IET Generation, Transmission & Distribution Research Article

Wind generation as a reserve provider

ISSN 1751-8687 Received on 16th June 2014 Accepted on 28th November 2014 doi: 10.1049/iet-gtd.2014.0614 www.ietdl.org

Yury Dvorkin ✉, Miguel A. Ortega-Vazuqez, Daniel S. Kirschen Department of Electrical Engineering, University of Washington, Seattle, Washington 98105-2500, USA ✉ E-mail: [email protected]

Abstract: Wind generation is notorious for its high intermittency and limited predictability. To account for the uncertainty induced by wind in day-ahead planning, system operators provide additional reserve by scheduling controllable generators at a less than optimal output, therefore increasing the operating cost and, in some cases, negating the benefits of relatively cheap wind generation. This study proposes a day-ahead decision-making framework that minimises the operating cost by derating the wind production and, consequently, reducing the reserve requirements. The headroom in wind generation that this deration creates is used to provide upward reserve. This deration also decreases the reserve requirement because it reduces the uncertainty of wind power generation. The proposed methodology has been tested on a modified 24-bus IEEE Reliability Test System with different reserve policies under different wind penetration levels. This case study is based on the mixed-integer linear unit commitment model, which enables exploring economic benefits of the proposed methodology while the technical constraints on the power system generation are duly enforced. The results demonstrate that the proposed methodology reduces the total operating cost, even when wind power producers are compensated for the reserve provision by their lost opportunity cost.

1

Introduction

Accommodation of high volumes of wind generation (WG) requires the provision of additional amounts of reserve to cope with the uncertainty and variability associated with deterministic wind forecasts [1] or stochastic wind scenarios [2]. This reserve is provided in the form of headroom that controllable generators can use to cope with intra-hour deviations between the actual value of the net load and its day-ahead forecast [3, 4]. To provide this additional headroom, the controllable generators are operated in a less than economically optimal manner [5]. As the penetration level of WG grows, so does the uncertainty on the net load and the reserve requirements [4, 6, 7]. Pinson [8] and Parks et al. [9] present a comprehensive literature review of existing and prospective forecasting tools and assess the impact of the forecast quality and value. The cost of operating a power system is sensitive to the amount of reserve provided [10], which is driven by the reserve policy chosen [11]. Even a marginal increment in reserve requirements may result in a sizable increase in the operating cost because it may require the synchronisation of additional generators, which then force other generators to operate at less than optimal efficiency [12]. As shown in [13], WG could be scheduled at a derated level to reduce the cost of additional reserve requirements. The resulting headroom could then be used to provide upward reserve. Existing wind turbine control systems have the technical ability to follow secondary and tertiary dispatch commands [14]. If a portion of the reserve requirements is contributed by WG, a lower cost of dispatch of controllable generators can be attained. On the other hand, wind generators must be compensated for the loss of opportunity that they would suffer by providing reserve rather than energy [15]. This loss of opportunity is typically larger for WG than for controllable generators because the WG has a lower, nearly zero operating cost. However, Saiz-Marin et al. [13] show that reserve provision by WG is justified in power systems with large wind penetration (WP) levels. As shown in [16], WG could increase its profit by bidding in both the energy and reserve markets, as opposed to bidding in the energy market only. This paper proposes a unit commitment (UC) model in which the deration of WG is scheduled to contribute to reserve

IET Gener. Transm. Distrib., pp. 1–9 & The Institution of Engineering and Technology 2015

