62 results about "Pattern vector" patented technology

The invention relates to the technical field of video behavior recognition, and discloses a behavior recognition and early warning system for a real-time video stream. The system comprises a spatio-temporal feature modeling module, a behavior fragment extraction module, an anomaly propagation modeling module, a risk area positioning module and an early warning strategy generation module. The spatial-temporal feature modeling module builds a dynamic model based on historical data, captures a skeleton key point three-dimensional coordinate sequence, a motion optical flow vector field and a micro-expression intensity spectrum, and outputs a theoretical behavior mode vector; the behavior fragment extraction module generates a multi-modal difference feature tensor through cross-modal difference analysis; the exception propagation modeling module generates an exception propagation path risk probability distribution cloud picture in combination with spatial constraint and trajectory information; the risk area positioning module identifies a high-risk area and marks a boundary; and the early warning strategy generation module dynamically configures monitoring parameters, starts high-frame-rate micro-expression capture for a high-risk area, and performs a track disturbance test on an adjacent area.

The invention discloses a love and marriage matching system based on hierarchical AI expert decision, relates to the technical field of hierarchical matching, and is used for solving the problems that hierarchical management and a refined hierarchical pushing mechanism are lacked, various recommendation targets are difficult to consider, and recommendation accuracy is reduced. Extracting a user pair feature vector, calling a plurality of field expert agents to generate a multi-dimensional score vector, performing similarity comparison on the feature vector and a success mode vector in the dynamic matching mode knowledge base, calculating a matching degree score, and synthesizing the matching degree score and the multi-dimensional score vector to obtain a matching recommendation level. And dividing the user pairs into corresponding recommendation pools according to the matching recommendation levels, and outputting a recommendation list according to a display strategy, so that multi-stage, layered and dynamic management of love and marriage matching is realized, the matching success rate and the interaction satisfaction degree are improved, and a layered recommendation decision-making mechanism based on multi-source expert knowledge driving is realized.

The invention relates to a real-time interaction violation detection method, system and device and a medium. The method comprises the following steps: acquiring an online interaction session original information flow containing a user text sequence, a voice signal, an image file and an interaction behavior timestamp; extracting text semantic vectors, voice acoustic features and image visual content description, and integrating to generate an initial feature vector set; based on the interaction behavior timestamps, constructing an interaction time sequence diagram by taking the initial vectors as nodes, calculating multi-modal association weights among the nodes and updating connection edges to obtain a multi-modal fusion diagram; and inputting the fused graph into a graph neural network, outputting a global graph embedded vector through message passing and node aggregation, matching a preset violation mode vector library to calculate a similarity score, determining a violation type, and generating a risk assessment conclusion containing the violation type and confidence. According to the method, cross-statement and cross-modal context violation association is effectively captured, violation judgment accuracy is improved, and an intervention basis is provided for a platform.

The invention discloses a Thangka generation method based on a structure and pattern two-channel constraint diffusion model, and belongs to the technical field of Thangka image generation, and the method comprises the steps: obtaining a Thangka image, and carrying out the structure labeling and pattern labeling of the Thangka image; constructing a structure vector and a pattern vector; constructing a dual-channel constraint diffusion model; constructing a joint loss function, and training the dual-channel constraint diffusion model; in the training and production process, a structure-pattern adaptive coupling mechanism is introduced, and a structure constraint weight and a pattern constraint weight are dynamically adjusted; and executing a back diffusion generation process, and generating a target Thangka image through iterative denoising under double constraints of a structure vector and a pattern vector. According to the method, the structure normalization, the pattern definition and the overall style uniformity of the generated Thangka image are remarkably improved, and the method can be used for Thangka digital protection, virtual restoration, digital recarving, cultural education, auxiliary drawing and intelligent content generation in the literature and blog industry.

The application discloses a mold control equipment operation state monitoring method and device and a medium, relates to the technical field of intelligent equipment monitoring, and comprises the following steps: carrying out denoising, normalization and outlier rejection processing on a mold operation data set to obtain clean operation data; performing correlation analysis and feature extraction on the clean operation data to generate a sound-vibration coupling mode vector set, and performing feature index calculation and mode deviation evaluation to obtain a sound-vibration state feature set; performing state coding on the sound-vibration state feature set to generate a mold operation state table, and performing health degree calculation and state evaluation to output a mold operation feature set; performing operation level judgment and trend analysis on the mold operation feature set to output an operation state result; and performing information arrangement and instruction generation on the operation state result to generate an operation state report. The mold operation state is quantitatively monitored with high precision.

