You are on page 1of 12

Received: 17 June 2021 Revised: 23 August 2021 Accepted: 5 September 2021 IET Generation, Transmission & Distribution

DOI: 10.1049/gtd2.12298

ORIGINAL RESEARCH PAPER

A variant of Newton–Raphson method with third-order


convergence for energy flow calculation of the integrated electric
power and natural gas system

J.H. Zheng C.Q. Wu K.S. Xiahou Zhigang Li Q.H. Wu

School of Electric Power Engineering, South China Abstract


University of Technology, Guangzhou, China
The energy flow calculation of the integrated electric power and natural gas system (IEGS)
is generally tackled by the classical Newton–Raphson (NR) method with second-order con-
Correspondence
Zhigang Li, School of Electric Power Engineering, vergence. However, this method may fail to converge or incur oscillating iterations lead-
South China University of Technology, Guangzhou ing to heavy computation burden if the initial point is not selected properly, especially in
510640, China.
heterogenous integrated energy systems containing the natural gas system. To handle this
Email: lizg16@scut.edu.cn
problem, a variant of Newton–Raphson method with third-order convergence is proposed,
Funding information which needs one function and two derivative evaluations per iteration without increasing
Guangdong Basic and Applied Basic Research Foun- the number of derivative evaluations. In order to verify the effectiveness of the proposed
dation, Grant/Award Number: 2020A1515111100; method, simulation studies are carried out on several different cases for the energy flow
National Key Research and Development Program
of China, Grant/Award Number: 2018YFE0208400; calculation of IEGS. Experiment results reveal that the proposed method is superior to the
Natural Science Foundation of Guangdong Province, classical Newton–Raphson method and its other variants in terms of computational effi-
Grant/Award Number: 2021A1515010584; Natural ciency.
Science Foundation of China, Grant/Award Number:
52007066; Talent Recruitment Project of Guangdong,
Grant/Award Number: 2017GC010467

1 INTRODUCTION method [2, 4, 11, 12]. The non-iterative technique is the analytic
method, that is, the holomorphic embedding method (HEM)
The restructuring of energy systems in several parts of the world [13] and homotopy method [14–16]. HEM was proposed first
has increased, and more and more attention is being paid to by Dr. Antonio Trias in 2012 [13, 17], and many applications
study the integrated energy system (IES), especially in the elec- based on HEM have been developed for EPS [18–21] and NGS
tric power system (EPS) coupled with NGS and district heat- [22].
ing network (DHS) [1–6]. However, energy flow study is a fun- As for the non-iterative technique, many scholars have paid
damental tool for heterogenous energy systems planning and attention to studying the trend of using non-iterative algorithm
operations [7–10]. Thus, to have an effective operating of het- to calculate the energy flow of EPS and NGS. In [15], a non-
erogenous integrated energy systems, it is essential to determine iterative method based on the homotopy method was proposed
which method is efficient and suitable to analyze the energy flow to solve load-flow problem of EPS. In [22], Zhang et al. pro-
of EPS, NGS, DHS and IES. posed a non-iterative method to calculate the energy flow of
It is well-known that the equations set describing the het- NGS which was based on the fast and flexible holomorphic
erogenous energy systems are non-linear, and there exist vari- embedding method proposed by Hsiao-Dong Chiang [23]. This
ous approaches for solving the energy flow problem in differ- is a novel idea to tackle the energy flow of NGS, but there is a
ent energy systems. At present, there are two common methods defect that the model pipeline of natural gas network contains a
to solve non-linear equations in energy systems: non-iterative square-root or close to square-root type terms and the direction
method and iterative method. The iterative method in energy of the gas flow in pipelines, which cannot be tackled effectively
systems mainly refers to the Newton–Raphson method and its by HEM at present. Moreover, if the initial guess is improper,
variants, Gauss–Seidel method and fast decoupled load flow it may not find the final solution. In [24, 25], the flow equation

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the
original work is properly cited, the use is non-commercial and no modifications or adaptations are made.
© 2021 The Authors. IET Generation, Transmission & Distribution published by John Wiley & Sons Ltd on behalf of The Institution of Engineering and Technology