requirements. This model is formulated using a mixed-integer linear framework, which has been shown to outperform other optimisation approaches and used by the majority of US independent system operators [17]. Unlike [13, 16], the proposed model assumes that the reserve requirement is a function of the expected wind power production, as proposed in the open literature [12]. Therefore, the reserve requirements are reduced as the wind deration increases, because this deration decreases the uncertainty on WG. Furthermore, the proposed methodology acknowledges wind integration policies that may limit the amount of wind power curtailment or deration to meet other constraints on flexibility [18]. Therefore the proposed methodology can restrict the wind deration in accordance with such policies at the expense of sacrificing economic benefits [19]. The proposed methodology achieves a lower operating cost not only through the provision of reserve from WG, as in [13, 16], but also through a reduction in the reserve requirements. The effectiveness of this approach is demonstrated for the ‘3.5σ’ [12] and ‘(3 + 5)%’ [20] reserve policies. The common thread of these reserve policies is that they are hourly based and have been derived from the statistical analysis of the historical load and wind measurements and forecasts. This data-driven analysis accounts for the non-linearity of the uncertainty on WG and makes these policies more cost efficient than the conventional (N − 1) criterion when the proportion of renewable generation is significant [12]. The ‘3.5σ’ policy sets the hourly reserve requirement in proportion to the standard deviation of the net load forecast, σ. On the other hand, the ‘(3 + 5)%’ [20] sets the hourly reserve requirement to the sum of 3% of the hourly load forecast and 5% of the hourly wind forecast. As compared with the ‘3.5σ’ policy, this policy treats the hourly load and wind forecasts separately.

2

Unit commitment formulation

In the following equations, the indices i, b and t refer to the sets of controllable generators I, buses B and time intervals T. The set of transmission lines is denoted by L. The objective function of the

1

UC is min



t[T

+

SCt,i xt,i + Fi



i[I





LOC up Ct,i (rt,i ) +

t[T i[I

up/down pt,i , rt,i







 (1)

wind Ct,b

operating hour t = 0. Parameters giup and gidown are the minimum upand Ldown,min are the and down-times of generator i, whereas Lup,min i i minimum numbers of hours that generator i is required to remain or Ldown,min can be equal to zero at on/off. Note that either Lup,min i i = 0, then the same time. For example, if Lup/down,min i Lidown/up,min = 0.

t[T b[B

2.2 The first term of the objective function (1) represents the start-up cost, SCt,i, and the running cost, Fi (·), of generator i at operating hour t. Binary variable xt,i is equal to 1 if generator i is being started at hour t, and 0 otherwise. The running cost, up/down ), accounts for the fuel cost of each generator with Fi (pt,i , rt,i up/down output pt, i and with upward/downward reserve procurement rt,i up/down up r,down down Fi (pt,i , rt,i ) = lei pt,i + lr,up rt,i i rt,i + li

The output of each generator, pt,i, is constrained as follows max ut,i , pmin i ut,i ≤ pt,i ≤ pi

(2)

are the marginal costs of energy and upward/ where lei and lr,up/down i downward reserve of the ith generator. Although the running cost in (2) is formulated as a linear function, the methodology proposed in this paper can be used with a quadratic running cost. This, however, would increase the computing time. If the procurement of upward reserve results in an out-of-merit-order dispatch of generator i, the loss of opportunity cost of generators is compensated by the term LOC up (rt,i ) Ct,i up LOC LOC up (rt,i ) = rt,i lt,i Ct,i

Constraints (5) and (6) are used to model binary cycling decisions of generators, through the auxiliary variables xt,i for the start-up decisions and yt,i for the shutdown decisions. Binary variables xt,i and yt,i are linked to the binary commitment status of generators, ut,i, as follows ∀t [ T ,

i[I

(5)

i[I

(6)

Constraints (7) and (8) enforce the minimum up- and down-times of the conventional generating units as explained in [21]




  ∀t [ 0, Lup,min + Ldown,min , i i

xr,i ≤ ur,i ,

2.3

pt−1,i − pt,i ≤ RDi ,

∀t [ T ,

i[I

(12)

Flow limits on transmission lines are modelled using a linear dc power flow as follows ∀{b, m} [ L,

t[T (13)

where Bb,m is the admittance of the line connecting buses b and m, δ is the voltage angle and f max is the maximum power flow limit. Constraint (11) enforces the nodal load-generation balance


pt,i + wnet t,b − wst,b −

i[Ib




+




Bbm (dt,b − dt,m )

{b,m}[L|m.b

Bmb (dt,m − dt,b ) = dt,b ,

∀i [ I,

b[B

(14)

  ∀t [ Lup,min ,T , i

i[I i[I

(7)

