76 results about "Sample vector" patented technology

The invention discloses a cross-language code semantic alignment method, which constructs a shared semantic space through a unified abstract syntax tree (AST) and a graph matching network (GMN) so as to reduce the difference of different programming languages in syntax structure and node representation. The method comprises the following steps: (1) mapping a multi-language AST node to a unified general label set and performing structure enhancement; (2) performing node feature coding on the unified AST, and realizing cross-language interaction in combination with a cross-graph attention mechanism; (3) node representation is generated through intra-graph loop updating, and an overall semantic vector is obtained through global attention pooling; and (4) through comparative learning training in the shared space, the distance between semantically equivalent positive sample vectors is shortened, and the distance between non-equivalent negative sample vectors is shortened, so that the discrimination capability of cross-language semantic representation is enhanced. According to the method, the semantic consistency of the functional level can be effectively captured, and the accuracy and efficiency of cross-language code understanding, multiplexing and retrieval are remarkably improved.

The invention discloses a document proofreading system and method based on artificial intelligence, and relates to the technical field of natural language processing, and the method comprises the steps: analyzing a document, dividing semantic units, and obtaining a document with a context label based on context type library labeling; performing initial error detection by using the corresponding rule set, and comparing the format of the document and the text quality characteristics with a standard sample vector space to obtain an initial error set; constructing a defect conduction chain network, and locating root cause error nodes causing a plurality of secondary errors by tracing the defect conduction chain network; generating an intelligent proofreading report; receiving a new-version document revised by the user, comparing the new-version document with the original-version document, positioning a change area, and analyzing a quality difference of the change area on a related quality dimension; based on the quality difference, an incremental proofreading report is output, and the incremental proofreading report comprises newly introduced errors and unsolved root cause errors.

The invention discloses an electric energy quality disturbance positioning identification method based on phase perception and multi-mode fusion, and relates to the technical field of electric power system monitoring and fault diagnosis. The electric energy quality disturbance positioning identification method based on phase perception and multi-modal fusion comprises the following steps: acquiring an original voltage signal of a three-phase electric power system; extracting a sample embedding vector of each phase based on the original voltage signal, comparing the sample embedding vector with a normal sample vector, and determining a disturbance phase with power quality disturbance in combination with an adaptive threshold value; performing time sequence feature vector extraction and bispectrum image feature extraction on the disturbance phase, and performing alignment and collaborative fusion to obtain fusion features; the electric energy quality disturbance type is identified based on fusion features, phase-level positioning of the three-phase PQD is achieved, through an innovative phase perception embedding and prototype comparison mechanism, accurate and rapid positioning of abnormal phases in a three-phase system is achieved for the first time under the condition of not depending on disturbance phase labels, and the core pain point of a traditional method is solved.

The invention relates to the technical field of power systems, in particular to a power grid fundamental harmonic and inter-harmonic detection method and device and a storage medium, and can collect power grid signals, arrange discrete samples into sample vectors, apply a Hanning window to the sample vectors for zero filling DFT, and obtain a frequency index of fundamental waves according to a peak value of frequency spectrum amplitude square; setting a harmonic wave search range according to the frequency index of the fundamental wave, and screening the harmonic waves according to the frequency related threshold to obtain a frequency index set of the fundamental wave and the harmonic waves; calculating a residual error between the power grid signal and the fundamental harmonic reconstruction signal, and calculating a residual error ratio; and performing secondary iteration on the residual error to obtain a TF coefficient corresponding to the inter-harmonic. According to the technical scheme, fundamental wave phasor measurement, harmonic detection and inter-harmonic detection can be completed at the same time, the precision is high, and the detection efficiency is high.

The application discloses a malicious traffic detection method, system and device and a storage medium, and belongs to the technical field of network traffic analysis and cyberspace security application, and the method comprises the following steps: obtaining traffic statistical information to be detected, and performing format preprocessing on the traffic statistical information to obtain a sample vector; inputting the sample vector into a pre-trained neural network partial framework search network model to obtain a prediction vector; the prediction vector comprises a plurality of prediction values, each prediction value comprises a classification label of itself, a classification label of a maximum prediction value is selected as a final classification label, if the final classification label is malicious, then traffic corresponding to the traffic statistical information is malicious traffic, otherwise, the traffic is non-malicious traffic; the category of the traffic can be determined without manual feature design; by using a relatively light model, the calculation amount is reduced, the model can be deployed on an edge computing node, the feature extraction capability and practicability are enhanced, and the problems of insufficient precision and insufficient universality are overcome.