The invention discloses a mold control equipment operation state monitoring method, equipment and a medium, and relates to the technical field of intelligent equipment monitoring, and the method comprises the steps: carrying out the denoising, normalization and abnormal value elimination processing of a mold operation data set, and obtaining clean operation data; performing correlation analysis and feature extraction on the clean operation data to generate a sound-vibration coupling mode vector set, and obtaining a sound-vibration state feature set through feature index calculation and mode deviation evaluation; state coding is carried out on the sound vibration state feature set, a mold operation state table is generated, health degree calculation and state evaluation are carried out, and a mold operation feature set is output; performing operation grade judgment and trend analysis on the mold operation feature set, and outputting an operation state result; and performing information arrangement and instruction generation on the operation state result to generate an operation state report. And high-precision quantitative monitoring of the operation state of the mold is realized.

The invention relates to the technical field of battery management, and discloses a battery fault diagnosis method and system for a CCS module, and the method comprises the steps: obtaining a multi-dimensional data sequence of the voltage, current and temperature of a battery pack, carrying out the abnormality detection, and obtaining a potential abnormal time period and a multi-dimensional data subset; performing clustering analysis on the multi-dimensional data subset to obtain an abnormal feature clustering center, and performing dynamic association analysis to obtain an association mode vector set; calculating the similarity of each single data and a fault propagation path, and matching the propagation path with a preset mode to obtain a fault propagation score; fusing the clustering center and the propagation score, positioning a fault monomer through a support vector machine regression model, and determining a fault monomer identifier; and carrying out segmented clustering and logic judgment on the data corresponding to the identifier, and determining a fault type. According to the invention, accurate positioning and type determination of fault monomers can be realized.

The invention provides a metro train dispatching optimization method based on a Markov decision process model. According to the method, a Markov decision model taking a state-action pair as a core is constructed, a train stop mode vector is introduced to accurately represent different jump stop schemes, and the method adapts to typical working conditions such as large and small intersections and non-stop passing. Meanwhile, simulation output data of the four typical parking modes are used for establishing a train running time query table, and the calculation efficiency of the train running time under the variable parking strategy is greatly improved. According to the method, a dual deep Q network algorithm is adopted, an'optimal / worst experience pool 'mechanism is innovatively introduced, and key state transition data which are optimal and worst in the training process are stored respectively, so that the learning capability of high-quality experience and the avoidance capability of low-quality strategies are enhanced. Compared with a traditional single-working-condition optimization scheme, the method shows higher comprehensive performance and practical application potential in a multi-working-condition fusion scheduling task.

The invention relates to the technical field of fire-fighting informatization, in particular to a fire-fighting remote control device based on the Internet of Things technology, and the device comprises a distributed abnormal signal monitoring module which collects non-alarm data such as communication abnormality, response timeout and voltage fluctuation; the abnormal feature decoding and risk identification module extracts features such as signal drift and delay vectors and generates risk mode vectors; the equipment layer stability prediction module predicts the stability level and the fault type in the future 30 minutes in combination with historical data and environmental parameters; the linkage consistency evaluation module aggregates prediction results of all the nodes and identifies potential consistency risks under a communication link or a power bus; and the intelligent disposal strategy generation module outputs a disposal suggestion package based on the consistency report. According to the invention, early recognition, trend prediction and cross-node linkage disposal of the fire control system on non-alarm faults are realized, and the system stability and the remote control intelligence level are significantly improved.

The invention discloses a multi-site collaborative energy storage centralized control scheduling method, which relates to the technical field of multi-site collaborative energy storage, and comprises the following steps of: after an ambiguity condition of a scheduling strategy is identified, constructing a response mode vector based on a site historical scheduling behavior, and utilizing a response prediction model fusing a multi-head attention mechanism and a behavior trajectory transformation network to obtain a multi-site collaborative energy storage centralized control scheduling strategy; determining a site response behavior difference under the condition that the scheduling strategy has ambiguity; according to a site response behavior difference result, a strategy analysis prompt instruction is generated through a behavior deviation value mapping mode, the strategy analysis prompt instruction and an original scheduling strategy are combined to construct a structured scheduling strategy with a consistent analysis path, and unified analysis and constraint control of scheduling strategy execution logic are achieved. According to the method, the problem of multi-site response behavior difference caused by an ambiguous scheduling strategy is solved, unified analysis and closed-loop control of the scheduling strategy are realized, and the consistency and stability of cooperative scheduling are guaranteed.

