www.ietdl.org Published in IET Renewable Power Generation Received on 16th March 2012 Revised on 8th October 2012 Accepted on 31st October 2012 doi: 10.1049/iet-rpg.2012.0085

ISSN 1752-1416

Impact of wind power on sizing and allocation of reserve requirements Kristof De Vos1, Joris Morbee2, Johan Driesen1, Ronnie Belmans1 1

Research group Electa (ESAT), University of Leuven, Kasteelpark Arenberg 10, 3001 Heverlee, Belgium European Commission, JRC, Institute for Energy and Transport, P.O. Box 2, 1755-ZG Petten, The Netherlands E-mail: [email protected]

2

Abstract: The increasing share of renewable energy sources for electricity, driven by variable output technologies such as wind and solar photovoltaics, is expected to have an impact on the operational reserve requirements of power systems. This study applies a probabilistic approach to estimate reserve requirements and establishes a methodology that makes it possible to distinguish between different categories of reserves based on the imbalance drivers of wind power. The methodology is based on sizing fast-response reserves based on the distribution of output fluctuations inside the settlement period, and sizing slowresponse reserves based on the distribution of the average prediction error over the settlement period. The main advantage of this methodology is a reduction of the fast-response reserves, which are generally assessed as expensive compared to slowresponse reserves. This approach is applied in a case study and compared with alternative strategies. The results for 500 MW of wind power installed in a North Sea country confirm these reductions and show that with the suggested approach the required fast-response and slow-response reserves, respectively, amount to 7 and 23–26% of the installed wind power capacity.

1

Introduction

The further growth of wind energy in Europe needs to be accommodated in an electricity system that requires a continuous balance between demand and generation. Divergence from this equilibrium results in frequency deviations that may disrupt the entire system by causing, for example, the disconnection of generation or even black-outs. Transmission system operators (TSOs) are responsible for balancing their control area, a task that they accomplish by contracting balancing services from grid users, mainly from generators in the form of power reserves. These power reserves represent power generation capacity that can be activated upward or downward when needed in order to restore system balance [1]. As different studies show, the ongoing integration of renewable energy sources for electricity such as wind is expected to amplify system imbalances [2], thereby increasing balancing costs [3, 4]. Indeed, the variability of these resources leads to limited controllability and predictability, and therefore requires the availability of additional power reserves [5]. These reserves are generally expensive and improving the performance of wind power prediction systems is the obvious solution to reduce reserve requirements and costs [6]. The focus of this article is on efficient reserve policies, meaning that a balance is struck between reliability and cost. The main drivers in this respect are sizing of reserves and allocation between different reserves categories. The existing literature presents three main methodologies for sizing reserve requirements: ‘heuristic’, ‘probabilistic’ and IET Renew. Power Gener., pp. 1–9 doi: 10.1049/iet-rpg.2012.0085

‘system simulation’ approaches. A typical example of a ‘heuristic’ approach is ‘N–1’, a rule of thumb that determines reserves as a contingency for the case of a sudden loss of the largest generation unit in the control area [7]. Although such approaches are still in use across the continental European synchronous zone managed by the European network of TSOs for electricity (ENTSO-E), and elsewhere [8], they may not be adequate to tackle the increasing complexity of uncertainty that modern power systems are coping with. ‘System simulations’ based on unit commitment and economic dispatch [9, 10] or dynamic system models [11] determine optimal power plant activation over time, and can be used to check the reliability of a given power system. Using multiple scenarios or Monte Carlo simulations [12], this allows, in principle, for an accurate computation of reliability and the costs of a predefined level of power reserves, but it is less suitable for determining the level of reserves itself. When specifically focusing on sizing power reserve requirements, the ‘probabilistic method’, as proposed by [13–17] is usually presented as the most efficient approach, trading off complexity and accuracy. In this methodology, statistical indicators, for example standard deviation, which describe wind power variability and uncertainty, are used as a measure for reserve requirements. This is the approach used in the remainder of this paper. Currently, reserve requirements are generally static as they are determined for a certain period, for example, year or season. State-of-the-art research emphasises the application of probabilistic forecasts to implement variable reserve requirements depending on the state of the generation 1

