https://orcid.org/0000-0001-6475-4870
Figures
Abstract
The ever-increasing demand for electricity has presented a grave threat to traditional energy sources, which are finite, rapidly depleting, and have a detrimental environmental impact. These shortcomings of conventional energy resources have caused the globe to switch from traditional to renewable energy sources. Wind power significantly contributes to carbon-free energy because it is widely accessible, inexpensive, and produces no harmful emissions. Better and more efficient renewable wind power production relies on accurate wind speed predictions. Accurate short-term wind speed forecasting is essential for effectively handling unsteady wind power generation and ensuring that wind turbines operate safely. The significant stochastic nature of the wind speed and its dynamic unpredictability makes it difficult to forecast. This paper develops a hybrid model, L-LG-S, for precise short-term wind speed forecasting to address problems in wind speed forecasting. In this research, state-of-the-art machine learning and deep learning algorithms employed in wind speed forecasting are compared with the proposed approach. The effectiveness of the proposed hybrid model is tested using real-world wind speed data from a wind turbine located in the city of Karachi, Pakistan. Moreover, the mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE) are used as accuracy evaluation indices. Experimental results show that the proposed model outperforms the state-of-the-art legacy models in terms of accuracy for short-term wind speed in training, validation and test predictions by 98% respectively.
Citation: Ashraf M, Raza B, Arshad M, Khan BM, Zaidi SSH (2024) Performance enhancement of short-term wind speed forecasting model using Realtime data. PLoS ONE 19(5): e0302664. https://doi.org/10.1371/journal.pone.0302664
Editor: Nasir Ayub, Air University, PAKISTAN
Received: December 19, 2023; Accepted: April 10, 2024; Published: May 31, 2024
Copyright: © 2024 Ashraf et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The data set contains wind speed against time and date. The data is from Jhimpir Power Karachi. The data is from a private company and all the R&D is done by the company to acquire the data set therefore University Ethics Committee impose restriction on the data set. However, the data set will be made available for research purposes on case to case basis after communicating with the head of the Ethics Committee at dir-rnd@pnec.nust.edu.pk. to gain access to data.
Funding: The author(s) received no specific funding for this work.
Competing interests: The authors have declared that no competing interests exist.
Ⅰ. Introduction
Energy demand is rising globally, especially regarding electrical energy, along with the escalating effects of problems like the decreasing supply of fossil fuels and their adverse impact on the environment, rising oil prices, and their combined products on civilizations [1, 2]. To meet global energy demand and reduce greenhouse gas emissions according to the Paris Agreement [3], the globe is undergoing a switch from fossil fuels to renewable energy sources (RESs). RESs are defined as clean forms of energy as they produce zero carbon emissions [4]. All energy that comes from natural movements and mechanisms of the environment is considered renewable energy. Renewable energy excludes fossil fuels, waste from fossil reserves, and trash from inorganic resources [5]. Fig 1 illustrates renewable energy supply types. RESs convert this natural source of energy into useful forms of energy like electricity. Fig 2 depicts the capacity of renewable energy sources to supply more than 3000 times the present global energy demand [6]. Among RESs, wind power is gaining attention considering its future potential to meet growing electricity demands at a reasonable price. In 2022, 50% of the world’s consumption of green energy came from wind, which grew at a rate of 13.5%, while 31.43% came from solar [7]. However, wind power production relies highly on wind speed, which is stochastic and fluctuating. For this reason, it is crucial to forecast wind speed accurately for the safe operation of wind turbines and the integration of wind power with the grid.
Machine learning is an emerging area and currently many applications are being targeted in order to solve practical problems ranging from Driver’s Diabetes Monitoring System, Vehicular Ad-Hoc Networks, Task offloading optimization and energy management system, and Offload micro services based applications [8–11]. However, in this paper we are implementing machine learning algorithms in wind energy domain which is an offshoot of machine learning to resolve more practical problem of wind speed forecast in wind power generation.
