60 results about "Correlation entropy" patented technology

The invention discloses a mesoscale vortex trajectory stationary sequence extraction and recurrent neural network prediction method. The method comprises the following steps: collecting related data and carrying out mathematical statistics; introducing variational mode decomposition applying correlation entropy, searching optimal parameters, and decomposing non-stationary mesoscale vortex propagation trajectory data into K stationary subsequences; establishing an improved multi-step prediction network model based on a double-stage attention recurrent neural network; constructing an improved regularization strategy training model; sequentially sending the decomposed stationary sub-sequence and the multi-feature variable sequence into a multi-step prediction network model, training the model through an improved regularization strategy, respectively predicting the sub-sequences, and finally obtaining a target prediction result. According to the method, prediction research of mesoscale vortex trajectory data is assisted through a satellite height measurement observation technology from the perspective of machine learning, and accurate prediction of mesoscale vortexes has important scientific and application values for understanding propagation and evolution characteristics of the mesoscale vortexes and improving the simulation capability of climate changes.

The invention discloses a variable step size SA self-adaptive filtering algorithm based on a correlation entropy. According to the provided algorithm, the complexity is relatively low, the rapid convergence can be realized and the relatively high precision can be achieved, so the algorithm is relatively easy to popularize and use in practical application. The problem that a traditional SA algorithm is difficult to popularize and use in the practical application due to the fact that demands for high precision and rapid convergence are difficult to satisfy is solved.

The invention discloses a load flow Jacobian matrix robust estimation method with the consideration to sparsity for an intelligent power distribution network, and the method comprises the following steps: 1) acquiring the number of nodes of the distribution network and numbering the nodes; 2) obtaining the measurement data of a synchronous phasor measurement device; 3) constructing a sensing matrix, and setting the line number of the load flow Jacobian matrix as m=1; 4) solving a least squares solution as an estimated solution; 5) switching to step 8) if an iteration termination condition is established, or else switching to step 6); 6) solving a minimized correlation entropy model as an estimated solution; 7) switching to step 8) if the iteration termination condition is established, or else updating the column number index set of the sensing matrix, and returning to step 6); 8) determining whether to complete the estimation of all rows of the Jacobian matrix or not: outputting an estimation result if the estimation of all rows of the Jacobian matrix is completed or else supposing m=m+1, and returning to step 4). The method effectively avoids the adverse impact on an estimation result from bad data in measurement while employing the sparsity of the load flow Jacobian matrix, and still can guarantee the estimation precision when the measurement data contains bad data.

The invention provides a hyperspectral image classification method and a system based on the correlation entropy principle. Firstly, the input image is preprocessed so that the element values of all spectral vectors are between 0 and 1; secondly, a small number of samples are selected as training samples, and the spatial features of hyperspectral images are extracted by using the designed hierarchical model of fusion dimension reduction method and correlation entropy principle. Then, the spatial spectrum characteristics of the test samples are learned by using the trained hierarchical structure. Finally, the characteristics of the test samples are input into the KNN classifier to get the class label. The invention fully combines the spectral and spatial characteristics of the hyperspectraldata by using the correlation entropy principle; by using the hierarchical model of the invention, more abstract spatial spectrum features can be obtained; due to the characteristics of LDA, the invention has small sample complexity, and only a small number of training samples are needed to obtain a better classification result, so the invention is more conducive to practical application.

The invention relates to a joint estimation method for the angle of an array antenna and the number of signal sources in a complex noise environment, and belongs to the technical field of radio positioning. The joint estimation method comprises the steps of 1) solving a cycle correlation entropy matrix V<alpha><y>(tau) of array signals under a condition that the cycle frequency is known; 2) building a cycle correlation entropy array linear prediction model V=[Phi]A which is applicable to broadband signals and narrow-band signals; 3) estimating the number K of interested signal sources and an error variance [sigma]<2>; 4) estimating a flow pattern matrix of the array model; and 5) performing DOA estimation by using spectrum peak searching. According to the invention, a cycle correlation entropy theory is organically applied to array signal processing, correlation characteristics of the cycle correlation entropy are innovatively proposed, and the array linear prediction model is built based on the characteristics. The algorithm is put forward according to actual requirements, and has the characteristics of high anti-noise performance, low computation complexity, small number of required snapshots, high angular resolution and the like.

