You are on page 1of 20

Page 1 of 20 IET Renewable Power Generation

This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

Modeling and Solutions of Coordinated Economic Dispatch with


Wind-Hydro-Thermal Complex Power Source Structure
Li-Bao Shi 1*, Ren Wang 1, Liang-Zhong Yao 2
1
Graduate School at Shenzhen, Tsinghua University; National Key Laboratory of Power
System in Shenzhen, Shenzhen 518055, P. R. China
2
China Electric Power Research Institute, Beijing 100192, P. R. China.
*
E-mail: shilb@sz.tsinghua.edu.cn

Abstract: This paper presents an efficient optimization strategy for solving coordinated economic
dispatch problem with wind-hydro-thermal complex power source structure. The wind-hydro-thermal
coordinated dispatch aims to minimize the total fuel costs of coal-fired thermal power units while
satisfying all kinds of operating constraints. In order to better handle the random variables in the
constraints introduced by wind power and load demand during analysis, a probabilistic analytical model is
employed to describe the uncertainty of wind farm power output firstly; and then an improved convolution
method is applied to calculate the total stochastic power consisting of load demand and power output of
wind farm. The two-stage stochastic linear programming method and stochastic chance constraints are
employed to further form a new deterministic objective function with penalty items taken into account. An
enhanced particle swarm optimization method is applied in the solution of the proposed model. Finally,
the simulations are performed on a 6-bus test system and a real-sized China power grid test system to
investigate the effectiveness and validity of the proposed optimization strategy.

1. Introduction
With the depletion of the fossil fuel supplies and the increasing concerns for the global
environmental problem, more and more attentions have been paid to the environment-friendly renewable
energy resources like wind energy, solar energy etc. Because of its mature technology, low cost
characteristics, the wind energy has becomes one of the fastest growing energy sources. While with the
gradually increasing of wind power penetration, its uncertainty has brought new problems and challenges
to power system operation [1][2]. In China, the coal-fired thermal power is dominant in generating
capacity, which accounts for 67.18% of total generating capacity of electric power [3]. In general, the
coal-fired thermal power generation is applied to accommodate the volatility of wind power. The
consequence is that the coal-fired thermal power units have to be committed frequently and will lead to a
rise in carbon emissions. In this situation, the medium or large scale hydro power plants need to be
committed to stabilize the volatility of wind power.
Currently, more and more interests and achievements [4-18] have intended and focused on how to
accommodate the intermittent wind power through dispatching conventional power sources elaborately. In
the existing economic dispatch models incorporating wind power, how to mathematically simulate the
uncertainties of wind power becomes the key points and difficulties. The concept of dynamic uncertainty
sets [4], the probabilistic wind power forecasting technique [5], the factors to account for both

IET Review Copy Only


1
IET Renewable Power Generation Page 2 of 20
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

overestimation and underestimation of available wind power [6][7], the scenario generation and reduction
technologies [8], the fuzzy model of wind power [9] and the chance constrained formulation [10] are
introduced to model the temporal and spatial characteristics of wind power uncertainty. In order to solve
the coordinated economic dispatch (CED) problem with wind-hydro-thermal complex power source
structure, many algorithms involving linear programming [11], interior point method [12], the class of
sampling-based decomposition algorithms [13], robust optimization method under uncertainty [4][14] as
well as the computational intelligence approaches including particle swarm optimization [15][16], genetic
algorithm [17], bacterial foraging method [18] etc have been employed for the solution of wind-hydro-
thermal ED issue. So far, most research works mainly focus on solution method itself. The coordinated
behavior or function among coal-fired thermal, hydro and wind plants has not been fully reflected during
analysis. In addition, how to comprehensively explore and exploit the uncertainties of wind power, load
and water inflow with complex power source structure still need to be studied further.
In this paper, an optimization strategy is proposed to implement the CED issue with wind-hydro-
thermal complex power source structure. In order to simulate the uncertainties of wind power and load, a
joint probabilistic analytical model is deduced via an improved convolution technique. For the built wind-
hydro-thermal CED model, an enhanced particle swarm optimization (EPSO) method combined with the
two-stage stochastic linear programming method is applied to implement the coordinate dispatch with
multiple power sources.

2. Formulation of coordinated economic dispatch

2.1. Original objective function

The optimization objective is to minimize the total fuel cost of all coal-fired thermal power units
while satisfying a series of equality and inequality constraints. The original objective function can be
represented as:
T
Min Fori = ∑∑ (ai + bi Ph,i (t ) + ci Ph2,i (t )) (1)
t =1 i∈H

where Fori is the total fuel cost; Ph,i(t) is power output of the ith coal-fired thermal generator in the hth time
period; ai, bi, and ci are the cost coefficients of the ith coal-fired thermal generator; T is the last time period;
H is the set of all coal-fired thermal generators.
2.2. Equality and inequality constraints

a) System Power Balance

2
IET Review Copy Only
Page 3 of 20 IET Renewable Power Generation
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

∑P
i∈H
h ,i (t ) + ∑P
i∈HY
hy ,i (t ) + ∑ P%wf (t ) = ∑ P%d ,i (t )
i∈D
(2)

