Professional Documents
Culture Documents
Electrical Power and Energy Systems: Shuai Li, Jiyuan Zhang, Jianzhong Xu, Chengyong Zhao T
Electrical Power and Energy Systems: Shuai Li, Jiyuan Zhang, Jianzhong Xu, Chengyong Zhao T
A R T I C LE I N FO A B S T R A C T
Keywords: With the high voltage direct current Transmission (HVDC) booming, HVDC grid has received wide attention. As
HVDC circuit breakers an essential component in HVDC grid, high voltage direct current circuit breakers (DCCB) requires urgent and
Current limiters intensive study. A novel topology for current limiting DCCB (CL-DCCB) is proposed in this paper. The topology
Over-voltage protection consists of several units which are divided into two classes: main circuit breaker (MCB) and branch circuit
Stability
breakers (BCBs). The number of inductor branches can be flexibly selected to enhance the current limiting effect.
The CL-DCCB can start current limiting operation when a suspected fault occurs. When the detection circuit
reveals what has happened, it can be determined whether to implement breaking operation or recover to normal
state. This mode allows longer fault detection time and circuit breaker operation delay while guaranteeing the
fault current within the maximum limit of the system. Finally, Simulation model and experiment prototype are
built to study the design criteria for CL-DCCB. It is observed that the maximum detection delay can be extended
to 12 ms. Moreover, the current limiting effect can be enhanced when the number of inductor branches or the
inductance of each branch is increased.
1. Introduction (SMs) with DC fault clearance capability [6,7]. Once short circuit fault
occurs, the fault current discharge path can be blocked by blocking
DC grid is a new type of power transmission system, which is ob- MMC SMs. Finally, the mechanical switch can be used to interrupt the
tained from voltage-source converter (VSC) type HVDC [1,2]. It inter- fault line, thus, the system fault characteristics can be greatly improved.
connects multiple AC and DC systems with higher reliability due to its However, the novel SMs will inevitably increase the cost and power loss
redundant DC lines. DC grid technology is especially suitable for large- of MMC. More importantly, all the converter station must be blocked
scale wind power or photovoltaic and other new energy integration, under fault condition. Thus, the loss of power transmission capacity will
which is the future direction of DC transmission technology [3,4]. affect the normal and stable operation of the whole grid. Taking Fig. 1
China is building a demonstration project named as Zhangbei DC grid. as an example, if a short-circuit fault occurs in the transmission line
The project is designed for the collection and transportation of large- among converter station C1-C4, all these converter station should be
scale wind power, photovoltaic, energy storage, and other energy blocked, then mechanical switch can be used to isolate faulty lines
forms. The rated voltage of this DC grid is ± 500 kV, with approxi- when the fault current drops to zero. The outage of those converter
mately 648 km overhead transmission lines [5]. stations means a quick stop of electricity transmission in the whole grid,
Due to the low damping of DC system and no zero crossing point of which is unacceptable.
the DC current, it is difficult to isolate the DC fault. Especially in large- The DC-DC converter can also be configured in the grid to isolate
scale power grid. Take a scenario of DC grid shown in Fig. 1 as an the DC fault, and the high frequency transformer of the DC-DC con-
example, the DC grid is divided into two parts by a DC-DC converter. verter is undoubtedly a good isolating device. Through the control of
The upper part C1-C4 is a cyclic and radiation mixed connection the DC-DC, the fault can be isolated quickly. However, the configura-
structure. The lower part is a point-to-point DC system. Obviously, the tion of DC-DC in the grid is limited with the issue of cost and power loss,
normal operation of the whole network will be hardest hit if the fault thus the protection area of the grid is limited [8,9].
cannot be cleared in time. SCFCL is featured with good current limiting effects and fast re-
One way to solve this problem is to use novel MMC sub-modules sponse, but the technology is not yet mature, and expensive [10,11].
☆
This work was supported in part by National Natural Science Foundation of China (51777072) and in part by Fundamental Research Funds for the Central
Universities (2017XS018). The authors are with the State Key Laboratory of Alternate Electrical Power System with Renewable Energy Sources, North China Electric
Power University (NCEPU).
⁎
Corresponding author at: 2 Beinong Road, Huilongguan Town, Changping District, Beijing, China.
