Search

Improving short-term wind speed prediction accuracy and stability remains a challenge for wind forecasting researchers. This paper proposes a new variational mode decomposition (VMD)-attention-based spatio-temporal network (VASTN) method that takes advantage of both temporal and spatial correlations of wind speed. First, VASTN is a hybrid wind speed prediction model that combines VMD, squeeze-and-excitation network (SENet), and attention mechanism (AM)-based bidirectional long short-term memory (BiLSTM). VASTN initially employs VMD to decompose the wind speed matrix into a series of intrinsic mode functions (IMF). Then, to extract the spatial features at the bottom of the model, each IMF employs an improved convolutional neural network algorithm based on channel AM, also known as SENet. Second, it combines BiLSTM and AM at the top layer to extract aggregated spatial features and capture temporal dependencies. Finally, VASTN accumulates the predictions of each IMF to obtain the predicted wind speed. This method employs VMD to reduce the randomness and instability of the original data before employing AM to improve prediction accuracy through mapping weight and parameter learning. Experimental results on real-world data demonstrate VASTN's superiority over previous related algorithms. [ABSTRACT FROM AUTHOR]

Reliable short-term wind speed prediction is critical for ensuring the rational exploitation and utilization of wind energy. However, due to complex characteristics (e.g., nonstationarity, nonlinearity, uncertainty, etc.) of natural winds, the realization of this task usually confronts a great challenge. For this purpose, an innovative method for forecasting short-term wind speeds is developed based on the principle of "decomposition-prediction-ensemble". Concretely, a new pretreatment technique, including boxplot figure based abnormal data diagnosis and correction, multivariate fast iterative filtering based time-frequency decomposition, and improved amplitude and frequency modulation based subseries reconstruction, is first developed to perform the high-quality data preprocessing. Then, three different algorithms in conjunction with the correntropy loss and consideration of model diversity are designed as high-performance predictors to capture more data characteristics. Further, a hybrid ensemble strategy combining stacking ensemble and hierarchical ensemble is developed to learn the potential interaction or nonlinear correlation among decomposed subseries as well as some uncertain information for high-reliability prediction. The eventual predictions are given in the form of deterministic point-value, interval, and real-time probability density function. Numerical examples based on four sets of multi-height wind speed data prove the effectiveness and superiority of the proposed method. For example, the average promotion obtained by this method compared with univariate conditional kernel density estimation in terms of mean absolute error is 33.87%, while the improvement in terms of coverage width criterion is 29.64%. • A new pretreatment technique is designed for high-quality data processing. • Three predictors with high robustness and strong generalization are developed. • Hybrid ensemble strategy can yield deterministic and probabilistic predictions. [ABSTRACT FROM AUTHOR]

Wind energy is important to the transformation and development of global energy, because it is clean and renewable. However, the productivity of wind power is low due to its volatility, randomness, and uncertainty. Therefore, a new hybrid prediction model based on combined Elman-radial basis function (RBF) and Lorenz disturbance is proposed, which can promote the productivity of wind power by better predicting wind speed, firstly, applying the variational mode decomposition (VMD) algorithm to original nonstationary wind speed data to obtain several relatively stationary intrinsic mode functions (IMF), so as to fully exploit its potential characteristics. Meanwhile, the sample entropy is introduced to determine the decomposition number K. Afterwards, different IMF components with different characteristics are used for training and prediction: Elman neural network with sensitivity to historical state data is used for wind speed trend components; RBF with strong nonlinear mapping capability is adopted for other stochastic modal components. Next, the first-step prediction values can be obtained by reconstructing the predicted results of the respective IMF components. Finally, the Lorenz equation is introduced in view of the effects of atmospheric disturbances on wind fluctuations, which can be used to revise the first-step prediction results to obtain more realistic prediction results. By experimenting with the real data from two different wind farms and comparing with other predictive models, we found that (1) VMD can solve the problem of modal aliasing in empirical mode decomposition, to obtain a better decomposition result; (2) the combined prediction method of Elman and RBF is used for modal components, that is, different algorithms are adopted for different components, which have better prediction effects; (3) in this research, the results of the proposed combination prediction model is more accurate by comparison with the other neural network models. This research work will help the power system to rationally formulate wind farm control strategies, enhance the self-regulation of wind farm, and further promote global energy innovation. [ABSTRACT FROM AUTHOR]

