a PV bus with reactive power generator limits enforced, assuming a power factor of 0.85. The DDSG was modelled as a PV bus with and without reactive power generator limits enforced. All PV models were attached to PV bus 6 of the 57-bus network. All load flow studies were performed using the BX version of the fast-decoupled load flow which was initialised to a flat start. A continuation power flow was implemented using a secant predictor and a bus voltage magnitude as the continuation parameter, as in [7], to trace the PV curve of the weakest bus in the system under study. Hence, the continuation power flow simulated a load increase on bus 31 of the 57-bus system for all PV curves traced. The load increase was performed using a loading factor, λ, whose value was obtained as part of the continuation power flow. The MATPOWER software package was used for all simulations due to the ease with which it can be modified to include additional generation which is entirely specified by the user, and also for its ease of use with continuation power flow analysis. It is a single-phase load flow package with many features including both ac and dc power flows. It also provides a variety of solvers for optimal power flow problems which make use of some of the features in MATLAB’s Optimization Toolbox. It was designed to be highly modifiable for use in power system analysis research. 6. R ESULTS 6.1. E FFECT OF M ODELS ON P OWER F LOW C ONVERGENCE Each of the WTG model types were implemented into the systems as aggregated wind farms with only one type of WTG modelled within the wind farm, i.e. all PQ, or all improved PQ, etc. The PQ-based WFs were in turn added to PQ bus 30 of the 57-bus system and the PV-based WFs were in turn added to PV bus 6. The effect of the models on the power flow convergence in terms of P θ and QV iterations at the bus at which they were applied was assessed. From 4. DYNAMIC M ODEL I NITIALISATION 5. S IMULATION S TUDIES Studies were carried out using the IEEE 57-bus test network, which was modified to incorporate the wind farms. In all cases, a wind farm capable of producing 30MW was assumed. The wind farm was comprised of 15 aggregated WTGs, each rated at 2MW. The WTG base values were as follows [6]: Vbase = 690V , Sbase = 2M W , ωbase = 2πfbase and fbase = 50Hz. The WTGs’ parameters were [6]: stator resistance (Rs ) = 0.0048pu, rotor resistance (Rr ) = 0.00549pu, stator reactance (Xls ) = 0.09241pu, rotor reactance (Xlr ) = 0.09955pu and magnetising reactance (Xm ) = 3.95279pu. The SCIG was modelled as a conventional PQ bus and as an improved PQ bus, as described in Section 3.1. The DFIG was modelled as a PQ bus with a constant power factor, the assumption being that the DFIG has local power factor control. Hence, a real power was assumed (nominal power output of each WTG within the wind farm), and the reactive power consumption was zero. All PQ models studied were −3 6 x 10 P mismatch Q mismatch 4 2 Real and reactive power mismatch The use of dynamic models of wind turbine generators in power system dynamic simulation software is becoming of increasing relevance as the amount of wind-generated electrical power on electrical grids worldwide increases. Dynamic models must be initialised with a valid steady-state operating point, otherwise during dynamic simulations time is wasted as the models try to find a valid operating point [1], [6]. 