Altmetrics
Downloads
262
Views
81
Comments
0
A peer-reviewed article of this preprint also exists.
This version is not peer-reviewed
Submitted:
25 April 2023
Posted:
25 April 2023
You are already at the latest version
A peer-reviewed article of this preprint also exists.
This version is not peer-reviewed
Submitted:
25 April 2023
Posted:
25 April 2023
You are already at the latest version
ANN | Artificial Neural Network |
BNN | Backpropagation Neural Network |
QR | Quantile Regression |
NWP | Numerical Weather Prediction |
WPF | Wind Power Forecasting |
WPPF | Wind Power Probabilistic Forecasting |
WSF | Wind Speed Forecasting |
WRF | Weather Research and Forecasting |
WPG | Wind Power Generation |
WFP | Wind Farm Parameterization |
RWPF | Regional Wind Power Forecasting |
QRNN | Quantile Regression Neural Network |
CSTWPP | Convolutional Spatial-temporal Wind Power Predictor |
STCN | Spatio-temporal Convolutional Network |
CEEMDAN | Complete Ensemble Empirical Mode Decomposition with Adaptive Noise |
IBA | Improved Backfill Algorithm |
GPR | Gaussian Process Regression |
MFF-SAM-GCN | Multi-Feature Fusion/Self-Attention Mechanism/Graph Convolutional Network |
WMTSM | Weighted Multivariate Time Series Motifs |
CLP | Conditional Linear Programming |
ABQs | Adaptive Boundary Quantiles |
WNN | Wavelet Neural Network |
EMD | Empirical Mode Decomposition |
EBSO | Enhanced Bee Swarm Optimization |
LSTM | Long-Short Term Memory |
CLSTM | Convolutional-Long Short Term Memory |
DOCLER | Deep Optimized Convolutional LSTM-Based Ensemble Reinforcement Learning |
SVM | Support Vector Machine |
DR-SVM | Distributionally-Robust Support Vector Machines |
SOM | Self-Organizing Map |
k-NN | k-Nearest Neighbors |
KNNR | K-Nearest Neighbour Based Routing Protocol |
KDE | Kernel Density Estimation |
ELM | Extreme Learning Machine |
KELM | Kernel Based Extreme Learning Machine |
Adaboost | Adaptive Boosting |
PSO | Particle Swarm Optimization |
LSSVM | Least Squares Support Vector Machine |
GMMN | Generative Moment Matching Network |
WindGMMN | Wind Power Using Generative Moment Matching Networks |
MSIN | Multi-Step Informer Network |
WPD | Wavelet Packet Decomposition |
VMD | Variational Mode Decomposition |
SSA | Salp Swarm Algorithms/Singular Spectrum Analysis |
IGWO | Improved Grey Wolf Optimization |
GRNN | Generalized Regression Neural Network |
SVR | Support Vector Regression |
HMMC | Higher-Order Multivariate Markov Chain |
MSTAN | Multi-Source and Temporal Attention Network |
ARMA | Auto-Regression Moving Average |
ARIMA | Autoregressive Integrated Moving Average |
MRE | Mean Relative Error |
MAE | Mean Absolute Error |
MBE | Mean Bias Error |
RMSE | Root Mean Squared Error |
MAPE | Mean Absolute Percent Error |
nMBE | Normalized Mean Bias Error |
nRMSE | Normalized Root Mean Squared Error |
R^{2} | Coefficient of Determination |
MG | Microgrid |
Name of forecasting system | R&D institutions | Methods |
---|---|---|
Prediktor | Danish National Laboratory | Physical methods |
SIPREÓLICO | University of Carlos III, Madrid, Spain | Physical methods |
HIRPOM | University College Cork, Ireland | Physical methods |
Previnto | University of Oldenburg, Germany | Physical methods |
WPFS Ver 1.0 system | China Electric Power Research Institute | Physical methods/ Meta-heuristic |
WPPT | Copenhagen University, Denmark | Statistical methods |
AWPPS | MINES ParisTech | Statistical methods, Fuzzy ANN |
RAL | Appleton Laboratory, Rutherford, UK | Statistical methods |
GH Forecaster | Garrad Hassan, UK | Statistical methods |
WPMS | Germany-ISET | Statistical methods, ANN |
Zephry | Risø National Laboratory | Statistical /Physical methods |
LocalPred-RegioPred | Spanish National Energy Center | Statistical /Physical methods |
ANEMOS | 23 scientific research institutions in 7 EU countries | Statistical /Physical methods |
eWind | True Wind USA, Inc. | Statistical /Physical methods |
WEPROG | University College Cork, Ireland | Statistical /Physical methods |
Time resolution | Reviewed works | Forecasting time scale |
---|---|---|
1 min | [22,29] | Ultra short term |
5 min | [15,16,29,48,50] | Ultra short term |
10 min | [15,28,30,43,50] | Ultra short term |
15 min | [12,15,20,21,23,30,40,41,44,47,50] | Ultra short term |
30 min | [15,16,23,26,30,50,51,52] | Ultra short term |
