Electrical Power and Energy Systems 106 (2019) 56–67

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Adaptive differential relay coordination for PV DC microgrid using a new T

kernel based time-frequency transform
⁎
Jyotirmayee Naika, Snehamoy Dhara, P.K. Dashb,
a
EE Dept., ITER, Siksha 'O' Anusandhan University, Bhubaneswar, India
b
Multidisciplinary Research Cell, Siksha 'O' Anusandhan University, Bhubaneswar, India

A R T I C LE I N FO A B S T R A C T

Keywords: A discrete differential current based unit protection and fault distance estimation is proposed for a 100 kW
DC microgrid Photovoltaic (PV) based Multi-Terminal Direct Current (MTDC) microgrid. Relay coordination and adaptive
Differential relay coordination relay settings are presented in terms of PV arc and DC cable faults. To distinguish PV arc and DC cable faults
PV based power system without nuisance tripping event, an improved energy density based time-frequency transform is presented in this
Arc fault detection
paper. A new frequency filtering Kernel based Discrete Time Frequency Transform (KDTFT) is proposed for
Fault location
Time-frequency transform
effective fault detection. DC microgrid protection is essential area to be research focus due to lack of precise grid
Kernels codes/standards. Pole-pole (PP) and pole-ground (PG): positive/negative faults are the most counted operational
threats for DC microgrids. As PV is considered to be integrated as primary DG via DC cables, highly resistive DC
arc faults (series, parallel: cross-string, intra-string) are also focused in this paper. The independent PV arc
protection is designed as backup to arc-fault circuit interrupters (AFCIs). The efficacy of proposed protection
measure is considered in terms of percentage error for fault distance estimation and Circuit Breakers trip time
(Ts). The adequacy study is presented while considering DC load diversity.

1. Introduction low voltage DC microgrid [5–7]. The PP type faults are common pro-
tection risks, but the PG/NPG events are complicated in nature (espe-
To cope with the thriving energy requirement the present electrical cially when DC arc faults are considered) and pose erroneous detection
distribution systems are towards improvement to a bidirectional active problem. A PV based DG is more prone to DC arcing issues [8,9] due to
network solution, where the idea of microgrid comes into picture [1,2]. the PV array construction in series/parallel manner.
In present era Direct Current (DC) based micrgrids are become realistic For any PV structure different possible arc fault events (i.e. series,
due to electronics based consumers' expansion [3] as well as DC oper- parallel: cross-string, intra-string and arcing ground) are recorded [8].
ated DGs like PV, fuel cell, etc. Among various DG applications Pho- In existing literature Time-Domain Reflectometry (TDR), Earth Capa-
tovoltaic (PV) with auxiliary storage/generators are become popular for citance Measurement (ECM) etc. are few well established techniques to
microgrid solutions lately. Effective DC microgrid operation is possible locate PV arc faults [9]. In accordance with the Article 690.11: National
by incorporating optimal converter control/management and co- Electrical Code (NEC), 2011 [10], Arc-fault circuit interrupter (AFCI)
ordination with protection measures [4]. apparatus is suggested to be installed as protection measure, with grid
DC distribution networks are advantageous as compared to con- interactive PV system greater than 80 V (rated). Commonly followed
ventional AC grids in terms of enhanced power flow ability (i.e. en- protection configurations [11] are from Underwriters Laboratory (UL:
hance power flow by √2 times [5]), reduced power losses (i.e. smaller 1699B), “Photovoltaic DC Arc-Fault Circuit Protection (STP: standard
distance DC cables are free from skin effect) and fast protection technical panel)”. AFCIs are primarily implemented for series arc fault
switching ([6]). Besides being beneficial over AC mocrogrids, low protection. Parallel (i.e. cross-string and intra-string) and ground arcs
voltage Multi-Terminal DC (MTDC) distribution networks are craving are required backup/supervisory arc protection for PV based microgrid.
for fault protection provisions like well-defined protection standards There is a need to protect/monitor fault occurrence at PV side and to
[6], expensiveness of fast, power electronics DC circuit breakers (CBs) provide adaptive relay settings according to PV injection level during
[7]. Pole-pole (PP) and Pole-ground (PG → positive pole to ground and fault, to rest of the distribution network. The fault detection techniques
NPG → negative pole to ground) are the two major fault challenges for a can be segregated into four major classes based on the domains to

⁎
Corresponding author.
E-mail address: [email protected] (P.K. Dash).

https://doi.org/10.1016/j.ijepes.2018.09.043
Received 16 May 2018; Received in revised form 28 August 2018; Accepted 20 September 2018
0142-0615/ © 2018 Elsevier Ltd. All rights reserved.
J. Naik et al. Electrical Power and Energy Systems 106 (2019) 56–67

analyze the fault current characteristic: Time-domain, Frequency-do- amplitude (for DC cable faults: PP, PG/NPG) and frequency (for PV
main, Time scale and Time Frequency domain [12]. Time domain arcs) variations by a suitable TFR. Conventional TFRs (i.e. Discrete
analysis is based on inconsistency in current signature, where trip signal Wavelet Transform/DWT [16], S-transform [17], chirplet transform
to DC CBs are generated by using CuSum [6], Mathematical Mor- [24], etc.) are effective for estimation of amplitude/frequency compo-
phology Gradient [13], etc. Active impedance estimation (AIE) [14], nents of any DC current signal. But they are ineffective to identify
probe power unit (PPU) [15] are studied for DC cable fault protection. various PV arc faults (i.e. series, parallel: intra-string, cross-string) from
These methods are effective due to less computation burden, but erro- frequency components of fault current. A kernel function based DTFT is
neous during PV arc fault consideration. Time scale (wavelet transform) adopted as a solution to this complexity, where Morlet Kernel function
based DC fault protection is reported in [16]. DC cable faults depict is opted for its strong energy concentration (by higher dimensional
significant fault current amplitude but DC arc faults show noisy low feature mapping). To identify arc faults accurately and to provide
amplitude current. Due to the narrow high frequency support, less in- protection coordination adequately for DC feeders feature mapping
terpretable feature resolution it is challenging to design an effective function (to discriminate different frequency scales) of kernel algorithm
fault protection scheme out of [16] while considering different types of is incorporated as frequency rotational and shifting operators in pro-
arc faults. TF responses [17] are popular choice when accuracy is posed KDTFT. The spectral energy elaboration proposed by kernel
needed to achieve from overlapped signal response. Time-frequency mechanism helps detecting different arc fault events from minimum of
response (TFR) based protection strategy is more promising for a PV their noise signature. The fastness and accuracy is ensured for proposed
based DC microgrid, where fault detection is challenging due to over- KDTFT by less computational frequency stack formation and decision
lapping nature of fault current magnitude (time domain analysis for DC making from vicinity of frequency components.
cable faults) and wide frequency spectrum (PV arc faults). As both low/ For a time series x(t), sampled with an interval of T seconds to get N
high frequency fault currents are involved, a new frequency filtering numbers of samples (i.e. for any window instance), the KDTFT can be
Kernel based Discrete Time Frequency Transform (KDTFT) is proposed defined as:
in this paper. The frequency selection feature of proposed KDTFT is
N −1
effective to reduce computational burden (fast detection), and thus n n jPkn ⎞
KDTFT ⎛jT , ⎞= ∑ z (kT ) w ⎡(j−k ) T , ⎤ exp ⎛−
become more attractive for DC microgrid applications. The fault loca- ⎝ NT ⎠ k=0 ⎣ NT ⎦ ⎝ N ⎠ (1)
tion identification is achieved by a non iterative (fast calculation) direct
Moore Penrose pseudo-inverse/MPI [18] technique. The frequency where j = 0,1, … , N − 1 is time point indexing, n = 0,1, … , (N/2) − 1
spectral analysis of proposed KDTFT is considered to detect PV side is frequency point indexing, and k is the time domain shift for the given
arcs, due to the difficulty of their low amplitude detection thresholds. time series. A convolution is considered with discrete time series, x (kT)
Independent PV protection leads to DGs disconnection and detection and window function (i.e. modified Gaussian window [17]), w as:
threshold variation at rest of the DC network side. To cope with this
n c
issue relay coordination (between PV arc and DC cable faults) is pre- n a + b NT n ⎞⎤ c 2

