EEG Feature Learning Model Based on Intrinsic Time-Scale Decomposition and Adaptive Huber Loss

According to the World Health Organization,about 50 million people worldwide suffer from epilepsy.The detection and treatment of epilepsy face great challenges.Electroencephalogram(EEG)is a significant research object widely used in diagnosis and treatment of epilepsy.In this paper,an adaptive feature learning model for EEG signals is proposed,which combines Huber loss function with adaptive weight penalty term.Firstly,each EEG signal is decomposed by intrinsic time-scale decomposition.Secondly,the statistical index values are calculated from the instantaneous amplitude and frequency of every component and fed into the proposed model.Finally,the discriminative features learned by the proposed model are used to detect seizures.Our main innovation is to consider a highly flexible penalization based on Huber loss function,which can set different weights according to the influence of different features on epilepsy detection.Besides,the new model can be solved by proximal alternating direction multiplier method,which can effectively ensure the convergence of the algorithm.The performance of the proposed method is evaluated on three public EEG datasets provided by the Bonn University,Childrens Hospital Boston-Massachusetts Institute of Technology,and Neurological and Sleep Center at Hauz Khas,New Delhi(New Delhi Epilepsy data).The recognition accuracy on these two datasets is 98%and 99.05%,respectively,indicating the application value of the new model.

基于Intrinsic Time-scale Decomposition算法的扰动信号特征量提取问题研究

由于电能质量扰动信号具有非线性的特点,当下算法无法有效提取电能质量扰动信号,采取Intrinsic Timescale Decomposition算法,可以有效提取电能质量扰动信号的频率、幅值、相位等特征量。利用Intrinsic Time-scale Decomposition算法提取电能质量扰动信号的Proper Rotation Component分量,后对Proper Rotation Component分量进行Hilbert变换以求得电能质量扰动信号的相位和瞬时频率,再根据电能质量扰动信号的高频突变点得到电能质量扰动信号的起止时间;根据envelope函数求得衰减因子。利用软件开发程序MATHEMATICA对电能质量扰动信号进行分析,结果表明,Intrinsic Time-scale Decomposition算法可以很好地提取电能质量扰动信号特征量,并且具有极佳的抗噪性能。

Real-time Subsynchronous Control Interaction Monitoring Using Improved Intrinsic Time-scale Decomposition

In recent years,subsynchronous control interaction(SSCI)has frequently taken place in renewable-connected power systems.To counter this issue,utilities have been seeking tools for fast and accurate identification of SSCI events.The main challenges of SSCI monitoring are the time-varying nature and uncertain modes of SSCI events.Accordingly,this paper presents a simple but effective method that takes advantage of intrinsic time-scale decomposition(ITD).The main purpose is to improve the accuracy and robustness of ITD by incorporating the least-squares method.Results show that the proposed method strikes a good balance between dynamic performance and estimation accuracy.More importantly,the method does not require any prior information,and its performance is therefore not affected by the frequency constitution of the SSCI.Comprehensive comparative studies are conducted to demonstrate the usefulness of the method through synthetic signals,electromagnetic temporary program(EMTP)simulations,and field-recorded SSCI data.Finally,real-time simulation tests are conducted to show the feasibility of the method for real-time monitoring.

Application of complete ensemble intrinsic time-scale decomposition and least-square SVM optimized using hybrid DE and PSO to fault diagnosis of diesel engines

Targeting the mode-mixing problem of intrinsic time-scale decomposition （ITD） and the parameter optimization problem of least-square support vector machine （LSSVM）, we propose a novel approach based on complete ensemble intrinsic time-scale decomposition （CEITD） and LSSVM optimized by the hybrid differential evolution and particle swarm optimization （HDEPSO） algorithm for the identification of the fault in a diesel engine. The approach consists mainly of three stages. First, to solve the mode-mixing problem of ITD, a novel CEITD method is proposed. Then the CEITD method is used to decompose the nonstationary vibration signal into a set of stationary proper rotation components （PRCs） and a residual signal. Second, three typical types of time-frequency features, namely singular values, PRCs energy and energy entropy, and AR model parameters, are extracted from the first several PRCs and used as the fault feature vectors. Finally, a HDEPSO algorithm is proposed for the parameter optimization of LSSVM, and the fault diagnosis results can be obtained by inputting the fault feature vectors into the HDEPSO-LSSVM classifier. Simulation and experimental results demonstrate that the proposed fault diagnosis approach can overcome the mode-mixing problem of ITD and accurately identify the fault patterns of diesel engines.