1 hr | [14,16,18,23,24,27,30,33,35,36,37,38,39,42,45,49,53] | Short term |
2 hr | [16,25,30] | Short term |
3 hr | [16,30] | Short term |
4 hr | [25,30] | Short term |
24 hr | [9,11,17,29,41] | Short term |
48 hr | [31,33] | Short term |
72 hr-1 week | [13,29] | Medium term |
1 month-years | [19] | Long term |
Ref | Method | Model Type | Parameters Used | Accuracy |
---|---|---|---|---|
[9] | -Other statistical analysis methods -Kernel density estimation |
Modified hidden Markov model | Wind speed, wind direction, wind power | RMSE=3.093 MAE=2.451 |
[10] | -Kernel density estimation | Distance weighted kernel density estimation (KDE) and regular vine (R-vine) copula | Wind power output, wind speed | RMSE=0.1089 MAE=0.075 |
[11] | -Kernel density estimation -Machine learning |
The k-NN and conditional KDE models | Historical wind power | MAE=3.18; RMSE=4.63; R^{2}=0.94 |
[12] | -Quantile regression method | A quantile passive–aggressive regression model for online convex optimization problems | Wind power | Pinball loss (PBL)=13.3 Average coverage error (ACE)=4.86%, Winkler score (WKS)=78.71 and Continuous ranked probability score (CRPS) =26.21 |
[13] | -Spatiotemporal forecasting -Quantile regression method |
Spatiotemporal quantile regression (SQR) | Wind power data | RMSE=16.62%; MAE=11.23% |
[14] | -Quantile regression method -AI or neural networks (NNs) |
A quantile regression neural network (QRNN) for regional wind power forecasting (RWPF) | Enhancing the abilities of nonlinear mapping and dealing with massive data | NMAE: DQR:9.086; QRNN:9.479 SBL:13.451; IFPA:13.967 NRMSE: DQR:10.917; QRNN:10.227 SBL:14.185; IFPA:14.538 |
[15] | -Spatiotemporal forecasting | A convolution-based spatial–temporal wind power predictor (CSTWPP) | Historical wind power | MASE= 190.02 RMSE=7.49 |
[16] | -Spatiotemporal forecasting | The spatiotemporal convolutional network (STCN) with a directed graph convolutional structure. | -Historical power data -STCN parameters selected by oneself |
MAEs =3.17% RMSEs =2.88%, |
[17] | -AI or neural networks (NNs) | Improved deep mixture density network model | Wind speed, wind direction, wind vector, wind power | NRMSE=0.138 |
[18] | -AI or neural networks (NNs) | New artificial neural network (ANN) models | Wind speed, wind direction, wind power output | Mean absolute relative error (MARE)=7.5%; R_{j}=5.4% (mean value of the Pearson correlation coefficient) |
[19] | -AI or neural networks (NNs) | A fuzzy logic approach for prediction of wind power output | Wind speed, air density | RMSE=1.04%; MAD=0.91% MSE=1.05% |
[20] | -AI or neural networks (NNs) -Hybrid model forecasting |
An ensemble neural forecast framework (ENFF) with three neural predictors for wind speed forecasting below. Elman neural network (ELM) Feedforward neural network (FNN) Radial basis function (RBF) neural network |
Wind speed, meteorological | Errors around 0.6 m/s |
[21] | -AI or Neural networks (NNs) | Day-ahead numerical weather prediction (NWP) with neural network |
The persistence method with BP three rolling prediction effect | The model accuracy improved by 7.61% and the RMSE reduced by 8.76% |
[22] | -AI or Neural networks (NNs) | -A classification model with the output wind power as the classification target -Use of Poisson re-sampling to replace the bootstrap method of the random forest to improve the training speed |
The random forest with Poisson re-sampling and set the parameters by oneself |
Mean square error (MSE) GBRT: 0.224; MLP: 0.117 Random forest with Bootstrap sampling: 0.111 Random forest with Poisson re-sampling: 0.096 |
[23] | -Ensemble methods | The CEEMDAN-IBA-GPR model | Historical wind power data | Stand deviation =10.42 |
[24] | -Ensemble methods -Hybrid model forecasting |
A multi-feature fusion self-attention mechanism graph convolutional network (MFF- SAM-GCN) forecasting model | Hyperparameter optimization of the predictive model by Bayesian optimization (BO) | RMSE of proposed (MFF-SAM- GCN) model is 0.0284, while the SMAPE is 9.453%, the MBE is 0.025, and R^{2} is 0.989. |
[25] | -Ensemble methods | Weighted multivariate time series motifs (WMTSM) and conditional LP (CLP) combined with the adaptive boundary quantiles (ABQs) | Wind speed, wind power | Both MAE and RMSE of less than 10% |