IET Gener. Transm. Distrib. 2021;1–11. wileyonlinelibrary.com/iet-gtd 1


2 ZHENG ET AL.

of natural gas network was modeled as a convex optimization the energy flow problem of the sub-system NGS in IES or
problem; then a linear program is employed to tightly linearize IEGS, the major equation used is Kirchhoff ’s first law, that is,
the non-linear parts to find the final solutions by the commer- Newton-nodal method, but the main disadvantage of Newton-
cial solver GUROBI. This is an original method to tackle the nodal method used to solve the energy flow of NGS or IEGS
energy flow of natural gas system, but the accuracy of the final is poor convergence characteristics for gas energy problems,
solution depends on the number of segments. Moreover, Wang due to the nodal equations contain square-root or close to
et al. [7] proposed an individual-based model for the modeling square-root type terms in natural gas network which model by
of large-scale heterogenous IES considering the physical char- Lacey model, Polyflo model, Panhandle A model , Panhandle
acteristics of EPS and DHS; this is an original idea for the model B model, or Weymouth model [1, 33]. Especially, if the ini-
and energy flow analysis of IES, but there is a defect that it tial guess of the pressures at all nodes of NGS or IEGS for
needs to tackle the issue of dimensionality reduction problem the iterative process are far away from the solution, the com-
to improve computing efficiency. In [26], a fast decomposed putational process may incur oscillating convergence or even
method, combining the non-iterative technique HEM and the diverge. Therefore, a reasonable initial guess of the pressure at
iterative method NR, was proposed to solve the energy flow of all nodes should be given or a useful iterative scheme based on
IES. Here, HEM was only used to tackle the energy flow of the Newton-nodal method should be proposed for finding the final
EPS rather than the NGS or DHS. solution.
The iterative method of generalized Newton–Raphson To improved computational efficiency and tackle this disad-
method and its variants has been widely used in EPS and other vantage of the Newton–Raphson method using energy flow cal-
heterogenous integrated energy systems, such as EPSs coupled culation of IEGS, we proposed an iterative scheme based on the
with DHSs, EPSs coupled with NGSs, or EPSs coupled with Newton–Raphson iterative method. The proposed method is
DHSs and NGSs. In [1], a method to calculate natural gas load- based on a conditional optimal two-step iterative method which
flow problems using electric loadflow techniques based on the is at least third-order with one function and two derivative eval-
Newton–Raphson method was presented. In [2], the Jacobian uations in each iteration.
matrix of the natural gas pipeline flow equation was calculated, Here, a model of IEGS considering bi-directional energy
and classical Newton–Raphson method is used to solve the conversion, that is, P2G and GF, is proposed, first. Subse-
energy flow of the IEGS. The authors in [4] proposed an inte- quently, in order to enhance the computational speed of energy
grated framework based on the Newton–Raphson technique to flow of IEGS, a variant of Newton–Raphson method with
solve the combined energy flow problem, and the interdepen- third-order convergence method is introduced. To verify the
dencies of the infrastructures and the non-linear part-load effi- efficiency of the proposed method, several cases of heteroge-
ciency performance for units were considered. In [27], a com- nous energy systems are taken, such as EPSs, NGSs, and IEGSs.
bined natural gas and electric optimal power flow problem was Moreover, a comparison is made; the proposed method is com-
proposed, and the variables in the NGS explained clearly, like pared with other Newton’s variant method, such as trapezoidal
the power flow calculation in the EPS, the known and unknown Newton’s method, Newton–Downhill method, midpoint New-
variables like PV buses and PQ buses were mentioned specifi- ton’s method and Jarratt method.
cally for the NGS flow calculation. Liu et al. [28] proposed the The remainder of this paper is organized as follows. Sec-
decomposed and integrated electrical–hydraulic–thermal calcu- tion 2 introduces the mathematical formulation of IEGS. Sec-
lation techniques to solve the energy flow problem of heteroge- tion 3 introduces a variant of NR method based on classical
nous integrated energy systems, and the main tackling tech- NR method to solve the energy flow of the IEGS. Section 4
nique is the generalized Newton–Raphson method. To ascer- analyses the proposed method. Finally, conclusions are drawn
tain the energy flow of IES, Li et al. [29] proposed a gradient in Section 5.
descent iterative method to tackle the energy flow of NGS based
on Newton–Raphson iterative method. The gradient descent
iterative method is effective for solving non-linear equations, 2 THE MODEL OF IEGS
and the key of the method is the derivation of the Jacobian
matrix [30]. In [31], a fast decoupled energy flow calculation The framework of IEGS system considered in this paper is
method for IES is proposed and its main idea is to replace the shown in Figure 1. As shown in the Figure 1, the IEGS con-
original Jacobian matrix of energy flow calculation based on sists electric power system, natural gas system and the coupling
Newton–Raphson method with a diagonal and constant Jaco- components. In the IEGS, the coupling units, that is, gas-fired
bian matrix by the transformation. In [32], Zeng et al. investi- generators, electricity-driven compressors and P2G units are the
gated the impacts and benefits of the bi-directional energy con- key components that result in the tight coupling of the EPS and
version of the IEGS; then a unified energy flow formulation NGS. The detail models of each part are shown as below.
was developed to describe the nodal balance and branch flow in
the two systems and it was solved with the classical Newton–
Raphson method. Here, it mainly considered the power-to-gas 2.1 Electric power system
(P2G) unit and gas-fired generation (GF).
As mentioned above, the iterative method of the classical The AC power flow equation is utilized for the EPS, and the
Newton–Raphson method and its other variants in solving balance equation of the active and reactive power injection at
ZHENG ET AL. 3

FIGURE 1 The brief framework of the proposed IEGS

bus i can be written as [4]: pipe


The gas flow fmn from the connected nodes m to n and is
∑ ( ) given in (4). The sign(𝜋m , 𝜋n ) by (5) indicates the direction of
Pi = Vi V j Gi j cos 𝜃i j + Bi j sin 𝜃i j , (1) the gas flow.
j ∈i
pipe
∑ ( ) fmn = sign(𝜋m , 𝜋n )Kmn (sign(𝜋m , 𝜋n ) ⋅ (𝜋m2 − 𝜋n2 ))0.5 (4)
Qi = Vi V j Gi j cos 𝜃i j − Bi j sin 𝜃i j , (2)
j =i
{
+1 𝜋 m ≥ 𝜋n
where Vi and V j are voltage magnitudes of bus i and j , while sign(𝜋m , 𝜋n ) = (5)
−1 𝜋 m < 𝜋n
𝜃i j = 𝜃i − 𝜃 j and 𝜃i and 𝜃 j are voltage angles of those buses.
Gi j and Bi j are the conductance and susceptance of this trans- 𝜋m is pressure at node m, and Kmn denotes the physical charac-
mission line, respectively. teristics of each pipeline which can be referenced in [34], [35].
The gas flow through the ith compressor between node m
and node n can be mathematically expressed in (6):
2.2 Natural gas system
comp
comp Hi
2.2.1 The physical model of NGS fmn = [ ( ) ], (6)
( )Zki ⋅ 𝛾−1
𝜋n 𝛾
Bk −1
The gas-flow balance equation at each node of natural gas net- 𝜋m
work can be written in matrix form as (3)
comp
( ) comp where Hi represents the horsepower consumption of ith
F g xg = Ag f pipe + U g f comp + f gas − T comp f need , (3)
compressor. Zki denotes the gas compressibility factor at com-
pressor inlet and Bk is the physical characteristics the of the
where xg = [𝝅; H], xg represents the variables of NGS’s equa- compressor that can be referenced [34].
tions, including pressure variable 𝝅 and the power of com-
pressor H. Ag is node-branch incidence matrix of NGS. f pipe
is the vector of natural gas flow within each pipeline. U g is 2.2.2 The variables of the NGS statement
incidence matrix between gas node and compressor which
describes the connection of compressors and nodes. f comp Generally speaking, the known and unknown variables of EPS
is the vector of natural gas flow through each compressor. are two types, that is, PV buses and PQ buses [11]. The vari-
f gas is a vector of gas injections at each node. The matrix ables of the NGS includes the known-injection nodes and
comp
T comp and the vector f need represent where gas is with- known-pressure nodes [1]. Generally, the source and load nodes,
drawn to power a gas turbine to operate the compressor. If and junctions with no gas injections belong to this node of
the compressor is driven by electric power, the T comp = 0 and the known-injection nodes type. For a node i of the known-
f comp = 0. injection nodes type, we know the gas flow injection fi , but
The gas flow is driven by the pressure difference between the pressure 𝜋i . For the known-pressure nodes type, we gen-
the two end notes of a pipeline. Here, the Weymouth equation erally assume that they are the references for others like the
[34], which is most often used for stead-state flow in a pipeline, slack buses in EPS, but the gas flow injection fi of a node i
is applied. is unknown.
4 ZHENG ET AL.