E-mail address: lishuaiwork@foxmail.com (S. Li).
https://doi.org/10.1016/j.ijepes.2018.07.042
Received 18 April 2018; Received in revised form 5 June 2018; Accepted 19 July 2018
0142-0615/ © 2018 Elsevier Ltd. All rights reserved.
S. Li et al. Electrical Power and Energy Systems 104 (2019) 933–942
Moreover, it does not have the fault isolation ability. It still relies on 2. Topology and working principle
other equipment to cut off the DC fault current.
The DCCB is a relatively safe solution because it can be flexibly 2.1. Derivation of the topology
arranged at any point in the grid and can interrupt the fault circuit
without affecting the operation of the converter and the normal DC line The proposed CL-DCCB consists of MCB and BCBs, as shown in
[12]. The hybrid DCCB has been deeply studied by [13]. The design Fig. 3(a), where L represents the reactors. The MCB includes three
idea of this type DCCB is to close the mechanical switch under normal branches: branch 1 is a low loss branch, branch 2 is a power electronic
condition to reduce the on-state power loss. The power electronic branch, branch 3 is an energy absorption circuit, as shown in Fig. 3(b).
branch only works under the condition of fault status. A 320 kV/2kA where, power electronic branch can break fault current rapidly. The
hybrid DCCB is manufactured by ABB, Ltd., in 2012, as shown in Fig. 2. function of energy absorption circuit is to protect the IGBTs from over-
On the fault status, firstly, block all the IGBTs of branch 1 and trigger all voltage, which consists of MOA (metal oxide arrestor). It should be
the IGBTs of branch 2, the fault current will be transferred from branch noted that IGBT modules displayed in Fig. 3(b) have been configured
1 to branch 2; secondly, open the UFD (ultra-fast disconnector) of with equalizing circuit to prevent partial overvoltage [15–17]. Con-
branch 1 when the fault current in this branch is attenuated to zero; sidering that each IGBT is configured with a diode rectifier circuit, the
thirdly, close the IGBTs of branch 2 when the UFD is fully opened, then fault current is converted to the same direction. Therefore, the MCB can
the fault current will be transferred to the MOA and attenuated to zero be able to cut bidirectional current on the premise of half the usage
quickly. amount of IGBTs.
However, the fault current may rise to the maximum limit of the The topology of BCB is shown in Fig. 3(c), which includes three
IGBTs in hybrid DCCB under the condition of larger capacity trans- branches: branch 1 is a low loss branch, which is composed of UFD and
mission occasion. Increasing the current limiting reactance can avoid several IGBTs, branch 2 is composed of capacitor banks, branch 3 is an
this problem but will affect the dynamic characteristics of the system. energy absorption circuit.
Therefore, many fault detection methods are unable to meet the needs
of DC grid protection due to the time delay, which can only be used as 2.2. Working principle
backup protection strategies. A feasible fault protection strategy in DC
grid: the measured ROCOV [14], can meet the needs of rapidity. But in With issue to MCB, when conducting UFD and all the IGBTs in
some cases it may cause misjudgment. In the case of small resistance branch 1 and blocking all the IGBTs in branch 2, the MCB will run
short circuit occasion, a local line fault may be mistaken as a remote under low power loss mode. As path A in Fig. 4(a). In order to quicken
line fault, because the local ROCOV value may be lower than the critical the breaking speed, fault current path should always be transferred to
threshold value to discriminate local and remote fault, and the mis- branch 2 when a suspect fault occurs. Then breaking operation can be
judgment cannot be amended through communication due to high executed quickly when the fault is determined. The process can be
delay. At the same time, the fault current rising-rate is still too large. If described in detail as follows: firstly, trigger all the IGBTs in branch 2
the current limiting operation cannot be taken in time, the fault current and then block all the IGBTs in branch 1; secondly, open the UFD in
will exceed the maximum breaking capacity of DCCB, and further lead branch 1 when the current in this branch drops to zero. The current
to the outage of the DC grid. Thus, this type of hybrid DCCB cannot fully path in this working mode is shown as path B. If the fault is confirmed,
meet the need of DC grid. the breaking operation can be executed by blocking all the IGBTs in
branch 2. As the current is interrupted by these IGBTs, the overvoltage
will be induced by system inductance, then the fault current will be
Branch 1 transferred to MOA when the over-voltage value exceeds the protection
voltage of the MOA, as current path C in Fig. 4(a).