[26] | -Ensemble methods -Machine learning |
Ensemble learning models (GRF, RF, XGB) | Wind power, wind speed, gearbox bearing temperatures | R^{2} =98.9; RMSE=50.36 ; MAE=23.63 |
[27] | -Ensemble methods -Machine learning |
The five algorithms include wavelet neural network (WNN) trained by improved clonal selection algorithm (ICSA), WNN trained by PSO, and extreme learning machine (ELM)-based neural network, etc. The best performing models are the WNN trained by ICSA and ELM-based NN models. | Selecting parameters by using particle swarm optimization | The average nRMSE for WNN trained by ISCA, ELM, RBF, MLP, WNN trained by PSO are 5.4059%, 6.925%, 10.294%, 12.407%, and 17.038%. The average nMAE for WNN trained by ISCA, ELM, RBF, MLP, and WNN trained by PSO, are 4.2893%, 5.4787%, 8.2527%, 9.5773%, and 13.4847%. |
[28] | -Ensemble methods -Machine learning |
Enhanced bee swarm optimization (EBSO) to perform the parameter optimization for least squares support vector machine (LSSVM) | Picking parameters for LSSVM by enhanced bee swarm optimization (EBSO) | DR-SVM VMED(m/s): 6.895 MAE (m/s) : 0.723 RMSE(m/s): 0.932 MAPE(%): 11.87 CPU time(s): 148.15 |
[29] | -Ensemble methods | An exhaustive review of the state of the art of wind speed and power forecasting models for wind turbines located in different segments of power systems | Data preprocessing (EMD and ICEEMDAN) and parameter optimization | No description |
[30] | -Machine learning | The Adaboost-PSO-ELM method | Wind speed, wind direction, wind power | MAPE=0.0372; NBE=0.4621 RMBE=0.2950; R2=0.9857 |
[31] | -Machine learning | Salp swarm algorithms–extreme learning machine (SSA-ELM) | Wind speed, wind direction, temperature, atmospheric pressure, and other data are sampled every 10 minutes | MAPE=1.2677 RMSE=0.2576 |
[32] | -Deep learning | A deep optimized convolutional LSTM-based ensemble reinforcement learning strategy (DOCLER) | Wind power | RMSE=7.1322% MAE=4.6713% |
[33] | -Deep learning | A variational mode decomposition (VMD) and convolutional long short-term memory network (Conv LSTM) model | Wind power | MRE(KW)=0.016 (should be %) MAE(KW)=792 (should be %) MSE(KW)=1568305.38 RMSE(KW)=1252.32 (should be %) |
[34] | -Deep learning | A multi-source and temporal attention network (MSTAN) | Wind speed, pressure, temperature, humidity, and wind direction | NRMSE=0.154 NMAE=0.110 |
[35] | -Deep learning | Two-dimensional convolution neural network trained by improved accidental floater PSO | Fine-tuning the weights of TDCNN by proposed AFPSO |
Average error of four seasons MAPE:3.76 NMAE:2.46 NRMSE:3.12 |
[36] | -Deep learning -Hybrid model forecasting |
The WD-IGFCM-LSTMS model for the accuracy of short-term wind power forecasting (WPF) approach | The best parameters determined by IGWO algorithm | Case A: NMAE 10.32%; NRMSE 14.59% CR: 85.41%; QR: 91.53% Case B: NMAE 10.18%; NRMSE 13.52% CR: 86.48%; QR: 91.53% |
[37] | -Deep learning | Deep neural network: LSTM method (best); MLP (second best) while using SVR, KNNR, and physical model with an expert correction |
More LSTM parameters and set these parameters by oneself | INT_OUT_EXT[GBT, RF, PHYS(v1&v2)→KNNR, MLP, LSTM] with additional expert SS:0.5925; nMAE[%]:11.3055 nRMSE:0.1618; nMBE:0.0146 |
[38] | -Deep learning | -Optimizing the hyperparameters of the LSTM network by the modified PSO algorithm -A PSO_LSTM model |
Selecting parameters by PSO | MPSO_ATT_LSTM MAPE: 4.6%; MAE: 211.5 kW Device capacity > 20000kW |
[39] | -Deep learning | Advanced deep learning techniques Encoder–Decoder LSTM |
Setting parameters by oneself | Annual and monthly errors |
[40] | -Deep learning | The CNN-MLSTMs-T Model | Wind power | RMSE=0.1998; MAE=0.1523 |
[41] | -Deep learning | Generative moment matching network (GMMN) | Historical wind power | PINAW=8.66MW; PICP=84% RMSE=127.10; MAE=0.6855MW |
[42] | -Deep learning | Bidirectional long short-term memory (Bi-LSTM) | Manual adjustment layers | Error can be divided into training, test and validation errors |
[43] | -Deep learning | Multi-step informer network (MSIN) | Manual selection of parameters | Multi-step informer network (MSIN) improves forecast accuracy by 29% compared with informer network for RMSE |