The invention belongs to the technical field of ocean exploration, and particularly relates to an ocean exploration wave field reconstruction method based on dictionary learning, and the method comprises the steps: carrying out the preprocessing of original wave field data, and obtaining the preprocessed wave field data; separating the preprocessed wave field data into non-overlapping block sample matrixes; converting all block sample matrixes into a sample vector set; selecting a training sample set from the sample vector set; adopting the training sample set to construct a double-sparse dictionary model; iteratively updating the double sparse dictionary model by adopting an alternate optimization mode to obtain a final sparse dictionary and a sparse coefficient matrix corresponding to the sparse dictionary; and carrying out sparse reconstruction on the actual observation wave field by adopting the final sparse dictionary and the sparse coefficient matrix corresponding to the final sparse dictionary. The method has high adaptability to noisy or under-sampled wave field data, and a continuous and high-fidelity wave field structure can still be recovered even under the condition that data acquisition conditions are limited.

The application discloses an anchor point guided clustering based data classification method, device and equipment, relates to the technical field of digital data processing, and comprises the following steps: obtaining an original data set to be classified, converting each sample data into a numerical vector, and constructing a data matrix; initializing a clustering center matrix and an anchor point matrix, and setting a fuzzy coefficient; obtaining the anchor point matrix and the clustering center matrix after joint iterative optimization; calculating the fuzzy membership degree of each sample vector; the category with the maximum fuzzy membership degree is taken as the final category label of the sample vector, and the classification results of all sample vectors are output. The application solves the problems that the existing fuzzy clustering method is sensitive to initial conditions, is easy to fall into a suboptimal solution, leads to unstable and inaccurate classification results, and cannot be directly solved by using a gradient descent algorithm, and is difficult to be applied to large-scale data sets, and realizes the enhancement of complex data classification precision and the applicability in different data scale scenes.

This application discloses a training method for a source evaluation model, a source evaluation method, and related products. Text sample sequences and frequency sample sequences are input into the evaluation model to be trained. The model encodes the text sample sequences and frequency sample sequences to obtain text sample vectors corresponding to the text sample sequences and frequency sample vectors corresponding to the frequency sample sequences. The evaluation model then performs prediction and evaluation processing on the text sample vectors and frequency sample vectors to obtain source evaluation prediction results. Based on the difference between the source evaluation result labels and the source evaluation prediction results, the parameters of the evaluation model are adjusted until the adjusted model meets the model training cutoff condition, and training ends to obtain the source evaluation model. Thus, this application can construct a source evaluation model based on the source sample name and the title, keywords, and publication frequency of the published sample text as features, thereby achieving the evaluation of the source.

The invention belongs to the field of magnetic sensors, and provides a nested neural network-based array current sensor rapid resolving method, which comprises the following steps of: S1, acquiring magnetic field detection sample vectors of a magnetic sensor array under different working conditions, and taking a lead standard current value as label data of the sample vectors, constructing a data set of a first neural network; s2, dividing the data set in the S1; s3, constructing a first neural network model by using the data set divided in the S2; s4, in an actual measurement environment, inputting an output signal of the magnetic sensor array on the measured wire into the first-layer neural network model to obtain a current of the measured wire; forming a second neural network data set, and dividing the second neural network data set; s5, constructing a second neural network model by using the data set divided in the step S4; and S6, obtaining a predicted current value of the current to be measured by using the second neural network model. According to the invention, high-precision rapid calculation of the current in a complex interference environment can be realized.

The invention relates to a permanent magnet synchronous motor model prediction control method and system, belongs to the technical field of permanent magnet synchronous motor driving, and solves the problem that parameter robustness and dynamic performance cannot be considered when parameter mismatch occurs in permanent magnet synchronous motor model prediction control in the prior art. Comprising the steps of obtaining sampling values of stator current, stator voltage and electrical angular velocity at a current sampling moment and two previous moments and a direct current bus voltage at the current sampling moment, further updating incremental current prediction errors and time sequence sampling data of incremental voltage, and further obtaining a first sample vector and a second sample vector at the current sampling moment; and then a small-batch stochastic gradient descent method is adopted to obtain an identification value of a parameter mismatch coefficient at the current sampling moment, then stator current prediction values of different voltage vectors at the next moment are obtained, then an optimal voltage vector is obtained, and permanent magnet synchronous motor model prediction control at the next moment is carried out based on the optimal voltage vector.

