Elsevier

Renewable Energy

Volume 35, Issue 6, June 2010, Pages 1236-1243
Renewable Energy

Wind forecasts for wind power generation using the Eta model

https://doi.org/10.1016/j.renene.2009.10.028Get rights and content

Abstract

The goal of this article is to apply the regional atmospheric numerical weather prediction Eta model and describe its performance in validation of the wind forecasts for wind power plants. Wind power generation depends on wind speed. Wind speed is converted into power through characteristic curve of a wind turbine. The forecasting of wind speed and wind power has the same principle.

Two sets of Eta model forecasts are made: one with a coarse resolution of 22 km, and another with a nested grid of 3.5 km, centered on the Nasudden power plants, (18.22°E, 57.07°N; 3 m) at island Gotland, Sweden. The coarse resolution forecasts were used for the boundary conditions of the nested runs. Verification is made for the nested grid model, for summers of 1996–1999, with a total number of 19 536 pairs of forecast and observed winds. The Eta model is compared against the wind observed at the nearest surface station and against the wind turbine tower 10 m wind. As a separate effort, the Eta model wind is compared against the wind from tower observations at a number of levels (38, 54, 75 and 96 m).

Four common measures of accuracy relative to observations - mean difference (bias), mean absolute difference, root mean square difference and correlation coefficient are evaluated. In addition, scatter plots of the observed and predicted pairs at 10 and 96 m are generated. Average overall results of the Eta model 10 m wind fits to tower observations are: mean difference (bias) of 0.48 m/s, mean absolute difference of 1.14 m/s, root mean square difference of 1.38 m/s, and the correlation coefficient of 0.79. Average values for the upper tower observation levels are the mean difference (bias) of 0.40 m/s; mean absolute difference of 1.46 m/s; root mean square difference of 1.84 m/s and the correlation coefficient of 0.80.

Introduction

Wind is “fuel” that drives wind turbines. However, this fuel cannot be simply switched on, because its availability is rather related to the prevailing meteorological conditions. Wind speed forecasts are important for the operation and maintenance of wind farms and their profitable integration into power grids. Integration of wind power into an electrical grid requires an estimate of the expected power from the wind farms at least one to two days in advance. Wind as the energy source has an intermittent nature. A grid operator must be able to balance the production and consumption of energy at very small time intervals. This is difficult, when the energy source that is feeding the electrical grid in unreliable. Economic value of the energy production among other components includes the prediction quality. These are all reasons why model predictions have an important role in handling of the wind power in electrical grids.

There are a lot of papers describing a wind-power forecasting systems using wind obtained from a numerical weather prediction (NWP) model. Lei et al. [1] in their paper give a bibliographical survey on the general background of research and developments in the fields of wind speed and wind power forecasting. Paper of Landberg et al. [2] gives an overview of the different methods used today for predicting the power output from wind farms on the 1–2 day time horizon. It describes the general set-up of such prediction systems and also gives examples of their performance. Landberg's paper [3] describes a model for prediction of the power produced by wind farms connected to the electrical grid. The physical basis of the model is the predictions generated from forecasts from the high-resolution limited area model (HIRLAM). Ramirez-Rosado et al. [4] in their paper present a comparison of two new advanced statistical short-term wind-power forecasting systems developed by two independent research teams. The input variables used in both systems were the same: forecasted meteorological variable values obtained from the NWP.

Wind is often considered as one of the most difficult meteorological parameters to forecast. Observed wind speed is a result of the complex interactions between large scale forcing mechanisms such as pressures and temperature gradients, the rotation of the earth and local characteristics of the surface [5].

We are aware of a number of studies that verify model prediction of winds, two of them verifying forecasts of the Eta model. O'Connor et al. [6] describe the operation and verification of a surge warning system for the Lake Erie, showing a high correlation of the predicted wind speed by the Eta model and the observed water levels. Nutter and Manobianco [7] describe an objective verification of the 29-km Eta model from May 1996 through January 1998. The evaluation was designed to assess the model's surface and upper-air point forecast accuracy at three selected locations during separate warm (May–August) and cool (October–January) season periods. Numerous studies are available concerning verification of the Eta model for a different regions and different variables (e.g., [8], [9], [10], [11], [12], [13], [14]).

The purpose of this study is to investigate how skillful is a regional atmospheric NWP, in our case the Eta model, in providing wind speed forecasts needed for wind power plants at a specific site. Verification is made over Scandinavia for a site in eastern Sweden, based on the data for summer seasons of 1996–1999. This study presents, to the authors' knowledge, the first evaluation of the Eta model in application for the wind energy.

The paper is organized as follows. Relationship between wind speed and wind power is described in section 2. A brief overview of the Eta model and its configuration as well as a verification tool are presented in section 3. Detailed statistical results describing the performance of Eta model wind forecast are presented in section 4 and the discussion is concluding the paper is summarized in section 5.

Section snippets

Relationship between wind speed and wind power

Wind power is directly related to the wind speed through a so-called power curve (Fig. 1). This is a simplified way of expressing the wind power in terms of atmospheric variables. Other atmospheric fields, such as wind shear, turbulence and air density have also impact on the actual power production for a given wind speed. However, for wind power verification wind speed is the most important parameter, because the bulk of the prediction error is caused by the wind speed prediction errors.

The

Model summary

Regional Eta model has been used for prediction of wind in this study. For the description of model dynamical part see Mesinger et al. [15] and references therein. The model physics, with Mellor-Yamada level 2.5 turbulence closure above the lowest model layer and Mellor-Yamada level 2 surface layer, is described in Janjić [16].

Experiments set-up

Two versions of the model geometry in the experiments over Scandinavia have run (Fig. 2). One with a larger area and a coarse resolution, and another one with a smaller

Results

We present verification results of 12–36 h wind speed forecasts from the nested grid model over Gotland island initialized daily over the summer periods 22 June to 10 October 1996–1999. The initial times of forecasts are 12 UTC. Surface wind observations, obtained by standard wind measurements (10 min averaging), are available hourly. Power tower wind data are available at every minute. Model wind data are available at every time step. Results in this study are verified at every 3 h using mean

Conclusions

We have examined an application and performance of the Eta model in forecasting wind for the wind power plants. Four common measures of accuracy relative to observations – mean difference (bias), mean absolute difference, root mean square difference and correlation coefficient – are evaluated, were calculated for the nested grid model over Scandinavia for the Nasudden power plants.

The wind at 10 m level obtained from the Eta model forecast has been compared to the observed wind from the surface

Acknowledgments

We have benefited considerably from discussions with and comments from Dr. Fedor Mesinger and Dr. Miodrag Rančić. The authors are grateful to the reviewer for useful comments and suggestions. This study is partly supported by the Serbian Academy of Science and Art under Grant F-147 and partly by the Ministry for Science and Ecology of Serbia under Grant 146006.

References (17)

There are more references available in the full text version of this article.

Cited by (0)

View full text