Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

2017 IEEE International Conference on Systems, Man, and Cybernetics (SMC)

Banff Center, Banff, Canada, October 5-8, 2017

Investigations of Distribution System Scheduling


with Photovoltaic Power and Load Variations
Shichao Liu∗† , Haikuo Shen∗ and Huanqing Wang∗ , Peter Xiaoping Liu∗†
∗ School of Mechanical, Electronic and Control Engineering
Beijing Jiaotong University, Beijing 100044, China
Email: scliu@bjtu.edu.cn
† Department of Systems and Computer Engineering

Carleton University, Ottawa, ON K1S5B6, Canada

Abstract—This paper investigates the uncertainty of the day- DGs. In [3], a MCS-based appraoch is used to investigate the
ahead distribution system scheduling considering the random cost effectiveness and the economic limitations for electrical
variations of both Photovoltaic-based distributed generator (PV- power systems with wind integration. In [4], a variance-
DG) output power and load. Instead of Monte-Carlo simulation
(MCS), a two-point estimation method (2PEM) is applied to reduction procedure is proposed for reducing sample numbers
obtain accurate and computation-efficient analysis results. Based used in MCS when assessing generating unit outages under
on the two-year real-world hourly weather and load data in the operating constraints. Although the MCS method can obtain
city of Ottawa, the estimation accuracy of the 2PEM has been reasonable results, the extensively repeated study is extremely
verified in an equivalent 44 kV distribution feeder system. In time-consuming even when some scenario reduction methods
terms of computational efficiency, the 2PEM can significantly
reduce the computation burden with comparison to MCS. By are used for speeding up simulation, such as two-step state
using the 2PEM, the impact of PV-DG output power and sampling [2] and clustering-based method [5]. Recently, a
load variations on the uncertainty of the distribution system number of analytical probabilistic methods have been reported
scheduling under different seasons is thoroughly studied. The for distribution system operation to improve the computational
analytical results show that the range of standard deviation of efficiency. For example, in [6], a novel interval analysis ap-
optimally scheduled DG generation for this distribution feeder
system is larger in summer than that in winter. proach is used to estimate the bounds of distribution reliability
Index Terms—Probabilistic optimal power flow, point estima- metrics considering the variation and unavailability in load and
tion, distribution system, correlation, Photovoltaic generation data. Other estimation methods have been reported
in terms of probabilistic optimal power flow study, including
I. I NTRODUCTION Cornish-Fisher expansion [7], cumulant-based method [8],
Currently, an increasing number of photovoltaic based dis- moment-based method [9], etc. However, distribution system
tributed generators (PV-DGs) are integrated into distribution scheduling uncertainties resulted from both PV output power
systems. For example, in 2010, the total PV installation capac- and load variations are not simultaneously considered in the
ity in Canada was 281.13 MW [1]. In 2013, the total Canadian above literature.
PV installation capacity was significantly increased to 1210.48
MW. This rapid growth of PV-DG penetration is making the
This paper analyzes the uncertainty of the day-ahead dis-
distribution system scheduling more difficult. One difficulty
tribution system scheduling. In specific, both PV-DG power
lies in the weather-dependent power production from PV-DGs
and load variations are considered. Due to Monte Carlo
are inherently variable. Another difficulty is that load demand
simulation (MCS) is really time-consuming, a two-point es-
in the distribution system is also randomly time-varying. The
timation method is applied instead. This 2PEM is able to fast
uncertainty related with PV-DG power production and load
obtain accurate probabilistic analysis results for the day-ahead
demand will consequently result in uncertainties of the day-
distribution system scheduling. By using the real-world PV
ahead distribution system scheduling. It is very important to
power and load data in the city of Ottawa, the impact of PV
develop fast and efficient probabilistic methods to analyze the
power variation and load change on the scheduling solution
uncertainty of the day-ahead distribution system scheduling
is thoroughly investigated. In summary, with comparison to
resulted from the variations associated with PV-DG power
MCS, the computation efficiency is greatly improved by the
production and load demand.
