Design and benchmark of a Multilevel Converter for large-scale wind power systems Benjamin Sahan, Samuel V. Ara´ujo, Alfred Engler
Fernando L. M. Antunes
Institut f¨ur Solare Energieversorgungstechnik, ISET e.V. University of Kassel K¨onigstor 59, D-34119 Kassel, Germany Email:
[email protected]
Federal University of Cear´a Electrical Engineering Department Campus do Pici s/n, Caixa Postal 6001 Fortaleza, CE 60.455-76, Brazil Email:
[email protected]
Abstract—This paper deals with the basic design and benchmark of a multilevel dc-ac conversion system for interconnecting a high power wind turbine to the grid. The benchmark has been performed for different system voltages and devices in terms of efficiency, grid filter design and cost. In this context the NPC loss calculation and basic filter design rules are presented. Finally the best-in-class configuration could be identified. Index Terms—Wind power generation, synchronous generator, multilevel converters, IGBT, efficiency.
traction [4]. However, the grid feed-in operation is different from the operation on a motor load. Thus there are different requirements, especially regarding power quality where regulations like the IEEE 519-1992 recommendation and the German Guideline VDEW apply. So, this report highlights the design and performance of the line side converter under consideration of efficiency, power dissipation and system voltage. The performance also depends on the semiconductors incorporated and the operating frequency of the power switches which mainly determines the dimension of the output filter.
I. I NTRODUCTION The trend towards multi-MW wind turbines has called up for new concepts in the design of wind energy conversion systems. Economic viability of offshore wind turbines clearly scales with power and efficiency of generators and power conversion systems. Increase of civil works and material cost will be well paid off by the additional revenues from the higher energy production per turbine. Within this trend, power electronic multilevel converters have been seen as an appropriate technology for the wind energy conversion system because they can operate at high power and high voltage. Multilevel converters can synthesize output waveforms using more then two voltage levels, so the spectrum of the output waveform is improved when compared to the one provided by conventional two level converters [1], [2]. Looking towards the multi-MW wind turbines, a 7MW unit is proposed. Focus will be on a full converter solution for synchronous generators in order to avoid dealing with slip rings and possible Low Voltage Ride Through (LVRT) difficulties [3]. The system can be divided into three sub-stages (see Fig. 1):
II. P ROPERTIES OF THE INVERTER TOPOLOGY Multilevel converters can extend power electronic systems to high power and high voltages ratio. The general concept of multilevel converters involves utilizing a higher number of active semiconductor switches to perform the power conversion in small voltage steps. An important feature of multilevel converters is that the semiconductors are wired in a series-type connection, which allows operation at higher voltages. However, the series connection is typically made by clamping diodes, which eliminates overvoltage concerns. Among several proposed multilevel topologies the three-level diode-clamped, or simply called Neutral-Point Clamped - NPC inverter was the first widely implemented by the industry and it continues to be extensively used in high voltage and high power applications.
T1 +Vdc/2
D5 Vdc
va0
Level 2
a
0
D6
Level 1
T2 T1 T2 T1 T2 T3 T2 T3 T2 T3 T2 T3 T4 T3 T4
T3 -Vdc/2
Level 3
T4
n,T Fig. 1.
P,Q
Full converter back-to-back system
The machine side converter provides torque control. The line side converter and its output filter perform active and reactive power control. The dc-link decouples both sides. The whole conversion system is then connected to an isolating transformer which steps up the voltage to the required level of the wind farms Point of Common Coupling (PCC) and it also provides additional short circuit power. Inverters in this power range are well known from mediumvoltage drives (MVDs) used in industrial applications such as rolling mills, fans, pumps, conveyors or marine appliances, mining and
Fig. 2. states
A phase-leg of the 3-Level NPC inverter and its principle switching
Fig. 2 shows one phase-leg of the NPC and its output voltage waveform. It is composed by four switches T1 to T4 and the antiparallel diodes D1 to D4. Two series connected capacitors provide the neutral point 0 on the dc side of the inverter. Voltages across the switches are only half of the dc input voltage due to the diodes connected to the neutral point. Since the voltages across the switches are limited by the conduction of the diodes D5 and D6 connected to the neutral point, this class of multilevel is termed diode-clamped multilevel inverter. Due to capacitor voltage balancing
III. P OWER SEMICONDUCTORS
cost related to switching power [€/kVA]
System performance highly depends on the semiconductors incorporated. This directly affects efficiency, heat transfer, system voltage and finally cost. It is desirable to use Insulated Gate Bipolar Transistors (IGBT) which are emerging in high power medium voltage applications. High power IGBTs are available in four main voltage classes: 1200V, 1700V which could be still regarded as low voltage switches and 3300V and 6500V, considered as high voltage switches. A survey on cost of high power IGBT modules is illustrated in Fig. 3.