For predicting wind speed, several approaches have been put forth; each has its own advantages and limitations and can be used with different datasets. It is difficult for general models to faithfully represent the development patterns buried in large and heavily sampled datasets. Physical model-based approaches and statistical modelling techniques are the two main groups into which wind speed prediction procedures can be classified [12]. A set of physical models with numerical parameters that characterize local meteorological and geographic characteristics such as temperature, atmospheric pressure, surface roughness, and barriers are the foundation of numerical weather prediction (NWP) systems [13, 14]. Supercomputers are used for performing NWPs because of the massive amount of processing power required. Due to the difficulties in gaining information quickly and the related high expenses, NWPs are typically conducted once or twice daily. When the weather is stable, these models work well for long-term forecasting [15]. The statistical method leverages the difference between the estimated and actual wind speeds in the recent past to fine-tune model parameters. It is based on training with measurement data. It’s simple to model, cheap, accurate and can make predictions in real-time. It is based on patterns rather than any pre-established mathematical formula. If prints are matched with historical ones, errors are reduced. Neural network (NN) and Time-series-based models based approaches are two subcategories of this method. These methods are considered accurate for short-term and medium to long-term wind speed prediction (WSP) [16].
With the explosion of interest in data science over the past few years, statistical models have come into greater focus [17]. Machine learning (ML) methods [18], such as artificial neural networks (ANN) [19], support vector machines (SVM) [20], extreme learning machines (ELM) [21], and deep learning (DL) [22], are developing remarkably quickly. In this paper, we are exploiting the power of machine learning in short-term wind speed forecasting and propose an ensemble L-LG-S model for accurate predictions. This model uses an LSTM component to analyze historical wind speed data and identify recurrent patterns. The LightGBM component focuses on integrating features and identifies intricate connections between wind speed and other pertinent variables. By considering both linear and non-linear correlations in the data, the SVR component adjusts the predictions given by the LSTM and LightGBM models. All three components’ projections are combined to form the final forecast. The rest of the article is divided into the following sections. The literature review is presented in Section II. The related methodologies are given in Section III. The proposed model establishment is presented in Section IV. Results and analysis are presented in Section V. Section VI presents the conclusion and future work is given in Section VII.
II. Related literature
The statistical methods are widely used in the literature for short-term wind speed prediction (WSP). Time series models such as ARIMA [23–25], ARMA [26, 27], AR [28], ARIMAX [29, 30], ARX [31], etc, have shown remarkable accuracy in predicting wind speed. However, they require high-order modeling and more computational power to capture nonlinearity and the stochastic nature of wind speed. Initially, time series models are best known for linear systems. On the contrary, with the advancement in ANN, ML, and DL techniques, they are getting more attention in the forecast area. Many researchers have used them single-handily for WSP and developed more advanced hybrid models to forecast short-term wind speed accurately. A few types of research are presented in Table 1 which shows the effectiveness of Machine learning models in wind speed forecast.
III. Related methodologies
Convolutional neural network algorithm
Convolutional neural networks, often called CNNs or ConvNets, were developed in the late 1990s and are a feed-forward artificial neural network whose connectivity topology is modeled after the structure of the animal visual cortex. Due to the input layer, output layer, several hidden layers, and millions of parameters that CNNs contain, they are able to learn complex objects and patterns. It pools and convolves the input to create smaller samples before applying the activation function. The output layer is the last completely linked layer after all the hidden layers, which were originally only partially connected [45, 46]. The general equation for CNN [47] is presented in Eq no 1:
(1)
where,
- x represents the input image or feature map,
- W is the set of learnable convolutional filters,
- b is the bias term,
- σ is the activation function, commonly ReLU (Rectified Linear Unit),
- y represents the output feature map
Recurrent neural network algorithm
RNN stands for a recurrent neural network. It is a neural network that processes time series, audio, and spoken language sequential data. RNNs, which analyze input data in a single pass, as opposed to feedforward neural networks, are designed to handle data sequences and can store information about earlier inputs. In sequential data, they can now spot patterns and temporal correlations [48]. A simple RNN model [49] can be described by the Eqs (2, 3):
(2)
(3)
where,
The hidden state ht at time step t is computed by applying the activation function σ to the sum of the weighted previous hidden state ht, the weighted current input xt, and the bias vector bh.