& The Institution of Engineering and Technology 2013

www.ietdl.org system. Information concerning the uncertainty of wind power predictions may allow system operators to size reserves day-ahead or intra-day based on a trade-off between risk and cost [18]. Furthermore, with stochastic programming, the uncertainty can also be implemented directly in system simulation models, allowing simultaneous calculation of optimal generation levels and reserve capacities [19, 20]. In this paper, a method is proposed for subdividing reserves into fast-response reserves (often referred to as ‘secondary reserves’) and slow-response reserves (often referred to as ‘tertiary reserves’). This division is important, because provision of slow-response reserves may be cheaper than provision of fast-response reserves, hence an optimal allocation may offer significant cost savings [17]. The trade-off between cost and reserves provision is already emphasised by Ortega-Vazquez and Kirschen [21], who determine optimal spinning reserves using an approach based on cost-benefit analysis. However, they do not subdivide the reserves further into different categories. Dany [11] provides an analysis of the impact of increased wind penetration on secondary and tertiary reserves, but the definition of tertiary reserves refers to a time horizon that is, far longer than what is considered in this paper. In general, little attention is devoted to primary reserves, since wind power variations are expected to be limited on a time scale of seconds because of the geographical distribution [22]. Likewise, primary reserves will not be considered in this paper. The method proposed in this paper for allocating reserves builds on the work of [14–17] and is conceptually and computationally simple, while allowing for a significant reduction of the fastest, and typically most expensive, reserves. To illustrate the method, a numerical simulation is performed for Belgium, a representative example of a control zone in the North Sea region. The approach presented in this paper is related to the subject of renewable power management. In [23], a number of strategies for integrating renewable energy sources into the European grid are described. In particular, there is a large role for optimising the communication and utilisation of information available at various stages in the power generation chain. In this context, for example, German TSOs are using the Wind Power Management System (WPMS) [24], an advanced wind power forecast tool that has become an integral part of the electricity supply system. The WPMS combines online power production measurements and day-ahead forecasts based on numerical weather prediction models, to generate optimised forecasts for up to 8 h ahead. Power management can also extend beyond forecasting. For example in [25] a smart energy management system (SEMS) for optimal microgrid operation is presented. The SEMS consists of a power forecasting module, an energy storage system management module and an optimisation module. Its purpose is to generate suitable set points for all the sources and storages in a microgrid in such a way that economically optimised power dispatch is maintained to fulfil certain power demand. Such approaches could be scaled up to applications beyond microgrids, for example, to the level of cities. On 10 July 2012, the European Commission launched the Smart Cities and Communities European Innovation Partnership [26]. This builds on the European Initiative on Smart Cities, which has specific objectives for electricity networks: smart grids, allowing renewable generation, electric vehicles charging, storage, demand response and grid balancing; smart metering and energy 2

& The Institution of Engineering and Technology 2013

management systems; smart appliances, lighting, equipment; and to foster electricity generated with renewable energy sources (RES) [24]. This would imply the realisation of advanced renewable power management at the level of cities or regions. This paper addresses a specific topic within renewable power management: in this paper, wind power forecasts are taken as an input, and this paper studies how the costs of additional reserves because of RES integration can be minimised. The remainder of the paper is structured as follows. Section 2 presents a general overview of the cost and procurement of reserves. Section 3 presents the methodology for determining the impact of wind on the different groups of reserves. In Sections 4 and 5, the methodology is applied to a case study for a North Sea country. Section 4 focuses on the characteristics of the prediction error and fluctuations, whereas Section 5 deals with the impact on reserve requirements and costs. Section 6 concludes this contribution with a summary of the main findings and recommendations for further research.

2

Reserve procurement

TSOs generally dimension the required reserve capacity based on their expectations of the system imbalance over a certain period. This is driven by two components. The first component consists of the imbalance volumes falling under the responsibility of the market by means of ‘imbalance settlement’ [1]. This concerns the averaged imbalances of generation and demand during each settlement period. Each market party, that is, balancing responsible party (BRP), is required to balance its average injections and off-takes in each settlement period. The length of this settlement period may vary between 15′ (Belgium, The Netherlands), 30′ (France, UK) and 60′ (Spain, Sweden). The second component consists of demand and generation fluctuations inside the settlement period, which do not fall under the responsibility of the BRPs. Reserve requirements are generally allocated towards different groups of reserves characterised by specific response times and characteristics. The classification, nomenclature and specific characteristics vary over different power systems, which is already dealt with extensively by literature [5, 28] and therefore not further discussed in this work. In the rest of this paper, focus is put on the distinction between fast- and slow-response reserves. Fast-response reserves are typically provided with generators synchronised to the grid (spinning) while slow-response reserves can be procured from off-line generators, which can be started in a time frame of minutes (standing). These include quick-start diesel or gas-turbine units as well as most hydro-units [7]. This distinction is logically not applicable to downward reserves, which are always spinning. The slower-response reserves can be provided with a quick shutdown of flexible power plants. When procured from conventional generators, these reserves are characterised with a reservation and activation cost [29]. Reservation costs result from the commitment of generators to keep their plants on-line: reserves may require a suboptimal scheduling of generators – for instance spreading the demand over additional generators – resulting in increased start-up costs and efficiency losses because of partial loading. The activation costs are incurred when the reserves are actually dispatched resulting in additional fuel costs or savings for upward or downward spinning reserves, IET Renew. Power Gener., pp. 1–9 doi: 10.1049/iet-rpg.2012.0085

www.ietdl.org respectively. For standing reserves, no reservation costs occur but the activation becomes generally more expensive because it includes a start-up and shutdown cost. Consequently, the allocation of reserve requirements over spinning and standing reserves becomes a trade-off between reservation and activation costs [17].

3 Quantification and allocation of reserve requirements 3.1

Total reserve requirements

In liberalised electricity markets, generators take positions in day-ahead and intra-day markets based on their expectations of the future. Consequently, real-time deviations of demand, fluctuations of variable RES-E and unexpected outages of generation units cause system imbalances. As explained in Section 2, these are balanced on the real-time market for reserve capacity and settled with the responsible market player. This real-time market is mainly coordinated by the TSO, as the final responsible for system security. Alternatively, market players with flexible assets might compensate the imbalances in their portfolio themselves with available flexibility. Moreover, they could foresee additional flexibility by scheduling their generators taking into account the uncertainty in their portfolio, so as to minimise imbalance volumes and costs. Such flexible systems would require less stringent reserve conditions. This is however not taken into account in the rest of this paper: all imbalances are assumed to be dealt with by the TSO, and market players offer their flexibility to the balancing market. To determine the required reserve capacity for a certain period, the probabilistic methodology is applied: the individual distribution functions of all imbalance drivers are combined into one single overall distribution function representing the total system imbalance. This approach is also implemented in other reserve sizing studies and generally referred to as recursive convolution [16]. The aggregated cumulative probability density curve of the total system imbalance allows the calculation of upward and downward reserve requirements by means of a predefined

