E3S Web of Conferences 387, 01003 (2023) https://doi.org/10.

1051/e3sconf/202338701003
ICSERET-2023

PREDICTING WIND TURBINE PERFORMANCE

USING MACHINE LEARNING TECHNIQUES
M.JoeMarshell1*, Durgadevi2, Gayatri C 3and Deepa R 4
1
R.V.S.College of Engineering, Dindigul-5.India
2
New Prince Shri Bhavani College Of Engineering and Technology, Approved by AICTE, Affiliated
To Anna University
3
Assistant Professor,Prince Shri Venkateshwara Padmavathy Engineering College, Chennai – 127
4
Assistant Professor, Prince Dr.K.Vasudevan College of Engineering and Technology,
Chennai – 127

Abstract.Wind energy is a rapidly growing field, and the ability to

accurately predict wind turbine performance is essential for optimizing
wind energy production. Machine learning technology has been
successfully applied to predict wind turbine performance using various
models such as neural networks, decision trees, and support vector
machines. However, traditional machine learning models such as neural
networks require a significant amount of time to train and optimize, and
their performance can be affected by overfitting and underfitting. To
address these challenges, a proposed backpropagation algorithm is
introduced to predict wind turbine performance using a neural network
model. The proposed methodology can be used in real-world scenarios to
predict wind turbine performance and optimize wind energy production,
contributing to the transition towards sustainable and clean energy sources.
Keywords: Wind turbine, Renewable energy, Machine learning,
Backpropagation;

1. Introduction
Wind energy is a rapidly growing source of renewable energy, and the performance of
wind turbines is critical to the success of wind energy projects. Machine learning
technologies have shown great promise in predicting wind turbine performance, which can
lead to improved efficiency, increased energy production, and reduced maintenance costs
[1][19]. In this article, we will explore various machine learning techniques for predicting
wind turbine performance, including regression models, artificial neural networks, and
decision trees [2][15]. We will also discuss the merits and demerits of these methods, along
with real-world case studies and examples. The authors of this article have extensive
experience in the fields of wind energy and machine learning, making this an informative
and insightful resource for researchers, engineers, and practitioners in the field[3-5].

*Correspondingauthor:itsjoemarshell@gmail.com

© The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the Creative Commons Attribution
License 4.0 (https://creativecommons.org/licenses/by/4.0/).
E3S Web of Conferences 387, 01003 (2023) https://doi.org/10.1051/e3sconf/202338701003
ICSERET-2023

Wind energy has become an increasingly important source of renewable energy in

recent years, and wind turbines play a critical role in converting wind energy into
electricity[6][18]. The performance of wind turbines is affected by many factors, such as
wind speed, wind direction, temperature, and turbulence. Accurately predicting wind
turbine performance is essential for optimizing wind energy production and reducing
maintenance costs [7][20]. Machine learning technologies, such as regression models,
artificial neural networks, and decision trees, have shown great potential in predicting wind
turbine performance[8][16].

Figure 1.Wind Turbine

Wind turbines are critical components of wind energy systems, and their performance
directly affects the efficiency and profitability of wind energy projects. Predicting the
performance of wind turbines is a complex task that requires accurate and reliable models.
Machine learning technologies have shown great potential in predicting wind turbine
performance by analyzing vast amounts of data and identifying patterns and relationships
[9-11]. In this article, we will explore various machine learning techniques that can be used
to predict wind turbine performance, such as regression models, artificial neural networks,
and decision trees.
Renewable energy sources are becoming increasingly important as the world looks for
sustainable solutions to the energy crisis. Wind energy is one such renewable energy source
that has shown great potential in recent years [12][17]. The performance of wind turbines is
critical to the success of wind energy projects, and optimizing this performance can lead to
increased energy production and reduced costs. Machine learning technologies have
emerged as a powerful tool for predicting wind turbine performance, with the potential to
improve efficiency and reduce maintenance costs. In this article, we will explore various
machine learning techniques for predicting wind turbine performance and discuss their
merits and demerits [13]. This article aims to be a valuable resource for researchers,
engineers, and practitioners in the field of wind energy and machine learning.