The invention discloses a method for predicting the performance degradation trend of an electronic device. A prediction model of an initial correlation entropy extreme learning machine is establishedoffline; the dynamic correlation entropy extreme learning machine is updated on line, a codebook mechanism combined with singular values is constructed, and then the singular values in historical dataare identified, so that the influence of noise and the singular values in the data on a prediction model is overcome, and the final prediction effect of the prediction model is improved.

The invention discloses a general integrated mixed pixel demixing framework model. A model solving method is provided. A hyperspectral image is divided into a plurality of sub-blocks; a proper overlapped area is reserved in adjacent areas of the sub-block; therefore, the problems of end member variation and non-linear non-uniform distribution are solved; the problem of abnormal noise is solved through sparse constraint reconstruction errors, the three problems of nonlinearity, end member variation and abnormal points are integrated into the same unmixing framework to be solved through a correlation entropy induction measurement criterion, and the unmixing performance of the proposed model mixed pixels is quantitatively evaluated through a Lagrangian function method and a KKT condition. Theinfluence of non-linearity, end member variation bars and abnormal points on hyperspectral image mixed pixel unmixing is overcome by constructing a universal integrated mixed pixel unmixing frameworkmodel. According to the characteristics of nonlinearity and end member variation heterogeneous distribution, and abnormal points including abnormal end members and abnormal noise, the model effectively simulates Gaussian mixture distribution, abandons sparse distribution of abnormal noise, and improves the unmixing performance of mixed pixels. And theoretical derivation is carried out on the proposed model, so that the feasibility and superiority of mixed pixel demixing based on a general integrated framework are proved.

The invention discloses a maneuvering target robust tracking method under a generalized correlation entropy criterion, and provides a robust framed solution for solving the tracking problem of a maneuvering target with sensor observation interference. According to the method, an interactive multi-model method serves as a basic framework, in the mixing step, a cost function with the target state filtering estimated value of a sub-model at the last moment as an independent variable is established with the maximized generalized correlation entropy as the criterion, and the mixed target state estimated value is found by optimizing the cost function; and in the fusion step, a cost function taking the target state filtering estimation value of the sub-model at the current moment as an independent variable is established by taking the maximized generalized correlation entropy as a criterion, estimation fusion of the model is achieved by optimizing the cost function, and a robust estimation value of the target state at the current moment is finally obtained.

The invention provides a coherent distributed non-circular signal angle parameter combined estimation method based on generalized complex correlation entropy, and belongs to the distributed source wireless positioning technical field in a multi-path environment; the method uses non-circular signal property to expand an array output matrix, and uses the generalized complex correlation entropy suppress noise interferences by pulse noises on a received signal. The method processes the expanded array output matrix, extracts an expanded generalized complex correlation entropy matrix and a steeringvector matrix, and uses a DSPE algorithm mind to extract angle parameter information contained in the steering vector matrix, thus estimating the center DOA and a diffusion angle. According to tests,the method is excellent in performance, and can accurately estimate the angle parameters of the distributed source signal under the pulse noise interference environment.

The invention belongs to the technical field of signal processing, and relates to a radio wave arrival direction estimation method under the condition of no direct signal, and especially can be used for estimating the wave arrival direction of a target under the coexistence of multipath, shielding and impulse noise interference in a radio propagation process with high precision. On the basis of asubspace theory, a distributed information source model is established based on a DSPE algorithm of generalized self-correlation entropy, the generalized self-correlation entropy is defined and expanded to the DSPE algorithm, for calculating a covariance matrix of the generalized correlation entropy of a linear array antenna receiving signal; and a two-dimensional spatial spectrum is built throughsingular value decomposition, and the direction of wave arrival of the distributed information source is estimated through spectral peak searching. The method disclosed by the invention has high robustness and is also suitable for gaussian noise.

