Complete Ensemble Research Articles

Machinery structural crack damage detection based on acoustic signal with ICEEMDAN and sensitive IMF fuzzy entropy

Abstract With the increasingly urgent demand on the reliability of mechanical equipment, in the process of production and service, it is of vital importance to apply precise and efficient crack damage detection on the critical structures. To overcome the shortcomings of existing damage detection methods and meet the urgent needs of engineering practice, by analyzing acoustic signal, a novel machinery structural crack damage detection method based on improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) and sensitive intrinsic mode function (IMF) fuzzy entropy is proposed in this paper. Firstly, redundant second-generation wavelet denoising strategy based on neighborhood correlation is applied on the raw acoustic signal. Then, the pre-denoised acoustic signal is decomposed by ICEEMDAN to obtain a set of IMFs, and the fuzzy entropies of the first eight IMFs are calculated to reflect the structural crack damage states. Finally, with distance evaluation technique, the most sensitive IMF fuzzy entropy is selected and defined as the damage index to assess the structural crack damage levels. Effectiveness of the proposed method is validated by two case studies as for the crack damage detection of different machinery structures. The results show that the defined damage index is not only sensitive to the occurrence of structural crack damage, but also decreases obviously with the increasing damage level and is not affected by damage location.

Data-Driven Soft Sensor Model Based on Multi-Timescale Feature Fusion for Crystal Quality Prediction in Czochralski Process

The accurate real-time prediction of the crystal quality index v/G is an important reference for the real-time monitoring of the growth quality status and the process optimization adjustment of semiconductor silicon single crystals. This paper proposes a data-driven crystal quality indicator v/G soft sensor prediction model based on multi-timescale feature fusion to achieve the effective prediction of the crystal quality indicator v/G. Firstly, the characteristics of the crystal quality index v/G in the growth process of Czochralski silicon single crystal are analyzed. Secondly, the crystal quality index v/G is broken down into several natural components using something called complete ensemble empirical mode decomposition with adaptive noise (CEEMDAD), which provides more stable data. On this basis, each intrinsic mode component is reconstructed according to the sample entropy. Then, the maximum mutual information coefficient (MIC) method is applied to identify the characteristic variables most closely associated with each reconstructed component of the crystal quality index v/G from the process-influencing factors. Then, a long short-term memory network with a self-attention mechanism is used to establish a prediction model of the reconstructed components to extract the multi-timescale feature information from the different components of the crystal quality index v/G. Finally, the prediction results of the crystal quality index v/G are obtained by fusing each subsequent prediction model. According to the actual field data, the comprehensive experimental results validate the efficacy of the proposed soft sensor modeling method for the crystal quality index v/G. Compared with the single model, the proposed prediction model has a smaller MAE, RMSE, and prediction performance index and a higher HR prediction hit rate.

Robust adaptive optimization for sustainable water demand prediction in water distribution systems

The advancement of the Internet of Things has positioned intelligent water demand forecasting as a critical component in the quest for sustainable water resource management. Despite the potential benefits, the inherent non-stationarity of water consumption data poses significant hurdles to the predictive accuracy of forecasting models. This study introduces a novel approach, the Robust Adaptive Optimization Decomposition (RAOD) strategy, which integrates a deep neural network to address these challenges. The RAOD strategy leverages the Complete Ensemble Empirical Mode Decomposition (CEEMD) to preprocess the water demand series, mitigating the effects of non-stationarity and non-linearity. To further enhance the model’s robustness, an innovative optimization algorithm is incorporated within the CEEMD process to minimize the variance in multi-scale arrangement entropy among the decomposed components, thereby improving the model’s generalization capabilities. The predictive power of the proposed model is harnessed through the construction of deep neural networks that utilize the decomposed data to forecast minutely water demand. To validate the effectiveness of the RAOD strategy, real-world datasets from four distinct geographical regions are employed for multi-step ahead predictions. The experimental outcomes demonstrate that the RAOD model outperforms existing models across all considered metrics, highlighting its suitability for accurate and reliable water demand forecasting in the context of sustainable energy management.