2PEM.
A variety of methods have been proposed for analyzing
uncertainties in the distribution system planning and operation.
One method is the Monte Carlo simulation (MCS). The MCS The remainder of this paper is organized as follows. The
tries to explore all possible operating scenarios when analyz- probabilistic day-ahead DG scheduling problem is formulated
ing the distribution system. In [2], a non-sequential MCS- in Section II. The MH-2PEM is introduced in Section III. Val-
based method is used for evaluating the reliability of active idate studies are performed in Section IV. Finally, Conclusions
distribution systems integrated with renewable-source-based are summarized in section V.
978-1-5386-1645-1/17/$31.00 ©2017 IEEE 793
II. P ROBABILISTIC DAY- AHEAD D ISTRIBUTION S YSTEM power production of noncontrollable DGs and load demand.
S CHEDULING The output vector Y is described as
The objective of the day-ahead DG scheduling is to assign Y = [PC , θ, V]T (4)
optimal active power references for all controllable DGs in the
distribution system at a minimum operation cost in the next where PC = [PC1 , · · · , PCN ], θ = [θ1,1 , · · · , θN,N ], and
24 hours. The scheduling is considered to be carried out with V = [V1 , · · · , VM ]
a time resolution of one hour. In this section, the deterministic As power production from uncontrollable DGs (such as
day-ahead scheduling problem is extended to the probabilistic wind and PV) and load demands are inherently variable, uncer-
model. tainties exist inevitably in PN Ck and PDr . These uncertainties
existing in the input vector X will consequently result in the
A. Deterministic Optimal Power Flow Formulation
random variation of the output vector Y of the OPF (1).
The deterministic day-ahead scheduling of a distribution Probabilistic optimal power flow (P-OPF) is to obtain the
system is formulated as the following constrained nonlinear statistical metrics of the random output vector Y (such as the
optimization problem. mean and the standard deviation (STD)) with known statistics
PN of random input vector X. In practice, the P-OPF will provide
minPCi FC = i=1 (ci PCi )
distribution system planners an efficient and reliable tool as
Constraints :
the variation range of these random output variables can be
PCi + PN Ck − PDr = Vi Vj (Gi,j cos(θi,j ) + Bi,j sin(θi,j )),
obtained when there are input uncertainties in the distribution
QCi + QN Ck − QDr = Vi Vj (Gi,j sin(θi,j ) − Bi,j cos(θi,j )),
system.
SLmin ≥ SLi,j ≤ SLmax ,
QCimin ≤ QCi,t ≤ QCimax , III. T WO -P OINT E STIMATION M ETHOD FOR S OLVING
PN imin ≤ PN i,t ≤ PN imax , P-OPF
QN imin ≤ QN i,t ≤ QN imax ,
Vimin ≥ Vi ≤ Vimax , Considering PV power and load variations, a two-point
θi,j,min ≥ θi,j ≤ θi,j,max . estimation method (2PEM) is applied to determine the range
(1) of output variable uncertainties in this work. The mean and
where, in the objective function, PCi and QCi are the active standard deviation are the dominant statistical metrics of these
and reactive power output of the ith controllable DG, ci is uncertain variables [10]. The procedure of computing the mean
the generation price of the ith controllable DG; in the two and STD of the output variables via the 2PEM method is
power flow equation constraints, PDr and QDr are the real summarized as the following steps.
and reactive power of the rth load, PN Ck and QN Ck are the • Step 1: determine the number of random input vari-
real and reactive power demand of the kth uncontrollable DGs, ables n. The uncertain input variable vector is X =
θi,j is the voltage angular difference between bus i and bus [X1 , X2 , · · · , Xn ].
j, with θi,j,min and θi,j,max are the lower and upper limits • Step 2: calculate the mean µXi and STD σXi of each
and Gi,j and Bi,j are elements of the system’s admittance random input variable Xi , i ∈ {1, 2, · · · , n} based on the
matrix; in the inequality constraints, SLi,j is the instantaneous obtained historical data.
power in the branch connecting bus i and bus j, with SLmin • Step 3: set the first two moments of the output variable
and SLmax are lower and upper limits, respectively; Vi is vector as E(Y) = 0, E(Y2 ) = 0.
the ith bus voltage with a lower limit Vimin and upper limit • Step 4: set k = 1
Vimax . The objective of this DG scheduling is to minimize • Step 5: determine locations of two concentrations ²k,1
the operation cost FC of the distribution system by assigning and ²k,2 and the concentration probabilities pk,1 and pk,2 ,
optimal power references PCi for all controllable DGs, with according to the following equations.
forecasted power production PN Ck from uncontrollable DGs √
and load demand PDr . ²k,1 = n (5)