reliability level, for example, 99%. This is done by fitting a probability density curve on the imbalances and choosing the reserves to be equal to the 99% percentile value of imbalances. In that way, reserves cover 99% of power system imbalances whereas the remaining 1% may be resolved by demand shedding or generation curtailment. The reliability level is generally referred to in literature as the loss of load expectation (LOLE), which measure the fraction of time within a certain period in which the load is not served [20]. It is stressed that the aggregated total system imbalance is determined by the aggregation of all drivers, and that it is not only important to take into account smoothing effects between different power locations, but also correlations between the different imbalance drivers, for example, demand and wind, solar and wind or demand and solar. By contrast, this work focuses only on wind as an imbalance driver, and does not consider any smoothing that may take place when wind variability is combined with the other drivers. Hence, in order to focus on the effect of wind only, this paper follows a conservative approach, which may lead to an overestimation of reserves. This approach is illustrated in Fig. 1. Total reserve requirements are determined based on the time series of total wind power imbalance, which is calculated as the difference between the real-time, actual wind power output (i.e. actual injection into the grid) and the average predicted wind power over the settlement period (Fig. 1, left). In a second step, the probability distribution function of total wind power imbalance is inferred, and by means of the predefined reliability level, the reserve requirements are determined (Fig. 1, right). In algebraic notation, we have Total wind power imbalance = Wind power output − Average wind power prediction (1) Since ‘Wind_power_output’ is uncertain, the quantity ‘Total_wind_power_imbalance’ is a stochastic variable. Total upward and downward reserve capacities are chosen

Fig. 1 Representation of the total wind power imbalance (left) and the reserve sizing process based on the predefined 99% reliability level (right) IET Renew. Power Gener., pp. 1–9 doi: 10.1049/iet-rpg.2012.0085

3

& The Institution of Engineering and Technology 2013

www.ietdl.org such that Prob[Total wind power imbalance] ≤ Total upward reserve capacity] = (1 − a)/2

(2)

Prob[Total wind power imbalance ≥ Total downward reserve capacity] = (1 − a)/2

(3)

in which α is the predefined reliability level (e.g. 99%), and Prob[·] is the function that maps events to probabilities. Note that for intuitive clarity, the sign of ‘Total_upward_ reserve_capacity’ is chosen to be positive and the sign of ‘Total_downward_reserve_capacity’ is chosen to be negative. This convention will be maintained throughout the remainder of the paper. 3.2

Allocation of reserve requirements

To ensure reliability, system imbalances are to be allocated towards flexible generators which meet the maximal ramping rates of the imbalances. More specifically, this paper is concerned with splitting the reserves into fast-response and slow-response reserves. One possible approach is to cover total reserve requirements with the fast-response reserves. This straightforward approach (which will be called ‘strategy A’) might be feasible in flexible generation systems but is expected to result in elevated costs because of the high reservation costs of spinning reserves. As mentioned in the introduction, one of the contributions of this paper is a convenient methodology for subdividing total reserves into fast-response reserves and slow-response reserves in a cost-effective way. The methodology is based on a decomposition of the system imbalances caused by wind power: indeed, the total imbalance has two main components, namely: (i) the prediction error over the settlement period and (ii) the fluctuations inside the settlement period. The prediction error over the settlement period is the difference between the prediction and the averaged output over the same period (Fig. 2, left). The aggregated generation deviations over the control zone are balanced by the TSO and charged to the responsible BRPs by means of the imbalance settlement. The fluctuations inside the settlement period remain the responsibility of the TSO (Fig. 2, right). In view of the total

imbalance, both components may reinforce (positive fluctuation together with positive prediction error) or may compensate (positive fluctuation together with negative prediction error) each other. The allocation of the total reserve requirements towards fast- and slow-response reserves is based on the above-mentioned decomposition. The fluctuations (real-time wind power generation minus average wind power output over settlement period) and prediction errors (average wind power generation minus average wind power predictions over settlement period) are calculated and displayed as time series (Fig. 2). These time series are now used to compose the two probability distribution curves and, similar to the total imbalance depicted in Fig. 1, calculate the fast- and slow-response reserve requirements by means of the LOLE. First, fast-response reserves are sized based on the fluctuations inside the settlement period as these are too fast to be covered with slow-response reserves. Second, the amount of slow-response reserves is chosen so as to be able to cover the remaining imbalances, that is, the prediction errors. This simple approach (which will be called ‘strategy B’) minimises the reserved capacity of fast-response reserves. This is a cost-effective solution in cases when fast-response reserves have high reservation costs compared with slow-response reserves. In this way, fast-response reserves are not allowed to participate in covering prediction errors, which is justified if the total activation cost of slow-response reserves is cheap compared to the activation costs of fast-response reserves. Owing to start-up or shutdown costs, this is generally not the case. Therefore an alternative solution (which will be called ‘strategy C’) is to determine the amount of slow-response reserves by subtracting the fast-response reserves from the total reserve requirement calculated above. In this way, fast-response reserves are also activated to cover for prediction errors, which results in a reduction of the required slow-response reserve capacity. In the activation phase, a choice can be made to maximise or minimise the use of fast-response reserves. The choice of strategy depends on the cost of activating fast- compared with slow-response reserves. The best strategy depends therefore on the power system characteristics and final costs can be tested in power system simulations.

