Artificial Neural Networks for Short-term Wind Power Estimation

Chaimae Zedak, Abdelaziz Belfqih, Faissal El Mariami, Jamal Boukherouaa, Abdelmajid Berdai and

Anass Lekbich

Energy and Electrical Systems Laboratory, National Higher School of Electricity and Mechanics, Hassan II University,

Casablanca, Morocco

Keywords: Artificial neural networks; performance; short-term forecasting; wind power; wind speed.

Abstract: Wind energy forecasting is an important part of the electrical system because of its intermittent nature. It

has become a challenge for many researchers to find the most accurate prediction method since an accurate,

reasonable and scientific forecasting of electrical power is a critical step in planning the electricity grid,

maintaining the supply-demand balance and more generally forming a scientific basis for the energy

planning. This paper presents the prediction of wind power by applying the technique of neural networks to

the power data of a wind farm in Spain with wind speed and wind direction data as these two parameters

have an influence on wind power. The performance of the proposed neural network was evaluated

according to the regression coefficient R and the Root Mean Square Error (RMSE) and by comparing the

one hour ahead predicted values of wind power for May 31 to the real available values.

1 INTRODUCTION

In recent years, and because of the depletion of

conventional sources of production and the

environmental constraints, renewable energies have

become the focus of interest for many researchers.

These energies have been strongly integrated into

the electricity sector as a kind of non-polluting

natural source.

Wind energy has experienced strong growth in

several areas and several countries. This energy

depends at all times on the speed of the wind and

therefore, knowing and predicting the wind potential

is related to changes in wind speed and many other

parameters.

Wind farm operators are still seeking to plan the

distribution of energy and have it available hours in

advance in order to be able to adapt wind generation

in an efficient manner at the right time and also to

determine reserve capacity and the penetration of

wind energy.

In the literature, many methods were used to

forecast wind energy in the long, medium and short

term. These methods can be divided into two

categories; conventional methods and artificial

intelligence. Some researchers have proven the

effectiveness of conventional methods in prediction

while others were curious about the performance of

artificial intelligence methods and there are even

those who have worked on hybrid models to

improve the accuracy of the predictive model.

In (Do-Young et al., 2016), a new approach of

forecasting with multi-variable inputs was proposed

where wind speed was estimated and used to predict

wind potential. The results show that the proposed

method is more accurate than the traditional

methods. A comparison between neural networks

and stochastic time-series model of ARIMA was

done in (Anurag and Deo, 2003) when forecasting

wind speed over varying periods of time. The neural

networks forecasting was much better and more

accurate than ARIMA models. Ma, L. (Ma et al.,

2009) gives a bibliographical review on the

researches done in the fields of wind and generated

power forecasting. Thanasis, G. B. (Thanasis et al.,

2006) employed three local recurrent neural

networks to predict the wind speed and power of a

wind park on the Greek island. Erasmo, C. (Erasmo

and Wilfrido, 2009) used different structures of

Zedak, C., Belfqih, A., El Mariami, F., Boukherouaa, J., Berdai, A. and Lekbich, A.

Artiﬁcial Neural Networks for Short-term Wind Power Estimation.

DOI: 10.5220/0009776400210025

In Proceedings of the 1st International Conference of Computer Science and Renewable Energies (ICCSRE 2018), pages 21-25

ISBN: 978-989-758-431-2

neural networks to predict short-term wind speed in

three regions in Mexico. The two layers model with

two inputs neurons and one output neuron gave

the best results. Hao, Q. (Hao et al., 2013)

implemented a neural network- based method for

constructing prediction intervals in order to forecast

short-term load and wind power. Kanna, B. (Kanna

and Sri, 2012) proposed a two-step approach; the

forecasting of wind speed up to 30 h ahead using

wavelet neural networks was phase I. During phase

II, a nonlinear mapping was performed between

wind speed and power to transform the predicted

values of speed into wind energy. The results show

that the proposed approach was persistent and

outperforms the benchmark models. João, P. S. C.

(João et al., 2009) evaluated the accuracy of the

approach he proposed, to predict short-term wind

power in Portugal using neural networks, for a real-

world case study and he finds that the desired

accuracy was achieved. Chinnawat, S. (Chinnawat

and Wanchen, 2015) used and tested ten different

neural networks to forecast wind speed for two

forecast times (3 and 6 hours ahead) and two

altitudes and the best neural network was chosen.

Shih-Hua, H. (Shih-Hua et al., 2015) used neural

networks to design a system for wind power

forecasting. Ankita, S. (Ankita et al., 2016) validated

two neural network models for wind speed and wind

power forecasting as accurate and performant

models. Lei, X. (Lei and Jiandong, 2016) applied

Particle Swarm Optimization to optimize the Elman

Neural Network for an effective model that gives

accurate forecasts of short-term wind power.

