Back to the programme printer.gif Print




Delegates are invited to meet and discuss with the poster presenters in this topic directly after the session 'How does the wind blow behind wind turbines and in wind farms?' taking place on Tuesday, 11 March 2014 at 16:30-18:00. The meet-the-authors will take place in the poster area.

Elvin Lemmens 3E, Belarus
Co-authors:
Elvin Lemmens (1) F P Rory Donnelly (1) Thanos Kyriazis (1)
(1) 3E, Brussels, Belarus

Printer friendly version: printer.gif Print

Abstract

Determining uncertainty in wind resource assessments using long term tall mast datasets

Introduction

The greatest hurdle standing in the way of effective determination of uncertainty of long term wind resource assessment techniques is the lack of long term data with which to validate the techniques.
This research identifies and quantifies uncertainties in the long term resource assessment process using actual long term datasets from various tall masts around the globe, including masts in Germany, Belgium, Spain, the United States, Hungary, the offshore FINO 1 mast and others.

Approach

Samples of each tall mast dataset are used as on-site measurements in resource assessment procedures. A variety of extrapolation methods including an in-house residual resampling technique and a neural network technique are investigated, including the Wind Index method, regression MCP with various residual resampling methods, and Weibull Scale method and combined with a range of reanalysis datasets (MERRA, MERRA 50 m, CFSR, ECMWF) and nearby meteorological station data. The reconstructed yearly energy for each calculation is compared to the computed average yearly energy of the remainder of the mast dataset and the errors are quantified. From these errors, the uncertainties related to the overlap period, extrapolation methodology and long-term reference source are determined.
Temporal granularity of both site and reference data are varied to determine the minimum granularity for resource assessment purposes.
Uncertainties are assessed according to availability of data, site complexity, data granularity, length of the overlapping period and the long term reference source used. The effect of conducting the same study in different years is shown by using different years of the tall mast data as the on-site data.
The error over all sites is shown for each combination of extrapolation methodology and reference dataset using the optimum overlap parameters determined by the methodolgy described above.
Multivariate analysis using a selection of the available tall mast datasets yields a methodology by which uncertainty can be determined analytically using a limited range of site-specific factors. The resulting relationships are tested against two additional sites. This relationship allows the most apropriate methodology for each site to be utilised, resulting in increased confidence for industry stakeholders.

Main body of abstract

The application of these methodlogies on a range of sites allows quantification of theuncertainties associated with each methodology in each area. The sites include offshore, close to shore, simple terrain, complex terrain and wooded areas. Numerate factors affecting the accuracy of the long term resource predictions, such as temporal granularity of on-site and reference source datasets, are varied to assess quantitatively their impact on resource estimate errors. Static factors, such as site location and complexity are considered only in the final multivariate analysis to allow consultants to choose the most appropriate methodology for each site/study as required. The analyses can be broadly grouped as follows.

Reference dataset
Results show that generally Merra at 50 m is the most reliable reference source, consistently yielding the lowest reconstruction RMSE. Investigations into the effect of varying the temporal granularity of the data on extrapolated yield accuracy show that the high time resolution is not the main cause of the improved performance of Merra 50 m compared to other long term reference data sources.

MCP methodology
Figure 1 shows the median and range of the RMSE of the reconstruction error over all tested mast datasets for each of the tested extrapolation methods and reference data sources (overlap duration is held constant). 3E’s bootstrap residual resampling method gives the lowest RMSE over all datasets . The spreads and median values of this method shows similar results to the Gaussian and advanced Gaussian residual resampling but gives the lowest spread and especially the lowest median using MERRA at 50m.
This method is shown to be more sensitive to shorter overlap periods compared to other resampling techniques.

Overlap Period
Results also indicate that there is a strong and significant relationship between the inverse of the site overlap period and the RMSE of the reconstruction. This relationship varies from test to test as one test is more sensitive to the number of data points than the other. This implies that the advantage of a longer site overlap quickly becomes marginal, typically between 1 and 2 years of overlap. Also, this relationship is slightly dependent of the length of the reference period and the granularity of the reference source.
Climate variability is quantified by the variation in long term resource assessment estimate for each site and method based on the overlap year. Figure 2 shows the error in the long term resource estimate as the overlap (simulated on-site measurement) period is varied for the Cabauw site using a selection of methodologies.

Uncertainty Quantification
Multivariate analysis testing reveals a very strong fit for the two test sites. Specific care has been taken to avoid overfitting with these limited datasets available for the study. The number of datasets available in the world appropriate for such a study is limited, but the authors have tried to utilise as many as are available to give the strudy as much scientific robustness as possible.
Nevertheless, these results should be treated with caution due to the low number of test samples available, although the strength of the fit is encouraging for the further application of this technique.

Conclusion

This paper presents analyses of a range of resource assessment procedures and reference datasets and compares these to long term tall mast data taken from a variety of sites around the globe. Not only are the length of overlap periods (simulating length of on-site measurement campaigns) varied, but also the effect of the granularity of the on-site and reference data is investigated. From the results of these analyses, the length of the on-site measurement period can be optimised as can the granularity of the measured data. Increasing measurements beyond 1 - 2 years shows minimal increase in resource assessment accuracy. Minimum granularity of 6 hours is required to get good agreement against measured data.

The effect of the number of overlapping points is shown to be particularly important for some techniques, particularly those utilising linear regression. The best performance was found over all sites for a linear regression methodology utilising a residual resampling technique with 1 hour granularity and 1 year overlap.

The best long term data source was found to be Merra 50 m, already a standard in the industry. The high temporal granularity is shown to not be the reason for the good performance of this dataset by resmpling the data to 6 hours; this also results in very good resource predictions.
Multivariate analysis using a handful of variables reveals a strong relationship by which respource assessment uncertainty can be determined.

The robust analysese presented show the first use of global tall mast datasets to assess the uncertainty of long term resource assessment procedures. The identification of the best MCP methodology and reference data source over such a broad range of sites gives clear guidance to stakeholders in the selection of these parameters. Quanitification of uncertainty based on site specific factors allows a numerate assessment of uncertainty in this procedure; addressing a fundamental need in the sector at this time.


Learning objectives
From the results in this study, clear guidance can be proposed for selection of the best reference source and MCP methodology, and determining resource estimate uncertainty based on the circumstances of their case study.