VALIDATION OF GH ENERGY AND UNCERTAINTY PREDICTIONS BY COMPARISON TO ACTUAL PRODUCTION

Transcription

1 VALIDATION OF GH ENERGY AND UNCERTAINTY PREDICTIONS BY COMPARISON TO ACTUAL PRODUCTION Andrew Tindal, Keir Harman, Clint Johnson, Adam Schwarz, Andrew Garrad, Garrad Hassan 1 INTRODUCTION Garrad Hassan (GH) has been predicting the energy production of wind farms for fourteen years. Predictions have now been produced for over 5, MW of plant internationally, and many of these projects have gone forward to construction and have now operated for considerable periods. In order to assess the accuracy of these predictions GH maintains an internal data base which allows the actual production of wind farms to be compared with pre-construction projections. Using the information within this data base GH has conducted a high level investigation of how these constructed wind farms have performed in relation to the original GH pre-construction predictions. This investigation has been designed to complement a range of more detailed validations that GH conducts on individual aspects of its methodologies and models. GH has previously published energy validation results [1]. This paper presents the latest validation results and it is GH s intention to continue to maintain the energy validation data base and to publish updated validation results. To overcome issues associated with different periods of data being available from the various wind farms, each year of actual production data has been considered separately, and compared against the GH net energy central estimate (P5) and 1 year 9 % probability of exceedence level (P9). It is the aim of this work to be able to evaluate as large a volume of validation data as possible. For many wind farms only high level data are available such as monthly sub-station meter readings with no detailed information on wind farm availability or performance. Wind farms with only high level data have been included within the analysis. However, where wind farms are known to have been affected by gross issues, for example very poor turbine or grid availability, or these issues are apparent from comparison with data from nearby wind farms, these wind farms have been excluded from the assessment. Such exclusion of wind farms from the data base is inevitably somewhat subjective, however, the results are also presented in this paper for the subset of data for which the availability is known. For historical reasons the vast majority of data available are from Europe and North America and these regions are therefore the focus of this assessment. 2 RESULTS FOR EUROPE AND NORTH AMERICA Results for the whole of Europe and North America have been considered. The data base includes results from 156 wind farms with operational periods which vary from 1 year to 14 years. There are currently a total of 51 wind farm years in the validation data base. The distribution of these data in terms of time and location is presented in Figure 1 below. It can be observed that the longest data sets are available from northern Europe, the greatest number of wind farm years is available from southern Europe and that there is an order of magnitude more data from Europe than from the USA. The specific breakdown within the data base is 59 % of wind farm years is from southern Europe, 31% from northern Europe and 1 % from the US. 1 of 12

2 22% % of population of wind farm production years 2% 18% 16% 14% 12% 1% 8% 6% 4% S Europe N Europe N America 2% % Year Figure 1 Distribution by region and year of validation information in the wind farm data base The distribution of annual energy production, relative to the GH central estimate, for the 51 wind farm years in the data base is presented in Figure 2. An equivalent distribution for the 322 wind farm years for which availability data are available and for which the energy production has been adjusted to reflect the pre-construction assumed availability level is presented in Figure 3. No of wind farm years Actual production GH Predicted distribution Wind farm years = 51 Average = 93.3% 5% 55% 6% 65% 7% 75% 8% 85% 9% 95% 1% 15% 11% 115% 12% 125% 13% 135% 14% 145% 15% Actual annual production / GH Predicted P5 Figure 2 Distribution of annual production relative to GH projected central estimates 2 of 12