An adaptive echo cancellation method based on correlation entropy is disclosed. The method comprises the following steps: A, collecting far-end signals, sampling far-end signals transmitted from far-end to obtain the discrete value x (n) of far-end input signals at the current time n, and the filter input signal vector is x (n) = [x (n), x (n-1),..., x (n-L+1)] T; B, sampling far-end signals transmitted from far-end to obtain the discrete value x (n) of far-end input signals at the current time n. B, estimate an echo signal, passing an input signal vector x (n) of that current time n through an adaptive filt, and outputting a value y (n) thereof, that is, an estimated value of the echo signal; C, canceling the echo, sampling with a near-end microphone to obtain the near-end signal d (n) ofthe current time n of the echo return, subtracting the estimated value y (n) of the echo signal; D, updating the tap weight coefficients of the filter, and calculating a tap weight vector w (n+1), w(n+1) = w (n) + Mu(n) U (n) (UT (n) U (n)) of n+1 at the next time of the filter; 1E(n)-C (n); E, let n=n+1, repeating the process of A, B, C, D, E until the end of the call. This method can obtain faster convergence rate and lower steady-state error, and the echo cancellation effect is good.

The invention discloses a blind adaptive multi-user detection method based on a generalized maximum correlation entropy criterion. The method comprises the following steps: S100, modeling a satelliteCDMA (Code Division Multiple Access) communication system, and connecting a plurality of ground users with satellites covering the areas of the ground users through an uplink of satellite communication; expressing a discrete-time dynamic state space multi-user detector as a combination of a process equation and an observation equation; S200, using the generalized maximum correlation entropy of thedifference between the observed value and the estimated value as a loss function; S300, according to a loss function of a generalized maximum correlation entropy criterion and partial derivatives relative to detector parameters, enabling the loss function to be equal to zero, and deriving an expression of the blind self-adaptive multi-user detector; and S400, solving an expression of the blind self-adaptive multi-user detector by using an immobile point iteration method, and converging an expression algorithm, thereby obtaining a solution of the blind self-adaptive multi-user detector. Compared with traditional blind self-adaptive multi-user detection, the method has better convergence performance and can have a better bit error rate under impact noise.

The invention discloses a classification method based on correlation entropy and transfer learning, in the Machine learning technology field, The method is used for solving the problems that in the prior art, dilution assumption needs to be conducted on unpredictable noise, and samples marked by a target data set are insufficient. Label information of a source data set is fully utilized, knowledgein the source data set is robustly migrated to a target data set in a severe noise environment through joint learning of the source data set and the target data set, and therefore the classificationrecognition effect on the target data set is remarkably improved.

The invention discloses a performance degradation trend prediction method based on a collaborative derivation related entropy extreme learning machine. The method comprises the following steps: firstly, calculating hidden layer output of an input sample based on an extreme learning machine of collaborative derivation correlation entropy; and corresponding prediction error, solving an optimal correlation entropy variance and an influence weight through a collaborative derivation algorithm; performing update iteration, until a global optimal solution [sigma]gbest in the particle swarm is found;q2 and the corresponding influence weight serving as the optimal correlation entropy variance and the influence weight, finally, under the condition that calculation convergence of the extreme learning machine is met, outputting a prediction value of the input sample, and therefore the performance degradation trend of the input sample is obtained, and the method has the advantages of being high inprediction precision, high in robustness and the like.

The invention discloses a power distribution network voltage power sensitivity robust estimation method based on model equivalence. The power distribution network voltage power sensitivity robust estimation method based on the model equivalence comprises the steps that firstly, a power distribution network equivalent model is established; data is measured by using synchronous phasor, and equivalent model parameters are estimated by using a maximized correlation entropy kalman filter algorithm; a power distribution network is simplified according to the obtained equivalent model parameters; andby calculating a power flow jacobian matrix, and then the jacobian matrix is inverted to obtain the voltage power sensitivity. The equivalent network model can ensure the consistency of the voltage power sensitivity of each node before and after the equivalence, a proposed equivalent model parameter estimation algorithm can ensure the accuracy of the estimation under the condition that the measurement data contains noise and bad data, and the robust estimation calculation of the voltage power sensitivity is realized.