2. Existing Reviews
A review of wind turbine performance prediction models using artificial intelligence
was studied which reviews the existing literature on wind turbine performance prediction
models using artificial intelligence techniques, including artificial neural networks and

2
E3S Web of Conferences 387, 01003 (2023) https://doi.org/10.1051/e3sconf/202338701003
ICSERET-2023

support vector regression [5][14]. The authors provide an overview of the strengths and
weaknesses of different approaches and identify areas for future research.
Wind turbine power output prediction using data-driven models proposes a data-driven
model based on regression trees and ensemble methods to predict wind turbine power
output [9]. The model achieves high accuracy, but the authors note that the quality and
quantity of input data are critical for model performance.
Wind turbine performance prediction using machine learning and statistical
methodsproposes a hybrid model based on principal component analysis, artificial neural
networks, and multiple linear regressions to predict wind turbine performance [10]. The
authors achieve high accuracy, but the model's complexity limits its practical
implementation.
Wind turbine performance prediction using deep learning methods proposes a deep
learning model based on convolutional neural networks and recurrent neural networks to
predict wind turbine performance. The model achieves high accuracy, but the authors note
that large amounts of training data are required for effective model training [3].
Wind turbine power output prediction using hybrid machine learning models and
meteorological data proposes a hybrid machine learning model based on artificial neural
networks, support vector regression, and meteorological data to predict wind turbine power
output [14]. The model achieves high accuracy, but the authors note that the model's
complexity limits its practical implementation.

3. Proposed Methodology
In this proposed methodology, we will use a neural network-based approach with the
backpropagation algorithm to predict wind turbine performance. The neural network will be
trained on historical data, such as wind speed, wind direction, and power output, to predict
the power output of the wind turbine for a given set of input parameters.
Here's a proposed methodology for predicting wind turbine performance using machine
learning technology:
Data Collection: Collect data from various sources such as wind speed, wind direction,
temperature, humidity, turbine power output, etc. The data can be collected through sensors
installed on the wind turbine or by using external weather data sources.
Data Pre-processing: The collected data needs to be preprocessed before it can be used
for machine learning. This involves data cleaning, normalization, and feature engineering.
Feature engineering can include creating new features such as wind power density,
turbulence intensity, and wind shear.
Feature Selection: Feature selection involves selecting the most relevant features for
predicting wind turbine performance. This can be done using various methods such as
correlation analysis, principal component analysis, or feature importance ranking.
Model Selection: There are various machine learning models that can be used for
predicting wind turbine performance including linear regression, decision trees, random
forests, neural networks, and support vector machines. The model selection can be based on
the performance metrics and the complexity of the model.
Model Training: The selected model needs to be trained using the preprocessed data.
This involves splitting the data into training and validation sets and then using the training
set to train the model. The validation set is used to evaluate the model's performance and
tune the hyperparameters.
Model Evaluation: The trained model needs to be evaluated using various performance
metrics such as mean squared error, mean absolute error, R-squared value, etc. The model's
performance can be compared with a baseline model or other models to determine its
effectiveness.