10 rel. switching losses Etot [Ws/kVA]
issues, practical implementation of diode-clamped inverters have been mostly limited to the originally proposed NPC structure [5]. And as the number of voltage levels of the inverter increases the THD of the output voltage waveform has little improvement [2]. To sum up, the major benefits over the common (low voltage) 2-Level inverter are: • Series-type connection of power switches allows operation at higher voltages • Usage of devices with lower voltage rating • Lower dv/dt and EMI • Enhanced power quality • Decreased common-mode voltage thus less problems with bearings As main drawbacks it should be mentioned: • Higher number of components • Voltage balancing issues • Operation at medium voltage level requires special trained maintenance personnel.
1
0,1
0,01 100
1000
10000
rated voltage Vce,max [V]
Fig. 4. Normalized switching losses of IGBTs, Etot /(VCE,max · IC,max ), Source: data sheets and IPOSIM data bank INFINEON diff. IGBT designs of the 2nd and 3rd generation [6]
IV. L OSS CALCULATION OF POWER SWITCHES A. Modulation Pulse Width Modulation (PWM) technique is widely used to synthesize the ac output voltage of inverters. It can be implemented by controlling the duty cycle D(t) which is the ratio between the time a switch is on and the inverter switching period. The inverter switches are operated at high frequency in a way to provide a controllable output voltage at a lower fundamental frequency. Without going into detail on PWM it is often necessary to use an overmodulation technique to have a higher utilisation rate of the dc-link voltage. This can be achieved by adding a 3rd harmonic to the reference duty-cycle signal [7] according to (2): D(t) = M · cos(ωN t) + 0, 16 · cos(3ωN t + 3π)
0,53
(2)
The modulation index M is defined by the ratio of dc input voltage and amplitude of the inverter output voltage Vi . It can be virtually increased by overmodulation to about 1,15. √ Vi · 2 M= (3) Vdc /2
0,43
0,33
0,23
0,13
0,03 1000
10000 rated voltage VCE,max [V]
Fig. 3. Survey on specific cost of high power IGBT modules (requested quantity:1k)
Data in Fig. 3 was obtained by dividing the price of an IGBT module by its switching power, i.e. the product of its rated voltage and current. Specific cost cleary increase with rated blocking voltage VCE,max . Moreover, low voltage switches naturally need more current carrying capability, what requires more silicon, to reach the same switching power as high voltage switches. Therefore it can be concluded that a single high voltage chip is much more expensive than a low voltage one. Another major indication in chip technology are the specific switching losses. As illustrated in Fig. 4 switching losses clearly scale non-linearly with the rated blocking voltage of the device. The relation is roughly given by [6]: Etot ∼ (VCE,max )1,4 (1) IC · VCE,max In other words, it is obvious to aim at using low voltage IGBTs to have less switching losses and lower cost, but this also implies increased conduction losses and increased mechanical and thermal stress on bus bars, cables and protection equipment (circuit breakers etc.). Hence follows that a trade-off needs to be figured out.
B. Conduction loss The conduction losses are composed of losses in the IGBTs and the diodes. They can be calculated for each switch with (4) based on the linearised forward voltage drop given in the datasheets. 2 PC = U0 · Iav + rF · Irms
(4)
The line side power factor shall be cos ϕ=1 and the output phase current iph (t) shall be regarded as an ideal current source for reasons of simplicity. See also [8],[9].
