Professional Documents
Culture Documents
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2018.2866486, IEEE
Transactions on Power Systems
0885-8950 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TPWRS.2018.2866486, IEEE
Transactions on Power Systems
P Electricity Power output [MW-e] Currently, many researches are concerning about the
P Cur Cur
/L Renewable energy /load curtailment [MW-e] strategic operation of CSP plants and analyzing the
cost-benefits of CSP [11]-[13]. Majority of them consider the
R/r Scheduled / deployed reserves [MW-e] strategic operation of CSP to maximize its profit in an
x ,u Unit status/ start-up binary variables electricity market as a price-taker [14],[15] or in co-operation
with VRE [16],[17]. Fewer studies analyze the CSP behaviors
I. INTRODUCTION from the prospective of power system operation. Chen et al. [18]
A. Background reformulated the power system scheduling model to take CSP
into consideration and analyzed the benefits of CSP in reducing
Global warming and fossil fuel depleting are pushing electric system operating costs and VRE curtailment, but the
energy power systems to be transformed from being thermal uncertainties of VRE and CSP were not considered. Jin et al.
generator dominated into renewable dominated systems [1]. [19] proposed a multi-day stochastic operation model for power
Some countries like Denmark and Ireland are already operating systems with CSP to make full use of TES. Xu and Zhang [20]
a power grid with more than 20% annual electricity generation utilized a stochastic unit commitment (SUC) and a look-ahead
from wind power and solar photovoltaics (PV). However, economic dispatch (ED) with considering wind/CSP
renewable energy production highly depends on the weather uncertainty to simulate the operation of a power system with
condition and thus is associated with large variability and CSP and quantify the value of CSP for the provision of energy
uncertainty[2]. This brings great challenges for the power and reserve services. Denholm et al. [21] compared the
system operation and necessitates considerable operational economic value among CSP, base-load units and PV, based on
flexibility. a whole year production simulation for the California power
In this light, concentrating solar power (CSP) is an appealing system with 33% renewable portfolio standard. Dominguez et
renewable generation technology due to its dispatch-ability al. [22] simulated the CSP operation in a fully renewable power
through the use of thermal energy storage (TES) and is thus system and illustrated the significant role of CSP to hedge the
expected to play a significant role in high renewable energy variability and uncertainty of VRE generation.
penetrated power systems [3]. Specifically, CSP transforms One of key challenges of operating a high penetrated power
solar irradiation into heat that can be stored in TES during the system is how to strategically schedule flexible generating
daytime and generates electricity as needed. Even if little solar sources under large scale uncertain renewable energy
irradiation is available on some cloudy days, a large enough generation supply. Zheng et al. [23] summarized the SUC
TES system allows CSP to shift electric production between approaches to operate power systems under uncertainty and
different days. Besides, TES also allows CSP to cooperate with illustrated the benefits of stochastic optimization. Morales et al.
electric heaters (EH) to convert electricity from the grid into [24] proposed a stochastic scheduling approach with
thermal energy that is re-used at a later time. Generally, CSP co-optimizing energy and reserves, and evaluated the economic
equipped with TES and EH provides an alternative choice to value of operating reserves in power systems with high wind
supply operational flexibility in power systems to help the power penetration. Bakirtzis et al. [25] proposed a multiple
balance between load and generation in an uncertainty time resolution unit commitment (UC) model to optimize the
environment. These attributes will help to facilitate the short-term generation schedule under high renewable enegy
integration of variable renewable energy (VRE) such as wind penetration. Denholm and Hand [26] analyzed the role of
power and PV in high renewable penetrated power systems. storage in renewable dominated power systems. Pozo et al. [27]
In this paper, we focus on the short-term operation of a high addressed how to incorporate a generic energy storage model
renewable penetrated power system with CSP plants and into the SUC model, considering the use of energy storage for
analyze the role of CSP in accommodating VRE generation. shifting generation and providing reserves. The look-ahead
B. Literature review scheduling model is widely used to manage the state-of- charge
Operation of high renewable energy penetrated power (SOC) of the storage [28],[29].
systems is attracting increasing attentions[4]-[7]. The National C. Contributions
Renewable Energy Laboratory (NREL) in U.S. has presented a In this paper, we propose an improved stochastic scheduling
report Renewable Electricity Future Study [8] to analyze the model to strategically operate a power system with CSP plants
scenario where a large part of electricity load demand is under high renewable energy penetrations. The proposed model
supplied by renewable energy sources for the U.S. power simultaneously optimizes energy and reserve scheduling
system in 2050. This study concludes that it is feasible to decisions, considering the uncertainties of wind speed and solar
achieve a renewable-dominated power system with 80% irradiation which are modeled by multiple scenarios. The CSP
generation capacity from renewable sources in 2050, and CSP plant equipped with TES and EH is modeled in detail and
is expected to be a significant part in the generation portfolio. operates strategically to provide both dispatchable generation
Another NREL report [9] specifically analyzed the value of and operating reserves to facilitate the integration of VRE.