2.2.3 The partial derivative of natural gas flow equations containing square-root or close to square-root [1]. As
can be seen from equations (7) and (8), there contains square-
Based on (4)–(6), the partial derivative of natural gas flow of root in the iterative Jacobian matrix of NGS based on the nodal
each pipeline with respect to the corresponding pressure and method. If the initial point is not selected properly, there may
horsepower of compressor are given as below [36]. incur oscillating iterations or even diverge. Therefore, an effi-
For pipeline, if 𝜋m > 𝜋n > 0 of (4), then cient iterative scheme based on nodal method should be pro-
posed for finding final solution of IEGS or NGS.

pipe
𝜕 fmn ∕𝜕𝜋m = Cmn 𝜋m ∕ 𝜋m2 − 𝜋n2 . (7)
2.3 Coupling components
If 𝜋n > 𝜋m > 0 of (4), then
√ IEGS is composed of fundamental units, such as multiple
𝜕
pipe
fmn ∕𝜕𝜋m = Cmn 𝜋m ∕ 𝜋n2 − 𝜋m2 . (8) energy supplies, multiple energy converters, and multiple energy
consumers. Here, we only consider the coupling units of GF,
P2G, compressors driven by electricity (15) and compressors
For compressor, if node m is the inlet of compressor of (6),
driven by gas (16).
then
𝛽
𝜋n
comp 𝛽Hi Bki 2.3.1 Compressor
𝜕 fmn 𝛽+1
𝜋m
=[ (( )]2 (9)
𝜕𝜋m )𝛽 comp
𝜋n The amount of consumed gas fi,need for a compressor to pro-
Bki −1
𝜋m vide horsepower Hi can be obtained by (15)

comp ( comp )2
𝛽−1
𝜋n comp comp comp comp
comp −𝛽Hi Bki fi,need = ai + bi Hi + ci Hi , (15)
𝜕 fmn 𝛽
𝜋m
=[ (( )]2 (10)
𝜕𝜋n )𝛽 comp comp comp
Bki
𝜋n
−1 where ai , bi and ci are the efficiency the ith compres-
𝜋m sor.
If the ith compressor between node m and node n is driven
comp 1 by electrical power, the relationship of power Pi
comp
𝜕 fmn ∕𝜕Hi = (( )𝛽 ). (11) and horse-
comp
Bki
𝜋n
−1 power Hi can be summarized as below:
𝜋m
comp comp comp
Pi = 𝜂i Hi , (16)
If node n is the inlet of compressor of (6), then
comp
𝛽−1 where 𝜂i is efficiency of converting horsepower into elec-
𝜋n
comp 𝛽Hi Bki tricity for the ith compressor.
𝜕 fmn m 𝜋
𝛽
=[ ( )]2 , (12)
𝜕𝜋m ( )𝛽
𝜋n
Bki −1 2.3.2 Gas-fired generator
𝜋m

𝛽 From the point of natural gas network, GF can be regarded as


𝜋n
comp −𝛽Hi Bki
𝜕 fmn 𝛽+1
𝜋m negative gas injections at the gas load nodes. The output of GF
=[ ( )]2 , (13) PiGF and the input natural gas flow fiGF has the following rela-
𝜕𝜋n ( )𝛽
𝜋n tionship:
Bki −1
𝜋m

𝛼i + 𝛽i PiGF + 𝛾i (PiGF )2
comp −1 fiGF = , (17)
𝜕 fmn ∕𝜕Hi = (( )𝛽 ), (14) GHVgas
𝜋n
Bki −1
𝜋m
where 𝛼i , 𝛽i , 𝛾i are determined by the heat rate curve coefficient
𝛾−1
of GF. GHVgas is gross heating value.
where 𝛽 = Zki ⋅ ( ). The detailed equations of (7)–(14)
𝛾
and the iterative Jacobian matrix of NGS can be found in
[36]. 2.3.3 P2G
When NGS is modeled by nodal method of (4), the main
disadvantage of Newton-nodal method is the poor conver- The presence of P2G technology provides an opportunity to
gence characteristics in solving gas energy flow, due to nodal convert surplus renewable energy to a gas fuel [37]. The model
ZHENG ET AL. 5

of P2G unit can be given by [32] where xn is the nth iterate, and f ′ (xn ) is the first partial deriva-
( ) tive.
𝜂P2G In this manner, (23) is extended to calculate the energy flow
fiP2G = × PiP2G , (18)
GHVgas for large-scale energy systems, that is, the classical NR method
[4]. The classical NR iterative method for solving non-linear sys-
where fiP2G is the output gas generation flow of the P2G unit, tem of IEGS with multiple dimensions is:
𝜂P2G denotes the energy conversion from electricity to the nat-
[ ]−1 ( )
ural gas in P2G unit, and PiP2G is electric power consumed by ΔX(n) = J (n) F X (n) , (25)
P2G unit.
X (n+1) = X (n) − ΔX (n) , (26)

3 A VARIANT OF NR METHOD FOR where J (n) is iterative Jacobian matrix, and J (n) = F ′ (X (n) ).
ENERGY FLOW CALCULATION OF IEGS

3.1 The unified energy flow equation of 3.3 Basic framework of the proposed
EGS iterative scheme for IEGS

Based on equations (1)–(18), the integrated model of IEGS can An iterative scheme based on nodal method for solving
be expressed by equations (19) and (20). non-linear system of IEGS with multiple dimensions is pro-
posed to find the final solution, which is based on a condi-
( )
∗ −P =0 tional optimal two-step iterative scheme proposed by Wang
( ) ⎡ Real V (YV ) S ⎤
F e xe , xg = ⎢PS = Pgen +(PGF − PeLD ) − P comp − P P2G
⎥, et al. [38]. The proposed iterative scheme for solving non-
⎢ ∗ ⎥ linear system of IEGS with multiple dimensions is shown as
⎣ Imag V (YV ) − QS = 0 ⎦
below.
(19) Let ΔX (n) of (26) as
comp ( )
A f pipe + U g f comp + f gas − T comp f need
( ) ⎡ g ⎤ ΔX = G X (n) , Y (n) dX , (27)
F g xe , xg = ⎢ f gas = f source + f P2G − f LD − f GF ⎥,
gas
⎢ ⎥
⎣ ⎦ and