3 No of wind farm years Actual production GH Predicted distribution Wind farm years = 322 Average = 93.6% 1 5% 55% 6% 65% 7% 75% 8% 85% 9% 95% 1% 15% 11% 115% 12% 125% 13% 135% 14% 145% 15% Actual annual production / GH Predicted P5 Figure 3 Distribution of annual production relative to GH projected central estimates including availability adjustment for a subset of 322 wind farm years The average ratio of actual to predicted wind farm energy production is presented within the figures and is summarised in Table 1 below which also includes the subset of the data base for which an availability adjustment can be made. The number of wind farm years which fell below P9 prediction levels was also assessed and these results are also included in Table 1 below. Whole data base 51 years Availability adjusted subset 322 years Average ratio actual / predicted 93.3% 93.6% Wind Farm Years below P9 energy level 21.2% 18.% Wind Farm Years below P95 energy level 11.6% 9.6% Table 1 Actual performance compared with pre-construction projections - Europe and North America It is apparent from the above results that for the validation data base as a whole, the currently available data indicates that the wind farms are, on average, underperforming compared with pre-construction estimates. Within this paper a detailed commentary is presented on the results observed for the US and UK subsets of the whole validation data base. It is considered that much of the commentary provided for the US and UK results will be relevant for the whole data base. However, as the results for the whole data base are strongly dependent on the energy production of wind farms in southern Europe, and GH has not yet attempted a more detailed reconciliation of this part of the data base, no detailed commentary is provided on the results for the whole population here. GH has submitted a paper for the 28 European Wind Energy Conference in which it is anticipated that a commentary on the southern European results will be provided and GH will then be in a position to provide a more detailed commentary on the results for the whole data base. 3 US VALIDATION RESULTS At an AWEA event in 26 Garrad Hassan presented similar validation results to those which are presented within this paper. It was noted at the time that there was a paucity of data from the US compared with the amount of European data in the data base. A request was made to the US industry players for more feedback of actual production data but substantial additional data have not been forthcoming. 3 of 12

4 In order to address this issue the validation reported here has used public domain data from the Energy Information Administration (EIA) [2] to augment data already in the GH data base. The use of the EIA data requires caution as the EIA data are very high level and do not include availability data or other detailed wind farm information. The EIA data were therefore carefully screened to include only data considered to be valid in the context of the energy validation. It is known that some wind farms have experienced severe grid curtailment issues or availability issues and a process was undertaken to identify such wind farms from the data trends and, where the data trends indicated likely strong influences from availability or grid curtailment such data were excluded from the analysis. It is noted, for several wind farms, that energy losses from such causes will have been subject to revenue reimbursement. It is accepted that without comprehensive and detailed availability data the exclusion and inclusion of data within the validation data base is somewhat subjective. The use of the EIA data in combination with the existing data base gave 55 wind farm years on which to base the US analysis. This quantity of data is somewhat disappointing given the 12, MW of installed wind capacity now operating in the US and considering GH has undertaken more than 6, MW of pre-construction energy assessments in the US. Figure 4 presents the cumulative installed capacity in the US and the pattern of installation helps to explain, to an extent, why there is less data in the data base than initially might be expected. The first 2 MW of installed capacity in the US is mostly old technology which is not particularly representative of modern wind farm installations. The value of such wind farms for this energy validation is reduced and indeed for many of these projects a suitable pre-construction assessment has not been undertaken by GH. Additionally, the most recently installed wind farms either have little or no production data to include in the assessment. They simply have not yet operated for a long enough period. A further issue is that the first year or years of operation of a wind farm are often accompanied by a ramp up in availability before reaching mature operation. Given these factors it is wind farms which were constructed in the 2 to 24 period which provide the majority of data within the data base. It is also noted that many recent assessments which GH has undertaken in the US have been based on wind farm operational data. Such wind farm analyses, while certainly of interest, are not the subject of this paper and are also subject to less uncertainty than the classical pre-construction calculations. US MW installed (MW) Data too old Key area for data Data too new Year Figure 4 Cumulative installed capacity of wind power in the US The distribution of annual energy production, relative to GH central estimate for the 55 US wind farm years in the data base is presented in Figure 5. Clearly, with only 55 wind farm years in the US data, caution needs to be exercised in the interpretation of these results. 4 of 12