The invention relates to the technical field of fault early warning, and discloses a CFB boiler operation safety domain fault early warning method based on multi-source data fusion, and the method comprises the steps: obtaining CFB boiler multi-source monitoring data, and generating a multi-dimensional spatio-temporal data matrix; analyzing the multi-dimensional spatio-temporal data matrix by using multi-resolution wavelet transform to generate a multi-scale feature tensor; analyzing the multi-scale feature tensor based on a tensor network, calculating a cross-scale coupling weight matrix according to a physical constraint equation, and extracting a cross-scale fault association mode vector through tensor contraction operation; performing similarity matching on the cross-scale fault association mode vector and a fault semantic prototype library to generate hierarchical fault situation information including a summary layer, a detail layer and a depth analysis layer; outputting a self-adaptive layered fault situation awareness result; according to the method, intelligent self-adaptive sensing of complex multi-scale fault situations is realized, the integrity of fault information is ensured, and the man-machine interaction efficiency is improved.

This invention relates to the field of bioinformatics processing and medical laboratory data analysis, specifically a method for classifying hepatitis B antibody patterns based on plasma protein profiles. The method includes: acquiring host hardware information of the execution environment and extracting central processing unit (CPU) cache parameters; acquiring a high-dimensional sparse one-dimensional array and extracting non-zero feature indices; performing mapping calculations using a locality-sensitive hashing (LSH) algorithm to reconstruct the data into a locally dense two-dimensional matrix; dynamically segmenting the data into independent sub-blocks according to cache parameters and initial segmentation dimensions; performing low-rank tensor decomposition on the independent sub-blocks to extract local latent feature vectors; concatenating the sub-blocks, weighting them through a single-layer attention network, and inputting the result into a classifier function to output a target classification pattern vector; and generating dimension update instructions based on a preset dynamic adjustment mechanism to adjust subsequent segmentation dimensions. This invention achieves lower memory peaks, higher cache hit rates, and stable multi-label pattern output capabilities.

A system and method for improving identity resolution match rates using deep learning techniques that enable privacy-preserving fuzzy matching of personally identifiable information (PII) employs a deep learning model trained with transformer architecture and contrastive learning on third-party identity graph data. Custom tokenizers process different Pll types by leveraging hierarchical structures and domain-specific characteristics. The trained model generates vector embeddings that enable fuzzy matching to account for variations in spellings, typographical errors, and data inconsistencies without requiring adherence to strict data schemas. A vector database stores embeddings for nearest neighbor searches to identify potential identity matches based on distance calculations between embeddings. A match filter applies logic to determine valid matches, and the system can operate without requiring movement of raw Pll data from secure environments.

The application discloses a Thangka generation method based on a structure and pattern double-channel constraint diffusion model, and belongs to the technical field of Thangka image generation, and comprises the following steps: acquiring a Thangka image, and performing structure labeling and pattern labeling on the Thangka image; constructing a structure vector and a pattern vector; constructing a double-channel constraint diffusion model; constructing a joint loss function, and training the double-channel constraint diffusion model; in the training and production process, introducing a structure-pattern adaptive coupling mechanism, and dynamically adjusting a structure constraint weight and a pattern constraint weight; and performing a reverse diffusion generation process, and generating a target Thangka image through iterative denoising under the double constraints of the structure vector and the pattern vector. The application significantly improves the structure standardization, pattern clarity and overall style uniformity of the generated Thangka image, and can be used for Thangka digital protection, virtual restoration, digital replication, cultural education, auxiliary drawing and intelligent content generation in the cultural and museum industry.