where subscripts h, hy, wf and d represent coal-fired thermal power, hydropower, wind power and load,
respectively.
b) Reservoir Storage
C i ≤ C i (t ) ≤ C i (i ∈ I ) (3)
CiT ≤ Ci (T ) (i ∈ I ) (4)
_
where Ci, Ci denote the upper and lower limits of water storage in the ith reservoir, respectively; CiT is the
lower limit of water storage in the ith reservoir at the last time period; I is the set of total reservoirs.
c) Water Discharge
0 ≤ Rk , j (t ) ≤ Rk , j (k ∈ K j , j ∈ I ) (5)
_
where Rk,j is the upper limit of water discharge of the kth hydro generator in the jth reservoir.
d) Thermal Power Output
Ph , i ≤ Ph , i (t ) ≤ Ph , i i∈ H (6)
_
where Ph,i, Ph,i denote the upper and lower limits of the coal-fired thermal power output, respectively.
e) Ramp rate
-ηi ∆T ≤ Ph,i (t +1) − Ph,i (t) ≤ ηi ∆T (i ∈ H , t = 1,2,L, T -1) (7)
_
where η, η are the ramp up rate limit and ramp down rate limit of the ith coal-fired thermal generator; ∆T
is the ramp time.
f) Network security
 ∑ Dl ,i ⋅ Ph ,i (t ) + ∑ Dl ,i ⋅ Phy ,i (t ) + Dl , wf ⋅ P%wf (t ) − ∑ Dl ,i ⋅ P%d ,i (t ) ≤ Pl
i∈TH i∈HY i∈D
 (l ∈ L) (8)
 ∑ Dl ,i ⋅ Ph ,i (t ) + ∑ Dl ,i ⋅ Phy ,i (t ) + Dl , wf ⋅ Pwf (t ) − ∑ Dl ,i ⋅ Pd ,i (t ) ≥ Pl
% %
i∈TH i∈HY i∈D

Where L is the set of all branches; Dl,i is the active power coefficient of the lth power line corresponding to
_
the nodes of wind power, coal-fired thermal power and hydropower; Pl, Pl denote the upper and lower
limits of the lth power line, respectively.

2.3. Modeling of hydropower plant

2.3.1 Modeling of water storage

3
IET Review Copy Only
IET Renewable Power Generation Page 4 of 20
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

Hydropower is considered as a kind of renewable, clean and efficient resource. However, the
generated electric power is limited to the reservoir storage and water discharge. In this paper, the multiple
hydropower plants and cascade reservoirs are involved during modeling. Taken time delay among
hydropower plants into account, the corresponding representation utilized to model hydropower plant is
shown as follows:
Ci (t ) = Ci (t −1) + ∑ R (t − d ) − ∑ R (t ) + W% (t)
k∈M i
k k
j∈Ni
j i (i ∈ I )
(9)

where Ci(t-1), Ci(t) are water storages of the ith reservoir at time periods t-1 and t; Rj(t) is the total water
~
release of the jth hydropower unit at the tth time period; Wi is the inflow of the ith reservoir at the tth time
period, which is subject to the normal distribution; Mi is the set of those hydropower plants that release
water to the ith reservoir; Ni is the set of those hydropower plants that intake water from the ith reservoir;
Rk(t- dk) is the water flowing from hydropower plant k to the downstream reservoir, dk is the time water
flowing to the downstream reservoir.
Ci(t-1) can be expressed using initial value of water storage. Thus, Ci(t) can be expressed through
following recursion:
t  
Ci (t ) = Ci (t0 ) + ∑ ∑ ∑ R k, j (t − d j ) − ∑ ∑R k, j (t ) + W%i (t1 )  (i ∈ I ) (10)
t1 =t0 +1  j∈Mi k∈K j j∈Ni k∈K j 

where Ci(t0) is the initial value of water storage; K is the set of hydropower unit in the jth hydropower
plant.
2.3.2 Water consumption characteristics
In general, the water consumption characteristics of most hydropower units are nonlinear and hard to
be modeled. In this paper, the change of water head is ignored. A kind of linear model as shown in (11) is
employed to represent the water consumption characteristics approximately:
Pk , j (t ) = η k , j (t ) ⋅ qk , j (t ) (11)

where ηk,j(t) is the efficiency coefficient of the kth unit of the jth hydropower plant at the tth time period.

3. Joint probabilistic analytical model considering wind power and load uncertainties
Firstly, an ad-hoc probabilistic analytical model [19] is applied to describe the uncertainty of wind
farm power output. In order to better reflect the effects of wind power and load uncertainties
simultaneously, in accordance with the convolution method and combined with the normal PDF of load, a
joint probabilistic analytical model for describing the uncertainties of wind farm power output and load
demand is deduced.
3.1. Probabilistic analytical model of wind power output

4
IET Review Copy Only
Page 5 of 20 IET Renewable Power Generation
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

On the basis of the unit impulse function, combined with the power output vs wind speed curve and
the wind speed PDF, the PDF of wind farm power output can be deduced analytically as shown in (12).
ϕ P ( Pwf ) =
0 Pwf < 0, Pwf > Pwfr
 Φ (0) ⋅ δ ( P ) Pwf =0
 P wf