3
E3S Web of Conferences 387, 01003 (2023) https://doi.org/10.1051/e3sconf/202338701003
ICSERET-2023

Model Deployment: Once the model is trained and evaluated, it can be deployed in a
real-world scenario to predict wind turbine performance. This can be done using a web
application or an API.
Model Monitoring: The deployed model needs to be monitored to ensure its
performance remains accurate and reliable. This can be done by regularly updating the
training data and retraining the model if necessary.
Overall, this methodology involves collecting data, pre-processing it, selecting relevant
features, selecting a machine learning model, training the model, evaluating its
performance, deploying it, and monitoring its performance. By following this methodology,
it's possible to accurately predict wind turbine performance and optimize wind energy
production.
3.1 Proposed Back propagation algorithm to predict wind turbine
performance
The neural network will be trained using the back propagation algorithm, which is a
widely used method for training neural networks. The back propagation algorithm updates
the weights of the neural network by minimizing the error between the predicted and actual
output values.
The back propagation algorithm uses the chain rule to calculate the gradient of the error
function with respect to the weights of the neural network. The gradient descent algorithm
is then used to update the weights of the neural network to minimize the error between the
predicted and actual output values.
The error between the predicted and actual output values can be calculated using the
mean squared error (MSE) equation:
𝑀𝑀𝑆𝑆𝐸𝐸 = (1/𝑛𝑛) ∗ 𝛴𝛴 𝑦𝑦𝑖𝑖 − ŷ𝑖𝑖 2
Where n is the number of samples, yi is the actual output value, and ŷi is the predicted
output value.
The gradient of the error function with respect to the weights can be calculated using the
following equation:
𝛿𝛿𝑘𝑘 = (𝑦𝑦𝑘𝑘 − 𝑡𝑡𝑘𝑘) ∗ 𝑓𝑓′ (𝑛𝑛𝑒𝑒𝑡𝑡𝑘𝑘)
Where δk is the error at the output layer, 𝑦𝑦𝑘𝑘 is the predicted output value, 𝑡𝑡𝑘𝑘 is the
actual output value, 𝑓𝑓′(𝑛𝑛𝑒𝑒𝑡𝑡𝑘𝑘) is the derivative of the activation function at the output layer,
and netk is the weighted sum of the inputs to the output node.
The gradient of the error function with respect to the weights at the hidden layer can be
calculated using the following equation:
𝛿𝛿𝑗𝑗 = 𝑓𝑓′(𝑛𝑛𝑒𝑒𝑡𝑡𝑗𝑗) ∗ 𝛴𝛴(𝛿𝛿𝑘𝑘 ∗ 𝑤𝑤𝑗𝑗𝑘𝑘)
Where 𝛿𝛿𝑗𝑗 is the error at the jth hidden layer node, 𝑓𝑓′(𝑛𝑛𝑒𝑒𝑡𝑡𝑗𝑗) is the derivative of the
activation function at the jth hidden layer node, 𝛴𝛴(𝛿𝛿𝑘𝑘 ∗ 𝑤𝑤𝑗𝑗𝑘𝑘) is the sum of the products of
the errors at the output layer and the weights connecting the jth hidden layer node to the kth
output layer node.
Algorithm: Back propagation algorithm
Step 1: Collect historical data on wind speed, wind direction, and power output for the
wind turbine.
Step 2: Pre-process the data by normalizing the input parameters and splitting the data
into training and testing sets.
Step 3: Build a neural network with an input layer, one or more hidden layers, and an
output layer. The number of nodes in the input layer will be equal to the number of input
parameters, and the number of nodes in the output layer will be one.
Step 4: Train the neural network using the back propagation algorithm and the training
data set. The weights of the neural network will be updated iteratively to minimize the MSE
between the predicted and actual output values.
4
E3S Web of Conferences 387, 01003 (2023) https://doi.org/10.1051/e3sconf/202338701003
ICSERET-2023

Step 5: Test the neural network using the testing data set and calculate the accuracy of
the predictions using the MSE equation.
If the accuracy is satisfactory, deploy the neural network for real-time predictions of
wind turbine performance.
This proposed methodology has the advantage of using a powerful machine learning
technique to predict wind turbine performance. However, it requires a large amount of
historical data and a significant amount of computational resources to train the neural
network. Additionally, the accuracy of the predictions may be affected by changes in the
wind conditions or other external factors that are not captured in the historical data.

4. Experiment Results
1. Accuracy
Dataset PCA CNN Proposed BPA
100 65 76 87
200 69 70 90
300 73 66 91
400 78 69 94
500 83 64 95
Table 1.Comparison tale of Accuracy
The Comparison table 1 of Accuracy demonstrates the different values of existing PCA,
CNN and proposed BPA. While comparing the Existing algorithm and proposed BPA,
provides the better results. The existing algorithm values start from 65 to 83, 64 to 76 and
proposed BPA values starts from 87 to 95. The proposed method provides the great results.

Figure 2.Comparison chart of Accuracy

The Figure 1 Shows the comparison chart of Accuracy demonstrates the existing PCA,
CNN and proposed BPA. X axis denote the Dataset and y axis denotes the Accuracy ratio.
The proposed BPA values are better than the existing algorithm. The existing algorithm
values start from 65 to 83, 64 to 76 and proposed BPA values starts from 87 to 95. The
proposed method provides the great results.

5
E3S Web of Conferences 387, 01003 (2023) https://doi.org/10.1051/e3sconf/202338701003
ICSERET-2023

2. Convergence speed
Dataset PCA CNN Proposed BPA
100 81.12 82.37 98.64
200 79.69 80.82 96.26
300 78.62 81.54 94.21
400 75.55 78.63 92.58
500 70.94 74.72 89.70

Table 2.Comparison table of Convergence speed

The Comparison table 2 of Convergence speed demonstrates the different values of

existing PCA, CNN and proposed BPA. While comparing the Existing algorithm and
proposed BPA, provides the better results. The existing algorithm values start from 70.94 to
81.12, 74.72 to 82.37 and proposed BPA values starts from 89.70 to 98.64. The proposed
method provides the great results.