Ultra Fast load commutation Due to the function of the diode rectifier bridge, unidirectional ar-
Disconnector switch rangement of the IGBTs can meet the requirement of the bidirectional
breaking operation. The forward and reverse current paths are shown in
Current Limiting
Fig. 4(b) separately. Considering that the capacity and cost character-
Reactor
istics of diodes under the same parameters are obviously better than
that of IGBTs, and the halve of IGBTs can also reduce the static and
Main breaker(Branch2) dynamic voltage balancing circuit, which can effectively reduce the
system cost.
Fig. 2. Generic structure of hybrid DCCB developed by ABB. For the BCB, the UFD and IGBTs of branch 1 are conducted under
934
S. Li et al. Electrical Power and Energy Systems 104 (2019) 933–942
C
branch
breaker 2
MOA2
L3
(a) The current paths of BCB
(a) Structure of CL-DCCB UFD
Open Gap
MOA
Branch 1 D1 D2
Contact
UFD
T1 T2
Branch 2 +
C
(b) The opening process of UFD
Branch 3 Fig. 5. Operation process of the BCB.
MOA1 MOAx
(b) Topology of MCB normal state, and the current path is shown as path A in Fig. 5(a). The
power loss is relatively low as there are only several IGBTs in this
MOA branch. When the current limiting operation is required, block all the
Branch 1 IGBTs in branch 1, After a short delay, open the UFD when the current
D1 D2
of branch A dropped to zero, then fault current path as shown by path
B. The opening process of UFD, as shown in Fig. 5(b), usually takes
about 2 ms. Considering that the capacitor can restrain the overvoltage
UFD
T1 T2 rising-rate, a high voltage capacitor bank is used as the auxiliary
switching device in BCB. The insulation voltage level and the gap dis-
Branch 2 tance of UFD are positively correlated [18]. Therefore, the voltage
C breakdown of UFD can be prevented by an appropriate capacitor. In
Branch 3 this case, the capacitor voltage is always less than the insulation voltage
of UFD during the operation process. After the UFD is extremely open,
MOA the capacitor voltage will gradually achieve the MOA protection value.
(c) Topology of BCB The function of MOA is to ensure that the voltage at both ends of the
capacitor during current limiting operation will not exceed its max-
Fig. 3. Topology of CL-DCCB. imum designed value. Then, fault current will be transferred to path C
and attenuated to zero quickly. Through this control, the branch in-
MOA1 ductors are connected in series, thus the current limiting effect is
Path A Path B Path C achieved.
D1 D2
The working mode of proposed CL-DCCB is as follows: 1) Normal
operation mode. In this mode, inductors of different branches are
UFD
T1 T2 connected in parallel with current split into N branches (N is the
MOA2 MOAx number of reactor branches); 2) Fault current limiting mode. In this
mode, the reactors of each branch are connected in series, with the
system’s inductance increased from L/N to N*L to limit the suspected
fault current rising-rate; 3) Current limiting plus circuit breaking mode.
In this mode, the branches series will be connected with each other to
(a) Current paths under different working mode reduce the current rising-rate before being shut down. Then the CL-
DCCB continues to break or recover according to the feedbacks. This
mode provided extra time for fault detection. However, conventional
DCCB may have an increased risk of misjudging as the time for fault
detection is relatively shorter.
The current directions in different modes are shown in the Fig. 6
(b) Current paths of power electronic branch Where, normal running state is shown in Fig. 6(a), fault current limiting
state is shown in Fig. 6(b), and the rising-rate of fault current under this
Fig. 4. Current paths of MCB. mode is effectively suppressed.
The operational process of mode “3″ is shown in Fig. 7. Its working
process is as follows: Close all the UFDs and IGBTs of branch 1 in the
935
S. Li et al. Electrical Power and Energy Systems 104 (2019) 933–942
Main L1
breaker + i i1 i2 i3 +
iLimit
branch
breaker 1 u L1 L2 L3 u L1 L2 L3
L2
- -
branch
breaker 2 Fig. 8. Current limiting transformation process.
L3 1 t
⎧ i1 = L1
∫−∞ u (ξ ) dξ
⎪ 1 t
(a) Normal working mode
i2 = L2
∫−∞ u (ξ ) dξ
⎨
⎪i = 1 t
Main L1 3 L3
∫−∞ u (ξ ) dξ (1)
⎩
breaker
where u(ξ) is the transient voltage at both ends of the CL-DCCB.