[44] | -Deep learning | Long short-term memory neural network (LSTM) with the improved particle swarm optimization algorithm (IPSO) | Determining the LSTM and DENSE layers, the number of neurons | VMD-CNN-IPSO-LSTM MAE:2.92668; RMSE:3.59604 MAPE:0.20147; adj-R^{2}:0.96639 |
[45] | -Hybrid model forecasting |
Generalized regression neural network (GRNN) and support vector machine (SVM) |
Turning GRNN and SVM parameters by oneself | The GRNN model gives the CC value of 0.956, RMSE of 28.82, and the SVR model gives the CC value of 0.965 and RMSE value of 44.40. |
[46] | -Hybrid model forecasting | The WPD-VMD-SSA-IGWO-KELM model | Wind speed |
NMAE=11.2% MAPE=4.2% |
[47] | -Other statistical analysis methods | Higher-order multivariate Markov chain (HMMC) | Wind power; PV power, Heat index |
NRMSE=2.59 |
[48] | -Other statistical analysis methods | Five minute-ahead wind power forecasts in terms of point forecast skill scores and calibration | To deduce the value of kernel methods for parameter adjustment | The error value is represented by a picture rather than a simple number. |
[49] | -Other statistical analysis methods | RL-Based ESS operation strategy | No description | 1% point analysis gap to the optimal solution, which requires complete information, including future values |
[50] | -Other statistical analysis methods | Regression and curve fitting by weather research and forecasting (WRF) and wind farm parameterization (WFP) | No description | No description |
[51] | -Other statistical analysis methods | Empirical dynamic modeling (EDM)-based probabilistic forecast | Historical wind turbine power | CRPS (%)=5.12 |
[52] | -Other statistical analysis methods | Multi-class autoregressive moving average (ARMA) | Historical wind power | RMSE=127.10 MAPE=1.25% |
[53] | -Other statistical analysis methods | Renewable energy is directly distributed to power dispatch | Incorporating renewable energy into the power flow | With an increase in power by 1.6 times, there is a decrease in energy of RES by 15-19. |
Work | Date of publication | Main contributions | Advantages | Disadvantages | Approaches |
---|---|---|---|---|---|
[9] | Jan./Feb. 2022 |
A wind power forecasting (WPF) system including WRF-based wind forecasting, modified HMM-based wind speed correction, and a kernel distribution estimation (KDE)-based WPF module | Enhancing the WPF accuracy from deterministic and probabilistic forecast | -Very time-consuming in huge computational burden -Complex configuration |
Classification and regression algorithms |
[10] | January 2020 |
The model is more accurate and flexible than the Gaussian copula model | Abundant bi-variate copula functions are available to make the model more accurate | -Complex structure and hardware requirement | Classification and regression algorithms |
[11] | November 2022 | Simple to improve the accuracy of aggregated point and wind power forecasts that can be derived from decentralized point forecasts | Providing system operators with a way of aggregating these forecasts while taking into account spatial and temporal correlations of wind power | -Being difficult in selecting good bandwidths in the presence of large datasets and high dimensionality | Classification and regression algorithms |
[12] | April 2022 |
Online ensemble learning framework for wind power forecasting that utilizes solid individual forecasting models and new information | Higher accuracy and lower computation complexity | -Time-consuming computation process -Excessive parameter adjustment |
Classification and regression algorithms |
[13] | Nov./Dec. 2020 |
An SQR model is proposed, which is a new nonparametric probabilistic prediction method | Providing an efficient framework for regional wind power probabilistic prediction with highly reliable performance | Complex nonlinear and high dimensional structure | Classification and regression algorithms |
[14] | October 2021 |
On basis of the QRNN, the structure of the DNN is improved to adapt regional wind power forecasting as well as constructing the DQR | -The deep quantile regression is proposed to improve the performance of the QRNN -The local-connected method is applied to the input layer of the neural network for tackling the challenge of the massive data |
Each test takes 72 hours, so it is impossible to clearly determine its effects with no parameters and time | Classification and regression algorithms |