sented by Piecewise Spline Lidstone Polynomial Interpolation, PSLPI w ⎛jT , ⎞= exp ⎡ (jT )2 ⎛a + b
⎝ NT ⎠ r P ⎢ ⎝ NT ⎠ ⎥ (2)
⎣ ⎦
[19], where adaptive relay setting (fault identification thresholds) for
present DC network is achieved. where r and b are named as scaling coefficients to control the number of
A differential current oriented rapid fault detection and accurate uncertainty response in a window; a and c are the positive constants
fault location identification for PV incorporated MTDC microgrid is contribution to damped hidden frequencies. P is periodicity coefficient
proposed in this paper. Independent PV arc protection (backup to AFCI) for DC signals (the value is considered near to zero as DC has no per-
and relay coordination according to it, for DC cable faults is im- iodicity). The kernel expanded discrete time series signal (z) in Eq. (1)
plemented by proposed time frequency transform (KDTFT). After em- can be defined as:
phasizing the recent trends and motivation of DC network protection
study in Section 1, the KDTFT formulation and its spectral energy cal- z (kT ) = X (kT ) ΦGR ((j, k ) T ) ΦGt0 ((j, k ) T ) (3)
culation is discussed in Section 2. In Section 3 the proposed unit pro-
where ΦGR and ΦGt0 are frequency rotational and shifting functions,
tection is presented in terms of DC cable parametric equivalence.
derived from a kernel function (KG). X (kT ) is calculated as Discrete
Independent PV arc fault protection scheme is discussed in this section.
Fourier Transform (DFT) of the actual sampled (i.e. DC currents' time-
Spectral energy (magnitude, frequency) based fault detection and
series components) signal x(nT) by Fast Fourier Transform (FFT) algo-
adaptive relay coordination concept (PSLP interpolation) between PV
rithm as:
arc and DC cables are presented in Section 4. Fast and accurate fault
distance calculation by MPI (non-iterative) approach is discussed in this N N
kn kn kn
section. A comprehensive case study is conferred in Section 5, where Xk = ∑ xn exp ⎛−iP N ⎞ = ∑ xn ⎡⎢cos ⎛P N ⎞−i sin ⎛P ⎞ ⎤
⎥
the effectiveness validation of proposed TFR is validated in MATLAB n=1 ⎝ ⎠ n=1 ⎣ ⎝ ⎠ ⎝ N ⎠⎦ (4)
and TMS320 C6713 DSP platform. Efficacy and the agility of the pro-
here i = −1 . An integrable kernel expression K G is derived by
posed method are evidenced based on two basic parameters: fault dis-
Lebesgue Measure, and is considered for rotation of ΦGR by
tance estimation (in terms of percentage error, ∊ in %) and fastness in
tan−1 [K G] ∈ L2 (R) in time-frequency plane; and frequency shift of ΦGt0 by
fault detection (in terms of CBs' trip time, Ts), respectively. K G (t0, T ) for each time instance. These rotational and shift functions
are expressed as:
2. Spectral energy calculation for fault detection
⎡ ⎛N 1 ⎞⎤
The microgrid operation exhibits nonlinear time varying (NTV) ΦGR ((j, k ) T ) = exp ⎢−i ∑ K G (XjT , XkT ) kT ⎥
⎜ j ⎟
nature as compared with conventional grids. Further PV based DG in- ⎢
⎣ ⎝ j=1 ⎠⎥⎦ (5)
tegration makes the system NTV response added with uncertainties,
nonlinear and non-stationary time frequency solution. For a DC mi- ΦGt0 ((j, k ) T ) = exp [iT (K G (X (j−t0), X (j−k )))] (6)
crogrid, various fault characteristics display noisy (various frequency
components) pattern, especially for arc faults. This scenario leads to a These kernel based higher dimensional operators are become
more complex protection challenge for DC feeders, while nuisance [ΦGR ]M × N and [ΦGt0]M × N for N number of samples for any particular
tripping taking place due to PV side arc faults (high frequency, low window instance. For each kth interval for any single window a Morlet
amplitude). Detection of these faults are effective by identifying the kernel based instantaneous patterns are obtained from X (k) as:

57
J. Naik et al. Electrical Power and Energy Systems 106 (2019) 56–67

Fig. 1. Equivalent circuit representation of DC cable: (a) Unit protection zone, (b) During PP fault, (c) During PG fault, (d) During PV arc fault.