The DC line current is shown in (2):
branch
breaker 1 i = i1 + i2 + i3
L2
= ( 1
L1
+
1
L2
+
1
L3 )∫t
−∞
u (ξ ) dξ
1 t
= L
∫−∞ u (ξ ) dξ (2)
The total inductance of the breaker is shown in (3):
branch
breaker 2 L1 L2 L3
L=
L2 L3 + L1 L3 + L1 L2 (3)
L3
The current of each branch can also be shown in (4)–(6):
L L 2 L3
(b) Current limiting mode i1 = i= i
L1 L2 L3 + L1 L3 + L1 L2 (4)
Fig. 6. Different current paths under normal mode and current limiting mode.
L L1 L3
i2 = i= i
L2 L2 L3 + L1 L3 + L1 L2 (5)
Yes
Breaking operation
No L L1 L2
i3 = i= i
Normal Permanent
L3 L2 L3 + L1 L3 + L1 L2 (6)
suspected fault
mode Fault
As shown in (4)–(6), when the inductance of these branches are
Yes
Recovering operation
equal, the current value of each branch will be consistent. Thus, the
Current limiting
No current value of each branch is 1/3 of the total system current.
mode
The value of flux linkages under normal state is shown in (7):
Fig. 7. Operational process under fault condition.
ψL (0−) = ψL1 (0−) + ψL2 (0−) + ψL3 (0−)
= L1 iL1 (0−) + L2 iL2 (0−) + L3 iL3 (0−) (7)
two BCBs and the MCB under normal conditions. When a suspected
failure occurs, current limiting mode will be executed: firstly, block all Under the current limiting mode, the reactors of each branch are
the IGBTs in branch 1 of all the BCBs and the MCB, then, open the UFD connected in series. The value of flux linkages is shown in (8).
of BCB when the fault current in branch 1 dropped to zero. In case of
ψL (0+) = ψL1 (0+) + ψL2 (0+) + ψL3 (0+)
permanent fault, all the IGBTs of branch 2 in the MCB should be
switched off to cut the fault current, otherwise, recover to normal op- = (L1 + L2 + L3 ) iLimit (0+) (8)
eration by closing all the UFDs in MCB and BCBs, then turn on the According to the conservation principle of flux linkages, (9) can be
IGBTs in branch 1 of MCB and BCBs. developed:
ψL (0−) = ψL (0+) (9)
2.3. Analysis of current limiting operation Thus, the current of the branches is limited to the value before the
current limiting operation. Therefore, the current limiting mode has a
In this section, the current limiting process under ideal conditions function of depressing fault current to a certain extent.
was analyzed firstly, and then the current limiting process under From the above analysis, the inductance value of the CL-DCCB can
practical conditions was analyzed. be greatly increased by the current limiting control, and then the in-
The proposed CL-DCCB can greatly increase the equivalent in- ductance value of the DC system can be increased to a degree. The
ductance by changing the connection mode of the reactor branches, as increase of inductance can effectively restrain the rising-rate of the
shown in Fig. 8. system fault current, thus the time margin for fault detection and
Before the current limiting operation is executed, each inductor is breaking operation can be enlarged.
connected in parallel. The fault current of each branch can be expressed The current limiting process can be analyzed by the equivalent
in (1): circuit shown in Fig. 9.
936
S. Li et al. Electrical Power and Energy Systems 104 (2019) 933–942
MCB is 1.5 times of the rated voltage of the DC cables [20]. Considering
the safety allowance and the characteristic of MOA, the two sides of
-
IGBT branches have to tolerate 800 kV, therefore, 270 IGBTs need to be
connected in series. on the condition of bidirectional breaking opera-
tion, the number of IGBTs is 540 in ABB’s scheme. Similarly, GEIRI’s
scheme also needs 540 IGBTs.
With the issue of proposed CL-DCCB, the protection voltage of MOA
in breaking valve should also be set to 800 kV. For a diode rectifier
circuit configured on each IGBT modules, single direction of the IGBTs
can meet the requirements of bidirectional breaking operation.
Therefore, breaking valve only needs 270 IGBTs and 1080 diodes.
