Intermittent and stochastic character of renewable energy sources: consequences, cost of intermittence and benefit of forecasting

HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés. Intermittent and stochastic character of renewable energy sources: consequences, cost of intermittence and benefit of forecasting Gilles Notton, Marie Laure Nivet, Cyril Voyant, Christophe Paoli, Christophe Darras, Fabrice Motte, Fouilloy Alexis


Introduction 30
The growth of the market of photovoltaic and wind energy systems over these last years is 31 always continuing with 50 GWp of PV plants and 62.7 GW of wind turbines installed in 2015 32 (+25% for PV and +22% for wind energy compared with 2014). Thus, the total capacity 33 respectively in Europe and in the World reached 94.6 GW and 227 GW for PV [1] and 141.7 34 GW and 432.56 GW for wind energy plants at the end of 2015 [2]. 35 As the part of electricity produced by PV and wind energy systems increases, the need for 36 these two intermittent and stochastic renewable energies systems (ISRES) to be fully integrated 37 into electricity grids arises. Thus, one of the main challenges for the near future global energy 38 supply is the high integration of renewable energy sources [3]. The stochastic and intermittent 39 behavior of solar and wind resources pose numerous problems to the electricity grid operator 40 which will be discussed in the first paragraph, these problems have then a negative impact on 41 the production cost. 42 As defined by the business dictionary in 2015 [4], "cost is usually a monetary valuation of 43 (1) effort, (2) material, (3) resources, (4) time and utilities consumed, (5) risks incurred, and (6) 44 opportunity forgone in production and delivery of a good or service". This definition may be 45 adapted to our problematic: cost is relative to an under or overproduction cost due to the random 46 and fluctuating variation of solar and wind resources what make less secure the electricity 47 production and distribution because not always available or non guaranteed. 48 Decreasing or smoothing these "unpredictable" variations need to use energy storages and 49 back-up energy production means able to compensate immediately the power variations; then, 50 backup generators must often stay switched-on for being able to maintain promptly the 51 production/consumption balance; moreover, PV and wind energy systems must sometimes be 52 switched off when their electrical production exceeds a certain percentage of the global 53 production. 54 It is obvious that such difficulties induced by the intermittence of wind speed and solar 55 radiation will lead to an additional production cost compared with conventional production. 56 Presenting costs is a very difficult task because it depends, on various parameters such as the 57 country and on legal incentives, on the situation of the electrical network (connected, partially 58 connected or remote grid), on meteorological conditions of the implementation site, etc. 59 The objective of this paper is to present an overview, affordable by non-economic 60 specialists, on intermittence extra-costs and on the positive influence of a reliable production 61 forecasting on the production cost for wind and solar production. This would allow to help to 62 justify future investments in the ISRES production forecasting in showing the benefits of 63 forecasting for utilities. Predicting with a good accuracy the electrical power produced by wind 64 or PV farms (and consumed by the load) allows to anticipate the actions of the electrical grid 65 operator, to improve the electricity balance management and especially to ensure better safety 66 of the electrical grid. 67 Predicting accurately the intermittence of renewable sources creates a cost-effective access 68 to these energy resources. The reasoning is as follows: the intermittence of solar and wind 69 resources is costly [5][6], sometimes very costly; a good forecasting of these intermittences 70 allows to manage more efficiently the overall electrical system; then, the negative cost impact 71 of these ISRES on the electrical network is decreased and at last, the cost effectiveness of PV 72 and wind energy systems is increased. 73 Evaluation and forecasting of ISRES power help developers of renewable energy power 74 plants to decide more easily where to install and how to operate them most efficiently by 75 reducing the use of conventional electricity production means as much as possible. 76 In this paper, we will answer to the following questions: 77 • Why does the integration of ISRES into an electrical grid pose technical problems to 78 the energy manager? 79 • Why is the price of the electricity not constant? 80 • Why do the variability and the behaviour of the solar and wind sources induce a cost  81 and what is the order of magnitude of this cost? 82 • Why does forecasting PV and wind production improve the management of the 83 electrical system and decrease the integration cost of ISRES? 84 This review paper syntheses the physical reasons to predict solar or wind fluctuations, it 85 shows that the variability and stochastic variation of ISRES have a cost, sometimes and often 86 high. It provides useful information on the intermittence cost and on the decreasing of this cost 87 due to an efficient forecasting of the renewable source fluctuation, for engineers and researchers 88 who are not necessarily familiar with the issue of the notions of cost and economy. 89