Yicong, W. (Yicong, 2014) proposed to predict wind

energy using genetic algorithms combined with

wavelet neural networks because of their high

accuracy and efficiency. Senthil, K. P. (Senthil and

Daphne, 2016) evaluated the performance of feature

selection and bagging neural networks in wind speed

forecasting. Qianyao, X. (Qianyao et al., 2015)

proposed a new model for short-term wind power

forecasting that adjusted the weather data inputs by

data mining. He proved that the proposed model

improves the forecast of wind energy.

This article studies the performance of neural

networks in predicting wind potential based on wind

speed and wind direction. And it is organized as

follows: Section 2 presents the method of neural

networks. The simulation and the results are

presented and discussed in section 3. And section 4

contains the conclusion of this work.

2 NEURAL NETWORKS

Inspired by the natural intelligence of the human

brain and its ability to solve complex problems,

artificial neural networks are structured as layers,

each layer contains several interconnected neurons

in order to transmit the signal to the output layer and

compare it to the desired output. They are like a

black box able to solve nonlinear problems between

inputs and outputs whose relationship is unknown.

Neural networks have the ability to learn from

past experiences and then develop the generalization

capacity from weight adjustment. The procedure is

as follows: firstly, a summation of the weighted

activation of the neurons is made, and then it goes

through the activation function to finally arrive at

the output of the neuron.

Figure 1: Neural Networks architecture

3 RESULTS AND DISCUSSION

In this paper, the hourly data used to forecast

wind power were collected from the Sotavento

experimental wind farm, consisting of 24 wind

turbines, from May 1

, 2018 to May 30, 2018. The

data has been normalized in the range [0, 1] in order

to minimize the error.

ICCSRE 2018 - International Conference of Computer Science and Renewable Energies

The following figures show the hourly data of

wind power, wind speed, and wind direction during

the month of May:

Figure 2: Hourly wind speed data

Figure 3: Hourly wind direction data

Figure 4: Hourly wind energy data

In this work, the multilayer feed-forward back

propagation network was the type of network used

with as inputs the wind speed and direction and wind

power as output. Only one hidden layer was

introduced with 20 hidden neurons. The activation

function used was the tan-sigmoid and the learning

algorithm was the Bayesian regulation.

Figure 5: The proposed neural network structure

After the training phase of the neural network,

the output values of the wind power have been

generated and therefore they can be compared to the

current values as shown in the following figure:

Figure 6: Actual vs. predicted values of wind

power

Artiﬁcial Neural Networks for Short-term Wind Power Estimation

In order to evaluate the model used for the wind

power forecasting, two performance indicators were

used:

• The determination coefficient R compares the

estimated values of the dependent variable against

its observed variables. It measures the predictive

quality of the model.

• The root mean square error RMSE is a

measure that determines the differences between the

predicted values and the actual ones. It is calculated

using the following equation:

(1)

Where n is the number of data (n=720)

The performance of the model is shown in the

following table:

Table 1: Performance of the model

Method R

MSE

Neural

Networks

0.8

0.0095

0.09

The value of the RMSE which is low and the

value of R which is close to 1, show that the model

is efficient. Moreover, the predicted values for the

31st day of May are close to the real values (see

Table II). This shows that neural networks give

promising results in the prediction of wind energy.

Table 2: Predicted values compared to the real values

Date

Real values

Predicted

values

(kwh) (kwh)

31/05/2018 00:00

900.05 835.63

31/05/2018 01:00

1164.8 1130.42

31/05/2018 02:00

1461.02 1035.29

31/05/2018 03:00

1031.95 1083.54

31/05/2018 04:00

858.73 639.84

31/05/2018 05:00

777.14 503.58

31/05/2018 06:00

376.04 211.02

31/05/2018 07:00 616.93 462.11

31/05/2018 08:00 104.24 176.22

31/05/2018 09:00 476.83 555.93

31/05/2018 10:00 1017.4 1196.3

31/05/2018 11:00 674.87 905.76

31/05/2018 12:00

1544.02 1216.94

31/05/2018 13:00

3017.44 1620.90

31/05/2018 14:00

1176.79 3357.58

31/05/2018 15:00

1041.41

3444.1

31/05/2018 16:00 142.55 143.49

31/05/2018 17:00 576.18 320.27

31/05/2018 18:00

1747.43

1351.003

31/05/2018 19:00

2067.49 1617.59

31/05/2018 20:00 2462.1 1726.7

31/05/2018 21:00

2497.02 2015.38

31/05/2018 22:00

1842.89 1077.47

31/05/2018 23:00 1857.6

1009.18

4 CONCLUSION

The integration of wind generation sources into

the electricity grid has grown in several countries

around the world. But with the intermittent nature of

the wind, several constraints have emerged, and

therefore, the forecast of wind energy is an essential

step that must be studied in order to manage the

electrical network. The purpose of this work is to

evaluate the performance of neural networks and

determine their predictive ability.