B. Probabilistic Optimal Power Flow Problem ²k,2 = − n (6)
The optimal solution of the above deterministic OPF and 1
the forecasted power production of noncontrollable DGs and pk,1 = (7)
2n
load demand have the following nonlinear relationship.
1
pk,2 = (8)
Y = h(X) (2) 2n
In this paper, the input vector X can be written as • Step 6: determine the two concentration vectors Xk,1 and
Xk,2 based on the following equations.
X = [PNC , PD ]T (3)
Xk,1 = µXk + ²k,1 σXk (9)
where PNC = [PN C1 , · · · , PN CM ] and PD =
[PD1 , · · · , PDM ]. The input vector includes the forecasted Xk,2 = µXk + ²k,2 σXk (10)
794
where Vk is the kth eigenvector vector of the correlation (a) Average Daily Load
2000

Load (MW)
matrix RX .
1500
• Step 7: run the deterministic OPF (1) for both concentra-
tions Xk,i , i ∈ {1, 2} 1000
• Step 8: update E(Y) and E(Y2 ), according to the
500
following equations. 0 5 10 15 20 25
Hours
n X
X 2 (b) Average Daily PV Power

PV Power (MW)
0.6
E(Y) ' (pk,i h(Xk,i )) (11) Winter
Spring
k=1 i=1 0.4 Summer
Fall
n X
X 2 0.2
E(Y2 ) ' (pk,i h(Xk,i )2 ) (12)
0
k=1 i=1 0 5 10 15 20 25
Hours
• Step 9: calculate the mean and standard deviation
q
Fig. 1. Daily average load and PV power during the two-year period of
µY = E(Y), σY = E(Y2 ) − µ2Y (13) 2003-11-01 1:00 AM to 2005-10-31 24:00PM

• Step 10: repeat Step 5 to Step 9 for k = k + 1 until all


the random input variables are calculated.
IV. C ASE S TUDIES AND VALIDATIONS
To investigate the impact of PV output power and load
variations on distribution system scheduling, an equivalent
distribution feeder system at the level of 44 kV in the city
of Ottawa is tested. The historical data of solar irradiance,
temperature, price of generation, system load demand are
collected for the studied system. Firstly, performance com-
parisons between 2PEM and the Monte Carlo simulations are Fig. 2. A distribution feeder system in the city of Ottawa
performed. Then, by using the 2PEM, probabilistic analysis of
the impact of PV output power and load variations on distri-
bution system scheduling in different seasons are performed from this figure that the average daily load demand in winter
to see the seasonal influences. is the highest among all the seasons, due to the heating need.
On the other hand, the average daily PV output in winter is
A. PV Power and Load Data the weakest among the four seasons. Moreover, the PV power
The historical solar irradiance, temperature and system load and load are varying from hour to hour during the day in all
data are studied for the city of Ottawa, Ontario, Canada for seasons.
two years (from 2003-11-01 1:00 AM to 2005-10-31 24:00
PM). The solar irradiance and temperature data are on an B. Validation of the 2PEM
hourly basis and obtained from Canadian Weather Energy and The equivalent 44 kV distribution feeder system in the city
Engineering Datasets (CWEEDS) [11]. Based on the solar of Ottawa (see Fig. 2) is tested to evaluate the effectiveness
irradiance and temperature data. The PV power are calculated of the 2PEM for the probabilistic analysis of the distribution
as follows [12]. system scheduling problem when the PV power and load are
both changing.
P = Ir Aη(1 + γ(Tref − T )) (14)
The distribution feeder system includes a 44 kV substation.
where Ir is solar irradiance, T is cell temperature, Tref is the The capacity of the substation transformer is 100 MVA. A 20
reference cell temperature, η is the PV generation efficiency, Mvar capacitor bank is located at the substation. A PV-DG
A is the area of PV array and γ is the temperature coefficient. is connected at bus 6. Two regular DGs are connected at bus
Parameters related to the equation (14) are chosen as Tref = 3 and bus 8 respectively. Loads are located at bus 3 to bus
25◦ C, η = 0.8 and r = 0.005/◦ C. The PV array area A is 8. Each load is seen as the equivalent total load in the bus.
assumed as 1000m2 . This A does not affect the results of the The feeder conductor parameters are R = 0.1156Ω/km and
correlation between the PV power variation and load change X = 0.3708Ω/km according to practical feeder parameters in
as the variation extent and directions of PV power and load the distribution system.
are the meaning of the correlation. The previous two-year PV power profile (equivalent PV
Fig. 1 shows the average daily PV power (with PV array array area of 1000m2 ) is used as the power output of the
area of 1000m2 ) and total load in the city of Ottawa (obtained PV-DG at bus 6. The load profile is scaled down by 1000
from IESO [13]) for different seasons during the period of times to fit the capacity level in the tested distribution feeder
2003-11-01 1:00 AM to 2005-10-31 24:00 PM. It can be seen system.
795
TABLE II
Average Daily Generation Price in Ontario M EAN AND S TANDARD D EVIATION OF ∗
PC IN S UMMER DAY
100
Winter
Spring
Summer Hour Method µ (MW) σ (MW)
Fall
80 MCS 2.5984 0.5857
10 Am
Price ($/MWh)