ISRES integration into an electrical grid 90
The uncertainty and variability of wind and solar resources pose problems for grid operators. 91 This variability requires additional and complex actions to balance the system. A greater 92 flexibility in the system is necessary to accommodate supply-side variability and the 93 relationship to generation levels and loads. 94 The electrical operator has often some difficulties to maintain the production/consumption 95 balance with conventional and manageable energy production means, mainly in small and/or 96 no interconnected electrical grid (as island ones). The reliability of the electrical system then 97 becomes dependent on its ability to accommodate expected and unexpected changes (in 98 production and consumption) and disturbances while maintaining quality and continuity of 99 service to the customers [7]. 100 Even if no ISRES are integrated in the electrical network, energy and power reserves are 101 needed, they can be divided in two categories: contingency reserve, used in case of specific 102 event (such as power plant switch-on) and no-event reserves used continuously (due, for 103 instance, to unreliable load prediction) [8]. These reserves (contingency and no-event ones) are 104 started at various time scales: within 1 minute (primary reserve) using spinning generators, from 105 1 min to 1 hour (secondary/tertiary reserves) and more than 1 hour [9]. ISRES introduction in 106 an electrical network only affects the non-event reserve particularly due to the imperfect 107 forecast of their production [8]. 108 Already, it appears that a predicted and anticipated event is easier to manage. The electrical 109 energy operator needs to know the future of the electrical production and consumption with 110 various temporal horizons ( Fig. 1) [10][11]. 111 Figure 1. Prediction scale for energy management in an electrical network [10][11]. 112 The integration of ISRES into an electrical network intensifies the complexity of the grid 113 management [10,[12][13]. The intermittence and the uncontrollability of ISRES production bring 114 also problems such as: voltages fluctuations, local power quality and stability issues [14][15][16]. 115 Sufficient energy resources in reserve are required to accommodate significant up or down 116 ramps in ISRES power generation to balance energy generated and energy consumed. When 117 ISRES power generation is available during low load levels, conventional generators need to 118 turn down to their minimum generation levels, with a bad efficiency and a high production cost. 119 Balancing the energy generated and the energy consumed at all times creates costs and even 120 more, if ISRES are integrated in the electrical network at a high level. 121 In case of a rapid decrease (or increase) of ISRES production, an instantaneous increase (or 122 decrease) of the delivered electrical power by a connected production mean has to occur and/or 123 a starting of a new production mean is needed; but the rise speed in power (ramp rate) of an 124 energy plant and its starting time is not instantaneous [17][18]. Then, an activation of a new 125 production system or a modification of the operating regime must be anticipated [7,17]. with an utilization of wind energy, conventional generators must meet the net load (net load = 128 demand minus wind energy) and, sometimes, this net load change or ramp is quicker than the 129 load alone; then, the remaining generators are operating at a low output level (called 130 "turndown") with a low efficiency [13,19], increasing the cost of electricity production, this is 131 another effect of intermittence on the extra cost. PV production is often more in line with load 132 [20] but during an evening load peak, the loss of a PV production after sunset increases the 133 ramping needs to balance the evening demand all the reserves must be increased by 7% (6-10%) of the installed wind capacity; 145 wind curtailments occur for a penetration rate up to 30% with a loss of production 146 between 0.4 and 3.5% of the wind energy production. 147 All these negative impacts have inevitably a consequence on the production cost. 148

Predicting ISRES production: a necessity for a better integration 149
Forecasting the output ISRES power systems is required for a good operating of the power 150 grid and for an optimal management of the energy fluxes occurring into the ISRES [22]. It is 151 necessary for estimating the reserves, for scheduling the power system, for congestion 152 management, for optimally managing the storage and for trading in the electricity market 153  (Table 1): 168 -Bulk energy storage or energy management storage used to decouple the timing of 169 generation and consumption. 170 -Distributed generation or bridging power, for peak shaving; the storage is used for 171 seconds to minutes and assures the continuity of service when switching from one 172 energy source to another. 173 -Power quality or end-use reliability. The stored energy is only applied for seconds 174 or less to assure continuity of quality power. 175 Similarly, the electrical operator needs to know the future production ( Thus, even in a conventional energy market, using only controllable energy means, the kWh 221 price varies greatly. It is already clear that knowing perfectly what will be the electrical 222 consumption (load) and production at various horizons will improve the management of the 223 various energy sources and will reduce the corresponding energy price. Centre (UKERC). A cost tag is lied to each of these characteristics, to compare them 247 economically [36] (Fig. 3). 248  Table 2. 255 Previous studies defined integration costs as "an increase in power system operating costs" 256  More spinning and stand-by-reserves ("uncertainty effect" Re-dispatch Market splitting à regional utilization/flexibility effects