At any operating hour t, the demand at bus b, dt,b, must be equal to the  sum of (i) the total output of the controllablenetgenerators at bus b, i[Ib pt,i , (ii) the net wind power injection, wt,b , at bus b, (iii) the wind spillage, wst, b, at bus b and (iv) the inflows and outflows on the transmission lines connected to bus b. The amount of wind spillage, wst, b, at every bus b is constrained by the availability of wind at bus b. The dc approximation of the transmission network used in constraints (13) and (14) neglects the reactive power and voltage support constraints, but preserves the linearity and, thus, tractability of the UC model at the expense of its accuracy. However, as shown in [22], the dc approximation adequately meets the needs of the active power analysis and day-ahead planning, which are considered in this paper. While the UC model proposed in this paper can be reformulated using an ac transmission model, this would increase the computational burden and would not change a qualitative effect on the results [22], because WG is not expected to provide reactive power and voltage services. 2.4

Wind power deration

The net wind power injection at bus b, wnet t,b , is (8) f wnet t,b = (1 − at,b )wt,b ,

up

yr,i ≤ 1 − ur,i ,

(11)

Nodal power balance constraints

r=t−gi +1




i[I

{b,m}[L|m,b

Constraints on binary variables

ut,i = gion/off ,

∀t [ T ,

(4)

where αt,b ∈ [0, 1] is a decision variable that determines the portion of the forecasted wind production, wft,b , to derated.

∀t [ T ,

(10)

pt,i − pt−1,i ≤ RUi ,

max max ≤ Bb,m (dt,b − dt,m ) ≤ f{b,m} , −f{b,m}

wind f = lLOC Ct,b t,b (at,b )wt,b

xt,i + yt,i ≤ 1,

i[I

(3)

is the marginal cost of the lost opportunity calculated for where each generator i at every operating hour t as explained in [15]. The wind , accounts for the lost third term of the objective function, Ct,b opportunity cost of WG because of either wind spillage or scheduled wind deration. In the context of this paper, the term wind deration refers to wind power generation that is reduced to provide reserve, whereas the term wind spillage is used for wind curtailments resulting from the enforcement of security criteria, for example, binding transmission generation constraints. Therefore wind can be computed as Ct,b

xt,i + yt,i = ut,i + ut−1,i ,

∀t [ T ,

max , Constraint (10) enforces the minimum, pmin i , and maximum, pi limits on the output of controllable generators. Constraints (11) and (12) enforce the ramp-up, RUi, and ramp-down, RDi, limits of the generators.

lLOC t,i

2.1

Dispatch constraints



∀t [ Ldown,min ,T , i

i[I

(9)

r=t−gidown +1

In these equations, gion/off denotes the on/off status of generator i at

where wft,b production the range production

∀t [ T ,

b[B

(15)

is the forecasted wind production and αt,b is the wind deration rate. This deration rate is a decision variable in from [0, 1]. If αt,b is zero, the forecasted wind is not derated and all the forecasted power is scheduled

IET Gener. Transm. Distrib., pp. 1–9

2

& The Institution of Engineering and Technology 2015

Fig. 1 Empirical CDF of the net wind power injection Central forecast is marked with a circle and the derated state is marked with a square Data: BPA [23]

f as power production, that is, wnet t,b = wt,b . On the other hand, if 0 < of the forecasted wind is scheduled as αt,b ≤ 1, only a portion

power production 0 , wft,b ≤ wft,b , whereas the rest constitutes a headroom that can be used for reserve. If αt, b = 1, the forecasted wind power is not used for power production, but used entirely for reserve and wnet t,b = 0. Fig. 1 illustrates wind deration using the cumulative distribution function (CDF) of the net wind power injection. This CDF is modelled for each operating hour based on fitting wind forecast error to the Skew–Laplace distribution, as explained in [24]. As the wind deration increases, that is, as αt,b approaches to 1, the probability that an actual realisation of wind production exceeds the value of the derated state increases. The difference between wind power production in the derated state and the wind forecast, illustrated as a grey area in Fig. 1, represents the maximum headroom that can be used for reserve provision. The expected amount of upward reserve that WG can provide is calculated as follows

w,up ≤ rt,b

wnet t,b wft,b

net Pr(wnet t,b )dwt,b ,

∀t [ T ,

b[B

(16)

where Pr(·) is the cummulative probability function, as illustrated in Fig. 1. Since this probability is not greater than 1, the amount of reserve that can be obtained from derating the wind production is lower than the amount of derated wind power w,up ≤ wft,b − wnet rt,b t,b ,