(20)
Y (n) = X (n) − 𝜃 dX , (28)
F (X ) = [F e ; F g ], (21) [ ]−1 ( )
dX = J (n) F X (n) , (29)
X = [xe ; xg ], (22)
where G (X (n) , Y (n) ) = [4J ∗ − 2J (n) ]−1 [J ∗ + J (n) ], and
∗ ′
where F e and F g represent equations of the sub-system of J = F (Y ). (n)

EPS and NGS in IEGS, respectively. xe represents the vari- The iterative scheme of (27)–(29) with one-dimensional form
ables of EPS’s equations, including voltage angle vector 𝜽 and is as below
voltage magnitude vector |V |; xg represents the variables of f (xn )
yn = xn − 𝜃
NGS’s equations, including pressure variable 𝝅 and compres- f ′ (xn )
gas f (xn ) (30)
sor horsepower H. PeLD and f LD are electrical loads and gas xn+1 = xn − G (xn , yn ) .
loads, respectively. Pgen denotes the active power of traditional f ′ (xn )
generator, that is, the output of coal-fired generators and hydro-
electric generators. QS is the reactive power injected into the Proof. Let en = xn − 𝛼, C j = (1∕ j !) f ( j ) (𝛼)∕ f (1) (𝛼), j =
EPS. f source is the gas flow which supplied by gas wells. 1, 2, 3, …, 𝛼 is the simple root of f (x). We denote G (𝛼, 𝛼) by
G , Gx (𝛼, 𝛼) by Gx , Gy (𝛼, 𝛼) by Gy , Gxx (𝛼, 𝛼) by Gxx , Gyy (𝛼, 𝛼)
by Gyy , Gxy (𝛼, 𝛼)by Gxy , Gxxx (𝛼, 𝛼) by Gxxx , Gxxy (𝛼, 𝛼) by
3.2 Basic framework of the classical NR Gxxy and Gxyy (𝛼, 𝛼) by Gxyy .
method for IEGS By Taylor’s expansion, we have

Consider the non-linear equation f (x), the classical Newton’s 1 (2)


f (xn ) = f (𝛼 + en ) = f (𝛼) + f (1) (𝛼)en + f (𝛼)en2
method for solving non-linear equations is: 2!
1 (3) ( )
xn+1 = xn − Δxn , (23) + f (𝛼)en3 + O en4
3!
[ ( )]
Δxn = f (xn )∕ f ′ (xn ), (24) = f (1) (𝛼) en + C2 en2 + C3 en3 + O en4 (31)
6 ZHENG ET AL.

TABLE 1 The processing scheme of the damping-factor 𝜆


f (1) (xn ) = f (1) (𝛼 + en ) = f (1) (𝛼) + f (2) (𝛼)en
set: 𝜆 (k) = 1, F 1 = ‖F (X (k) )‖2 ,
1 (3) ( ) T = X (k) − 𝜆 (k) ΔX (k) , eps = ‖F (T )‖2 − F 1
+ f (𝛼)en2 + O en3
2! While eps > 0;
[ ( )]
= f (1) (𝛼) 1 + 2C2 en + 3C3 en2 + 4C4 en3 + O en4 (32) 𝜆 (k) = 𝜆 (k) ∕2
T = X (k) − 𝜆 (k) ΔX (k)
and dividing (31) by (32), eps = ‖F (T )‖2 − F 1
End
f (xn ) [ ( )]
= en + C2 en2 + C3 en3 + O en4 Output 𝜆 (k)
f (xn )
(1)

[ ( )]−1
∗ 1 + 2C2 en + 3C3 en2 + O en3
( ) ( )
= en − C2 en2 + 2C22 − 2C3 en3 + O en4 . (33) 1
If G = 1; and Gx + Gy (1 − 𝜃) = C2 , that is, 𝜃 = ,
3
It follows that f ′ (x)+ f ′ (y)
G (x, y) = , then (38) can be written as below,
−2 f ′ (x)+4 f ′ (y)
f (xn )
yn − 𝛼 = xn − 𝜃 (1) −𝛼 1
f (xn ) . en+1 = {2C3 − C22 − [𝜃GyC2 + Gxx
( ).
2
( ) ( ) 1
(39)
= (1 − 𝜃)en + 𝜃C2 en2 + 𝜃 2C3 − 2C22 en3 + O en4 +Gxy (1 − 𝜃) + Gyy (1 − 𝜃) ]} + O en4
2
2
(34)
Applying Taylor’s expansion again for the two-variable func- Therefore, the order of convergence of the iterative scheme
tion G (xn , yn ) around the point (𝛼, 𝛼) gives (26)–(29) is at least three. □
[ ]
G (xn , yn ) = G + Gx + Gy (1 − 𝜃) en The proposed iterative scheme (26)–(29) in processing the
[ ] energy flow of IEGS or NGS may incur oscillating iterations
1 1
+ 𝜃GyC2 + Gxx + Gxy (1 − 𝜃) + Gyy (1 − 𝜃)2 en2 if initial guess is not selected properly; thus a damping-factor
2 2
[ ( ) 𝜆 is introduced to solve this problem in case of the oscillating
+ 𝜃Gy 2C3 − 2C2 + Gxy 𝜃C2 + Gyy (1 − 𝜃)𝜃C2
2
(35) phenomenon, which is shown as (40)
1 1 1
+ Gxxx + Gxxy (1 − 𝜃) + Gxyy (1 − 𝜃)2
6 2 2 X (k+1) = X (k) − 𝜆 (k) ΔX (k) , (40)
1 3]( )
+ Gyyy (1 − 𝜃) en3 + O en4
6 where the process of damping-factor 𝜆 is shown in Table 1.
by (33) and (35)