An apparatus and method for determining a frequency reference offset between a transmitter and a receiver. Time sample vectors are created, where each time sample vector has a sequence of time samples of a received chirp period within a sequence of chirp periods. For each pair of time sample vectors in the time sample vectors, a product vector of each pair of time sample vectors is computed. Each product vector of each pair of time sample vectors is added to result in an accumulated value vector. A discrete Fourier transform of the accumulated value vector is computed. A detected frequency is determined based on a frequency corresponding to an element of the discrete Fourier transform that has a higher magnitude than other elements of the discrete Fourier transform. A frequency reference offset between a transmitter of the received chirp period and a receiver of the received chirp period is determined based on the detected frequency.

The invention discloses a second-hand car price evaluation model training method and device based on deep learning, and the method comprises the steps: obtaining the historical transaction data of a second-hand car, and carrying out the data preprocessing, and obtaining a training sample; selecting samples from the training samples to construct a sample pair of the to-be-evaluated sample and the real sample; respectively inputting the real sample and the to-be-evaluated sample into a first sub-network and a second sub-network of the double-tower model to obtain a first sample vector and a second sample vector; wherein the first sub-network and the second sub-network both adopt a deep cross network architecture; performing element-level product on the first sample vector and the second sample vector to obtain an interactive feature vector, and mapping the interactive feature vector into a scalar through an output layer; all parameters of the double-tower model are updated through back propagation according to the scalar and the loss function of the sample price difference, a trained second-hand car price evaluation model is obtained, and the problems that a second-hand car price prediction method is weak in feature intersection and insufficient in sample relevance utilization are solved.

The application relates to a computing power cloud large language model training optimization method, which comprises the following steps of: in an offline tool software, Tokenizer internal order is disturbed to obtain a new Token sequence; an Embedding matrix is adjusted according to corresponding positions according to the new Token sequence to obtain a new Embedding matrix corresponding to the new Token sequence; the Tokenizer is retrained according to the new Token sequence and the new Embedding matrix; a corresponding Tokenized sample vector is generated according to the retrained Tokenizer; a large language model after modification is tested, sample vectors and corresponding Embedding matrices are uploaded, and a supervised fine-tuning process is performed. The application can reduce the risk of data leakage, and it is beneficial for computing power service providers and users to reduce the risk of data leakage. The application can reduce the possibility of data leakage for service providers and make users more confident in using cloud services. No additional calculation and storage are introduced. Generally, users renting computing power cloud need to use GPU or TPU acceleration functions, and the Tokenize process generally uses the calculation capacity of CPU, so that Tokenize on the user side does not introduce additional calculation and storage.

The invention relates to the field of data processing, in particular to a heart age determination method and device, electronic equipment and a storage medium, and the method comprises the steps: minimizing the distance between heart health characterization vectors of the same heart health state output by an initial heart health encoder; performing updating iteration on the initial heart health encoder to obtain a first determination model by taking maximization of the distance between the heart health characterization vectors of different heart health states as a target; updating and iterating the first heart health encoder according to the heart fusion feature sample vector and the heart activity state and the relative heart age thereof to obtain a second determination model; and taking the heart fusion feature sample vector as a sample and the heart biological age as a label to update and iterate an age prediction module to obtain a target determination model so as to determine the heart age. According to the method and the device, the heart age can be accurately and efficiently determined through the trained determination model.

This invention discloses a training method for an equipment fault diagnosis model and a method for equipment fault diagnosis, relating to the fields of fault diagnosis and deep learning technologies. The method includes: acquiring at least one sound sample data and corresponding label data; processing the sound sample data through a background noise suppression channel and an impulse interference suppression channel to obtain a denoised sound sample vector; performing dual-tree complex wavelet packet decomposition on the denoised sound sample vector to obtain a sample time-frequency feature matrix; processing the denoised sound sample vector to obtain a sample operating condition auxiliary feature vector; the sample operating condition auxiliary feature vector is used to characterize the equipment operating condition state corresponding to the sound sample data; and training a deep learning model based on the sample time-frequency feature matrix, the sample operating condition auxiliary feature vector, and the label data to obtain an equipment fault diagnosis model. The above technical solution can improve the accuracy of equipment fault diagnosis.