CSP in integrating large scale renewable energy generation Compared with a co-located VRE and storage design, CSP
through a case study on the U.S. southwestern electric system. essentially can be regarded as a storage with a variable and
Dominguez et al. [10] explored whether it is possible to achieve uncertain charging source. TES can be built large enough to
a fully renewable power system. The results show that CSP will allow CSP to shift generation between different days to hedge
play a significant role to ensure the power balance. the intermittency of solar irradiation. Hence, the day-ahead
operation schedule of CSP should not only consider the
0885-8950 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TPWRS.2018.2866486, IEEE
Transactions on Power Systems
operation situation of the next operating day, but also take into
account how much energy should be stored in the TES at the
end of next operating day for the operation at future days. Electric Heater
In order to address the above issue, a look-ahead SUC model
is proposed. It has a three-stage structure. The first stage Curtailment Loss
Solar
optimizes the scheduling decisions including both energy and power Solar Field Power Block Electricity
reserves at the day-ahead time scale. The second stage employs HTF
(SF) Thermal Thermal (PB)
a real-time re-dispatch process to eliminate the power power power
releasing storing
imbalance between day-ahead schedules and RT actuals for
each realizable scenario. The third-stage accounts for Thermal Energy
Storage (TES)
look-ahead operation in future operating days to optimize the
SOC of TES in each scenario at the end of next operating day. Heat loss
Compared with existing multi-stage stochastic programming
Fig. 2. Simplified energy flow in CSP with TES and EH
methods to optimize the operation of energy storage devices,
such as [30] and [31], the proposed model is more of an Many references have presented various CSP operation
extension of a two-stage SUC model considering look-ahead models. Detailed CSP operation model for the self-operation/
operation. control study can be found in [32]. Simplified steady-state
Taking into account the above background and literature dispatch model of CSP is proposed in [15] and [22]. Based on
review, the contributions of this paper are three-fold: these results, a more generic and comprehensive dispatch
1) Proposing a generic operation model of CSP with TES and model of CSP is proposed and can be easily incorporated into
EH towards the short-term power system operation scheduling. power system scheduling problems. A configuration of CSP
2) Incorporating the CSP operation model into a look-ahead equipped with TES and EH is considered.
stochastic unit commitment model that makes full use of the Through analyzing the thermal energy balance in each
flexibility of CSP in high renewable penetrated systems. encapsulated block in CSP, a mixed-integer linear constraint set
3) Analyzing the benefits of CSP in facilitating the is proposed to describe the feasible operating space of CSP.
integration of VRE generation.
The rest of the paper is organized as follows. Section II
CSP
c
SFcFore CSP EH
,t , t Pc ,t , Pc ,t , Pc ,t
TES , cha
, PcTES
,t
, dis
,t , xc ,t , t
, EcTES CSP
describes the operation model of CSP with TES and EH. SF Fore
c ,t PcCSP
,t
,Cur
/cPB cEH PcEH
,t =
Section III formulates the proposed look-ahead stochastic unit ,t (1)
commitment model. A case study on a modified IEEE RTS-79 P TES , cha
c ,t / cTES ,cha PcTES
,t c
, dis TES , dis
PcPB,t ,in
system is provided in section IV to verify the effectiveness of
PcPB ,t ) Pc ,t /c +uc ,t Ec
f ( PcCSP ,t (2)
,in CSP PB PB PB , SU
proposed model. Section V concludes the paper. ,t
,t (1 c ) Ec ,t -1 Pc ,t
EcTES PcTES t ,t 1 (3)
TES TES TES , cha , dis
II. CSP MODEL WITH TES AND EH ,t
,t 1 Pc ,t
PcCSP xcPB,t ( PcPB , Min PcRd , PB )
Hot tank CSP
,t 1 (13)
Pump
xcPB,t 1 ( PcPB , Max PcPB , Min ) PcPB , Max
Pump Pump
Pump
Cold tank
0 PcEH
,t Pc
PB , Max
,t (14)
Air cooled condenser
Net
P =P
c ,t
CSP
c ,t P EH
c ,t ,t (15)
Fig. 1. Configuration of a CSP plant
0885-8950 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TPWRS.2018.2866486, IEEE
Transactions on Power Systems
Constraint (1) represents the instantaneous thermal power scenario at the end of next operating day. The SUC model is
balance in CSP plants. The total thermal energy input is equal optimized to minimize the overall expected operating cost.