The invention provides a law case recommendation method and system based on a large model, and relates to the technical field of intellectual property, and the method comprises the steps: obtaining document data of an intellectual property case, a user operation sequence corresponding to a to-be-analyzed technical scheme, and law text information of a to-be-queried case, and constructing a technical knowledge graph based on the document data; inputting the technical knowledge graph and the user operation sequence into a time sequence graph neural network, and generating a mode vector and weight information; generating technical association data based on the mode vector and the weight information, and processing the technical association data by using a variational auto-encoder to obtain enhanced technical feature data; and finally, inputting the enhanced technical feature data and the legal text information into a large legal model, generating a joint query vector, and performing case retrieval to obtain a class case recommendation result. According to the method, the association accuracy of the class case recommendation result and the technical facts is improved.

The invention provides a mobile printing robot based on Mecanum wheels. The mobile printing robot aims at solving the problems that an existing automatic printing device is limited in printing breadth, poor in moving flexibility and low in complex path printing precision. The robot comprises a robot body, Mecanum wheels, an ink-jet printing assembly, a position sensor, an AI trajectory planning module and a linkage control module. The four Mecanum wheels at the bottom of the robot body are arranged in a four-wheel rectangle mode and are connected with independent driving motors correspondingly, and omnidirectional movement is achieved. And the ink jet head is vertically aligned with the center of motion of the Mecanum wheel. The AI trajectory planning module disassembles a path based on target pattern vector data, plans movement parameters by combining with equipment parameters and performs real-time correction according to feedback of a position sensor; and the linkage control module realizes linkage of the AI algorithm, the driving motor and the ink jet head. The method breaks through the limitation of the printing breadth, effectively reduces the printing error of a complex path, greatly improves the printing efficiency, and is suitable for automatic printing of scenes such as large-area paper surfaces, grounds and wall surfaces.

The application relates to the technical field of fault early warning and discloses a CFB boiler operation safety domain fault early warning method based on multi-source data fusion, which comprises the following steps: acquiring multi-source monitoring data of a CFB boiler to generate a multi-dimensional space-time data matrix; analyzing the multi-dimensional space-time data matrix by using multi-resolution wavelet transform to generate a multi-scale feature tensor; analyzing the multi-scale feature tensor based on a tensor network, calculating a cross-scale coupling weight matrix according to a physical constraint equation, and extracting a cross-scale fault correlation mode vector through tensor contraction operation; carrying out similarity matching on the cross-scale fault correlation mode vector and a fault semantic prototype library to generate layered fault situation information comprising a summary layer, a detail layer and a deep analysis layer; and outputting an adaptive layered fault situation awareness result; the application realizes intelligent adaptive perception of a complex multi-scale fault situation, guarantees the integrity of fault information, and improves the efficiency of man-machine interaction.

The invention relates to the technical field of computer data processing, and discloses a dynamic market research sample intelligent matching method based on multi-source data fusion, which comprises the following steps: performing sentiment analysis processing on social media data to obtain a sentiment feature vector, performing activeness analysis processing on player behavior data to obtain a behavior pattern vector, and matching the behavior pattern vector with the sentiment feature vector; the method comprises the following steps: performing purchase frequency analysis processing on a game purchase record to obtain consumption behavior vectors, uniformly mapping three types of feature vectors to the same dimension space, performing normalization and vector alignment fusion, constructing an initial fusion matrix, and further generating a fusion vector diagram, so that the relation between different data sources under a behavior tag index is subjected to structured expression, and the fusion efficiency is improved. Guiding weighted dynamic fusion processing through a fusion vector diagram, constructing a fusion feature matrix, and inputting the fusion feature matrix and the three types of original data into a deep neural network model to obtain a survey sample data set.

The invention discloses an electromechanical coupling system fault diagnosis method and system based on a multi-level vector library, and belongs to the technical field of fault diagnosis. The electromechanical coupling system is sequentially divided into a plurality of hierarchies according to functional composition of the electromechanical coupling system, each hierarchy comprises a plurality of units with different functions, and the method comprises the following steps: converting a fault mode into a fault mode vector x through an embedded model; the adjacent agent An evaluates and awards the fault reason of the corresponding unit retrieved by the agent An-1 from the PHM multi-level vector library, and retrieves the fault reason of the unit in the PHM multi-level vector library according to the fault mode vector x and the fault reason; and fusing the N fault causes of the intelligent body search through a fusion device. According to the invention, the fault part of the electromechanical coupling system can be accurately positioned.

