170 results about "Frequency matrix" patented technology

The invention provides a method of automatically discovering a hot news theme (or a hot event) on the internet, which can automatically discover various news events on the internet and automatically and briefly explain the events. The method comprises the following steps: downloading webpages of recent news, blogs, microblogs and the like from the internet; extracting titles and texts from the news, blogs, microblogs and the like; carrying out word segmentation for the extracted titles and texts to acquire a word frequency vector, and combining all the texts to acquire a word frequency matrix; carrying out cluster analysis for the word frequency matrix with clustering algorithm so that texts with the same theme are gathered and clusters of various themes are acquired; and extracting a title of a central text and keywords from each cluster to explain the cluster, i.e., to explain the news theme.

The invention relates to a distribution network fault classification method based on a convolution depth confidence network. The method comprises the steps of firstly collecting the three-phase voltage, zero-sequence voltage and three-phase current of a low-voltage bus of a main transformer and a low-voltage side of the main transformer, and respectively interceptting the signal waveform data of one cycle wave before and after each fault condition as training samples; secondly, carrying out the time-frequency decomposition on the training sample data of step S1 by using the discrete wavelet packet transform, and obtaining the time-frequency matrix, then constructing the pixel matrix of the time-frequency spectrum map, and constructing the time-frequency spectrum map as the input of the subsequent CDBN model; then constructing the CDBN model to train two convolution-constrained Boltzmann machines in unsupervised learning mode, and adding the softmax classifier after the second CRBM to train the network model to effectively extract and automatically classify the fault features, and finally, using the trained model to realize the fault classification of distribution network. The method of the invention can realize accurate fault location.

The invention discloses a large-scale MIMO uplink transmission channel estimation method with low complexity. The method mainly comprises the following steps: a base station generates a large-scale beam set to cover the whole cell through beamforming, and all users synchronously send pilot frequency signals to the base station; after acquiring pilot frequency information, the base station constructs a beam-domain compressed channel estimation problem according to a pilot frequency matrix, and estimates a normalized angle delay domain channel response matrix through a way of weighting a sensingmatrix by using a priority vector according to a structural sparsity and energy concentration feature of the beam domain channel; and the estimation on the space-frequency domain channel response matrix is accomplished through a matrix multiplying way after obtain the normalized angle delay domain channel response matrix. The power leakage is reduced by using an over-complete discrete Fourier matrix, the sparsity of the angle delay domain is increased, the recovery precision of the compressed channel is improved, the retrieval frequency of the channel vector estimation process is lowered through the weight vector and the priority matrix, and the complexity is reduced.

The embodiment of the invention discloses a channel estimation method and device, relates to the technical field of communication, and helps to reduce indication overhead. The method can comprise thesteps that indication information is generated and sent, the indication information is used for indicating M N-dimensional precoding vectors, each precoding vector is applied to one frequency band inthe M frequency bands, the M N-dimensional precoding vectors form a space frequency matrix, and the space frequency matrix is generated by weighting and combining a plurality of space frequency component matrixes, the space-frequency matrix is an M * N-dimensional space-frequency vector or an X * Y space-frequency matrix, X and Y are respectively one and the other of M and N, the M is greater thanor equal to 1, the N is greater than or equal to 2, and the M and N are integers.

The invention discloses a method for fusing multi-focus images in an anti-noise environment, which can be used for fusing the multi-focus images, has noise resisting capacity, and can be applied to a common optical system. The method particularly comprises the following steps of: (1) performing Laplacian pyramid decomposition on a source image; (2) performing iterated operation on a decomposed sub-band coefficient by a pulse coupling neural network to obtain a corresponding ignition frequency matrix; (3) performing judgment operation by a judgment operator operational unit to acquire a fusion coefficient; and (4) performing LP inverse transformation on the fusion coefficient by adopting a pseudo-inverse reconstruction method to acquire the final fusion image. The method has the capacity of resisting additive noises and JPEG compression system noises, has a high objective evaluating indicator for the fusion image, and effectively weakens 'false Gibss' phenomena at the edge of the fusion image of the conventional LP method; in addition, the method has high adaptability, is convenient for calculation and is favorable for hardware implementation.

