Measurement 164 (2020) 107915

Contents lists available at ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

Dynamic test signal modelling and a compressed sensing based test for
electric energy meter errors
Xuewei Wang a, Di Wu a, Ruiming Yuan b, Zhenyu Jiang b, Jing Yang a
a
Beijing University of Chemical Technology, Beijing, China
b
State Grid Jibei Electric Power Company Limited Power Research Institute, Beijing, China

a r t i c l e i n f o a b s t r a c t

Article history: The problem of dynamic error evaluation for smart energy meter under the condition of large power
Received 5 June 2019 dynamic load is investigated. A general model of dynamic test signal is established and two forms of
Received in revised form 10 April 2020 the test signal are given. Then two models of the test signal are decomposed to obtain the prior informa-
Accepted 28 April 2020
tion of the dynamic intrinsic signal, and the Compressed Sensing (CS) measurement matrix is con-
Available online 24 May 2020
structed. The CS indirect measurement model of the dynamic test signal is deduced to accurately
measure the value of the dynamic electrical energy. Based on the CS indirect measurement model, a
Keywords:
new indirect measuring method for evaluating dynamic error of smart energy meter is proposed, and
Dynamic testing signals
Compressed sensing
meanwhile a testing system of smart energy meter for evaluating dynamic error is given. Finally, exper-
Indirect measurement method iment results and uncertainty analysis show that the CS indirect measurement method can be used to
Smart energy meter test the dynamic error of the smart energy meter under different signal conditions and the measurement
Testing system for evaluating dynamic uncertainty (k = 2) is better than 0.05%.
errors Ó 2020 Published by Elsevier Ltd.

1. Introduction ation of load current or power amplitude [11]. However, these

models are deterministic and cannot represent the characteristics
With the more prevalent use of renewable energy sources and of transient-time mode, short-time mode and rapidly random fluc-
the Large Power Dynamic (LPD) loads in smart grid, such as photo- tuation of the dynamic current and power. On the other hand, for
voltaic power supply, and electric arc furnaces, high-speed railway the method of evaluating dynamic error of smart energy meters,
train, etc., smart grid has shown two important dynamic character- there have been proposed two methods including direct measuring
istics from traditional power grid. The two dynamic characteristics method with uncertainty 3.0% [11] and On-Off keying indirect
include rapidly random fluctuation of dynamic load current [1–3], comparison measuring method with the uncertainty 0.38% [10],
and large range fluctuation of the output power of the renewable but no uncertainty of dynamic electric energy measuring is given
energy electric source [4]. In such instances, the dynamic charac- in [5,6]. Unfortunately, these measuring methods are only suitable
teristics lead to the enormously increase in error of smart energy for deterministic DTS rather than random DTS. Until now, there
meter [5,6]. Because of the above practical problem, the investiga- have not been a method that can accurately measure random
tion of Dynamic Test Signal (DTS) model with the above dynamic DTS reference energy in error evaluation of smart energy meter.
characteristics and the measuring method of DTS electrical energy With the development of signal processing theory in recent
as a reference value become two important challenges for evaluat- years, the problem of accurately measuring random DTS electrical
ing dynamic error of smart energy meter, and the challenges have energy can be resolved by using different ideas. Compressing Sens-
gained concerns from many researchers [7–11]. ing (CS) is a novel signal processing method [12,13], which is used
During the past few years, several DTS models have been sug- in reconstructing signal from less samples through designing mea-
gested, such as test signal model with subharmonic component surement matrices. At present, CS signal processing theory has
[7], test signal model with sinusoidal or trapezoidal wave envelope been investigated in many fields, such as image compressing
[8], amplitude-frequency modulated DTS model [9], and On-Off [14], channel estimation [15], power quality analysis [16], etc.
Keying (OOK) DTS model [10]. The proposed DTS models focused However, signal reconstruction is not necessary in many signal
on reflecting the characteristics of long-time mode periodic fluctu- processing situations, such as signal detection. In order to avoid
highly complex signal reconstruction, Compressive Measurement
(CM) theory has been proposed after CS theory. And the proposed
E-mail address: [email protected] (X. Wang)

https://doi.org/10.1016/j.measurement.2020.107915
0263-2241/Ó 2020 Published by Elsevier Ltd.
2 X. Wang et al. / Measurement 164 (2020) 107915

CM methods are used to measure random communication binary where U k is the amplitude of the voltage and Ik is that of the current,
signals [17] and identify feature of machine faults [18]. These f 1 is the fundamental frequency, X1 is the angular frequency and
applications reflect that the CM methods are now only suitable X1 ¼ 2pf 1 , T is the period and T ¼ 1=f 1 , /k and uk are the initial
for distinguishing the state of random signal, e.g. if the received phase of voltage and current respectively, and k = a, b, c represents
signal exists or not [19,20]. Unfortunately, up to now neither CS the a-phase, b-phase, and c-phase, respectively.
nor CM method gives accurate measuring value of the DTS electri- In order to represent the fluctuation characteristics of dynamic
s
cal energy. load voltage and current, steady-state signals usk ðt Þ and ik ðtÞ are
In this paper, we are motivated to solve the above two impor- modulated, and then voltageudk ðt Þ and current ik ðtÞ signals of LPD
d

tant challenges problem. The main contribution of this paper is

load are expressed as
to establish the general models of DTS that can reflect the charac-
teristics of the LPD load and propose a CS indirect measurement udk ðt Þ ¼ v 1 ðt Þ  usk ðtÞ ¼ v 1 ðt Þ  U k sinðX1 t þ /k Þ ð3Þ
method and a system that can be used to testing smart energy
meter for accurately evaluating dynamic errors. The paper is orga-
ik ðt Þ ¼ v 2 ðt Þ  ik ðt Þ ¼ v 2 ðt Þ  Ik sinðX1 t þ uk Þ
d s
ð4Þ
nized as below. Section 2 establishes m-sequence pseudorandom
test signal model and OOK Deterministic test signal model, and where v 1 ðtÞ and v 2 ðt Þ are modulating signal and used to reflect the
then gives a ON-OFF keying generation scheme. Section 3 deduce amplitude fluctuation characteristics of the LPD load voltage udk ðt Þ
a CS indirect measuring model of dynamic reference electrical d
and current ik ðtÞrespectively. As above Deterministic characteris-
energy of DTS for calculating reference electrical energy. Section 4
tics (a) (b) and (c) stated, the amplitude of load voltage has stable
shows experiment results and analysis. Section 4 gives instrumen-
characteristics, and the amplitude of load current is short-time sta-
tal measurement uncertainty analysist. And the end, the conclu-
bility within each period and cyclically fluctuates with large range,
sion is given.
so we define
X
1 X
1
2. The general model of dynamic test signal v 1 ðtÞ ¼ bn g ðt  nT Þ ; v 2 ðtÞ ¼ an g ðt  nT Þ
n¼0 n¼0