An innovative lost circulation forecasting framework utilizing multivariate feature trend analysis

The prompt and precise prediction of lost circulation is essential for safeguarding the security of drilling operations in the field. This study introduces a lost circulation prediction model convolutional neural networks-long short-term memory-feature-time graph attention network-transformer (CL-FTGTR) that combines improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) data decomposition and trend reconstruction. A notable feature of this model is the utilization of an innovative logging data analysis technique for processing drilling fluid and engineering parameters, and the synthesis of two consecutive encoding modules: Feature-GAN-transformer (FGTR) and time-GAN-transformer (TGTR). Experimental results confirm the following: ① The ICEEMDAN algorithm can effectively filter out noise in logging data and extract trend components, minimizing the impact of noise on prediction outcomes. ② Convolutional neural networks-long short-term memory (CLSTM) position encoding module, substituting traditional sin-cos encoding, significantly improves the model's ability to encapsulate global information within the input data. ③ The FGTR and TGTR modules are capable of efficiently handling feature and time dimension information in logging data, leading to a significant enhancement in the performance of the lost circulation prediction model. The CL-FTGTR model was experimentally tested across four wells in the same block, with the essentiality of its modules confirmed by five metrics. The CL-FTGTR model attained peak precision, recall, F1PA%K, and area under the curve values of 0.908, 0.948, 0.967, and 0.927, respectively. The findings demonstrate that the CL-FTGTR model for predicting lost circulation boasts high precision and dependability.

Real-time identification of foundation damage in high-Pile Wharves: Nonlinear feature change point analysis in dynamic characteristics under wave excitation

Waves are a common excitation form for wharves. The responses of pile foundation in high-pile wharves under such excitation exhibit pronounced nonlinear and non-stationary characteristics, compromising the robustness of damage detection methods. To address this challenge, this paper first introduces a method using Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and spectral feature analysis, aimed at extracting the overall characteristic sub-signals to resolve multi-type signal aliasing. Secondly, the Generalized Autoregressive Conditional Heteroskedasticity (GARCH) model is employed to isolate the nonlinear dynamic features of the response under wave excitation. Thirdly, the Weighted Residual Cumulative Sum (WRCUSUM) method is developed to monitor changes in response characteristics online, thereby enabling the identification of damage occurrence and its timing. Further, the efficacy of the proposed method has been validated through high-pile wharf experiments conducted at various damage levels. This approach not only offers a novel solution for online damage detection under wave excitation but also introduces a new strategy for non-destructive testing across offshore engineering applications.

A novel decomposition-reconstruction-prediction hybrid model to predict precipitable water vapor from GNSS Observations

Abstract Precipitable water vapor (PWV) is one of the key factors in weather disaster preparedness and water forecasting. Due to the nonlinearity and nonstationarity of the PWV sequence, the existing models cannot achieve stable prediction accuracy. Therefore, this paper proposes a novel decomposition-reconstruction-prediction hybrid prediction model, named ICEEMDAN-PE-ARIMAX-1D CNN-BiLSTM, for predicting PWV. Firstly, the PWV sequence derived from the Global Navigation Satellite System (GNSS) is decomposed using the improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN). Secondly, the permutation entropy (PE) of each decomposed modal is calculated and the PWV sequence is reconstructed into low- and high-frequency components. Then, considering spatio-temporal information, geographic information and meteorological data (longitude, latitude, altitude, Doy, Hod, surface air pressure and temperature) aided modelling, the low- and high-frequency components are predicted using autoregressive integrated moving average with exogenous inputs (ARIMAX) and one-dimensional convolution neural network-bidirectional long-short-term memory (1D CNN-BiLSTM), respectively, and superimposed to obtain the predicted value and prediction intervals. Finally, the proposed model is validated for performance using the PWV derived from six GNSS stations. Compared with other models, the results show that ICEEMDAN-PE-ARIMAX-1D CNN-BiLSTM significantly improves PWV prediction performance, the mean RMSEs for the six stations are 0.3765 mm, 0.7517 mm, and 1.4696 mm for the 1-hour, 12-hour, and 24-hour forecasts, respectively.