to the sum of thermal energy absorbed from the SF( SFcFore
,t
) and Only the decisions for the coming operating day are put into
the thermal energy transferred from EH( cEH PcEH ). The possible practice. The decisions for future operating days are for
,t
look-ahead operation evaluation.
spill is modeled by variable PcCSP ,t
,Cur
. The thermal energy is The variables associated with each stage are explained as
consumed via the energy exchange in TES and the consumption follows. For the first stage, variables include 1) the start-up and
by PB ( PcPB ,in
). Constraint (2) denotes the function between the shut-down decisions of each unit, 2) scheduled power output of
,t
each unit, 3) scheduled down/up spinning reserve of each unit
thermal energy input in PB ( PcPB ,in
) and the electricity output of
,t and 4) power flow on each branch. These variables are called
PB ( PcCSP
,t
)[15] . This function can be approximated by here and now decisions and do not depend on any particular
piece-wise linear functions. For simplicity, a constant scenario realization. For the second stage, the variables pertain
generating efficiency cPB is assumed, and EcPB , SU denotes the to each particular scenario for real-time re-dispatch, including:
1) the deployment of down/up spinning reserve of each unit, 2)
start-up energy needed to begin generating electricity[22].
the renewable energy curtailment, 3) the load shedding at each
Constraint (3) models the thermal energy balance in the TES
node bus, 4) actual power flow on each branch. These variables
system. The heat dissipation of TES is modeled by introducing
are called wait and see decisions. For the third stage, the
a dissipation factor cTES [18]. Constraint (4) depicts the feasible variables also pertain to each particular scenario but are related
operation interval of the SOC of TES. Constraints (5) and (6) to the look-ahead operation in future days, including: 1)
set the charging and discharging rates within the operation generation schedule of each unit, 2) power flow on each branch.
bound 0, PcTES , Max if only if the corresponding state variable is These variables are called look-ahead decisions.
1. Constraint (7) guarantees that the charging and discharging
Stage 1 Stage 2 Stage 3
states will not occur simultaneously and that the discharging
state will occur only when the PB is on. The PB module has
similar operational constraints as a conventional unit [33]. Realizable scenarios
Real-time realizable
Specifically, constraints (8) and (9) formulate the minimum s1.m1 for future days
scenarios
on/off time limit. Constraint (10) describes the relationship
s1 s1.m2
between the on/off state variables and the starting-up variable.
Constraint (11) sets the generation output within the operation
s2.m1
bound PcPB, Min , PcPB, Max if only if the PB is on-line. Constraints Day-ahead s2
(12) and (13), respectively, formulate the up/down ramping rate point forecasts s2.m2
limits. Constraint (14) assumes that the electricity consumption
s3 s3.m1
of EH is not more than the electricity generation capacity of
CSP. Constraint (15) formulates the net output of CSP with s3.m2
Net Day-ahead Real-time
TES and EH, Pc ,t , which locates in the interval Schedule Dispatch Look-ahead Operation
P
c
PB , Max
,P
c
PB , Max
.
Fig. 3. Uncertainty representation: a three-stage scenario tree.
The produced thermal energy in the SF, SFcFore , is modeled as
,t
B. Model Formulation
the input parameter. If given the value of solar irradiation,
1) Objective function
software tool SAM (Solar Advisor Model, developed by NREL)
[34] can be used to convert solar irradiation into produced solar minimize Cost Sys =Cost DA Cost RT Cost LA
thermal power in SF ( SFcFore
,t
) for CSP plants. where:
NT N G
III. UNIT-COMMITMENT MODEL FORMULATION
Cost DA SU gG u gG,t +C gG PgG,t +C Ru
g Rg ,t C g Rg ,t
Ru Rd Rd
t 1 g 1
A. Model Framework
N T G N N NCSP
Cost RT s C gG RgRu,t , s RgRd,t , s C