f (x ) [ ] 4 SIMULATION STUDIES
G (xn , yn ) ′ n = G en + Gx + Gy (1 − 𝜃) − GC2 en2
f (xn )
[ ( ) In this section, eight simulated test systems are used to verify
+ G 2C22 − 2C3 − C2 Gx − C2 Gy (1 − 𝜃) + 𝜃GyC2 (36)
the proposed method. Those simulated test systems of EPSs,
1 1 2] ( )
+ Gxx + Gxy (1 − 𝜃) + Gyy (1 − 𝜃) en3 + O en4 NGSs and IEGSs are as below:
2 2
Case A: only, a modified 15-node NGS [34] is considered.
Thus Case B: only, a modified 20-node NGS [39] is considered.
Case C: only, a modified 48-node NGS [40] is considered.
f (xn ) Case D: only, a modified 130-node NGS which is based on
xn+1 = xn − G (xn , yn )
f ′ (xn ) 48-node NGS and 20-node NGS is considered.
f (xn ) Case E: only, a case24_ieee_rts available in MATPOWER of
⇒ en+1 + 𝛼 = en + 𝛼 − G (xn , yn ) (37)
f ′ (xn ) EPS is considered.
f (xn )
⇒ en+1 = en − G (xn , yn ) Case F: only, a case300 available in MATPOWER of EPS
f ′ (xn )
is considered.
[ ] Case G: The first IEGS consists of a case33bw-node EPS
en+1 = (1 − G )en − Gx + Gy (1 − 𝜃) − GC2 en2 available in MATPOWER and a 15-node NGS [34], considering
[ ( ) the renewable energy sources. Two small distributed-wind farms
− G 2C22 − 2C3 − C2 Gx − C2 Gy (1 − 𝜃) + 𝜃GyC2 (38) installed at bus 18 and 22, six small distributed-photovoltaic
1 1 2] ( ) generators at bus 15, 25, 26, 29, 31, and 33. Moreover, two P2G
+ Gxx + Gxy (1 − 𝜃) + Gyy (1 − 𝜃) en3 + O en4
2 2 units are installed at bus 18, 22 of EPS, and node 5, 14 of NGS.
ZHENG ET AL. 7

TABLE 2 Comparisons of energy flow results of the pressure (psia) at all


nodes

Node NDM MNM TNM JM ToCNR1 ToCNR2

1 0 0 0 0 0 0
2 0 0 0 0 0 0
FIGURE 2 Schematic of the 15-node NGS 3 0 0 0 1.14−13 0 0
4 0 0 0 0 1.14−13 1.14−13
5 0 0 0 1.14−13 1.14−13 1.14−13
The two compressors driven by electric power are connected at
6 0 0 0 2.27−13 2.27−13 2.27−13
bus 4 and 11.
Case H: The second IEGS consists of a case118-node EPS 7 0 1.14−13 1.14−13 1.14−13 1.14−13 1.14−13
available in MATPOWER and a 20-node NGS [39]. In this 8 0 2.27−13 2.27−13 2.27−13 2.27−13 2.27−13
case, four wind farms and four photovoltaic generators are 9 0 2.27−13 0 0 2.27−13 2.27−13
considered, which are installed at bus 12, 46, 54, 103, 25, 10 0 2.27−13 0 0 2.27−13 2.27−13
36, 40 and 49, respectively. Moreover, the four GF units are 11 0 2.27−13 0 0 2.27−13 2.27−13
installed at bus 25, 46, 54, 103 of EPS, and node 3, 12,
12 0 2.27−13 0 0 2.27−13 2.27−13
15, 20 of NGS. Three P2G units are set at bus 7, 23, 113
of EPS, and node 2, 5, 14 of NGS. The two compressors 13 0 2.27−13 1.14−13 2.27−13 4.55−13 4.55−13
driven by electric power are connected at bus 73 and 102 of 14 0 2.27−13 0 1.14−13 4.55−13 4.55−13
EPS. 15 0 2.27−13 0 1.14−13 3.41−13 3.41−13
The referred tested electricity network data of EPS above-
mentioned are all from the MATPOWER 7.1 version which can
be referenced in [41, 42].
Let Z be a root of the function F (X ) and suppose that parameters of the test system can be found in [34]. In this sub-
X (k+1) , X (k) , and X (k−1) are three consecutive iterations closer section, the initial point is the same with [34], but the compres-
to the root Z. Then, the computational order of convergence sors driven by natural gas are set to run at a constant com-
COC can be approximated using the formula [43]: pression ratio. The solution obtained by CNRM is used to be
the benchmark. Then, results obtained by different methods are
( ) ( )
⎡ ln || X k+1 − Z ∕ X k − Z || ⎤ compared with CNRM, that is,
COC ≈ MAX⎢ |( |⎥
⎢ ln || X k − Z)∕(X k−1 − Z)|| ⎥
(41)
⎣ | |⎦ 𝜀i = |MCNRM − Mi |, (42)

Here, our proposed method, that is, the variant of Newton– where 𝜀i is the mismatch of the ith method compared with
Raphson method with third-order convergence, in dealing the CNRM.
with the energy flow of (26)–(29) without the damping- The mismatch 𝜀i of the energy flow results of the pres-
factor 𝜆, called ToCNR1, that 𝜆 in (40) is 1. The pro- sure (psia) at all nodes and the gas flow of each pipeline
posed iterative scheme of (40) is named ToCNR2. Results (106 ×SCF/hr) can be seen from Tables 2 and 3. It can con-
obtained by ToCNR1 and ToCNR2 are compared with clude that the solution obtained by ToCNR1 and ToCNR2 are
those of classical Newton–Raphson method (CNRM) [36], the same with CNRM.
trapezoidal Newton’s method (TNM) proposed by Fernando
et al. [43], Newton–Downhill method (NDM) [44], mid-
point Newton’s method (MNM) [45], and Jarratt method 4.2 Comparisons of computational speed
(JM) [46]. Z in (41) is the final solution corresponding to and efficiency
the maximum convergence limit of the iterative calculation
process. In order to show the efficiency of the proposed method in solv-
ing the energy flow of heterogenous energy systems, the conver-
gence curves, computational time, and the computational order
4.1 Comparisons of computational accuracy of convergence COC are compared.
The convergence curves in different cases of NGSs, EPSs,
To verify the computational accuracy of the ToCNR1 and and IEGSs solved by abovementioned iterative methods is given
ToCNR2 method in solving the energy flow of the IEGSs, EPSs in Figure 3. Moreover, the computational order of convergence
and NGSs, the energy flow results solved by different meth- COC in all cases are presented in Table 4.
ods are compared. In this subsection, we only show the result In Table 4, the numerical results indicate that ToCNR1 and
of NGS, that is, the system structure and simulation parame- ToCNR2 are at least third order rate of convergence in solving
ters of the modified 15-node NGS [34] is introduced, and the the nonlinear systems of NGSs, EPSs and IEGSs. Furthermore,
schematic of the 15-node NGS is shown in Figure 2. Detailed the computational order of convergence COC is even more than
8 ZHENG ET AL.