Short-Medium-Term Solar Irradiance Forecasting with a CEEMDAN-CNN-ATT-LSTM Hybrid Model Using Meteorological Data

In recent years, the adverse effects of climate change have increased rapidly worldwide, driving countries to transition to clean energy sources such as solar and wind. However, these energies face challenges such as cloud cover, precipitation, wind speed, and temperature, which introduce variability and intermittency in power generation, making integration into the interconnected grid difficult. To achieve this, we present a novel hybrid deep learning model, CEEMDAN-CNN-ATT-LSTM, for short- and medium-term solar irradiance prediction. The model utilizes complete empirical ensemble modal decomposition with adaptive noise (CEEMDAN) to extract intrinsic seasonal patterns in solar irradiance. In addition, it employs a hybrid encoder-decoder framework that combines convolutional neural networks (CNN) to capture spatial relationships between variables, an attention mechanism (ATT) to identify long-term patterns, and a long short-term memory (LSTM) network to capture short-term dependencies in time series data. This model has been validated using meteorological data in a more than 2400 masl region characterized by complex climatic conditions south of Ecuador. It was able to predict irradiance at 1, 6, and 12 h horizons, with a mean absolute error (MAE) of 99.89 W/m2 in winter and 110.13 W/m2 in summer, outperforming the reference methods of this study. These results demonstrate that our model represents progress in contributing to the scientific community in the field of solar energy in environments with high climatic variability and its applicability in real scenarios.

Research on the rapid diagnosis method for hunting of high-speed trains

PurposeHigh-speed turnouts are more complex in structure and thus may cause abnormal vibration of high-speed train car body, affecting driving safety and passenger riding experience. Therefore, it is necessary to analyze the data characteristics of continuous hunting of high-speed trains passing through turnouts and propose a diagnostic method for engineering applications.Design/methodology/approachFirst, Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) is performed to determine the first characteristic component of the car body’s lateral acceleration. Then, the Short-Time Fourier Transform (STFT) is performed to calculate the marginal spectra. Finally, the presence of a continuous hunting problem is determined based on the results of the comparison calculations and diagnostic thresholds. To improve computational efficiency, permutation entropy (PE) is used as a fast indicator to identify turnouts with potential problems.FindingsUnder continuous hunting conditions, the PE is less than 0.90; the ratio of the maximum peak value of the signal component to the original signal peak value exceeded 0.7, and there is an energy band in the STFT time-frequency map, which corresponds to a frequency distribution range of 1–2 Hz.Originality/valueThe research results have revealed the lateral vibration characteristics of the high-speed train’s car body during continuous hunting when passing through turnouts. On this basis, an effective diagnostic method has been proposed. With a focus on practical engineering applications, a rapid screening index for identifying potential issues has been proposed, significantly enhancing the efficiency of diagnostic processes.

Predicting EV battery state of health using long short term degradation feature extraction and FEA TimeMixer

Accurately predicting the State of Health (SOH) of new energy vehicle batteries is critical for ensuring their reliable operation and extending battery’s service life. To address the issue of low SOH prediction accuracy across different prediction lengths, this paper proposes a prediction method based on long-short-term battery degradation feature extraction and FEA-TimeMixer model. First, a novel automatic SOH extraction algorithm for offline charging data is introduced to label the battery SOH degradation data. Then, the autoencoder is utilized to fuse the features of long-term and short-term SOH degradation trends extracted by empirical degradation models and Complete Ensemble Empirical Mode Decomposition with Adaptive Noise to improve the prediction accuracy over different prediction lengths. Finally, a Frequency Enhanced Attention (FEA) mechanism is introduced to improve the TimeMixer model, which integrates time-domain and frequency-domain information to overcome the limitations of the original model in capturing frequency-domain features. Experimental results show that the proposed method achieves a Mean Absolute Error of less than 0.0219 for short-term SOH predictions and less than 0.1007 for long-term SOH predictions, outperforming other deep learning models in prediction accuracy over multiple prediction lengths.