1.75 ∥ (Xj −Xk ) ∥ ⎞ distinguishably from DC feeder faults (lower frequency components:
K G = K (Xj , Xk ) | = cos ⎛ exp ⎜⎛
j, k = 1,2.. N
few hundred Hz) a frequency correction filter is introduced. The max-
⎜ ⎟

⎝ cK ⎠ ⎝
imal amplitude signature of any specific frequency can be obtained
∥ (Xj −Xk ) ∥2 ⎞
− ⎟ from the DFT of the time-series DC current (Eq. (8)). The higher range
2cK2 ⎠ (7) threshold frequencies (by cutoff amplitude signature from Eq. (11)) are
considered for PV arc detection. Thus Eq. (11) can be reconsidered by
Addition of the nonlinear feature mapping with spectral energy
element-wise calculation as:
calculation gives effective energy concentration (ck = concentration
coefficient) by which the PV side arc fault detection is simplified by m ⩾ Fmin ⎞
avoiding overlapping with DC feeder faults (i.e. PP/PG/NPG) for pro- IFFT{Zm × n × Wm × n} if ⎛ ⎜ ⎟

[EKDTFT ]M × N = ⎝ and m ⩽ Fmax ⎠

posed KDTFT.
0 otherwise
The z (kT) containing N samples is rotated and concatenated to get a
et1, f 1 . . . etj, f 1 . . . etN , f 1
matrix Z as: ⎡ ... ... ... ... ... ⎤
z z3 .. . zN z1 =⎢ e ... e ... e
⎢ .t1,. fn. . . . .tj., fn. . . . .tN. , fn
⎥
⎡ z2 ⎤ . ⎥
⎢ et1, fM . . . etj, fM . . . etN , fM ⎥
⎢ . .3. .z.4. ..
..
. z1
. ... ...
z2 ⎥
⎣ ⎦ (12)
[Z ]M × N =⎢
z zM + 1 .. . zM − 2 zM − 1⎥
⎢ M ⎥ here m = 1, 2, … , M and n = 1, 2, … , N. The computational burden
⎣ zM + 1 zM + 2 .. . zM − 1 zM ⎦ (8)
for proposed KDTFT is reduced further by adopting this selective fre-
where M = (N/2) obtained from Nyquist sampling theorem. N is the quency elimination approach. The amplitude and frequency spectrums
number of samples and considered to be positive even integer. For real- are obtained from EKDTFT as:
time implementation it is advisable to consider less number of samples
(rather power of 2 for digital signal processors) out of total samples |EKDTFT | = (Re(EKDTFT )2 + Im(EKDTFT )2)
(Ns). To obtain localization in both time and frequency domains a two Im(E )
∠EKDTFT = tan−1 Re(EKDTFT ) (13)
KDTFT
dimensional window is derived as:
w .. . wtj, f 1 .. . wtN , f 1 For present study the DC cable faults are adequately identified by
⎡ . t.1, f. 1 .. . ... .. . ... ⎤ amplitude spectrum where classification of various low amplitude
[W ]M × N ⎢
= ⎢ wt1, fn .. . wtj, fn .. . wtN , fn ⎥ (more noisy property) PV arc is achieved by frequency spectrum.
... .. . ... .. . ... ⎥
⎢ wt1, fM .. . wtj, fM .. . wtN , fM ⎥
⎣ ⎦ (9)
3. Differential current based unit protection
where individual components is calculated as derived in Eq. (2). To
obtain fast TFR appropriate frequency scaling (i.e. dyadic scaling [17]) 3.1. DC cable equivalent representation
is selected as:

Df = [20 , 21, ..., 2l] where 2l < N Fast fault identification and effective fault location estimation are
(10)
the two desired constrains of any unit protection scheme. In this paper a
The frequency index, n (in Eq. (1)) is thus reduced from (N/2) discrete unit protection: fast fault detection by proposed KDTFT, and
tolog 2 (N /2) for proposed KDTFT. The spectral energy for proposed fault distance calculation by a non-iterative, direct solution is proposed
KDTFT is calculated by considering Inverse Fast Fourier Transform for PV based DC microgrids [6]. Here discrete differential current based
(IFFT) as: fault distance calculation is achieved by Moore–Penrose pseudo inverse
[18] technique with minimum input samples. Equivalent circuit re-
[EKDTFT ]M × N = IFFT{ZM × N × WM × N } (11)
presentation (Fig. 1a) of DC cable is achieved to formulate the unit
Higher order harmonics (usually few hundred Hz to few kHz) are protection. In this section various DC fault scenarios are expressed in
more significant to detect DC fault events (evidenced in Section 5.3). To terms of DC cable (π model) equivalent circuit [20]. For proposed unit
identify the PV side arc faults (higher frequency components: few kHz) protection both side parametric measurements are considered for a DC

58
J. Naik et al. Electrical Power and Energy Systems 106 (2019) 56–67

cable segment as: Vdc1(t) → bus entering side DC voltage, Idc1(t) → bus ∂Warc
= Pin−Pnom
entering side DC current, Vdc2(t) → bus exit/load side DC voltage and ∂t (20)
Idc2(t) → bus exit/load side DC current. In Fig. 1 F1 is fault occurrence
Now the independent PV arc detection is implemented through DC
point inside protection zone; and R1, L1, R2, L2 are the equivalent series
cable formulation as:
resistances, inductances of the DC cable and C1 and Rf are the parallel
capacitance (cable coupling location) and fault resistance, respectively. ⎛ xc ⎞ Iarc (t ) + R1 ⎛IPV (t ) + L1 (dIPV (t )/ dt ) ⎞
⎜ ⎟ ⎜ ⎟

Generally Rf is very small (0.5–3 Ω) for DC cable faults (PP and PG/ ⎝ x c + x arc ⎠ ⎝ R1 ⎠
NPG) but appeared large (25–30 Ω or more) for PV arc faults. By ap-
= VPV (t )−sign(Varc )× |Varc| (21)
plying Kirchhoff's voltage law (KVL) for Fig. 1 the following relation-
ship can be achieved: where x c / x c + x arc is arc path reactance. As the arcs are considered
at PV side the voltage and current components are represented as VPV
L 1
R1 ⎛Idc1 + 1 (dIdc1/ dt ) ⎞ = Vdc1−
⎜

R
⎟

C1
∫ ic dt (14)
and IPV respectively. VPV and IPV are PV generated voltage and current
⎝ 1 ⎠ at Maximum Power Point (MPP). To represent the above in discrete
where ic is current (instantaneous) against parallel capacitor at coupling time-frame by assuming piecewise approximation as:
point. This shunt arrangement during fault event leads to a differential
current expression (i.e. Idc3(t) = Idc1(t) − Idc2(t)) for Eq. (14) as: Iarc (jT ) × ⎛
⎜
xc ⎞ + ⎛ IPV (j )−IPV (j−1) ⎞ L1 + IPV (jT ) R1
⎟

⎝ x c + x arc ⎠ ⎝ T ⎠
⎛ Rf ⎞ L1 = VPV (jT )−|Varc (jT )| × sign[Varc (jT )] (22)
⎜
C (R + x )
⎟ ∫ Idc3 dt + R1 ⎛Idc1 +
⎜