2.1. The general mathematical model of DTS where bn is a constant within each period T and a deterministic data
sequence during different periods, so let bn ¼ 1. an is a random or
In order to investigate the dynamic error test method of the deterministic data sequence during consecutive fundamental peri-
electric energy meter (this paper refers specifically to smart energy ods, but a constant within each period. g(t) is a rectangular window
meter), it is necessary to give two definitions and to establish a d
function. Hence, the expressions of udk ðt Þ and ik ðt Þin Eqs. (3) and (4)
mathematical model of the DTS based on the essential characteris-
can be rewritten as follows
tics of LPD load.
L0
X
Definition 1.. Dynamic test signal udk ðt Þ ¼ bn g ðt  nT Þ  usk ðtÞ ; bkðLþnÞ : ¼ bkn ð5Þ
n¼0
Dynamic test signals are voltage, current and power signals
which can not only reflect the cyclic fluctuation and rapidly ran- L0
X
dom fluctuation characteristics of LPD load signal amplitude but d
ik ðt Þ ¼ akn g ðt  nT Þ  isk ðtÞ ; akðLþnÞ : ¼ akn ð6Þ
also can be used for evaluation of dynamic errors of electric energy n¼0
meter and achieve the traceability of the dynamic electrical energy

value to the steady electrical energy value. 1 nT 6 t 6 ðn þ 1ÞT; n 2 N
g ðt  nT Þ ¼ ð7Þ
0 else
Definition 2.. Dynamic error of electric energy meter
where g ðt  nT Þ is the window function used to cut off steady peri-
The errors of electric energy meter under dynamic test signal odic voltage and current signals in each period, L is cyclic period of
condition. the sequence an and bn, L0 is used to represent the length of the
According to the LPD load characteristics which were obtained
dynamic signal udk ðt Þ or ik ðtÞ, L0 > L, n; L; L0 2 N (integer set), and
d
by analyzing typical waveforms from real world, e.g. electric arc
furnace and high-speed railway traction substation, etc., the five L0 T is the test time interval. Let T d s ¼ LT, and T d s means the cyclic
d
essential characteristics of LPD load signal can be given as the period of the dynamic signal udk ðt Þ or ik ðt Þ and is an integer multiple
following. of the period of steady-state signal.
(1) Deterministic characteristics: (a) The amplitude of load Now we express the above sequences and window functions in
voltage has stable characteristics and fluctuate very slightly [1,2], vector form as following
(b) The amplitude of load current is short-time stability within
BL0 ¼ ½b0 ; b1 ;    ; bL0 1  ¼ ½1; 1;    ; 11L0 ð8Þ
each period [11], (c) The load current amplitude cyclically fluctu-
ates with large range [3] and presents the three types of run modes h i
of transient, short-time and long-time [11]. AkL0 ¼ ak0 ; ak1 ;    ; akðL0 1Þ ð9Þ
(2) Random characteristics: (d) LPD load current amplitude
fluctuates randomly [3] and approximately obeys Gaussian distri-   
GL0 ðt Þ ¼ g ðt Þ; g ðt  T Þ;    ; g t  L0 T ð10Þ
bution [21,22], (e) The LPD load current and active power are
cyclostationary signals [3]. where AkL0 is the feature vector of dynamic current test signal, the
 
Let us consider a three-phase steady-state voltage signal usk ðt Þ element akn n ¼ 0; 1;    ; L0  1 of AkL0 is deterministic or random
s
and steady-state current signalik ðt Þ sequence with binary (0, +1) or ternary (1,0, +1) for reflecting
the load current amplitude random characteristics. BL0 is the feature
usk ðt Þ ¼ U k sinðX1 t þ /k Þ ð1Þ  
vector of dynamic voltage test signal, and bn n ¼ 0; 1;    ; L0  1 is
ik ðt Þ ¼ Ik sinðX1 t þ uk Þ
s
ð2Þ constant ‘‘100 for reflecting the load voltage amplitude stable
characteristics.
 X. Wang et al. / Measurement 164 (2020) 107915 3

By using the above feature vectors, Eqs. (5) and (6) can also be 2l  1, and let T m ¼ Lm T which is the cyclic period of the m-
represented as matrix form sequence signal. The primitive polynomial for constructing the
h i m-sequence is
udk ðt Þ ¼ BL0  GTL0 ðt Þ  U k sinðX1 t þ /k Þ ð11Þ
X
l
h i f ð xÞ ¼ C 0 þ C 1 x þ C 2 x2 þ    þ C l xl ¼ C i xi ð16Þ
ik ðtÞ ¼ AkL0  GTL0 ðt Þ  Ik sinðX1 t þ uk Þ
d
ð12Þ i¼0

According to the parameterC i and Lm , we can construct a long

From Eqs. (3), (4), (11) and (12), the dynamic test power signal
can be expressed as m-sequence (Lm ¼ 212  1) and a short m-sequence (Lm ¼ 29  1)
h i to make that the m-sequence DTS consists of short-time,
d
pdk ðtÞ ¼ udk ðt Þ  ik ðt Þ ¼ AkL0  BTL0  GTL0 ðtÞ  psk ðt Þ transient-time, long-time On-Off run modes.
h i Substitute Eq. (14) into Eqs. (11)–(13), an m-sequence pseudo-
¼ AkL0  BTL0  GTL0 ðt Þ  Uk2Ik ½cosð/k  uk Þ  cosð2X1 t þ /k þ uk Þ random DTS model during a cyclic period can be constructed as
0
LP 1
h i
k ðt Þ ¼ BL  GLm ðt Þ  U k sinðX1 t þ /k Þ
T
¼ Uk2Ik an bn g ðt  nTÞ  ½cosð/k  uk Þ  cosð2X1 t þ /k þ uk Þ um m ð17Þ
n¼0