Prediction of FeO content in sintered ore based on ICEEMDAN and CNN-BiLSTM-AM

FeO content of sintered ore is an important reference index for measuring the performance of sintered ore. It significantly impacts the ironmaking process, iron quality, and energy consumption. Aiming at the current problem of delayed and poor accuracy of sintered ore FeO content detection results, this article proposes a hybrid network model that incorporates improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN), convolutional neural network (CNN), bidirectional long short-term memory (BiLSTM), and attention mechanism (AM) for FeO content prediction. First, the FeO content time series were decomposed using ICEEMDAN to obtain sub-layers with different frequencies. Then, the features with higher correlation with FeO content were selected by feature selection as model inputs, followed by predicting the decomposition sequences of FeO content using CNN-BiLSTM-AM for the feature-selected variables, respectively. Finally, all predicted sublayer predictions were reconstructed into the final prediction by summation. The proposed model effectively captures the essence of the sequence through the ICEEMDAN algorithm, extracts deep features from FeO content data using CNN, captures contextual information through BiLSTM, and enhances feature extraction capability with AM. The experimental results show that the collaboration of CNN, BiLSTM, and AM effectively enhances the modelling capability of the model and significantly improves the prediction accuracy. Additionally, the ICEEMDAN decomposition algorithm is employed to enhance the prediction performance further, offering advantages over other decomposition techniques. The MAE, MAPE, RMSE, RRMSE, and R² of the new ICEEMDAN-CNN-BiLSTM-AM (ICBA) model are 0.0751, 0.846%, 0.0937, 1.0500%, and 0.9646, respectively, demonstrating a significant improvement in prediction accuracy and outperforming the relevant comparison models.

A hybrid stock prediction method based on periodic/non-periodic features analyses

Stock investment is an economic activity characterized by high risks and high returns. Therefore, the prediction of stock prices or fluctuations is of great importance to investors. Stock price prediction is a challenging task due to the nonlinearity and high volatility of stock time series. Existing deep learning models may not capture the periodic and non-periodic features of stock data effectively. In this paper, we propose a novel model that leverages Complete Ensemble Empirical Mode Decomposition (CEEMD), Time2Vec, and Transformer to better capture and utilize various patterns in stock data for enhanced prediction performance, and we call it ETT. CEEMD decomposes the stock data into different frequency components based on their intrinsic scales. Time2Vec provides a time vector representation that captures both periodic and non-periodic patterns while being invariant to time scaling. Transformer learns the long-term dependencies and global information from the data. We apply ETT to predict stock prices in the Chinese A-share market and compare it with several baseline models. The results show that ETT reduces the mean squared error (MSE) by an average of 4% and increases the average cumulative return by 58% on the CSI 100 and Hushen 300 datasets.

Study on joint de-noising method of ECG signals with improved CEEMDAN-wavelet thresholding and singular spectrum analysis

The electrocardiogram (ECG) signal is often contaminated by various noises during acquisition and transmission, and the denoising effect directly affects the diagnosis of heart diseases. To improve the denoising effect, this paper proposes an improved Complete Ensemble Empirical Mode Decomposition (CEEMDAN)-wavelet thresholding and singular spectrum analysis (SSA) joint denoising method. First, the noise-dominated high-frequency intrinsic mode function (IMF) is identified using CEEMDAN; second, it is denoised to reduce high-frequency noise by the improved wavelet thresholding technique; and then, SSA is used to remove baseline wander (BW); finally, to verify the effectiveness of this denoising method, the root mean square error (RMSE) and signal-to-noise ratio (SNR) are used as the evaluation criteria for the denoising effect and the standard MIT-BIH ECG noise database signals are used for verification. The validation results show that compared with the existing ECG denoising methods, the denoising method improves the SNR by 39%∼ 55% and reduces the RMSE by 18%∼ 42%. The method can be used as an effective tool for ECG signal denoising and provide a better basis for diagnosis in computerized automated medical systems.

Research on short-term precipitation forecasting method based on CEEMDAN-GRU algorithm

Precipitation forecasting is vital for managing disasters, urban traffic, and agriculture. This study develops an improved model for short-term precipitation forecasting by combining Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Gated Recurrent Unit (GRU). Using precipitation data from January 1, 2019, to December 31, 2022, as a sample, the model capitalizes on CEEMDAN’s superior signal decomposition capabilities and GRU’s ability to capture nonlinear dynamic patterns in time series. To assess the model’s effectiveness, comparisons were conducted with 12 benchmark models, including CEEMDAN-LSTM, EMD-GRU, EMD-LSTM, BI-LSTM, GRU, LSTM, and TCN. The results demonstrate that the CEEMDAN-GRU model achieves higher accuracy and stability in short-term precipitation forecasting. Leveraging an Adam optimizer with adaptive learning rate reduction enhances convergence and ensures reliable predictions, achieving an R²of 0.7915, MAE of 0.05382, and MSE of 0.09081.

