43 results about "Propagation matrix" patented technology

The invention is suitable for the technical field of optical design parameter optimization, and provides a design parameter optimization method and system for a long-optical-path gas absorption cell, and the method comprises the following steps: carrying out the theoretical verification of the feasibility of the lens parameters of a spherical reflector based on a theoretical model of a light propagation matrix, and obtaining the design parameters of the spherical reflector; determining an initial lens parameter and a key design parameter range; an optical cavity model composed of two spherical reflectors is established, and automatic ray tracing simulation is carried out according to the initial lens parameters, the key design parameter range and a preset stepping value; the simulation data are verified, the number of light spots is determined, and the optimal key design parameter combination enabling the optical path to be maximized is screened out according to the number of the light spots. The method is mainly used for designing a long-optical-path gas absorption cell in infrared absorption spectrum trace gas sensor detection, has practical significance, realizes verification of optimal design parameters, reduces development difficulty and improves development efficiency.

The invention discloses a robust robust sequence extraterrestrial celestial body terrain reference surface fitting method and system, and relates to the technical field of terrain reference surface fitting, and the method comprises the steps: collecting three-dimensional laser point cloud data to construct a reference plane geometric model, and building a joint error model of an observation value error and a design matrix error; adopting a sequence PEIV total least square framework to carry out grouping iteration adjustment on the point cloud data, and recursively updating parameter estimation by utilizing an error propagation matrix and a matrix inversion formula; introducing a two-factor robust mechanism, dynamically calculating a weight according to a residual error and estimating a robust scale; and circularly optimizing the parameters through an iterative strategy until a convergence condition is met, and outputting reference plane parameters. According to the method, different terrain complexity, data scales and error distribution scenes can be covered, the adaptability and engineering practicability of terrain datum plane fitting are remarkably improved, and reliable geometric reference is provided for subsequent obstacle detection and safe area decision making.

The invention discloses a water transfer project risk cascade propagation and risk assessment method and system based on a multi-level topological structure, and the method comprises the steps: obtaining a water transfer project topological network, and generating a node comprehensive risk vector of each node based on multi-source operation data; constructing a basic risk propagation matrix according to the physical connection relationship, and performing structured modulation on the basic risk propagation matrix based on the node comprehensive risk vector to generate a comprehensive propagation operator; performing cascade propagation iterative evolution on the node comprehensive risk vector based on the comprehensive propagation operator, and continuously calculating network dynamics stability characteristic parameters of the comprehensive propagation operator in the evolution process; when it is judged that the system is in a divergent phase state according to the parameters, a comprehensive propagation operator is reconstructed, iterative evolution continues till convergence, and steady-state risk distribution is obtained; and outputting a comprehensive risk assessment result of the system based on the steady-state risk distribution. The method can improve the operation safety management and control capability of the water regulation project under complex operation conditions and external disturbance conditions.

This invention belongs to the field of building energy consumption monitoring technology and provides an anomaly detection method based on ARIMA residuals. It utilizes an ARIMA model to extract the residual sequence after the linear trend of data, learns normal patterns using LSTM-AE, and identifies anomalies. Simultaneously, it combines correlation coefficient matrices and anomaly propagation matrices to analyze fault propagation paths, featuring high detection accuracy and strong interpretability. This invention improves anomaly detection accuracy and effectively reduces false alarm and false negative rates. It also effectively handles sparse data, making it suitable for scenarios such as building energy consumption monitoring. By constructing correlation coefficient matrices and anomaly propagation matrices, it analyzes the propagation paths and impact mechanisms of anomalies in multi-parameter systems, providing a reliable basis for anomaly tracing and system optimization. Through practical application, this invention helps enterprises promptly identify and resolve energy waste problems, save energy costs, optimize energy management systems, and improve energy utilization efficiency.

The invention belongs to the field of power system automation and information security, and discloses a robust observer-based power distribution network control input attack detection and state estimation method, which comprises the following steps of: constructing a power distribution network linear discrete dynamic model based on Droop control, and improving the observability of a system by utilizing a multi-step observation structure and differential modeling; based on a disturbance propagation matrix and a null space projection technology, an attack detection problem is converted into a sparse optimization problem; designing an unknown input observer (UIO), and combining a dynamic compensation matrix and feedback gain to enhance robustness; and an online feedback mechanism is introduced to realize parameter adaptive updating. The method can realize the real-time identification of the control instruction tampering attack and the high-precision estimation of the system state, and meets the high requirements of the operation of the power distribution network on the safety, the real-time performance and the reliability.