Z
5TN /4
IT 1,av = IT 4,av
1 = TN
|iph (t)| · |D(t)| dt
(5)
3TN /4
IT 1,rms = IT 4,rms
v u u u 1 =t
Z
5TN /4
iph (t)2 · |D(t)| dt
TN
(6)
3TN /4
Z
5TN /4
IT 2,av = IT 3,av
1 = TN
|iph (t)| dt 3TN /4
(7)
IT 2,rms = IT 3,rms
v u u u 1 =t
Z
5TN /4
iph (t)2 dt
TN
(8)
The junction temperature at the outer IGBT T1/T4 is most critical since there are the highest power losses. The maximum allowed power dissipation PV,max in T1 can be calculated as follows: ≈0
3TN /4
z }| {
ID5,av = ID6,av = IT 2,av − IT 1,av
(9)
q IT2 2,rms − IT2 1,rms
ID5,rms = ID6,rms =
(10)
In case of cos ϕ=1 the IGBT freewheeling diodes D1-D4 do not conduct and therefore are not included in the loss calculation. C. Switching loss The switching energy functions are given in the IGBT datasheet depending on the current of IGBT and diode. The turn-on switching energy usually also contains losses induced by the free-wheeling diode’s reverse-recovery charge at the instant of commutation. The average switching losses over one half period of the output phase voltage are caused by the commutation of T1 and D5 while T2 is permanently on [8], [9]:
Z
5TN /4
PS,T 1
Vdc /2 1 = · fs · VCE,test TN
Eon+of f (|iph (t)|)dt
(11)
3TN /4
PV,T 1 max =
∆Tj,T 1 max − PV,D1 Rth,HA Rth,JC T + Rth,CH T + Rth,HA
(15)
In practical design a maximum power dissipation of 5 kW per module is a realistic value. V. O UTPUT FILTER CONSIDERATIONS The quality of the current injected into the grid is based on the requirements of the IEEE 519-1992 recommendation [11] and the German Guideline VDEW [12], [13] for connection and parallel operation of generators at the medium voltage grid. According to the IEEE recommendation, a maximum current distortion is given as a percentage of the equipment nominal current, independently of the size of the grid for power generation devices. Meanwhile, the VDEW guideline establishes limits based on a current-power ratio constrained by the grid voltage level depending on the short circuit capability at the Point of Common Coupling (PCC). The line side inductors shape the pulsed width modulated currents from the inverter into a continious sinusodial waveform (see Fig. 6). These inductors tends to be heavy, bulky and costly.
Z
5TN /4
PS,D5
Vdc /2 1 = · fs · VCE,test TN
Erec (|iph (t)|)dt
(12)
3TN /4
D5
vph(t)
vi(t)
0
(13)
Fig. 6. NPC simplified circuit with L-filter (positive halfwave), without common mode voltage
PS,D5 ≈ kD · fs · (IT 1,av + ID5,av )
(14)
As a general rule of thumb the inverter side inductance can be assessed by the maximum ac current ripple. It should not exceed 1025% of the phase current amplitude at rated power [14]. A rough approximation of the current ripple time function can be calcuated ! assuming no common mode voltage and no overmodulation (D(t) = M · cos(ωN t)) according to equation (16) which is derived from Fig 6.
The transfer of dissipated power in the semiconductors is a crucial issue. The datasheet contains the thermal resistances of IGBT modules. The thermal resistance between heat-sink and ambient Rth,HA varies with the type of module and the cooling method. According to [10], typically the best value which can be achieved by water cooling (6-10 l/min) is 4,5 K/kW while forced air cooling is almost an order of magnitude worse. junction
case
Rth,JC_T
heat-sink
Rth,JC_D
1000
800
ambient 2 'Iph t 780V
Rth,CH_T
2 'Iph t 550V
Rth,HA
ǻTj,T1 Rth,CH_D
ǻTj,D1
Fig. 5.