(a) Equivalent circuit under normal mode Considering the auxiliary capacitor is configured to achieve the
breaking effect in BCB, only a few of the IGBTs are needed in the low
loss branch. Therefore, the BCB is mainly composed of capacitor banks,
MOAs and UFDs. As the voltage across the capacitor banks or UFD are
+
+
set to 800 kV, the voltage of capacitor of BCB should also be designed to
this voltage value, considering the rated voltage of a DryDCap type
-
rising-rate of fault currents, thus more time for fault detection and
isolation is permitted.
(b) Equivalent circuit under current limiting mode
Fig. 9. Equivalent circuit of proposed CL-DCCB.
3.2. Performance analysis
When C1 and C2 are inserted in the circuit, they will be charged With the increase of the voltage and current in the transmission line
immediately, in which the voltage of C1 is determined by the voltage of of large-scale HVDC grid in future, the short circuit current and its
L1 and L2. While the voltage of C2 is determined by the voltage of L2 and rising-rate will also be increased. Although the rate can be suppressed
L3. The followings can be deduced by increasing the line reactor, the dynamic behavior of the grid will also
be affected. Because of the large reactance, the voltage source converter
di
⎧udc = uc1 + L dt3 + iR will be closed to the current source converter. Therefore, the char-
⎪ di
⎪udc = uc 2 + L dt1 + iR acteristics of the voltage source converter will be counteracted. What’s
⎪ more, at present, many fault detection methods are unable to meet the
⎪udc = uc1 + uc 2−L di2 + iR needs of DC grid protection due to the time delay. Most of them can
dt
⎨i = i1 + i2 + i3 only be used as backup protection strategies except ROCOV, etc.
⎪
⎪uc1 = L ddit1 −L ddit2 When ROCOV method is adopted, the width of the boundary be-
⎪ tween the local and remote faults is limited due to the consideration of
⎪uc 2 = L di3 −L di2 the short circuit through impedance and the bipolar short circuit. So the
⎩ dt dt (10)
ROCOV method is probably to miss what the real state is.
Considering that the voltage of the reactor will drops to zero after A 4-terminal DC grid simulation model, as shown in Fig. 10, is used
entering steady state, C1 and C2 will have a charging and discharging to test the ROCOV data. When line inductor is set to 50mH, the fault
process, and the rising-rate of charging voltage is relatively lower, thus point is at F1, which 10 km away from MMC4. The length of the line1,
the effective opening of the UFD is ensured. For the storage energy of C1 line2, line3, line4 is 227 km, 126 km, 219 km, 66 km, respectively. Other
and C2 are limited, the subsequent discharge process will not cause a parameters are shown in Tables A1 and A2, and the ROCOV data of
sharp increase of the system fault current. point A (local ROCOV) and point B (remote ROCOV) under different
types of faults are shown in Table 2:
3. Comparison of the three solutions It can be seen from Table 2 that, the ROCOV level of local grounding
fault is 290 kV, the same as the remote bipolar short circuit when the
3.1. Economic analysis grounding fault resistance is 50 Ω. At this time, it is easy to misjudge the
local fault as the remote fault. There are two solutions to solve this
This Section will make a comparison between the proposed to-
pology and the existing typical DCCB schemes including the ABB’s Table 1
Parameters of simulated DC grid.
DCCB scheme and the global energy interconnection research institute’s
(GEIRI) DCCB scheme. schemes IGBTs Diodes Capacitors UFDs
IGBT model 5SNA 2000 K450300 is adopted to analyze the schemes.
ABB 540 0 0 1
The nominal parameter of the model is 4.5 kV/2kA [19], the design
GEIRI 540 0 270 1
voltage is 3 kV considering the security margin, and the peak inter- Proposed 270 1080 540 3
rupting current is 9kA when applied to DCCB [13]. Diode model 5SDD
937
S. Li et al. Electrical Power and Energy Systems 104 (2019) 933–942
Line1 5.0
/kA
(a)
0.0
Line4
Line2
-1.0
-2.0
-3.0
B 0.998 1.000 1.002 1.004 1.006 1.008 1.010 1.012 1.014
A t/s
MMC4 Line3 MMC3
10.0
/kA
(b) Branch 2
Table 2 2.0 Branch 3
Parameters of simulated DC grid.