• A new improved RBF neural network model is proposed. • VMD-WT is used to extract features and remove noise of wind speed data. • The PCA-BP method is proposed to screen out the model input vectors. • The validity and applicability of the model are tested by data from two wind farms. Accurate short-term wind power forecasting is significant for rational dispatching of the power grid and ensuring the power supply quality. In order to enhance the accuracy of short-term wind speed prediction, a hybrid model based on VMD-WT and PCA-BP-RBF neural network is proposed. In data pre-processing period, the non-stationary wind speed sequence is decomposed into a number of relatively stationary intrinsic mode functions (IMF) by variational mode decomposition (VMD); then WT algorithm is used to perform secondary denoising on each IMF. At the same time, several factors affecting wind speed are introduced, from which the input features that participated in the prediction are selected by PCA-BP method. Next, the RBF neural network is utilized to predict each IMF. Finally, all IMF prediction results are aggregated to obtain the final wind speed value. Combining the data of Spanish and Chinese wind farms, the experiment results show that: (1) compared with EMD, VMD-WT can better solve the problems of modal aliasing and endpoint effect, which can make the periodic characteristics of each IMF more obvious, then promote the forecasting performance; (2) using PCA-BP method to filter the model input data, the redundant and irrelevant information is eliminated, the complexity of the model is reduced, and the predictive performance of RBF model is improved; (3) compared with other traditional models, the hybrid model proposed in this paper has greatly improved the accuracy in short-term wind speed forecasting. [ABSTRACT FROM AUTHOR]

The safe and stable operation of wind power systems requires the support of wind speed prediction. To ensure the controllability and stability of smart grid dispatching, a novel hybrid model consisting of data-adaptive decomposition, reinforcement learning ensemble, and improved error correction is established for short-term wind speed forecasting. In decomposition module, empirical wavelet transform algorithm is used to adaptively disassemble and reconstruct the wind speed series. In ensemble module, Q-learning is utilized to integrate gated recurrent unit, bidirectional long short-term memory, and deep belief network. In error correction module, wavelet packet decomposition and outlier-robust extreme learning machine are combined to developing predictable components. An appropriate correction shrinkage rate is used to obtain the best correction effect. Ljung-Box Q-Test is utilized to judge the termination of the error correction iteration. Four real data are utilized to validate model performance in the case study. Experimental results show that: (a) The proposed hybrid model can accurately capture the changes of wind data. Taking 1-step prediction results as an example, the mean absolute errors for site #1, #2, #3, and #4 are 0.0829 m/s, 0.0661 m/s, 0.0906 m/s, and 0.0803 m/s, respectively; (b) Compared with several state-of-the-art models, the proposed model has the best prediction performance. • A reinforcement learning ensemble strategy is developed to determine the best model. • Three deep learning benchmark predictors are combined to learn the law of wind speed. • An improved error correction technique is proposed. • A correction shrinkage rate is introduced to reduce the risk of overfitting. [ABSTRACT FROM AUTHOR]

The Canary Islands are a well known tourist destination with generally stable and clement weather conditions. However, occasionally extreme weather conditions occur, which although very unusual, may cause severe damage to the local economy. The ViMetRi-MAC EU funded project has among its goals, managing climate-change-associated risks. The Spanish National Meteorology Agency (AEMET) has a network of weather stations across the eight Canary Islands. Using data from those stations, we propose a novel methodology for the prediction of maximum wind speed in order to trigger an early alert for extreme weather conditions. The methodology proposed has the added value of using an innovative kind of machine learning that is based on the data stream mining paradigm. This type of machine learning system relies on two important features: models are learned incrementally and adaptively. That means the learner tunes the models gradually and endlessly as new observations are received and also modifies it when there is concept drift (statistical instability), in the modeled phenomenon. The results presented seem to prove that this data stream mining approach is a good fit for this kind of problem, clearly improving the results obtained with the accumulative non-adaptive version of the methodology. [ABSTRACT FROM AUTHOR]