The invention discloses a chip performance degradation trend prediction method based on a multi-step robust prediction learning machine, which combines an extreme learning machine and a recurrent neural network, has extremely high information fusion capability and rapid information processing capability, constructs similarity based on correlation entropy by establishing an error code book, establishes real-time prediction model updating according to chip degradation diversity and dynamics, and overcomes the influence of interference on a prediction result. Therefore, compared with the existing method, the method has higher online prediction precision, and the multi-step prediction result is more accurate compared with the existing method.

The invention discloses a weak feature extraction method based on combination of singular value decomposition and a correlation entropy theory. The method comprises a singular value decomposition module, a correlation entropy induction measure calculation module, a signal reconstruction module and a cyclic correlation entropy spectrum estimation module. The method comprises the following steps: firstly, performing two-dimensional conversion on signals to obtain a Hankel matrix, and performing singular value decomposition on the Hankel matrix formed by the signals to obtain singular values and corresponding singular components; then calculating a correlation entropy induction measurement value of the original signal and each singular component, determining a threshold value according to the distribution of the correlation entropy induction measurement values, and selecting the singular component of which the correlation entropy induction measurement value is lower than the threshold value; performing signal reconstruction on the selected singular component to obtain a noise-reduced signal; finally estimating the cyclic correlation entropy spectrum of the reconstructed signal, and observing the signal characteristics. The method can effectively discriminate the singular components, and has a good noise reduction effect, thereby extracting the weak features of the signals.

The invention discloses a high-order extended Kalman filter design method based on maximum correlation entropy. According to the method, the method comprises the steps: two one-dimensional random variables are given, and the correlation entropy of a random variable pair under limited data driving is obtained; then a state model and a measurement model of the motion of the unmanned aerial vehicle are given; a high-order polynomial in a state model is defined as a hidden variable of a system, the state model of the system is represented in a pseudo-linearization mode, a measurement model is represented in a pseudo-linearization mode in the same way, and linear forms of the state model and the measurement model are obtained; for the state model and the measurement model in the linear form, a high-order extended Kalman filter is obtained by using a recursive filter design thought; and the high-order extended Kalman filter based on the maximum correlation entropy is designed by using the correlation entropy form of the multi-dimensional independent vector and the obtained high-order extended Kalman filter. According to the method, the problems of reduction and divergence of filtering performance under the condition of a nonlinear non-Gaussian system can be solved, and the method can be applied to the fields of real-time estimation and target tracking.

The invention relates to a rapid nondestructive testing method for a large-size composite material, which is characterized by comprising the following steps: randomly sampling according to a standard component to obtain a rapid sparse representation vector of a defect-free reference signal sample set; performing fast sparse characterization, the sparse characterization of the reference signal and the sampling signal adopting parallel noise reduction and sparse representation of signal singularity measurement based on a wavelet domain modulus maximum value, feature clustering of a nonlinear support vector machine, an adaptive observation model of correlation entropy and normalization of a Sigmoid function; using a maskless compressed sensing (CS) topological distribution optimization strategy of a correlation entropy adaptive model, and characterizing a result and the correlation entropy adaptive model rapidly; judging the homogeneity and heterogeneity of signals through weighting, designing a structured measurement matrix easy to store under a CS theoretical framework in combination with a block random form of a structured thought, and providing effective data for reconstruction after rapid detection. The method is closely related to detection object parameters, the implementation method is simple and easy to implement, and the engineering practicability is high.

The invention relates to the technical field of data processing, in particular to a robust local and global regularized non-negative matrix factorization clustering method, which comprises the following steps of: acquiring an image clustering sample; constructing a nearest adjacency graph on the local scattering of the sample and introducing smooth regularization; using transformation to represent a global geometric structure of the space, and taking the global geometric structure as an additional principal component graph regularization item to be incorporated into an NMF algorithm; graph regularization term constraint is applied to the original NMF model through joint modeling, and the basis matrix is constrained by using LP smoothness constraint; in error measurement, correlation entropy is used to replace Euclidean norm, so that a robust local and global regularized non-negative matrix factorization objective function is obtained; iteration is carried out for preset times by using an iterative weighting method according to the target function, the variables U and V are updated, and robust local and global regularized non-negative matrix factorization is completed; and carrying out clustering analysis on the coefficient matrix by adopting a K-means clustering algorithm.