ð13Þ  
kLm  GL ðt Þ  Ik sinðX1 t þ uk Þ
m
ik ðtÞ ¼ Am m ð18Þ
where  is Hadamard product of matrices. Eqs. (11), (12), and h i
(13) are the general mathematical models of DTS. And a block dia- pkm ðt Þ ¼ Am T T
kLm  BLm  GLm ðt Þ
gram of the DTS keying generation scheme is shown in Fig. 1. The
U k Ik
dynamic test current signal is generated via Silicon Controlled Rec-  ½cosð/k  uk Þ  cosð2X1 t þ /k þ uk Þ ð19Þ
tifier (SCR) On/Off switching which is controlled by AkL0 . 2
In next section, based on the general mathematical models of where the continuous ‘‘100 of the mðnÞ in AmkL0 represent an m-
DTS, two kinds of practical signal models will be constructed and sequence DTS ‘‘On” run. In contrast, the continuous ”0‘‘ of the
the signal waveforms will be given. mðnÞ represent an m-sequence DTS ‘‘Off” run. Based on the
experimental scheme in Fig. 1, a m-sequence modulated dynamic
2.2. m-Sequence pseudorandom DTS model test current and power signals have been generated as shown in
Fig. 2.
It is well known that m-sequence signal is a binary pseudoran-
dom cyclostationary signal with random amplitude fluctuation 2.3. OOK deterministic DTS model
within each cyclic period [23], and the ‘‘100 or ‘‘0” run lengths of
m-sequence signals follow approximate Gaussian distribution in In order to reflect the above Deterministic characteristics (b)
each cyclic period [24]. These random characteristics are similar (c) of the LPD load, AkL0 in the DTS model can be defined as
to the above Random characteristics (d) (e) of the LPD load signal. 8 h 9
Thus, the m-sequence is especially suitable to be the element of   < M 1 ; M2 2 1; 2;    ; LOOK =
AOOK
kLOOK ¼ AkM1 ; AkM2 : ð20Þ
feature vector AkL0 of dynamic test current signal in the DTS mod- : ;
els. Let
M 1 þ M 2 ¼ LOOK ; k ¼ a; b; c
  0  where,
Am
kL0 ¼ mð0Þ; mð1Þ;    ; m L  1 ð14Þ
 
AkM1 ¼ ak0 ; ak1 ;    ; akðM1 1Þ ¼ ½1; 1;    ; 11M1 ð21Þ
where mðnÞ is an m-sequence and defined as follows [24,25]
mðnÞ ¼ C 1 mðn  1Þ  C 2 mðn  2Þ      C n mð0Þ  
AkM2 ¼ akM1 ; akðM1 þ1Þ ;    ; akðM1 þM2 1Þ ¼ ½0; 0;    ; 01M2 ð22Þ
P
l ð15Þ
¼ C i mðn  iÞð mod 2Þ
i¼1

where the element of mðnÞ is equal ‘‘100 or ‘‘0”, l is the number of

the register stages, and parameters fC i ¼ 0 or 1 : i ¼ 0; :::; ng rep-
resent the state of the feedback line off or on between shift regis-
ters respectively. Let Lm be the m-sequence length which is equal
uam (t )

uks HE 5020 DTS generation Tested smart iam (t )

Three-phase Steady- device energy meter
state Voltage

Three-phase Steady-
iks
pam (t )
state Current Synchronous
Detection

Sequence Generator
A kn′
1
Steady Power Standard
Fluke 6100
0 ikd
s Control switch d
i k
i
k

Fig. 2. Waveforms of a-phase m-sequence modulated dynamic test current and

Fig. 1. DTS keying generation scheme. power signals.
4 X. Wang et al. / Measurement 164 (2020) 107915

or, mode signals, i.e., the steady- state periodic intrinsic mode signal
  psk ðt Þ (abbreviation, steady signal) and two (or triple) value ampli-
AkM2 ¼ akM1 ; akðM1 þ1Þ ;    ; akðM1 þM2 1Þ 0
tude dynamic intrinsic mode signal pLk ðtÞ (abbreviation, dynamic
¼ ½1; 1;    ; 11M2 ð23Þ intrinsic signal), and they can be expressed as

whereLOOK is OOK sequence length, M1 and M 2 represent the num- 1

psk ðt Þ ¼ U k Ik ½cosð/k  uk Þ  cosð2X1 t þ /k þ uk Þ ð27Þ
ber of On or Off periods of the OOK dynamic test current signal 2
OOK
within L . Meanwhile, the different value of M1 and M2 can reflect
0
h i
transient-time modes (Rapid amplitude shock of testing current), pLk ðtÞ ¼ AkL0  BTL0  GTL0 ðtÞ ð28Þ
short-time On-Off run modes (Quick fluctuating amplitude of test-
ing current) or long-time On-Off run modes (Medium fluctuating The steady signal psk ðt Þ has higher stability and is easy to trace
amplitude of testing current) [11]. Substitute Eqs. (20)–(23) into its value of electrical energy under steady conditions. Oppositely,
0
Eqs. (11)–(13), the OOK deterministic DTS during a cyclic period dynamic intrinsic signalpLk ðtÞ performance is related to AkL0 which
can be constructed as reflects the essential characteristics of DTS, and the mode signal
h i waveforms are shown in Fig. 4.
uOOK
k ðtÞ ¼ BLOOK  GTLOOK ðtÞ  U k sinðX1 t þ /k Þ ð24Þ According to the definition of active energy and mode decom-
position, the measurement of the electrical energy during one per-
 
ðtÞ ¼ AkM1 ; AkM2  GTL ðt Þ  Ik sinðX1 t þ uk Þ iod T is calculated as
OOK
ik ð25Þ
Z ðnþ1ÞT Z ðnþ1ÞT
h i 0
Ekn ¼ pdk ðt Þdt ¼ psk ðtÞ  pLk ðt Þdt ð29Þ
pOOK
k ðt Þ ¼ AkM  BTLOOK  GTLOOK ðt Þ nT nT


