6806 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 16, NO. 11, NOVEMBER 2020
A Novel Wind Speed Interval Prediction
Based on Error Prediction Method
Geng Tang, Yifan Wu, Chaoshun Li , Member, IEEE, Pak Kin Wong , Zhihuai Xiao, and Xueli An
Abstract—Wind speed interval prediction plays an impor-
tant role in wind power generation. In this article, a new
interval construction model based on error prediction is
proposed. The variational mode decomposition is used to
decompose the complex wind speed time series into sim-
plified modes. Two types of GRU models are built for wind
speed prediction and error prediction. Prediction error for
each mode is given a weight and accumulated to obtain
the width of the prediction interval. The particle swarm
optimization algorithm is applied to search for the optimal
weights of the prediction errors. Experiments considering
eight cases from two wind fields are conducted by using
methods of interval construction in the literature for com-
parison with the proposed model. The result shows that the
proposed model can obtain prediction intervals with higher
quality.
Index Terms—Deep learning, error prediction, gated re-
current unit, interval prediction, wind speed prediction.
NOMENCLATURE
ANN Artificial neural network.
CWC Coverage width criterion.
EMD Empirical mode decomposition.
GD Gradient descent.
GRU Gated recurrent unit.
LSTM Long short-term memory.
LUBE Lower upper bound estimation.
MOO Multiple objective optimization.
Manuscript received June 9, 2019; revised December 18, 2019;
accepted January 27, 2020. Date of publication February 12, 2020;
date of current version July 29, 2020. This work was supported in
part by the National Natural Science Foundation of China under Grant
51879111 and Grant 51679095, in part by the Applied Fundamental
Frontier Project of Wuhan Science and Technology Bureau under Grant
2018010401011269, in part by the Hubei Provincial Natural Science
Foundation of China under Grant 2019CFA068, and the Fundamental
Research Funds for the Central Universities under Grant 2019kfyR-
CPY072. Paper no. TII-19-2429. (Corresponding author: Chaoshun Li.)
Geng Tang, Yifan Wu, and Chaoshun Li are with the School of Hy-
dropower and Information Engineering, Huazhong University of Science
and Technology, Wuhan 430074, China (e-mail: 2280608510@qq.com;
1095108207@qq.com; csli@hust.edu.cn).
Pak Kin Wong is with the Department of Electromechanical Engineer-
ing, Faculty of Science and Technology, University of Macau, Macau
999078, China (e-mail: fstpkw@um.edu.mo).
Zhihuai Xiao is with the College of Power Mechanical Engineer-
ing, Wuhan University, Wuhan 430072, China (e-mail: xiaozhihuai@
126.com).
Xueli An is with the Department of Hydraulic Machinery, China Insti-
tute of Water Resources and Hydropower Research, Beijing 100038,
China (e-mail: an_xueli@163.com).
Color versions of one or more of the figures in this article are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TII.2020.2973413
NN Neural network.
PI Prediction interval.
PICP PI coverage probability.
PINC PI nominal confidence.
PINRW PI normalized root-mean-square width.
PSO Particle swarm optimization.
RNN Recurrent neural network.
SA Simulated annealing.
SOO Single-objective optimization.
VMD Variational mode decomposition.
WSIP Wind speed interval prediction.
WSPP Wind speed point prediction.
WT Wavelet transform.
I. I
NTRODUCTION
O
VER the past few decades, renewable energy draws sig-
nificant attention owing to the gradual depletion of fossil
fuels and a serious rise in air pollution. One of the popular
alternative sources is wind energy. It is clean and recyclable,
but exhibits intermittent and stochastically fluctuating nature
at the same time, which brings challenges in harnessing wind
energy and to the reliability of wind power systems [1]–[4].
To optimize power generation, wind speed prediction is very
important. Currently, many research works on wind speed pre-
diction methods are pointing forecasting-based approaches [5],
[6]. This is, however, difficult to conquer owing to the difficulties
in eliminating the forecasting errors. For a more comprehen-
sive reference to the planning and operation of power systems,
quantifying the uncertainty associated with wind power, in other
words, using interval prediction is a more reasonable approach
in wind forecasting [7].
A number of conventional probabilistic interval prediction
approaches, such as the bootstrap quantile regression method
[8], the kernel density forecast method [9], and the Gaussian
process [10], [11], are available in the literature. However,
these methods are restricted with data distribution assumptions
and suffer from massive computational loads. To solve these
problems, Khosravi proposed a model called lower upper bound
estimation, which is an NN with two outputs for estimating
PI bounds [12]. By minimizing a PI-based objective function
that covers both interval width and coverage probability, the
SOO-based LUBE method is faster and more reliable [13], [14].
Some scholars adopted both the indices of coverage probability
and interval width as objectives, and these MOO-based meth-
ods show improvement over single objective LUBE methods
[15], [16].
1551-3203 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: East China Jiaotong University. Downloaded on February 09,2021 at 13:48:27 UTC from IEEE Xplore. Restrictions apply.