[15] | June 2020 |
The deep architecture and nonlinearity of CSTWPP, spatial–temporal features inside the power of multiple wind farms can be effectively extracted. The accuracy of this short-term forecasting approach is significantly higher than existing models. | -The powerful ability of CSTWPP to extract spatial–temporal features from multiple wind farms -Superiority over other competing methods |
- Time-consuming computation process -Graphics processing units (GPU) is used to speed up the training process | Deep Learning algorithms |
[16] | March 2022 |
-A deep learning architecture STCN based on a graph model for spatiotemporal wind power forecasting -A novel directed graph model and the corresponding GCN structure |
Fewer input time steps cause the STCN model to learn temporal features insufficiently, affecting the forecasting results | The STCN model exhibit certain interpret-ability, which is not available in traditional deep learning models |
Deep Learning algorithms |
[17] | July 2020 |
The improved deep mixture density network (IDMDN) has better function approximation and density estimation ability than conventional shallow MDN | It is not necessary to obtain the deterministic prediction result firstly and acquire the probabilistic result by post-processing | -Slow convergence speed | Feed-forward neural network algorithms |
[18] | May 2020 |
Improvement in the efficiency and stability of ANN models by varying the number of prior 1 h periods | Improvement of model performance, efficiency and stability for the performance of ANN- based WPF models | -Possible disadvantages of ANN Model in short term prediction of wind power generation | Feed-forward neural network algorithms |
[19] | February 2021 |
Developing fuzzy model and model predictive control for prediction of wind power for the particularly selected location in India | The proposed models can be employed for the estimation of wind speed and wind power generation of any location in the world with having the complete information | -The time for calculating the wind power is 1 second. It is difficult to estimate the wind power during the summer period in which wind speed is very low. | Rule-based algorithms |
[20] | January 2021 |
Development of a new ensemble neural forecast framework (ENFF) to accurately forecast the wind speed | -Enhancing the utilization of super-capacitor energy storage (SCES) as the N-1 contingency events -Being easily extended to N-1-1 contingency |
It is not economical to deploy energy storage only for VIS as these events are less frequent in power systems | Feed-forward neural network algorithms |
[21] | August 2020 |
Located by the NWP information and time windows to improve the low forecasting accuracy of rolling WPP | The hybrid approach combined with neural network and persistence method | -The relevance of the day ahead is doubted due to the great change of wind -The setting of neural network parameters is a big issue |
Feed-forward neural network algorithms |
[22] | January 2021 |
An improved random forest short-term prediction model based on the hierarchical output power | -Discretizing the power data to divide the large- scale training data and remove abnormal data -Fewer regression trees -Better performance |
The tree size has a great impact. -It is slow to run if there are more trees -It is not accurate if there are fewer trees |
Classification and regression algorithms |
[23] | March 2020 |
The probabilistic wind power forecasting (WPF) results are utilized as one part in the micro-grid (MG) system for optimal dispatching | Automatically generate optimal compromise solutions for decision makers | Longer training duration | Classification and regression algorithms |
[24] | April 2022 |
-A Bi-LSTM network and1D-CNN in parallel connection to form a multi-feature fusion (MFF) framework, which can extract spatiotemporal correlation features of the load data -The eigenvalue can be found to reduce the data |
-Enhancing the feature extraction capability of the 1D-CNN network by a self- attention mechanism | More LSTM parameter settings require to be adjusted |
Deep Learning algorithms |
[25] | January 2022 |