U k Ik
fcosð/k  uk Þ  cosð2X1 t þ /k þ uk Þg ð26Þ where n ¼ 0; 1;    ; L0  1. Let EX0 be the three-phase dynamic elec-
2 trical energy value of the DTS, thus, within the test time interval
where ðM1 þ M2 ÞT ¼ T OOK is the cyclic period of the OOK DTS. L0 T, EX0 can be calculated as
Through the experimental scheme with OOK sequence generator L0 1
X X
in Fig. 1, an OOK deterministic dynamic test current and power sig- EX0 ¼ Ekn ð30Þ
nals (M1 ¼ 3 and M2 ¼ 1) have been generated as shown in Fig. 3. k¼a;b;c n¼0
In summary, the above general models of DTS can reflect the
Substitute Eqs. (27) and (28) into Eq. (30). And during the test-
deterministic and random characteristics of LPD load efficiently.
ing time interval L0 T, the three-phase dynamic electrical energy
value EX0 can be expressed as follows
3. The Cs indirect measurement method of DTS electrical energy
value P 0
LP 1 P 0
LP 1 R ðnþ1ÞT 0
EX0 ¼ Ekn ¼ nT
psk ðtÞ  pLk ðt Þdt
k¼a;b;c n¼0 k¼a;b;c n¼0
3.1. Mode decomposition and measurement of DTS n h io
P 0
LP 1 R ðnþ1ÞT
¼ nT
psk ðt Þ  AkL0  BTL0  GTL0 ðt Þ dt
k¼a;b;c n¼0
For the sake of more accurate measurement of DTS electrical ð31Þ
energy which is used to calculate the dynamic error of the smart h i P 0
LP 1 R ðnþ1ÞT
energy meter, this paper proposes a novel indirect measurement
¼ AL0  BTL0  1
L0 nT
psk ðtÞg ðt  nT Þdt
k¼a;b;c n¼0
method. At first, the DTS is processed by mode decomposition to h i P R L0 T
get the two mode signals, then a CS measurement matrix is con- ¼ AL0  BTL0  L10 0
psk ðtÞ dt
k¼a;b;c
structed based on the prior structure information of AkL0 . Finally,
the reference electrical energy value of DTS is measured by an indi- where AL0 ¼ AaL0 ¼ AbL0 ¼ AcL0 . Let ES0 be steady signal electrical
rect measurement model. energy value. ES0 can be calculated as
Through analyzing the establishment process of the general DTS
model and Eq. (13), it is easy to see that the signal consists of two

uas (t )

uaOOK (t ) iad (t )
L′T
OOK d
i (t ) p a (t )
a

paOOK (t ) pkL′ (t ) 1 1 1 1 1

0 0 0 0
-1 -1 -1

s
p (t )
k

Fig. 4. Waveforms of DTS mode decomposition. Single-phase signal and ternary

0
Fig. 3. Waveforms of a-phase OOK dynamic test current and power signals. (+1, 0–1) signal pLk ðtÞ are chosen as an example.
 X. Wang et al. / Measurement 164 (2020) 107915 5

X Z L0 T (2) U only consists discrete values(such as‘‘000 or ”100 ), so it is

ES0 ¼ psk ðtÞ dt ð32Þ simple to measure the cumulative eigenvalue ED0 of dynamic
k¼a;b;c 0
intrinsic signal;
Let ED0 be dynamic intrinsic signal cumulative eigenvalue. ED0 (3) U has low computational complexity, and it is easy to build
can be expressed as indirect measurement model.
h i
ED0 ¼ AL0  BTL0 ð33Þ 3.3. CS indirect measurement model for measuring dynamic electrical
0
energy value of DTS
It is well known that the steady electrical energy ES0 during L T
can be measured accurately by standard electric energy meter. The The dynamic electrical energy value EX0 can be deduced by sub-
challenge is to design an appropriate operator U to calculate the stituting Eq. (39) into (35) as follows
cumulative eigenvalue ED0 of dynamic intrinsic signal AL0 . Let U " þ 0  #
be a matrix operator 1 B L ;t
EX0 ¼ E  ½ 1; 1     ATL0
L0
S0
ED0 ¼ U½AL0  ð34Þ B L0 ; t

Substitute Eq. (33) into Eq. (31), the calculation of dynamic 2 P þ

3
" # bk ðnÞ
energy value EX0 during L0 T can be simplified as the product of þ
B ðn; t Þ; AL0 6 bþ ðnÞ¼1 7
ES0 and ED0 , ¼ L10 ES0  ½1; 1 ¼ ES0  L10 ½1; 16
4 P
k 7
hB ðn; t Þ; AL0 i bk ðnÞ 5


1 b
k ðnÞ¼1
EX0 ¼ ED0  ES0 ð35Þ " #
L0 Mþ ðMþ M Þ
¼ ES0 L10  ½1; 1  ¼ ES0  L0
As mentioned above thatAL0 can be either deterministic M
sequence or pseudorandom one. Thus, the measuring method of
ð40Þ
ED0 is the key of accurate measurement of DTS electrical energy
value. According to the CM theory of non-reconstruction signal where
processing, we measure cumulative eigenvalue ED0 accurately by X þ
constructing a novel CS measurement matrix. Mþ ¼ bk ðnÞ ð41Þ
bþ
k
ðnÞ¼1

3.2. CS measurement matrix construction based on the prior structure X 

information of DTS M ¼ bk ðnÞ ð42Þ
b
k ðnÞ¼1

For measuring ED0 , a suitable matrix operator U is needed to be

where Mþ is the positive cumulative eigenvalue of dynamic current
constructed. According to the prior structure information of
h i signal, similarly M  is the negative cumulative eigenvalue of
AkL0 ¼ ak0 ; ak1 ;    ; akðL0 1Þ :akðLþnÞ ¼ akn , L0 > L, we construct a CS dynamic current signal.
matched filter bank as follows Then, we get the CS indirect measuring model of dynamic refer-
( þ þ   ) ence electrical energy of DTS as
þ
 0
 bk ðn; t Þ ¼ bk ðnÞg t  nT 0 :