R1
(dIdc1/ dt ) ⎞ = Vdc1
⎟

⎝ 1 f c ⎠ ⎝ ⎠ (15)
Now Eqs. (16) and (22) are discrete solutions studied for proposed
In the above equation, xc is reactance (instantaneous during fault DC protection (KDTFT), where PV arc protection (backup to AFCI) is
transient) of C1. The reactance of this has a negligible value during designed to coordinate with differential DC cable protection (adaptive
normal operation. Eq. (15) can be formulated in terms of discrete detection threshold estimation).
equation, by considering zero-order hold. The sampling frequency is
studied as 5 kHz for present focus, and the second order discrete dif- 3.3. DC microgrid design: PV based DG operation
ferential system (at kth sampling instant) is presented as:
I (j )−Idc1 (j−1) ⎞ The PV based DG integrated MTDC network is shown in Fig. 2(a).
xd
(Idc3 (j ) + Idc3 (j−1)) T + ⎛ dc1 L1 + Idc1 (j ) R1 = Vdc1 (j ) The DC network side of the network is considered for present study. The
C1 ⎝ T ⎠
DC parameter measurement points (feeders) are expanded (i.e. 1 km as
(16) in Fig. 2b) up to the load centres. These feeders are equipped with hall
where xd = the coefficient of differential current (derived from shunt sensors and DC CBs [21] at both ends (current entering and exiting
capacitance, fault resistance and instantaneous reactance); and T = the sides in a zone). The detailed parametric description of this PV-DC
sampling time interval (considered 0.2 ms) and j = 1, 2,… N. This microgrid is given in Table 1. The cable parametric description is shown
transitory reactance coefficient (xd) is critical for investigating various in Table 2. The CBs at both sides in a zone will trip dependent on
DC arc faults (Fig. 1d). Due to the resistive nature of DC cable the fault KDTFT spectrum of fault current.
distance can be estimated from resistance measurement by an accurate, For DC cables arc faults are most treacherous phenomena, because
non-repetitive MPI method, discussed in Section 4.3), after the fault of rise in temperature (i.e. fire risk) against high stored energy at arc
detection accomplished by KDTFT method. Though resistive in steady point. PV systems are potential for arc event because of its degradation
state, DC cables during abrupt fault event shows transient response, of connection/coupling points (in between PV modules, PV array-
which can be expressed in terms of: inductive series reactance (xL) and combiner box, and PV system-converter station). Increased inter-
capacitive parallel reactance (xC). This xC is important during high re- connections (e.g. torn connection, two nearby nude cables) is primary
actance path of DC arc fault. reason of series arc event (Fig. 2c). Voltage mismatch between various
cable points within a specific PV module/array (especially with poor
3.2. Arc fault design for PV system shunt insulation) is possible source of parallel arcs. Parallel arcs can
take place in between two PV string and is more risky (unlike series arc)
A model [8] for PV side DC cable arc is formulated for proposed PV due to its continuous fault path after disconnection of PV arrays from
system, as in Fig. 1d, while considering Mayr reactance (xmayr) and combiner box [8]. Only series arc protection is required according to
corona reactance (xcor). The arc reactance (xarc = xmayr + xcor) is con- 2011 National Electrical Code® (NEC). A 24-string combiner box with
sisting two shunt current diode with minimum current threshold (Is). Is an AFCI [9] is introduced by Sandia National Laboratories (SNL) to
is derived to describe the relation between the corona and arc dis- protect PV systems from series, parallel arc faults.
charge, and may vary within: −Is,nom ⩽ Is ⩽Is,nom, according to arc The backup (to AFCI, UL1699B) PV arc protection will compel the
voltage (Varc) polarity. Varc is obtained (Fig. 1d) from Ayrton model: CBs to de-energize the PV system, which will effectively reduce the PV
penetration to the grid. Due to change in PV penetration level the de-
Pnom
Varc = + Vnom tection threshold for other faults (PP and PG) will change significantly
Iarc (17)
(especially during islanded operation). Thus a relay coordination,
here, Pnom = cooling power and Vnom = arc potential constant. The adaptive threshold based DC fault detection is proposed in this paper.
voltage at F1 during arc is expressed as:
Vdc = Iarc × x arc −Varc (18) 4. Fault detection with relay coordination and distance
measurement
The arc reactance (xarc) is obtained as:
4.1. Fast KDTFT based fault classification
x arc (k ) = x cor (k ) + ∫ xarc (ρj−1) ⎛1− PPnom
⎜
in ⎞
⎟

(19)
⎝ ⎠
A significant change in measured Idc1 and Idc2 during faulty condi-
where ρ is time constant (arcing delay) and Pin is injected power to tion is important to calculate the differential current (Idc3 = Idc1 − Idc2).
arc path. The arc path stored energy (Warc) during fault can be ex- To enable the proposed KDTFT for DC fault detection two types of
pressed as: sampling can be achieved: (a) sample by sample approach: here

59
J. Naik et al. Electrical Power and Energy Systems 106 (2019) 56–67

Fig. 2. (a) Multiple DGs based DC microgrid integrated with utility (AC) bus, (b) Detailed unit protection zones representation of proposed multi-PV DC microgrid
and (c) Various types of arc fault configuration for PV system.

Table 1
Considered network parameters with Generation unit.
PV based DC bus parameters
PV system power rated (kW) 4 × 100 kW (individual) = 400 kW
DC converter (buck type) rated 4 × 108 kW (+8% IEC 6210); 470 V (DC)
DC Bus rated 500 V; Rdc → 121 mΩ/km; Ldc → 0.97 mH/km; Cdc 12.1 → nF/km; Rground → 0.5 Ω

VSC parameters
Voltage Converter parameters Operational frequency → 60 Hz; 400 kW, 260 V (AC) to step-up. DC link Capacitor → 100 μF
VSC PLL control (2nd order) Switching frequency → 5 kHz; proportional (Kp) gains 7, 9.8; integral (Ki) gains 32, 20
Line parameters Line voltage(L-L) → 25 kV (AC) after step-up; R1, X1 (2.5 km) → 0.074 Ω, 2.61 mH; R2, X2 (3.2 km) → 0.0947 Ω, 3.34 mH; R3, X3
(2 km) → 0.0592 Ω, 2.08 mH

Load parameters
DC Lamp load PL,dc → 50 kW; 450 V
DC motor Load PL2,dc → 20 kW, 450 V
PV (DER1) local load (load 1) PL1 → 100 kW; QL1 → 5 kVAR
Grid side non-linear load (load 2) PL2,0 → 275 kW; QL2,0 → 25 kVAR
Diesel generator (DG2)
Diesel generator (DG2) parameters Vdg2,rated → 480 V; DG rated revolution 750 rpm
DG2 Governor control gains K1,dg2 → 1.85; K1,dg2 → 2.6; K1,dg2 → 0.85