The output ht at time step t is computed by applying the softmax function to the sum of the weighted current hidden state ht, the weighted matrix Why, and the bias vector bh.
Lazyprophet algorithm
LazyProphet is a potent time series prediction approach [50]. It functions by connecting the start of the time series to a spot halfway through, and then connecting the midpoint spot to the end of the series. This procedure is repeated multiple times, and the position of the”kink” is simultaneously altered (intermediate node). The slope of each line from the midway to the final spot is further penalized by the addition of the”decay” component. Thus, this model is known as”LazyProphet” because it is less demanding.
LightGBM algorithm
In 2017, Microsoft developed the LightGBM gradient boosting framework. The decision-tree-based algorithm emphasizes high efficiency, scalability, and prediction accuracy [51]. To make predictions, LightGBM builds an ensemble of gradient-boosting decision trees, where each tree is trained to reduce the residual error of the prior tree. The training procedure employs a leaf-wise method, where each tree is formed by recursively splitting the data at the point that leads to a significant reduction in the loss function. LightGBM additionally uses regularization, early halting, and feature subsampling to minimize overfitting. LightGBM model [52] equation is described in Eq 4:
(4)
where, fm(x) is the output of the mth weak regression tree and F(x) is the ultimate result.
LSTM algorithm
Hochreiter and Schmidhuber first introduced the Long Short-Term Memory (LSTM) technique as a sort of recurrent neural network (RNN) architecture in 1997 [53]. Traditional RNNs have a problem with vanishing gradients that can make it challenging to learn long-term dependencies in time series data. LSTMs are made to solve this issue. The input gate, forget gate, output gate, memory cell, and other components make up the LSTM architecture. The gates control the flow of information into and out of the memory cell, which is in charge of storing and updating data over time. Moreover, the LSTM architecture has optional dropout layers to avoid overfitting. The loss function, commonly mean squared error, is minimized by training the LSTM model via backpropagation through time (BPTT) (MSE). An optimization technique like Adam or RMSprop is used to iteratively change the weights and biases of the LSTM layers. Once trained, the LSTM model can be used to predict outcomes in the future. The LSTM model receives historical time series values as input and forecasts a single value for the following time step. The Eqs (5–10) describe the LSTM components [54]:
(5)
(6)
(7)
(8)
(9)
(10)
Where Eq 4 is the forget cell equation, Eq 5 is the input gate equation,
Eq 6 is the candidate cell state equation, Eq 7 is the cell state equation,
Eq 8 is the output gate equation, Eq 9 is the hidden state equation, and ht−1 is the previous hidden state, xt is the input at time step t, σ is the sigmoid
activation function, tanh is the hyperbolic tangent activation function, Vf, Vi, VC, Vo
are weight matrices and af, ai, aC, ao are bias vectors.
Support vector regression algorithm
Two forms of support vector machines (SVMs) are Support Vector Regression (SVR) and Support Vector Classification (SVC), respectively utilized for regression analysis and classification [55]. Comparable to other regression models. SVR uses a margin of error (epsilon) to control the error level in the model, and a function called kernel is used to convert the input data into a higher dimensional space where the data is more separable. SVR aims to discover a hyperplane in the higher dimensional space to acquire the maximum margin of error while still obtaining the maximum amount of data. The prediction of new data points is made using a regression hyperplane [56]. The mathematical expressions for SVR [57] are described in Eqs (11–16):
1. The primal optimization problem:
minimize:
The objective function is composed of two parts:
(11)
The first component is the regularization term , which minimises the weight vector w’s squared norm in order to penalize the model’s complexity. The penalty term, which is the product of the slack variables ξi and the product of ξi* and a regularization parameter C. This variable regulates the trade-off between mistake minimization and margin maximization.
These are the restrictions that govern the primary optimization problem. By using slack variables ξi to penalize violations, they guarantee that the model makes predictions within a margin of error ∈ of the actual target values additionally ξi*.
2. The dual optimization problem:
With the goal of maximizing the margin between support vectors, the dual problem is constructed. In this case, ai and are the Lagrange multipliers linked to the primal constraints, and 〈∅(xj),∅(xj)〉 denotes the inner product in the feature space caused by the kernel function.