iph(t)
PS,T 1 ≈ kT · fs · (IT 1,av + ID5,av )
D. Heat transfer
Pv,D1
Li
Vdc/2
If a first order approximation of the switching energy functions is applied the switching losses are only dependant on the switching frequency fs and the average switched current:
Pv,T1
T1
Equivalent steady-state thermal circuit of IGBT module
600
D t 780V 1000 D t 550V 1000 400
200
0
0
0.005
0.01
0.015
t
Fig. 7. Phase current ripple ∆Iph (peak-to-peak) at different phase voltages Vph ; Li =170 µH, fs = 2kHz, Vdc =2200V, M=1, P =7MW
(16)
The current ripple at the output is the largest for a duty cycle of 50%. Therefore: Vdc Li ≈ (17) 16 · fs · ∆Iph,max Another important aspect that needs to be taken into account is the ac voltage drop across the inductor. Equation 19 includes a 10% tolerance on the grid voltage and as mentioned before the modulation index can not exceed Mmax =1,15 with overmodulation. Vi
=
Vdc = s 2
¡
Li,max
q 2 2 · L21 · I12 + Vph ωN
M
=
1.15 V2dc
¢2
−
¡
2 1.1Vph
MW) efficiency actually slightly increases as conduction loss become smaller (Fig. 9). IGBT conduction
IGBT switching
(18)
20 15 10 5 0 1700V(2kHz) 3300V(1.5kHz) 6500V(800Hz) Voltage class (switching frequency)
(19)
2 2 ωN I1
TABLE I PARAMETERS OF CHOSEN LINE SIDE INDUCTORS Volt. class [V]
fs [Hz]
Li [µH]
∆Iph /ˆiph
WLi [Ws]
1700
2000
170
0,097
723,75
3300
1500
403
0,2
469,84
6500
800
1482
0,3
618,31
A more detailed analysis of LC and LCL-filter designs will be published in the near future. VI. B ENCHMARK OF VOLTAGE CLASSES AND DEVICES The power losses of several IGBTs from different manufacturers were calculated. It turned out that effciency was quite similar (+0,4%) throughout all voltage classes and devices.
Fig. 9. Best-in-class IGBT power loss calculation for line side inverter at half load
An important issue concerns the heat transfer of the disspated power. Due to the nature of the NPC the loss distribution is unequal. While the power loss of inner switches T2,T3 mainly consist of conduction loss, the outer switches T1,T4 have both conduction loss and the main part of the switching loss. As illustrated in Fig. 10 this issue becomes even worse at higher voltage classes where switching losses are dominating. Pv_T11,max
Pv_T12,max
Diode conduction
Pv_max
8,00 7,00 6,00 5,00 4,00 3,00 2,00 1,00 0,00 3300V(1.5kHz)
6500V(800Hz)
Diode switching
Fig. 10. Power dissipation across the switches at full load (7MW) and max. allowed power dissipation per module Pv,max
80 Total semi losses [kW]
Pv_D5,max
9,00
1700V(2kHz)
IGBT switching
Diode switching
25
¢
Based on this the switching frequency and the value of the inverter side inductor were chosen for the different voltage classes. The physical size of the inductors can be assessed by the peak energy 2 stored in the inductor WLi ≈ Li Iph .
IGBT conduction
Diode conduction
30 Total semi losses [kW]
√ − ( 2Vph cos(ωN t)) D(t) · 2 · Li fs
dissipated power per module [kW]
∆Iph (t) =
Vdc 2
70
However, a solution to this problem has been proposed in [4] at the cost of additional switches.