The application belongs to the field of power system automation and information security, and discloses a distribution network control input attack detection and state estimation method based on a robust observer, which constructs a linear discrete dynamic model of a distribution network based on Droop control, improves the observability of the system by using a multi-step observation structure and differential modeling, converts the attack detection problem into a sparse optimization problem based on a disturbance propagation matrix and zero space projection technology, designs an unknown input observer (UIO) combined with a dynamic compensation matrix and a feedback gain to enhance robustness, and introduces an online feedback mechanism to realize parameter adaptive updating. The method can realize real-time identification of control instruction tampering attacks and high-precision estimation of the system state, and meets the high requirements of the distribution network operation on safety, real-time performance and reliability.

The invention discloses an electromagnetic pulse propagation calculation method based on a generalized propagation matrix and a transfer function method, and the method comprises the steps: determining the explosion point height and the receiving point height of an electromagnetic pulse, and constructing an ionosphere parameter on a propagation path; calculating a wave vector of the position of an electromagnetic pulse explosion point, and calculating a reflection coefficient matrix and a transmission coefficient matrix of the electromagnetic pulse which upwards penetrates through the ionized layer and downwards radiates to the ground through a generalized propagation matrix method; defining the type of the electromagnetic pulse, and calculating the amplitude of the superposed uplink and downlink electromagnetic waves at the explosion point position through the reflection coefficient matrix; in combination with a generalized propagation matrix method and a transfer function method, calculating the frequency spectrum and time domain waveform of the electromagnetic wave penetrating through the ionosphere at the explosion point position through inverse Fourier transform; time-frequency characteristic analysis is carried out according to the frequency spectrum and the time-domain waveform of the electromagnetic waves penetrating through the ionosphere, a time-frequency graph is obtained, the experiment cost is remarkably reduced, and efficient and low-cost theoretical support is provided for propagation of the electromagnetic pulses in the ionosphere.

The invention relates to a federal recommendation method based on a sharable space diagram and a hypergraph diffusion model. The federal recommendation method comprises the following steps: a client constructs a personalized hypergraph and a hypergraph propagation matrix by using local data; enhancing user characterization by means of a diffusion model and then performing multi-dimensional characterization fusion; then collaborating with global geographic knowledge provided by a server, and constructing a loss function to update a recommendation model; clustering the clients, and carrying out data self-adaptive aggregation and retraining on edge ends corresponding to each category after clustering; then the server performs weighted aggregation on the trained training parameter set on each edge end again; and repeatedly iterating for multiple rounds to obtain a final optimal client recommendation model. According to the method, the accuracy of federal POI recommendation can be remarkably improved on the premise of strictly protecting user data privacy.

The invention relates to a microservice system abnormal root cause positioning method based on a space-time diagram convolutional neural network and a thermal diffusion model, and the method mainly comprises the following steps: S1, collecting the performance index data of a service node and calling metadata through Prometheus, carrying out the anomaly detection through a sliding time window mechanism, and outputting an abnormal node and a time window; s2, constructing a node cascade propagation feature vector, and obtaining an abnormal time window service call directed weighted dependency graph in combination with a GAT mechanism; s3, constructing a space-time diagram convolutional neural network, extracting space-time features of node anomalies, and depicting an anomaly cascade propagation effect by adopting a thermal diffusion mechanism to obtain space-time diffusion features; and S4, calculating an abnormal score based on the node cascade propagation characteristics and the space-time diffusion characteristics, constructing a propagation matrix by adopting PageRank, and outputting an abnormal root cause node list. According to the method, the abnormal spatial-temporal characteristics and cascade propagation effects of the micro-service nodes are modeled, so that the root cause positioning accuracy in a complex service scene is effectively improved.