IV. Wind speed forecast model establishment
Gharo wind turbine and data collection
Gharo is a small town in Pakistan’s Thatta district, Sindh province. The Gharo wind mast is located on the town’s outskirts. The meteorological mast has an NRG data collecting system installed. The atmospheric sensors’ specifications are displayed in Table 2. The site’s topographical coordinates are 24°35’48”N and 67°26’39”E. The wind speed data is collected at a time stamp of ten minutes for fifteen days.
Data preprocessing
The wind speed information comes from a Gharo wind turbine in Karachi, Pakistan’s city. The dataset spans September 15, 2016, to September 30, 2016, with a total of 2300 data samples, and the statistical summary of the dataset are shown in Table 3. As can be seen from the table, the experimental data fluctuates, ranging from 14.180 to 0.000, and the mean of the data set is 8.466088. Each data reading is taken at a timestamp of ten minutes. The plot of wind speed is shown in Fig 3. The data set is split into training, validation, and testing data for forecasting wind speed. The proposed model is trained on 1600 data samples, and verification and testing are done using 350 data samples. The platform used for wind speed prediction is Jupyter Notebook and programming language is Python 3.8.0 on HP Notebook laptop.
Construction of the proposed model
Fig 4 describes the detail overview of an experimental setup of the proposed system. At first in the data is extracted from wind turbines, which is then preprocessed before being fed into our L-LG-S model. L-LG-S model is develop by combining LSTM, LightGBM and SVR using an ensemble method. The data is first preprocessed, and the LSTM, LightGBM, and SVR models are given a training dataset with 1600 samples. Sequential patterns are captured by the LSTM model; minute, hour, and day characteristics are captured by LightGBM; and both linear and nonlinear patterns are captured by SVR, which also creates hyperplanes to predict.
These models go under a hyperparameter tuning process using the Grid search method to achieve the optimal parameter value, as shown in Table 5. The ensemble technique is then utilized to fuse these models and validate the proposed model training process; the validation dataset of 350 samples is used to predict. Finally, the trained model is saved and deployed to test prediction after validation. As shown in Fig 4, the suggested model’s complexity and interpret ability are more tolerable, outperforming the benchmark models’ accuracy. The evaluation metrics of training, validation, and test prediction of the proposed model are given in Table 7.
Evaluation indicator
The root mean square error (RMSE), mean squared error (MSE), and mean absolute error (MAE) are the most popular evaluation metrics in regression analysis [58, 59]. They are frequently employed in regression analyses to assess the prediction model’s performance. The difference between what was expected and what happened is less when the RMSE, MSE, and MAE are smaller. Table 4 contains a list of the RMSE, MSE, and MAE mathematical expressions. In these equations, n is the number of samples, xi is the true value of the ith sample, and is the predicted value of the ith sample.
Hyperparameter tuning
To improve the accuracy of predictions made by machine learning models, hyperparameter tuning is performed. The ideal value takes advantage of the benchmark models to operate more effectively inside the search domain and aids in avoiding underperformance due to incorrect parameters. For hyperparameter optimization in this paper, the grid search approach is used. This technique is frequently employed while optimizing hyperparameters. It offers a tailored search range and is simple to apply for. In this study, the hybrid models use default parameters while the individual test models are modified. Tables 5 and 6 show the search range and the best value for each individual benchmark model, respectively.
V. Result and discussion
Experimental results
An accurate wind speed forecast is essential for efficient wind farm operation and dependable wind power generation. However, due to the stochastic and erratic character of wind speed, time series WSP is a challenging problem in forecasting. The short-term WSP is essential for managing wind turbines and extending their lifespan and power generation efficiency. This research employs the proposed model L-LG-S for short-term wind speed prediction, whose data source is mentioned in Section IV. In this paper, the proposed model is compared with popularly known state of the art prediction models available in literature, CNN, Lazyprophet, LSTM, RNN, and SARIMAX-RNN-SVR to determine its effectiveness. Figs 5–10 depicts the forecasting results of models. Among the benchmark models, LSTM network show promising results in wind speed prediction due to their ability to capture long-range dependencies. LazyProphet is a promising time series predictive model in the past few years, RNN, and CNN are representative deep neural network models with sound prediction effects for time series. SARIMAX-RNN-SVR is a hybrid model which has shown in commendable results in author’s previous work. These comparison models are more efficient and can verify how well the suggested model predicts the wind speed.