The invention provides an error parameter identification co-localization method based on a factor graph. The method comprises the following steps: firstly, establishing an error parameter identification factor graph model, then carrying out error parameter identification on a process error containing a speed error and a course error, and carrying out information transmission between a function node and a variable node in a factor graph by utilizing a maximum correlation entropy criterion to realize error compensation of the speed and the course of a slave boat, and finally, filtering fusion estimation of the position state information of the slave boat is realized. Under the condition that the measurement precision of an inertial device in the system is not changed, the process error is identified and compensated, so that the co-positioning error is reduced, and the positioning capability of the co-positioning system is improved.

The invention designs a factor graph co-localization algorithm based on maximum correlation entropy. The method comprises the following steps: firstly, establishing a co-localization algorithm factor graph model based on the maximum correlation entropy, then utilizing the correlation entropy as reliability information, calculating the correlation entropy of variable nodes in the factor graph model, and calculating position state information when the correlation entropy is maximum through differential operation according to a maximum correlation entropy criterion, and realizing filtering fusion estimation of the position state information of the slave boat. According to the method, under the condition of not changing the measurement precision of the inertial device in the system, the error formula of the correlation entropy is used as the cost function of the system, so that the influence of an abnormal value on the system is effectively inhibited, and the robustness of the system is enhanced.

The invention belongs to the technical field of signal processing, relates to a radio signal carrier frequency estimation method based on a generalized cyclic correlation entropy and is used for solving the problem of estimating radio signal carrier frequency in radio signal monitoring. The method includes: utilizing a known signal to respectively calculate a correlation function of the signal in European space and correlation entropy of the signal in Hilbert space; when the known signal has second-order cyclic stable characteristics, acquiring a composite correlation entropy based on limited European space and Hilberty space; calculating the generalized cyclic correlation entropy of the known signal, establishing a three-dimensional stereogram, and acquiring carrier frequency of an original signal through frequency spectrum peak. By the method, robustness to pulse noise, Gaussian noise and noising mixing the two can be improved, and higher-accuracy radio signal carrier frequency estimation can be realized.

The invention provides a maximum cross-correlation entropy Kalman filtering method based on a random weighting criterion. Comprising the following steps: constructing a linear system equation and a measurement equation, selecting a proper kernel width, initializing a system state and a covariance, updating one-step prediction of the state and the covariance according to the system equation, and re-initializing a state value at an initial iteration moment of a fixed point, the method comprises the following steps: performing system model deformation according to an initial system and a measurement equation, calculating an error after deformation and a kernel function of the error, obtaining two diagonal matrixes according to a random weighting criterion and the kernel function, and correcting a one-step prediction covariance and a measurement error by the two diagonal matrixes so as to correct a gain matrix; according to the method, for the non-Gaussian heavy tail impact noise problem of a linear model, the performance better than that of Kalman filtering and maximum cross-correlation entropy Kalman filtering can be obtained, the method can be widely applied to the situation that the noise of a linear system is non-Gaussian, and the filtering estimation precision under the non-Gaussian noise situation is improved.

The invention discloses a wind power prediction method based on a hybrid cross-correlation entropy long short-term memory network. The method comprises the following steps: decomposing an original wind power sequence into K subsequences; carrying out complexity evaluation on the K subsequences, and reconstructing to obtain M reconstructed feature sequences; normalizing the wind speed, the wind direction cosine value, the temperature and the M reconstructed feature sequences; respectively establishing prediction models based on a hybrid cross-correlation entropy long short-term memory network for the processed M reconstructed feature sequences, and optimizing related parameters of the prediction models to obtain predicted values of the M reconstructed feature sequences; and linearly superposing predicted values of the M reconstructed feature sequences to obtain a wind power prediction result. The prediction precision can be effectively improved, and the purposes of high precision and strong robustness are achieved.