The invention discloses a speech enhancement method based on the non-negative low-rank and sparse matrix decomposition principle. The method includes the first step of firstly carrying out smoothing, framing and discrete Fourier transformation on noisy speech signals to obtain noisy speech frequency spectra, the second step of allowing the noisy speech magnitude spectra of frames to serve as column vectors which are arranged in chronological order to form a noisy speech time-frequency matrix and then carrying out non-negative low-rank and sparse matrix decomposition on the noisy speech time-frequency matrix to obtain a non-negative low-rank and sparse matrix, and the third step of utilizing the sparse matrix and reconstruction of noisy speech phase positions to enhance the speech spectra and finally obtaining the enhanced speech in a time domain form through inverse Fourier transformation. By the adoption of the method, noise adaptability is high, endpoint detection and model training are not needed, parameters are fewer and easy to regulate, strong noise environmental performance is good, and therefore the method has a good application prospect.

The invention discloses a position social network-oriented geographic position regularization-based interest point recommendation method. The method comprises the following steps of: establishing a sign-in frequency matrix of users and interest points and calculating preference confidence coefficients, for the interest points, of the users; calculating similarity degrees between the interest points on the basis of geographic position coordinates between the interest points; establishing an interest point neighborhood relationship matrix, and representing feature vectors of target interest points by feature points of adjacent interest points; generating a target function on the basis of a weight probability matrix decomposition model; generating a final potential feature matrix of the usersand the interest points; generating a predicted scoring matrix according to the final potential feature matrix of the users and the interest points; and carrying out personalized recommendation on the users. According to the method, geographic relationships between interest points are explored from the angle of position, and the geographic relationships are fused into weight probability matrix decomposition models in a regular term manner, so that recommendation targets are optimized, the correctness and recall rate are improved, better recommendation effect is obtained, and the method more accords with position social network characteristics.

The invention discloses a method for recognizing the fault type of a feed line of a power distribution line, and the method comprises the steps: obtaining waveform sampling data, and sequentially carrying out the Hilbert-Huang transformation and band-pass filtering of the waveform sampling data; reconfiguring a time frequency matrix according to the band-pass filtering data, solving the singular value of the time frequency matrix, and forming feature vector matrixes; carrying out the normalization processing of all feature vector matrixes, and enabling the feature vector matrixes to serve as the input samples of a multi-stage support vector machine after normalization processing, so as to recognize the fault type of the feed line of the power distribution line. The invention also provides a device for recognizing the fault type of the feed line of the power distribution line. The multi-stage support vector machine is good in performance, is clear in logic, is simple, and can recognize four types of power grid faults: single-phase grounding faults, two-phase grounding faults, two-phase short-circuit faults and three-phase short-circuit faults. The method provided by the invention is stronger in adaptive capability, and still has a high recognition rate of fault types under the impact of noise.

The invention discloses a personalized commodity recommendation method based on a frequency matrix and text similarity. The method comprises the following steps: data acquisition, data cleaning, access user identification, session identification and transaction identification are carried out during pre-processing to obtain data of unified format; and the personalized commodity recommendation method based on a frequency matrix and text similarity is used to perform calculation to obtain a commodity candidate set and perform grading on the basis of the commodity candidate set, and a final result is presented to a user. A commodity recommendation module is constructed and implemented by the use of an access frequency matrix and text similarity calculation, the complexity of a recommendation system is reduced as far as possible, the requirement for real-time recommendation is met, and high coverage rate and high matching rate are maintained.

The invention discloses a fast frequency hopping synchronization method based on time of day (TOD) information. The method comprises the following steps of: forming a frequency matrix consisting of n groups of working frequency produced by a transmitting end according to time of day of a high section (TODH) and front and back (n-1)/2 values and an agreed algorithm, selecting frequencies on diagonals as working frequencies, and transmitting synchronous information by adopting frequency modulation; producing a group of waiting frequencies by a receiving end according to the TODH and the same algorithm, wherein the waiting frequencies and received signals undergo frequency mixing, band-pass filtration and square law detection, and the number of the waiting frequency with the maximum value reflects deviation of TOD information of the receiving end and the transmitting end; and adjusting the TODH of the TOD information of the receiving end according to the number of the waiting frequency. Therefore, the synchronous frequencies can be changed together with change of the TOD, anti-interference performance and intercepting resistance of a frequency hopping system are ensured, meanwhile, the TODH of the TOD is adjusted under the condition that related codes are not transmitted, and the aim of synchronization of the frequency hopping system is quickly and accurately fulfilled.