Economic importance
Electricity is not a homogeneous good over-time (storage constraints)

Short-term response is costly
Electricity is not a homogeneous good across space (grid constraints)

Day-ahead spot market
Intraday and balancing power markets Nodal spot markets (or grid fees)

Price impact
Hourly price structure changes (e.g. lower prices during times of high NPRE in-feed) Regulating power/ balancing price increases Locational price structure changes (e.g. lower prices at nodes with much ISRES in-feed)

Profiles costs
Balancing cost Grid-related costs * Impacts on the power system and thermal plant operation for large-scale ISRES deployment. At small scale, the effect could be the opposite, These over-costs can also be divided into costs due to "system balancing impacts" and 278 "reliability impacts", the first one relative to rapid short term adjustments for managing 279 fluctuations from minute to hour and the second one to the uncertainties of production [13,53]. 280 The effect of the merit order on the ISRES kWh price vas analysed by Hirth [57] who shows 281 that the kWh price is all the more decreased than the installed ISRES capacity is high. or balancing costs is large: from 0 to 6 €/kWh for costs estimated from models with a 299 moderate increase with the ISRES penetration rate and from 0 to 13 €/MWh for 300 observed costs with no influence of the penetration rate. These gaps seems to be lied to 301 the peculiarities of the national markets; the need of an improvement of forecasting is 302 underlined for reducing these costs. 303 The ranges of integration costs are quasi similar for the three reviews: 0-6 €/kWh. 304 Higher costs were found: at high penetration rates, 30-40%, ISRES integration costs are 305 found to be between 25 and 35 €/MWh, i.e. up to 50% of generation costs [50].

Predicting for increasing the benefit of ISRES systems production. 344
As said in paragraph 2, the random production of ISRES systems causes stresses on the fossil 345 fuel generators, increasing the fuel generator cycling, decreasing their efficiency at low 346 operating regime and increasing the electricity production cost. Coal-fired thermal plants have 347 the highest cycling costs and many combustion turbines can have significant costs as well. 348 Hydropower turbines, internal combustion engines, and specially designed combustion turbines 349 have the lowest cycling costs [16]. Combustion turbine are well adapted for peak production 350 and can be started rapidly [7]. 351 Wind and solar power forecasting allows to reduce the uncertainty of variable renewable 352 generation. The use of forecasts helps grid operators more efficiently to commit or de-commit 353 generators to accommodate changes in ISRES generation and react to extreme events (ISRES 354 production or load consumption unusually high or low). Forecasts reduce too the amount of 355 operating reserves needed for the system, reducing costs of balancing the system. 356 Thus, using variable generation forecasts, grid operators can schedule and operate other 357 generating capacity efficiently, reducing fuel consumption, operation and maintenance costs, 358 and emissions as compared to simply letting variable generation "show up" [67]. 359

