Short-Term Wind Energy Production Forecasting and Target Plant

Selection Based on Meteorological Data Using Empirical Mode

Decomposition

İsrafil Karadöl

Department of Electrical-Electronics Engineering, Faculty of Engineering-Architecture,

Kilis 7 Aralık University, Kilis, Turkey

Keywords: Wind Energy, XGBoost, EMD, İzmir.

Abstract: This study aims to identify the most suitable target wind power plant (WPP) for short-term wind energy

production forecasting. Hourly meteorological data for 2022 from İzmir Province were processed using the

Empirical Mode Decomposition (EMD) method to generate 56 Intrinsic Mode Function (IMF) signals, which

were used as input variables for the XGBoost model. As output, production data from 52 different WPPs

located within the same provincial boundaries were individually used as target variables. The model’s

performance was evaluated using R², MAE, and MSE metrics. The results indicated that while high prediction

accuracy was achieved for some plants, the model's performance was limited for others. The best forecast

accuracy was obtained using data from WPP35, whereas the poorest performance was observed with WPP7.

These findings suggest that, despite being within the same province, differences in the geographical locations

of meteorological stations and WPPs, as well as region-specific meteorological characteristics, can

significantly affect prediction accuracy.

1 INTRODUCTION

Globally, energy demand is increasing day by

day(Renewable Energy Agency, 2025). Meeting the

increasing energy demand from traditional energy sources

causes many environmental consequences. The depletion of

traditional energy sources is also recognized as a major

problem by energy producers(Karadöl & Şekkeli, 2022).

For these reasons, states have turned to renewable energy

sources as an alternative to traditional energy sources(Irena,

2025). The fact that renewable energy sources are

environmentally friendly and sustainable is seen as a great

advantage for governments. However, the fact that these

sources have random generation characteristics is

recognized as an economic and technical problem by

electricity grid operators(Karadöl et al., 2021). To

overcome this problem, forecasting energy production is of

great importance.

When we examine the research in the field of short-term

forecasting of wind energy, it is seen that many studies have

been carried out with various machine learning based

methods(Joseph et al., 2023; Liu et al., 2019; Mustaqeem et

al., 2022; Shukla & Pasari, 2025).In addition to these

studies, Li et al. proposed the XGBoost regression model

optimized by Genetic Algorithm (GA) to solve the accuracy

and speed problems in wind power forecasting(X. Li et al.,

2023). Ma et al. proposed the XGBoost algorithm for very

short-term wind power forecasting due to the variability of

wind power(Ma et al., 2020).

There is no research in the literature on wind

power generation forecasting for different targets at

1-hour horizon. To fill this gap in the literature, this

study examines the forecasting performance of the

XGBoost model for different target allocations at 1-

hour horizon. The meteorological data and WPP

generation data of Izmir province for the year 2022

are used to perform this investigation. All data sets

used are hourly resolution and real-time data. All

meteorological data obtained were decomposed into

signals by EMD method. As a result of this process,

56 signals were generated. The generated signals

were defined as input to the XGBoost model. With

this model, 52 WPP productions were defined as

targets in order to perform forecasts for a 1-hour

period. For each target, the forecasting performance

of the XGBoost model was analyzed. As a result of

these examinations, the best facilities for 1-hour

forecasts were determined.

164

Karadöl,

Short-Term Wind Energy Production Forecasting and Target Plant Selection Based on Meteorological Data Using Empirical Mode Decomposition.

DOI: 10.5220/0014299300004848

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 2nd International Conference on Advances in Electrical, Electronics, Energy, and Computer Sciences (ICEEECS 2025), pages 164-168

ISBN: 978-989-758-783-2

2 MATERIAL METHOD

In this study, it is aimed to determine the most

suitable target facility to realize wind power

generation forecasting with the XGBoost model at 1-

hour time horizon. For this purpose, firstly,

meteorological measurement data and WPP

generation data of Izmir province were obtained. The

obtained meteorological data were preprocessed by

EMD method and decomposed into IMF signals. As

a result of this decomposition, 56×8760 signals were

generated. These signals were used as input data for

the XGBoost model. With the XGBoost model, 52

WPP productions were selected as separate targets in

order to make WPP production forecasts with a time

horizon of 1 hour. R

, MAE, and MSE metrics were

used to evaluate the WPP production forecasting

performance according to the selected target plants.