Fig. 2 Representation of the prediction errors (left) and short-term fluctuations (right) 4

& The Institution of Engineering and Technology 2013

IET Renew. Power Gener., pp. 1–9 doi: 10.1049/iet-rpg.2012.0085

www.ietdl.org One drawback of these methodologies is that the fast-response reserves may be underestimated because of the way markets are currently organised. The hourly or 15′ minutely designs result in jumps in prediction errors when moving between different settlement periods, leading to additional ramps in system imbalance, which require additional fast-response reserves. This flaw can be corrected easily by analysing the imbalance ramps and reallocating slow-response reserves to fast-response in order to meet the imbalance ramps. This leads to modified versions of strategies B and C, which will be called B′ and C′, respectively.

4 Numerical model: characteristics of wind power imbalance drivers The suggested approach in Section 3 sizes the reserve requirements based on a stochastic analysis of time series of wind power imbalances, prediction errors and short-term fluctuations. In this section, a numerical model is presented which is able to generate these data series for a small North Sea country, for example, Belgium. The installed wind power capacity at five different locations (offshore, coastal, inland) is defined as an input parameter which allows us to scale the total capacity or the distribution across the country. 4.1

Data and assumptions

The numerical model is based on measurements and predictions of wind speeds acquired from the Royal Dutch Meteorological Institute, KNMI [30] and the Energy Research Centre of the Netherlands, ECN [31], respectively. In total, five locations were studied among which two offshore (Vlakte van de Raan, Lichteiland Goeree), one coastal (Stavenisse) and two inland (Woensdrecht, Eindhoven) (Fig. 3). These five locations cover a geographical area of 160 × 100 km² in the North Sea area. The distribution of the locations over zones with different wind characteristics allows capturing a large part of the aggregation effects which occur in a control zone. One year of historical wind speed prediction data is acquired from ECN for 2011 which was generated with a performing prediction model. The dataset contains four prediction runs with a resolution of 10′, which can be processed towards seven prediction horizons: 00, 06, 12, 18 h day-ahead and 00, 06, 12 h intra-day with a resolution of 1 h. These predictions can be compared with the

Table 1 CF for wind per location Location

DR

GO

ST

WOE

EI

total

wind speed, m/s CF, %

9.4 37.8

9.8 40.6

7.5 33.5

5.8 22.4

5.8 21.4

– 32.7

real-time wind speed measurements acquired from the KNMI for the same resolution and locations. These wind speed time series are subsequently transformed into power output time series by means of a non-linear power curve representing the output characteristics of a wind turbine, farm or park. For this study, normalised, regional aggregated power curves from the TradeWind project are used to represent future technology [32]. For this study, installed wind power capacities are fixed at 130 MW per off-shore and 80 MW per inland or coastal location, resulting in a total capacity of 500 MW. This illustrative model allows drawing conclusions concerning the impact of wind on prediction errors and power fluctuations inside a balancing area. Real-time output data permit the calculation of the capacity factor (CF) of wind for 2011 for the five locations. These results show that wind resources decrease when going from offshore to inland (Table 1). The overall CF is 32.7%. As described above, three time series of wind power data are calculated for each location: † PRED 1 h: hourly predictions. † KNMI 10′: 10 min measurements. † KNMI 1 h: averaged hourly measurement. After aggregating the predictions and output for the entire control zone, three time series of wind power data are obtained: PRED 1 h, RT 1 h, RT 10′. These are the main inputs in Section 5 for determining the reserve requirements. Predictions are studied over a period of 1 h and fluctuations over 10′. 4.2

Results

4.2.1 Output prediction error: The imbalances of a single wind power plant are calculated by subtracting the nominated (day-ahead, intra-day) active power output from the predicted active power output as these nominations are generally based on wind speed predictions. When assuming that market players bid their nominations and are not able to rebalance their portfolio real-time, this imbalance driver is therefore calculated as the difference between the hourly averaged measurements and predictions Prediction error = RT 1 h − PRED 1 h

Fig. 3 Geographical representation of five wind measurement locations IET Renew. Power Gener., pp. 1–9 doi: 10.1049/iet-rpg.2012.0085

(4)

As shown in Fig. 4, the prediction error over the entire control zone can be described by means of a Laplace distribution. As can be seen, this distribution is in fact more realistic than the generally assumed normal distribution as it is better at describing the peak and the heavier tails relevant for the sizing of the reserves. This is true for all locations and prediction horizons studied and this distribution is assumed in the rest of this study. The prediction error can be characterised by comparing the predicted values to the real-time measured values. Three common indicators are the bias, mean absolute error (MAE) and root-mean-square error (RMSE). To make results 5

& The Institution of Engineering and Technology 2013

www.ietdl.org Table 3 Aggregated RMSE of different prediction horizons expressed as a % installed wind power of capacity 00hda 06hda 12hda 18hda 00hid 06hid 12hid total RMSE

14.39

13.47

12.81

11.94

11.59 11.04 10.85

Table 4 Confidence intervals for the prediction error expressed as % of the installed wind power capacity ( − overestimation; + underestimation) Reliability, %