The invention discloses a data classification method, device and equipment based on anchor guide clustering, and relates to the technical field of electrical digital data processing, the method comprises the following steps: obtaining a to-be-classified original data set, converting each sample data into a numerical vector, and constructing a data matrix; initializing a clustering center matrix and an anchor point matrix, and setting a fuzzy coefficient; obtaining an anchor point matrix and a clustering center matrix after joint iterative optimization; calculating the fuzzy membership degree of each sample vector; the category with the maximum fuzzy membership degree value serves as a final category label of the sample vectors, and classification results of all the sample vectors are output. The problems that an existing fuzzy clustering method is sensitive to initial conditions and is prone to falling into a suboptimal solution, so that a classification result is unstable and inaccurate, and a gradient descent algorithm cannot be directly used for solving and is difficult to adapt to a large-scale data set are solved; the method achieves the enhancement of the classification precision of complex data and the applicability in different data scale scenes.

The application discloses a kind of magnetic sensor array anti-interference current measurement method and system based on neural network, the method of the application includes respectively collecting the magnetic field detection value sample vector of magnetic sensor array under different wire current, wire eccentricity, crosstalk current interference, stray magnetic field interference and external temperature and humidity variation;Wire current is used as label data to construct training data set;According to the preset proportion, training set and test set are divided;Machine learning model is constructed, and machine learning model is trained using training set and test set in combination with preset loss function and optimization algorithm;The magnetic field detection value sample vector detected by the magnetic sensor array on the measured wire is input into the trained machine learning model to obtain the wire current of the measured wire.The application aims at the problem that current sensor is easily subjected to external current magnetic field interference, resulting in the decline of current measurement accuracy, and provides a kind of current measurement technical scheme for effectively resisting external magnetic field interference using artificial neural network.

The application discloses a comment information classification method and device, electronic equipment and storage medium, the method comprises the following steps: obtaining the text vector representation of sample comment information; mining the entity and theme of the text vector representation, and constructing a heterogeneous information graph according to the entity, theme and text vector representation; performing convolution operation on the heterogeneous information graph to obtain the text node representation of the sample comment information; training a model using the text node representation to obtain a comment information classification model; and classifying target comment information using the comment information classification model. That is, the application can vectorize the sample, construct a heterogeneous information graph according to the entity and theme mined from the sample, realize semantic enhancement of the sample from multiple dimensions, solve the problems of feature sparsity and high noise, obtain the features required for model training based on the convolution operation on the heterogeneous information graph, train the features to obtain a model for classifying comment information, and improve the accuracy of the classification result.

An apparatus and method for determining a frequency reference offset between a transmitter and a receiver. Time sample vectors are created with each having a time sample sequence of a received chirp period within a sequence of chirp periods. A product vector for each pair of time sample vectors is calculated for each pair of time sample vectors in the time sample vectors. Each product vector of each pair of time sample vectors is summed into an accumulated value vector. A Discrete Fourier Transform of the accumulated value vector is calculated. A detected frequency is determined based on a frequency corresponding to an element of the Discrete Fourier Transform that has an amplitude higher than other elements of the Discrete Fourier Transform. A frequency reference offset between a transmitter of the received chirp period and a receiver of the received chirp period is determined based on the detected frequency.

Embodiments of the present application disclose a picture analysis model training method, an advertisement picture selection method and an electronic device. The picture analysis model training method comprises: inputting at least one group of sample vectors into a picture analysis model to obtain a first probability corresponding to each group of sample vectors; each group of sample vectors is generated based on a first advertisement and a second advertisement with the same advertisement language on a same advertisement position, and comprises the advertisement language, a first picture used by the first advertisement, a first text used by the first picture, a second picture used by the second advertisement, and a second text used by the second picture; the first probability represents a probability that the first picture is better than the second picture in the corresponding sample vector; a click rate of the first advertisement is higher than a click rate of the second advertisement; a loss value corresponding to each group of sample vectors in the at least one group of sample vectors is calculated based on the first probability corresponding to each group of sample vectors; and a weight parameter of the picture analysis model is updated according to the calculated loss value.