-A hierarchical clustering method based on weighted multivariate time series motifs (WMTSM) is used to analyze the static, dynamic, and meteorological characteristics of regional wind power -Based on the clustering analysis, the correlation coefficients are formulated as the weights for the accuracy of samples to optimize the cost function of conditional LP (CLP) |
The result of clustering, the CLP for each cluster is quantified, which can improve the accuracy of sample utilization, and further enhance the performance of CNQR | -Highly affected by the accuracy of NWP and the static relationship between the wind power and speed | Feed-forward neural network algorithms |
[26] | April 2020 |
Ensemble learning models provide a better prediction of wind power | Better performance of wind power prediction by the ensemble models considering lagged data | Spatiotemporal dependencies are not considered in ensemble learning models and machine learning models | Classification and regression algorithms |
[27] | August 2020 |
-A novel hybrid neural network (NN)-based day-ahead (24 hour horizon) wind speed forecasting is proposed -Single- and multi-features and their effect on the accuracy of wind speed prediction are analyzed |
-Very effective for day-ahead wind speed prediction - Only need wind speed as a feature |
It is not clear how to select the eigenvalues of historical data | Feed-forward neural network algorithms |
[28] | November 2020 |
The data regression (DR) algorithm gets meaningful training data to reduce the number of modeling data and improve the computing efficiency | Effectively reduce computing time by data regression algorithm | It is difficult to assess errors for wind speed forecasts due to large variations | Classification and regression algorithms |
[29] | September 2022 | A comprehensive review of the state of the art of wind speed and power forecasting models is presented for wind turbines located in different segments of power systems | Due to the variable nature of wind speed and its relationship to meteorological variables, it is possible to study the accuracy of integrating physical forecasting methods into hybrid models | Individual forecasting algorithm has been replaced by hybrid algorithms combining mainly AI-based and statistical methods | Feed-forward neural network algorithms |
[30] | July 2021 |
-This model has good generalization ability and robustness -Providing a more reliable basis for power grid dispatch |
Higher accuracy and better generalization ability by Adaboost-PSO-ELM wind power prediction model | The training samples are selected based on experience | Feed-forward neural network algorithms |
[31] | March 2020 |
Improving the accuracy of ultra-short-term wind power prediction | Better performance based on SSA analog integrated circuit design | Longer training time | Feed-forward neural network algorithms |
[32] | February 2022 |
A combined deep neural network model by integrating the advantages of the CNN and LSTM neural network | Excellent performance of the proposed algorithm in comparison to several state of-the-art WPF models | Low computational ability of the algorithm in selecting CLSTM hyperparameters | Deep Learning algorithms |
[33] | July 2020 |
A short-term wind power forecasting model including VMD decomposition, ConvLSTM predictor and error series modeling | Removing the nonstationary features of the raw wind power series |
A large computational cost to obtain the optimal parameters | Deep Learning algorithms |
[34] | October 2021 |
Multi-source NWP is used in WPPF, and its long-term temporal error pattern is discussed | -Higher deterministic prediction accuracy -Better probabilistic evaluation score |
-Time-consuming computation process -Complex configuration |
Deep Learning algorithms |
[35] | September 2020 |
-The proposed forecasting engine composed two-dimensional convolution neural network (TDCNN) -Trained by improved optimization algorithm based on particle swarm optimization |
Fine-tuning the weights of TDCNN to increase the prediction accuracy of the forecast engine | Longer time for model training due to high requirement of data quality | Deep Learning algorithms |
[36] | March/April 2022 | -The WD-IGFCM-LSTMS model -Six fluctuation features that reflect the shape characteristics are extracted to quantify the partitioned waves |