 1 [ Pwf ( v 3 − v 3 ) + v 3 ]− 3 ⋅ ( vr − vci ) ⋅
2 3 3

3 P r ci ci
Pwfr
 wfr

  
k −1
k 1 P
  3 wf ( vr3 − vci3 ) + vci3  ⋅ 0< Pwf < Pwfr
 λ  λ Pwfr 


   P  
k

 exp  −  1 3 wf ( v r3 − vci3 ) + vci3  


   λ Pwfr  
    

 Pwf = Pwfr
 Φ P ( Pwfr ) ⋅ δ ( Pwf − Pwfr )
(t = 1L T ) (12)
3.2. Joint probabilistic analytical model
In our work, the normal distribution is applied to model the uncertainty of load [20]. According to
the probability theory, if the independent random variables are normally distributed, then their sum will
also be normally distributed. Therefore, the PDF of loads that follow normal distribution with different
parameters can be expressed as:
1  ( P − µ − µ L − µn )2 
ϕ D ( PD ) = exp  − D 2 1 2 2  (13)
2π (σ + σ L + σ )
1
2 2
2 
2
n
2(σ 1 + σ 2 L + σ n2 ) 

where PD is the load value; µ1 to µn are means of different loads; σ21 to σ2n are variances of different loads.
Thus, the PDF of loads can always be expressed by formula (13) with the appropriate assumptions and
simplifications.
In the power system, the power balance should always be satisfied. When the wind power is
integrated into the existing power grid, the system power balance can be expressed as:
PG =P%D − P%wf = P%all (14)
where PG denotes the total power output from traditional power sources like the coal-fired thermal power,
~ ~
hydropower etc. PD and Pwf represent the load demand, power output of wind farm respectively, and they
~
are random variables. Pall is the defined total stochastic power.
According to the relevant knowledge of probability theory, the PDF of the sum of two independent
~
random variables is equal to their convolution [21]. Therefore, the PDF of the total stochastic power Pall
can be calculated as
ϕall ( Pall ) = ϕD ( PD ) ∗ ϕP (− Pwf ) (15)
Substitute (12) and (13) into (15), the PDF of the total stochastic power can be deduced as:

5
IET Review Copy Only
IET Renewable Power Generation Page 6 of 20
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.
ϕall ( Pall ) = ϕD ( PD ) ∗ ϕ P (− Pwf )
= Φ P (0) ⋅ ϕ D ( Pall ) + Φ P ( Pwfr ) ⋅ ϕ D ( Pall + Pwfr ) (16)
0 1  ( Pall + Pwf − µ )2  1 vr3 − vci3
+∫ exp  −  ⋅ ⋅ ⋅
− Pwfr
σ 2π  2σ 2
  3 Pwfr
k −1
− Pwf −
2
k1 − Pwf 
(3 (vr3 − vci3 ) + vci3 ) 3 ⋅  3 (vr3 − vci3 ) + vci3  ⋅
Pwfr λ  λ Pwfr 

  − Pwf 
k

1
exp  −  3 (v − v ) + v 
3 3 3  dPwf
  λ Pwfr
r ci


ci

 

In formula (16), the first and second items can be analytically expressed. However, the third item is
hard to be analytically expressed in theory. Here, we apply the following multi-dimensional Gaussian
function to fit the third item with enough accuracy (coefficient of determination equals to 1 when n=6):
 (P − b )2   (P −b )2   (P −b )2 
a1 ⋅ exp − all 1  + a2 ⋅ exp − all 2  +L+ an ⋅ exp − all n  (17)
 c1   c2   cn 

The corresponding curve of the proposed model is given in Fig.1.


8

6
Probability density

0
0 0.2 0.4 0.6 0.8
Power output(p.u)

Fig. 1. Joint PDF of the total stochastic power with consideration of wind farm power output and load uncertainties

3.3. Comparisons with numerical simulation method


In this paper, the numerical simulation method combined with the Monte Carlo technique and the
discrete convolution method [22] is applied for comparisons to demonstrate the effectiveness of the
proposed model. Fig. 2 shows the results of proposed joint probabilistic analytical model and the
numerical simulation method with different sample sizes.
From Fig.2, it can be seen that when the sample size is set to be 100000, the curves from the
proposed analytical model and the numerical simulation are matched very well. Table 1 gives the
corresponding CPU time.

6
IET Review Copy Only
Page 7 of 20 IET Renewable Power Generation
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

Fig. 2. Comparisons of proposed model and numerical simulation method with different sample sizes

Table 1 CPU time applying proposed joint probabilistic analytical model and numerical simulation method with different
sample sizes
Numerical simulation method
Proposed analytical method
n=100 n=1000 n=10000 n=100000
19s 19s 23s 28s 0.2s

4. Treatment strategies for constraints

4.1. Treatment for equality constraint through two-stage stochastic linear programming

In our work, the two-stage stochastic linear programming method [23] will applied to deal with the
equality constraint including random variables.
4.1.1 Two-stage Stochastic Linear Programming
Consider the stochastic linear programming problem as given below:
Min {c x Ax = b%, x ≥ 0}
T
(18)
~
where Ax = b% is an equality with a random variable b and its distribution function is known.
It is impossible to find a deterministic value to satisfy the equality constraint before determining the
values of random variables. In other words, there must exists a certain deviation when the optimum is
substituted into the equality constraint. Here, a new variable y and the relevant matrix B are introduced to
represent the deviation, i.e.
Ax* + By = b* , y ≥ 0 (19)

7
IET Review Copy Only
IET Renewable Power Generation Page 8 of 20
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

In (19), the item By denotes the degree of deviation due to the violation of constraints for x* in the
first stage. The penalty item of the deviation is represented by dTy. Then the (18) can be converted to a
deterministic linear programing problem through minimizing the penalty item dTy as shown in (20) in the
second stage.
 Min d T y
 (20)
S.T . Ax + By = b
* *

 y≥0

~
Because the random variable b bears the pre-known probability distribution, the sample value of b
~
can be used to replace its actual value. Thus, the optimal solution becomes the function of x and b, which
~
can be defined as S(x, b). In this way, the two-stage stochastic linear programming problem mentioned
above can be converted to a deterministic programming problem.