0 −2 −4 −6 −8 −10 −12 1 2 3 4 5 6 Number of iterations 7 8 9 10 Fig. 4. Convergence of Pθ iterations for PQ model of SCIG on bus 30 of 57 bus system Figures 4-7, it can be seen that using the improved PQ bus model for the SCIG results in a higher number of P and Q iterations than the conventional PQ model of the SCIG. This can be attributed to the calculation of the specified reactive power as a function of bus voltage, which is therefore updated within the complex power injection at each load flow iteration. The PQ model of the DFIG had a similar convergence behaviour to the conventional PQ model of the SCIG, as shown in Figures 8 and 9. Figures 10 and 11 show −3 0.005 P mismatch Q mismatch 0 4 2 −0.005 Real and reactive power mismatch Real and reactive power mismatch x 10 6 P mismatch Q mismatch −0.01 −0.015 −0.02 0 −2 −4 −6 −8 −0.025 −10 −0.03 1 2 3 4 5 6 Number of iterations 7 8 9 −12 1 2 Fig. 5. Convergence of QV iterations for PQ model of SCIG on bus 30 of 57 bus system 3 4 5 6 Number of iterations 7 8 9 Fig. 8. Convergence of Pθ iterations for PQ model of DFIG on bus 30 of 57 bus system −3 6 x 10 0.005 P mismatch Q mismatch P mismatch Q mismatch 4 0 2 Real and reactive power mismatch Real and reactive power mismatch −0.005 0 −2 −4 −6 −0.01 −0.015 −0.02 −0.025 −8 −0.03 −10 −12 −0.035 0 5 10 15 1 2 3 4 Number of iterations 5 6 Number of iterations 7 8 9 Fig. 9. Convergence of QV iterations for PQ model of DFIG on bus 30 of 57 bus system Fig. 6. Convergence of Pθ iterations for improved PQ model of SCIG on bus 30 of 57 bus system 1.4 0.005 P mismatch Q mismatch P mismatch Q mismatch 1.2 0 Real and reactive power mismatch Real and reactive power mismatch 1 −0.005 −0.01 −0.015 −0.02 0.6 0.4 0.2 −0.025 −0.03 0.8 0 −0.2 0 2 4 6 8 Number of iterations 10 12 14 Fig. 7. Convergence of QV iterations for improved PQ model of SCIG on bus 30 of 57 bus system that the PV model with reactive power limits applied had approximately the same convergence as the PQ model for the DFIG and SCIG. Figures 12-15 show the convergence characteristic of the load flow calculation for the DDSG models. There was no effect on convergence of applying the reactive power generation limits; PV buses whose limits are reached are converted into PQ buses which doesn’t affect the load flow convergence. It should be mentioned, however, that during this process several attempts at the load flow calculation itself may be made in order to satisfy all generators’ reactive power generation limits. A summary of 1 2 3 4 5 6 Number of iterations 7 8 9 10 Fig. 10. Convergence of Pθ iterations for PV model with Q limits of DFIG on bus 6 of 57 bus system the results of the load flow convergence is shown in Tables 1 and 2. 6.2. E FFECT OF M ODELS ON VOLTAGE S TABILITY L IMIT The effect of the models on the VSL of bus 31 of the the 57-bus system can be seen in Figures 16-21. Figures 1619 illustrate the impact of the different induction generator models on the voltage stability limit (VSL) of the 57-bus system. The conventional PQ model of the SCIG gives a VSL of 4.4, as does the improved PQ model. Hence, the increased accuracy of the reactive power consumption of the machine −3 0.01 1 P mismatch Q mismatch P mismatch Q mismatch 0.5 0 0 Real and reactive power mismatch Real and reactive power mismatch x 10 −0.01 −0.02 −0.03 −0.5 −1 −1.5 −2 −0.04 −2.5 −0.05 1 2 3 4 5 6 Number of iterations 7 8 9 −3 1 Fig. 11. Convergence of QV iterations for PV model with Q limits of DFIG on bus 6 of 57 bus system 2 3 4 5 6 Number of iterations 7 8 9 Fig. 14. Convergence of Pθ iterations for PV model with Q limits of DDSG on bus 6 of 57 bus system −3 14 x 10 0.09 P mismatch Q mismatch P mismatch Q mismatch 0.08 12 0.07 Real and reactive power mismatch Real and reactive power mismatch 10 8 6 4 0.06 0.05 0.04 0.03 0.02 2 0.01 0 −2 0 −0.01 1 2 3 4 5 6 Number of iterations 7 8 9 2 3 4 5 Number of iterations 6 7 8 Fig. 15. Convergence of QV iterations for PV model with Q limits of DDSG on bus 6 of 57 bus system Fig. 12. Convergence of Pθ iterations for PV model of DDSG on bus 6 of 57 bus system 0.95 −3 1 1 x 10 0.9 P mismatch Q mismatch 0.85 0 −1 Voltage (p.u.) Real