∀t [ T ,

b[B

(17)

dividing it across its vertical axis in a number of segments, as proposed in [12, 25] and illustrated in Fig. 2. The accuracy of this approximation depends on the number of segments and can be adjusted for each particular application. A larger number of segments would not only increase the accuracy of the model, but also result in longer computing times. Each segment j is defined by two parameters: its probability, πt,b,j, and range of the net wind power injection, Dwnet t,b,j . A binary variable, vt,b,j, is assigned to each segment j. Constraint (16) can then be reformulated in a linear manner as follows w,up ≤ rt,b




pt,b,j Dwnet t,b,j vt,b,j ,

∀t [ T ,

b[B

(18)

∀t [ T ,

b[B

(19)

j f wnet t,b ≤ wt,b −




Dwt,b,j vt,b,j ,

j

Constraint (18) calculates the expected amount of reserve provided by WG by integrating the intervals, which have non-zero binary variable vt,b,j. Constraint (19) ensures that the net wind power injection, wnet t,b , is less than the difference between the wind  f forecast, wt,b , and the derated wind power, j (Dwt,b,j vt,b,j ). Wind integration policies that restrict the amount of deferred wind can be enforced as follows wft,b − wnet t,b ≤ Vt,b ,

∀t [ T ,

b[B

(20)

wind . Ct,b

The This reserve procurement is remunerated in (1) through amount of this compensation is based on the full amount of wind deration, that is, wft,b − wnet t,b , since this quantity represents an actual loss of opportunity cost of WG. Constraint (16) is non-linear not only because of the non-linear nature of the CDF, but also because of the product of this function with the continuous variable wnet t,b . This CDF can be linearised by

Constraint (20) limits the total amount of wind deration using parameter Ωt,b, which can be determined by a particular wind integration policy. If Ωt,b = 0, no wind deration can be scheduled, that is, WG provides no reserve. Alternatively, if Ωt,b is large, economic benefits can be attained if WG is used as a reserve provider.

Fig. 2 Empirical and linearised CDF of the net wind power injection Data: BPA [23]

IET Gener. Transm. Distrib., pp. 1–9 & The Institution of Engineering and Technology 2015

3

2.5

Reserve constraints

up Controllable generator i provides ri,t of upward reserve at time t

 up max  up ≤ min Dri,t , pi ut,i − pt,i , ri,t

∀t [ T ,

i[I

(21)

up is the upward rampable capacity of that generator. Unlike where Dri,t reserve provided by WG, the ability to deploy reserve provided by controllable generators is subject to less uncertainty, for example, failure to synchronise an offline generator providing non-spinning

reserve [26]. This uncertainty is outside the scope of this paper, since (21) assumes that only synchronised controllable generators (ut,i = 1) can participate in the provision of reserve. The constraint on the total upward reserve requirement can then be expressed as follows
i[I

up rt,i +


b[B

w,up rt,b ≥


req net

R (wt,b ) + Rreq (dt,b ) ,

∀t [ T (22)

b[B

The first term on the left-hand side of (22) represents the amount of

Fig. 3 Modified 24-bus IEEE Reliability Test System (RTS) [27]

IET Gener. Transm. Distrib., pp. 1–9

4

& The Institution of Engineering and Technology 2015

Fig. 4 Wind forecast profiles with positive and negative correlations with load Data: BPA [23]

Table 1 Day-ahead cost for the ‘3.5σ’-rule (in 103, $) WP, %

0 10 20 30 40 50

Positive correlation

Negative correlation

Case I

Case II

Case III

Case I

Case II

Case III

1032.0 816.2 639.3 502.2 471.7 493.9

1032.0 816.2 639.3 502.0 456.8 471.1

1032.0 816.2 639.3 502.0 445.7 428.8

1032.0 894.5 791.4 743.8 710.4 718.9

1032.0 894.5 791.4 730.5 698.4 663.4

1032.0 894.5 790.9 729.3 689.7 638.5

0 10 20 30 40 50

Positive correlation

f Rreq (wnet t,b ) = 3.5(1 − at,b )s(wt,b ),

∀t [ T,

b[B

(23)

where s(wft,b ) is the standard deviation of the wind forecast. On the other hand, if the ‘(3 + 5)%’-rule [20] is used, the reserve requirement because of WG is given by