( ( ))
Min cT x + E S x, b%
(21)
S.T. x ≥ 0

Error will exist in the original equality when the fixed value b* is replaced by arbitrary numbers.
Positive and negative deviation vectors y + = ( y1+ , y2+ ,L , ym+ )T and y − = ( y1− , y2− ,L , ym− )T are introduced to
describe the error.
b − Ai x bi − Ai x ≥ 0 (22)
yi+ =  i
0 bi − Ai x < 0

 A x − bi bi − Ai x ≤ 0 (23)
yi− =  i
0 bi − Ai x > 0

where Ai is the vector of the ith column of A.


For the solution of (21), it can be equivalent to the following mathematical programming problem:
  m
− 
 Min c x + E  Min ∑ ( f i yi + g i yi ) 
T +

  i =1  (24)
 + −
S.T . Ax + y − y = b%
 x, y + , y − ≥ 0



where fi, gi are penalty coefficients of positive and negative deviation vectors. In our work, the two
coefficients reveal the corresponding physical meanings of the power shortage and power surplus.
According to formulas (22) and (23), the positive and negative values describe the deviation between the
actual equivalent load and planned power output. This can also be explained as the amount and probability
of the actual demand of spinning reserves. Thus, the greater the value, the higher the cost of invoking the
spinning reserves. In other words, the value of penalty coefficients reflects the degree of difficulty in

8
IET Review Copy Only
Page 9 of 20 IET Renewable Power Generation
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

invoking the spinning reserves. Thus, the insufficient of the positive and negative spinning reserves may
lead to the increase of the coefficients fi and gi. The range of the coefficients can be obtained through local
actual price and experience. Besides, the value of coefficients can also be seen as the incentive to spinning
reserves. If the decision makers hope to reduce the possibility of wind curtailment in actual situations,
consequently, the penalty coefficient fi is set to be larger. In this paper, to make the model more realistic
and for the purpose of simplified analysis, we refer to the price of spinning reserves and set the penalty
coefficients as constant values.
~
F b% ( y ) is the cumulative distribution function of random variable b. According to the knowledge of

stochastic linear programming, formula (25) is applicable to any value of y:


y y
∫−∞
t ⋅ dFb% (t ) = Fb% ( y ) ⋅ y − ∫ Fb% (t ) dt
−∞
(25)

Applying the method of formula (25) to the expectation in formula (24), (24) can be finally
converted to the form of formula (26):

(( f + g ) ∫ )
m

Min c x + ∑
Ai x
T
i i Fb% (t ) dt + f i E (b%i ) − f i Ai x (26)
 i =1
−∞ i

S.T. x ≥ 0

E (b%i ) is a constant term when Fb% ( y ) is determined.

The power balance equality constraint includes the wind-load joint probabilistic model, and its
deviation can be calculated through formula (26).
4.1.2 Augmented objective function

Ph ,i

1 Ph(2)
,i

(1)
Ph ,i
Ph ,i

Power output (MW)

Fig. 3. Piece-wise linear approximation


In order to apply for the mature stochastic linear programming method, the objective function as
shown in (1) needs to be reconstructed to a linear formulation. In our work, the piece-wise linear
approximation method is employed to rebuild the objective function as shown in following:

9
IET Review Copy Only
IET Renewable Power Generation Page 10 of 20
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.
T K
Min F = ∑∑∑ bik Ph ,ik (t ) (27)
t =1 i∈H k =1

where K is the segment number; Fig.3 is the fuel consumption curve with 3 sections.
For the proposed original model as described in (27), and in accordance with (26), the augmented
objective function with penalty items can be obtained.
T  K  K 
Min Z = ∑  ∑∑ bik ⋅ Ph ,ik (t ) − ft ⋅  ∑∑ Ph,ik (t ) + ∑ ∑ Phy ,ik (t )  + (28)

t =1  i∈H k =1  i∈H k =1 i∈HY k ∈Ki 
~
 3 
 ∑∑ Ph ,ik ( t ) + ∑ ∑ Phy ,ik ( t )  
ft ⋅ E ( Pall (t )) + ( ft + g t ) ⋅ ∫  i∈H k =1  Ft ( y ) dy 
i∈HY k∈Ki

−∞ 

where Ft(y) is the probability distribution function of random variable ∑ P%d ,i (t ) − P%wf (t ) . ft and gt are penalty
i∈ D

coefficients of deviation vectors, which have the same meanings as explained for fi and gi in Eq (24). The
first item on the right side of the formula (28) is the rebuilt total fuel cost using the piece-wise linear
approximation method. The rest items can be seen as penalty of power imbalance caused by system
randomness.
4.2. Treatment for inequality constraints through stochastic chance constraints
~
Consider the following inequality with random variable b:

ax ≥ b% (29)
To make (29) remain satisfied with a certain probability α, it can be written as:
Φ ( ax ≥ b% ) ≥ α (30)
Equation (30) can be converted to the deterministic inequality through the corresponding theory of
stochastic chance constraints as given in following.
 Φ ( ax ≥ b% ) ≥ α ⇔ ax ≥ Bα
 (31)
 Φ ( ax ≤ b% ) ≥ α ⇔ ax ≤ B1−α

where Bα is the one-sided quantile of the probability α.