The invention discloses a permanent magnet synchronous motor phase current reconstruction method and device for reducing harmonic waves, and belongs to the technical field of motor control. According to the method, a sampling topological structure is optimized, a sampling resistor is shifted from a direct-current bus side to a position between upper bridge arms of two phases A and B of an inverter, and low-harmonic and high-precision phase current reconstruction is realized by combining a novel switch state phase shift method (N switch state phase shift strategy) and a dynamic zero vector distribution strategy. The method specifically comprises the steps that sectors are divided according to a space vector diagram, and sampling vectors are selected; phase shift operation is carried out in a low modulation degree area through minimization of PWM waveform modification, and the average voltage conservation principle is followed; zero vector time is dynamically allocated in a high modulation degree region to extend a linear modulation range. The phase current total harmonic distortion (THD) is remarkably reduced, the modulation range is expanded, meanwhile, the cost effectiveness and the real-time performance are considered, and the method is suitable for low-cost permanent magnet synchronous motor driving systems such as electric automobile fans and water pumps.

The invention discloses a population data generation method and device and a computer readable storage medium, and the method comprises the steps: carrying out the conversion of an initial population sample from an attribute domain to a copula domain, obtaining a copula domain sample, carrying out the vectorization coding representation of the copula domain sample, and obtaining a matrix sample; performing diffusion model training and sample generation based on the matrix sample to obtain a new matrix sample; and decoding the new matrix sample to obtain a copula domain vector sample, and performing quantile function mapping on the copula domain vector sample to obtain an attribute domain demographic data generation sample. According to the method, the initial population sample is converted to the copula domain, the correlation mode between the attribute variables is reflected through the copula density function, the influence of edge distribution is avoided, the intrinsic dependency constraint relation between the variables can be effectively captured, and the population sample data with high feasibility and diversity is generated.

The application discloses a kind of finger vein image recognition methods based on multi-index fusion pre-evaluation, comprising: obtaining image quality evaluation index, carrying out normalization processing to evaluation index, generating sample vector, determining classifier core parameter and generating SVM classifier, evaluating and screening picture to generate database to image, carrying out HOG feature extraction and gray level co-occurrence matrix feature extraction to the image processed, fusion two kinds of features form fusion vector, identity matching.The application can effectively reduce the problems such as recognition rate reduction caused by uneven image quality.

The invention discloses an image feature reconstruction and identification method based on fractional multi-view nonlinear coherence. The method comprises the following steps: 1) modeling a multi-resolution image sample; 2) projecting the data to a high-dimensional feature space through nonlinear mapping; 3) adopting dual representation and converting an optimization problem; 4) introducing fractional order modeling; 5) re-estimating the intra-group covariance matrix and the inter-group covariance matrix; 6) gradually solving a plurality of projection directions through an iteration recursion mode; 7) performing principal component analysis dimension reduction on each sample vector; 8) mapping the low-resolution input image to a high-resolution feature space; and 9) carrying out image super-resolution feature reconstruction to realize super-resolution feature recovery. According to the method, a fractional order embedding thought is introduced under a nonlinear multi-set partial least square framework, fractional order re-estimation is carried out on intra-group and inter-group covariance matrixes, so that more stable feature representation is obtained under the condition that construction samples are insufficient, and higher precision and robustness are achieved.

This invention relates to the fields of computer vision and intelligent video analysis technology, and in particular to a cross-viewpoint human body re-identification and matching method based on a corner domain vector generation mechanism, comprising the following steps: S1, constructing a multi-corner domain sample collection set for the same identity; S2, extracting an initial identity vector from each human body image using an unsupervised re-identification model; S3, generating the identity vector of the human body target in the target corner domain; S4, applying corner domain consistency constraints to the identity vector in the target corner domain; S5, completing the cross-angle and cross-viewpoint human body identity matching determination based on the similarity comparison between the multi-corner domain identity vector set of the query target and the candidate sample vector; This invention solves the problem of decreased cross-viewpoint matching accuracy in existing technologies by directly generating the representation of the same identity in multiple target angles within the feature space, without the need for complex image generation.