60
J. Naik et al. Electrical Power and Energy Systems 106 (2019) 56–67

Table 2 samples is considered for single window, and two successive window's
DC cable parameters for proposed utility interactive MTDC network. kth sample values (e.g. 1st value) can be compared to enable (NTfault is a
Component Data (km) Parameters value near to zero) the KDTFTflag during sudden changes, Fig. 3(a).
Once KDTFTflag gets enabled the time-frequency scheme will start cal-
Resistive Inductive Capacitive culating spectral energy by collecting N number samples (for KDTFT
(Ω/km) (mH/km) (µF/km) window as discussed in Section 2) from buffer (delay incorporated).
Rectifier cable 1.02 0.125 0.20 0.5 The amplitude spectrum (|EKDTFT |) from KDTFT is investigated to be
(diesel within detection threshold limits (maximum and minimum thresholds:
generator) Thmax and Thmin respectively) in absent of PV arc fault flag (ARCflag = 0)
PV cables 1 0.641 0.48 0.32 and the CBs at both side of protection zone (CB1 and CB2) will get
DC load cables 1 0.407 0.34 0.11
disconnected once fault is detected. The fault flag will get enabled
(DC_Fault_flag = 1) to convey acknowledgment to fault distance calcu-
differential current (Idc3) samples at kth instant are compared with past lation algorithm (Section 4.3). The PV penetration level is directly de-
k − 1th instant value for enabling the protection flag (KDTFTflag) pendent on PV internal arc faults. Thus proposed backup (to AFCI) PV
during sudden changes; (b) window by window approach: a n chunk of arc protection enables adaptive threshold flag (ARCflag) as shown in

Fig. 3. (a) Proposed fault detection and relay coordination by KDTFT, (b) Actual fault duration and fault duration to calculate distance for PG fault; Adaptive
threshold calculation for utility interactive DC microgrid protection: (c) threshold estimation during PP faults, (d) threshold variation for simultaneous PP and PV
series arc faults.

61
J. Naik et al. Electrical Power and Energy Systems 106 (2019) 56–67

Fig. 4. The effective fault protection during DC feeder faults (i.e. PP, PG and NPG) to the DC motor/lightning load: (a) trip time (Ts) of DC switches with fault current
(Idc3), (b) KDTFT amplitude spectral response for differential current (Idc3) and average current (Idc,avg) for internal zone fault, (c) actual Idc3 and Idc,avg for
internal zone fault and (d) |KDTFT| of Idc3 < |KDTFT|of Idc,avg for external fault.

Fig. 3(a). A detailed DG penetration level study is considered for pro- code for plug and play incorporation to avoid false tripping due to lack
posed PSLP interpolation based adaptive threshold estimation. of synchronisation. Suitably for a limit of 64 × 103 bits/s data transi-
A selection ability between internal (fault point F1 and F2 in tion rate can be able to adopt in accordance with IEC-61850, via at least
Fig. 2b) and external faults (Fext) is presented for proposed unit pro- of 16 bit channel. The DC switching delay is considered for present trip-
tection. The amplitude spectral response (|EKDTFT |) from KDTFT for time where communication delay is neglected.
differential current (Idc3) will be greater than the average fault current
(Idc,avg) for internal fault protection zone. For external fault zone that
4.2. Adaptive detection threshold calculation
formulation will be inversed. The Idc,avg is calculated in terms of Idc1 and
Idc2 as:
Arc fault at PV side is primary reason to de-energize the DGs from
(I (jT ) + Idc 2 (jT )) grid (penetration level variation). The steady state constant detection
Idc, avg (jT ) = dc1 threshold settings will vary due to this. PV penetration (rated) is as-
2 (23)
sumed for standard test condition, STC (1000 W/m2, 25 °C). As Idc3 level
The fault classification for PP and PG types are based on polarity of varies due to change in DG penetration, the KDTFT amplitude spectrum
Idc3 (discussed in Section 5, Figs. 4 and 5). Ts is calculated as: threshold (Thmax and Thmin) will vary accordingly. Relay coordination
Ts = Cd + DC swiching delay [22]. Ts compared with actual fault between independent PV arc protection and DC cable faults is proposed
duration for PG fault is shown in Fig. 3(b). The requirement for an in this paper (Fig. 3a). Various penetration levels for considered multi-
adequate communication channel is opted via IEC-61850, open grid DG (PV systems) based DC grid is studied (Table 3) for both PP and PG

62
J. Naik et al. Electrical Power and Energy Systems 106 (2019) 56–67

Fig. 5. Various PV arc faults and proposed backup protection performance: (a) PV parallel/intra-string fault characteristic, (b) Fault detection and trip time (Ts) for
PV parallel/intra-string, (c) Fault detection and trip time (Ts) for PV parallel/cross-string, (d) Fault detection and trip time (Ts) for PV series arc fault.

Table 3 variation in PV penetration levels, a Piecewise Spline Lidstone

Different PV penetration due to arc faults applied for DC power variation. Polynomial Interpolation (PSLPI) approach is selected. PV level varia-
Sl no. PV System 1 PV System 2 PV System 3 PV System 4 tion (due to arc at PV side) reflects through |EKDTFT | for four shunt
connected PV systems (Fig. 2a). To estimate the maximum and
Data 1 ON ON ON ON minimum relay setting values the PCHPI scheme is incorporated with 2
Data 2 ON ON ON OFF
nos. of multi-order (13th) polynomials. The relationship is obtained
Data 3 OFF ON ON ON
Data 4 ON OFF ON ON
from |EKDTFT | detection threshold as functions of DC power (Pdc) as
Data 5 ON ON OFF ON shown in Table 3.
Data 6 ON ON OFF OFF For a cubic Lidstone interpolation every element is expanded as 3rd
Data 7 OFF ON ON OFF degree polynomial in Peano’s form, where derived spline is continuous
Data 8 OFF OFF ON ON
with first derivative and be able to be expressed for given interval (Sk,
Data 9 ON OFF OFF ON
Data 10 ON OFF OFF OFF Sk+1) as in Eq. (24):
Data 11 OFF ON OFF OFF
Data 12 OFF OFF ON OFF f (Sk ) = L00 (m) Zsk + L10 (m)(Sk + 1−Sk ) Tk + L01 (m) Zsk + 1
Data 13 OFF OFF OFF ON
+ L11 (m)(Sk + 1−Sk ) Tk + 1 (24)