00hda

06hda

18hda

00hid

06hid

12hid

99.9

−61.54 57.31 −44.53 43.97 −33.07 29.89 −26.08 23.16

−55.19 56.32 −42.46 43.30 −30.54 27.32 −23.91 20.79

−50.47 50.86 −41.28 36.32 −29.40 25.72 −22.64 20.47

−50.06 49.66 −38.34 35.77 −26.15 25.85 −20.62 19.57

−52.98 50.75 −38.85 32.69 −25.93 22.75 −20.46 18.44

−46.48 41.33 −35.03 31.53 −24.25 22.36 −19.40 18.07

99.0 95.0

Fig. 4 Histogram of the total prediction error for a 00 h day-ahead horizon

independent from the wind farm or park size, the normalised error measures can be used by dividing the prediction error by the installed capacity [33]. Tables 2 and 3 represent the day-ahead indicators for the five locations and seven prediction horizons. The RMSE varies between 17–22% day-ahead and 13–19% intra-day. It can also be remarked that this improvement slows down for prediction horizons close to real time. This is partially explained by the lower impact as only the last nominations of the day can be adapted (18–24 h). A second observation is the relatively large impact of aggregation even for a relatively small geographical area inside one climate zone. An equally distributed capacity over the five locations results in a reduction towards 14% day-ahead and 11% intra-day. The results for different confidence intervals are shown in Table 4. These results show that for a day-ahead prediction, 99% of the prediction errors remain in a bandwidth of [ −45; + 44]% of the installed wind power capacity. Further, these results show a significant impact of the prediction horizon and the confidence level. 4.2.2 Output fluctuations: As discussed in Section 3, fluctuations inside the settlement period are an important parameter to determine the amount of required fast-response reserves. Therefore a time series of fluctuations is generated by comparing the 10′ resolution output with the hourly averaged output Fluctuations = RT 10′ − RT 1 h

(5)

Based on the assumptions when generating the RT 10′ time series, it is found that the fluctuations are Laplace distributed and entirely symmetric. Table 5 shows that for a

90.0

99% confidence level, the fluctuations remain in an interval of 8% of the total installed wind capacity. This reduces firmly with the confidence level. 4.2.3 System imbalance: Total reserves requirements are based on the total system imbalance, which results from the aggregation of the short-term fluctuations and the prediction errors. These can reinforce or weaken each other based on their direction. A time series of system imbalances is constructed by means of Imbalance = RT 10′ − PRED 1 h

(6)

These results show the confidence intervals between which the system imbalance is expected to be for a certain percentage of the time. These are again based on the Laplace distribution fit of the imbalance time series. These results show more or less symmetric imbalances with a large impact of the prediction horizon and the confidence level. Total imbalances for a 99% interval amount up to [ −45; 45]% of the installed wind power capacity, which is reduced up to [ −35; 32]% with intra-day predictions.

5 Sizing and allocation of reserve requirements 5.1

Scenarios

To illustrate the impact of wind power imbalances on reserve requirements, the 500 MW numerical model presented in Section 4 is presented in three scenarios:

Table 2 BIAS, RMSE and MAE for the 00 h day-ahead predictions expressed as % of the installed wind power capacity 00 h D-1

De Raan

Goeree

Stavenisse

Woensdrecht

Eindhoven

Total

BIAS MAE RMSE

2.37 13.16 18.95

3.84 11.54 17.18

−2.95 14.02 19.99

−2.37 13.91 19.21

−8.71 15.99 21.85

−0.63 10.27 14.39

3.20 10.95 15.72

4.34 8.95 13.24

−2.36 11.63 16.59

−1.35 12.12 16.63

−7.51 13.78 18.88

0.16 7.89 10.85

12 h D BIAS MAE RMSE

6

& The Institution of Engineering and Technology 2013

IET Renew. Power Gener., pp. 1–9 doi: 10.1049/iet-rpg.2012.0085

www.ietdl.org Table 5 Confidence intervals for the fluctuations expressed as % of the installed wind power capacity ( − overestimation; + underestimation) Reliability, %

99.9

99.0

95.0

90.0

fluctuation, %

−13.49 13.48

−8.05 8.08

−4.79 4.84

−3.59 3.61

† 99.9% reliability, day-ahead predictions (00hda). † 99.9% reliability, intra-day predictions (12hid). † 99.0% reliability, intra-day predictions (12hid). The second scenario assumes that market players are able to use reliable intra-day predictions on well-functioning intra-day markets to correct their nominations. The third scenario assumes that the contracted amount of reserves is lower than the predefined reliability level of 99.9% and implies additional balancing actions from variable renewable sources (curtailment) or demand (shedding). 5.2

Results

Table 6 shows the reserve requirements when implementing the strategies discussed in Section 3. The main inputs of the calculation are the time series representing the total wind power imbalance, the short-term fluctuations and prediction errors, as described in the previous section. The Laplace distribution is used to determine the reserve capacity that is needed to cover these imbalance drivers with a predefined reliability. Note that in order to calculate the tertiary reserve requirements for strategy C, the experimental distribution is used as the time series representing the difference between the total imbalance and the fluctuation is not Laplace distributed. The reserve requirements are allocated between fast-response reserves (R2) and slow-response reserves (R3), according to the strategies A, B, C, B′ and C′ described in Section 3.2. The reserve requirements are expressed as a percentage of the installed wind power capacity. The two main reserve sizing strategies suggested in this paper (strategies B and C) are compared with a strategy that covers the entire imbalance with fast-response reserves (strategy A). The calculations are repeated for the two alternative strategies B′ and C′, in which the secondary reserves are based on the ramp of the system imbalance instead of the short-term fluctuations. The results in Table 6 show that large fast-response reserve capacities are required when covering all wind imbalances with fast-response reserves: In a scenario with 99.9% reliability and in the absence of intra-day trading possibilities, total fast-response reserve requirements would amount to [ + 73; −71]%. These values decrease to [ + 37;

