Portal de Periódicos da CAPES

Energy storage and wind power: sensitivity of revenue to future market uncertainties

2016; Institution of Engineering and Technology; Volume: 10; Issue: 10 Linguagem: Inglês

10.1049/iet-rpg.2016.0024

1752-1424

Anna Dunbar, A. R. Wallace, Gareth Harrison,

Smart Grid Energy Management

IET Renewable Power GenerationVolume 10, Issue 10 p. 1535-1542 Special Issue: Selected Papers from the Offshore Energy & Storage Symposium (OSES 2015)Free Access Energy storage and wind power: sensitivity of revenue to future market uncertainties Anna Dunbar, Anna Dunbar School of Engineering, Institute for Energy Systems, University of Edinburgh, Edinburgh, UKSearch for more papers by this authorA. Robin Wallace, A. Robin Wallace School of Engineering, Institute for Energy Systems, University of Edinburgh, Edinburgh, UKSearch for more papers by this authorGareth P. Harrison, Corresponding Author Gareth P. Harrison gareth.harrison@ed.ac.uk School of Engineering, Institute for Energy Systems, University of Edinburgh, Edinburgh, UKSearch for more papers by this author Anna Dunbar, Anna Dunbar School of Engineering, Institute for Energy Systems, University of Edinburgh, Edinburgh, UKSearch for more papers by this authorA. Robin Wallace, A. Robin Wallace School of Engineering, Institute for Energy Systems, University of Edinburgh, Edinburgh, UKSearch for more papers by this authorGareth P. Harrison, Corresponding Author Gareth P. Harrison gareth.harrison@ed.ac.uk School of Engineering, Institute for Energy Systems, University of Edinburgh, Edinburgh, UKSearch for more papers by this author First published: 09 September 2016 https://doi.org/10.1049/iet-rpg.2016.0024Citations: 6AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onFacebookTwitterLinkedInRedditWechat Abstract Grid connected electrical energy storage is expected to enable the integration of variable renewable generation in the future. As the electricity sector develops wholesale electricity prices will change, which will change the way in which storage technologies are operated. This study investigates the sensitivity of storage revenue to uncertain market variables. Results indicate that higher gas prices, carbon prices and average demand would increase peak electricity prices, leading to larger daily price spreads and increased storage revenue. Increased wind generation, however, would reduce opportunities for price arbitrage and lessen storage revenue. Wind power also affects the way in which devices are operated and changes the characteristics which are rewarded by the market. With increased wind capacity, storage devices cycle less regularly as operation is driven by substantial changes in wind power output, rather than daily demand patterns. As a result, slower discharge times are more favourable and revenue is more sensitive to rates of self-discharge. Furthermore, there is less variation in wholesale electricity price and consequently conversion efficiency is more critical to performance. 1 Introduction Electrical energy storage (EES) is regarded as a potential solution to the challenge of the Energy Trilemma in facilitating the grid integration of renewable energy. Many benefits of storage have been identified including improved system control, reduced network congestion and avoided curtailment of renewable output [1]. In the coming decades deployment of renewable energy capacity is expected to increase significantly leading to a greater requirement for flexibility and a higher value to be placed on these benefits. There are expected to be bespoke applications, such as in islanded or heavily constrained networks, where business cases for storage would exist [2]. Storing renewable electricity until it could be consumed locally would be a more attractive option than upgrading transmission or distribution network connections in these cases. However, many renewable energy projects will not have the benefit of local consumers. These projects, particularly those located offshore, will be network-connected and will generate into a centralised energy system. Grid connected EES will be required to aggregate revenue streams from a range of markets if it is to be commercially viable [3, 4]. One recognised revenue stream is price arbitrage – purchasing electricity when it is cheap and selling it back to electricity suppliers during periods of peak demand when the price is high. Several studies have investigated the revenue available to a storage operator through price arbitrage with [5-9] using historic electricity prices to estimate revenue available under existing market conditions. Barbour et al. [5] compared the performance of pumped hydro, hydrogen and battery storage devices in the Great Britain (GB) market from 2005 to 2010. The results demonstrated increasing revenue with charging rate and technology performance improving with efficiency. Other studies have investigated a