where m = S−Sk / Sk + 1−Sk ; and S = Lidstone basis functions derived

faults. The DG penetration data are collected in terms of proposed TFR from Cauchy-Schwarz inequility; Tk = starting tangent and
response, as relay settings. The deviation in threshold level for different Tk+1 = ending tangent.
PV penetration is shown in Fig. 3(c) for PP faults. The threshold var- The polynomial coefficients are calculated as concurrent results
iation during simultaneous PP fault and series arc is shown in Fig. 3(d). from the system of plynomial relation. The estimated relay settings
In this figure the first PP fault (duration: t = 1–1.5 s) got identified by from existing variation band are shown in Fig. 3(c) and (d). The pro-
maximum threshold level, estimated when all 4 PV systems are ON posed adaptive relay settings, is estimated accurately by the PHCPI
(Table 3). Further PV series arc event is subjected at t = 2 s for PV algorithm. After ability to change the relay settings (i.e. maximum and
system 1, and got detected at t = 2.61 s by the proposed backup arc minimum detection thresholds) according to PV side availability (in-
fault protection. Thus PV penetration changes and new detection dependent PV arc protection) the fault distance estimation is proposed
threshold is estimated as described in Fig. 3(a) and (c). To validate the for this discrete DC unit protection.
new threshold (in terms of fault current) again same (i.e. feeder dis-
tance → 0.5 km, Rf → 0.55 Ω) PP fault is considered at t = 4–4.5 s. From
Fig. 3(d) it is clearly evidenced that the fault detection threshold is 4.3. Non-iterative fault distance calculation
changed when DGs penetration changes and thus adaptive coordination
of CBs are important to establish effective protection measures. From Fig. 2(b), the discrete differential relation obtained from both
To calculate detection threshold in a real-time manner during side current measurements of the fault point (F1, inside protection
zone) is considered to calculate fault distance via non-iterative MPI

63
J. Naik et al. Electrical Power and Energy Systems 106 (2019) 56–67

technique. The fault location in the proposed DC microgrid is obtained Table 4

from the discrete differential equation given in Eq. (16). Electrical characteristics data of WS-140 (WSBYW03101693) solar panel at
This formulation can be used for fault location estimation while Standard test condition (STC).
getting positive signature from decision flag (DC_Faultflag =1, Fig. 3a). Parameters Values
The second order system from differential current, in Eq. (16) can be
rewritten in matrix form with N numbers of measured sample (Vdc1(kT), Maximum power (Pmax) 140 W
Voltage at Pmax (Vmpp) 17 V
Vdc2(kT), Idc1(kT) and Idc2(kT)):
Current at Pmax (Impp) 8.52 amps
Open circuit voltage (Voc) 21 V
⎡ Idc3 (j−1) + Idc1 (j ) Idc1 (j )−Idc1 (j−1) Idc1 (j ) ⎤ x
⎡C T ⎤ Short circuit current (Isc) 8.89 amps
⎢ Idc3 (j ) + Idc1 (j + 1) Idc1 (j + 1)−Idc1 (j ) Idc1 (j + 1) ⎥ ⎢ L1 ⎥ Temperature coefficient of Isc (KI) (0.065 ± 0.015)%/°C
⎢
⋮ ⋮ ⋮ ⎥ ⎢ 1 ⎥ Temperature coefficient of Voc (KV) −(80 ± 10)mV/°C
⎢ ⎥⎢ T ⎥
RT NOCT (4 7± 2) °C
⎣ Idc3 (N −1) + Idc1 (N ) Idc1 (N )−Idc1 (N −1) Idc1 (N ) ⎥
⎢ ⎦⎣ 1 ⎦
⎡ Vdc1 ((j−1) T ) ⎤
⎢ V ((j ) T ) ⎥ Table 5
= ⎢ dc1 ⎥
⎢ ⋮ ⎥ Percentage error in fault location calculation for pole to pole fault.
⎢ V
⎣ dc1 ((N − 1) T ) ⎥
⎦ (25) Fault distance (kms) Fault resistance (Ω)

For arc fault at PV side, Eq. (22) can be obtained equivalently with n
0.5 1 1.5 2
number of sample values: ∊ (%) ∊ (%) ∊ (%) ∊ (%)

⎡ Iarc (j ) Idc1 (j )−Idc1 (j−1) Idc1 (j ) ⎤ x T

2 ⎤
0.50 0.302 1.081 3.206 5.880
⎢ Iarc (j + 1) Idc1 (j + 1)−Idc1 (j ) Idc1 (j + 1) ⎥ ⎡⎢ L1 ⎥ 0.55 0.362 1.721 3.387 4.388
⎢ ⎥ T ⎥ 0.60 0.211 0.615 1.646 3.238
⎢ ⋮ ⋮ ⋮ ⎥⎢⎢ R1 T ⎥ 0.65 0.197 0.735 1.640 2.825
⎢
⎣ Iarc (N ) Idc 1 (N )−Idc 1 (N −1) Idc 1 (N ) ⎥
⎦ ⎣ ⎦
0.70 0.310 0.601 1.486 2.706
−Varc (jT ) × sign (Varc (jT )) 0.75 0.091 0.864 1.363 2.363
⎡ ⎤ 0.80 0.090 0.732 1.360 2.360
⎢Vdc1 ((j + 1) T )−Varc ((j + 1) T ) × sign (Varc ((j + 1) T )) ⎥
=⎢ ⎥ 0.85 0.086 0.642 1.286 2.219
⎢ ⋮ ⎥ 0.90 0.076 0.555 1.238 2.140
⎢
⎣ Vdc1 (NT )− V arc (NT ) × sign (Varc (NT )) ⎥
⎦ (26) 0.95 0.064 0.418 1.224 1.923
1.00 0.262 0.360 1.182 1.826
where x2 = ( xc
x c + x arc ) and I DC Є Rn×3 = matrix with measured current
parameters, VDC Є R = corresponding DC voltage vector, and ZDC Є
n×1
Table 6
R3×1 = unkown DC cable parameters to be obtained. From these ex-
Percentage error in fault location calculation for pole to ground fault.
pressions the unknown parameters (R1 and L1) can be estimated by MPI
technique. Thus Eqs. (25) & (26) can be written as: Fault distance (kms) Fault resistance (Ω)