and reactive power mismatch 0.8 −2 0.75 0.7 0.65 0.6 −3 0.55 0.5 −4 0.45 −5 1 2 3 4 5 Number of iterations 6 7 1 1.5 2 2.5 3 Loading factor, λ 3.5 4 4.5 8 Fig. 13. Convergence of QV iterations for PV model of DDSG on bus 6 of 57 bus system in the improved PQ model makes no difference to the VSL obtained. As the DFIG operating in power factor controlled mode doesn’t consume reactive power, it yields a higher VSL, at 4.92, and also has a less steep gradient over the first few incremental load increases than the SCIG models. The decline in bus voltage in this region is particularly important, as a large decrease may often mean that the voltage is below its minimum allowable value. The PV representation of the DFIG gives a lower voltage stability margin at bus 31 than any of the PQ models; its placement in the network Fig. 16. PV curve for PQ model of SCIG on 57 bus system is electrically distant from bus 31, hence having little effect on its VSL, but causing a more rapid decline in bus voltage than any of the other models. The results are summarised in Table 3. The VSLs for the two DDSG models differed by only 0.07, as the effect of applying the reactive power limits had virtually no impact at bus 31, again due to the distance between it and the wind farm bus; the results are shown in Table 4. 6.3. DYNAMIC M ODEL I NITIALISATION DATA FROM L OAD F LOW Model # of Iterations for P V Convergence 10 15 9 10 PQ (SCIG) Improved PQ PQ (DFIG) PV with Q limits # of Iterations for Qθ Convergence 9 14 9 9 1 0.9 0.8 Model # of Iterations for P V Convergence 9 9 PV PV with Q limits Voltage (p.u.) Table 1. Comparison of convergence characteristics for induction generator models # of Iterations for Qθ Convergence 8 8 0.7 0.6 0.5 Table 2. Comparison of convergence characteristics for DDSG models 0.4 1 1.5 2 2.5 3 Loading factor, λ 3.5 4 4.5 0.95 Fig. 19. PV curve for PV model with Q limits of DFIG on 57 bus system 0.9 0.85 Model PQ (SCIG) Improved PQ PQ (DFIG) PV with Q limits Voltage (p.u.) 0.8 0.75 0.7 Critical Point 4.4 4.4 4.92 4.02 Table 3. Comparison of voltage stability limits for induction generator models 0.65 0.6 0.55 1 0.5 0.45 1 1.5 2 2.5 3 Loading factor, λ 3.5 4 4.5 0.9 Fig. 17. PV curve for improved PQ model of SCIG on 57 bus system Voltage (p.u.) 0.8 1 0.9 0.6 0.8 Voltage (p.u.) 0.7 0.5 0.7 0.4 1 1.5 2 2.5 3 Loading factor, λ 3.5 4 4.5 0.6 Fig. 20. PV curve for PV model of DDSG on 57 bus system 0.5 1 0.4 1 1.5 2 2.5 3 Loading factor, λ 3.5 4 4.5 5 0.9 Fig. 18. PV curve for PQ model of DFIG on 57 bus system 6.3.1. S QUIRREL C AGE I NDUCTION G ENERATOR : The SCIG is an induction generator with a zero rotor voltage. The voltage equations of the machine can be written as [1]: vds = −Rs ids + ωs ((Lsσ + Lm )iqs + Lm iqr ) 0 = −Rr iqr − sωs ((Lrσ + Lm )idr + Lm ids ) The real power generated, P , and the reactive power consumed, Q, are [1]: = vds ids + vqs iqs Q = vqs ids − vds iqs 0.7 0.6 0.5 vqs = −Rs iqs − ωs ((Lsσ + Lm )ids + Lm idr ) (2) 0 = −Rr idr + sωs ((Lrσ + Lm )iqr + Lm iqs ) P Voltage (p.u.) 0.8 (3) 0.4 1 1.5 2 2.5 3 Loading factor, λ 3.5 4 4.5 Fig. 21. PV curve for PV model with Q limits of DDSG on 57 bus system where v is voltage, i is current, R is resistance, L is inductance, ω is frequency and s is slip. The equations are Model PV PV with Q limits Critical Point 4.08 4.01 different to the current. Hence, the necessity of applying these limits is evident, otherwise unrealistic values would be obtained to initialise the dynamic models. Table 4. Comparison of voltage stability limits for DDSG models 7. C ONCLUSIONS expressed in terms of their direct (d) and quadrature (q) components. The subscripts m, s and r stand for mutual, stator and rotor, respectively. The symbol σ symbolises leakage. 