FIGURE 3 The mismatch of each iteration for all cases


ZHENG ET AL. 9

TABLE 3 Comparisons of energy flow results of gas flow (106 ×SCF/hr) solving the energy flow of EPSs compared with ToCNR1 and
in each pipeline ToCNR2, which is no square-root in Jacobian matrix of EPSs.
From To NDM MNM TNM JM ToCNR1 ToCNR2 The iteration curves of ToCNR1 and ToCNR2 are the same
when there is no oscillation in each iteration. When oscil-
1 3 0 0 0 2.66−15 0 0
lation occurs, the iterations of ToCNR2 is much less than
2 4 0 0 0 0 8.88−16 8.88−16 ToCNR1 because of the damping-factor 𝜆 introduced into
3 4 0 0 0 1.62−14 8.22−15 8.22−15 ToCNR2.
3 5 0 0 0 5.33−15 1.78−15 1.78−15 As shown in Figure 3c, the number of iterations of CNRM
4 7 0 2.66−15 2.66−15 2.66−15 8.88−16 8.88−16 and TNM are more than 40 in solving the energy flow of NGSs.
ToCNR1 and ToCNR2 can significantly outperform above two
6 9 0 5.33−15 0 7.99−15 2.66−15 2.66−15
which only need 6 and 8, respectively, but it only need six itera-
8 11 0 0 1.78−15 1.78−15 1.78−15 1.78−15
tions in Case E and Case F of EPS, in which the flat start calcu-
10 13 0 0 8.88−16 1.78−15 8.88−16 8.88−16 lation method is used. Moreover, as can be seen from Figure 3d,
12 14 0 8.88−16 0 8.88−16 1.78−15 1.78−15 compared with Case A to Case C and Case G to Case H, the
13 14 0 0 3.06−14 3.06−14 0 0 order of convergence of JM is four, but its iterative efficiency
13 15 0 1.08−14 2.18−14 3.26−14 2.18−14 2.18−14 is not much better than ToCNR1 and ToCNR2. In larger test
system, like 130-node NGS, JM may even diverge, but the JM
14 15 0 1.37−14 0 1.37−14 2.74−14 2.74−14
outperforms ToCNR1 and ToCNR2 in solving the energy flow
5 6 0 1.78−15 1.78−15 8.88−16 7.11−15 7.11−15
of EPS.
7 8 0 8.88−16 8.88−16 8.88−16 8.88−16 8.88−16 For all cases, those four methods of NDM, MNM, ToCNR1,
9 10 0 8.88−16 1.78−15 8.88−16 8.88−16 8.88−16 and ToCNR2 methods are the most efficient and stable. In
11 12 0 8.88−16 0 1.78−15 5.33−15 5.33−15 terms of the convergence rate, we observe that the (max)
mismatch decreases higher of ToCNR1 and ToCNR2 com-
pared with NDM and MNM. Table 5 summarizes the num-
ber of iterations and runtime obtained by the NDM, MNM,
TABLE 4 The computational order of convergence COC in different ToCNR1, and ToCNR2 methods. As shown in the table,
cases obtained by various methods
ToCNR1 and ToCNR2 outperform NDM and MNM in terms
Method Case A Case B Case C Case D Case E Case F Case G Case H of the runtime and the number of iterations. Compared with
CNRM 2.012 - 2.026 - 2.533 5.841 - - ToCNR1, ToCNR2 is much superior in the number of itera-
tions, but there is no advantage in runtime if the number of
NDM 2.089 2.183 2.014 2.007 2.533 5.841 2.392 1.992
iterations is not too different from the two methods. This is
MNM 2.660 2.562 2.562 2.827 6.837 4.488 2.490 3.441
because judgment process with the damping-factor 𝜆 is intro-
TNM 3.243 - 2.961 - 4.734 4.431 - - duced in the iterative process, which spends more time to
JM 3.748 3.371 3.001 - 4.681 4.505 2.235 4.580 calculate the mismatch and update state variables of natural
ToCNR1 5.986 3.954 3.336 3.751 4.580 4.381 3.667 3.542 gas system; the detail of iterative process with the damping-
ToCNR2 5.986 3.293 3.336 4.026 4.580 4.884 3.667 3.542 factor 𝜆 is shown in Table 1. Moreover, from Figure 3 of
Case D and Table 5, the number of iterations and runtime
of ToCNR1 are 11 steps, 6.0334 s, respectively. However,
the number of iterations and runtime of ToCNR2 are nine
three in terms of some cases. We know that CNRM, NDM used steps and 5.7774 s, which concludes that ToCNR2 is much
in solving energy flow of non-linear system can reach second- superior.
order convergence, and MNM is at least third-order conver- In addition to the above analysis, to demonstrate the supe-
gence in theory [45] but it did not reach third-order convergence riority of the proposed methods of ToCNR1 and ToCNR2 in
in solving energy flow of NGSs, but EPSs and EGSs if the initial solving the large-scale energy flow in IEGS, specially in NGS,
point is select properly. Although the order of convergence of different initial guess of the pressure at all nodes are set to
JM is four [46], the computational order of convergence COC simulate analysis based on Case C. The schematic of the 48-
is not four-order in solving the energy flow of NGSs, which node NGS and the detailed parameters be found in [40]. The
contains square-root in Jacobian matrix of NGSs. However, number of iterations with tolerance of 10−10 solved by differ-
JM can reach four-order convergence in solving energy flow of ent methods are shown in Table 6. As shown in Table 6, we
EPSs. know that the initial guess of the pressure at all nodes (with
As shown in Figure 3a,b,g,h, the CNRM and TNM diverge no including the slack node) given from 50% to 150% of solu-
when the initial guess is the same with others. Moreover, tions, the number of iterations ToCNR1 and ToCNR2 do not
the max mismatch decreases rapidly after a few iterations of increase, obviously. Moreover, our proposed method has the
ToCNR1 and ToCNR2 compared with CNRM and above least number of iterations compared with NDM, MNM, and JM
ones, but from Figure 3e,f, the JM is much better in methods.
10 ZHENG ET AL.