[IDC ]N × 3 × [ZDC ]3 × 1 = [VDC ]N × 1 (27) 0.5 1 1.5 2.0

∊ (%) ∊ (%) ∊ (%) ∊ (%)
To obtain unknown values of ZDC from Eq. (27), the pseudo inverse
[IDC†] is incorporated for direct solution (instead of sample by sample). 0.50 0.476 1.181 3.361 5.242
0.65 0.411 0.835 2.424 5.380
The total data samples are considered to be Ns, where N is the number
0.70 0.358 0.748 2.210 4.862
of samples used for this chunk by chunk based incremental learning. To 0.85 0.331 0.776 1.894 4.321
calculate fault location effectively adaptive updation of ZDC is adopted 0.90 0.312 0.575 1.468 3.452
with minimum learning cycles (2–6 cycles). The calculation of pseudo 1.00 0.284 0.402 1.310 2.424
inverse term, [IDC†] from Moore Penrose approach the relationship is
written as:
prominent for distance calculation error (∊ in %), which get escalated
+ T T
IDC (n) = (ηI + IDC (N ) × IDC (N )) × IDC (N ) × VDC (N ) (28) for proposed protection calculation during noisy channel condition. The
way out to it is more number of total samples (N), which may lead to a
where IT = conjugate transpose of IDC, I = identity matrix, &
higher Ts time. This issue will be covered in future scope of this paper.
η = coefficient with minimum value. The R1 is calculated very accu-
rately during fault, by proposed method. The calculation error (∊ in %)
might be high for introductory chunks depending on noise, and non-
5. Result analysis
linear pattern Idc3. For this chunk by chunk incremental learning, M can
be expressed for kth chunk as:
In this section the efficacy of proposed time-frequency based DC
T T protection is evidenced for considered multi-PV-DC microgrid, simu-
M (k ) = M (k−1)−M (k−1) IDC (k ) [I + IDC (k ) M (k−1) IDC (k )]−1IDC (k ) M (k
lated in MATLAB platform. Different DC loading (DC lightning/motor
−1) (30)
as in Table 1) condition based case studies are presented accordingly.
The unknown DC cable values are updated flexibly as: An encouraging fastness in fault identification and accuracy in fault
T location estimation is presented in terms of Ts and ∊ (%) respectively,
ZDC (k ) = ZDC (k−1) + M (k ) IDC (k ) [VDC (k )−(IDC (k ) ZDC (k−1))] (31)
throughout these case studies. The proposed DC protection is ad-
Fault location can be identified from measured R1 and Rcab/ vantageous over [24], as reduction of equipment (PPU units) and lo-
km = DC cable unit resistance (Table 2). As discussed in Section 3.1 DC cation accuracy. The advanced relay coordination during PV arc backup
cable fault path, involved with C1 is majorly contributing transient protection makes this proposed study more dispatchable for PV based
snubber reactance. DC networks.
This reactance is less effective during PP faults, but limiting the fault
calculation accuracy majorly during PG faults. This is evidenced clearly
in Section 5 (Tables 5 & 6). For DC cable faults the effect of noise is

64
J. Naik et al. Electrical Power and Energy Systems 106 (2019) 56–67

Fig. 6. Frequency Contours from proposed KDTFT: Fault Current and Frequency of DC cable faults: (a) PP fault, (b) PG fault, (c) PV series arc and (d) PV parallel arc
(inter string).

5.1. Case 1: DC feeder faults Table 1). The distance variation for PG and NPG fault location (F1) and
the considered fault resistance limits are similar to previous (case 1). As
5.1.1. A. PP fault for DC motor loading in Fig. 4(a) Ts is recorded: 110–145 ms for PG faults, with fault loca-
PV based water pumping application for rural/suburban population tion/resistance variations. The proposed KDTFT based time-frequency
is very popular and selection inspiration for present loading (DC amplitude calculation and its effective detection threshold crossing is
Motor). The rated parameters for this loading are mentioned in Table 1. evidenced in Fig. 4(b). The zone wise unit protection is further verified
The DC cable span is considered to be 1 km from common DC bus to this in Fig. 4(c) and (d), where for internal zone faults Idc3 > Idc,avg
load. Fault location F2 is differed from 0.5 kms to 1 km, to validate the (index wise) and for external zone Idc3 < Idc,avg.
potency of proposed protection. The Rf is considered for the range of
0.5–2 Ω for this study. At time instance t = 1–1.5 s the PP fault has
5.2. Case 2. PV side DC arc event
occurred. The DC cable fault response and proposed KDTFT is shown in
Fig. 4. As shown in Fig. 4(a) Ts are recorded: 105–125 ms (ms), for PP
Because of the small amplitude (detection threshold) the DC arc
faults in accordance with fault distance and resistance variation. As PV
faults are challenging to be detected. The arc fault DC equivalence is
based DC arcs are needed to be distinguished from DC cable faults,
discussed in Section 3.2. Both series and parallel (cross-string and intra-
KDTFT amplitude spectrum (|KDTFT|) is used for PP and PG fault de-
string) are studied in this case. The arc fault detector is implemented as
tection, Fig. 4(b). The zone wise unit protection is validated as in
backup to primary AFCI (UL 1699B) and a relay coordinator for pro-
Fig. 4(d) where Idc3 is evidenced less than Idc,avg for external fault zone
posed DC unit protection. As in Table 1, the 400 kW PV generation is
(at Fext).
comprised of individual shunt connected 100 kW PV system (Fig. 2a).
The elemental module is considered as WS-140/WSBYW03101693
5.1.2. PG/NPG faults for DC light loading which is rated as 140 W, 17 V, as in Table 4.
Because of low power utilization, prolong life etc. light-emitting The 100 kW PV system is designed from this module by connecting
diode (LED) based lightning is quite popular in present era. Hence LED 24 parallel PV strings (each string contains 29 PV modules). Now
based DC lightning load is considered for this case (details are in among these module connection series and parallel itra-string arcs are

65
J. Naik et al. Electrical Power and Energy Systems 106 (2019) 56–67

created within a string. The fault characteristic and Ts by proposed output data acquisition for series and parallel arc fault models. This
KDFTF is shown in Fig. 5. emulator is interfaced with Simulink based DG dynamics via IEEE 1284
The arc fault (low amplitude, more noisy to create heat) detection is to USB cable. A delay buffer is required to make the data transition after
based on frequency sprectrum of KDTFT (∠EKDTFT ) as discussed earlier, completion of every iteration/generation. This is achieved by
in Section 3. The reason is clearly depicted in Fig. 6, where DC cable DSK6713_waitusec (td), where td is the delay time. The KDTFT fault
fault frequency responses are showing low frequency variation during detection is recorded on Digital Storage Oscilloscope/DSO (Tektronix
PP faults (100–340 Hz, Fig. 6a) and PG faults (110–415 Hz, Fig. 6b). But TBS1022) through a Toshiba TLP 250 gate driver. The efficacy is evi-
during PV arc faults the response is high enough to distinguish denced in Fig. 7 (for PV parallel cross-string). Ts is recorded as
(760 Hz − 4 kHz, Fig. 6c and d). The fault location estimation errors are 180–200 ms (Fig. 7a) which is almost similar ( ± 15%) with simulation
shown in Tables 5 and 6. studies.