B L ;t ¼ ð36Þ   Mþ M
bk ðnÞ ¼ jakn j 1þsgn2ðakn Þ ; n ¼ 0L~0  1
þ
EX0 ¼ f M þ =L0 ; M  =L0 ; ES0 ¼ ES0   0 ð43Þ
L0 L
(     )
  bk ðn; t Þ ¼ bk ðnÞg t  nT 0 : It is worth noticing that the model has three indirect input
B L0 ; t ¼  1sgnðakn Þ
ð37Þ parameters including ES0 , M þ =L0 , and M  =L0 . ES0 is the electrical
bk ðnÞ ¼ ja kn j 2
;n 0
¼ 0L~  1
energy value of steady signal and can be accurately measured by
where Bþ ðn; t Þ is the positive eigenvectors of filter bank,B ðn; t Þ is a standard electric energy meter. M þ =L0 and M  =L0 are the occur-
the negative eigenvectors, and b(n) is related to the element value rence probabilities of the positive and negative eigenvalues of
of akn in AL0 . Thus, the CS measurement matrix U is constructed from dynamic current signal during the testing time interval respec-
the above CS matched filters as follows tively. They can be easily calculated by means of computer pro-
"  # " þ þ þ # gram based on Eqs. (41) and (42). The model output EX0 is the
Bþ L0 ; t b ð0; tÞ b ð1; tÞ    b L0  1; t value of the dynamic reference electrical energy and the dynamic
U¼   ¼      ð38Þ
B L0 ; t b ð0; tÞ b ð1; tÞ    b L0  1; t errors of smart energy meter can be calculated as follows

From Eq. (38), we built a operator U ¼ ½1 ; 1U and then Eq. EX - EX0
e¼ ð44Þ
(34) can be calculated as EX0
ED0 ¼ U ½AL0  ¼ ½1 ; 1U  ATL0 where EX is the energy of tested smart energy meter during test
þ
" þ þ # time interval.
b ð0; t Þ b ð1; t Þ    b L0  1; t ð39Þ
¼ ½1; 1       ATL0 The model shows the advantage to make the measurement of
b ð0; t Þ b ð1; t Þ    b L0  1; t EX0 simple and easier for application. Meanwhile, the CS indirect
measurement model gives the relationship between EX0 and ES0 ,
The novel CS measurement matrixU in Eq. (38) is based on the
which achieves the traceability of the dynamic electrical energy
prior structure information ofAL0 and has three advantages as
value to the steady electrical energy value.
follows
The CS indirect measuring method and a testing system of
smart energy meter for evaluating dynamic error based on the
(1) U applies to measure both pseudorandom sequence Am
kLm and
DTS general model and the CS indirect measurement model is
deterministic sequence AOOK
kLOOK ; given in Fig. 5.
6 X. Wang et al. / Measurement 164 (2020) 107915

d
Electric Power Standard Dynamic current ik is fed to smart meter for testing the smart
d
Fluke 6100A Tested Smart meter. The dynamic current ik flows out of the smart meter and

Energy Meter d
Voltage Current converges with another current ik (see Fig. 5), and the converged
s
ikd current becomes a steady-state current ik ðt Þ again, this steady-
s
uks iks d
state current ik ðtÞ flows back to the Fluke 6100A electric power
ik
standard. On contrary, The voltage loop is parallel.
HE5020 DTS generation
In the testing experiments, the test time interval is not the inte-
device
ger times of the DTS cyclic period which is called asynchronous
iks ikd testing. And we adopted two kinds of DTS with the following
conditions:
(1) The first kind of DTS is m-sequence pseudorandom DTS
Reference Electric energy meter RD33 which includes short m-sequence DTS and long m-sequence DTS.
Energy In order to reflect the different modes of signal and improve the
pulse A kn
detection efficiency, we selected primitive polynomials which are
output used to generate m-sequences as follows
energy pulse (a) The adopted primitive polynomial of short m-sequence is
f ðxÞ ¼ x9 þ x5 þ 1(l ¼ 9) with the parame-
Dynamic electrical energy tersfC i ¼ 0 : i ¼ 1; 2; 3; 4; 6; 7; 8g fC 9 ¼ C 5 ¼ C 0 ¼ 1g. Meanwhile,
value calculation unit
+ the frequency of synchronous signal for generating the short m-
Energy pulse Indirect Measurement M CS s
sequence is a quarter of the signal (usk or ik ) frequency. Thus, the
accumulation Model M − Measurement short m-sequence DTS consists short-time and transient-time
ES 0 EX 0 = f (M + / L′, M − / L′, ES 0 ) Algorithm
On-Off run modes.
(b) The adopted primitive polynomial of long m-sequence is
f ðxÞ ¼ x12 þ x11 þ x8 þ x6 þ 1(l ¼ 12) with the parameters
Dynamic error ε Energy pulse energy pulse
fC i ¼ 0 : i ¼ 1; 2; 3; 4; 5; 7; 9; 10gfC 12 ¼ C 11 ¼ C 8 ¼ C 6 ¼ C 0 ¼ 1g.
Calculation unit accumulation
And the frequency of synchronous signal for generating the long
ε = ( Eχ -E X 0 ) / E X 0 unit Eχ s
m-sequence is twenty-fourth of the signal (usk or ik ) frequency.
Fig. 5. The CS indirect measuring method and a testing system of smart energy Thus, the long m-sequence DTS consists short-time and long-
meter for evaluating dynamic error. time On-Off run modes.
(2) The second kind of DTS is OOK deterministic DTS. We
adopted five different typical OOK On-Off run modes, i.e.
transient-time run ratios: 4/20, short-time run ratios: 30/50,
4. Test results and analysis 40/40, and long-time run ratios: 100/200, 300/300. OOK sequences
represent three types of typical run modes of LPD load current
With the aim of verifying the effectiveness of the above pro- amplitude in real world respectively.
posed general DTS models and the CS indirect measurement model Other test conditions include test voltage: 3  220 V, current:
of dynamic reference electrical energy of DTS, a testing system of 3  5 A, power factor: 1.0, and frequency: 50 Hz.
smart energy meter for dynamic errors was built (see Fig. 5). The In the above experiment conditions (1) and (2), the errors of the
system is composed of five units which include Fluke 6100A elec- tested energy meter are measured repeatedly five times and the
tric power standard, HE5020 DTS generation device (developed by mean of the errors are given in Table 1. In order to make the results
our team), RD33 reference electric energy meter, dynamic electri- more intuitive, we convert the measured number of energy pulses
cal energy value calculation unit and dynamic error calculation into Wh value.
unit (developed by our team). The testing steps are as follows In Table 1, dynamic errors of tested meter increase drastically
from 0.053% and 0.007% (Small errors) to 12.47% and 4.228%
Step 1: the RD33 is used to measure steady signal energy ES0 ; (Large errors), this means that dynamic errors of tested meter
Step 2: the CS measurement algorithm is used to calculate the are related to On-Off run modes of testing current signal. The long
cumulative eigenvalue M þ and M  ; m-sequence contains both long-time On-Off run modes (Medium
Step 3: the indirect measurement model is used to calculate fluctuating amplitude of testing current) and short-time On-Off
dynamic reference electrical energy of DTS EX0 ; run modes (Quick fluctuating amplitude of testing current), so
Step 4: the dynamic errors of tested smart energy meter are cal- the dynamic errors of the tested meter in this case are between
culated based on Eq. (44) through the dynamic error calculation the above large error and small error. The possible reasons for
unit. dynamic errors of smart energy meter are electric power and
energy measurement algorithms which is designed at steady-
Tested smart energy meter is a three-phase four- wire smart state signal condition, in addition range switching algorithm of
energy meter with accuracy class 0.5. Experiments were carried current amplifier in smart energy meter should be another possible
out respectively in National Institute of Metrology (NIM) of China reason.
and North China Electric Power Research Institute (NCEPRI). The above experiments indicate that the CS indirect measure-
s
In the Fig. 5, red solid lines denote steady-state current ik ðtÞ, ment model can be used to accurately calculate the dynamic error
which is output from Fluke 6100A electric power standard and of the smart energy meter under the different DTS test conditions.
fed to RD33 reference electric energy meter and then fed to
HE5020 DTS generation device(RD33 and HE5020 in Series in Cur-
s
5. Uncertainty analysis
rent loop). In HE5020 DTS generation device, the current ik ðt Þ is
switched to ON or OFF by SCR switch (see Fig. 1), and the ON cur- Now we evaluate the uncertainty of the testing system in Fig. 5
d
rent is the dynamic current ik that is denoted by a red dotted line. based on the ISO GUM (guide to the expression of uncertainty in
 X. Wang et al. / Measurement 164 (2020) 107915 7