A COST Action (European Cooperation in Science and Technology) [68] on Weather 360
Intelligence for Renewable Energies (WIRE, ES1002) realized a bibliographical study; 361 concerning wind forecasting, the final document underlined "even though the necessity and 362 advantages of wind power forecasting are generally accepted, there are not many analyses that 363 have looked in detail into the benefits of forecasting for a utility". However, some positive and 364 important impacts were found in literature. 365 The uncertainty and/or forecasting error is a significant parameter in the integration costs 366 [69]. The lack of a good forecasting implies to use larger energy reserves which cannot be used 367 for other utilizations [70]. 368 Today, forecast errors generally range from 3% to 6% of rated capacity for a prediction one 369 hour ahead and 6% to 8% for a day ahead on a regional basis (higher errors for a single plant 370 due to the aggregate effect). In comparison, errors for forecasting load typically range from 1% 371 to 3% day-ahead [71], some progress stay to do. Day-ahead forecasts are used to make day-372 ahead unit commitment decisions and thus drive operational efficiency and cost savings. Short-373 term forecasts are used to take decision concerning a quick-start generator, demand response, 374 or other mitigating option and thus drive reliability. 375 When forecasting errors are reduced, ISRES production is predicted with more confidence, 376 then fewer reserves will be needed, reducing integration costs [67,72]. 377 The importance of a good forecasts was stated by the operations manager, Carl Hilger, from 378 Eltra forecasts their production on a day-to-day basis and they perceive the benefits around 4.50 396 £/MWh (6.93 €/MWh). The minimum size to justify the forecasting expense was 100 MW but 397 will be able to reach 10 MW rapidly. 398 For a 35% ISRES penetration, using a day-ahead generation forecasting reduces annual 399 operating costs by up to 5 G$ annually (3.6 G€), or 12 to 17 $ (8.64-12.2 €) per MWh of 400 renewable energy [79]. 401 The influence of an improvement of the forecasting reliability in the integration cost have 402 been studied in numerous papers: 403 a 1% MAE (Mean Absolute Error) improvement in a 6 h-ahead forecast had relatively 404 modest influence with an reduction of 972 k$ (748 k€) on 6 months (0.05% of the total 405 system cost) and a decrease of wind curtailments of about 35 GWh [80]. 406 a similar study realized on the basis of the Irish electricity system with a wind 407 penetration of 33% [81], concluded that an improvement from 8% to 4% in MAE saved 408 0.5% to 1.64% the total system costs and induces a curtailment reduction of 9%. 409 a wind forecasting improvements of 20% doubled the savings compared with a 10% 410 improvement [71] (Fig 4). Moreover, at low penetration levels (up to 15%), savings are 411 modest and for higher penetration levels (e.g., 24%); the savings is not linear versus the 412 forecasting improvement as noted also in [79]. In Fig 5,  The effects of a 100% perfect forecasting was sometimes studied and can be used as a 421 reference: 422 operating costs were reduced by 5 billion $/year by using a forecasting method and an 423 additional reduction of 500 million $/year (345 million €/year) [77]   forecasts are rare; a study [84] was realized for the 50 MW CSP system Andasol 3 in Spain and 464 concluded that the use of a statistical forecast model reduced the amount of penalties (due to 465 day-ahead market) by 47.6% compared with the use of a simple persistence model. 466

Conclusion 467
Solar and wind forecasting should be the first response to manage the variable nature of solar 468 or wind energy production, before the more costly strategies of energy storage and demand 469 response systems would be put in place. Furthermore, once a forecasting system is in place, it 470 provides additional benefits through the optimized use of these demand-side resources. 471 Even if the various studies analysed in this paper show a wide disparity about the integration 472 costs, due to definition of costs and calculation methods, due to applications to various 473 situations, various back-up systems, various integration rates, various meteorological 474 conditions, some general conclusions can be drawn: 475 the integration costs due to intermittence and variability of the production result from 476 the non guaranteed ISRES production imposes to electrical grid manager to take specific 477 measures for maintaining the production/load equilibrium. Some of these measures have 478 a negative impact on the operation of other energy production means; 479 these integration costs includes various sub-costs for which a good prediction of the 480 production has not the same influence; 481 these integration costs depend on the ISRES integration rate in the electrical network: 482 more the integration rate is high, more the integration cost is important and more the 483 influence of a good forecasting will benefit. 484 A reliable forecasting method both for wind and solar production will have very positive 485 influence on: 486 -the reduction of the integration costs; 487 -the decrease of the average annual operating costs; 488 -the decrease of the reserve shortfalls; 489 -the increase of the percentage reduction in curtailments of PV systems or wind turbines. 490 The improvements effects of a good forecasting depend of the integration level of the 491 renewable systems in the electrical network. 492 The improvement of the adequacy of the forecasting methodology was also studied (from 0 493 to the theoretical value of 100%): beyond a given percentage of improvement of the forecasting 494 model, his influence is reduced. 495 This review illustrates too that current state-of-the-art forecasts are likely to achieve most of 496 the economic benefits possible and that the interest for forecasting is increasing even for small 497 or medium ISRES. The energy storage development needs specific operating strategies for an 498 optimal management which cannot be developed without a good knowledge of the future input 499 and output energies.   Table  754 755