S
A look-ahead SUC model is proposed to optimize the CSP CSP
c P
c ,t , s
generation schedule of high renewable penetrated power s 1 t 1 g 1 c 1
systems with CSP plants. The proposed model simultaneously N
VoLL LCur
N
optimizes energy and reserve scheduling decisions, considering n ,t , s
the uncertainties of wind speed and solar irradiation. These n 1
uncertain inputs are modeled by a three-stage scenario tree NS N M K G
N N NCSP
shown in Fig.3. The proposed model has a three-stage structure. Cost LA s s , m C gG PˆgG, k , s , m C CSP ˆ CSP
c Pc , k , s , m
The first stage optimizes the UC decisions in a day-ahead s 1 m 1 k 1 g 1 c 1
framework based on point forecasting results, the second stage
NN
eliminates the imbalance power for all possible realizations in VoLL LˆCur
n , k , s , m
the real-time based on economic dispatch (ED) operation, and n1
the third stage accounts for the look-ahead ED operation in (16)
future operating days to optimize the SOC of TES in each The objective function (16) minimizes the overall expected
0885-8950 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TPWRS.2018.2866486, IEEE
Transactions on Power Systems
operating cost, Cost Sys . The first part is the day-ahead 0 PbB,t,dis 1 IbB,t,cha PbB, Max ,0 PbB,t,cha IbB,t,cha PbB,Max ,t ,b (26)
generation scheduling cost Cost DA , including the start-up cost,
fuel cost, and up/down spinning reserve scheduling cost for all EbB,t EbB,t 1 PbB,t,cha / bB,cha PbB,t,disbB,dis t ,t 1,b (27)
conventional units. The second part is the expected real-time
EbB,t EbB,ini PbB,t,cha / bB,cha PbB,t,disbB,dis t ,t =1,b (28)
dispatch cost Cost RT , including reserve deployment cost of
conventional units, the generating cost of CSP plants, and the 0 EbB,t H bB PbB , Max ,t ,b (29)
penalty cost of involuntary load shedding. The third part is the
PbB,t,cha PbB,t, dis RbB,t, Ru PbB , Max ; RbB,t, Ru 0 ,t ,b (30)
expected look-ahead operating cost Cost LA in future days.
2) First-stage constraints P B , dis
b ,t P B , cha
b ,t R B , Rd
b ,t P
,t ,b (31)
b
B , Max
;R
B , Rd
b ,t 0
First-stage constraints are formulated with day-ahead point For each CSP plant, the operating position should be located
forecasts of wind power and solar power. Constraint (17)
ensures the system load-generation balance at each time slot.
in the feasible operation space CSP
c SFcFore
,t , t , formulated
NG NW N PV NB by constraint (32). Constraints (33) to (36) formulate the
P G
g ,t PwW,t PvPV
,t Pb ,t
B , dis
PbB,t,cha up/down spinning reserve scheduling constraints. Specifically,
g 1 w 1 v 1 b 1 shown in constraint (33), when the PB is on-line, the up reserve
,t (17)
PcCSP
NCSP NN
D ,
provided by the PB is not more than the room ( PcPB , Max
)
P CSP
c ,t P EH
c ,t n ,t L Cur
n ,t L Cur
n ,t [0, Dn,t ] ,t ,t
the online conventional unit is scheduled to offer both energy 0 RcPB,t , Ru xcPB,t * min PcPB
,t
, Max
,t , Ec ,t Ec
PcCSP TES TES
cPB ,t ,c (33)
PgG,t and up (down) reserve RgRu,t ( RgRd,t ) without violating its
0 RcEH,t , Ru PcEH
,t ,t ,c (34)
operating bound P .
, EcTES EcTES
,t c
G , Min G , Max
,P g g 0 RcPB,t , Rd xcPB,t * min PcCSP PcPB , Min PB
,t ,c (35)
,t ,t
t 1
f l ,t DFl , g PgG,t DFl , w PwW,t DFl ,v PvPV
,t
x G
g ,t
xgG,t 1 TgOn
,min
,G
xgG, 0 ,t 1,g (20)
NB
g 1 w 1
NCSP
v 1
DF P PbB,t,cha DF P ,t
t TgOn
,min 1
,t ,l (37)
,G
l ,b
B , dis
b ,t l ,c
CSP
c ,t PcEH
t 1
x G
g ,t 1 xgG,t TgOff
,min
,G
1 x 0 G
g,
,t 1,g (21) b 1
NN
c 1
t TgOff
,min 1
DFl , n Dn ,t LCur
n ,t [ Fl
,G
Max
, Fl Max ]
xgG,t xgG,t 1 ugG,t , ugG,t xgG,t , ugG,t 1 xgG,t 1 ,t 1,g (22) NG
n 1
NB NCSP
P R x P
G
g ,t
Rd
g ,t
G G , Min
g ,t g ; gG,tP R x P Ru
g ,t
G G , Max
g ,t g ; R , R 0 , t ,g (23)
Ru
g ,t
Rd
g ,t RgG,,tRu RbB,t, Ru
g 1 b 1
R
c 1
PB , Ru
c ,t RcEH
,t
, Ru
R Ru,Sys ,t (38)
As for renewable generating units, constraints (24) and (25) NG NB NCSP
limit the schedule of a wind or PV plant not more than its
forecasting output value.