Modal identification of multi-degree-of-freedom structures based on intrinsic chirp component decomposition method

Modal parameter identification is a mature technology.However,there are some challenges in its practical applications such as the identification of vibration systems involving closely spaced modes and intensive noise contamination.This paper proposes a new time-frequency method based on intrinsic chirp component decomposition(ICCD)to address these issues.In this method,a redundant Fourier model is used to ameliorate border distortions and improve the accuracy of signal reconstruction.The effectiveness and accuracy of the proposed method are illustrated using three examples:a cantilever beam structure with intensive noise contamination or environmental interference,a four-degree-of-freedom structure with two closely spaced modes,and an impact test on a cantilever rectangular plate.By comparison with the identification method based on the empirical wavelet transform(EWT),it is shown that the presented method is effective,even in a high-noise environment,and the dynamic characteristics of closely spaced modes are accurately determined.

Non-saturation Throughput of S-ALOHA Using the Time-Scale Decomposition Technique

S-ALOHA （Slotted ALOHA） random access protocol is a widely used protocol mainly for the transmission of short packets in wireless networks. Most papers consider either an infinite population model where the impact of the backoff protocol cannot be adequately evaluated or a finite population model where the number of nodes is fixed. In this letter, a combination of both models is proposed using the time-scale decomposition technique. This methodology allows to study the system under more realistic conditions where the dynamics of users enter and leaving the system are reflected on the performance of the system as well as the impact of the backoff protocol. Also, it allows studying the system in non-saturation conditions. The proposed methodology divides the analysis in two parts： packet-level and connection-level. This analysis renders suitable results when the time scale of the packet level and connection level statistics is different. On the other hand, when these scales are similar, the proposed methodology is no longer suited.

Adaptive Variational Mode Decomposition for Bearing Fault Detection

Rolling element bearings are commonly used in rotary mechanical and electrical equipment. According to investigation, more than half of rotating machinery defects are related to bearing faults. However, reliable bearing fault detection still remains a challenging task, especially in industrial applications. The objective of this work is to propose an adaptive variational mode decomposition (AVMD) technique for non-stationary signal analysis and bearing fault detection. The AVMD includes several steps in processing: 1) Signal characteristics are analyzed to determine the signal center frequency and the related parameters. 2) The ensemble-kurtosis index is suggested to decompose the target signal and select the most representative intrinsic mode functions (IMFs). 3) The envelope spectrum analysis is performed using the selected IMFs to identify the characteristic features for bearing fault detection. The effectiveness of the proposed AVMD technique is examined by experimental tests under different bearing conditions, with the comparison of other related bearing fault techniques.

Random noise attenuation by f–x spatial projection-based complex empirical mode decomposition predictive filtering

The frequency–space（f–x） empirical mode decomposition（EMD） denoising method has two limitations when applied to nonstationary seismic data. First, subtracting the first intrinsic mode function（IMF） results in signal damage and limited denoising. Second, decomposing the real and imaginary parts of complex data may lead to inconsistent decomposition numbers. Thus, we propose a new method named f–x spatial projection-based complex empirical mode decomposition（CEMD） prediction filtering. The proposed approach directly decomposes complex seismic data into a series of complex IMFs（CIMFs） using the spatial projection-based CEMD algorithm and then applies f–x predictive filtering to the stationary CIMFs to improve the signal-to-noise ratio. Synthetic and real data examples were used to demonstrate the performance of the new method in random noise attenuation and seismic signal preservation.

Regional features of the temperature trend in China based on Empirical Mode Decomposition

By the Empirical Mode Decomposition method, we analyzed the observed monthly average temperature in more than 700 stations from 1951-2001 over China. Simultaneously, the temperature variability of each station is calculated by this method, and classification chart of long term trend and temperature variability distributing chart of China are obtained, supported by GIS, 1 kmxl km resolution. The results show that： in recent 50 years, the temperature has increased by more than 0.4~C/10a in most parts of northern China, while in Southwest China and the middle and lower Yangtze Valley, the increase is not significant. The areas with a negative temperature change rate are distributed sporadically in Southwest China. Meanwhile, the temperature data from 1881 to 2001 in nine study regions in China are also analyzed, indicating that in the past 100 years, the temperature has been increasing all the way in Northeast China, North China, South China, Northwest China and Xinjiang and declining in Southwest China. An inverse ‘V-shaped’ trend is also found in Central China. But in Tibet the change is less significant.