Forecasting Influenza Trends Using Decomposition Technique and LightGBM Optimized by Grey Wolf Optimizer Algorithm

Influenza is an acute respiratory infectious disease marked by its high contagiousness and rapid spread, caused by influenza viruses. Accurate influenza prediction is a critical issue in public health and serves as an essential tool for epidemiological studies. This paper seeks to improve the prediction accuracy of influenza-like illness (ILI) proportions by proposing a novel predictive model that integrates a data decomposition technique with the Grey Wolf Optimizer (GWO) algorithm, aiming to overcome the limitations of current prediction methods. Firstly, the most suitable indicators were selected using Spearman correlation coefficient. Secondly, a GWO-LightGBM model was established to obtain the residuals between the predicted and actual values. The residual sequence from the GWO-LightGBM model was then decomposed and corrected using the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) method, which led to the development of the GWO-LightGBM-CEEMDAN model. The incorporation of the Baidu Index was shown to enhance the precision of the proposed model’s predictions. The proposed model outperforms comparison models in terms of evaluation metrics such as RMSE and MAPE. Additionally, our study found that the revised Baidu Index indicators show a notable association with ILI trends.

Plant-Scale Biogas Production Based on Integrating of CEEMDAN Decomposition with PSO Optimized Multilayer Perceptron Neural Network

Accurate and dependable forecasting of biogas production is vital for optimizing process parameters and maintaining stable operation in large-scale anaerobic digestion projects. In this study, a novel hybrid approach (CEE-PMLP) integrating complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and a multilayer perceptron (MLP) neural network optimized by particle swarm optimization (PSO) were proposed for predicting biogas production in large-scale anaerobic digesters (ADs). The methodology involves extracting Intrinsic Mode Function (IMF) components using CEEMDAN and subsequently employing MLP optimized by particle swarm optimization (PSO) to predict each component. The performance of the models was evaluated using root mean square error (RMSE), mean squared error (MSE), mean absolute error (MAE), and fitting determination coefficient (R2). The findings revealed that the prediction errors of the proposed CEE-PMLP model were consistently lower than those of other comparative models. Notably, the model achieved the highest R2 value of 98%, indicating an exceptionally high accuracy in prediction. The validation experiment confirmed the high accuracy of the CEE-PMLP model, further demonstrating its superiority in biogas production prediction.

Photovoltaic Short-Term Output Power Forecast Model Based on Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise–Kernel Principal Component Analysis–Long Short-Term Memory

To solve the problem of photovoltaic power prediction in areas with large climate changes, this article proposes a hybrid Long Short-Term Memory method to improve the prediction accuracy and noise resistance. It combines the improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) and kernel principal component analysis (KPCA) algorithm. The ICEEMDAN algorithm reduces the instability of the environmental factor sequence. The KPCA algorithm reduces the input dimensions of the model. LSTM performs dynamic time modeling of the multivariate feature sequences to predict the output PV power. The adaptability of the ICEEMDAN-KPCA-LSTM model is assessed with datasets from a PV plant in west China and evaluated by root mean squared error (RMSE), mean absolute percentage error (MAPE), and R-squared metrics. Using 70% of the datasets for output PV power estimation, the results show a good performance, with an RMSE of 4.3715, MAPE of 8.9264%, and R-squared value of 89.973%. By comparing with other prediction models, the ICEEMDAN-KPCA-LSTM photovoltaic output power model outperforms other models.

TFR: A Temporal Feature‐Refined Multi‐Stage Carbon Price Forecasting

ABSTRACTAccurate carbon price forecasting is crucial for effective carbon market analysis and decision‐making. We propose a novel Temporal Feature‐Refined (TFR) model to address the challenges of complex dependencies and high noise levels in carbon price time series data. The TFR model integrates Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) for signal decomposition, an Autoencoder for feature optimization, and a Temporal Convolutional Network (TCN) for capturing long‐range temporal dependencies. It incorporates both traditional economic factors and unconventional determinants such as air quality, policy uncertainty, and public sentiment. Experiments on the Shanghai carbon trading market demonstrate that the TFR model significantly outperforms existing methods, achieving an 83.96% improvement in MAE over Support Vector Regression (SVR) and up to a 65.56% improvement over Long Short‐Term Memory (LSTM) networks. Further analyses, including comparisons with different decomposition models and ablation studies, confirm the effectiveness of each component and the overall model.

Identification Method for Railway Rail Corrugation Utilizing CEEMDAN-PE-SPWVD.