Experimental analysis
The prediction performance of proposed and benchmark models are assessed using evaluation metrics, including RSME, MSE, and MAE. In Table 7, the evaluation of unique models shows that LSTM performs the best in training, validation, and test predictions, surpassing the other individual models. CNN achieves the second-best performance, though it outperforms LSTM in training, but in validation and test prediction, it stands after LSTM. The remaining models can be ranked in terms of their performance based on validation and test predictions as follows: SARIMAX-RNN-SVR, RNN, Lazyprophet. Although these models perform well in predicting wind speed. However, to capture wind speed’s stochastic nature and be more accurate in forecasting, this paper proposes a new hybrid model, L-LG-S, to enhance the accuracy and reliability of wind speed forecasts. This proposed model harnesses the strengths of LSTM, LightGBM, and SVR algorithms where the bold text indicates the predictions made by the model we have proposed. It is evident from Table 7 that the proposed model surpasses other state of the art models with train, validation, and test MSE, RSME, and MAE, 0.13, 0.36 and 0.28, 0.11, 0.34 and 0.27, 0.30, 0.55, and 0.43 respectively. The slightest errors in the predictions of the proposed model establish it as a reliable and accurate short-term wind speed prediction model.
VI. Conclusion
Climate change and rising demand for electricity have led to the rapid development of clean energies like wind power in recent years. For certain power systems, wind power generation in particular becomes a major energy source. However, wind power generation forecasting and scheduling can be exceedingly challenging due to the unpredictable and irregular behavior of wind speed. This research presents the development and effectiveness of a novel hybrid model, L-LG-S, for wind speed prediction at a small wind farm located in Pakistan. In order to represent the stochastic nature of wind speed in the short-term horizon, the suggested model combines the strengths of LightGBM, SVR, and LSTM. An extensive evaluation of the proposed model’s performance is carried out utilizing statistical error indicators, namely MAE, RMSE, and MSE. The L-LG-S findings show that the train, validation, and test prediction errors for MSE, MAE, and RMSE are closer to zero at 0.13, 0.36, 0.28, 0.11, 0.34, and 0.27, 0.30, 0.55, and 0.43, respectively. To validate the suggested framework, comparisons are made using real wind speed plot, LSTM, CNN, SARIMAX-RNN-SVR, Lazyprophet, and RNN models. The simulation results demonstrated that the suggested model is suitable for short-term wind speed forecasting and outperforms legacy models.
VII. Future work
This research work is focused on developing an accurate forecasting model to predict wind speed in short-term horizon and it has limitations for long-term predictions. The reduction in computational complexity and time will be the future work for this research work to become commercial system.
References
- 1. Al-Hamadani S. Solar energy as a potential contributor to help bridge the gap between electricity supply and growing demand in Iraq: A review. Int J Adv Appl Sci ISSN. 2020;2252(8814):8814.
- 2. Dotson DL, Eddy J, Swan P. Climate action and growing electricity demand: Meeting both challenges in the 21st century with space-based solar power delivered by space elevator. Acta Astronautica. 2022;198:761–766. https://doi.org/10.1016/j.actaastro.2022.05.029.
- 3. Lima MA, Mendes LFR, Moth´e GA, Linhares FG, de Castro MPP da Silva MG, et al. Renewable energy in reducing greenhouse gas emissions: Reaching the goals of the Paris agreement in Brazil. Environmental Development. 2020;33:100504. https://doi.org/10.1016/j.envdev.2020.100504.
- 4. Ashraf M, Gulraiz A, Zaidi SSH, Ashraf F, Khan BM. Wind’s Data Analysis for its Accurate Prediction in Smart Grid Systems. In: 2022 Third International Conference on Latest trends in Electrical Engineering and Computing Technologies (INTELLECT); 2022. p. 1–5.