According to these metrics, the best target WPPs in

Izmir province were determined for the 1-hour

horizon.

2.1 Meteorological Data

For the 1-hour time horizon, empirical mode-

disaggregated meteorological data sets are used to

determine the best target facility for wind power

generation forecasting. The meteorological data used

in the study are hourly resolution and have a 1-year

period (01.01.2022-31.12.2022). These data sets

consist of solar radiation, humidity, cloudiness, air

temperature, soil temperature, rainfall, wind direction,

and wind speed. Some statistical properties of these

meteorological parameters are given in Table 1.

Table 1: Statistical properties of meteorological parameters

Meteorological

parameters

Mean

Standard

Deviation

Maks. value Min value

Number

of data

Period

Cloudiness 2.34 2.56 8 0

8760

01.01.2022-

31.12.2022

Humidity 58.59 17.08 99 13

Radiation 211.19 300.86 1036.4 0

Wind rota 192.1 100.55 360 1

Wind speed 2.86 1.47 11.5 0

Air temperature 18.9 8.31 38.6 -0.1

Soil temperature 20.64 9.8 42.9 0.7

Rainfall 0.053 0.57 24 0

2.2 WPP Generation Data

In the research, 52 WPP production data located

within the borders of Izmir province were used. These

data are real-time and hourly resolution for the year

2022. The total installed capacity of the facilities from

which the data were obtained is 1742 MW. Within

this installed capacity, there are facilities with

different installed capacities, with a minimum

capacity of 3 MW and a maximum capacity of 226

MW. Co.

Figure 1: Total WPP production at 1 hour resolution

The total generation graph of these plants at hourly

resolution is given in Figure 1. WPP generation data used

in the study were obtained from Turkish Electricity

Transmission and Distribution

2.3 Empirical Mode Decomposition

(EMD)

The EMD method is based on the principle that

nonlinear and non-stationary complex signals can be

decomposed into sub-components with different

oscillation characteristics

(Jiang & Liu, 2023).

Based on this approach, Huang et al. proposed that a

complex signal can be decomposed into a residual

signal with a finite number of intrinsic mode

functions (IMFs)

(Huang et al., 1998). This method

has found wide application in many fields from

engineering to climate science, from biomedical

signal processing to energy systems, thanks to its

structure suitable for time-frequency analysis.

The EMD method is based on the assumption that

complex signals are composed of components that

oscillate at various frequencies and

amplitudes

(Shukla & Pasari, 2025). According to

this approach, the EMD decomposes a signal

sequentially into subcomponents called Intrinsic

Mode Functions (IMF) and a residual signal

(Shang

et al., 2022)

. IMF components reflect the short-term

Short-Term Wind Energy Production Forecasting and Target Plant Selection Based on Meteorological Data Using Empirical Mode

Decomposition

165

and high-frequency fluctuations of the signal, while

the residual signal refers to the lower frequency and

overall trend of the signal

(Aladağ, 2023; Yuzgec et

al., 2024). The mathematical expression of the

EMD method is given in Equation

1(RajasundrapandiyanLeebanon et al., 2025).In

this equation, Y

is the input signal, N is the

number of IMF signals, and R is the residual

signal.