5 No of wind farm years Actual production GH Predicted distribution Wind farm years = 55 Average = 92.1% 4 2 5% 55% 6% 65% 7% 75% 8% 85% 9% 95% 1% 15% 11% 115% 12% 125% 13% 135% 14% 145% 15% Actual annual production / GH Predicted P5 Figure 5 Distribution of annual production relative to GH projected central estimates for US wind farms The above result along with the number of wind farm years below the P9 level is summarised in Table 2 below. US Wind Farms 55 years Average ratio actual / predicted 92.1% Wind Farm Years below P9 energy level 23.6% Table 2 Actual performance compared with pre-construction projections US wind farms It is clear from the above that the validation data base results indicate that estimates have typically been overpredictions with the average production being 92.1 % of the GH central estimate. Also 23.6 % of wind farm years rather than the expected 1 % of wind farm years lie below the P9 value. GH does not consider there are sufficiently detailed wind farm data to allow an attempt to provide a comprehensive reconciliation of the causes of the observed result. However, there follows a discussion of what are likely to have been key issues which have contributed to the observed result. Availability GH does not have the necessary detailed availability data to reconcile fully the impact of individual wind farm availability on the validation result shown. However, there are indications from a broader data base of US wind farm production data, which includes wind farms not in the validation data set reported here, that in recent years many US wind farms have experienced poor availability and the indications are from these data that the average result may be approximately 93 %. This observation is also supported by anecdotal comments. If the 93 % average availability estimate is accurate, it starkly contrasts with the typical experience of wind farms in Europe where substantially higher availability levels are the norm. Figure 6 below presents actual wind farm availability data against time for approximately 8 wind farms, the majority of which are in Europe. The data are available for up to 8 years although it is noted that the number of wind farms included is less for the longer periods than for the shorter periods. A ramp up in availability can be observed in the first year or so. Thereafter average availability levels typically exceed 97 %. 5 of 12

6 1% % 7 Average availability 98% 97% 96% Number of wind farms in sample Average monthly availability 3 month moving average of availability 12 month moving average of availability No of wind farms in sample 2 95% 1 94% Years of operation since commissioning Figure 6 Variation of average wind farm availability with time from commissioning The above result shows that wind farms routinely can and do achieve availability levels of 97 %. This figure illustrates that a key issue for the US industry is to improve the average availability which is being achieved at wind farm projects. It is also appropriate to review loss factor assumptions in the light of the observed results. Power performance Most wind turbine models have a sales power curve which is supported by one or more examples of a power curve measured on a test site by an independent third party to international standards. However, there is more to be learnt about how turbines, which may well have passed such tests, perform on sites where the terrain is complex and/or the meteorological conditions are significantly different from those at typical wind turbine power performance test sites. In addition the influence of dirt accretion on blades, degradation of blades and performance being influenced by ice accretion in winter need to be considered. Modern wind turbines are controlled by control algorithms and changes and upgrades to software are common. Small changes in the control algorithms of machines can significantly influence the power curve. An example is given below showing the power curve of a wind turbine on a monthly basis for a period of about a year. For several months degraded performance is observed as a result of an altered control algorithm. Also shown is an example of severe insect accretion on the blades of a turbine. 6 of 12

7 Power [kw] Feb-2 Mar-2 Apr-2 May-2 Jun-2 Jul-2 Monthly Aug-2 power curves Sep-2 using Oct-2 standard SCADA Nov-2 Dec-2 Jan-3 data incorrect Feb-3 Mar-3 change Apr-3 to control May-3 algorithm Jun-3revealed Jul-3 Aug-3 Sep-3 Oct-3 Nov-3 Dec wind speed [m/s] Figure 7 Real world issues for turbine power performance It is certain that such real world issues will have influenced the energy production of some wind farms in the data base. The data within the validation data base do not allow the magnitude of such influences to be quantified and compared with pre-construction loss factor estimates. Wind measurements anemometer mounting As indicated above it is wind farms which were constructed in the 2 to 24 period which have provided the majority of data in the validation data base. Energy assessment reports will have been produced for such wind farms before their construction and the wind data on which those reports were based were recorded earlier still. Therefore a significant amount of data in the energy validation data base is base on wind data dating back into the 199s. Wind speed measurement practices in the US have improved substantially in recent years. In particular new data about the influence of mounting arrangement of sensors came to light in 22. It has long been known that it is best practice to mount anemometers well clear from the support structure be that the met mast itself or a side mount boom. However, it was thought that the degradation in the accuracy of the anemometer with poor mounting arrangements was small. New measured data published in 22 indicated that the so called stub mount arrangement, where an anemometer is mounted on a small vertical stub above the tower, lead to an upward bias in wind speed recorded that increased the wind speed by to 4 % depending on the exact length of the stub, the exact location of lightning spikes and wind vanes and also dependant on the local wind regime. This influence is exacerbated as, if the mast is not a hub height mast, and the shear is estimated using a top stub mount anemometer the shear can be overestimated. A typical stub mount and a modern best practice arrangement are shown in Figure 8 below. 7 of 12