Improving the global searching ability of the GWO to select the initial clustering center of fuzzy C-means more effectively | -Longer time to calculate the parameters of fuzzy C-means, modeling and prediction of LSTM | Deep Learning algorithms |
[37] | February 2022 |
Using meteorological forecasts from two NWP models (ECMWF and GFS) as input data yields better results than using a single NWP model | -Presenting real wind speed data and obtain higher accuracy of models with more numerical weather prediction (NWP) points | -We have tested it to find that too much input data will result in slow operation -Only one hour can be predicted | Deep Learning algorithms |
[38] | June 2022 |
-Speeding up the convergence of the model dramatically to avoids falling into local optima. -Reducing the influence of man-made random selection of LSTM network hyperparameters on the prediction results | Lower influence of parameters | - LSTM has more parameters than SVM, GRMM, RBF, and other modeling methods, so it will become a big issue in computation process - The computation times will increase significantly when the amount of training data is large -The accuracy needs to be confirmed |
Deep Learning algorithms |
[39] | June 2022 |
The encoder–decoder LSTM for medium-term wind speed based on a real-time measurement dataset, which were compared with two well- known conventional methods | Easy to determine the eigenvalue by using encoder–decoder | Improper correlation and weight ratio of coded data | Deep Learning algorithms |
[40] | May 2021 |
The sample classification features mined by the CNN are submitted to the numerical prediction task as supplementary knowledge to help the training of the LSTM prediction models | -Effectively improve the accuracy of WPF based on sample similarity analysis | Exploring whether mode classification task can provide valuable knowledge for numerical prediction | Deep Learning algorithms |
[41] | October 2022 |
-Based on historical observations, combined with deterministic point forecasts, WindGMMN is developed to generate a large number of realistic wind power scenarios with similar characteristics to real wind power -The proposed WindGMMN is unbound from statistical hypotheses -Producing a series of possible forecasting scenarios without a time horizon and number restrictions by simply adjusting parameters |
-Capturing the probability distribution characteristics of actual wind power scenarios -Reflecting the temporal correlations of wind power scenarios |
It is difficult to distinguish which prediction interval is better | Deep Learning algorithms |
[42] | April 2021 |
Significant improvements in the peak value forecasting have been observed by using the fused network of short and long Bi-LSTM networks with DRNets | Bi-LSTM network can improve performance by eliminating propagated errors | -The test for bidirectional long short-term memory (Bi-LSTM) layers will be more troublesome -The LSTM prone to overfitting due to the increasing depth of DNN, which degrades the performance of the deep learning model |
Deep Learning algorithms |
[43] | September 2022 | -Accurate mid- and long-term wind power forecasting can provide an important basis for power distribution and energy storage configuration after wind power grid-connected -A dynamic pressure model is introduced in MSIN to modify wind power generation forecast with having highly correlated physical characteristics |
-Multi-step informer network improves forecast accuracy by 29% compared with informer network -The multi-step process is beneficial to the anti-risk ability and security of the network |
In the case of ignoring meteorological factors such as surface temperature and relative humidity, the coupling factors between multiple wind turbines need to be considered in the research | Deep Learning algorithms |
[44] | September 2022 | Considering the nonlinear and fluctuating characteristics of wind speed and wind power series, a hybrid short-term wind power forecasting model based on data decomposition (VMD) and joint deep neural network (CNN-LSTM) is proposed | The wind speed and wind power sequences in the input data are decomposed by variational mode decomposition to reduce the noise in the raw signal | Using IPSO to select the layers and neurons of LSTM would cause the issue of too long calculation time | Deep Learning algorithms |