6.3.2. D OUBLY-F ED I NDUCTION G ENERATOR : The voltage equations required to initialise the reduced order dynamic DFIG are [1]: vds vqs = −Rs ids + ωs ((Lsσ + Lm )iqs + Lm iqr ) = −Rs iqs − ωs ((Lsσ + Lm )ids + Lm idr ) vdr vqr = −Rr idr + sωs ((Lrσ + Lm )iqr + Lm iqs ) = −Rr iqr − sωs ((Lrσ + Lm )idr + Lm ids ) (4) and the real and reactive power of the generator are given by [1]: P = Ps + Pr = vds ids + vqs iqs + vds idc + vqs iqc Q = vqs ids − vds iqs (5) (6) where the subscript c means converter. All other symbols are as described earlier. 6.3.3. D IRECT-D RIVE S YNCHRONOUS G ENERATOR : As seen from the power grid, the DDSG’s real and reactive power generated is that provided by the voltage source converter, as in Figure 3, [1]: Pc Qc = vdc idc + vqc iqc = vqc idc − vdc iqc (7) and the voltage equations are given by: q 2 + v2 vg = vds qs vds = ωm (Lqm + Lsσ )iqs − Rs ids vqs vf d = ωm (−(Ldm + Lsσ )ids + Ldm if d ) − Rs iqs = R f d if d (8) The real and reactive power generated is expressed as [1]: Pg = vds ids + vqs iqs Qg = vqs ids − vds iqs (9) Using the load flow data obtained for each of the models, the initial conditions of stator voltages and stator currents in the d-q reference frame were calculated. It can be seen in Table 5 that depending on the model used, a different initial condition is obtained. The SCIG models gave similar The effect of steady-state models of fixed-speed and variablespeed wind turbine generators (WTGs) on power flow convergence, voltage stability limit and suitability for dynamic model initialisation was studied in this paper. The squirrelcage induction generator (SCIG) was modelled as a conventional PQ bus and as an improved PQ bus, where the dependence of reactive power on nodal voltage was taken into account during the load flow iterations. The doubly-fed induction generator (DFIG) was modelled in power factor controlled mode as a conventional PQ bus, and in voltage controlled mode as a PV bus with reactive power generator limits enforced. The direct-drive synchronous generator (DDSG) was modelled as a PV bus with and without reactive power limits enforced. If its reactive power generation was at a limit, it was converted into a PQ bus. The improved PQ model required more load flow iterations to converge, however it yielded an identical voltage stability limit (VSL) to the conventional PQ model, meaning that its costly computation provided no advantage over the conventional PQ model when calculating the VSL. Its use would be more suited to conventional load flow studies, where an accurate calculation of an induction generator’s reactive power consumption is required. The PQ model of the DFIG gave a higher VSL than the other PQ models, as the nature of the DFIG operating in power factor controlled mode means that it does consume any reactive power, and hence doesn’t negatively affect the VSL. It also provided the slowest decline in bus voltage for the continuation power flow study, meaning a higer bus voltage was maintained over the load increase. The effect of enforcing reactive power limits for the PV models for the DFIG and DDSG made no difference to either the convergence of the load flow iterations or to the VSL. If the wind farm were placed closer to the weak load bus being monitored, a decrease in that bus’s VSL would be expected. It can therefore be concluded that the placement of such wind farms is preferably some distance from weak buses. The importance of correct initialisation of dynamic models was highlighted in this paper, and the results of each of the models were used to calculate the initial conditions of some sample SCIG, DFIG and DDSG dynamic models. The importance of applying generator reactive power limits was evident; without doing so the dynamic models could be initialised with unrealistic or