RgG,,tRd RbB,t, Rd
g 1 b 1
R
c 1
PB , Rd
c ,t RcEH
,t
, Rd
R Rd ,Sys ,t (39)
PwW,t PwW,t,Cur PwW,t, Fore ; PwW,t , PwW,t,Cur 0 ,t ,w (24) 3) Second-stage constraints
To represent the wind and solar power uncertainty, N S
,t Pv ,t
PvPV PvPV 0 ,t ,v (25)
PV , Cur , Fore PV PV ,Cur
,t ; Pv ,t , Pv ,t
scenarios representing real-time possible realizations of
For energy storage systems (ESS), constraint (26) limits the renewable energy production are considered in second-stage
charging and discharging rates within the operation bound constraints. These constraints describe how the scheduled
0, PcB, Max . Constraints (27) and (28) formulate the energy reserves in the first stage are deployed to eliminate the power
balance in ESS. Constraint (29) limits the SOC level not more imbalance between day-ahead schedules and real-time actuals.
B B , Max
than the energy storage capacity ( H b Pb ). The ESS is able Specifically, constraint (40) represents the load-generation
balance in each scenario. The branch power flow constraint in
to provide operating reserves. Constraints (30) and (31), (41) is the same as constraint (37) but modified for the second
respectively, schedule the up/down reserves provided by ESS stage variables.
with not violating the charging/discharging power capacity.
0885-8950 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TPWRS.2018.2866486, IEEE
Transactions on Power Systems
Modified version of (37) (41) 0 PˆgG,k , s,m PgG, Max ,k ,g , s, m (58)
For each conventional unit, the operating constraints are the
same as those in the first-stage. The dispatched power output is PgG, Rd PˆgG,k , s,m PˆgG,k -1,s ,m PgG, Rd ,k 1,g , s, m (59)
equal to its day-ahead scheduled output plus the real-time
deployed reserves. PgG, Rd PˆgG,k , s,m PgG,t T , s PgG , Rd ,k =1,g , s, m (60)
Modified versions of (18) and (19) (42) Constraints (61)-(68) formulate a simplified operational
xgG,t PgG , Min PgG,t , s xgG,t PgG , Max , t ,g ,s (43) model for CSP plants. Specifically, constraints (61)-(64)
simulate the operation of the power block and EH. Constraints
PgG,t , s =PgG,t rgG,t,,Ru
s rg ,t , s
G , Rd
, t ,g ,s (44)
(65) formulates the thermal energy balance in TES. The
0 rgG,t,,Ru
s Rg ,t ,0 rg ,t , s Rg ,t
G , Ru G , Rd G , Rd
, t ,g ,s (45) discharging rate of TES is related to the electricity generation
For each wind and PV plant: of CSP PˆcCSP
, k , s , m , while the charging rate of TES includes the
Modified versions of (24)-(25) (46) absorbed solar thermal energy from SF and the transferred
For each EES, the operating constraints are the same as those thermal energy from EH. Constraint (66) links the SOC of TES
in the first-stage. Constraints (48) and (49) link the scheduled in the second-stage with that in the third stage. The SOC of TES
ESS operation and actual ESS operation. is limited within [ EcTES , EcTES ] via constraint (67). Constraint
Modified versions of (26)-(28) (47)
(68) ensures the SOC of TES to go back to its initial value
, s pb ,t , s Pb , t
pbB,t, dis PbB,t,cha rbB,t,,Ru
s rb , t , s , t ,b,s (48)
B , cha B , dis B , Rd
EcTES , Ini at the end of the simulating time horizon.
0 rbB,t,,Ru
s Rb ,t ,0 rb ,t , s Rb ,t
B , Ru B , Rd B , Rd
, t ,b,s (49)
PcRd PˆcCSP ˆ CSP
, k , s , m Pc , k -1, s , m Pc
Rd
,k 1,c, s, m (61)
For each CSP unit, the PB status is fixed. The real-time
dispatch position should be located in the feasible operation PcRd PˆcCSP
, k , s , m Pc ,t NT , s Pc
CSP Rd
,k =1,c, s, m (62)
space. The dispatched power output is equal to the day-ahead
scheduled output plus the real-time deployed reserves. 0 PˆcCSP
, k , s , m Pc
PB , Max
,k ,c, s, m (63)
Deployed reserves from the power block and EH are limited not
exceeding the scheduled values in the first stage.