In this work, the prediction of wind power was

made by the neural networks method. The

performance of the model was evaluated by the

regression coefficient and the RMSE and the

prediction results were compared to the actual

values.

The results show that neural networks can be

considered as a fairly efficient forecasting method

with promising predicted values. These values can

be improved by combining the neural network with

other prediction methods to give more accurate

ICCSRE 2018 - International Conference of Computer Science and Renewable Energies

results and this would be our next work to study and

eventually manage the electrical network.

REFERENCES

Do-Young, H., Teng-Yu, J., Lili, Z., Mai Suan, L., and

QH, W. (2016). An Indirect Short-Term Wind Power

Forecast Approach with Multi-variable Inputs. In 2016

IEEE Innovative Smart Grid Technologies - Asia

(ISGT-Asia).

Anurag, M. and Deo, M. C. (2003). Forecasting wind with

neural networks. In Marine Structures 16 (2003) 35–

49.

Ma, L., Luan, S., Jiang, C., Liu, H., and Zhang, Y. (2009).

A review on the forecasting of wind speed and

generated power. In Renewable and Sustainable

Energy Reviews 13 (2009) 915–920.

Thanasis, G. B., John, B. T., Minas, C. A., and Petros, S.

D., (2006). Long-Term Wind Speed and Power

Forecasting Using Local Recurrent Neural Network

Models. In IEEE Transactions on Energy Conversion,

VOL. 21, NO. 1, March 2006.

Erasmo, C. and Wilfrido, R. (2009). Short term wind

speed forecasting in La Venta, Oaxaca, México, using

artificial neural networks. In Renewable Energy 34

(2009) 274–278.

Hao, Q., Dipti, S., and Abbas, K. (2013). Short-Term Load

and Wind Power Forecasting Using Neural Network-

Based Prediction Intervals. In IEEE

TRANSACTIONS ON NEURAL NETWORKS AND

LEARNING SYSTEMS.

Kanna, B., and Sri, N. S. (2012). AWNN-Assisted Wind

Power Forecasting Using Feed-Forward Neural

Network. In IEEE TRANSACTIONS ON

SUSTAINABLE ENERGY, VOL.3, NO.2,

APRIL2012.

João, P. S. C., Hugo, M. I. P., and Mendes, V. M. F.

(2009). An Artificial Neural Network Approach for

Short-Term Wind Power Forecasting in Portugal. In

2009 15th International Conference on Intelligent

System Applications to Power Systems.

Chinnawat, S., and Wanchen, W. (2015). A NEURAL

NETWORK-BASED WIND FORECASTING MODEL

FOR WIND POWER MANAGEMENT IN

NORTHEASTERN THAILAND. In 2015 IEEE

International Geoscience and Remote Sensing

Symposium (IGARSS).

Shih-Hua, H., Ko-Ming, M., Ping-Yuan, L., Chao-Yang,

T., Yih-Guang, L., and Li-Fen, C. (2015). The

Application of Neural Network in Wind Speed

Forecasting. In Proceedings of 2015 IEEE 12th

International Conference on Networking, Sensing and

Control.

Ankita, S., Gurtej, K., Gourav, J., Faraz, N., and

Markandey, M. T. (2016). Short Term Wind Speed

and Power Forecasting in Indian and UK Wind Power

Farms. In 2016 IEEE 7th Power India International

Conference (PIICON).

Lei, X., and Jiandong, M. (2016). Short-Term Wind Power

Forecasting Based on Elman Neural Network with

Particle Swarm Optimization. In 2016 Chinese Control

and Decision Conference (CCDC).

Yicong, W. (2014). Short-term Wind Power Forecasting

by Genetic Algorithm of Wavelet Neural Network. In

2014 International Conference on Information Science,

Electronics and Electrical Engineering.

Senthil, K. P., and Daphne, L. (2016). Forecasting of

Wind Speed Using Feature Selection and Neural

Networks. In INTERNATIONAL JOURNAL of

RENEWABLE ENERGY RESEARCH, Vol.6, No.3,

2016.

Qianyao, X., Dawei, H., and Ning, Z. (2015). A Short-

Term Wind Power Forecasting Approach With

Adjustment of Numerical Weather Prediction Input by

Data Mining. In IEEE TRANSACTIONS ON

SUSTAINABLE ENERGY, VOL. 6, NO. 4,

OCTOBER 2015.

Artiﬁcial Neural Networks for Short-term Wind Power Estimation