5.3. Case 3. Validation of proposed KDTFT in TMS320 C6713 DSP 6. Conclusion

platform
A new discrete differential current based unit protection and loca-
A real time validation is achieved for the proposed KDTFT based CBs tion estimation for PV incorporated DC distribution network is pre-
trpping, through efficient Digital Signal Processor (DSP), TMS320 [23]. sented in this paper. PV systems are potential for DC arc faults, and thus
TMS 320C6713 is a 32 bit floating point digital signal processor, ef- a PV system arc (backup to AFCI, UL 1699B) protection is designed for
fective for complex computational algorithms (clock frequency up to relay coordination (according to PV penetration) for proposed DC mi-
225 MHz). The DSP kit is interfaced with MATLAB Simulink with the crogrid. As DC cable faults exhibit nonlinear time varying (NTV) nature,
help of Embedded USB JTAG controller and plug and play drivers (USB a new frequency filtering based KDTFT is proposed for fault detection.
interfaced). The software interface is designed with the help of Code DC cable faults are distinguishable from arc faults in terms of amplitude
Composer Studio (i.e. CCStudio v3.1) from Texas Instruments. The test and noise signature in measured current parameters. Thus to increase
bench is incorporated with three major parts: PV Emulator (Ecosense the energy density for both amplitude and frequency spectrum of pro-
mini-Solar PV Plant Prototype), MATLAB Simulink based DC cable fault posed TFR, a Morlet Kernel function based feature mapped rotational
simulation, TMS320 DSP Starter Kit (DSK) for validation of proposed and shift function is introduced here. The discrete KDTFT is less com-
KDTFT scheme (shown in Fig. 7c). The PV Emulator is used for PV putational design achieved by band pass filtering based spectral energy

Fig. 7. Validation of Proposed KDTFT in TMS320 C6713 DSP platform:(a) Fault detection and trip time (Ts) for PV parallel/cross-string, (b) Actual fault current
characteristic and (c) Test bench validation setup for proposed KDTFT based DC protection.

66
J. Naik et al. Electrical Power and Energy Systems 106 (2019) 56–67

calculation. Further a fast, non-iterative distance calculation by tests. Technical Report, SAND2013-5916, Sandia National Laboratories,
Moore–Penrose pseudo inverse technique is presented in this paper. The Albuquerque, NM, USA; 2013.
[10] Brooks Bill RJ, White Sean PR. Article 690 photovoltaic systems part II circuit re-
comprehensive DC fault protection is studied through rigorous case quirements. Photovoltaic systems and the National Electric Code. ROUTLEDGE in
study in MATLAB and TMS320C6713 platform, to validate effective- association with GSE Research; 2018. p. 26–57.
ness. The trip time (Ts) calculated in terms of fast acting DC CBs' delay, [11] Zgonena T, Ji L, Dini D. Photovoltaic DC arc-fault circuit protection and UL Subject
1699B. Golden, CO: Photovoltaic module reliability workshop; 2011.
algorithm response time and communication delay; distance calculation [12] Ghaderi A, Mohammadpour HA, Ginn HL, et al. High-impedance fault detection in
error (∊ in %) are considered as two major performance index for the distribution network using the time-frequency-based algorithm. IEEE Trans
present study. The proposed technique is effective during all DC faults Power Deliv 2015;30(3):1260–8.
[13] Sarlak M, Shahrtash SM. High impedance fault detection using combination of
with fast (less than 150 ms) detection, and distance calculation error multi-layer perceptron neural networks based on multi-resolution morphological
(almost less than 5%). gradient features of current waveform. IET Gener Transm Distrib
2011;5(5):588–95.
[14] Christopher E, Sumner M, Thomas DW, et al. Fault location in a zonal DC marine
Appendix A. Supplementary material
power system using active impedance estimation. IEEE Trans Ind Appl
2013;49(2):860–5.
Supplementary data to this article can be found online at https:// [15] Mohanty R, Balaji UM, Pradhan AK. An accurate non-iterative fault location tech-
doi.org/10.1016/j.ijepes.2018.09.043. nique for low voltage DC microgrid. IEEE Trans Power Deliv 2016;31(2):475–81.
[16] De Kerf K, Srivastava K, Reza M, et al. Wavelet-based protection strategy for DC
faults in multi-terminal VSC HVDC systems. IET Gener Transm Distrib
References 2011;5(4):496–503.
[17] Tripathy LN, Samantaray SR, Dash PK. A fast time–frequency transform based
differential relaying scheme for UPFC based double-circuit transmission line. Int J
[1] Lasseter RH. Microgrids. IEEE power engineering society winter meeting. 2002. p. Electr Power Energy Syst 2016;77:404–17.
305–8. [18] Barata JCA, Hussein MS. The Moore-Penrose pseudoinverse: a tutorial review of the
[2] Ackermann T, Andersson G, Söder L. Distributed generation: a definition. Electr theory. Braz J Phys 2012;42(1–2):146–65.
Power Syst Res 2001;57(3):195–204. [19] Gilman A, Bailey DG, Marsland SR. Interpolation models for image super-resolu-
[3] Salomonsson D, Soder L, Sannino A. Protection of low-voltage DC microgrids. IEEE tion. In: Electronic design, test and applications, DELTA 2008, 4th IEEE interna-
Trans Power Deliv 2009;24(3):1045–53. tional symposium; 2008. p. 55–60.
[4] Kaipia T, Salonen P Lassila J. et al. Application of low voltage DC-distribution [20] Karlsson P, Svensson J. DC bus voltage control for a distributed power. IEEE Trans
system—a techno-economical study. In: Proceedings of CIRED, p. 21–4. Power Electron 2003;18(6):1405–12.
[5] Park J-D, Candelaria J, Ma L, et al. DC ring-bus microgrid fault protection and [21] Rogne T. Wartsila Norway As. Electronic DC circuit breaker. U.S. Patent 8,716,905;
identification of fault location. IEEE Trans Power Deliv 2013;28(4):2574–84. 2014.
[6] Dhar Snehamoy, Patnaik RK, Dash PK. Fault detection and location of photovoltaic [22] Technical catalogue 2010 S800/S500: the high performance MCB, ABB Group.
based DC microgrid using differential protection strategy. IEEE Transactions on [23] Chassaing R. Digital signal processing and applications with the C6713 and C6416
Smart Grid. Early Access; 2017. DSK. New York: Wiley; 2004.
[7] Tang L, Ooi B-T. Locating and isolating DC faults in multi-terminal DC systems. IEEE [24] Angrisani L, D'Arco M. A measurement method based on an improved version of the
Trans Power Deliv 2007;22(3):1877–84. chirplet transform for instantaneous frequency estimation. In: Instrumentation and
[8] Flicker J, Johnson J. Electrical simulations of series and parallel PV arc- faults. 2013 measurement technology conference, 2001. IMTC 2001. Proceedings of the 18th
IEEE 39th photovoltaic specialists conf (PVSC). 2013. p. 3165–72. IEEE; 2001, vol. 2, p. 1123–9. IEEE.
[9] Johnson J, Gudgel B, Meares A, et al. Series and parallel arc-fault circuit interrupter

67