- 5. Ellabban O, Abu-Rub H, Blaabjerg F. Renewable energy resources: Current status, future prospects and their enabling technology. Renewable and Sustainable Energy Reviews. 2014;39:748–764. https://doi.org/10.1016/j.rser.2014.07.113.
- 6.
Re-thinking 2050: Making EU 100 Available from: http://pr.euractiv.com/pr/re-thinking-2050-making-eu-100-renewables-based-89763.
- 7. Bp B. Statistical review of world energy 2022; 2023.
- 8. Sohail R., Saeed Y., Ali A., Alkanhel R., Jamil H., Muthanna A., et al. (2023). A Machine Learning-Based Intelligent Vehicular System (IVS) for Driver’s Diabetes Monitoring in Vehicular Ad-Hoc Networks (VANETs). Applied Sciences, 13(5), 3326.
- 9. Rashid K., Saeed Y., Ali A., Jamil F., Alkanhel R., & Muthanna A. (2023). An Adaptive Realtime Malicious Node Detection Framework Using Machine Learning in Vehicular Ad-Hoc Networks (VANETs). Sensors, 23(5), 2594. pmid:36904798
- 10. Ali A., Iqbal M. M., Jamil H., Qayyum F., Jabbar S., Cheikhrouhou O., et al. (2021). An efficient dynamic-decision-based task scheduler for task offloading optimization and energy management in mobile cloud computing. Sensors, 21(13), 4527. pmid:34282786
- 11. Ali A., & Iqbal M. M. (2022). A cost and energy-efficient task scheduling technique to offload microservices-based applications in mobile cloud computing. IEEE Access, 10, 46633–46651.
- 12. Han Y, Mi L, Shen L, Cai CS, Liu Y, Li K, et al. A short-term wind speed prediction method utilizing novel hybrid deep learning algorithms to correct numerical weather forecasting. Applied Energy. 2022;312:118777. https://doi.org/10.1016/j.apenergy.2022.118777.
- 13. Costa A, Crespo A, Navarro J, Lizcano G, Madsen H, Feitosa E. A review on the young history of the wind power short-term prediction. Renewable and Sustainable Energy Reviews. 2008;12(6):1725–1744. https://doi.org/10.1016/j.rser.2007.01.015.
- 14. Lynch P. The origins of computer weather prediction and climate modeling. Journal of Computational Physics. 2008;227(7):3431–3444. https://doi.org/10.1016/j.jcp.2007.02.034.
- 15. Dupr´e A, Drobinski P, Alonzo B, Badosa J, Briard C, Plougonven R. Sub-hourly forecasting of wind speed and wind energy. Renewable Energy. 2020;145:2373–2379. https://doi.org/10.1016/j.renene.2019.07.161.
- 16. Soman SS, Zareipour H, Malik O, Mandal P. A review of wind power and wind speed forecasting methods with different time horizons. In: North American Power Symposium 2010; 2010. p. 1–8.
- 17. Ma Z, Chen H, Wang J, Yang X, Yan R, Jia J, et al. Application of hybrid model based on double decomposition, error correction and deep learning in short-term wind speed prediction. Energy Conversion and Management. 2020;205:112345. https://doi.org/10.1016/j.enconman.2019.112345.
- 18. Gupta D, Natarajan N, Berlin M. Short-term wind speed prediction using hybrid machine learning techniques. Environmental Science and Pollution Research.2021; p. 1–19. pmid:34251573
- 19. Zhang Y, Pan G, Chen B, Han J, Zhao Y, Zhang C. Short-term wind speed prediction model based on GA-ANN improved by VMD. Renewable Energy. 2020;156:1373–1388.
- 20. Chen N, Sun H, Zhang Q, Li S. A Short-Term Wind Speed Forecasting Model Based on EMD/CEEMD and ARIMA-SVM Algorithms. Applied Sciences.2022;12(12):6085.
- 21. Tian Z, Li S, Wang Y. A prediction approach using ensemble empirical mode decomposition-permutation entropy and regularized extreme learning machine for short-term wind speed. Wind Energy. 2020;23(2):177–206.