Rail corrugation intensifies wheel-rail vibrations, often leading to damage in vehicle-track system components within affected sections. This paper proposes a novel method for identifying rail corrugation, which combines Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), permutation entropy (PE), and Smoothed Pseudo Wigner-Ville Distribution (SPWVD). Initially, vertical acceleration data from the axle box are decomposed using CEEMDAN to extract intrinsic mode functions (IMFs) with distinct frequencies. PE is used to evaluate the randomness of each IMF component, discarding those with high permutation entropy values. Subsequently, correlation analysis is performed on the retained IMFs to identify the component most strongly correlated with the original signal. The selected component is subjected to SPWVD time-frequency analysis to identify the location and wavelength of the corrugation occurrence. Filtering is applied to the IMF based on the frequency concentration observed in the time-frequency analysis results. Then, frequency-domain integration is performed to estimate the rail's corrugation depth. Finally, the algorithm is validated and analyzed using both simulated data and measured data. Validation results show that this approach reliably identifies the wavelength and depth characteristics of rail corrugation. Additionally, the time-frequency analysis results reveal variations in the severity of corrugation damage at different locations.

Charging Strategies for Electric Vehicles Using a Machine Learning Load Forecasting Approach for Residential Buildings in Canada

The global electric vehicle (EV) market is experiencing exponential growth, driven by technological advancements, environmental awareness, and government incentives. As EV adoption accelerates, it introduces opportunities and challenges for power systems worldwide due to the large battery capacity, uncertain charging behaviors of EV users, and seasonal variations. This could result in significant peak–valley differences in load in featured time slots, particularly during winter periods when EVs’ heating systems use increases. This paper proposes three future charging strategies, namely the Overnight, Workplace/Other Charging Sites, and Overnight Workplace/Other Charging Sites, to reduce overall charging in peak periods. The charging strategies are based on predicted load utilizing a hybrid machine learning (ML) approach to reduce overall charging in peak periods. The hybrid ML method combines similar day selection, complete ensemble empirical mode decomposition with adaptive noise, and deep neural networks. The dataset utilized in this study was gathered from 1000 EVs across nine provinces in Canada between 2017 and 2019, encompassing charging loads for thirty-five vehicle models, and charging locations and levels. The analysis revealed that the aggregated charging power of EV fleets aligns and overlaps with the peak periods of residential buildings energy consumption. The proposed Overnight Workplace/Other Charging Sites strategy can significantly reduce the Peak-to-Average Ratio (PAR) and energy cost during the day by leveraging predictions made three days in advance. It showed that the PAR values were approximately half those on the predicted load profile (50% and 51%), while charging costs were reduced by 54% and 56% in spring and winter, respectively. The proposed strategies can be implemented using incentive programs to motivate EV owners to charge in the workplace and at home during off-peak times.

MIVNDN: Ultra-Short-Term Wind Power Prediction Method with MSDBO-ICEEMDAN-VMD-Nons-DCTransformer Net

Wind speed, wind direction, humidity, temperature, altitude, and other factors affect wind power generation, and the uncertainty and instability of the above factors bring challenges to the regulation and control of wind power generation, which requires flexible management and scheduling strategies. Therefore, it is crucial to improve the accuracy of ultra-short-term wind power prediction. To solve this problem, this paper proposes an ultra-short-term wind power prediction method with MIVNDN. Firstly, the Spearman’s and Kendall’s correlation coefficients are integrated to select the appropriate features. Secondly, the multi-strategy dung beetle optimization algorithm (MSDBO) is used to optimize the parameter combinations in the improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) method, and the optimized decomposition method is used to decompose the historical wind power sequence to obtain a series of intrinsic modal function (IMF) components with different frequency ranges. Then, the high-frequency band IMF components and low-frequency band IMF components are reconstructed using the t-mean test and sample entropy, and the reconstructed high-frequency IMF component is decomposed quadratically using the variational modal decomposition (VMD) to obtain a new set of IMF components. Finally, the Nons-Transformer model is improved by adding dilated causal convolution to its encoder, and the new set of IMF components, as well as the unreconstructed mid-frequency band IMF components and the reconstructed low-frequency IMF, component are used as inputs to the model to obtain the prediction results and perform error analysis. The experimental results show that our proposed model outperforms other single and combined models.