0.0
Fault type Local ROCOV Remote ROCOV -2.0
0.998 1.000 1.002 1.004 1.006 1.008 1.010 1.012 1.014
Ground fault 0 Ω 350 kV 240 kV
Ground fault 50 Ω 290 kV 200 kV
Ground fault 100 Ω 250 kV 165 V Fig. 11. Currents of MCB and BCB under current limiting operation.
Bipolar short circuit 525 kV 290 kV
938
S. Li et al. Electrical Power and Energy Systems 104 (2019) 933–942
1.0k 100k
0.8k UC1 80k
Voltage /kV
UC2
Energy/kJ
0.6k 60k
0.4k 40k
0.2k 20k
0.0 0.0
0.980 0.990 1.000 1.010 1.020 1.030 1.040
-0.2k t/s
0.998 0.999 1.000 1.001 1.002 1.003 1.004 1.005 1.006
t/s Fig. 15. Energy absorption value of MCB.
Current /kA
t1 t2 t3 t4 Branch 1
8.0
6.0
Current /kA
Branch 2
6.0
Branch 3 4.0 Current 1
4.0
2.0 Current 2
2.0
0.0
0.0 0.998 1.000 1.002 1.004 1.006 1.008 1.010 1.012 1.014
-2.0
0.980 0.990 1.000 1.010 1.020 1.030 1.040
Fig. 16. Comparison of current limiting effect, current 1: fault current wave-
t/s form without current limiting function, current 2: fault current waveform under
current limiting mode.
Fig. 13. Currents of MCB.
4.0k
4.0
3.0k
2.0k 2.0
1.0k 0.0
0.0 -2.0
0.980 0.990 1.000 1.010 1.020 1.030 1.040 0.990 0.995 1.000 1.005 1.010 1.015 1.020 1.025 1.030 1.035 1.040
t/s t/s
Fig. 14. Energy absorption value of BCB. Fig. 17. DC fault current waveform of ABB’s DCCB.
939
S. Li et al. Electrical Power and Energy Systems 104 (2019) 933–942
140k
120k
100k
80k
Energy/kJ
60k
40k
20k
0.0
0.990 0.995 1.000 1.005 1.010 1.015 1.020 1.025 1.030 1.035 1.040
t/s
Fig. 18. Energy absorption value of ABB’s DCCB. Fig. 21. DC fault current waveform under current limiting mode.
5. Conclusions
940
S. Li et al. Electrical Power and Energy Systems 104 (2019) 933–942
Appendix
Table A1
Parameters of the current limiting CL-DCCB.
Table A2
Parameters of simulated DC grid.
Table A3
Parameters of the Prototype.
Branch inductance/mH L 25
Branch numbers N 3
Threshold current of IGBT/kA IMAX 0.1
DC voltage/kV Udc2 0.1
DC reactor/mH Ldc 20
BCB capacitance/µF Cb 50
941
S. Li et al. Electrical Power and Energy Systems 104 (2019) 933–942
respectively. Currently he is a Ph.D. student at NCEPU. His research interests include 2017, he was respectively a visiting Ph.D. student and Post-Doctoral Fellow (PDF) at the
MMC based dc grid protection and control. University of Manitoba. He is now working on the high-speed electromagnetic transient
(EMT) modeling and control & protection of MMC-HVdc and dc grid.
Jiyuan Zhang was born in Zhejiang, China in 1995. He received his B.Sc. degree in
Hangzhou Dianzi University, Hangzhou, China in 2017. Currently he is pursuing his Chengyong Zhao (M’05-SM’15) was born in Zhejiang, China. He received the B.S., M.S.
Master degree at NCEPU, his research interests include MMC-HVDC and dc grid. and Ph.D. degrees in power system and its automation from NCEPU in 1988, 1993 and
2001, respectively. He was a visiting professor at the University of Manitoba from Jan.
Jianzhong Xu (M’14) was born in Shanxi, China. He received the B.S. and Ph.D. degrees 2013 to Apr. 2013 and Sep. 2016 to Oct. 2016. Currently, he is a professor at the School of
from North China Electric Power University (NCEPU) in 2009 and 2014 respectively. Electrical and Electronic Engineering, NCEPU. His research interests include HVdc system
Currently, he is an associate professor of the State Key Laboratory of Alternate Electrical and dc grid.
Power System with Renewable Energy Sources, NCEPU. From 2012 to 2013 and 2016 to
942