- 22. Neshat M, Nezhad MM, Abbasnejad E, Mirjalili S, Tjernberg LB, Garcia DA, et al. A deep learning-based evolutionary model for short-term wind speed forecasting: A case study of the Lillgrund offshore wind farm. Energy conversion and management. 2021;236:114002.
- 23. Grigonytė E, Butkevičiūtė E. Short-term wind speed forecasting using ARIMA model. Energetika. 2016;62(1–2).
- 24. Radziukynas V, Klementavicius A. Short-term wind speed forecasting with ARIMA model. In: 2014 55th International Scientific Conference on Power and Electrical Engineering of Riga Technical University (RTUCON). IEEE; 2014. p.145–149.
- 25. Singh S, Mohapatra A, et al. Repeated wavelet transform based ARIMA model for very short-term wind speed forecasting. Renewable energy. 2019;136:758–768.
- 26. Lydia M, Kumar SS, Selvakumar AI, Kumar GEP. Linear and non-linear autoregressive models for short-term wind speed forecasting. Energy conversion and management. 2016;112:115–124.
- 27. Ak¸cay H, Filik T. Short-term wind speed forecasting by spectral analysis from long-term observations with missing values. Applied energy. 2017;191:653–662.
- 28. Ailliot P, Monbet V. Markov-switching autoregressive models for wind time series. Environmental Modelling & Software. 2012;30:92–101.
- 29. Zhao E, Zhao J, Liu L, Su Z, An N. Hybrid wind speed prediction based on a self-adaptive ARIMAX model with an exogenous WRF simulation. Energies.2015;9(1):7.
- 30. Xu Q, Li W, Kong D, Zhao X, Wang X, Li Y, et al. Ultra-short-term wind speed forecast based on WD-ARIMAX-GARCH model. In: 2019 IEEE 2nd International Conference on Automation, Electronics and Electrical Engineering (AUTEEE). IEEE; 2019. p. 219–222.
- 31. Duran MJ, Cros D, Riquelme J. Short-term wind power forecast based on ARX models. Journal of Energy Engineering. 2007;133(3):172–180.
- 32. Chen K, Yu J. Short-term wind speed prediction using an unscented Kalman filter based state-space support vector regression approach. Applied Energy.2014;113:690–705. https://doi.org/10.1016/j.apenergy.2013.08.025.
- 33. Li M, Zhang Z, Ji T, Wu QH. Ultra-short term wind speed prediction using mathematical morphology decomposition and long short-term memory. CSEE Journal of Power and Energy Systems. 2020;6(4):890–900.
- 34. Elsaraiti M, Merabet A. Application of Long-Short-Term-Memory Recurrent Neural Networks to Forecast Wind Speed. Applied Sciences. 2021;11(5).
- 35. Chen Y, Dong Z, Wang Y, Su J, Han Z, Zhou D, et al. Short-term wind speed predicting framework based on EEMD-GA-LSTM method under large scaled wind history. Energy Conversion and Management. 2021;227:113559. https://doi.org/10.1016/j.enconman.2020.113559.
- 36. Duan J, Zuo H, Bai Y, Duan J, Chang M, Chen B. Short-term wind speed forecasting using recurrent neural networks with error correction. Energy.2021;217:119397. https://doi.org/10.1016/j.energy.2020.119397.
- 37. Dhakal R, Sedai A, Pol S, Parameswaran S, Nejat A, Moussa H. A Novel Hybrid Method for Short-Term Wind Speed Prediction Based on Wind Probability Distribution Function and Machine Learning Models. Applied Sciences.2022;12(18).
- 38. Wang Y, Gui R. A Hybrid Model for GRU Ultra-Short-Term Wind Speed Prediction Based on Tsfresh and Sparse PCA. Energies. 2022;15(20).
- 39. Ashraf M, Raza B, Arshad M, Ahmed A, Zaidi SSH. A Hybrid Statistical Model for Ultra Short Term Wind Speed Prediction. In: 2023 7th International Multi-Topic ICT Conference (IMTIC); 2023. p. 1–8.
- 40. Duan J, Chang M, Chen X, Wang W, Zuo H, Bai Y, et al. A combined short-term wind speed forecasting model based on CNN–RNN and linear regression optimization considering error. Renewable Energy. 2022;200:788–808. https://doi.org/10.1016/j.renene.2022.09.114.