Table 2 Day-ahead cost for the ‘(3 + 5)%’-rule (in 103, $) WP, %

req by WG, Rreq (wnet t,b ), and by the load, R (dt,b). Transmission constraints on reserve deployment are enforced as in (13) and ensure that if this reserve is fully deployed, the power flow in every line remains in the range [− f max, f max]. Note that Rreq (wnet t,b ) is a function of the wind deration at bus b. If the ‘3.5σ’-rule [12] is adopted, the hourly amount of reserve required because of WG in constraint (20) can be calculated as follows

Negative correlation

Case I

Case II

Case III

Case I

Case II

Case III

1057.1 830.7 653.8 516.1 469.2 488.1

1057.1 830.7 653.8 513.2 451.8 462.1

1057.1 830.7 653.8 511.6 448.1 426.6

1057.1 917.9 811.5 755.4 748.3 719.9

1057.1 917.9 811.5 714.0 702.3 699.3

1057.1 917.9 811.4 669.5 655.1 648.4

reserve provided by all controllable generators. The second term represents the expected amount of reserve provided by WG. The right-hand side of (22) is the sum of reserve requirements induced

f Rreq (wnet t,b ) = 0.05(1 − at,b )s(wt,b ),

∀t [ T ,

b[B

(24)

Both the ‘3.5σ’-rule and ‘(3 + 5)%’-rule decrease the reserve requirement linearly as the wind deration increases. For both reserve policies, the deration of wind production has a two-fold impact on the reserve requirements. First, a lesser amount of reserve is provided by the controllable generators, as shown in (22). Second, as shown in (23) and (24), the reserve requirement decreases as the wind deration increases because of the reduced uncertainty. Downward reserve requirements are enforced in a similar way as in (22) and controllable generators are the only provider of this reserve.

Fig. 5 Wind usage for the positively correlated wind and load profiles with the ‘(3 + 5)%’ reserve policy Dark grey area denotes wind spillage, the light grey area represents wind deration, the white area stands for the energy produced by WG

IET Gener. Transm. Distrib., pp. 1–9 & The Institution of Engineering and Technology 2015

5

Fig. 6 Wind usage for the negatively correlated wind and load profiles with the ‘(3 + 5)%’ reserve policy Dark grey area denotes wind spillage, the light grey area represents wind deration, the white area stands for the energy produced by WG

3

Case study

The proposed case study uses a modified version of the 24-bus IEEE RTS [27] illustrated in Fig. 3. The proposed UC model is tested for different WP levels and for wind forecast profiles that are positively and negatively correlated with the load. Fig. 4 illustrates these wind profiles normalised based on the nameplate capacity. The WP level is

defined as the percentage of energy produced by WG system wide. The loss of opportunity cost is compensated for controllable generators that provide reserve as explained in [8]. Similarly, wind spillage and wind deration are compensated based on the marginal cost of energy in the unconstrained dispatch. Load shedding is monetised through the value of loss load, which is set at $5000/MWh. To compare the proposed UC formulation with other

Fig. 7 Fulfillment of the reserve requirements for the positively correlated wind and load profiles under the ‘(3 + 5)%’ reserve policy Light grey area is the portion provided by conventional generators, the dark grey area is the portion provided by wind, the white area represents the reduction in the reserve requirement achieved with the proposed methodology

IET Gener. Transm. Distrib., pp. 1–9

6

& The Institution of Engineering and Technology 2015

Fig. 8 Fulfillment of the reserve requirements for the negatively correlated wind and load profiles under the ‘(3 + 5)%’ reserve policy Light grey area is the portion provided by conventional generators, the dark grey area is the portion provided by wind, the white area represents the reduction in the reserve requirement achieved with the proposed methodology

Table 3 Expected operating cost for the ‘3.5σ’-rule (in 103, $)