The invention relates to network flow multi-type confrontation sample detection based on residual enhancement and evidence conflict weight reduction, and belongs to the field of network security. According to the method, firstly, normal sample training is utilized to generate an intrusion detection source model and an auto-encoder, layer copying and full connection layer adding are sequentially carried out on the source model to form k models, and then adversarial samples are utilized to finely adjust the k models to generate a detector; then calculating a residual error between a to-be-detected sample vector and an auto-encoder reconstruction vector, selecting a residual error component by adopting a gating network to increase disturbance, and then detecting an addition result of the to-be-detected sample vector and an enhanced residual error; and constructing D-S evidences by using detection confidence, calculating an evidence conflict penalty weight through a Jousseme distance between the evidences, and finally detecting whether the sample is an adversarial sample by using a D-S fusion algorithm. Aiming at the problems that only specific types of confrontation samples are detected and multi-detector decision conflicts exist in an existing method, multi-type confrontation sample detection is supported through residual enhancement, and a multi-detector evidence conflict weight reduction strategy is designed to improve the detection accuracy.

This application relates to the field of data processing, and in particular to a method, apparatus, electronic device, and storage medium for determining cardiac age. The method includes: updating and iterating an initial cardiac health encoder to obtain a first determining model, with the objective of minimizing the distance between cardiac health representation vectors of the same cardiac health state output by an initial cardiac health encoder and maximizing the distance between cardiac health representation vectors of different cardiac health states; updating and iterating the first cardiac health encoder based on cardiac fusion feature sample vectors, their cardiac activity states, and relative cardiac age to obtain a second determining model; and updating and iterating an age prediction module using cardiac fusion feature sample vectors as samples and cardiac biological age as a label to obtain a target determining model, thereby determining cardiac age. This application enables the determination model, once trained, to accurately and efficiently determine cardiac age.

The invention discloses a patrol process doubtful point screening method based on artificial intelligence, and the method comprises the following steps: S1, obtaining process data during patrol, and carrying out the preprocessing; s2, constructing a heterogeneous graph model, and generating a node feature vector set and an edge connection relation matrix; s3, introducing a double attention mechanism to form a feature weight set; s4, performing graph convolution operation on the feature weight set, and fusing to generate doubtful point feature vectors; s5, constructing a comparative learning network, and calculating the similarity between the comparative learning network and a known sample vector; s6, performing density clustering and time sequence behavior modeling on the feature vectors in the doubtful point candidate set, and constructing a doubtful point behavior chain graph; and S7, performing confidence scoring and risk sorting, and outputting a doubtful point screening result set. According to the method, intelligent modeling and efficient screening of multi-source heterogeneous data in the patrol process can be realized, and the accuracy of doubtful point identification and the automation level of risk judgment are remarkably improved.

A method for selecting ground motions considering spectral shape and duration parameters includes the following steps: S1, performing probabilistic seismic hazard analysis on the target site to determine the target peak ground acceleration (PGA) value at a specified exceedance probability level; S2, extracting the spectral shape parameter vector and duration parameter for each actual ground motion record based on a selected candidate strong earthquake database; S3, constructing a conditional statistical distribution model; S4, determining the desired target parameter sample set; S5, linearly scaling the actual ground motion records in the candidate strong earthquake database to the PGA target value determined in S1, using each sample vector in the desired target parameter sample set determined in S4 as a matching target, selecting the actual ground motion record with the smallest mismatch, and forming a final ground motion set consistent with the target hazard level. This invention can more accurately evaluate the seismic performance of structures.

The application discloses a multi-sample comparison and fusion multi-type vulnerability detection method and system, and the method comprises the following steps: processing an original data set to generate a data set of multiple vulnerability types; performing a single code symbolization operation on samples in the data set to obtain symbolized samples; constructing a special word vector model for a specific program language, performing a vectorization operation on the symbolized samples to obtain sample vectors; constructing a same-type sample fusion matrix and a different-type sample comparison matrix for the sample vectors respectively, and generating a same-type sample fusion vector and a different-type sample comparison vector; training a deep learning model by using the same-type sample fusion vector and the different-type sample comparison vector, and performing vulnerability detection by using the trained deep learning model. The application can reduce the complexity of the model, and improve the vulnerability identification capability and accuracy of the model.