single technology in multiple markets. Connolly et al. [10] compared arbitrage value of pumped hydro plant in 13 different regions highlighting the dependence of revenue on local market conditions. Sioshansi et al. [11] concluded that arbitrage value was dependent on the specific generation mix and fuel costs. As larger number of wind farms are deployed, the generation mix will change substantially and wholesale electricity prices will increasingly be driven by wind power output in addition to demand cycles. Gas and carbon prices will also change in the future affecting the daily price spread. These may change the way in which storage devices are operated. Few authors have modelled arbitrage revenue in future electricity markets, however, Grünewald et al. [12] proposed a method to investigate this. Four storage technologies were examined in markets with increasing renewable energy capacity and the sensitivity of net present value was tested against a range of variables. The trading margin, or daily price spread, was highlighted as one of the most sensitive parameters. The model was not, however, capable of reflecting variations in gas and carbon prices in the price spread and consequently the effect of these on storage revenue. Furthermore, wind power was attributed a marginal value equivalent to the opportunity cost of a Renewable Obligation Certificate, which had the effect of driving costs negative. Barbour et al. [13] investigated the impact of negative electricity prices on arbitrage revenue for storage and concluded that whilst creating some opportunities to gain additional revenue during periods of charging, their occurrence would be infrequent and would probably not impact technology choices for storage. Dunbar et al. [14] used a similar approach to [12] implementing an alternative wind model, price function and including sufficient detail to reflect the impact of changing gas and carbon prices on arbitrage revenue. A single set of storage characteristics was investigated to show the changing annual revenue from 2020 to 2025 in the National Grid 2014 'Gone Green' Future Energy Scenario (FES). This scenario exhibited increasing wind capacity and higher gas and carbon prices, among other changes. The results suggested that increased wind power may lead to reduced arbitrage revenue, while increasing gas and carbon prices may increase revenue. However, these factors were investigated with a single scenario and the impact of their individual effects was not explicitly identified. Furthermore, the impact of these changes on the storage operation strategy was not investigated, nor the implications of this on the device characteristics which would be most favourable in these conditions. Using the model described in [14], this paper investigates the sensitivity of arbitrage revenue to changes in gas price, carbon price, capacity margin and wind power capacity in the GB market and the impact of these variables on the preferred device characteristics. The potential future value of these variables is highly uncertain and investors will need to understand how a storage investment will perform across a range of outcomes. Scenarios are a useful method for appraising uncertainty, but sensitivity studies allow the influence of individual factors to be investigated. It is critical that the relative importance of key storage characteristics is understood in the context of uncertain market variables. This will enable technology choices to be developed which are robust to changing market conditions instead of solutions which are optimal for today's market, but which may become redundant as the sector evolves. Table 1 compares the approach used in this study with other work investigating arbitrage revenue to highlight its contribution. Table 1. Comparison of scientific literature Paper Market Electricity prices Negative prices Sensitivities investigated Implications Figueiredo et al. [8] various historic no alternative historic markets Revenue varies significantly between markets, dependent on specific generation mix, market design and participant behaviour Connolly et al. [10] various historic no market, year, optimisation strategy Highlights variation in revenue between historic years, markets and optimisation strategies Sioshansi et al. [11] PJM historic no storage characteristics, forecasting, year Highlights influence of gas price on arbitrage revenue from historic prices. Justifies use of perfect foresight Grünewald et al. [12] GB future scenarios – varies wind/solar yes storage characteristics, generation mix Concludes arbitrage may be commercially viable for low cost, long duration storage in future with large renewable capacity Barbour et al. [13] GB historic (modified) yes Storage efficiency and capacity Negative pricing demonstrated to be beneficial to storage, but unlikely to have a major impact on technology choices Imperial College London [4] GB future