= I𝑀𝐹







+𝑅(𝑡)

(1)

2.4 The Extreme Gradient Boosting

(XGBoost) Model

XGBoost is a supervised machine learning algorithm

based on decision trees, providing high accuracy and

high computational efficiency(Kanji & Das, 2025). It

is widely used in the literature due to its superior

performance, especially in classification and

regression problems (Hakkal & Lahcen, 2024; Joshi

et al., 2024; M. Li et al., 2020; Rathore et al., 2023;

Shi et al., 2021; Yelgeç & Bingöl, 2022; Zhang et al.,

2024). While models in traditional boosting

algorithms try to directly reduce the error values, in

the extreme gradient boosting approach, the model

focuses on the pseudo-residuals that are derivatives of

these errors(Demirer et al., 2024).In other words,

XGBoost is defined as an algorithm based on gradient

boosting, which is faster, more accurate, and more

robust to overlearning.

The gradient boosting method is based on

incrementally training successive decision trees in a

way that reduces the error rates in their previous

models(Demirer et al., 2024; X. Li et al., 2022). Each

new model focuses on the prediction errors of the

previous model and tries to correct them(Yan et al.,

2022). In this way, the generalization capability of the

model is increased, and more accurate predictions are

obtained.

In the literature analysis, it was found that the

XGBoost model is a highly effective approach for

modelling complex and dynamic systems such as

wind power generation forecasting due to its high

accuracy, flexibility, and computational

efficiency(Guan et al., 2023; W. Li et al., 2020; Phan

et al., 2021; Zheng & Wu, 2019).In this study,

XGBoost is used as the main modelling tool for short-

term power generation forecasts. The parameters of

the model used are given in Table 2.

Table 2: XGBoost model parameters

Parameters Value

estimators 400

Learnin

rate 0.1

Max_depth 6

Subsample 0.8

Colsample_bytree 0.8

Random

state 42

ective re

uarederro

3 RESULTS

3.1 Meteorological IMF Signals

The first stage of the research is the conversion of

meteorological data into IMF signals by the EMD

method. At this stage, each meteorological parameter

was converted into 7 IMF signals. A total of 56 IMF

signals were obtained from all meteorological

parameters. IMF signals of wind speed data are given

in Figure 2 as an example.

Figure 2: IMF signals of the wind speed data.

3.2 XGBoost Model WPP Forecasting

Results

In this study, the XGBoost model is used to predict

wind power plant (WPP) generation at 1-hour

forecast horizon and to determine the most suitable

target facility within the same province by using

meteorological measurement data. In this context,

meteorological data of Izmir province were

transformed into Internal Mode Functions (IMF)

using the Empirical Mode Decomposition (EMD)

method, and these signals were used as input data to

the XGBoost model.

To evaluate the performance of the model, 52

different WPP production data sets, each of which is

treated as a different target variable, are used

individually as the output of the model. For these

targets, the forecast performance metrics obtained

with the XGBoost model are presented in Table 3.

When the table is analyzed, it is seen that for some

power plants, forecasts with very high accuracy are

obtained, while for some power plants, the

ICEEECS 2025 - International Conference on Advances in Electrical, Electronics, Energy, and Computer Sciences

166

performance of the model is poor. Forecast results

with R² values of 0.90 and above are shown in bold

in the table.

Table 3: WPP production forecast performance metrics

over a 1-hour horizon.