−37]% when assuming lower reliability and intra-day trading. It is expected that these percentages increase when more wind is concentrated in one location. Focusing on the optimistic scenario, that is, the last row in Table 6, it can be observed that strategy B reduces fast-response reserves to [ + 7; −7]% of the installed wind power capacity. This value may increase to [ + 9; − 9]% when taking into account imbalance ramps (strategy B′). In both strategies B and B′, the amount of slow-response reserves remains high: the bandwidth of [ + 36; −36]% is nearly as large as the total imbalance (R2A). This is because the fast-response reserves are not used to cover part of the prediction error, thereby leading to oversized slow-response reserves. As discussed in Section 3, this is cost-effective when reservation costs for fast-response reserves are comparatively high. Alternatively, if activation cost of fast-response reserves are relatively cheap compared with those of slow-response reserves, the fast-response reserves can be used to help cover prediction errors (i.e. strategy C), thereby reducing the slow-response reserve requirement to [+ 28; − 25]%. Table 7 shows the yearly estimated activation volumes when the proposed strategies are used to cover the wind power imbalances. Again, these results are derived from the imbalance time series presented in the previous section. However, maximum activated volumes to cover these imbalances are capped by the reserved capacity. For reserve sizing strategy C, there are two possible activation strategies, which will be called C1 and C2. In strategy C1, activation priority is given to slow-response reserves. In strategy C2, activation priority is given to fast-response reserves. These results show that the total reserve activation to cover imbalances (R2A) amounts to [172; −180] GWh in the optimistic scenario. It is found that a part can be allocated towards fast-response reserves when its reserve capacity is minimised (i.e. in strategy B). By contrast, the levels of slow-response reserves activation remain more or less at the level of the total imbalance, which implies that the total volume of activated reserves exceeds the total volume of wind imbalances. This can be explained by the fact that fast- and slow-response reserves may be moving in opposite directions. Again, this may be justified by cost differences between both. When applying reserve sizing strategy C, it is seen that a part of the activation of slow-response reserves is indeed transferred to fast-response reserve activations (see activation strategy C1). This effect becomes even stronger when the fast-response reserves are used to the maximum for covering prediction errors (activation strategy C2): slow-response reserve activations are reduced significantly at the cost of increased fast-response reserve activation.

Table 6 Yearly reserve capacity requirements (expressed in % of wind capacity) ( + upward reserves, − downward reserves; R2 = fast-response, R3 = slow-response; subscripts A, B, C refer to the reserve sizing strategy chosen) [%]

Strategy A

B

C

B′

C′

Scenario

R2A

R2B = R2C

R3B

R3C

R2B′ = R2C′

R3B′

R3C′

500 MW 99.9% DA

72.59 −71.32 55.45 −55.78 36.91 −37.24

10.45 −10.45 10.45 −10.45 6.97 −6.97

71.65 −70.38 54.31 −54.64 36.15 −36.48

48.35 −47.80 35.99 −31.25 28.09 −24.89

13.33 −13.33 13.19 −13.19 8.80 −8.80

71.65 −70.38 54.31 −54.64 36.15 −36.48

45.47 −44.92 33.25 −28.50 26.26 −23.06

500 MW 99.9% ID 500 MW 99.0% ID

IET Renew. Power Gener., pp. 1–9 doi: 10.1049/iet-rpg.2012.0085

7

& The Institution of Engineering and Technology 2013

www.ietdl.org Table 7 Yearly reserve activation volumes in GWh ( + upward reserves, − downward reserves) [GWh]

Strategy A

B

C1

C2

Scenario

R2

R2B

R3B

R2C1

R3C1

R2C2

R3C2

99.9% DA

242 −214

33 −33

239 −211

33 −33

238 −210

143 −141

99 −84

99.9% ID

173 −180

33 −33

169 −176

34 −33

168 −176

121 −139

52 −51

99.0% ID

172 −180

32 −32

168 −176

34 −34

166 −173

92 −111

79 −81

The argument for strategies B and C in this paper is cost-effectiveness. To demonstrate this point, Table 8 shows a simplified calculation of the annual cost of reserves provision (including both reservation and activation costs) of strategies A, B, C1 and C2. For the sake of simplicity, only upward reserves are considered. Fast-response reserves are assumed to be provided through spinning combined cycle gas turbine capacity, whereas slow-response reserves are assumed to be provided through standing open cycle gas turbine capacity. Cost data for both technologies are taken from the comprehensive cost overview in [34]. Reservation costs are based on capital investment costs and operating and maintenance cost. Activation costs are based on fuel price and efficiency. With strategy A, total annual costs are 23.9 million EUR, or 20.1 EUR/MWh of wind power output. As expected, strategy B drastically lowers the reservation costs, because expensive spinning reserves are replaced by cheaper standing reserves. However, part of the cost advantage is offset by higher activation costs. Strategy C reduces reservation costs even further. In particular, with the cost parameters assumed here, strategy C2 is the most attractive, because it lowers reservation costs without increasing activation costs much. In the simplified calculation in this paper, it achieves a 27% cost reduction compared with strategy A.