0 Pˆ EH
c,k , s ,m P c
PB , Max
,k ,c, s, m (64)
P CSP
c ,t , s
,t , s c
, PcEH CSP
SF Fore
c ,t , s
, t ,t ,c, s (50) Eˆ TES
c,k , s ,m EˆTES
c , k -1, s , m
Pˆ CSP
c ,k , s ,m / PB
c t / TES , dis
c
,t , s xc ,t
xcCSP CSP
,t ,c, s (51) SF Fore
c,k , s,m PˆcCSP ,Cur PB ˆ EH
, k , s , m / c Pc , k , s , m / EH c
TES , cha
t
P CSP
c ,t , s P CSP
c ,t rPB , Ru
c ,t , s r PB , Rd
c ,t , s ,t ,c, s (52) ,k 1,c, s, m (65)
0 rcPB
,t , s
, Ru
RcPB
,t
, Ru
,0 rcPB
,t , s
, Rd
RcPB
,t
, Rd
,t ,c, s (53) EˆTES
c,k , s ,m E TES
c ,t NT , s
Pˆ CSP
c ,k , s ,m / PB
c t / TES , dis
c
,t , s Pc ,t rc ,t , s
PcEH EH EH , Ru
rcEH
,t , s
, Rd
,t ,c, s (54) SF Fore
c,k , s,m PˆcCSP
, k , s , m / c
,Cur PB
PˆcEH
, k , s , m / EH c
TES , cha
t
0 rcEH
,t , s
, Ru
RcEH
,t
, Ru
,0 rcEH
,t , s
, Rd
RcEH
,t
, Rd
,t ,c, s (55) ,k =1,c, s, m (66)
4) Third-stage constraints E TES
c Eˆ TES
c,k , s,m E TES
c ,k ,c, s, m (67)
The third-stage model simulates the look-ahead operation for
future days, and the generation schedule is not put into practice. Eˆ E TES
c,k , s,m,k =N K ,c, s, m (68)
TES , Ini
c
Thus, a simplified linear operation model is employed and Constraints (69)-(73) formulate a simplified operational
binary UC decisions are not considered, in order to improve the model for ESS. Specifically, constraint (69) limits the
computation performance. Specifically, ED operational charging/discharging rate of ESS. Constraints (70) formulates
constraints are incorporated in the third-stage to optimize the the energy balance in ESS. Constraint (71) links the SOC of
residual SOC of TES and ESS at the end of the next operating ESS in the second-stage with that in the third stage. The SOC of
day for each scenario. B B , Max
ESS is limited within [0, H b Pb ] via constraint (72).
Constraint (56) represents the system load-generation
Constraint (73) ensures that the SOC of ESS will go back to its
balance. Constraint (57) includes the branch power flow limit B , Ini
and the operating constraints for VRE units. initial value Eb at the end of the simulating time horizon.
ˆ , Pˆ B,cha P B, Max
0P ,k ,b, s, m (69)
NG NW N PV NB B , dis
Pˆ G
PˆwW, k , s , m PˆvPV
, k , s , m Pb , k , s , m Pb , k , s , m
ˆ B ,dis ˆ B ,cha b, k , s , m b, k , s , m b
g ,k , s,m
g 1 w 1 v 1 b 1
Eˆ bB, k , s , m EˆbB, k -1, s ,m PˆbB,k,cha
, s , m / b
B , cha
PˆbB,k,dis
, s , mb
B , dis
t
NCSP NN
, k , s , m Pc , k , s , m Dn , k , s , m Ln , k , s , m , Ln , k , s , m [0, Dn , k , s , m ]
PˆcCSP ˆ EH ˆCur ˆCur ,k 1,b, s, m (70)
c 1 n 1
0885-8950 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TPWRS.2018.2866486, IEEE
Transactions on Power Systems
Eˆ bB, k , s , m EbB,t NT , s PˆbB,k,cha
, s , m / b
B , cha
PˆbB,k, dis
, s , mb
B , dis
t SUC model and the simplified look-ahead SUC model.
Value of Look-ahead Operation:
,k =1,b, s, m (71)
(77)
C Three arg min C Three (u, p) C Three arg min C Two
Full (u, p)
0 Eˆ
Full Full Full
B
b, k , s , m H P B B , Max
b b ,k ,b, s, m (72)
Cost of Simplification:
Eˆ cB, k , s , m EbB , Ini ,k =N K ,b, s, m (73)
(78)
C Three
Full arg min C Three
Full (u, p) C Three
Full arg min C Three
Simp (u, p)
Overall, constraints (16) to (73) compose the proposed
look-ahead three-stage stochastic operation model, which is a
stochastic mixed integer linear programming (MILP) model. IV. CASE STUDY
minimize C Three
Full (u, p)=Cost DA Cost RT Cost LA B3 B9 B10
uU ,pP
the cost difference of UC solution between the full look-ahead TES capacity 8 hours dis ( cha ) 98% (98%)
0885-8950 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TPWRS.2018.2866486, IEEE
Transactions on Power Systems
Wind Power Scenarios PV Power Scenarios the variability of wind power and PV generation. Fig. 6(b)
depicts the operating schedule of CSP plants. The absorbed
solar thermal energy during the daytime is stored in the TES,
instead of generating electricity. The TES system allows CSP
to shift generation to the periods of sunrise and sunset.