Plant Number MAE MSE R

WPP1 0.075 0.012 0.892 WPP27 0.062 0.008 0.931

WPP2 0.078 0.013 0.907 WPP28 0.064 0.008 0.913

WPP3 0.073 0.010 0.897 WPP29 0.074 0.011 0.904

WPP4 0.065 0.009 0.934 WPP30 0.060 0.007 0.924

WPP5 0.091 0.016 0.856 WPP31 0.064 0.008 0.911

WPP6 0.089 0.016 0.860 WPP32 0.057 0.006 0.937

WPP7 0.048 0.005 0.555 WPP33 0.065 0.009 0.936

WPP8 0.070 0.010 0.916 WPP34 0.058 0.007 0.920

WPP9 0.063 0.007 0.933 WPP35 0.059 0.007 0.938

WPP10 0.079 0.013 0.858 WPP36 0.067 0.009 0.892

WPP11 0.069 0.010 0.924 WPP37 0.100 0.020 0.867

WPP12 0.070 0.009 0.878 WPP38 0.081 0.013 0.889

WPP13 0.068 0.009 0.922 WPP39 0.065 0.009 0.928

WPP14 0.079 0.011 0.888 WPP40 0.059 0.007 0.897

WPP15 0.062 0.008 0.906 WPP41 0.068 0.009 0.912

WPP16 0.061 0.007 0.936 WPP42 0.064 0.008 0.901

WPP17 0.063 0.008 0.880 WPP43 0.067 0.010 0.915

WPP18 0.076 0.011 0.887 WPP44 0.062 0.008 0.929

WPP19 0.063 0.014 0.885 WPP45 0.067 0.009 0.904

WPP20 0.069 0.009 0.875 WPP46 0.061 0.007 0.930

WPP21 0.082 0.013 0.898 WPP47 0.071 0.009 0.918

WPP22 0.060 0.007 0.934 WPP48 0.065 0.008 0.928

WPP23 0.067 0.009 0.920 WPP49 0.058 0.007 0.926

WPP24 0.072 0.010 0.917 WPP50 0.079 0.012 0.894

WPP25 0.071 0.011 0.889 WPP51 0.068 0.009 0.917

WPP26 0.067 0.009 0.923 WPP52 0.079 0.013 0.902

According to the findings, the most accurate

generation forecasts for the 1-hour forecast horizon

were realized with the data from WPP35. On the other

hand, the lowest forecast performance was obtained

with the production data of plant WPP7.

4 CONCLUSIONS

As a result of the research, it was concluded that it is

not always possible to use meteorological

measurement data and all power plant production data

within the same provincial borders together in

forecasting studies. One of the main reasons for this

situation is that meteorological measurement stations

and wind power plants are located in different

locations. Moreover, even if the measurement station

and the power plant are located in the same province,

the fact that some power plants have endemic

meteorological conditions specific to their region is

another reason for the differences in forecasting

performance.

REFERENCES

Aladağ, E. (2023). Ampirik Mod Ayrıştırmasına Dayalı

ARIMA Modeli Kullanılarak Van İli Hava Kirliliğinin

Tahmini. Yüzüncü Yıl Üniversitesi Fen Bilimleri

Enstitüsü Dergisi, 28(2), 495–509.

https://doi.org/10.53433/yyufbed.1220578

Demirer, H., Ayci, D., Ozturk, S. A., Karabal, C., & Valles,

J. A. (2024). A Machine Learning-Based Reference

Baseline Consumption Model for Energy Efficiency

Measurement and Verification. Electrical-Electronics

and Biomedical Engineering Conference, ELECO 2024

- Proceedings, 1–5.

https://doi.org/10.1109/ELECO64362.2024.10847193

Guan, S., Wang, Y., Liu, L., Gao, J., Xu, Z., & Kan, S.

(2023). Ultra-short-term wind power prediction method

combining financial technology feature engineering

and XGBoost algorithm. Heliyon, 9(6), e16938.

https://doi.org/10.1016/j.heliyon.2023.e16938

Hakkal, S., & Lahcen, A. A. (2024). XGBoost To Enhance

Learner Performance Prediction. Computers and

Education: Artificial Intelligence, 7(January), 100254.

https://doi.org/10.1016/j.caeai.2024.100254

Huang, N. E., Shen, Z., Long, S. R., Wu, M. C., Snin, H.

H., Zheng, Q., Yen, N. C., Tung, C. C., & Liu, H. H.

(1998). The empirical mode decomposition and the

Hubert spectrum for nonlinear and non-stationary time

series analysis. Proceedings of the Royal Society A:

Mathematical, Physical and Engineering Sciences,

454(1971), 903–995.

https://doi.org/10.1098/rspa.1998.0193

Irena. (2025). Renewable capacity statistics 2025

International Renewable Energy Agency. In Irena.

https://irena.org/-

/media/Files/IRENA/Agency/Publication/2020/Mar/IR

ENA_RE_Capacity_Statistics_2020.pdf

Jiang, T., & Liu, Y. (2023). A short-term wind power

prediction approach based on ensemble empirical mode

decomposition and improved long short-term memory.