3.1

WP, %

Tables 1 and 2 show the day-ahead operating cost, calculated using the objective function (1), for different reserve policies and WP levels. If the WP is under 10%, all three cases for both wind profiles result in the same schedule and day-ahead operating cost. As the WP increases, the proposed formulation (case III) results in a cheaper schedule as compared with the operating costs obtained with cases I and II. Case III enables wind deration at the 20% WP level, whereas cases I and II apply wind deration at the 30% WP level. Therefore the methodology proposed in this paper produces a cheaper solution than that obtained with the method proposed in [13]. Regardless of the reserve policy chosen, the day-ahead cost in cases I and II goes through a minimum for a 40% WP level. Further wind integration increases the operating cost. On the other hand, the cost of case III strictly decreases as the WP increases. Thus, the proposed approach would allow larger penetrations of WG as compared with the traditional approaches and approaches that do not account for wind deration in reserve requirements [13].

0 10 20 30 40 50

Positive correlation

Negative correlation

Case I

Case II

Case III

Case I

Case II

Case III

1041.3 829.2 671.0 524.2 483.6 501.0

1041.3 829.0 664.3 516.0 477.3 473.9

1041.3 829.0 660.3 508.7 450.1 436.2

1041.3 917.2 830.9 761.7 724.2 723.7

1041.3 917.2 825.6 751.1 706.3 682.5

1041.3 917.2 815.8 744.1 693.4 641.3

methodologies, this case study considers three cases. In case I, the WG is not derated and therefore does not provide reserve. In this case, constraints (15) and (19) are enforced with αt, b = 0. Case II implements conditions reported in [13], that is, wind deration is enabled, but is not considered when setting the reserve requirements. Therefore constraint (15) is enforced with 0 ≤ αt,b ≤ 1, but constraint (19) is enforced with αt,b = 0. In case III, wind deration is enabled and taken into account in the reserve requirements, that is, constraints (15) and (19) are enforced with 0 ≤ αt,b ≤ 1. Cases I and II are enforced with no limits on the amount of wind that could be derrated, that is, constraint (20) is unbounded.

Table 4 Expected operating cost for the ‘(3 + 5)%’-rule (in 103, $) WP, %

0 10 20 30 40 50

Positive correlation

Negative correlation

Case I

Case II

Case III

Case I

Case II

Case III

1059.3 844.2 684.8 529.1 478.9 499.1

1059.3 844.2 671.4 527.9 464.7 477.8

1059.3 844.2 671.4 526.8 456.0 439.4

1059.3 936.0 840.1 767.4 759.1 764.3

1059.3 936.0 831.6 733.4 709.4 734.6

1059.3 936.0 827.4 704.6 661.6 654.3

IET Gener. Transm. Distrib., pp. 1–9 & The Institution of Engineering and Technology 2015

3.2

Day-ahead cost

Wind utilisation

This subsection discusses the utilisation of available WG and its allocation between energy production, reserve provision and wind spillage. Figs. 5 and 6 illustrate the wind utilisation in the different schedules obtained for both wind profiles. The amount of wind deration increases as the WP grows. A higher WP also results in wind spillage, which is due to constraints on the transmission lines connected to the WG facility. For the negative correlation between load and wind, wind spillage occurs even for relatively low WP. This spillage is driven not only by transmission constraints, but also by the minimum down time constraints of the generators. Thereby system operators curtail wind to avoid shutting down base load generators U350 and U400. Figs. 7 and 8 show the itemised contributions to the reserve requirements for both wind profiles. The reserve requirements decrease when wind deration is applied. This typically happens during the peak wind production periods (i.e. during daytime hours for the correlated profiles and during the nighttime hours for the negatively correlated profiles). Figs. 7

7

Fig. 9 Expected cost of the Monte Carlo simulations a For the positively correlated wind profile and ‘3.5σ’-rule b For the positively correlated wind profile and ‘(3 + 5)%’-rule c For the negatively correlated wind profile and ‘3.5σ’-rule d For the negatively correlated wind profile and ‘(3 + 5)%’-rule

and 8 also illustrate that wind deration leads to 100% reserve procurement from WG during some operating hours if WP is high. 3.3