The invention discloses a bearing fault diagnosis method based on cyclic correlation entropy and a one-dimensional shallow convolutional neural network. The bearing fault diagnosis method comprises the following steps: collecting N vibration signal samples, and calculating a time-varying correlation entropy matrix, a cyclic correlation entropy spectral density matrix and generalized cyclostationarity DCS[sigma] of each signal sample xi; dividing the vibration signal samples into a training set, a verification set and a test set according to a certain proportion; and taking the generalized cyclostationarity DCS[sigma] as the input of the CCe-1D SCNN, performing one-dimensional shallow convolutional neural network parameter training and testing, and performing bearing fault mode classification. According to the method, the influence of Gaussian noise and non-Gaussian noise can be effectively suppressed, the fault features of the bearing are automatically extracted, the structure of the convolutional neural network is simplified, and the defects that the deep convolutional neural network is high in calculation complexity and prone to over-fitting and gradient disappearance are effectively overcome.

The invention discloses a self-adaptive filtering method and system based on a generalized maximum asymmetric correlation entropy criterion, and the method comprises the steps: 1, carrying out the modeling of an input time sequence signal based on an autoregression model, enabling the input signal to pass through a filter, obtaining an output signal, taking the output signal as a predicted value of the signal at the moment, and carrying out the calculation of the predicted value; obtaining a prediction error of the autoregression model according to the prediction value; 2, according to the prediction error of the autoregression model, constructing an objective function based on a generalized maximum asymmetric correlation entropy criterion, and calculating the loss corresponding to the prediction error; 3, according to the prediction error loss, a gradient descent method is adopted to obtain a filter parameter updating expression, and the filter parameters are updated in real time; 4, the robustness of the adaptive filter under the asymmetric and non-Gaussian noise is analyzed; and 5, carrying out steady-state performance analysis and verification on the adaptive filter. According to the method, the robustness, the performance and the adaptive capacity of the adaptive filter under the non-Gaussian noise can be effectively improved.

The invention discloses a construction method of a mixed pixel demixing model based on a correlation entropy induction measurement criterion. The method comprises the following steps of providing a sparse constraint reconstruction error approximation method unmixing model; deriving an unmixing result of the proposed model by adopting semi-quadratic minimization and a conjugate function theory; andquantitatively comparing and evaluating the unmixing performance of mixed pixels of the proposed model by using a multiplication iteration rule. According to the method, the influence of noise, nonlinearity and abnormal points on the hyperspectral image mixed pixel unmixing is overcome by constructing the mixed pixel unmixing model based on the correlation entropy induction measurement criterion;the model effectively overcomes the influence of various interferences in a hyperspectral image on demixing according to the characteristics that the correlation entropy induction measurement is insensitive to the abnormal points, and the nonlinear problem can be solved, so that a reconstruction error is accurately approximated, a guarantee is provided for improving the demixing performance of the mixed pixels, theoretical derivation is carried out on a proposed model, and the feasibility and the superiority of the mixed pixel demixing based on a correlation entropy induction measurement model are proved.

The invention discloses a multi-label image classification method based on direct push type semi-supervised deep learning, and the method specifically comprises the steps: 1) preparing a multi-label image training sample set, and dividing the training sample set into a labeled training sample set and an unlabeled training sample set; 2) constructing a class symbiosis relation graph; 3) building a network model according to the class symbiosis relation graph; 4) constructing a cross-correlation entropy classification loss function with confidence; 5) constructing a classification probability constraint function with confidence; 6) constructing a category-feature correlation objective function with confidence; 7) constructing a total objective function; (8) training the network model built in the step (3) according to the total objective function constructed in the step (7) in combination with a direct push type semi-supervised learning principle; and 9) inputting a to-be-classified multi-label image into the network model trained in the step 8), and obtaining a prediction category of the image in a classification layer. According to the method, the feature vectors with higher category correlation can be learned.