Computers and Electrical Engineering, 110(March),

108830.

https://doi.org/10.1016/j.compeleceng.2023.108830

Joseph, L. P., Deo, R. C., Prasad, R., Salcedo-Sanz, S., Raj,

N., & Soar, J. (2023). Near real-time wind speed

forecast model with bidirectional LSTM networks.

Renewable Energy, 204(May 2022), 39–58.

https://doi.org/10.1016/j.renene.2022.12.123

Joshi, A., Vishnu, C., Mohan, C. K., & Raman, B. (2024).

Application of XGBoost model for early prediction of

earthquake magnitude from waveform data. Journal of

Earth System Science, 133(1).

https://doi.org/10.1007/s12040-023-02210-1

Kanji, S., & Das, S. (2025). Assessing groundwater

potentialities and replenishment feasibility using

machine learning and MCDM models considering

hydro-geological aspects and water quality

constituents. Environmental Earth Sciences, 84(1), 1–

18. https://doi.org/10.1007/s12665-024-11996-2

Karadöl, İ., & Şekkeli, M. (2022). Ramp Events in River

Type Hydroelectric Power Plants: The Case of Turkey.

Fırat Üniversitesi Mühendislik Bilimleri Dergisi, 34(1),

361–375. https://doi.org/10.35234/fumbd.1029081

Karadöl, İ., Yıldız, C., & Şekkeli, M. (2021). Determining

optimal spatial and temporal complementarity between

wind and hydropower. Energy, 230.

https://doi.org/10.1016/j.energy.2021.120790

Short-Term Wind Energy Production Forecasting and Target Plant Selection Based on Meteorological Data Using Empirical Mode

Decomposition

167

Li, M., Fu, X., & Li, D. (2020). Diabetes Prediction Based

on XGBoost Algorithm. IOP Conference Series:

Materials Science and Engineering, 768(7).

https://doi.org/10.1088/1757-899X/768/7/072093

Li, W., Peng, X., Cheng, K., Wang, H., Xu, Q., Wang, B.,

& Che, J. (2020). A Short-Term Regional Wind Power

Prediction Method Based on XGBoost and Multi-stage

Features Selection. 2020 IEEE Student Conference on

Electric Machines and Systems, SCEMS 2020, 614–

618.

https://doi.org/10.1109/SCEMS48876.2020.9352249

Li, X., Leung, F. H. F., Su, S., & Ling, S. H. (2022). Sleep

Apnea Detection Using Multi-Error-Reduction

Classification System with Multiple Bio-Signals.

Sensors, 22(15), 1–17.

https://doi.org/10.3390/s22155560

Li, X., Wang, J., Geng, Z., Jin, Y., & Xu, J. (2023). Short-

term Wind Power Prediction Method Based on Genetic

Algorithm Optimized XGBoost Regression Model.

Journal of Physics: Conference Series, 2527(1).

https://doi.org/10.1088/1742-6596/2527/1/012061

Liu, G., Zhou, J., Jia, B., He, F., Yang, Y., & Sun, N.

(2019). Advance short-term wind energy quality

assessment based on instantaneous standard deviation

and variogram of wind speed by a hybrid method.