scenarios no Storage size relative to wind farm size Focus on 'value of storage'; simple case study of wind farm this paper GB future scenarios – varies wind, gas and carbon prices no 'Wind year', gas and carbon prices, wind capacity, storage characteristics Sensitivity to wind power capacity, gas and carbon prices investigated (independently). Implications of wind-driven price profile on storage operation investigated 2 Storage revenue model The model comprises several components which estimate storage revenue in a simulated electricity system: an electricity market price model, a wind generation model and a storage arbitrage revenue model. Each of these components is explained fully in [14] which should be referred to for further details regarding the model assumptions and validation. A summary of the key features is given below. 2.1 Electricity market model The electricity market price model was established on the assumption of perfect competition. This approach has been shown to be representative of the power exchange in GB and is commonly used for modelling scenarios of future electricity prices where variables are significantly different from historic levels [15]. The pricing model operates using estimated aggregate supply and demand functions where the price of electricity for each half hour time period is determined by the market clearing price. Thermal generators were grouped into four classes: nuclear, coal, combined cycle gas turbines (CCGTs) and open cycle gas turbines (OCGTs). The aggregate supply function was formed by stacking the generator classes in merit order of increasing marginal cost, C, which was calculated for each technology using: (1) where η is the thermal efficiency, F is the fuel cost, ν is the carbon emitted from combustion, Fcar is the carbon price, V is the variable generation cost, a is a conversion coefficient and e is the cost of nuclear fuel enrichment. Competitive prices were assumed with marginal generators bidding a price between their own marginal generation cost and the cost of the next class of generator in the merit order stack (the fundamental costs and characteristics of each generator type are given in Table 2). Between these two values, a hyperbolic function was used to smooth the discontinuities in the step function and better represent the complexities of the supply curve, such as differing ages and efficiencies of plant within each generation type [14]. An exponential uplift in price which applies to OCGT to represent their ability to set high prices at extreme demand levels (and recover their fixed costs) [14]; similar approaches have been used by other authors [16, 17]. An example of the supply curve is shown in Fig. 1 where the merit order was nuclear, coal, CCGT then OCGT. The supply curve is notably flat in the regions where base load and mid merit generators fulfil demand. When peaking capacity is required, there is a sharp increase in price. Table 2. Thermal generator data [18-20] Generator type Thermal efficiency η, % Carbon emissions ν, kg/MWh Variable operating costs V, £/MWh Enrichment cost e, £/MWh Availability, % Conversion coefficient a nuclear 36 0 1.8 2.5 78 8.24 × 10−3 coal 36 285 2.0 0 86 150 CCGT 60 185 2.2 0 87 34.128 OCGT 46 185 2.7 0 95 34.128 Fig. 1Open in figure viewerPowerPoint Example electricity supply curve Historic demand data from National Grid is used to drive the model and this defines the power which conventional generation must serve in each time period. It is common when modelling markets with wind generation to deduct aggregate wind output time series (Section 2.2) from the underlying electricity demand time series; it is this 'net demand' that the remaining generation is dispatched to meet and the resulting intersection with the supply curve defines the market price. In practice only relatively small, embedded wind generation behaves as negative load with larger wind farms forecasting their output and trading in forward markets. As such, wind farm output would tend to adjust the supply function for each half-hour period. Thermal generation would be required to respond not only to changes in wind and demand, as well as forecast errors, indicating that the assumption that thermal plant is dispatched in merit order is a simplification. With significant market share of wind there is potential for market prices to not only be suppressed, but in certain cases to become negative [21]. First, wind may be the marginal generator and when in receipt of subsidy it may offer negative bids up to the subsidy level to avoid curtailment. Second, an inflexible baseload plant that otherwise would be shut down and re-started may seek to avoid the costs of doing so by offering negative bids to keep generating. The extent and occurrence of negative prices is, however, strongly dependent on a range of factors including the extent of wind generation, levels of demand and the specific subsidy