Regarding the reservoir storage constraints (t≠T), it can be converted to the following deterministic
representations:
 t  
 ∑  ∑ R j (t1 − d j ) − ∑R j (t1 )  ≤ Fi1−1 (1 − α i ) (32)
 t1 = t0 +1  j∈M i j∈ N i 
 t (i ∈ I )
  
 ∑  ∑ R j (t1 − d j ) − ∑R j (t1 )  ≥ Fi 2−1 (α i )
 t1 = t 0 +1  j∈ M i j∈ N i 

Similarly, the other constraints with random variables can be treated in the same way.

10
IET Review Copy Only
Page 11 of 20 IET Renewable Power Generation
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

5. Solution methodology
From (28), it can be seen that the proposed model bears non-convex and highly nonlinear
characteristics. The conventional mathematical programming techniques are hard to be applied to solve
this issue. In this paper, an enhanced particle swarm optimization (EPSO) method is designed to obtain the
optimal solution. Considering that the regulation speed of coal-fired thermal power is too slow to
accommodate the fast fluctuation of wind power. In our work, a hydropower priority based strategy to
restrain the fluctuation of wind power is proposed during applying EPSO method.
5.1. Strategy for conventional units to restrain the fluctuation of wind power
The regulation speed of hydropower is generally much more quickly than that of the coal-fired
thermal power. The specific strategies for hydropower to stabilize the fluctuation of load can be simply
summarized as: the variation of hydropower must be consistent with the variation of wind-load, and the
variation of hydropower must be less than the variation of wind-load, i.e.
∆Phy
0≤ ≤1 (33)
∆Pall

where ∆Phy, ∆Pall are the variations of hydropower and wind-load, respectively.
Under the conditions of multi-hydropower units, these hydropower units which have bigger
efficiency coefficient will be assigned more load tasks. The remaining tasks will be undertaken by the
coal-fired thermal power units with faster regulation speed.
5.2. Reserve capacity selection
In general, the actual power output of wind farm may be very different from the forecasting value.
To stabilize the fast fluctuation of wind power, the hydropower should be considered firstly. Therefore, a
certain reserve capacity should be given to the hydropower units. In our work, the hydropower must meet
the following inequality constraints:
 Phy _ max − α ⋅ Ptl _ max_ t +1 + α ⋅ Pth _ t
 Phy _ t +1 < min( , Phy _ max )
 1−α

Phy _ min − β ⋅ Ptl _ min_ t +1 + β ⋅ Pth _ t (34)
P > max( , Phy _ min )
 hy _ t +1 1− β

where Phy_t , Phy_t+1 are the power outputs of all hydropower units at time period t and t+1; Pth_t is the
power output of all coal-fired thermal power units at time period t; Phy_max and Phy_min denote the upper and
lower limits of hydropower output, respectively; Ptl_max_t+1 and Ptl_min_t+1 denote the upper and lower limits
of the total wind-load power at time period t+1; α and β are the variation coefficients of the total wind-
load variation (mainly refer to the wind power output variation).

11
IET Review Copy Only
IET Renewable Power Generation Page 12 of 20
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

If the variations of hydropower have to be consistent with variations of wind-load, and the
hydropower outputs go beyond this range, the excess part will be distributed according to the available
output capacities of coal-fired thermal power units.
5.3. Global optimization strategy
The flowchart of the solution for the proposed model based on the EPSO method is given in Fig. 4.
The first step is to initialize the velocity of particles, historical best positions pbest and the global optimal
position gbest. The next step is to generate a group of new particles and meet the solution space. Strategies
for conventional units to restrain the variation of wind-load power and reserve capacity calculation are
applied after that. pbest and gbest will be updated after each iteration. The program will stop at the
maximum number of iterations.
Initialize the velocity of particles

Initialize historical best positions pbest and


the global optimal position gbest

Generate a group of new particles

Compute the margin of power increasing and


decreasing using constraints of ramp rate and
power output

No
Meet the conditions for units
to track the variation of
equivalent load power?
Redistribute the
mismatched load
Yes according to the
strategy
Calculate reserve capacity of each
hydropower unit according to the strategy

No
Satisfy the upper and lower limit
of hydropower?

Redistribute the
Yes mismatched load

Update history best solution and


global optimal position

No
Reach iteration limit?

Yes

END

Fig. 4. The solution for the proposed model based on the EPSO method

6. Application example
Two case studies are carried out on the MATLABTM environment to investigate the effectiveness and
validity of the proposed optimization strategy. All simulations were run in a computer with Intel(R)
Core(TM) i5-3230M CPU @2.60GHz processor and 4GB of RAM memory.
6.1. Case study 1

12
IET Review Copy Only
Page 13 of 20 IET Renewable Power Generation
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

The 6-bus test system is depicted in Fig.5. In this case, ft, gt are set to be 2 and 1 respectively. α, β
are set to be 0.6. The rating of wind farm power output is 0.2 (the base value is 100MW). The number of
particles is 30 and the iteration number is 100.