TABLE 5 The required numbers of iterations and runtime by NDM, MNM, ToCNR1, and ToCNR2 methods for different cases

Case A Case B Case C Case D Case E Case F Case G Case H

Method Iter. Time(s) Iter. Time(s) Iter. Time(s) Iter. Time(s) Iter. Time(s) Iter. Time(s) Iter. Time(s) Iter. Time(s)

NDM 13 0.5943 15 0.5277 11 0.9526 15 7.2441 6 0.0094 6 0.0184 12 0.3997 8 0.3445


MNM 7 0.2096 17 0.5673 7 0.6676 21 11.5115 4 0.0086 5 0.0235 19 0.6192 6 0.2663
ToCNR1 5 0.1449 9 0.2901 8 0.7780 11 6.0334 4 0.0106 4 0.1093 8 0.2536 4 0.1901
ToCNR2 5 0.1901 8 0.3546 8 0.9570 9 5.7774 4 0.0114 4 0.1086 8 0.3298 4 0.2379

TABLE 6 The number of iterations and runtime with different initial guess of pressures at all nodes based on Case C

Method 50% 60% 70% 80% 90% 100% 110% 120% 130% 140% 150%

CNRM - - - 195 160 3 160 186 - - -


NDM 13 12 11 12 9 3 10 12 13 13 -
MNM 11 12 14 15 13 4 13 13 12 15 14
TNM - - 119 94 85 4 87 101 - - -
JM 12 11 11 12 11 4 11 11 12 12 12
ToCNR1 8 7 6 7 7 4 7 8 7 8 7
ToCNR2 8 7 6 7 7 4 7 7 7 8 7

5 CONCLUSION DATA AVAILABILITY STATEMENT


The authors declare that all the data in this paper are available.
Here, several cases are used to test proposed method, the
numerical results show that the number of iterations and ORCID
the computational time are significantly reduced under the J.H. Zheng https://orcid.org/0000-0003-4928-6524
same mismatch tolerance and initial guess, compared with Zhigang Li https://orcid.org/0000-0003-0464-8800
classical NR method and its existing variants. The proposed
method for energy flow calculation of IEGS is at least third- REFERENCES
order with one function and two derivative evaluations in 1. Li, Q., Gedra, T.: Solving natural gas loadflow problems using electric load-
each iteration. Moreover, the results also verify the efficiency flow techniques. (2003)
2. Martinez-Mares, A., Fuerte-Esquivel, C.R.: A unified gas and power flow
and superiority of the proposed method in solving energy
analysis in natural gas and electricity coupled networks. IEEE Trans. Power
flow of natural gas system compared with the classical NR Syst. 27(4), 2156–2166 (2012)
method and its other variants in terms of computational 3. Zheng, J.H., Wu, C.Q., Huang, J., Liu, Y., Wu, Q.H.: Multi-objective opti-
efficiency. mization for coordinated day-ahead scheduling problem of integrated
electricity-natural gas system with microgrid. IEEE Access 8, 86788–
86796 (2020)
ACKNOWLEDGEMENTS
4. Shabanpour-Haghighi, A., Seifi, A.R.: An integrated steady-state operation
This work was supported by the National Key Research assessment of electrical, natural gas, and district heating networks. IEEE
and Development Program of China (Grant No. Trans. Power Syst. 31(5), 3636–3647 (2016)
2018YFE0208400), the Science and Technology Project of 5. Zheng, J., Chen, J., Wu, Q., Jing, Z.: Multi-objective optimization and deci-
State Grid Corporation of China (Key Technologies of Novel sion making for power dispatch of a large-scale integrated energy system
with distributed DHCs embedded. Appl. Energy 154, 369–379 (2015)
Integrated Energy System Considering Cross-border Inter-
6. Chen, Y., Yao, Y., Lin, Y., Yang, X.: Dynamic state estimation for integrated
connection), the Natural Science Foundation of China (No. electricity-gas systems based on Kalman filter. CSEE J. Power Energy Syst.
52007066), the Natural Science Foundation of Guangdong 1(1), 1–11 (2020)
Province, China (No. 2021A1515010584), the Guangdong 7. Wang, L., Zheng, J., Li, M., Lin, X., Jing, Z., Wu, P., Wu, Q., Zhou,
Basic and Applied Basic Research Foundation under Grand X.: Multi-time scale dynamic analysis of integrated energy systems: An
individual-based model. Appl. Energy 237, 848–861 (2019)
(No. 2020A1515111100), and the Talent Recruitment Project
8. Zheng, W., Hou, Y., Li, Z.: A dynamic equivalent model for district heating
of Guangdong (Grant No. 2017GC010467). networks: Formulation, existence and application in distributed electricity-
heat operation. IEEE Trans. Smart Grid 12(3), 2685–2695 (2021)
9. Chen, Y., Yao, Y., Zhang, Y.: A robust state estimation method based
CONFLICT OF INTEREST on SOCP for integrated electricity-heat system. IEEE Trans. Smart Grid
The authors have declared no conflict of interest. 12(1), 810–820 (2021)
ZHENG ET AL. 11