6

Conclusions

This contribution estimates the additional reserve requirements caused by increasing wind power integration and proposes a cost-effective allocation between fast-response (secondary) and slow-response (tertiary) Table 8 Annual total reserve costs (reservation and activation) for upward reserves in scenario ‘500 MW 99.0% ID’ when applying different sizing and activation strategies [Million EUR]

reservation costs activation costs total costs compared with A in EUR/y/kW of wind power in EUR/MWh of wind power

8

reserves. The methodology is based on a probabilistic approach and a decomposition of the different imbalance drivers of wind: output fluctuations inside the settlement period and prediction errors over the settlement period. This methodology is implemented by means of a numerical model based on a small North Sea country. These results show that when maintaining current static reserve policies, total reserve capacity requirements for a 99.9% reliability level may amount to [ + 73; − 71]% of the installed wind power capacity. This can be reduced to [ + 37; − 37]% when lowering the reliability level to 99% and enabling intra-day trading possibilities. Moreover, reserve requirements are influenced by the geographical distribution of wind power. When the fast-response reserves are sized based on the short-term fluctuations, the required fast-response reserve capacity amounts to [ + 7; − 7]% of the installed wind power capacity. The required slow reserve capacity, sized on the prediction errors, remains around [ + 36; − 36]% of the installed wind power capacity, for a scenario with intra-day trading and 99% reliability. This methodology minimises the expensive reservation costs of fast-response reserves. In case of cheap activation of fast-response reserves, an alternative approach is to maximise the use of available fast-response reserves. In doing so, the required capacity of slow-response reserves can be reduced to [+28; − 25]% of the installed wind power capacity. Activation of fast-response reserves can be maximised or minimised in this case depending on the activation cost ratio of fast-response against slow-response reserves. In an illustrative cost calculation the best of the proposed strategies achieves a 27% cost reduction compared to a baseline in which all reserves are simply assumed to be provided by fast-response reserves. As shown by the significant amounts of reserve capacity and activation volumes, amounting up to several hundreds of GWhs in this illustrative case, this assessment is highly relevant for future power system reliability and operating costs. A next step is therefore the validation of the reliability and cost performance of the proposed methodologies in system simulations and in real-life. In order to further reduce the costs of reserves, research is to be directed towards methodologies minimising reserve requirements while maintaining reliability, for instance with probabilistic forecasts. Moreover, new technologies, for example, storage or demand-response are expected to play a role in providing cost-effective balancing capacity. Finally, international market developments are expected to decrease total balancing requirements and costs.

Strategy

R2 R3 R2 R3

A

B

17.5

47.8

3.3 8.0 1.2 9.5 21.9 −8% 43.9

3.3 6.2 1.3 9.4 20.1 −16% 40.3

3.3 6.2 3.4 4.5 17.4 −27% 34.7

20.1

18.5

17.0

14.6

6.4 23.9

C1

& The Institution of Engineering and Technology 2013

C2

7

Acknowledgments

The authors acnowledge the European Commission, JRC, Institute for Energy and Transport for the sponsorship of the study. This work was carried out within the multi-annual work programme of the ‘Assessment of Energy Technologies and Systems’ (ASSETS) Action of the European Commission’s Joint Research Centre. Any interpretations or opinions contained in this paper are those of the authors and do not necessarily represent the view of the European Commission. IET Renew. Power Gener., pp. 1–9 doi: 10.1049/iet-rpg.2012.0085

www.ietdl.org 8

References

1 Tractebel Engineering and Katholieke Universiteit Leuven: ‘Study of the interactions and dependencies of balancing markets, intraday trade and automatically activated reserves,’ DG-TREN (TREN/C2/84/2007), 2009 2 Holttinen, H.: ‘Impact of hourly wind power variations on the system operation in the Nordic countries’, Wind Energy, 2005, 8, (2), pp. 197–218 3 UKERC UK Energy Research Centre: ‘The costs and impacts of intermittency: an assessment of the evidence on the costs and impacts of intermittent generation on the British electricity network,’ London, UK, October 2006, available at http://www.ukerc.ac.uk 4 VTT Technical Research Centre of Finland: ‘Design and operation of power systems with large amounts of wind power, final report phase one 2006–08,’ IEA Wind Task 25, 2009, available at http://www. ieawind.org/ 5 Ela, E., Kirby, B., Lannoye, E., et al.: ‘Evolution of operating reserve determination in wind power integration studies’. IEEE PES General Meeting, 2010, pp. 1–8 6 Giebel, G., Brownsword, R., Kariniotakis, G., et al.: ‘The state of the art in short-term prediction of wind power – a literature overview, 2nd edition,’ Deliverable Report, on behalf of ANEMOS.plus & SafeWind, 2011 7 Wood, A., Wollenberg, B.: ‘Power generation, operation and control’ (Wiley, New York, 1996, 2nd edn.), p. 569 8 Dragoon, K., Milligan, M.: ‘Assessing wind integration costs with dispatch models: a case study of PacifiCorp’. National Renewable Energy Laboratory (NREL/CP-500-34022), 2003 9 Tuohy, A., Meibom, P., Denny, E., O’Malley, M.: ‘Unit commitment for systems with significant wind penetration’, IEEE Trans. Power Syst., 2009, 24, (2), pp. 592–601 10 Padhy, N.: ‘Unit commitment – a bibliographical survey’, IEEE Trans. Power Syst., 2004, 19, (2), pp. 1196–1205 11 Dany, G.: ‘Power reserve in interconnected systems with high wind power production’. IEEE Porto Power Tech Proc., Porto, Portugal, 10–13 September 2001 12 Wang, J., Shahidehpour, M., Li, Z.: ‘Security-constrained unit commitment with volatile wind power generation’, Power, 2008, 23, (3), pp. 1319–1327 13 Anstine, L., Burke, R., Casey, J., et al.: ‘Application of probability methods to the determination of spinning reserve requirements for the Pennsylvania–New Jersey–Maryland interconnection’, IEEE Trans. Power Appar. Syst., 1963, 82, (68), pp. 726–735 14 Maurer, C., Krahl, S., Weber, H.: ‘Dimensioning of secondary and tertiary control reserve by probabilistic methods’, Eur. Trans. Electr. Power, 2009, 19, pp. 544–552 15 Holttinen, H., Milligan, M., Kirby, B., et al.: ‘Using standard deviation as a measure of increased operational reserve requirement for wind power’, Wind Eng., 2008, 32, (4), pp. 355–377 16 Deutsche Energie-Agentur GmbH: ‘Dena grid study II. Integration of renewable energy sources in the German power supply system from 2015–2020 with an outlook to 2025,’ 2010, available at http://www. dena.de/