60 50 40
TABLE II PARAMETERS AND RESULTS
30 20 10 0 1700V(2kHz) 3300V(1.5kHz) 6500V(800Hz) Voltage class (switching frequency)
Fig. 8. Best-in-class IGBT power loss calculation for line side inverter at full load
The results of the best-in-class devices are shown in Fig. 8 for full load conditions. While efficiency is fairly equal among the voltage classes, low voltage devices are operated at higher switching frequency and despite high operating current the conduction losses remain acceptable. This is due to the high number of chips in the module and thus low on-state resistance. At partial load (3.5
Volt. class [V]
Vdc [kV]
Vph−ph [kV]
ηmax (7MW)
1700
2,2
1,4
99,24
3300
4,2
2,6
99,14
6500
7,2
4,5
98,99
VII. C ONCLUSION To assess the applicability of a 3-Level Neutral Point Clamped (NPC) Inverter in large-scale wind power systems, a benchmark of devices and system voltages has been carried out. For this purpose it was necessary to calculate the efficiency of the power semiconductors and to consider the output ac filter of the line side inverter in order to find an appropriate switching frequency. It can be concluded that low
voltage switches (1.7kV) offer the best cost-performance ratio and due to operation at higher switching frequency the size of bulky filter inductors can be reduced. When a higher system voltage is desired, e.g. to put the power conversion system outside of the nacelle, the usage of 3.3kV devices would be a good option. Alternatively, the direct series connection of two 1.7 kV IGBTs per branch would provide a higher system voltage with low voltage switches only. This method is quite common in MV applications [15] but it needs additional efforts on synchronisation and symmetry. In any case the unbalanced loss distribution needs to be taken into account as this hampers cooling efforts and severely limits further increase of the converters output power. ACKNOWLEDGMENT The authors would like to thank the European Commission and the work package partners for their support in the project ”Upwind” (Contract No. SES6-019945). This article reflects only the author’s views and the Community is not liable for any use that may be made of the information contained therein. R EFERENCES [1] J. Pou, ”Modulation and control of three-phase PWM multilevel converters”, PhD Dissertation, Universidat Politecnica de Catalunya, 2002. [2] T. Skavarenina ”The Power Electronics Handbook - Cap 6 by Corzine, Keith” CRC Press, 2002. [3] L. Rouco, F. Fernandez, J.L. Zamora y P. Garcia, ”Comparison of the dynamic response of wind power generators of different technologies in case of voltage dips”, Cigre, Paris, 2006 [4] Th. Br¨uckner, St. Bernet, H. G¨uldner ”The Active NPC Converter and its Loss-Balancing Control”, IEEE Transaction on Industrial Electronics, Vol. 52, No. 3, June 2005 [5] D. Gr. Holmes, T. A. Lipo ”Pulse Width Modulation for Power Converters - Principle and practice”. IEEE Press, 2003. [6] P. Zacharias, B.Burger, ”Overview of recent Inverter developments for grid-connected PV systems”, Proceedings of the 21st European Photovoltaic Solar Energy Conference and Exhibition, Dresden, Germany, 0408 Sept. 2006. [7] F. Jenni , D. West, ”Steuerverfahren fuer selbstgefuehrte Stromrichter” vdf Hochschulverlag AG an der ETH Zrich, 1995 [8] G. Tomta, R. Nielsen, ”Analytical Equations for Three Level NPC Converters”. 9th European Conference on Power Electronics and Applications, EPE, 2001 [9] T. J. Kim, D. W. Kang, Y. H. Lee, and D. S. Hyun, ”The analysis of conduction and switching losses in multi-level inverter system,” in PESC Record - IEEE Annual Power Electronics Specialists Conference, 2001, pp. 1363. [10] Infineon Technical Documentation, ”Dimensioning program IPOSIM for loss and thermal calculation of Eupec IGBT modules”, www.infineon.com, last checked Jan 2007 [11] IEEE Std. 519-1992 - IEEE Recommended Practices and Requirement for Harmonic Control in Electrical Power Systems - IEEE Industry Applications Society/ Power Engineering Society. [12] VDEW Eigenerzeugungsanlagen am Mittelspannungsnetz - Richtlinie fr Anschluss und Parallelbetrieb von Eigenerzeugungsanlagen am Mittelspannungsnetz. [13] DIN EN 61400-21 - Windenergieanlagen - Teil 21 : Messung und Bewertung von netz-gekoppelten Windenergieanlagen - Deutsches Institut fr Normung e.V. und VDE Verband der Elektrotechnik Elektronik Informationstechnik e.V. [14] Wang C.T, Ye Z., Sinha G, Yuan X., ”Output Filter Design for a Grid-Interconnected Three-Phase Inverter”, Power Electronics Specialist Conference PESC, 2003 [15] R. Sommer, A. Mertens, M. Griggs, H.-J. Conraths, M. Bruckmann, T. Greif, ”New medium voltage drive systems using three-level neutral point clamped inverter with high voltage IGBT”, Industry Applications Conference, 1999. Thirty-Fourth IAS Annual Meeting. Conference Record of the 1999 IEEE, Vol.3, Iss., 1999