Validation using Monte Carlo simulation

Since the actual wind energy production differs from its forecast, the day-ahead schedules must be compared using Monte Carlo simulations reflecting the uncertainty on WG. For each trial of these Monte Carlo simulations, the day-ahead schedules are dispatched in pseudo-real time to follow randomly generated wind and load profiles. Additional commitments are allowed at this stage if the day-ahead constraints (5)–(9) are not violated. Tables 3 and 4 show the numerical results for the expected operating cost calculated using these Monte Carlo simulations. Fig. 9 illustrates these results. This operating cost includes the day-ahead and real-time start-up cost, the running cost, the cost of lost opportunity, the penalty cost of wind spillage and the social cost of load shedding. As WP grows, case I consistently produces the most expensive solution. Case III results in the least expensive solution for both reserve procurement rules and both WG profiles. The difference in the expected cost of cases II and III is driven by the reduction in the reserve requirements because of decreased wind uncertainty, as shown in (22).

4

Conclusion

This paper proposes a UC model that schedules WG to participate in both the energy and reserve procurement. The performance of this model has been tested on the 24-bus IEEE Reliability Test System using Monte Carlo simulations for different WP levels and different correlation of wind energy production and load. According to these simulations, the proposed methodology achieves a lower operating cost by derating WG to simultaneously reduce uncertainty on WG and the reserve requirements. The conventional generators are therefore dispatched more efficiently

because their contribution to the reserve requirements is reduced. Additionally, this methodology reveals that the cost savings achieved using the proposed method increases with the WP, even though WG receives its full lost opportunity cost for reducing its production. For high WP levels, the proposed method avoids intra-day cycling of inflexible generators. The wind deration also reduces the wind spillage because of transmission constraints and, thus, increases the capacity factor of WG. Therefore the proposed method facilitates further integration of WG in existing power systems in a cost-efficient, reliable and sustainable manner.

5

References

1 Ela, E., O’Malley, M.: ‘Studying the variability and uncertainty impacts of variable generation at multiple timescales’, IEEE Trans. Power Syst., 2012, 27, (3), pp. 1324–1333 2 Dvorkin, Y., Pandzic, H., Wang, Y., Kirschen, D.S.: ‘Comparison of scenario reduction techniques for the stochastic unit commitment’. Proc. of the 2014 IEEE Power and Energy Society General Meeting, 2014, pp. 1–5 3 Pandzic, H., Dvorkin, Y., Wang, Y., Ting, Q., Kirschen, D.S.: ‘Effect of time resolution on unit commitment decisions in systems with high wind penetration’. Proc. of the 2014 IEEE Power and Energy Society General Meeting, 2014, pp. 1–5 4 Dvorkin, Y., Kirschen, D.S., Ortega-Vazquez, M.A.: ‘Assessing flexibility requirements in power systems’, IET Gener. Transm. Distrib., 2014, 8, pp. 1820–1830 5 Wood, A.J., Wollenberg, B.F.: ‘Power generation, operation, and control’ (Wiley, 1984) 6 Makarov, Y.V., Loutan, C., Jian, M., de Mello, P.: ‘Operational impacts of wind generation on California power systems’, IEEE Trans. Power Syst., 2009, 24, (2), pp. 1039–1050 7 Dvorkin, Y., Pandzic, H., Ortega-Vazquez, M.A., Kirschen, D.S.: ‘A hybrid stochastic/interval approach to transmission-constrained unit commitment’, IEEE Trans. Power Syst., 2014, 99, pp. 1–11 8 Pinson, P.: ‘Wind energy: forecasting challenges for its operational management’, Stat. Sci., 2013, 28, (4), pp. 564–585 9 Parks, K., Wan, Y.H., Wiener, G., Liu, Y.: ‘Wind energy forecasting: a collaboration of the national center for atmospheric research (NCAR) and xcel energy’. Technical Report NREL/SR-5500-52233, NREL, 2011 10 Ortega-Vazquez, M.A., Kirschen, D.S.: ‘Optimizing the spinning reserve requirements using a cost/benefit analysis’, IEEE Trans. Power Syst., 2007, 22, (1), pp. 24–33