2PEM 2.5096 0.5868


60 MCS 2.4111 0.6316
12 PM 2PEM 2.3219 0.6323
MCS 2.5623 0.5734
40 14 PM
2PEM 2.4735 0.5735
MCS 2.9753 0.4298
20 16 PM
0 5 10 15 20 25 2PEM 2.8871 0.4337
Hours

Fig. 3. Average hourly generation price in Ontario during the two-year period
of 2003-11-01 1:00 AM to 2005-10-31 24:00PM
As the Monte Carlo simulation results (Case 1) are con-
sidered as accurate, the errors for the mean µ and standard
deviation σ of the OPF outputs in Case 2 are calculated as
The two-year average hourly prices of generation in Ontario |(µM CS − µ)|
are shown in Fig. 3, based on the data in IESO (the ISO of ²µ = (15)
µM CS
Ontario) [13]. The average hourly generation price is used
from day to day in the same season. |(σM CS − σ)|
²σ = (16)
For the purpose of validating the 2PEM accuracy and σM CS
performance, only the DG at bus 8 is considered as the
where µM CS and σM CS are the mean and deviation of the
controllable DG to achieve the minimum cost. The mean and
OPF solution for Case 1 as the reference. As seen from Table
STD of the OPF solution are calculated by the following two
probabilistic methods: TABLE III
C OMPARISON OF ²µ
• Case 1-Monte Carlo Simulation (MCS) Method: By using
the 2-year historical PV and load data as the random
Season Method 10 AM 12 PM 14 PM 16 PM
inputs, the deterministic OPF is calculated for each ran-
dom input. Then, the mean and standard deviation of the Winter 2PEM 2.3934% 2.6221% 2.5447% 2.2462%
OPF outputs are calculated. As discussed in the literature, Summer 2PEM 3.4174% 3.6995% 3.4656% 2.9644%
MCS is able to provide accurate results. Its mean and
standard deviation are considered as references.
• Case 2-Original 2PEM: Based on the same historical PV TABLE IV
and load data, the original 2PEM is used for estimating C OMPARISON OF ²σ
the mean and standard deviation of the OPF outputs.
Season Method 10 AM 12 PM 14 PM 16 PM
By using these two methods, the mean and standard deviation
Winter 2PEM 6.3220% 3.8742% 2.9925% 6.3154%
of the scheduled DG power generation PC∗ at bus 8 of the
distribution feeder system in selected hours in winter and Summer 2PEM 0.1878% 0.1108% 0.0174% 0.9073%
summer are listed in Table I and Table II.
TABLE I III and Table IV, the 2PEM can keep estimation error for the

M EAN AND S TANDARD D EVIATION OF PC IN W INTER DAY mean value of the scheduled DG power generation PC∗ within
4%. For the STD, the 2PEM has better estimation accuracy
Hour Method µ (MW) σ (MW) in summer than that in winter. In terms of the computational
MCS 3.6391 0.2515 efficiency, times spent in calculating the solution of the for-
10 Am mulated P-OPF are listed in Table V for both 2PEM and MCS
2PEM 3.5520 0.2674
methods. The table shows that the 2PEM significantly reduces
MCS 3.3408 0.3975
12 PM the calculation time compared with the MCS method. To see
2PEM 3.2532 0.4129
the impact of PV output power and load variations, means and
MCS 3.4384 0.3609 STDs of the optimal scheduling solution PC∗ based on hourly
14 PM
2PEM 3.3509 0.3717 generation prices shown in Fig. 3 from 10AM to 16PM in
MCS 3.8909 0.1547 winter and summer are investigated and shown in Fig. 4 and
16 PM
2PEM 3.8035 0.16447 Fig. 5, respectively. It can be seen from Fig. 5 that the variation
range of scheduled controllable DG generation PC∗ in summer
is larger that that in winter.
796
TABLE V
C OMPARISONS OF C ALCULATING T IME ACKNOWLEDGMENT
This work was partly supported by the Fundamental Re-
Method Time Spent (seconds) search Funds for the Central Universities, Beijing Natural
2PEM 4.24 Science Foundation under project 4172048 and Jiangxi Provin-
MCS 121.01 cial Department of Science and Technology under grants
20121BBE50023,20133BCB22002 and 612049302056. The
work of Prof. Shichao Liu was partly supported by Funda-
4
Mean of Scheduled DG Powers mental Research Funds of Beijing Jiaotong University under
Winter the grant M17RC00010.
3.8 Summer