The invention discloses a transmission line fault fast phase selection method based on S conversion. The transmission line fault fast phase selection method based on S conversion includes following steps: (1) collecting the current of the protection element at the line side at real time; (2) executing phase-mode transformation on the fault component current and extracting the model component of the fault component current after filtering the power frequency component for the corrected current; (3) executing S conversion on the model component current signal to obtain complex time-frequency matrix of each model, obtaining the model and phase angle, judging the ground fault or non-ground fault according to the characteristic of the zero model phasor, using the relationship among the line model phasors for having fast phase selection operation; (4) respectively integrating for the wave form of the phase current at two sides of the time axis, describing the wave form characteristic and recognizing the fault phase while the fault is dual-phase ground fault. The method adopts the S conversion for extracting the single frequency model component of each model component at special time, the fault type is determined according to the relationship among the model phasors, the method has the advantages of high sensitivity, fast action speed and exact recognition.

The invention relates to a power distribution network feeder fault type identification method and device. According to the method and device, the transient information of post-fault related electric quantity is mainly utilized to realize fault identification. The method includes the following steps that: waveform sampling data are obtained at first, local feature scale decomposition, Hilbert transform and band-pass filtering are sequentially performed on the waveform sampling data, so that a reconstructed time-frequency matrix can be obtained; and the singular spectrum of the time-frequency matrix is solved, and the distribution parameters of the singular spectrum are extracted so as to construct feature vector matrixes, normalized processing is performed on all the feature vector matrixes, and the normalized feature vector matrixes are adopted as the input samples of a multi-level support vector machine, so that feeder fault type identification can be realized. The power distribution network feeder fault type identification method and device of the invention have the advantages of high fault type recognition accuracy and strong adaptability under conditions such as a noise interference condition.

The invention discloses an FDD large-scale MIMO channel estimation pilot frequency optimization method based on compressed sensing, and the method comprises the steps: firstly enabling a channel to be modeled into a formula in a large-scale MIMO system: Y=HX+N, wherein H (shown in the description) is a channel matrix, X (shown in the description) is a pilot frequency matrix, Y (shown in the description) is a receiving signal matrix, and N (shown in the description) is channel noise, M is the number of transmitting antennas, and T is the number of pilot frequencies; secondly carrying out the conversion of the channel matrix, and solving the conjugate matrix (shown in the description) of Y, wherein the conjugate matrix (shown in the description) of the channel matrix represents the conversion form of the channel matrix, the conjugate matrix (shown in the description) of the pilot frequency matrix represents the conversion form of the pilot frequency matrix, and the conjugate matrix (shown in the description) of the receiving signal matrix represents the conversion form of the receiving signals of a receiving end; and finally solving an optimal pilot frequency matrix. Because the conjugate matrix (shown in the description) of the channel matrix is a sparse vector, a channel estimation problem can be modeled into a compressed sensing reconstruction problem shown in the description, wherein ||*||<1> represents 1-norm, ||*||<2> represents 2-norm, and epsilon is greater than zero and less than one. The method can guarantee that the FDD MIMO downlink channel estimation based on compressed sensing can remarkably reduce the mean square error of channel estimation, and improves the channel estimation performance.

The invention discloses an overvoltage identification method based on a time-frequency matrix singular value. The overvoltage identification method comprises the following steps of obtaining an overvoltage signal; preprocessing the overvoltage signal; decomposing the preprocessed overvoltage signal, and obtaining signal components of different frequency bands; according to the signal components of the different frequency bands, constructing a multi-scale time-frequency matrix; conducting singular value decomposition on the matrix, and calculating the singular values in different stages; and according to the weighted energy contribution rate of the singular values of the different stages, selecting characteristic quantity of overvoltage. The overvoltage identification method extracts characteristic parameters which effectively reflect the essence of the switching overvoltage according to wave shapes of transient state switching overvoltage obtained by monitoring equipment, and therefore achieves effective identification on the switching overvoltage of a power grid.

The invention provides a method for discriminating re-package of an application based on a keyword context frequency matrix, which is applied to an android system. The method comprises the following steps of firstly processing an application program file to obtain a smali code file, processing smali codes, extracting an operator sequence, counting keyword information, constructing context-related characteristic triple to each specific type keyword to generate a characteristic matrix based on context frequency, performing comparison in pairs to the characteristic matrixes of application programs, and obtaining the similarity degree of two application programs according to the similarity degree of the characteristic matrixes; finally combining contents such as writer information to judge whetherthe re-package relationshipexists inthe application programsor not. By utilizing the technical scheme of the method provided by the invention, re-packed android application programs can be judged, and meanwhile, the additional expense for performing mega string hash process to a whole application program is avoided; no dependency on a binary code sequenceof an original file exists; by limiting the sizes of the characteristic matrixes, the space expense is reduced; the performing efficiency of android application program re-package judgment is improved.