Table 1
Dynamic Error of smart meter under OOK, m-sequence test signals and asynchronous test conditions.

DTS On-Off run OOK On-Off run Tested smart energy meter Ev Dynamic reference electrical energy Dynamic error of tested meter
modes ratios (W.h) EX0 (W.h) e(%)
OOK DTS transient-time 4/20 112.50 112.440 0.053
short-time 30/50 55.000 55.004 0.007
40/40 72.500 72.544 0.061
long-time 100/200 75.000 85.684 12.47
300/300 112.50 117.466 4.228
m-sequen-ce m-sequence / / /
DTS length
short 29  1 150.00 150.039 0.026
long 212  1 900.00 949.075 5.17

Table 2
 
Uncertainty of uB2 M þ =L0 .
 
DTS On-Off run modes OOK On/Off run ratios or m-sequence length L0 Mþ uB2 M þ =L0 (%)

OOK DTS transient-time 4/20 36,781 6132 0.0190

short-time 30/50 23,962 9000 0.0153
40/40 23,710 11,870 0.0126
long-time 100/200 41,830 14,000 0.0095
300/300 38,429 19,229 0.0078
m-sequence DTS m-sequence length / / /
short 29  1 16,395 8183 0.0183
long 212  1 103,128 51,767 0.0029

measurement) [26]. Considering that the SCR switch in HE5020 The uncertainty budget of the indirect measuring system is
device leads to leakage current during generating DTS and negative shown in Table 3.
cumulative eigenvalue M  is zero in the experiment conditions (1) It can be seen obviously that the expanded uncertainties (k = 2)
and (2), thus, the Eq. (43) can be rewritten as follows of the CS indirect measuring system decrease with the increase of
M þ length. When M þ ¼ 6132, the expanded uncertainty has max-
Mþ
EX0 ¼ ðES0 þ DEP Þ ð45Þ imum value of 0.048%. When M þ ¼ 51767, the expanded uncer-
L0 tainty decreases to the value of 0.03%.
where ES0 is measured by RD33 reference electric energy meter. DEP The performance comparison is given in Table 4. It is worth
is electrical energy loss caused by leakage current during generating pointing out that the CS indirect measuring system based on the
  new CS indirect measuring method has obviously smaller uncer-
DTS. And Mþ =L0 is calculated by means of computer program
tainty than that of the other measurement methods with the
based on algorithms given in Eq. (41).
uncertainties of 3.0% [11], and 0.38% [10].
Based on the ISO GUM, the components of the measuring sys-
tem uncertainty are as follows.
The uncertainty of repeatability uA1 is calculated as 0.0040%
6. Conclusions
based on the Bessel formula.
uA2 ðDEP Þ is caused by the leakage current from SCR switching on
In this paper, the general models of dynamic test signal
or off in the HE5020 device. uB1 ðES0 Þ is caused by the RD33 refer-
  have been established based on the essential characteristics of
ence electric energy uB2 M þ =L0 is caused by dynamic electrical the LPD loads. Meanwhile, a three-phase pseudo-random
 
energy value calculation unit, and uB2 M þ =L0 ¼ 1=Mþ þ 1=L0 . M þ m-sequence DTS has been proposed, which can reflect the
0
and L are obtained by testing, and the uncertainties caused by dif- dynamic current amplitude fluctuation and the run distribution
ferent M þ and L0 (M þ <L0 ) are shown in Table 2. of the LPD load. A three-phase OOK DTS has been also given

Table 3
Uncertainty budge.

Source of uncertainty Symbol Uncertainty contributions (%)

Repeatability of test uA1 0.0040
HE5020 SCR switch uA2 0.0133
Reference electric energy meter RD33 uB1 0.0058
The CS indirect measurement
model run mode transient/short-time long-time short m-sequence long m-sequence
Mþ 6132 14,000 8183 51,767
uB2 0.0190 0.0095 0.0183 0.0029
combined uncertainty 0.024 0.017 0.023 0.015
Expanded uncertainty (coverage factor k = 2) 0.048 0.034 0.046 0.030
8 X. Wang et al. / Measurement 164 (2020) 107915

Table 4
The performance comparison.