IET Renew. Power Gener., pp. 1–9 doi: 10.1049/iet-rpg.2012.0085

17 Strbac, G., Shakoor, A., Black, M., Pudjiando, D., Bopp, T.: ‘Impact of wind generation on the operation and development of the UK electricity systems’, Electr. Power Syst. Res., 2007, 77, (9), pp. 1214–1227 18 Matos, M., Bessa, R.: ‘Setting the operating reserve using probabilistic wind power forecasts’, IEEE Trans. Power Syst., 2011, 26, (2), pp. 594–603 19 Bouffard, F., Galiana, F.: ‘Stochastic security for operations planning with significant wind power generation’, IEEE Trans. Power Syst., 2008, 23, (2), pp. 306–316 20 Meibom, P., Barth, R., Hasche, B., et al.: ‘Stochastic optimization model to study the operational impacts of high wind penetrations in Ireland’, IEEE Trans. Power Syst., 2010, 26, (3), pp. 1367–1379 21 Ortega-Vazquez, M., Kirschen, D.: ‘Estimating the spinning reserve requirements in systems with significant wind power generation penetration’, IEEE Trans. Power Syst., 2009, 24, (1), pp. 114–124 22 Ackermann, T.: ‘Wind power in power systems’ (John Wiley & Sons, Ltd, Chichester, UK, 2005), p. 675 23 Hammons, T.J.: ‘Integrating renewable energy sources into European grids’, Electr. Power Energy Syst., 2008, 30, (8), pp. 462–475 24 Lange, B., Rohrig, K., Ernst, B., et al.: ‘Wind power prediction in Germany – recent advances and future challenges’, Zeitschrift für Energiewirtschaft, 2006, 30, (2), pp. 115–120 25 Chen, C., Duan, S., Cai, T., Liu, B., Hu, G.: ‘Smart energy management system for optimal microgrid economic operation’, IET Renew. Power Gener., 2011, 5, (3), pp. 258–267 26 European Commission: ‘Smart cities and communities – European innovation partnership’, communication from the commission, Brussels, 10.7.2012, C(2012) 4701 final 27 SETIS (Strategic Energy Technologies Information System): ‘European initiative on smart cities – indicative roadmap’, 2012, available at http ://www.setis.ec.europa.eu/about-setis/technology-roadmap/europeaninitiative-on-smart-cities 28 CIGRE: ‘Ancillary services: an overview of international practices’, CIGRE Technical Brochure, Working Group C5–6, 2009 29 Singh, H., Papalexopoulos, A.: ‘Competitive procurement of ancillary services by an independent system operator’, IEEE Trans. Power Syst., 1999, 14, (2), pp. 498–504 30 KNMI Royal Dutch Meteorological Institute: ‘Potential wind speeds of the Netherlands’, 2009. Available at http://www.knmi.be, [accessed 15 May 2009] 31 Brand, A., Kok, J.: ‘Aanbodvoorspeller duurzame energie – Deel 2: Korte-termijn prognose van windvermogen’, ECN Nederland (ECN-C-03-049), 2003 32 Mc Lean, J.R., Hassan, G., Partners Ltd.: ‘TradeWind WP2.6 – equivalent wind power curves,’ on behalf of TradeWind, 2008, available at http://www.trade-wind.eu/ 33 Madsen, H., Kariniotakis, G., Nielsen, H., et al.: ‘A protocol for standardizing the performance evaluation of short-term wind power prediction models’, on behalf of project ANEMOS, December 2004, available at http://www.anemos.cma.fr/ 34 European Commission: ‘Energy sources, production costs and performance of technologies for power generation, heating and transport’. Commission staff working document accompanying the second strategic energy review, an EU energy security and solidarity action plan. SEC (2008) 2872. Brussels, 2008

9

& The Institution of Engineering and Technology 2013