Improvement of the Mirror Extending in Empirical Mode Decomposition Method and the Technology for Eliminating Frequency Mixing

The mirror extending approach proposed by Zhao and Huang in EMD method is improved in this paper. Mirror extending manner of data is kept unchanged, but the approach for determining envelopes is changed. When the end of data is obviously not extremum, the envelope is determined by the first inner extremum and the image value in the mirror, ignoring the value on the end. This improvement eliminates the frequency compression near the end and decreases the error. Meanwhile, tridiagonal equations are used and the calculation speed is much increased. The temporal process curve is more important in reflecting the real physical process and comparable with other phenomena. Frequency mixing in IMFs makes it impossible. A high frequency reconstruction (HFR) approach is proposed to eliminate common frequency mixing and reconstruct an IMF with all high frequency portions. By this approach, the IMFs without frequency mixing are obtained to express significative processes. The high frequency information restored in high frequency IMF can be extracted by general spectrum method. After obtaining IMFs by EMD method, some of the theoretical and technological issues still exist when using the IMFs. The consistency of IMFs with real physical process is discussed in detail. By virtue of the approach proposed in this paper, the EMD method can be widely used in various fields.

Pressure fluctuation signal analysis of pump based on ensemble empirical mode decomposition method

Pressure fluctuations, which are inevitable in the operation of pumps, have a strong non-stationary characteristic and contain a great deal of important information representing the operation conditions. With an axial-flow pump as an example, a new method for time-frequency analysis based on the ensemble empirical mode decomposition （EEMD） method is proposed for research on the characteristics of pressure fluctuations. First, the pressure fluctuation signals are preprocessed with the empirical mode decomposition （EMD） method, and intrinsic mode functions （IMFs） are extracted. Second, the EEMD method is used to extract more precise decomposition results, and the number of iterations is determined according to the number of IMFs produced by the EMD method. Third, correlation coefficients between IMFs produced by the EMD and EEMD methods and the original signal are calculated, and the most sensitive IMFs are chosen to analyze the frequency spectrum. Finally, the operation conditions of the pump are identified with the frequency features. The results show that, compared with the EMD method, the EEMD method can improve the time-frequency resolution and extract main vibration components from pressure fluctuation signals.

GEARBOX FAULTDIAGNOSIS BASED ON EMPIRICAL MODE DECOMPOSITION

Time synchronous averaging of vibration data is a fundament technique forgearbox diagnosis. Currently, this technique relies on hardware tachometer to give phase synchronousinformation. Empirical mode decomposition (EMD) is introduced to replace time synchronous averagingof gearbox vibration signal. With it, any complicated dataset can be decomposed into a finite andoften small number of intrinsic mode functions (IMF). The key problem is how to assure thatvibration signals deduced by gear defects could be sifted out by EMD. The characteristic vibrationsignals of gear defects are proved IMFs, which makes it possible to utilize EMD for the diagnosis ofgearbox faults. The method is validated by data from recordings of the vibration of a single-stagespiral bevel gearbox with fatigue pitting. The results show EMD is powerful to extractcharacteristic information from noisy vibration signals.

A method for extracting human gait series from accelerometer signals based on the ensemble empirical mode decomposition

In this paper, the ensemble empirical mode decomposition （EEMD） is applied to analyse accelerometer signals collected during normal human walking. First, the self-adaptive feature of EEMD is utilised to decompose the ac- celerometer signals, thus sifting out several intrinsic mode functions （IMFs） at disparate scales. Then, gait series can be extracted through peak detection from the eigen IMF that best represents gait rhythmicity. Compared with the method based on the empirical mode decomposition （EMD）, the EEMD-based method has the following advantages： it remarkably improves the detection rate of peak values hidden in the original accelerometer signal, even when the signal is severely contaminated by the intermittent noises; this method effectively prevents the phenomenon of mode mixing found in the process of EMD. And a reasonable selection of parameters for the stop-filtering criteria can improve the calculation speed of the EEMD-based method. Meanwhile, the endpoint effect can be suppressed by using the auto regressive and moving average model to extend a short-time series in dual directions. The results suggest that EEMD is a powerful tool for extraction of gait rhythmicity and it also provides valuable clues for extracting eigen rhythm of other physiological signals.

Anchoring Bolt Detection Based on Morphological Filtering and Variational Modal Decomposition

The pull test is a damaging detection method that fails to measure the actual length of a bolt.Thus,the ultrasonic echo is an important non?destructive testing method for bolt quality detection.In this research,the variational modal decomposition(VMD)method is introduced into the bolt detection signal analysis.On the basis of morphological filtering(MF)and the VMD method,a VMD?combined MF principle is established into a bolt detection signal analysis method(MF?VMD).MF?VMD is used to analyze the vibration and actual bolt detection signals of the simulation.Results show that MF?VMD effectively separates intrinsic mode function,even under strong interference.In comparison with conventional VMD method,the proposed method can remove noise interference.An intrinsic mode function of the field detection signal can be effectively identified by reflecting the signal at the bottom of the bolt.