regime in place [21]; they would be expected only at high levels of installed capacity and would tend to be relatively infrequent [21]. As the merit order model applied here does not account for the dynamics of generation dispatch, the assumption that no subsidies are paid for renewable generation means the minimum price of electricity never falls below zero. As such, the shape of the supply curve does not fundamentally change, remaining shallow at low net demand and steep during periods of high net demand. Historic Market Index Prices from the UK power exchange [22] were used to calibrate and validate the market model. Data from 2005 to 2007 [14] was used as it represents a period prior to the major increase in wind generation in GB. Time series of historic fuel and carbon prices were sourced from [23-26]. Price data at as high a temporal resolution as possible was used; in the case of gas, daily prices were found to substantially better capture underlying electricity market price behaviour. This was deemed credible as although generators will purchase most of their fuel at fixed prices on forward markets, they may also trade on daily gas markets to adjust their position; this leads daily gas prices to better represent the marginal behaviour of gas generation. The calibration was aimed at ensuring that the intraday spread in electricity prices – of critical importance to arbitrage revenue – was represented well by the model. The mean daily peak and trough prices are much more important for arbitrage than the extreme values and these were found to be captured well [14]. Across the 3 years, the absolute error (bias) for mean peak prices was £2.64/MWh and £5.64/MWh for mean trough values. When applied to the storage model (Section 2.3), the difference between the modelled and historic electricity prices resulted in revenue differing by at worst 10% in any one year and virtually zero on average. The quality of the fit is well demonstrated in Fig. 2 which shows the historic and simulated electricity prices for the first week in August 2007. Fig. 2Open in figure viewerPowerPoint Historic market prices and simulated electricity prices for first week in August 2007 2.2 Wind power production Aggregate wind production time series were based on high resolution hourly wind simulations for the UK and surrounding waters produced by Hawkins [27]. The DECC RESTATS planning database [28] was used to identify the location and capacity of existing and planned wind farms in GB with the site-specific wind speed series extracted at each location. Power output from each wind farm was calculated using an equivalent aggregate power curve described in [27]. Depending on the assumptions about wind deployment this allows aggregate production for onshore and offshore wind fleets to reflect the diversity of wind speeds across the UK. Aggregate production was reduced by 10% to account for availability, a conservative assumption onshore but early offshore availability was less than 90% [29]. The data was interpolated linearly to obtain a time series of wind power output at half-hourly intervals. To represent scenarios with larger amounts of wind generation, the capacities at existing wind farm locations were scaled up; while this does not fully reflect the spatial diversity promoted by larger, more distributed wind fleets it is adequate for the purposes of this paper and not expected to substantially alter the results. Although the analysis presented here focuses on wind, it could conceivably be extended to other variable renewable generation such as solar PV, wave and tidal using similar atmospheric or oceanographic modelling techniques. 2.3 Energy arbitrage model The time series of electricity prices formed an input for the storage arbitrage revenue model. The revenue was calculated using linear optimisation [30] which determines the quantity of electricity bought and sold during each period, subject to the constraints of the storage capacity, maximum charging/discharging rates as well as efficiencies for conversion (the round-trip ratio of energy delivered to energy consumed) and storage (which measures self-discharge of the device). The model assumes the storage operator has perfect foresight of electricity prices; previous work showed minimal reduction in revenue using practical operating strategies compared to perfect foresight [31]. The storage device was assumed to be small relative to the total capacity in the market and its operation did not affect the price of electricity. Further details on the optimisation can be found in [14]. 