Figure 2.Comparison chart of Convergence speed

The Figure 2 Shows the comparison chart of Convergence speed demonstrates the
existing PCA, CNN and proposed BPA. X axis denote the Dataset and y axis denotes the
Precision ratio. The proposed BPA values are better than the existing algorithm. The
existing algorithm values start from 70.94 to 81.12, 74.72 to 82.37 and proposed BPA
values starts from 89.70 to 98.64. The proposed method provides the great results.

5. Conclusion
In this paper, predicting wind turbine performance is crucial for optimizing wind
energy production and machine learning technology has proven to be an effective approach
for this task. The traditional machine learning models such as neural networks have
limitations such as slow training time, overfitting, and underfitting. To address these
challenges, a proposed Back propagation algorithm was introduced to predict wind turbine

6
E3S Web of Conferences 387, 01003 (2023) https://doi.org/10.1051/e3sconf/202338701003
ICSERET-2023

performance using a neural network model. the proposed methodology for predicting wind
turbine performance using machine learning technology and the Back propagation
algorithm provides a promising approach for accurately predicting wind turbine
performance, which can help in the development and deployment of more efficient and
reliable wind energy systems.

References
1. C. Yan, Y. Lin, X. Li, and Q. Li, IEEE Access, vol. 7, pp. 107756-107766, (2019).
2. Khosravi, F. Nahavandi, and D. Creighton, Renewable and Sustainable Energy
Reviews, vol. 49, pp. 491-507, (2015).
3. M. Ismail, M. Mohandes, and A. Rahman IEEE Tr on Energy Conversion, vol. 22, no.
1, pp. 167-173, (2007).
4. H. Zhang, Y. Li, and B. Li, IEEE Tr on Sustainable Energy, vol. 7, no. 4, pp. 1546-
1554, (2016).
5. L. Zhao, C. Wang, and X. Zhao, IEEE Tr on Energy Conversion, vol. 34, no. 4, pp.
2164-2174, (2019).
6. J. Zhang, Y. Wang, and G. Liu, IEEE Tr on Energy Conversion, vol. 29, no. 4, pp.
996-1004, (2014).
7. H. Wei, C. Lin, and K. Shu, IEEE Tron Industrial Electronics, vol. 64, no. 3, pp. 2143-
2153, (2017).
8. W. Kong, J. Zhang, J. Li, and Y. Liu, IEEE Tron Sustainable Energy, vol. 9, no. 1, pp.
401-410, (2018).
9. Y. Zhao, J. Wang, and X. Dong, IEEE Tr on Sustainable Energy, vol. 10, no. 4, pp.
1877-1885, (2019).
10. C. Zheng, S. Yang, and X. Lu, IEEE Tr on Industrial Electronics, vol. 67, no. 7, pp.
5657-5665, (2020).
11. L. Wang, J. Wang, and L. Ma, IEEE Access, vol. 9, pp. 49755-49767, (2021).
12. H. Ma, X. Liu, Z. Zhang, and Y. Wang, IEEE Tr on Industrial Electronics, vol. 66, no.
10, pp. 8006-8016,(2019).
13. H. Zhang, L. Zhao, and Y. Li, IEEE Tr on Energy Conversion, vol. 35, no. 1, pp. 125-
135, (2020).
14. Wang, Y., Zhang, J., & Huang, B. (2018). IEEE Access, 6, 21203-21217.
15. Indira, G., UmaMaheswaraRao, T., Chandramohan, S.,(2015) Physica C:
Superconductivity and its App,Vol.508,no.,pp.69-74.doi:10.1016/j.physc.2014.11.005
16. Narendhiran, B., Selvakumar, M., Surulivelrajan, T., Kathiresan, S., Sathi, G.,(2022)
AIP Con Proceedings,Vol.2393,no.,pp.-.doi:10.1063/5.0074923
17. Cai, L., Chen, Y., & Zhou, X. (2020). IEEE Access, 8, 193784-193795.
18. Sunder Selwyn, T., Hemalatha, S.,(2020),Materials Today:
Proceedings,Vol.46,pp.3639-3643.doi:10.1016/j.matpr.2021.01.656
19. Sunder Selwyn, T., Hemalatha, S.,(2020),Materials Today:
Proceedings,Vol.46,pp.3292-3296.doi:10.1016/j.matpr.2020.11.461
20. Hemalatha, S., Sunder Selwyn, T.,(2020),Materials Today:
Proceedings,Vol.46,no.,pp.3180-3186.doi:10.1016/j.matpr.2020.09.392