Short-Time Wind Speed Forecast Using Artificial Learning-Based Algorithms

Table 1

Main characteristics of the existing wind speed forecasting schemes.

Aim	Technique	Merits/outcomes	Demerits	Dataset
Hybrid wind speed prediction [13]	Empirical wavelet transformation (EWT), long short-term memory neural network, and a deep learning algorithm.	The proposed model has the satisfactory multistep forecasting results.	The performance of the EWT for the wind speed multistep forecasting has not been studied	Four sets of original wind speed series including 700 samples.
Wind speed forecasting [14]	Unscented Kalman filter (UKF) is integrated with support vector regression (SVR) model	The proposed method has better performance in both one-step-ahead and multistep-ahead predictions than ANNs, SVR, autoregressive, and autoregressive integrated with Kalman filter models	Needs to develop the predictive model-based control and optimization strategies for wind farm operation.	Center for Energy Efficiency and Renewable Energy at University of Massachusetts
Wind speed forecasting [15]	Long short-term memory neural networks, support vector regression machine, and extremal optimization algorithm.	The proposed model can achieve a better forecasting performance than ARIMA, SVR, ANN, KNN, and GBRT models.	Needs to consider more interrelated features like weather conditions, human factors, and power system status.	A wind farm in Inner Mongolia, China
A hybrid short-term wind speed forecasting [16]	Wavelet transform (WT), genetic algorithm (GA), and support vector machines (SVMs)	The proposed method is more efficient than a persistent model and a SVM-GA model without WT	Needs to augment external information such as the air pressure, precipitation, and air humidity besides the temperature.	The wind speed data every 0.5 h in a wind farm of North China in September 2012
Short-term wind speed prediction [18].	Support vector regression (SVR) and artificial neural network (ANN) with backpropagation	The proposed SVR and ANN models are able to predict wind speed with more than 99% accuracy.	Computationally expensive	Historical dataset (2008–2014) of wind speed of Chittagong costal area from Bangladesh Meteorological Division (BMD)
Hybrid wind speed forecasting [19]	Variational mode decomposition (VMD), the partial autocorrelation function (PACF), and weighted regularized extreme learning machine (WRELM)	(i) The VMD reduces the influences of randomness and volatility of wind speed.; (ii) PACF reduces the feature dimension and complexity of the model.; (iii) ELM improves the prediction accuracy.	The forecasting accuracy of two-step-ahead and three-step-ahead predictions declined to different degrees.	USA National Renewable Energy Laboratory (NREL) in 2004.
Short-term wind speed forecasting [20].	Wavelet analysis and AdaBoosting neural network.	(i) Benefits the analysis of the wind speed’s randomness and optimal neural network’s structure.; (ii) It can be used to promote the model’s configuration and show the confidence in high-accuracy forecasting.	Needs to consider the dynamical model with ability of error correction and adaptive adjustment.	USA National Renewable Energy Laboratory (NREL) in 2004.
Short-term wind speed forecasting [21].	Support vector machine (SVM) with particle swarm optimization (PSO)	The proposed model has the best forecasting accuracy compared to classical SVM and backpropagation neural network models.	Needs to consider additional information for efficient forecasting such as season and weather variables.	Wind farm data in China in 2011.
Wind speed predictions [22].	Recurrent neural network (RNN) with long short-term memory (LSTM).	The model provides 92.7% accuracy for training data and 91.6% for new data.	High rate epochs increased the process time and eventually provided low accuracy performance.	Nganjuk Meteorology and Geophysics Agency (BMKG), East Java (2008–2017).
Forecasting multistep-ahead wind speed [23]	NARNET model to forecast hourly wind speed using an artificial neural network (ANN).	The model is cost effective and can work with minimum availability of statistical data	(i) Faulty measurements of inputs are likely to affect the model parameters.; (ii) Removing rapid changes using a low-pass filter might result in neglecting important information.	Meteorological data from the National Oceanic and Atmospheric Administration (NOAA) located in Dodge City, Kansas (January 2010 -December 2010).
Short-term wind speed prediction [24]	Backpropagation (BP) neural network based on improved artificial bee colony algorithm (ABC-BP).	The model has high precision and fast convergence rate compared with traditional and genetic BP neural networks.	Sensitive for noisy data. Therefore, data should be filtered, which may affect the nature of data.	Wind farm in Tianjin, China (December 2013–January 2014).
Short-term wind speed forecasting [25] 2019	Fuzzy C-means clustering (FCM) and improved mind evolutionary algorithm-BP (IMEA-BP).	The proposed model is suitable for one-step forecasting and enhances the accuracy of multistep forecasting.	The accuracy of multistep forecasting needs to be further improved.	Wind farm in China
Predicting wind speed [26].	Artificial neural network and decision tree algorithms	The platform has the ability of mass storage of meteorological data, and efficient query and analysis of weather forecasting.	Needs improvement in order to forecast more realistic weather parameters.	Meteorological data provided by the Dalian Meteorological Bureau (2011–2015)
Our scheme	Employing multi-lags-one-step (MLOS) ahead forecasting technique with artificial learning-based algorithms	The provided results suggest that the ConvLSTM model has the best performance as compared to ANN, CNN, LSTM, and SVM models.	Increasing the number of hidden layers may increase the computational time exponentially.	National Wind Institution, West Texas Mesonet (2012–2015)