[45] | May 2022 |
-The GRNN model is better than the SVR model regarding the RMSE value -The inclusion of average electrical load data is possible when the forecasting system can obtain near real-time observation data of electricity load |
Which weather parameters affect the electricity load? Primary impact is temperature; secondary impact is wind speed. |
The temperature affects the power demand in terms of load forecasting | Feed-forward neural network algorithms |
[46] | April 2020 |
Significantly increase the accuracy in short term wind power prediction | -Greatly extract the trend information of wind power -Better accuracy in short-term wind power prediction |
The structure of proposed nonlinear combination model | Feed-forward neural network algorithms |
[47] | April 2020 |
A flexible framework for forecasting wind generated power in the case of disjointed batch of historical data to enhance the accuracy of wind power output modeling | Reaching better performance of the forecast algorithm | -Complex computation process -Increasing dimensions of the input vector |
Classification and regression algorithms |
[48] | August 2021 |
Identifying changes in the time series, avoiding abrupt loss of information, and maintaining a controlled number of examples, since there is adaptive selection of the active kernel | Dealing with the increasing kernel matrix size associated with time and memory complexities, and the overfitting problem | -An accurate very-short term forecasts for one or multiple wind farms | Classification and regression algorithms |
[49] | February 2020 |
-Improving learning performance by reducing the variance of the WPG forecast uncertainty -Extensive simulations based on practical WPG generation data and forecasting |
Managing the wind power generation (WPG) forecast uncertainty by a reinforcement leaning-based ESS operation strategy | -Requiring complete information, including future values, to achieve cross period prediction | Rule-based algorithms |
[50] | June 2022 |
The weather research and forecasting (WRF) model is utilized with the wind farm parameterization (WFP) method for short-range wind power forecasting simulation | -The horizontal downsize method can prolong more than 10 km in the velocity field, especially for higher incoming wind velocity -Improving the accuracy of the power forecast for higher wind speed simulations | -Adding some complexity of the correlation works -There may be time and space difference between real-time data and geographical location. -Leading to uncertainty of data |
Classification and regression algorithms |
[51] | April 2020 |
The proposed approach is effective on achieving probabilistic WTPF with high reliability as well as satisfactory sharpness | -Developed for the wind power forecasting by the EDM method -Applied to estimate the uncertain behavior of WTP |
-Obtaining the highest CRPS values -The poorest performance. |
Classification and regression algorithms |
[52] | September 2022 | Lower training complexity in the proposed model to ensure prediction accuracy compared with traditional models | -Tackling seasonality and randomness of wind power with moderate model complexity -Effectively guarantees the convergence speed and efficiency of the training process |
If the input data are nonstationary so that the proposed data preprocessing fails, the proposed model may not be able to obtain accurate prediction results | Classification and regression algorithms |
[53] | September 2022 | The load schedule, electricity consumption, use of installed power, boundary conditions of generation, and ensuring energy balance were taken into account | For the average monthly values of renewable energy sources generation with an increase in power by 1.6 times, there was a decrease in electricity consumption by 1.57–4 times. | The selection of the WG installation location for load object is difficult to comprehensively take into account, since the system load will vary from day to day | Rule-based algorithms |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
Moniki Ferreira
et al.
,
2018
Saeed Samadianfard
et al.
,
2020
Md. Alamgir Hossain
et al.
,
2020
© 2024 MDPI (Basel, Switzerland) unless otherwise stated