Multi-scale spatial relationships between soil total nitrogen and influencing factors in a basin landscape based on multivariate empirical mode decomposition

The relationships between soil total nitrogen(STN)and influencing factors are scale-dependent.The objective of this study was to identify the multi-scale spatial relationships of STN with selected environmental factors(elevation,slope and topographic wetness index),intrinsic soil factors(soil bulk density,sand content,silt content,and clay content)and combined environmental factors(including the first two principal components(PC1 and PC2)of the Vis-NIR soil spectra)along three sampling transects located at the upstream,midstream and downstream of Taiyuan Basin on the Chinese Loess Plateau.We separated the multivariate data series of STN and influencing factors at each transect into six intrinsic mode functions(IMFs)and one residue by multivariate empirical mode decomposition(MEMD).Meanwhile,we obtained the predicted equations of STN based on MEMD by stepwise multiple linear regression(SMLR).The results indicated that the dominant scales of explained variance in STN were at scale 995 m for transect 1,at scales 956 and 8852 m for transect 2,and at scales 972,5716 and 12,317 m for transect 3.Multi-scale correlation coefficients between STN and influencing factors were less significant in transect 3 than in transects 1 and 2.The goodness of fit root mean square error(RMSE),normalized root mean square error(NRMSE),and coefficient of determination(R2)indicated that the prediction of STN at the sampling scale by summing all of the predicted IMFs and residue was more accurate than that by SMLR directly.Therefore,the multi-scale method of MEMD has a good potential in characterizing the multi-scale spatial relationships between STN and influencing factors at the basin landscape scale.

Segmented second algorithm of empirical mode decomposition

A new algorithm, named segmented second empirical mode decomposition （EMD） algorithm, is proposed in this paper in order to reduce the computing time of EMD and make EMD algorithm available to online time-frequency analysis. The original data is divided into some segments with the same length. Each segment data is processed based on the principle of the first-level EMD decomposition. The algorithm is compared with the traditional EMD and results show that it is more useful and effective for analyzing nonlinear and non-stationary signals.

Weighing axle weight of moving vehicle based on empirical mode decomposition

Dynanfic forces are the main factor that influences the axle weight measurement accuracy of moving vehicle. Empirical mode decomposition （EMD） is presented to separate the dynamic forces contained in the axle weight signal. The concept and algorithm of EMD are introduced. The characteristic of the axle weight signal is analyzed. The method of judging pseudo intrinsic mode function （pseudo-IMF） is presented to improve the weighing accuracy. Numerical simulation and field experiments are conducted to evaluate the performance of EMD. The result shows effectiveness of the proposed method. Maximum weighing errors of the front axle, the rear axle and the gross weight at the speed of 15 km/h or lower are 2.22%, 6.26% and 4.11% respectively.

Signal prediction based on empirical mode decomposition and artificial neural networks

In view of the usefulness of Empirical Mode Decomposition （EMD）, Artificial Neural Networks （ ANN）, and Most Relevant Matching Extension （MRME） methods in dealing with nonlinear signals, we pro- pose a new way of combining these methods to deal with signal prediction. We found the results of combining EMD with either ANN or MRME to have higher prediction precision for a time series than the result of using EMD alone.

Discriminative feature encoding for intrinsic image decomposition

Intrinsic image decomposition is an important and long-standing computer vision problem.Given an input image,recovering the physical scene properties is ill-posed.Several physically motivated priors have been used to restrict the solution space of the optimization problem for intrinsic image decomposition.This work takes advantage of deep learning,and shows that it can solve this challenging computer vision problem with high efficiency.The focus lies in the feature encoding phase to extract discriminative features for different intrinsic layers from an input image.To achieve this goal,we explore the distinctive characteristics of different intrinsic components in the high-dimensional feature embedding space.We define feature distribution divergence to efficiently separate the feature vectors of different intrinsic components.The feature distributions are also constrained to fit the real ones through a feature distribution consistency.In addition,a data refinement approach is provided to remove data inconsistency from the Sintel dataset,making it more suitable for intrinsic image decomposition.Our method is also extended to intrinsic video decomposition based on pixel-wise correspondences between adjacent frames.Experimental results indicate that our proposed network structure can outperform the existing state-of-the-art.

混凝土缺陷信号变分模态分解与超声成像方法

混凝土的强散射特性限制了其中缺陷声波成像的分辨率。该文采用一种依据变分模态分解与全聚焦成像相结合的方法,将接收信号分解成多个本征模态函数,计算各本征模态函数与激励信号的相关系数,对信号加权重构以实现对特征信号的提取,从而提高成像算法对混凝土缺陷间散射波互干扰的鲁棒性。通过设置对比试验,研究了不同缺陷混凝土结构中该信号处理方式对于成像结果的影响。试验结果表明,该方法对于弱散射及散射干扰具有更好的鲁棒性,相比基于原始数据的成像方法能够更好地还原混凝土内部结构。