8 Figure 8 Stub mount anemometry mounting (left) and an example of current best practice mounting (right) When the bias in the wind speed, the impact on wind shear and sensitivity of wind turbine energy production to wind speed are considered, the impact of a severe stub mount effect could be to increase the energy production estimate by up to 5 to 1 %. It would not be true to say that all assessments of energy and production prior to 22 were influenced by stub mount effects, or where there were non-optimal mounting arrangements used, that the impact on wind speed and energy production was always as large as the worst case range of 5 to 1 % defined above. However such poor mounting arrangements are likely to have had an upward bias on some of the energy assessment included in the data base and will account for some of the discrepancy seen. The stub mount issue is a good illustration of the care which is required to make accurate wind speed measurements at a wind farm site and it is also an area where the industry has improved substantially over the past five years. Generally poor mounting arrangements are now a thing of the past although some examples of poor mounting are still encountered in the field. Wind measurements Number and heights of masts and duration of measurements The accuracy of the prediction of the energy production of a wind farm is strongly dependent on the number and height of masts and the duration of measurements available. In GH s experience over the past several years significant improvements are being made in the number, heights, and duration of measurements available for the assessment of wind farms. The availability of improved data sets allow some confidence that the results presented in this paper, which present assessments of the energy production of wind farms which includes data based on assessments more than five years old, will not necessarily be representative of current assessments. The industry is generally improving in this area although there are still exceptions. Windiness of recent years The relative windiness of recent years will certainly influence the validation result observed. The wind farms in the data base are located in many different regions of the US which experience very different wind regimes. It has not been possible to attempt to reliably correct for the observed windiness in different regions for the years over which actual production data are available. However, anecdotally in particular 25 was a low wind speed year in many areas relevant to the data base. As an example the wind data in recent years from a met station in Oklahoma is presented in Figure 9 below. It is considered that recent windiness in the US may explain a part of the underperformance of US wind farms presented above. 8 of 12

9 18% 16% Annual mean wind speed (%) 14% 12% 1% 98% 96% 94% 92% Year Figure 9 Normalized annual mean wind speed at a met station in Oklahoma 4 UK VALIDATION RESULTS A study was undertaken to validate predictions undertaken in the UK. The assessment of the UK data is of considerable interest as: GH s first energy predictions were for wind farms in the UK and therefore the longest potential data sets are available; GH has developed a windiness index for the UK and a single windiness index is reasonably representative of the whole of the UK which can be used to apply a windiness adjustment; Significant detailed production and performance data for UK wind farms are available; Some of the older UK wind farms have experienced good availability throughout their lifetimes. It is therefore possible to attempt to reconcile the causes behind any discrepancies observed in a more sophisticated way than is currently possible for other regions owing to the availability of the necessary information for this process. Data on annual substation metered energy from 27 wind farms have been compiled. The wind farms have been operational for between 1 and 14 years and are located across the UK. There is a total of 113 UK wind farm years in the data base of which for 34 wind farm years there is availability data. It is noted this is currently approximately twice the amount of data available in the US. A few wind farms have been excluded from the validation as they have experienced gross issues, such as extremely low availability. Also from some of the earliest assessments either very short met towers or very short periods of data were available for the original analysis. These were not considered to provide a useful comparison for modern assessments and were excluded from the assessment. The distributions of annual energy production, relative to GH central estimate for the 113 wind farm years in the data base have been derived and are presented in Figure 1 below. Summary results for the whole data base and also for a subset of the data base where availability and windiness adjustment can be applied ate presented in Table 3 below. 9 of 12

10 3 Actual metered production GH predicted distribution Number of wind farm years Wind farm years = 113 Average = 96.% 5 Figure 1 5% 55% 6% 65% 7% 75% 8% 85% 9% 95% 1% 15% 11% 115% 12% 125% 13% 135% 14% 145% 15% Actual production / GH P5 Distribution of annual production relative to GH projected central estimates for screened UK data set Whole UK data base 113 years Availability + windiness adjusted subset 34 years Average ratio actual / predicted 96.% 11.7% Wind Farm Years below P9 energy level 13.3% 2.% Table 3 UK wind farms - Actual performance compared with pre-construction projections for whole data base and for subset with availability and windiness adjustment It is apparent that for the UK data set the average production has been 96% of the expected production for the 113 wind farm year data base with 13.3% of wind farm years below the P9 value compared with the ideal 1% value. When the data set is adjusted for availability and windiness the actual production is 11.7% of expected with only 2% of wind farm years below P9 values. More details of the UK validation results will be presented in the 27 BWEA conference in Glasgow in early October. For the UK data set it can be concluded that if the data are screened to include only data representative of modern assessments and corrected for both availability and windiness then the actual energy data is in close agreement with the original pre-construction projections. While clearly it is the US data which are most relevant to the assessment of the energy prediction accuracy in the US, GH considers that some comfort can be gained from the UK data set in the context of other markets. 5 OVERALL CONCLUSIONS WITH A FOCUS ON US RESULTS GH maintains an internal energy production validation data base which contains actual wind farm production data and GH pre-construction energy projections. The data base contains only high level information which, as a minimum, includes monthly wind farm production and in some cases more detailed information such as availability. 1 of 12