NO Dynamic test current signal waveform Fluctuation of Test signal characteristics Measuring method Measurent
current amplitude for evaluating uncertainty
dynamic error
Other literatures 1
methods Sine wave Reflecting slow cyclic period direct measuring 3%
in the amplitude characteristics method in
modulation reference [11]

2 Trapezoidal wave Reflecting slow cyclic period ON/OFF ratio direct 3%

amplitude characteristics measuring method
modulation in reference [11]

3 ON-OFF Keying Reflecting transient, short- indirect 0.38%

amplitude term (fast) or long-term comparison
modulation (slow) cyclic period measuring method
characteristics in reference [10]

The proposed method in the paper 4

m- Reflecting fast or slow The CS indirect (short m-
sequenceamplitude random dynamic measuring method sequence);
modulation characteristics in the paper 0.030% (long m-
sequence)
5 ON-OFF Keying Reflecting transient, short- The CS indirect 0.048%
amplitude term (fast) or long-term measuring method (transient/short-
modulation (slow) cyclic period in the paper time); 0.034%
characteristics (long-time)

as another form of the general model, which can reflect three Appendix
kinds of current amplitude run modes and cyclical fluctuation
of the LPD loads. A. Derivation of EX0
Then a novel CS indirect measurement model have been pro- In Eq. (31), the specific calculation processes of the dynamic
posed for calculating the dynamic reference electrical energy of energy value EX0 can be given as
DTS. Based on the two DTS models (m-sequence and OOK) and
P 0
LP 1 P 0
LP 1 R ðnþ1ÞT 0
CS indirect measurement model, a CS indirect measuring method EX0 ¼ Ekn ¼ psk ðtÞ  pLk ðt Þdt
nT
and a testing system of smart energy meter for evaluating k¼a;b;c n¼0 k¼a;b;c n¼0

dynamic error has been given. Finally, the testing results of a P 0

LP 1 R ðnþ1ÞT h h ii
smart energy meter for dynamic error have been presented under ¼ nT
psk ðt Þ  AkL0 BTL0  GTL0 ðt Þ dt ðA:1Þ
k¼a;b;c n¼0
the condition of the m-sequence and OOK DTS. The test results n h io
show that the novel CS indirect measurement method can P 0
LP 1 R ðnþ1ÞT
¼ nT
AkL0  psk ðt Þ  BTL0  GTL0 ðtÞ dt
measure the dynamic errors of the smart energy meter effec- k¼a;b;c n¼0
tively. And the measurement uncertainty (k = 2) of the CS indirect
measuring system is better than 0.05% which is higher than the P 0
LP 1 R ðnþ1ÞT n h io
other proposed methods. ¼ nT
AkL0  BTL0  psk ðt Þ GTL0 ðt Þ dt
k¼a;b;c n¼0

P 0
LP 1 n R ðnþ1ÞT s o
Declaration of Competing Interest ¼ AkL0  BTL0  nT pk ðt Þ GTL0 ðtÞdt
k¼a;b;c n¼0
8 2 3 9
> g ðt Þ >
None. >
> >
>
< 6 7 > >
P 0
LP 1 R ðnþ1ÞT 6 g ðt  T Þ 7 =
¼ AkL0  BTL0  psk ðtÞ 6
6 .. 7 dt
7 >
Acknowledgment >
>
nT
4 . 5 >
k¼a;b;c >
>
n¼0
  0   >
>
: ;
g t L 1 T
This work is supported by the National Natural Science Founda-
tion of China (No. NSFC-51577006). ðA:1Þ
 X. Wang et al. / Measurement 164 (2020) 107915 9