Fig. 5. One line diagram of the 6-bus test system


Fig.6 shows the convergence process of the objective function value. The objective function value
decreases from 3.077 to 3.011 through optimization.
3.08

3.07

3.06
Objective function value

3.05

3.04

3.03

3.02

3.01
0 10 20 30 40 50 60 70 80 90 100
Iteration times

Fig. 6. Convergence process of the objective function value


The simulation results are given in Table 2 and in Fig.7.
Compared with the coal-fired thermal power units, the hydropower units can rapidly reduce or
increase power to meet the wind-load variation. Take time periods 2 and 3 as an example, the variations of
total wind-load power, coal-fired thermal power and hydropower are 0.05458, 0.00066 and 0.05392
respectively. Hydropower undertakes the majority of load variation, while the thermal power output is
stable. In this case, we found that the total maximum load accounts for only a small proportion (52% or so)
of the maximum power output of the whole units. Furthermore, the maximum power output of all
hydropower units accounts for approximately 48%. It should be noted that the power outputs of

13
IET Review Copy Only
IET Renewable Power Generation Page 14 of 20
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

hydropower units show the relatively large variation range. Consequently, the reserve capacity of
hydropower unit can be fully used to stabilize the fluctuation of the wind farm power output and load
variation.
As we can see, the power output of hydropower unit 1 is larger and more volatile than that of
hydropower unit 2. This is because the hydropower unit 1 can generate more power. Accordingly, the
hydropower unit 1 should preferably be utilized.

Table 2 Power output of units in 6-bus test system

t U1 U2 U3 U4
1 0.06171 0.09000 0.07042 0.04607
2 0.06789 0.10711 0.09217 0.02660
3 0.06943 0.10623 0.12085 0.05184
4 0.07323 0.13328 0.14435 0.05533
5 0.09312 0.11262 0.18397 0.05372
6 0.07807 0.13879 0.20101 0.06570

0.5
thermal power units
0.45 hydropower units
hydropower unit 1
0.4 hydropower unit 2
wind-load
0.35
Power output(p.u)

0.3

0.25

0.2

0.15

0.1

0.05

0
1 2 3 4 5 6
time period

Fig. 7. Power output of units in 6-bus test system

6.2. Case study 2


A real-sized regional power grid in China is analyzed to further illustrate how to implement the
wind-hydro-thermal coordinated dispatch. The 33-bus regional power system depicted in Fig.8 has five
coal-fired thermal power units at bus 14, 26, 28, 30, 33 and two hydropower units at bus 25, 31, and a
wind farm at bus 4. ft, gt are chosen as 4 and 0.5 respectively. α, β are chosen as 0.1. The rating of wind
farm power output is 1.

14
IET Review Copy Only
Page 15 of 20 IET Renewable Power Generation
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

21
29
3

15

17
5

8
10
7

2
6

11

18

Fig. 8. One line diagram of the 33-bus real-sized power system

In this case, the maximum load is 92.1% of the maximum power output of units. The maximum
power output of hydropower units is 21.3% of the maximum power output of all units. Thus, the
corresponding reserve capacity used to stabilize the fluctuation of the wind farm power output restricts the
power output of hydropower unit. Two kinds of dispatch modes are discussed below. The corresponding
CPU time of solving the CED model by applying EPSO method is roughly 351 seconds within 100
iterations
6.2.1 Dispatch considering the coordinated strategy
The objective function value decreases from 351.5608 to 341.8424 after 100 iterations. The
convergence process of the objective function value is shown in Fig. 9.
352

350
Objective function value

348

346

344

342

340
0 10 20 30 40 50 60 70 80 90 100
Iteration times

Fig. 9. Convergence process of the objective function value (33-bus power system, considering the coordinated strategy)

The power output of each unit after optimization is shown in Table 3. U1-U5 are coal-fired thermal
power units, U6 and U7 are hydropower units. The corresponding power output curve of each unit is
shown in Fig.10.
Table 3 Power output of units in 33-bus real-sized power system (with consideration of coordinated strategy)

15
IET Review Copy Only
IET Renewable Power Generation Page 16 of 20
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

t U1 U2 U3 U4 U5 U6 U7
1 3.41 1.90 1.76 1.42 1.41 1.51 1.03
2 2.99 1.85 2.14 1.49 1.56 1.03 0.34
3 3.76 1.70 2.17 1.61 1.26 1.02 0.34
4 3.37 2.00 1.82 1.42 1.44 0.89 0.34
5 4.19 2.20 1.82 1.40 1.44 1.85 1.48
6 3.66 1.79 2.47 1.57 1.63 1.85 1.48
7 4.87 2.26 2.17 1.63 1.37 1.85 1.48
8 5.34 2.18 2.32 1.45 1.71 1.89 1.50
9 4.48 2.31 2.36 1.47 1.42 1.45 1.11
10 6.03 2.66 2.26 1.62 1.69 1.49 1.17
11 5.52 2.53 2.27 1.86 1.80 1.85 1.38
12 5.17 2.76 2.17 1.53 1.66 1.54 1.38

18

16

14

12
Power output(p.u)

10
wind-load
8 thermal power units
hydropower units
6 hydropower unit 1
hydropower unit 2
4

0
0 2 4 6 8 10 12
time period

Fig. 10 Power output of units in 33-bus real-sized power system


(with consideration of coordinated strategy)