11 It is only by looking at large volumes of data that a scientific view on the typical accuracy of predictions can be taken. This paper has presented the current results from the energy validation data base containing 51 wind farm years with 55 wind farm years in the US. Some wind farms where there have been gross issues such as grid curtailment or very poor availability have been excluded. The raw results show that predictions have, on average, been over estimates. For the full data base of 51 wind farm years the average actual production has been 93.3 % of pre-construction projections. For the US there are 55 wind farm years included within the data base and these indicate that average actual production has been 92.1 % of pre-construction projections. The main text provides a detailed commentary on likely causes of the observed discrepancy between actual and projected results. Perhaps the two most significant issues considered within the commentary are turbine availability and wind speed measurement quality. There is some significant evidence that typical availability levels in the US have been, on average, approximately 93 %. If accurate, such an availability is substantially lower than the availability level typically assumed in preconstruction energy assessments at the time of the original analysis and would, in part, explain the observed result. While this discrepancy indicates loss factor assumptions for current projects should be reviewed, the availability levels achieved in the US are in stark contrast to 97 % availability levels which are routinely being met in Europe. The European experience indicates a significant industry focus on achieving improved wind farm availability is merited and, if improved availability levels are achieved, the observed discrepancy in energy can be expected to narrow in future years. The energy validation data base includes assessments which are based on wind data recorded from 5 to 1 years ago. At this time it was common practice to mount anemometers insufficiently far from supporting booms and structures to measure the undisturbed free wind speed often known as the stub mount effect. In 22 new data came to light which indicated that poor mounting arrangements did not simply add a small degree of uncertainty to the measurements but could introduce a substantial upward bias into the wind measurements. The industry has now improved in this area. As well as better mounting arrangements in GH s experience (for most developments) there has also been an improvement in the number and heights of masts on sites. The quality of wind data may therefore be considered to have improved compared to the measurements for some of the earlier wind farms in the validation data base. A validation has been undertaken for UK wind farms where more data are available than from the US, better availability levels have been experienced and where it is possible to attempt to apply a windiness adjustment. It was found that for 113 wind farm years the average actual production to pre-construction prediction was 96. %, but for the subset of wind farm years for which an availability and windiness adjustment could be applied the actual to predicted result was 11.7 %. The UK results therefore give some confidence that for US wind farms when the influence of the issue of the stub mount effect reduces with more data in the validation data base and when existing and new wind farms experience a better match between actual and projected availability the discrepancy between the actual and projected energy production results should substantially reduce. Differences in typical wind farm size and wind regime between the US and the UK mean that it is important not to read too much into the UK in the UK result in the context of the US. The results of this validation exercise certainly demonstrate that the industry should not be complacent about wind farm energy projection and the need for high quality measurements. GH will continue to maintain and publish updates of the energy validation data base. Clearly the more data, and specifically the more detailed US data which are available, the better the validation and the more methods and models can be improved. The help of the US industry in achieving this is requested. There are many areas in which further work is merited; however key areas on which the industry (and GH) continue to work include: 1) Continue promoting high standards for monitoring campaigns; 2) Understanding loss factors through rigorous analysis of the data; 3) Validating advanced measurement techniques; 4) Understanding power performance in complex terrain; 11 of 12

12 5) Increasingly sophisticated wind flow and wake models. REFERENCES 1 Raftery P, Tindal A, Garrad A, Validation of GH energy and uncertainty predictions by comparison to actual production, Proceedings of the European Wind Energy Conference, London of 12