10. Li, Z., Li, Y., Liu, Y., Wang, P., Lu, R., Gooi, H.B.: Deep learning based 30. Meng, Q., Guan, Q., Jia, N., Zhang, L., Wang, Y.: An improved sequen-
densely connected network for load forecasting. IEEE Trans. Power Syst. tial energy flow analysis method based on multiple balance nodes in
36(4), 2829–2840 (2021) gas-electricity interconnection systems. IEEE Access 7, 95487–95495
11. Tinney, W.F., Hart, C.E.: Power flow solution by Newton’s method. IEEE (2019)
Trans. Power Apparatus Syst. PAS-86(11), 1449–1460 (1967) 31. Chen, Y., Zhao, J., Ma, J.: Fast decoupled multi-energy flow calculation for
12. Stott, B.: Decoupled Newton load flow. IEEE Trans. Power Apparatus integrated energy system. J. Modern Power Syst. Clean Energy 8(5), 951–
Syst. PAS-91(5), 1955–1959 (1972) 960 (2020)
13. Trias, A.: The holomorphic embedding load flow method. In: 2012 IEEE 32. Zeng, Q., Fang, J., Li, J., Chen, Z.: Steady-state analysis of the integrated
Power and Energy Society General Meeting, pp. 1–8. IEEE, Piscataway, natural gas and electric power system with bi-directional energy conver-
NJ (2012) sion. Appl. Energy 184, 1483–1492 (2016)
14. Xu, J.S.T.L.W., Liu, Y., Chang, G.: Series load flow: A novel non-iterative 33. Osiadacz, A.: Simulation and Analysis of Gas Networks. Gulf Publishing
load flow method. IEEE Proc. Gener. Transm. Distrib. 145(5), 251–256 Company, Houston, TX.
(1998) 34. An, S.: Natural gas and electricity optimal power flow. Ph.D. thesis, Okla-
15. deSouza, A.Z., Junior, C.R., Lopes, B.I.L., Leme, R., Carpinteiro, O.: Non- homa State University.
iterative load-flow method as a tool for voltage stability studies. IET 35. Zheng, J., Wu, Q., Jing, Z.: Coordinated scheduling strategy to optimize
Gener., Transm. Distrib. 1(6), 499–505 (2007) conflicting benefits for daily operation of integrated electricity and gas net-
16. Chiang, H.-D., Zhao, T.-Q., Deng, J.-J., Koyanagi, K.: Homotopy-enhanced works. Appl. Energy 192, 370–381 (2017)
power flow methods for general distribution networks with distributed 36. Liu, C., Shahidehpour, M., Fu, Y., Li, Z.: Security-constrained unit commit-
generators. IEEE Trans. Power Syst. 29(1), 93–100 (2014) ment with natural gas transmission constraints. IEEE Trans. Power Syst.
17. Rao, S., Feng, Y., Tylavsky, D.J., Subramanian, M.K.: The holomorphic 24(3), 1523–1536 (2009)
embedding method applied to the power-flow problem. IEEE Trans. 37. Jentsch, M., Trost, T., Sterner, M.: Optimal use of power-to-gas energy
Power Syst. 31(5), 3816–3828 (2016) storage systems in an 85% renewable energy scenario. In: 8th International
18. Abbasi, A.R.: Probabilistic load flow based on holomorphic embedding, Renewable Energy Storage Conference and Exhibition, pp. 85254–261.
kernel density estimator and saddle point approximation including corre- (IRES 2013)
lated uncertainty variables. Electr. Power Syst. Res. 183, 106178 38. Wang, X., Liu, X.: A conditional optimal two-step iterative method and a
19. Liu, C., Bak, C.L., Zhu, Y., Sun, K.: Analytical solutions for power flow variant of Jarratt method. J. Yanbian University: Natural Sci. 43(4), 314–
equations based on the multivariate quotient-difference method. IEEE, 320+349 (2017 (in Chinese))
pp. 1–6. IEEE, Piscataway, NJ (2019) 39. DeWolf, D., Smeers, Y.: The gas transmission problem solved by an exten-
20. Wang, T., Chiang, H.-D.: On the holomorphic and conjugate properties sion of the simplex algorithm. Manage. Sci. 46, 1454–1465 (2000)
for holomorphic embedding methods for solving power flow equations. 40. Wu, S., Rĺłos-Mercado, R., Boyd, E., Scott, L.: Model relaxations for the
IEEE Trans. Power Syst. 35(4), 2506–2515 fuel cost minimization of steady-state gas pipeline networks. Math. Com-
21. Liu, C., Sun, K., Wang, B., Ju, W.: Probabilistic power flow analysis using put. Modell. 31(2), 197–220 (2000)
multidimensional holomorphic embedding and generalized cumulants. 41. Zimmerman, R.D., Murillo-Sĺćnchez, C.E., Thomas, R.J.: Matpower:
IEEE Trans. Power Syst. 33(6), 7132–7142 Steady-state operations, planning, and analysis tools for power systems
22. Zhang, T., Li, Z., Zheng, J., Wu, Q., Zhou, X.: Power flow analysis of inte- research and education. IEEE Trans. Power Syst. 26(1), 12–19 (2011)
grated gas and electricity systems using the fast and flexible holomorphic 42. Zimmerman, R.D., Murillo-Sĺćnchez, C.E., Thomas, R.J.: Matpower (ver-
embedding method. In: 2020 IEEE Power Energy Society General Meet- sion 7.1) [software]. https://matpower.org
ing (PESGM), pp. 1–5. IEEE, Piscataway, NJ (2020) 43. Weerakoon, S., Fernando, T.: A variant of Newton’s method with acceler-
23. Chiang, H.-D., Wang, T., Sheng, H.: A novel fast and flexible holomorphic ated third-order convergence. Appl. Math. Lett. 13(8), 87–93 (2000)
embedding power flow method. IEEE Trans. Power Syst. 33(3), 2551– 44. Yang, H., Wen, F., Wang, L., Singh, S.N.: Newton-downhill algorithm for
2562 (2018) distribution power flow analysis. In: 2008 IEEE 2nd International Power
24. Ding, T., Xu, Y., Yang, Y., Li, Z., Zhang, X., Blaabjerg, F.: A tight linear and Energy Conference, pp. 1628–1632. IEEE, Piscataway, NJ (2008)
program for feasibility check and solutions to natural gas flow equations. 45. Özban, A.: Some new variants of Newton’s method. Appl. Math. Lett.
IEEE Trans. Power Syst. 34(3), 2441–2444 (2019) 17(6), 677–682 (2004)
25. Jia, W., Ding, T., Huang, C., Wang, Z., Zhou, Q., Shahidehpour, M.: Convex 46. Jarratt, P.: Some fourth order multipoint iterative methods for solving
optimization of integrated power-gas energy flow model with applications equations. Math. Comput. 20(95), 434–437 (1966). http://www.jstor.org/
to probabilistic energy flow. IEEE Trans. Power Syst. 36(2), 1432–1441 stable/2003602
(2021)
26. Massrur, H.R., Niknam, T., Aghaei, J., Shafie-khah, M., Catal?o, J.P.S.:
Fast decomposed energy flow in large-scale integrated electricity–gas–heat
energy systems. IEEE Trans. Sustainable Energy 9(4), 1565–1577 (2018)
How to cite this article: Zheng, J.H., Wu, C.Q.,
27. Seungwon, A., Qing, L., Gedra, T.W.: Natural gas and electricity optimal
power flow. In: 2003 IEEE PES Transmission and Distribution Confer- Xiahou, K.S., Li, Z., Wu, Q.H. A variant of
ence and Exposition (IEEE Cat. No.03CH37495), Vol. 1, pp. 138–143. Newton–Raphson method with third-order
(2003) convergence for energy flow calculation of the
28. Liu, X., Wu, J., Jenkins, N., Bagdanavicius, A.: Combined analysis of elec- integrated electric power and natural gas system. IET
tricity and heat network. Appl. Energy 162, 1238–1250 (2016)
Gener. Transm. Distrib. 1–11 (2021).
29. Li, J., Huang, Y., Zhu, M.: Gradient descent iterative method for energy
flow of integrated energy system considering multiple modes of compres- https://doi.org/10.1049/gtd2.12298
sors. Energy Conv. Manage. 207, 112534 (2020)

You might also like