3000
Conventional Unit
CSP
CSP Solar Thermal Power Scenarios Load Scenarios 2500 Wind Power
PV
Generation / MW
2000 RES_Cur
Load
1500
1000
t10
t11
t12
t13
t14
t15
t16
t17
t18
t19
t20
t21
t22
t23
t24
t1
t2
t3
t4
t5
t6
t7
t8
t9
Three models, respectively represented by formulations (74) Time / Hour
to (76), are tested on the above test system. Results are (a) generation schedule
summarized in Table III. The estimated performance is the CSP generation from SF CSP generation from TES
results coming out of the model itself. The true cost Stored Energy from SF Storage Level of TES
500 8000
Generation / MW
model involves the minimum overall cost and drives the best 350
UC solution. However, the solving time is too long to be 300 5000
acceptable. The conventional two-stage SUC model has much 250 4000
200
better computational performance, but involves the worst 3000
150
solution, since the look-ahead future operation is not 100
2000
considered. The economic loss of not considering future 50 1000
operation is quantified as $83.0k. Compared with these two 0 0
models, our proposed simplified look-ahead SUC model t1 t5 t9 t13 t17 t21
Time / Hour
incorporates an optimistic estimation of look-ahead future (b) operation schedule of CSP plants
operation. The solving time dramatically drops to the level Fig. 6 Day-ahead generation schedules optimized by proposed method
similar with the two-stage SUC model.
The storage curves of TES in each scenario for different
According to the formulations (77) and (78), the saved
methods are compared in Fig.7. Obviously, for the two-stage
benefit of considering future operation is $28.8k, and the cost
SUC method, the storage level of TES in every scenario goes
of the simplification of employing look-ahead ED operation for
back to the initial level at the end of next operating day. For two
future days in the third-stage is $54.2k.
look-ahead SUC models, the residual SOC of TES at the end of
TABLE III next operating day is pertained to each scenario and optimized
PERFORMANCE COMPARISON AMONG THREE MODELS through considering the look-ahead future operation.
Simplified Full
Two-stage
Compared with two-stage SUC method, this brings a multi-day
look-ahead Look-ahead operation coordination.
SUC
SUC SUC
Expected Operating Residual SOC of TES
Estim- 216.05 221.12 222.42 Two Stage SUC
Cost in Day1 (k$) at the end of day1 /
ated
Expected Cost MWh-t
Perfor- 739.20 801.26 0
in Day2-Day4 (k$)
mance
Overall Cost 955.25 1022.4 222.42
SOC of TES / MWh-t
Expected Operating
216.05 221.12 222.42
True Cost in Day1 (k$)
Perfor- Expected Cost in Simplified Look-ahead SUC
860.56 801.26 882.98
mance Day2-Day4 (k$)
Overall Cost 1076.6 1022.4 1105.4
Benefits of Look-ahead (k$) 28.8 83.0 -
Cost of Simplification (k$) 54.2 - -
Solving Time (s) 136.2 32844 80.1
Full Look-ahead SUC
According the results from proposed SUC method, Fig. 6(a)
shows how generation units are scheduled to supply demand on
day-ahead stage. The renewable energy penetration level
reaches up to 51.1%. The production of CSP is limited by the Time / Hour
available solar irradiation, but is flexible in compensating for Fig. 7 Storage level of TES in each scenarios
0885-8950 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TPWRS.2018.2866486, IEEE
Transactions on Power Systems
C. Evaluating the Benefits of CSP Plants replaced by 710MW PV with keeping the same renewable
In this section, a whole year of simulation is made to quantify energy penetration level), the total investment cost of CSP in
the benefit of incorporating look-ahead operation in the SUC case 1 is higher than the investment cost of PV in case 3 by
model and to preliminarily analyze the cost-effectiveness of $11.20M-yr, while the save annual operating cost is $22.34M.
CSP plants. From this regard, investing CSP is cost-efficient. Detailed
Case 1 (Base case): CSP plants are equipped with TES. The analysis method for evaluating the investment economy of CSP
proposed look-ahead UC model is performed to strategically is left for the future study.
operate CSP to obtain a minimum-cost generation schedule. D. Sensitivity Analysis
Case 2 (no LA): Conventional two-stage SUC model is used.
