Elsevier

Renewable Energy

Volume 50, February 2013, Pages 590-595
Renewable Energy

A hybrid strategy of short term wind power prediction

https://doi.org/10.1016/j.renene.2012.07.022Get rights and content

Abstract

Two different prediction methods are investigated for short term wind power prediction of a wind farm in this paper. The adopted strategies are individual artificial neural network (ANN) and hybrid strategy based on the physical and the statistical methods. The performance of two prediction methods is comprehensively compared. The calculated results show that the individual ANN prediction method can yield the prediction results quickly. The prediction accuracy is low and the root mean squared error (RMSE) is 10.67%. By contrast the hybrid prediction method operates costly and slowly. However, the prediction accuracy is high and the RMSE is 2.01%, less than 1/5 of that by individual ANN method. Meanwhile, it is found that the errors of the prediction have some relation with the wind speeds. The prediction errors are small when the wind speeds lower than 5 m/s or higher than 15 m/s. The reasons for such phenomena are also investigated.

Highlights

► The physical strategy and ANN technique are effectively integrated. ► The wind sources at each turbine, the variation of generator power curve to time and environment are considered. ► The final results of the whole wind farm are significantly reduced due to the averaging effect. ► Under the ANN based prediction, with the MAE values more than 4000 kW and the NRMSE values about 11%. ► Under the hybrid method, with the MAE less than 760 kW and the NRMSE around 2%.

Introduction

Due to the intermittent and irregular characteristics of wind, the generated wind power is usually uncontrollable and introduces detrimental effects to the power grid at the case of high penetration [1]. Several matured strategies of wind power prediction and management have been developed, especially in the countries with a high portion of wind power in its total capacity. Predictor of Risoe National Lab (Denmark) [2], WPMS of ISET (Germany) [3], AWS Turewind of ewind (America) [4] and so on represent the state of arts, while the current research of China is mainly focusing on short term wind speed prediction for a wind farm [5], [6], [7]. The research on wind power prediction is in the initial stage [8], [9], [10], [11]. In [10] and [11], the artificial neural network method was applied on the data from numerical weather prediction for short wind power prediction. The main reason for less research on wind power prediction in China is summarized as: (1) the numerical weather prediction system has not been widely installed in most wind farms and the accuracy of obtained wind speed, wind direction and other key factors is low both in time and space resolution; (2) the SCADA of some wind farms are not well developed and the data of every 10 min or per hour cannot be acquired.

The wind information like speed and direction are usually employed to predict the output power of a wind generator. However, there is no matured numerical weather prediction data available. Therefore, this paper employs the real wind speed, wind direction and local temperature of a certain wind farm in Inner Mongolia in replace of the numerical weather prediction data for prediction. In the meantime, the past output power of the wind farm is used for the training of artificial network and prediction of future output power.

Section snippets

Review of basic prediction strategies

The wind power prediction method can be classified as physical and statistics types. There is no need of historical operation data in physical strategy, and the prediction can be started just after the installation of the wind farm. However, the demerits are that the detailed topographic map, coordinates of the wind turbine locations, and power curves of generators are required. Moreover, the program is considerably complicated. The typical operation system is Predictor [2]. The obvious

Prediction strategy

In order to judge the optimal input model, the general two error criteria of MAE and NRMSE are employed for comparison [9]. The MAE can be defined asMAE=1Nn=1N|PmeasurePforecast|

The NRMSE is determined byNRMSE=1Nn=1N(PmeasurePforecastPrated)2Where Pmeasure,Pforecast andPrated represent the measured power, predicted power and rated power of all the wind turbine generators of a wind farm; N is the predicted numbers.

Analysis of prediction error

Beside the prediction models, the following three factors [14] also contribute to the accuracy of the prediction results: (1) the prediction accuracy of wind speed, wind direction and air density (or temperature); (2) the amplification or attenuation effect of the nonlinear generator power curve to wind speed; (3) the wind farm efficiency, including the each wind generator performance and the wind utilization.

By analyzing the prediction errors under different wind speed, the prediction error of

Conclusions

In this paper, short term wind power prediction is performed for a certain wind farm with 40 wind generators and the following conclusions can be obtained:

  • (1)

    The ANN based strategy is able to rapidly predict the output power of a wind farm and the mean square root error is approximate to 10%.

  • (2)

    By combining the advantages of physical prediction and statistical prediction techniques, the hybrid strategy exhibits high accuracy with mean square root error around 2% which is only 20% of that of ANN based

Acknowledgements

Project was supported by the National Natural Science Foundation of China (Grant No. 51007026).

References (14)

  • V. Kurkova

    Komlogorov's theorem and multi-layer neural networks

    Neural Networks

    (1992)
  • Liu Yongqian et al.

    Review on short-term wind power prediction

    Modern Electric Power

    (2007)
  • Landberg L. Prediktor: an on-line prediction system. EUWEC special topic conference wind power for the 21st century,...
  • Lange B, Rohrig K, Ernst B, Schlögl F, Cali Ü, Jursa R, et al. Wind power prediction in Germany: recent advances and...
  • Zack JW, Brower MC, Bailey BH. Validating of the for ewind model in wind forecasting application. EUWEC special topic...
  • Pan Difu et al.

    A wind speed forecasting optimization model for wind farms based on time series analysis and Kalman filter algorithm

    Power System Technology

    (2008)
  • Erasmo Cadenas et al.

    Short term wind speed forecasting in La Venta, Oaxaca, Mexico, using artificial neural networks

    Renewable Energy

    (2009)
There are more references available in the full text version of this article.

Cited by (104)

View all citing articles on Scopus
View full text