The invention provides a script analysis method and device, electronic equipment and a storage medium. The method comprises the steps of obtaining a to-be-processed script text; a thermodynamic map is constructed according to the to-be-processed script text, and the thermodynamic map comprises a character node set used for representing all character entities in a script; the heat propagation matrix is used for recording a heat propagation value between every two figure nodes; determining core character nodes and non-core character nodes from the character node set; for any two non-core character nodes in the thermodynamic map, when the thermodynamic propagation values between the two non-core character nodes and all core character nodes are lower than a first threshold value and the thermodynamic propagation value between the two non-core character nodes is higher than a second threshold value, the thermodynamic map is determined to be the thermodynamic map; and determining a plot line corresponding to the two non-core character nodes as a water injection branch line. Therefore, accurate identification of the water injection branch line is realized.

The invention provides a fault propagation analysis method fusing a rotating mechanical system fault mode library and a causal chain explanatory structure model, and belongs to the field of fault diagnosis and reliability engineering, and the method comprises the steps: constructing a fault propagation mode library, calculating a causal weight matrix, and constructing the causal chain explanatory structure model. Constructing a fault propagation mode library to extract the characteristics of the rotating mechanical system so as to reduce the error influence of physical connection characteristics on a causal weight matrix; a direct influence matrix, a forward propagation matrix and a reverse propagation matrix need to be calculated for calculating the causal weight matrix so as to help to identify a fault propagation path in the system; a causal chain explanatory structure model is constructed, a dynamic causal chain network is constructed, causal relationship strength in a fault propagation path is quantified, and a key propagation path and a feedback loop are identified, so that nonlinear interaction of multiple fault modes in the rotating machinery is more accurately modeled.

The invention discloses a multi-modal preference-driven graph convolution combinatorial optimization learning path generation method, which comprises the following steps: acquiring multi-modal learning feature data, and carrying out multi-modal feature data fusion processing to obtain a learning resource initial feature matrix; a multi-relation adjacent matrix is constructed, normalization processing is carried out, and a normalized propagation matrix is constructed; updating the convolution features of the graph, and obtaining a learning gain score; calculating a comprehensive utility based on the learning gain score; integer programming is established under the condition of considering multiple constraints, and optimized candidate resources are obtained through screening; an edge cost function and a learning path objective function are adopted to generate an optimal learning path in the optimized candidate resources; in the optimal learning path generation process, dynamic re-planning is adopted; the dynamic re-planning comprises grasp updating, coverage demand decreasing and duration budget updating. By means of the scheme, the method has the advantages of being simple in logic, reliable in multi-mode fusion and the like.

The invention relates to a contact network power supply high-voltage system insulation detection method, device and equipment and a medium. The method comprises the following steps: constructing a network topology model based on electrical connection data of a convergence area, and generating a topology matrix for describing structural characteristics; accurately simulating a test signal propagation path by using the matrix in combination with a time domain reflection method, and quantifying reflection and attenuation characteristics to form a signal propagation matrix; based on this, the insulation sensitivity of each component is calculated, and a high-priority monitoring area set is screened; iteratively generating a least observation point layout scheme meeting monitoring requirements through a coverage optimization algorithm; a virtual sensor is deployed, measurement data are dynamically processed by adopting a particle filtering algorithm, and the insulation degradation trend is tracked in real time; and finally, when an anomaly is detected, accurately determining a fault position through a triangulation positioning principle by using the time delay difference data of the multiple observation points. According to the technology, the fault positioning precision and the response speed in a dynamic environment are remarkably improved, so that the system resource utilization efficiency is optimized while the detection comprehensiveness is guaranteed.

This application discloses a method, apparatus, storage medium, and product for determining the distribution probability of oil and gas reservoirs based on seismic frequency-varying fluid factors, relating to the field of oil and gas exploration technology. The method includes: acquiring pre-stack data of the exploration area and performing pre-stack AVO inversion to obtain initial frequency-varying velocity vectors; constructing a posterior probability distribution of the objective function based on the convolution model and propagation matrix model in seismic wave propagation theory and Bayesian theory; obtaining the posterior probability of the objective function using the Markov chain Monte Carlo method; and determining the frequency-varying velocity vector at the maximum of the posterior probability density function; thereby determining the dispersion gradient of new fluid factors at different frequencies at the dominant seismic wave frequency to determine the distribution probability of oil and gas reservoirs; the dispersion gradient of the new fluid factors is positively correlated with the probability of the existence of oil and gas reservoirs; the parameters constituting the new fluid factors include: Poisson's ratio, P-wave velocity impedance and S-wave velocity impedance, Lamé coefficient, and density. This method can significantly improve the accuracy of oil and gas reservoir identification.