The invention relates to a novel disturbance identification method used during intrusion of lightning waves in a transformer substation, which comprises the following steps of: (1) monitoring a voltage signal of a busbar in real time; (2) performing S transformation on the voltage signal in failure, and establishing a module time-frequency matrix of the voltage signal; and (3) calculating the amplitude of fundamental waves, the times of primary harmonic waves, and the maximum amplitude of the primary harmonic waves by using the module time-frequency matrix, wherein the drop of the amplitude of the fundamental waves of the voltage of the busbar serves as the evidence for judging faulted and non-faulted lighting or other interferences, and the times of the primary harmonic waves and the corresponding amplitudes thereof serve as the evidence for judging whether lighting stroke occurs. Characteristics of a disturbance signal are extracted by using an S transformation technology suitable for analyzing an instable signal; and identification data are established according to the times of the primary harmonic waves of a transient voltage signal and the corresponding amplitudes thereof, aswell as the drop difference of the voltages of the fundamental waves under different disturbance conditions; and moreover, the method has the advantages of simplicity, convenience, high sensibility and accuracy for identification, and is easy for engineering realization.

The invention belongs to the technical field of big data, and discloses a multi-source data aggregation sampling method and system based on a big data environment, and the method comprises the steps:collecting a plurality of original data sources, wherein each original data source comprises a data source name and at least one association domain; cleaning and identifying the acquired data source,and removing redundant of the acquired data source; obtaining an original strategy list by utilizing a construction program according to the original data source, and sorting the original strategies in the original strategy list to form a strategy list between the data sources; carrying out fusion processing on different source data sets by utilizing a fusion program; carrying out word segmentation on the fused file to form a two-dimensional word frequency matrix of file words; setting a balance verification numerical value, circularly matching each word, and carrying out snowball sampling; and displaying the acquired multi-source data by using a display. According to the method, distributed computing is completed by scheduling the computing nodes by the Spark through the preprocessing module, more efficient data preprocessing can be achieved, practicability is high, and the application range is wide.

The invention discloses a time-rearrangement compression transformation-based time-frequency analysis and reconstruction method of an impact-type signal. The method includes: 1) calculating short-timeFourier transform (STFT) of a to-be-analyzed discrete signal to obtain a corresponding time-frequency complex-matrix S<x>[n,k]; 2) carrying out short-time Fourier transform on the product of a time variable and the signal to obtain a time-frequency complex-matrix S<tx>[n,k]; 3) calculating a group delay estimation operator; 4) only rearranging the time-frequency complex-matrix S<x>[n,k], which isobtained in the step 1, along a time direction to obtain a time-frequency matrix V<x>[m,k] after time-rearrangement compression transformation; and 5) adding all rows of elements of the time-frequency matrix V<x>[m,k], which is obtained after time-rearrangement compression transformation, together to obtain a one-dimensional column vector, and then dividing the column vector by a mean value of awindow function, which is used by short-time Fourier transform, to obtain a frequency spectrum of a reconstructed signal. For analyzing the impact-type signal, the time-frequency graph aggregation obtained by time-rearrangement compression transformation to which the invention relates is higher than that of synchronous compression transformation, and good anti-noise performance is realized. Compared with traditional time-frequency rearrangement, the method has the advantages of reconstructability and a fast calculation speed.

The invention belongs to the technical field of speech separation and particularly relates to a memory and attention model-based auditory selection method and device, so as to solve the problems of supervision label arrangement, speaker aliasing number uncertainty and memory unit dimension fixation in the prior art. The memory and attention model-based auditory selection method comprises steps: original speech signals are coded as a time frequency matrix, the time frequency matrix is subjected to coding and transformation, the time frequency matrix is converted to speech vectors, a long term memory unit is used to store speakers and corresponding speech vectors, the speech vector of a target speaker is acquired, and target speech is separated from the original speech signals through an attention selection model. The method provided in the invention does not need to fix or designate the number of speakers but can separate the target speech from the original speech signals.