According to the interval additivity of the definite integral, EX0 [4] S. Blaifi, S. Moulahoum, I. Colak, W. Merrouche, An enhanced dynamic model of
battery using genetic algorithm suitable for photovoltaic applications, Appl.
can be expressed as
Energy 169 (2016) 888–898, https://doi.org/10.1016/j.apenergy.2016.02.062.
8 2 3 9 [5] Z.H. Lin, X.G. Peng, K.D. Lin, Y. Liu, Dynamic Error Estimation for Power Energy
> psk ðtÞg ðtÞ >
>
> >
> Meters Based on Wavelet Decomposition and Outlier Robust ELM, in: 2018
>
< Z LT6 psk ðtÞg ðt  T Þ 7 >
X 7 =
0 International Conference on Power System Technology (POWERCON),
6
EX0 ¼ AkL0  BTL0  6 .. 7 dt Guangzhou, 2018, pp. 1950–1956. http://dx.doi.org/10.1109/
> 6 7 >
>
> 0 4 . 5 > >
powercon.2018.8602166.
k¼a;b;c
>
:     >
; [6] A. Ortiz, M. Lehtonen, M. Mañana, et al., Evaluation of energy meters’ accuracy
0
pk ðt Þg t  L  1 T
s
based on a power quality test platform, Electric Power Comp. Sys. 35 (2007)
221–237, https://doi.org/10.1080/15325000600891267.
8 2 R L0 T s 39 [7] A. Domijan, E. Embriz-Santander, et al., Watthour meter accuracy under
>
> pk ðtÞg ðtÞdt >
> controlled unbalanced harmonic voltage and current conditions, IEEE Trans.
>
> 6 0 7> >
>
> 6 R L0 T s 7> > Power Del. 11 (1996) 64–70, https://doi.org/10.1109/61.484002.
P < 6 p ð t Þg ð t  T Þdt 7 =
T 6 0 k 7 [8] Z.L. Lu, E. So, A proposal for verifying the performance specifications of certain
¼ AkL  BL0  6
0
7
>
> 6 .. 7> > functions of smart meters in distribution power line networks, in: Proc. IEEE
k¼a;b;c >
> 4 . 5> >
>
> R L0 T s   0   >
> CPEM, Daejeon, Korea, 2010, pp. 271–272. http://dx.doi.org/10.1109/
: ;
0
p k ð t Þg t  L  1 T dt CPEM.2010.5544816.
8 2 R L0 T s 39 [9] H.T. Huang, Z.L. Lu, L. Wang, et al., Dynamical waveforms and the dynamical
> 2 3 >
>
> pk ðt Þg ðt Þdt > source for electricity meter dynamical experiment, in: Proc. IEEE CPEM,
>
>
1 6 0 7>
>
>
< 617 6 R L0 T s 7>
> Ottawa, ON, Canada. (2016)1-2. http://dx.doi.org/10.1109/
P h i 6 7 6 pk ðt Þg ðt  T Þdt 7=
¼ ak0 ; ak1 ;   ; akðL0 1Þ  6 6 .
.
7
7  6
6
0 7
7
CPEM.2016.7540737.
>
> 4.5 6 .
.. 7>
> [10] X.W. Wang, J.X. Chen, et al., OOK power model based dynamic error testing for
k¼a;b;c >
> 4 5>
>
>
> R L0 T s   0   >
> smart electricity meter, Meas. Sci. Technol. 28 (2017), https://doi.org/10.1088/
: 1 L0 1 ;
0
pk ðt Þg t  L  1 T dt 1361-6501/aa5067 025015.
[11] Z.L. Lu, M. Li et al., Evaluation of the Dynamic Performance Characteristic of
ðA:2Þ Electrical Energy Meters, in: Proc. IEEE CPEM, Washington, DC, USA. (2012)
0 120-121. http://dx.doi.org/10.1109/CPEM.2012.6250690.
Within the test time L T 1 , the steady signal electrical energy [12] E.J. Candès, J. Romberg, T. Tao, Robust uncertainty principle: exact signal
value ES0 can be calculated as reconstruction from highly incomplete frequency information, IEEE Trans. Inf.
Theory. 52 (2006) 489–509, https://doi.org/10.1109/TIT.2005.862083.
Z L0 T Z ðnþ1ÞT [13] D.L. Donoho, Compressed sensing, IEEE Trans. Inf. Theory. 52 (2006) 1289–
psk ðt Þdt ¼ L0 psk ðt Þg ðt  nT Þdt ¼ ES0 ðA:3Þ 1306, https://doi.org/10.1109/TIT.2006.871582.
0 nT [14] R. Obermeier, J.A. Martinez-Lorenzo, Sensing Matrix Design via Mutual
Coherence Minimization for Electromagnetic Compressive Imaging
where Applications, IEEE Trans, Comput. Imag. 3 (2017) 217–229, https://doi.org/
R ðnþ1ÞT 10.1109/TCI.2017.2671398.
nT
psk ðt Þg ðt  nT Þdt [15] Z. Gao, L.L. Dai, et al., Structured compressive sensing-based spatio-temporal
R L0 T joint channel estimation for FDD Massive MIMO, IEEE Trans. Commun. 64
¼ psk ðtÞg ðt  nT Þdt ðA:4Þ (2016) 601–617, https://doi.org/10.1109/TCOMM.2015.2508809.
0
¼ L10 ES0 ¼ EnS0 ; n ¼ 0; 1;    L0  1 [16] Francesco Bonavolontà, Massimo D’Apuzzo, A. Liccardo, et al., Harmonic and
interharmonic measurements through a compressed sampling approach,
Measurement 77 (2016) 1–15, https://doi.org/10.1016/j.
Then, EX0 can be simplified by substituting Eq. (A.4) into (A.2) as measurement.2015.08.022.
follows [17] M. Davenport, P. Boufounos, M. Wakin, Signal processing with compressive
8 2 0 39 8 2 3 9 measurements, IEEE J. Sel. Top. Signal Process. 4 (2010) 445–460, https://doi.
>
> ES0 > > > 1 > org/10.1109/jstsp.2009.2039178.
>
> 6 7 >
> >
> >
>
>
< 6 E1S0 7> = >
< 617 >
=
[18] Z.H. Du, X.F. Chen, H. Zhang, et al., Feature identification with compressive
P 6 7 P 6 7 measurements for machine fault diagnosis, IEEE Trans. Instrum. Meas. 65
EX0 ¼ T
AkL0  BL0  6 . 7 ¼ AkL0  BL0  6
T
6 . 7  0 ES0
7
1
(2016) 977–987, https://doi.org/10.1109/I2MTC.2015.7151334.
> 6 7 > > . L >
k¼a;b;c >
>
> 4 .. 5> >
>
k¼a;b;c >
>
>
4.5 >
>
> [19] D.V. Luca, A. Octavia, A compressive sampling-based method for classification
>
: 0
L 1
>
; : ; and parameter estimation of FSK signals, Measurement (2017) 439–444,
ES0 1
https://doi.org/10.1016/j.measurement.2015.12.038.
P h i h i P R L0 T s [20] M.R. Rao, S. Chatterjee, B. Ottersten, Detection of Sparse Random Signals Using
¼ AL0  BTL0 L10 ES0 ¼ AL0  BTL0 L10 0
pk ðtÞdt Compressive Measurements, in: Proc. IEEEICASSP,Kyoto,Japan, 2012. http://
k¼a;b;c k¼a;b;c dx.doi.org/10.1109/ICASSP.2012.6288610.
ðA:5Þ [21] X.W. Wang, J. Wang, R.M. Yuan, Z.Y. Jiang, Dynamic error testing method of
electricity meters by a pseudo random distorted test signal, Appl. Energy 249
(2019) 67–78, https://doi.org/10.1016/j.apenergy.2019.04.054.
[22] Y. Yang, S. Li, W. Li, M. Qu, Power load probability density forecasting using
References Gaussian process quantile regression, Appl. Energy 213 (2018) 499–509,
https://doi.org/10.1016/j.apenergy.2017.11.035.
[1] A. Bernieri, G. Betta, L. Ferrigno, et al., Electrical energy metering: Some [23] J.G. Proakis et al., PN sequence generation, in Digital Communications, 5th ed.,
challenges of the European Directive on Measuring Instruments (MID), McGraw-Hill, New York NY, USA, 2011, pp. 554–558.
Measurement 46 (2013) 3347–3354, https://doi.org/10.1016/j. [24] H. J. Zepernick, A. Finger, Pseudo random signal processing, Theory and
measurement.2013.06.025. application, New York: Wiley, 2005. http://dx.doi.org/10.1002/
[2] M. Rylander, W.M. Grady, A. Arapostathis, et al., Power electronic transient 9780470866597.ch7.
load model for use in stability studies of electric power grids, IEEE Trans. [25] Y.F. Luo, J.P. Gao, et al., A test method of winding deformation excited by
Power Syst. 25 (2010) 914–921, https://doi.org/10.1109/ pseudorandom m-sequences-Part I: theory and simulation, IEEE Trans.
TPWRS.2009.2032351. Dielectr. Electr. Insul. 23 (2016) 1605–1612, https://doi.org/10.1109/
[3] M. Renmu, D.J. Jin, Hill, Composite load modeling via measurement approach, tdei.2016.005681.
IEEE Trans. Power Syst. 21 (2006) 663–672, https://doi.org/10.1109/ [26] Guide to the expression of uncertainty in measurements, International
tpwrs.2006.873130. Organization for Standardization, Switzerland, Geneva, 1993.