The invention relates to the technical field of energy storage equipment monitoring, and discloses a fault prediction and early warning system for an energy storage box-type substation. The multi-modal data acquisition and preprocessing module, the macroscopic operation mode vector extraction module, the real-time causal calculation engine module, the real-time dynamic causal entropy map construction module and the working condition causal baseline dynamic mapping module generate a theoretical health baseline map corresponding to the current working condition in real time according to the macroscopic operation mode vector; an atlas topology difference and anomaly detection module generates a difference atlas by comparing the real-time dynamic causal entropy atlas with the theoretical health baseline atlas; the topology fingerprint and fault diagnosis module analyzes the difference atlas to extract topology fingerprint vectors and matches a fault knowledge base to output a diagnosis conclusion. By constructing the health base line which is dynamically adjusted along with the working condition, normal operation fluctuation can be effectively filtered out, the accuracy of fault detection is improved, and diagnosis from abnormal detection to specific fault attribution is realized.

The application discloses an AGV digital twin monitoring method based on a three-dimensional simulation model, relates to the technical field of digital twin monitoring, and comprises the following steps: performing multi-dimensional feature extraction on a target AGV to form an AGV behavior mode vector set; generating a nonlinear coupling matrix representing the correlation strength between AGV motion state parameters and environmental constraint parameters based on the vector set and environmental geometric information; dividing a three-dimensional digital twin model into a plurality of simulation units, and independently simulating local AGV disturbance responses of the units according to the matrix; mapping high-frequency motion events and low-frequency task events into the simulation units based on the local AGV disturbance responses to form a nonlinear disturbance coupling field; and dynamically adjusting a virtual AGV state based on the disturbance propagation trend of the disturbance coupling field. The application realizes nonlinear behavior synchronization of a virtual AGV and a physical AGV, thereby solving the problem that virtual and real behaviors of an AGV are difficult to be consistent in the prior art.

The invention relates to a method for recognizing an object (2) in a search image (1), comprising the following steps: a) providing a pattern vector (5) describing the object (2) by means of coordinates of characteristic pixels (7); b) moving the pattern vector (5) over different positions of the search image (1); c) determining a respective success value (15) at the different positions; and d) recognizing the object (2) at the positions depending on the success values (15); wherein a first direction (8) and a second direction (9) different from the first direction (8) are associated with each characteristic pixel (7), wherein a first total intensity value (10) of a first number (11) of one-dimensionally arranged pixels in the first direction (8) and a second total intensity value (12) of a second number (13) of one-dimensionally arranged pixels in the second direction (9) are determined, respectively, wherein a difference value (14) between the first total intensity value (10) and the second total intensity value (12) is determined, respectively, wherein the success value (15) is determined depending on the respective difference value (14).

The invention provides a distributed data storage calling method and system supporting an Internet of Things device, and relates to the technical field of data storage calling, and the method comprises the steps: executing local data collection and blocking processing at a node side of the Internet of Things device; performing node self-inspection of the Internet of Things equipment; identifying a candidate node subset corresponding to the target data block based on the communication mode vector and a distributed storage-call collaborative graph constructed by the block data, and establishing an access path sequence based on the candidate node subset, the mapping edge weight and the transient link index; and carrying out hybrid multi-state communication matching on self-adaptive block data fusion results returned by different access paths and superior equipment, and carrying out calling management by utilizing a matching result. According to the method and the device, the technical problem of unbalanced node resource utilization rate caused by lack of dynamic adaptability of a distributed data storage calling mechanism in the prior art can be solved, and the technical effect of improving the node resource utilization efficiency is achieved.

This invention discloses a multi-behavior pattern collaborative training method for semi-supervised semantic segmentation training, comprising: learning models R1 and R2 using labeled images and generating pseudo-labels for each subnet from unlabeled data; adding a behavior pattern vector extraction module to each model; introducing a new data stream for each subnet and applying different feature perturbations to the data stream; and integrating the data streams to obtain the total loss function for each model during training. This invention improves segmentation performance by enhancing the diversity of subnets in the collaborative training framework.