3 Sensitivity study The sensitivity study was conducted by individually adjusting key 'external' parameters from initial baseline values to investigate the impact of each factor on arbitrage revenue. These parameters included gas and carbon prices, average capacity margin and installed wind capacity. In a future energy system these would not vary independently of each other and additional variables, such as thermal generation capacity and underlying patterns of demand, would also change; however, these were kept constant to investigate each effect in isolation and gauge its significance. The baseline case used historic data from 2006, including time series of fuel [25] and carbon [26] prices, generation capacity [32], demand times series [33] and wind speed time series [27]. This ensured a degree of coherence in the underlying data. The installed capacity of each class of generator is: 12 GW nuclear, 26 GW coal, 22.6 GW CCGT, 12 GW OCGT, 1.9 GW onshore wind and 300 MW offshore wind [32]. Installed wind capacity was less than 3% of the total generation capacity and typically, coal generation was dispatched before CCGT in the merit order. Initially, the storage characteristics were fixed at the baseline values listed in Table 3. These depict a moderate scale device with a power-to-storage ratio of 1:10, reasonable round trip efficiency and no other losses. Table 3. Baseline storage characteristics Storage constraint Unit Value maximum storage capacity MWh 200 maximum charging/discharging rate MW 20 conversion efficiency (round trip) % 75 storage efficiency %/day 100 4 Market variables 4.1 Gas price The 2014 National Grid FES [34] estimates that, in a high price scenario, gas prices would be slightly less than £1/therm by 2035. The arbitrage algorithm was therefore run for simulated electricity prices with average gas prices increasing from 10 p/therm (£3.41/MWh) to £1/therm (£34.13/MWh). Gas prices are volatile and have varied over this range of values in the last ten years. For reference, the average gas price in 2015 was ∼50 p/therm (£17.07/MWh) [25]. For each average gas price, the remaining inputs from the baseline year were used and the electricity price simulated at each half hour for 365 days to enable the annual revenue to be determined. The time series of historic gas prices from the 2006 baseline year was scaled in each case to maintain a constant intra-annual volatility of gas prices for each run. Fig. 3a shows that for gas prices greater than 30 p/therm the arbitrage revenue increased approximately linearly with gas price. Gas turbines were the most expensive thermal generators dispatched and their marginal prices set the daily peak electricity prices. Fig. 3b shows the storage device state of charge over a two-week period with an average gas price of 40 p/therm and £1/therm. This shows that the optimum operating schedule in both cases was almost identical. The device charged and discharged on a daily basis in line with daily demand cycles. Wind power output had little influence on electricity prices compared with variations in demand as the baseline installed capacity was small. Despite the similar storage operational pattern, the higher gas price led to a larger daily price spread enabling more revenue to be made during each cycle. Fig. 3Open in figure viewerPowerPoint Impact of gas prices on (a) Annual storage revenue with range of prices and, (b) Storage state of charge for two winter weeks with gas price of 40 p/therm and £1/therm Interestingly, for the lowest gas prices, revenue increased. This was because the lowest gas prices reduced the marginal generation cost of gas sufficiently that it became cheaper than coal for some periods. During these periods, coal was the marginal generator and CCGT contributed to base load during some off peak hours. Lower gas prices reduced the price of off peak generation which increased the daily price spread, enabling more revenue to be achieved during a storage cycle. Fig. 4a shows the marginal generation costs across the year with an average gas price of 10 p/therm. This shows periods where coal had the highest marginal generation cost and was dispatched as peaking plant. Fig. 4b shows the marginal generation costs with an average gas price of £1/therm, which shows that OCGTs were the most expensive generator for all periods of the year. The variation in marginal costs for each generator type is a result of the time series of fuel prices used (daily gas, monthly coal and quarterly nuclear prices). Fig. 4Open in figure viewerPowerPoint Marginal generation costs with average gas prices of (a) 10 p/therm and, (b) £1/therm 4.2 Carbon price Historically, the carbon price has always been below £20/tonne [26] and the 2014 National Grid FES [34] estimate that by 2035 it could increase to between £30/tonne and £75/tonne. The analysis was repeated for average carbon prices from £10/tonne (£0.01/kg) to £100/tonne (£0.1/kg). The time series of carbon prices within the year was scaled from 2006 to the average value. Fig. 5a shows that revenue increased with carbon price, albeit at a diminishing rate. Increasing the carbon price increased both gas and coal marginal generation costs, but did not affect nuclear generation costs. For many periods, increasing