IET Gener. Transm. Distrib., pp. 1–9

8

& The Institution of Engineering and Technology 2015

11

12

13

14

15

16

17

18

Ortega-Vazquez, M.A., Kirschen, D.S.: ‘Should the spinning reserve procurement in systems with wind power generation be deterministic or probabilistic?’. Proc. of SUPERGEN ’09 Int. Conf. on Sustainable Power Generation and Supply, 2009, pp. 1–9 Ortega-Vazquez, M.A., Kirschen, D.S.: ‘Estimating the spinning reserve requirements in systems with significant wind power generation penetration’, IEEE Trans. Power Syst., 2009, 24, (1), pp. 114–124 Saiz-Marin, E., Garcia-Gonzalez, J., Barquin, J., Lobato, E.: ‘Economic assessment of the participation of wind generation in the secondary regulation market’, IEEE Trans. Power Syst., 2012, 27, (2), pp. 866–874 Aho, J., Buckspan, A., Pao, L., Fleming, P.: ‘Active power control system for wind turbines capable of primary and secondary frequency control for supporting grid reliability’. Proc. of 2012 AIAA/ASME Wind Symp., Grapevine, TX, 2012 Deqiang, J.G., Litvinov, E.: ‘Energy and reserve market designs with explicit consideration to lost opportunity costs’, IEEE Trans. Power Syst., 2003, 18, (1), pp. 53–59 Zhang, Z.S., Sun, Y.Z., Cheng, L.: ‘Potential of trading wind power as regulation services in the California short-term electricity market’, Energy Policy, 2013, 59, pp. 885–897 Federal Energy Regulatory Commission: ‘Recent ISO software enhancements of the last decade and future software and modeling plans’. [Online]. Available at: http://www.ferc.gov/industries/electric/indus-act/rto/rto-iso-soft-2011.pdf Bird, L., Cochran, J., Wang, X.: ‘Wind and solar energy curtailment: experience and practices in the United States’. NREL Report No. TP-6A2060983, 2014, pp. 1–58. [Online]. Available at: http://www.nrel.gov/docs/ fy14osti/60983.pdf

IET Gener. Transm. Distrib., pp. 1–9 & The Institution of Engineering and Technology 2015

19

20

21

22

23

24

25 26

27

Lew, D., Bird, L., Milligan, M., et al.: ‘Wind and solar curtailment’. Presented at the Int. Workshop on Large-Scale Integration of Wind Power Into Power Systems, 2013, pp. 1–9 Ela, E., Milligan, M., Kirby, B.: ‘Operating reserves and variable generation’. NREL Technical Report, 2011. [Online]. Available at: http://www.nrel.gov/docs/ fy11osti/51978.pdf Rajan, D., Takriti, S.: ‘Minimum up/down polytopes of the unit commitment problem with start-up costs’. IBM Research Report, 2005. [Online]. Available at: http://domino.research.ibm.com/library/cyberdig.nsf/papers/CDCB02A7C809D89 E8525702300502AC0/File/rc23628.pdf Purchala, K., Meeus, L., Van Dommelen, D., Berlmans, R.: ‘Usefulness of DC power flow for active power flow analysis’. Proc. of the 2005 IEEE Power Engineering Society General Meeting, 2005, pp. 1–5 Bonneville Power Administration: ‘BPA balancing authority load and total wind generation’. [Online]. Available at: http://www.transmission.bpa.gov/Business/ Operations/Wind/default.aspx Nazir, M.S., Bouffard, F.: ‘Intra-hour wind power characteristics for flexible operations’. 2012 IEEE Power and Energy Society General Meeting, San Diego, CA, 2012, pp. 1–8 Liu, G., Tomsovic, K.: ‘Quantifying spinning reserve in systems with significant wind power penetration’, IEEE Trans. Power Syst., 2012, 27, (4), pp. 2385–2393 Ortega-Vazquez, M.A., Kirschen, D.S.: ‘Optimising the spinning reserve requirements considering failures to synchronise’, IET Gener., Transm. Distrib., 2008, 2, (5), pp. 655–665 Pandzic, H., Dvorkin, Y., Wang, Y., Kirschen, D.S.: ‘Unit commitment under uncertainty – GAMS models’. Renewable Energy Analysis Laboratory, University of Washington. [Online]. Available at: http://www.ee.washington.edu/ research/real/Library/Code/UnitCommitmentunderUncertainty-GAMSModels.pdf

9