Embodiments of the present disclosure provide a model training method and device of a question and answer system, an electronic device, and a storage medium, wherein the model training method of the question and answer system comprises: obtaining a question text sample; inputting the question text sample into an initial question analysis model for iterative training to obtain a question analysis model; the initial question analysis model comprises a first encoding layer and a conversion layer; the first encoding layer is configured to perform encoding processing according to the question text sample to obtain a corresponding first sentence pattern vector; the conversion layer is configured to generate a preset number of initial intent vectors under the condition of receiving the first sentence pattern vector, perform filling processing on each initial intent vector according to the first sentence pattern vector to obtain a corresponding target intent vector, and convert the target intent vector into a corresponding text segment; and the text segment is configured to query an answer corresponding to the question text sample in the question and answer system, thereby improving the intent recognition capability of the question and answer system.

This provides a method for comparing die systems and alignment vectors. [Solution] The die system 100 includes a plurality of dies 101 arranged in a desired pattern. Alignment vectors, such as die vectors, can be determined from the edge features of the dies. Alignment vectors can be compared with other dies or die patterns 111 in the same system. A method for comparing dies and die patterns includes comparing die vectors and / or pattern vectors. By comparing the alignment vectors, the die pattern can be modified for the next processing round. The provided method enables accurate comparison between deposited edge features so that precise die splicing can be achieved.

This invention belongs to the fields of data governance and artificial intelligence technology, and discloses an intelligent data cataloging method and system for cross-modal fusion and continuous learning guided by heterogeneous data. The method encodes structured metadata, unstructured text, and unstructured images into structured pattern vectors, text content vectors, and image visual vectors, respectively; it performs cross-modal attention fusion using the structured pattern vector as the query vector and the combined vectors formed by concatenating the other two along the feature dimensions as key-value vectors; it classifies the fused semantic tensor through a classification model; when learning new categories, it constructs a parameter update offset penalty term based on the diagonal elements of the Fisher information matrix of each parameter to update the parameter offset; and it uses the classification results as anchors to determine the business subject domain, sensitivity level, and field-level data lineage to form multi-dimensional cataloging information. This invention improves cross-modal semantic alignment accuracy, maintains the performance of old categories, and reduces manual intervention in multi-dimensional cataloging.

The present disclosure relates to a pattern type recognition method and device, electronic equipment and readable storage medium. The method comprises: obtaining an initial image to be recognized, the initial image comprising a pattern to be recognized; obtaining a feature vector of the pattern to be recognized in the initial image based on a preset pattern vector extraction model; obtaining a similarity between the feature vector and each candidate feature vector in a preset database to obtain at least one target candidate feature vector; and determining a type of the pattern to be recognized according to a pattern type of the at least one target candidate feature vector. In this embodiment, the problem that an existing pattern recognition model needs to be retrained for recognition when the pattern to be recognized is a new type can be solved without retraining the model when the pattern to be recognized is a new type, which is beneficial to improving the use efficiency of pattern type recognition.

The invention provides a transformer substation flood prevention material configuration method and device, and the method comprises the steps: obtaining a predefined triple mode which is a head entity type-relation type-tail entity type; the method comprises the following steps: acquiring multi-source static data of a transformer substation, and extracting triple data of the multi-source static data by taking a triple mode as a constraint so as to construct a static flood database; obtaining dynamic meteorological data, constructing a meteorological mode vector based on the dynamic meteorological data and the static disaster resistance level in the multi-source static data, and embedding the meteorological mode vector into the static flood database to obtain a dynamic flood database; and performing scene matching and reasoning decision based on the dynamic flood database, and generating flood prevention material configuration for the target substation. According to the invention, the fundamental transformation from a static plan to a dynamic context awareness decision is realized. In other words, the adaptability of the flood prevention material allocation decision process and the real-time scene is improved.

Computerized methods and systems of de novo sequencing from a mass spectrometer and identifying a biological polymer using mass invariant charge patterns in the spectrometer data by transforming spectra to a natural logarithmic space where peaks arising from the same analyte mass align along a predictable pattern defined solely by charge state. In some embodiments, the computerized method employs an operation that iterates the residue mass in the transformed natural logarithmic space, e.g., minimizing charge state difference errors between corresponding isotopologues assigned to different charge states. In some embodiments, the de novo sequencing of the present disclosure also allows for viewing the mass-to-charge (m / z) spectrum in a natural logarithmic manner (e.g., Equation 1—ln(m / z−q)) to provide confidence in any reassignment of peaks in an observed charge pattern vector.