the carbon price raised the daily peak electricity prices, increasing the price spread and enabling the storage device to gain additional revenue. For other off peak periods the second marginal generator – commonly coal – was required, which set the off peak electricity prices. Increasing the carbon price increased coal generation costs more significantly than gas generation costs, reducing the price spread during periods where coal was required for off peak generation. As the carbon price increased further the price spread – and opportunity for arbitrage – was reduced during these periods, leading to diminishing gains in revenue. Fig. 5Open in figure viewerPowerPoint Variation in annual storage revenue with (a) Carbon price and, (b) Average annual demand as a proxy for capacity margin The storage device followed a similar strategy to that shown in Fig. 3b over the range of carbon prices investigated. The storage revenue was significantly less sensitive to carbon price than to gas price. There was an increase in revenue of less than 30% with an order of magnitude increase in carbon price (from £10 to £100/tonne). This compared with an increase in revenue of over 125% for an order of magnitude increase in gas price (from 10 p to £1/therm). This demonstrates the relatively modest influence of the current range of expected carbon prices on arbitrage revenue compared to the impact of gas prices. 4.3 Capacity margin Retaining the underlying pattern of demand from 2006 and with peak demand kept constant, average demand was varied from 30 to 50 GW to investigate the impact on storage revenue. Generation capacity was fixed at 2006 levels, so increasing demand represented a reduction in the average capacity margin. In a competitive market, this would lead to increased electricity prices incentivising investors to build more generators. This would, in turn, restore a greater average capacity margin and reduce prices restoring market equilibrium. The static market model used does not reflect these changes, but allows variations in demand to be a proxy for the capacity margin. Fig. 5b indicates annual revenue increasing as capacity margin falls. For low average demand, representing a high average capacity margin, commonly the low merit order generators were able to serve demand throughout the day. This was delivered by the left-hand side of the supply curve shown in Fig. 1. In this region, prices are low and price elasticity of supply is also low, demonstrated by the shallow curve, resulting in a small price spread. As demand grew, reducing the capacity margin, the higher merit order generators including peaking plant were required. This was delivered through generation represented by the right-hand side of the supply curve. Here, prices are higher, but price elasticity of supply is also higher, demonstrated by the steep shape of the curve. As a result, for the same daily variation in demand, the price spread was increasingly larger enabling higher revenue to be achieved. As the pattern of demand remained unchanged, the optimum operation strategy was similar across the range of average demand investigated, comparable to that shown in Fig. 3b. 4.4 Installed wind capacity The arbitrage model was run for installed wind capacity increasing from 0 to 40 GW. The ratio of offshore to onshore capacity was fixed at 3:2. Again, the remaining inputs, including the wind speed distributions, were taken from the 2006 baseline year. Retaining all other generation capacity as per 2006, 40 GW of installed wind represents 35% of the total generation capacity in GB. Fig. 6a shows that the revenue reduced as the wind capacity increased. This was due to lower variation in electricity price with increased wind power output. To illustrate this Fig. 6b shows the wind power output for two winter weeks with 40 GW of installed wind capacity and Fig. 6c the resulting electricity prices for cases of 40 GW and no installed wind. Prices were similar for both scenarios between days 8 and 9 when the wind power output was nearest to zero. With 40 GW of installed wind capacity, peak prices were significantly reduced during periods of high wind power output, which led generally to lower price variation. This is a result of the shape of the supply curve (Fig. 1) which was steep during periods of low wind production, but shallow during periods of high wind production. 40 GW of wind capacity reduced scarcity of supply, leading to reduced average prices and although there was increased variation in thermal output there was reduced variation in price; these led to decreasing arbitrage revenue. This is shown clearly in the annual price duration curves (Fig. 6d) with 0 and 40 GW of installed wind capacity. Prices are generally suppressed with 40 GW wind including at extreme low net demand. Higher peak prices have been suggested to be a natural outcome in a system with high penetrations of wind capacity wher