The invention discloses a single source point detection-based underdetermined blind source separation method. The method comprises the steps of firstly performing short-time Fourier transform on to-be-analyzed observation signals to obtain corresponding time frequency domain complex matrixes; secondly vectorizing and normalizing the time frequency complex matrix of each observation signal; thirdlydetecting out all equal column vectors in the normalized time frequency matrix by utilizing cosine included angle criteria of the vectors, wherein the extracted column vectors are namely single source points; fourthly performing hierarchical clustering on the extracted single source points to obtain a clustering center, thereby realizing estimation of a mixed matrix, wherein the center of each type corresponds to a column of the mixed matrix; and finally by utilizing the estimated mixed matrix, realizing time frequency estimation of all sources at all time frequency points through a least square method, and obtaining time domain forms of the sources through time domain inverse transform. According to the method disclosed by the invention, a linear relationship among different single source points is considered, and the detection of the single source points can be realized only by judging whether the vectors are same or not, so that efficient and high-precision estimation of the mixedmatrix and source signals can be realized under the underdetermined condition.

The embodiment of the invention provides an S conversion based SDPQD (Short Duration Power Quality Disturbance) signal identifying method, a device and a system. The method comprises the following steps: collecting SDPQD signals on a substation; performing dispersing S conversion on the collected SDPQD signals so as to generate a time frequency matrix; performing scaling on the time frequency matrix; confirming a similarity value of each row in the time frequency matrix according to a preset standard time frequency matrix; confirming an average similarity value of the time frequency matrix according to the similarity value of each row; and identifying the SDPQD signal according to the average similarity value of the time frequency matrix. The method provided by the invention is used for solving the problems of the identifying method in the prior art that only simple disturbance signals can be identified and the complicated disturbance signals, such as transient oscillation, short-time harmonic wave, cannot be identified.

The invention discloses a chaos-based method for detecting and classifying early single-point faults of a mechanical component. The method comprises the following steps of: processing conventional sample fault signals in different states of the mechanical component to establish check intervals for different fault types; acquiring fault characteristic frequencies corresponding to all single-point fault states of the mechanical component to construct a frequency matrix of a Duffing chaotic oscillator; solving critical thresholds of periodic driving force amplitudes corresponding to different fault characteristic frequencies to construct a frequency-threshold matrix; and finally, adding a signal to be detected to calculate the maximum Lyapunov exponent matrix M, checking according to data inthe M, calculating correlation dimension of the signal to be detected if a fault signal is available, and classifying the faults to determine a fault mode in comparison with the established correlation dimension intervals for different fault types. By adopting the method, the early single-point faults of the mechanical component are detected and classified; and the method has high noise resistance capacity and extremely high fault detection success rate.

The invention discloses a method and a device for extracting domain keywords. The method comprises the following steps of: generating a domain term frequency matrix consisting of term frequencies of words in each domain description text; according to a set algorithm, decoupling the domain term frequency matrix into a sum of a low-rank background word term frequency matrix and a sparse keyword term frequency matrix; and according to the keyword term frequency matrix obtained by decoupling, extracting the keywords in the corresponding domains from the words in each domain description text. According to the technical scheme disclosed by the invention, extraction of the domain keywords can be carried out on the basis of distribution of the frequencies of the words in each domain text in all the domain text and the keywords with representativeness and discrimination in the corresponding domains can be accurately and effectively extracted from each domain text.

The invention discloses a multi-source data aggregation sampling strategy based on a big data environment. The method comprises the following steps that on the basis of a multi-source data set encoded by GBK, fusion of multiple attributes in one data set and fusion of different source data sets are achieved, and multi-source data fusion operation is completed; fused files are segmented to form a two-dimensional word frequency matrix of file words; high-frequency words are displayed to a user for reference, and random migratory selection of seed root node words required by snowball sampling is conducted; business-goal-oriented seed root node keywords are selected, and the depth of snowball sampling is input; on the basis of seed root node data, a balance check value is set, and the words are circularly matched for snowball sampling; a directed acyclic graph and an adjacent matrix are constructed; a root node clustering network diagram and a logical reasoning diagram which are related to business goals can be obtained. The interference of sample noise on subsequent reasoning can be reduced through the strategy.

The present invention provides a method and system for transmitting data. The system includes an encoder for converting the data into one or more binary sequences using a binary encoding scheme, selecting one or more sets of frequencies from a frequency matrix based on the one or more binary sequences, and prepending a starter frequency to the one or more sets of frequencies. The system further includes an audio generator for generating one or more audio signals based on the one or more sets of frequencies, a storage module and an output module having one or more an electro acoustic transducers for generating an acoustic output based on the one or more audio signals.