1) Impact of the CSP flexibility
Thus, the look-ahead (LA) future operation is not considered.
The operating flexibility of CSP varies with the capacity of
Case 3 (no CSP): Compared with case 1, CSP plants are
the TES system. Table V shows the results of cases with
removed.
different TES capacity. With the increase in TES capacity, the
Case 4 (PV): Compared with case 1, 300MW CSP plants are
operational flexibility of CSP is enhanced, and both the overall
replaced by 710MW PV, keeping the same annual available
operating cost and the ratio of renewable energy curtailment
electricity generation.
decrease. Fig.8 demonstrates that the benefit of considering
Case 5 (with EH): Compared with case 1, CSP plants are
look-ahead operation and the benefit of incorporating EH are
equipped with EH to enhance the operational flexibility and
both increased with the increase of TES capacity.
make use of curtailed renewable energy generation. The TABLE V
transfer efficiency HE of EH is assumed to be 90%. PERFORMANCE COMPARISON WITH DIFFERENT TES CAPACITY
Case 6 (Storage): Compared with case 1, CSP plants are TES Capacity 0 Hour 4 Hour 8 Hour 12Hour
replaced by ESS with the same capacity. The cycle efficiency Case 1 (Base case) 88.06 69.24 65.72 64.50
Operating
of ESS is assumed to be 75%, a typical cycle efficiency of large Cost (M$)
Case 2 (no LA) 88.06 70.18 66.93 65.95
pumped hydro station. Case 5 (with EH) 88.06 69.03 65.25 63.84
The operating cost and renewable energy curtailment during Case 1 (Base case) 19.89% 14.56% 11.68% 10.70%
Renewable
a whole year in each case are compared in Table IV. Several Curtailment
Case 2 (no LA) 19.89% 14.69% 11.99% 11.10%
findings are summarized. 1) Comparing case 1 with case 2, Case 5 (with EH) 19.89% 12.90% 9.24% 7.56%
considering look-ahead future operation brings $0.91M saving
on operating cost and reduces renewable energy curtailment by Reduction on renewable
1.60 Reduction on operating cost 3.50%
0.028TWh. 2) Comparing case 1 with case 3, the introduction curtailment
Reduced Operation Cost / k$
flexibility, we compare case 1 with case 4. Results show that 0.40 1.00%
the flexibility of CSP brings $22.34M savings and 0.772TWh 0.20 0.50%
0885-8950 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TPWRS.2018.2866486, IEEE
Transactions on Power Systems
Reduction on renewable [9] P Denholm and M. Mehos. “Enabling greater penetration of solar power
Reduction on operating cost via the use of csp with thermal energy storage,” NREL/TP-6A20-52978,
curtailment
1.40 3.50% Golden, CO: National Renewable Energy Laboratory, 2013.
20% Decrease 20% Decrease
nguez, A J. Conejo, and M. Carrión. “Toward fully renewable
0885-8950 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
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. Citation information: DOI 10.1109/TPWRS.2018.2866486, IEEE
Transactions on Power Systems
[32] M. J. Wagner and P. Gilmann, “Technical Manual for the SAM Physical Benjamin Kroposki (M’92, SM’00, F’14) received his
Trough Model,” NREL/TP-5500-51825. Golden, CO: National BS and MS in Electrical Engineering from Virginia Tech
Renewable Energy Laboratory, 2011. and PhD from the Colorado School of Mines. He is now
[33] M Mehrtash, M Raoofat, M Mohammadi, et al. “Fast stochastic the Director of the Power Systems Engineering Center at
security-constrained unit commitment using point estimation method”. the National Renewable Energy Laboratory (NREL). His
International Transactions on Electrical Energy Systems, vol.26, no.3, expertise is in the design, testing, and integration of
pp:671-688, 2016. renewable and distributed power systems.
[34] System Advisor Model (SAM). National Renewable Energy Laboratory.
[EB/OL] https://sam.nrel.gov/
[35] Probability Methods Subcommittee, “IEEE reliability test system,”
IEEE Trans. Power App. Syst., vol.PAS-98, no.6, pp.2047–2054, 1979. Qing Xia (M’01, SM’08) received his Ph.D. from the
Electrical Engineering Department of Tsinghua
Ershun Du (S’13) received both a B.S and Ph.D. from University, China in 1989. He is now a Professor at the
the Electrical Engineering Department of Tsinghua same university. His research interests include electricity
University in 2013 and 2018, respectively. His research markets, generation schedule optimization and power
interests include renewable energy uncertainty analysis, system planning.
power system operation and planning with wind power,
photovoltaic, and concentrating solar power.
0885-8950 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.