impossible values. Future directions of this research include dynamic model validation and verification. A PPENDIX Generator Type SCIG SCIG DFIG DFIG DDSG DDSG Model PQ Improved PQ PQ PV with Q limits PV PV with Q limits vds , vqs (pu) 0.98, -0.067 0.95, -0.058 1.03, -0.09 0.97, -0.11 1.03, -0.14 0.97, -0.11 ids , iqs (pu) 0.26, -0.95 0.27, -0.12 0.25, -0.39 0.29, -0.10 0.43, 1.08 0.29, -0.10 Table 5. Comparison of initial conditions for all models results for the stator voltages and currents, whilst the PQ model of the DFIG showed a higher value for the voltage. The application of generator reactive power limits to the PV model made a difference to the voltage and a significant The equivalent circuits and equations for the wind turbine generator models used in this paper are presented below. The following equations represent the SCIG: Zin = Rs + jX1 + (jXm||(( Rr ) + jX2 )) s (10) Va Vs Zin Va = Z2 = Vs + Is (Rs + jX1 ) (13) P = 3|Vs ||Is |cosθ (14) Is Ir = (11) (12) ACKNOWLEDGEMENT The authors would like to thank the A.E. Rowden White Foundation for their generous, ongoing financial support of this research. The use of the MATPOWER software package is gratefully acknowledged: http://www.pserc.cornell.edu/matpower/. R EFERENCES Fig. 22. Per-phase equivalent circuit of SCIG Q = 3|Vs ||Is |sinθ (15) where Zin is the input impedance, Rs is the stator resistance, Rr is the rotor resistance, X1 is the stator reactance, X2 is the rotor reactance, s is the slip, Is is the stator current, Ir is the rotor current, Z2 is the rotor impedance and Va is the voltage across the magnetizing reactance. P and Q are the real and reactive power, respectively. Fig. 23. Per-phase equivalent circuit of DFIG The following equations represent the DFIG: Vs = Rs Is + jX1 Is + jXm (Is + Ir ) Vr Rr = Ir + jX2 Ir + jXm (Is + Ir ) s s (16) (17) The power generated is the sum of the stator and rotor generated powers: Ps + jQs = 3Vs Is∗ Pr + jQr = 3Vr Ir∗ (18) (19) Pe = P s + P r (20) where the subscripts are as described earlier. The following Fig. 24. Per-phase equivalent circuit of DDSG equations represent the DDSG: Vφ = EA − jXs IA − RA IA (21) where Vφ is the per-phase terminal voltage, EA is the internal generated voltage, XS is the synchronous reactance, IA is the current, and RA is the stator resistance. The power generated can be expressed as: P = 3Vφ IA cosθ Q = 3Vφ IA sinθ (22) (23) [1] J. Slootweg, H. Polinder, and W. Kling, “Initialization of wind turbine models in power system dynamics simulations,” in IEEE Porto Power Tech Proceedings, vol. 4, p. 6pp, 2001. [2] J. Slootweg and W. Kling, “Is the answer blowing in the wind?,” IEEE Power and Energy Magazine, vol. 1, pp. 26–33, November–December 2003. [3] R. Koessler, S. Pillutla, L. Trinh, and D. Dickmander, “Integration of large wind farms into utility grids pt. i - modeling of dfig,” in Proceedings of the IEEE Power Engineering Society General Meeting, vol. 3, 2003. [4] A. FeijoĢo and J. Cidras, “Modeling of wind farms in the load flow analysis,” IEEE Transactions on Power Systems, vol. 15, no. 1, pp. 110– 115, 2000. [5] G. Coath, M. Al-Dabbagh, and S. Halgamuge, “Particle swarm optimisation for reactive power and voltage control with grid-integrated wind farms,” in Proceedings of the IEEE Power Engineering Society General Meeting, vol. 1, 2004. [6] L. Holdsworth, X. Wu, J. Ekanayake, and N. Jenkins, “Direct solution method for initialising doubly-fed induction wind turbines in power system dynamic models,” IEE Proceedings on Generation, Transmission and Distribution, vol. 150, no. 3, pp. 334–342, 2003. [7] D. Alves, d. L.C.P., and V. Castro, C.A. da Costa, “Continuation fast decoupled power flow with secant predictor,” IEEE Transactions on Power Systems, vol. 18, no. 3, pp. 1078–1085, 2003.