3.6 R EFERENCES
3.4 [1] Global market outlook: for photovoltaic 2014-2018.
[Online]. Available: http://www.cleanenergybusinesscouncil.com/
µ P* (MW)

3.2
global-market-outlook-for-photovoltaics-2014-2018-epia-2014
c

3
[2] Z. Bie, P. Zhang, G. Li, B. Hua, M. Meehan, and X. Wang, “Reliability
2.8 evaluation of active distribution systems including microgrids,” IEEE
Trans. Power Syst., vol. 27, no. 4, pp. 2342–2350, 2012.
2.6
[3] G. Desrochers, M. Blanchard, and S. Sud, “A monte-carlo simulation
2.4 method for the economic assessment of the contribution of wind energy
to power systems,” IEEE Transactions on Energy Conversion, vol. EC-1,
2.2
10 11 12 13 14 15 16 no. 4, pp. 50–56, 1986.
Hour [4] J. Valenzuela and M. Mazumdar, “Monte carlo computation of power
generation production costs under operating constraints,” IEEE Trans-
Fig. 4. Means of scheduled DG power actions on Power Systems, vol. 16, no. 4, pp. 671–677, 2001.
[5] H. Keko and V. Miranda, “Impact of clustering-based scenario reduction
on the perception of risk in unit commitment problem,” in Intelligent
System Application to Power Systems (ISAP), 2015 18th International
Conference on, 2015, pp. 1–6.
V. C ONCLUSIONS [6] P. Zhang, W. Li, and shouxiang Wang, “Reliability-oriented distribution
network reconfiguration considering uncertainties of data by interval
analysis,” Electr. Power and Energ. Syst., vol. 34, pp. 138–144, 2012.
In this paper, the probabilistic analysis of the day-ahead [7] F. R. Rodriguez, J. Hernandez, and F. Juradoa, “Probabilistic load
distribution system scheduling with both PV-DG power and flow for photovoltaic distributed generation using the cornishfisher
expansion,” Electr. Power Syst. Research, vol. 89, pp. 129–138, 2012.
load variations is performed. A two-point estimation method [8] A. Schellenberg, W. Rosehart, and J. Aguado, “Cumulant-based proba-
(2PEM) is applied to solve the formulated probabilistic power bilistic optimal power flow (p-opf) with gaussian and gamma distribu-
flow in a computation-efficient way. By using the two-year tions,” IEEE Trans. Power Syst., vol. 20, no. 2, pp. 773–781, 2005.
[9] D. K. Molzahn and I. A. Hiskens, “Sparsity-exploiting moment-based
historical data of solar irradiance, temperature and load in the relaxations of the optimal power flow problem,” IEEE Trans. Power
city of Ottawa, an equivalent 44 kV distribution feeder system Syst., vol. 30, no. 6, pp. 3168–3180, 2015.
in the Ottawa grid is tested to validate the effectiveness of the [10] E. Rosenblueth, “Point estimates for probability moments,” Proc. Nat.
Acad. Sci. USA, vol. 72, no. 10, pp. 3812–3814, 1975.
2PEM. The comparison results show the 2PEM can estimate [11] Canadian weather energy and engineering datasets (cweeds). [Online].
the means of the scheduled DG production with estimation Available: http://climate.weather.gc.ca/prods servs/engineering e.html
errors less than 3%. In terms of the computational efficiency, [12] Pvwatts version 5 manual. [Online]. Available: http://www.nrel.gov/
docs/fy14osti/62641.pdf
the formulated P-OPF is solved via the 2PEM within 2.5 [13] Independent electricity system operator (ieso). [Online]. Available:
seconds while the Monte Carlo simulation (MCS) method gets http://www.ieso.ca/Pages/Power-Data/Data-Directory.aspx
the solution after spending more than 2 minutes.

STD of Scheduled DG Powers


0.7
Winter
Summer
0.6

0.5
σ P* (MW)

0.4
c

0.3

0.2

0.1
10 11 12 13 14 15 16
Hour

Fig. 5. STDs of scheduled DG power


797

You might also like