It can be seen from Fig.10 that the hydropower units can follow the variation of wind-load power
well at any moment. Because of the limitation of the maximum hydropower output, the thermal power
units have to undertake most of load demands. At time periods 1-5, the hydro power is not be fully utilized.
It is mainly because the lower limits of the thermal power have to be satisfied. At time periods 5-8, the
wind-load power continues to increase. But the hydropower output has reached the upper limit. In Fig.11,
the dotted lines denote the upper and lower limits of power output of hydropower units which are limited
by the reserve capacity. The solid line denotes the power output of hydropower units after optimization.
The lower limit increases at the time period 4, while the upper limit decreases at the time periods 4-12.
This leads to the results that the output range of hydropower units turns smaller at the time periods 4-12.
The hydropower outputs exceed the upper limit at the time periods 6, 7 and 8. This is because that the
wind-load increases at these time periods, and the variation of hydropower has to be consistent with the

16
IET Review Copy Only
Page 17 of 20 IET Renewable Power Generation
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

variation of wind-load. The excess part is distributed according to the available output capacity of thermal
power units. Thus, the hydropower is fixed to the upper limit and the thermal power undertakes the
variation of wind-load power.
4

3.5

3
Power output(p.u)

2.5

1.5

0.5
0 2 4 6 8 10 12
time period

Fig. 11. Limitation of hydropower output in 33-bus real system

6.2.2 Dispatch without the coordinated strategy


The objective function value decreases from 353.6408 to 347.1151 after 100 iterations. The
convergence process of the objective function value is shown in Fig. 12. The convergence speed of the
case without considering the coordinated strategy is smaller than the speed considering coordinated
strategy. This is because the power balance is easier to be satisfied when considering the coordinated
strategy. And the hydropower output has been artificially increased during iterations.
354

353

352
Objective function value

351

350

349

348

347
0 10 20 30 40 50 60 70 80 90 100
Iteration times

Fig. 12. Convergence process of the objective function value (33-bus power system, without the coordinated strategy)

The power output result of each unit after optimization is shown in Table 4 and Fig.13.
Table 4 Power output of units in 33-bus real-sized power system (without consideration of coordinated strategy)

17
IET Review Copy Only
IET Renewable Power Generation Page 18 of 20
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

t U1 U2 U3 U4 U5 U6 U7
1 2.94 1.86 2.55 1.38 1.34 1.14 0.80
2 3.79 2.25 2.59 1.42 1.26 0.69 0.97
3 2.97 1.94 2.03 1.29 1.29 1.58 0.45
4 3.25 1.50 2.09 1.53 1.51 1.33 0.80
5 4.19 2.50 1.59 1.44 1.37 1.41 1.12
6 4.67 1.96 2.73 1.15 1.87 1.37 1.19
7 4.86 2.55 2.38 1.76 1.43 1.39 1.29
8 5.36 2.75 2.01 1.63 1.65 1.51 1.13
9 4.40 2.60 2.65 1.36 1.21 1.45 0.81
10 6.13 2.88 2.78 1.54 1.43 1.32 0.83
11 5.98 2.42 2.59 1.47 1.87 1.51 1.09
12 4.46 2.84 2.61 1.79 1.76 1.62 1.38

18

16

14

12
Power output(p.u)

10
wind-load
8 thermal power units
hydropower units
6 hydropower unit 1
hydropower unit 2
4

0
0 2 4 6 8 10 12
time period

Fig. 13 Power output of units in 33-bus real-sized power system


(without consideration of coordinated strategy)

As we can see from Fig.13, the variations of hydropower outputs are different from the variation of
wind-load power at the time periods 2, 3, 8, 10, 11 and 12. Take time periods 4 and 5 as an example, in the
case considering the coordinated strategy, the variation of hydropower accounts for 68% of the total wind-
load power variation, while the maximum power output of hydropower units is only 21.3%. In the case
without considering the coordinated strategy, the variation of hydropower accounts for only 24.8% of the
total wind-load power variation. Compared with the dispatch mode considering the coordinated strategy,
the case without considering the coordinated strategy can hardly see the regularity of hydropower output.

7. Conclusion
In this paper, a wind-load power analytical model is deduced via the convolution method and
combined with the normal PDF of load. To deal with the random variables in CED problem, the stochastic
chance constraints and two-stage stochastic linear programming method are employed. Moreover, to better

18
IET Review Copy Only
Page 19 of 20 IET Renewable Power Generation
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

accommodate the intermittent wind power through dispatching thermal power and hydro power plants
elaborately, an EPSO method with a cooperative dispatch strategy is proposed.
The proposed method is tested on a 6-bus test system and a 33-bus real-sized power system.
According to the dispatch mode considering coordinated strategy, the level of load and the size of
hydropower capacity can lead to different performance of units. Also, units with higher efficiency have
power output priority. Specially, some unreasonable units power output can be avoided. And comparing
with the dispatch mode without the coordinated strategy, the EPSO method has faster convergence speed,
better fluctuation tracking ability and more reasonable reserve capacity selection.

8. Acknowledgments
This work is partially sponsored by the National Basic Research Program of China 973 Program
(2013CB228203).