- 41. Wu H, Meng K, Fan D, Zhang Z, Liu Q. Multistep short-term wind speed forecasting using transformer. Energy. 2022;261:125231. https://doi.org/10.1016/j.energy.2022.125231.
- 42. ho Hur S. Short-term wind speed prediction using Extended Kalman filter and machine learning. Energy Reports. 2021;7:1046–1054. https://doi.org/10.1016/j.egyr.2020.12.020.
- 43. Liu M, Cao Z, Zhang J, Wang L, Huang C, Luo X. Short-term wind speed forecasting based on the Jaya-SVM model. International Journal of Electrical Power Energy Systems. 2020;121:106056. https://doi.org/10.1016/j.ijepes.2020.106056.
- 44. Chen Y, Wang Y, Dong Z, Su J, Han Z, Zhou D, et al. 2-D regional short-term wind speed forecast based on CNN-LSTM deep learning model. Energy Conversion and Management. 2021;244:114451. https://doi.org/10.1016/j.enconman.2021.114451.
- 45. Gu J, Wang Z, Kuen J, Ma L, Shahroudy A, Shuai B, et al. Recent advances in convolutional neural networks. Pattern Recognition. 2018;77:354–377. https://doi.org/10.1016/j.patcog.2017.10.013.
- 46. Li Z, Liu F, Yang W, Peng S, Zhou J. A Survey of Convolutional Neural Networks: Analysis, Applications, and Prospects. IEEE Transactions on Neural Networks and Learning Systems. 2022;33(12):6999–7019. pmid:34111009
- 47. LeCun Y, Bengio Y, Hinton G. Deep learning. nature. 2015;521(7553):436–444.
- 48. Liu MD, Ding L, Bai YL. Application of hybrid model based on empirical mode decomposition, novel recurrent neural networks and the ARIMA to wind speed prediction. Energy Conversion and Management. 2021;233:113917.
- 49. Hochreiter S, Schmidhuber J. Long Short-Term Memory. Neural Computation.1997;9(8):1735–1780. pmid:9377276
- 50. Löning M, Király F. Forecasting with sktime: Designing sktime’s new forecasting api and applying it to replicate and extend the m4 study. arXiv preprint arXiv:200508067. 2020;.
- 51. Ke G, Meng Q, Finley T, Wang T, Chen W, Ma W, et al. Lightgbm: A highly efficient gradient boosting decision tree. Advances in neural information processing systems. 2017;30.
- 52. Ren J, Yu Z, Gao G, Yu G, Yu J. A CNN-LSTM-LightGBM based short-term wind power prediction method based on attention mechanism. Energy Reports.2022;8:437–443. https://doi.org/10.1016/j.egyr.2022.02.206
- 53. Hochreiter S, Schmidhuber J. Long Short-term Memory. Neural computation.1997;9:1735–80. pmid:9377276
- 54. Jaseena KU, Kovoor BC. Decomposition-based hybrid wind speed forecasting model using deep bidirectional LSTM networks. Energy Conversion and Management. 2021;234:113944.doi:https://doi.org/https://doi.org/10.1016/j.enconman.2021.113944
- 55. Vapnik V. The support vector method of function estimation. Nonlinear modeling: Advanced black-box techniques. 1998; p. 55–85.
- 56. Awad M, Khanna R, Awad M, Khanna R. Support vector regression. Efficient learning machines: Theories, concepts, and applications for engineers and system designers. 2015; p. 67–80.
- 57. Suykens JA, Vandewalle J. Least squares support vector machine classifiers. Neural processing letters. 1999;9:293–300.
- 58. Zhou L, Zhao P, Wu D, Cheng C, Huang H. Time series model for forecasting the number of new admission inpatients. BMC medical informatics and decision making. 2018;18:1–11.
- 59. Liang S, Nguyen L, Jin F. A Multi-variable Stacked Long-Short Term Memory Network for Wind Speed Forecasting. In: 2018 IEEE International Conference on Big Data (Big Data); 2018. p. 4561–4564.