Applied Energy, 238(January), 643–667.

https://doi.org/10.1016/j.apenergy.2019.01.105

Ma, Z., Chang, H., Sun, Z., Liu, F., Li, W., Zhao, D., &

Chen, C. (2020). Very Short-Term Renewable Energy

Power Prediction Using XGBoost Optimized by TPE

Algorithm. 2020 4th International Conference on

HVDC, HVDC 2020, 1236–1241.

https://doi.org/10.1109/HVDC50696.2020.9292870

Mustaqeem, Ishaq, M., & Kwon, S. (2022). A CNN-

Assisted deep echo state network using multiple Time-

Scale dynamic learning reservoirs for generating Short-

Term solar energy forecasting. Sustainable Energy

Technologies and Assessments, 52(PC), 102275.

https://doi.org/10.1016/j.seta.2022.102275

Phan, Q. T., Wu, Y. K., & Phan, Q. D. (2021). A hybrid

wind power forecasting model with xgboost, data

preprocessing considering different nwps. Applied

Sciences (Switzerland), 11(3), 1–19.

https://doi.org/10.3390/app11031100

RajasundrapandiyanLeebanon, T., Murugan, N. S. S.,

Kumaresan, K., & Jeyabose, A. (2025). Long-term

solar radiation forecasting in India using EMD, EEMD,

and advanced machine learning algorithms. In

Environmental Monitoring and Assessment (Vol. 197,

Issue 3). https://doi.org/10.1007/s10661-025-13738-8

Rathore, H., Meena, H. K., & Jain, P. (2023). Prediction of

EV Energy consumption Using Random Forest And

XGBoost. Proceedings - 2nd International Conference

on Power Electronics and Energy, ICPEE 2023, 1–6.

https://doi.org/10.1109/ICPEE54198.2023.10060798

Renewable Energy Agency, I. (2025). Renewable Capacity

Statistics 2025. In Irena. www.irena.org

Shang, Y., Miao, L., Shan, Y., Gnyawali, K. R., Zhang, J.,

& Kattel, G. (2022). A Hybrid Ultra-Short-Term and

Short-Term Wind Speed Forecasting Method Based on

CEEMDAN and GA-BPNN. Weather and Forecasting,

37(4), 415–428. https://doi.org/10.1175/WAF-D-21-

0047.1

Shi, R., Xu, X., Li, J., & Li, Y. (2021). Prediction and

analysis of train arrival delay based on XGBoost and

Bayesian optimization. Applied Soft Computing, 109,

107538. https://doi.org/10.1016/j.asoc.2021.107538

Shukla, S., & Pasari, S. (2025). Short-term wind speed

prediction with adaptive signal processing based hybrid

statistical models. In Energy Systems (Issue

0123456789). Springer Berlin Heidelberg.

https://doi.org/10.1007/s12667-025-00727-6

Yan, X., Wang, L., Zhu, J., Wang, S., Zhang, Q., & Xin, Y.

(2022). Automatic Obstructive Sleep Apnea Detection

Based on Respiratory Parameters in Physiological

Signals. 2022 IEEE International Conference on

Mechatronics and Automation, ICMA 2022, 461–466.

https://doi.org/10.1109/ICMA54519.2022.9856347

Yelgeç, M. A., & Bingöl, O. (2022). Ayrık dalgacık

dönüşümü ve Xgboost ile rüzgâr gücü tahmini.

Uluslararası Teknolojik Bilimler Dergisi, 14(2), 58–65.

https://doi.org/10.55974/utbd.1132336

Yuzgec, U., Dokur, E., & Balci, M. (2024). A novel hybrid

model based on Empirical Mode Decomposition and

Echo State Network for wind power forecasting.

Energy, 300(April), 131546.

https://doi.org/10.1016/j.energy.2024.131546

Zhang, T., Zhang, X., Rubasinghe, O., Liu, Y., Chow, Y.

H., Iu, H. H. C., & Fernando, T. (2024). Long-Term

Energy and Peak Power Demand Forecasting Based on

Sequential-XGBoost. IEEE Transactions on Power

Systems, 39(2), 3088–3104.

https://doi.org/10.1109/TPWRS.2023.3289400

Zheng, H., & Wu, Y. (2019). A XGBoost Model with

Weather Similarity Analysis and Feature Engineering

for Short-Term Wind Power Forecasting. Applied

Sciences (Switzerland).

ICEEECS 2025 - International Conference on Advances in Electrical, Electronics, Energy, and Computer Sciences

168