9. References

[1] T.H.M, El-Fouly, H.H., Zeineldin, E.F., El-Saadany, M.M.A., Salama,: 'Impact of wind generation control
strategies, penetration level and installation location on electricity market prices', IET Renew. Power Gener., 2008,
2, (3), pp. 162–169
[2] S., Martin-Martinez, E., Gomez-Lazaro, A., Molina-Garcia, A., Honrubia-Escribano,: 'Impact of wind power
curtailments on the Spanish Power System operation', IEEE PES General Meeting, Washington D.C, USA, July
2014, pp. 1–5
[3] China Electricity Council, ‘National Electric Power Industry Statistics 2015 Express’, (2015), pp. 1
[4] A., Lorca, X.A., Sun,: ‘Adaptive Robust Optimization With Dynamic Uncertainty Sets for Multi-Period
Economic Dispatch Under Significant Wind’, IEEE Trans. Power Syst., 2015, 30, (4), pp. 1702-1713
[5] A., Botterud, Z., Zhou, J.H., Wang, J., Sumaili, H., Keko, J., Mendes, R.J., Bessa, V., Miranda,: ‘Demand
Dispatch and Probabilistic Wind Power Forecasting in Unit Commitment and Economic Dispatch: A Case Study of
Illinois’, IEEE Trans. Sustainable Energy, 2013, 4, (1), pp. 250-261
[6] J., Hetzer, D.C. Yu, K. Bhattarai,: ‘An Economic Dispatch Model Incorporating Wind Power’, IEEE Trans.
Energy Conversion, 2008, 23, (2), pp. 603-611
[7] L.B., Shi, C., Wang, L.Z., Yao, Y.X., Ni, M., Bazargan,: ‘Optimal Power Flow Solution Incorporating Wind
Power’, IEEE Systems Journal, 2012, 6, (2), pp. 233-241
[8] V.S., Pappala, I., Erlich, K., Rohrig, J., Dobschinski,: ‘A stochastic model for the optimal operation of a wind-
thermal power system’, IEEE Trans. Power Syst., 2009, 24, (2), pp. 940-950
[9] V., Miranda, P.S., Hang,: ‘Economic dispatch model with fuzzy wind constraints and attitudes of dispatchers’,
IEEE Trans. Power Syst., 2005, 20, (4), pp. 2143-2145
[10] M.H., Albadi, E.F., El-Saadany,: ‘Comparative study on impacts of wind profiles on thermal units scheduling
costs’, IET Renew. Power Gener., 2011, 5, (1), pp. 26-35
[11] Y., Bai, Y., Wang, Q., Xia, X., Sun, M.H., Yang, J., Zhang,: ‘A Full-scenario SCED With Coordinative
Optimization of Hydro-thermal-wind Power’, Proceedings of the CSEE, May 2013, 33, (13)
[12] Y. Fu, M. Liu, L. Li,: ‘Multiobjective Stochastic Economic Dispatch With Variable Wind Generation Using
Scenario-Based Decomposition and Asynchronous Block Iteration’, IEEE Trans. Sustainable Energy, 2016, 7, (1),
pp. 139-149
[13] S.R., Silva, de Queiroz A.R., Lima, L.M.M., Lima, J.W.M,: ‘Effects of wind penetration in the scheduling of a
hydro-dominant power system’, IEEE PES General Meeting, 2014, pp.1-5

19
IET Review Copy Only
IET Renewable Power Generation Page 20 of 20
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.
[14] B.Q., Hu, L., W, M., Marwali,: ‘On the Robust Solution to SCUC With Load and Wind Uncertainty
Correlations’, IEEE Trans. Power Syst., 2014, 29, (6), pp. 2952-2964
[15] J.L., Wu, B.H., Zhang, K., Wang, J., Shao, J.G., Yao, D., Zeng, T.Y., Ge,: ‘Optimal economic dispatch model
based on risk management for wind-integrated power system’, IET Gener., Transm., Distrib., 2015, 9, (15), pp.
2152-2158
[16] J.B., Park, Y.W., Jeong, J.R., Shin, K.Y., Lee, ‘An Improved Particle Swarm Optimization for Nonconvex
Economic Dispatch Problems’, IEEE Trans. Power Syst., 2010, 25, (1), pp. 156-166
[17] M., Moeini-Aghtaie, P., Dehghanian, M., Fotuhi-Firuzabad, A., Abbaspour,: ‘Multiagent Genetic Algorithm:
An Online Probabilistic View on Economic Dispatch of Energy Hubs Constrained by Wind Availability’, IEEE
Trans. Sustainable Energy, 2014, 5, (2), pp. 699-708
[18] I.A., Farhat, M.E., El-Hawary,: ‘Dynamic adaptive bacterial foraging algorithm for optimum economic
dispatch with valve-point effects and wind power’, IET Gener., Transm., Distrib., 2010, 4, (9), pp. 989-999
[19] L.B., Shi, Z.X., Weng, L.Z., Yao, Y.X., Ni,: ‘An Analytical Solution for Wind Farm Power Output’, IEEE
Trans. Power Syst., 2014, 29, (6), pp. 3122-3123
[20] R., Bo, F.X., Li,: ‘Probabilistic LMP Forecasting Considering Load Uncertainty’, IEEE Trans. Power Syst.,
2009, 24, (3), pp. 1279-1289
[21] A.V., Oppenheim, A.S., Willsky, S., Hamid,: ‘Signals and Systems 2nd Edition’ (Prentice Hall, 1997)
[22] J.W., Pierre,: ‘A novel method for calculating the convolution sum of two finite length sequences’, IEEE Trans.
Education, 1996, 39, (1), pp. 77-80
[23] John R Birge, François Louveaux,: ‘Introduction to Stochastic Programming’ (Springer, 1997)

20
IET Review Copy Only

You might also like