This invention provides a method for optimizing and constructing a human-machine collaborative full-process quality management system, comprising: real-time collection of multi-source heterogeneous industrial data from each process node in the production process to construct a quality data resource pool for human-machine collaboration; constructing a multi-process quality deviation coupling and propagation model based on a deviation propagation matrix, and introducing an expert experience correction mechanism to form a human-machine collaborative deviation propagation model, so as to trace and quantitatively characterize the propagation of quality deviations in the entire process chain; constructing an industrial digital twin for full-process quality management, generating collaborative optimization decision schemes through quality risk quantitative assessment and human-machine two-way decision fusion, and implementing dynamic adaptive control of process parameters throughout the entire process; and constructing a full-process quality feedback mechanism and a continuous iterative optimization mechanism. This invention aims to solve the technical problems in the prior art where quality prediction models rely on training data, expert experience is difficult to quantify and integrate, and the full-process deviation propagation mechanism is missing.

The application discloses a method and device for determining a lithologic trap boundary, electronic equipment and a storage medium. The method comprises the following steps: simulating a pseudo-well vertical lithologic combination based on a continuous-time Markov chain model, determining a pseudo-well lithofacies curve of different lithologic combinations according to a simulation result; determining a target elastic parameter through the pseudo-well lithofacies curve of the different lithologic combinations based on elastic parameter Monte Carlo random simulation with lithofacies constraints; determining pseudo-well simulation seismic data of the different lithologic combinations through the target elastic parameter based on a wave equation forward modeling method of a propagation matrix; and determining a lithologic trap boundary according to the pseudo-well simulation seismic data of the different lithologic combinations. The technical scheme of the application avoids the problems of strong subjectivity, large deviation of boundaries delineated by different personnel, and insufficient objective basis in the determination of a traditional lithologic trap boundary, thereby avoiding the situation of deviation of the delineated boundary caused by human factors.

The application discloses an intelligent experiment table multi-sensing data distributed storage method based on edge calculation and relates to the technical field of edge calculation, which comprises the following steps: generating a double-layer segment matrix based on continuous sampling data of an intelligent experiment table; obtaining a tail trace residual matrix based on the double-layer segment matrix; constructing a shift matrix based on the tail trace residual matrix and generating a tail trace adhesion-propagation matrix based on the shift matrix; calculating a tail trace cutting cost based on the tail trace adhesion-propagation matrix; calculating the fitness of each particle in a particle swarm based on the tail trace cutting cost; iteratively updating the speed and position of each particle in the particle swarm based on the fitness to determine an optimal distributed write matrix; and obtaining a distributed storage result based on the optimal distributed write matrix; and the application improves the experiment data recovery efficiency in a distributed storage environment.

The application is suitable for the technical field of optical design parameter optimization, and provides a design parameter optimization method and system for a long optical path gas absorption cell, which comprises the following steps: based on a theoretical model of a light propagation matrix, theoretically verifying the feasibility of mirror parameters of a spherical mirror, determining initial mirror parameters and a key design parameter range; establishing an optical cavity model composed of two spherical mirrors, and performing automatic light ray tracing simulation according to the initial mirror parameters, the key design parameter range and a preset step value; verifying simulation data, determining the number of light spots, and selecting an optimal key design parameter combination for maximizing the optical path according to the number of light spots. The method is mainly used for long optical path gas absorption cell design in an infrared absorption spectrum trace gas sensor detection, has practical significance, realizes verification of optimal design parameters, reduces development difficulty and improves development efficiency.

The invention discloses a method and system for testing and analyzing magneto-optical parameters of a magnetic multilayer film, and belongs to the technical field of photoelectron data analysis, and the method comprises the following steps: carrying out an ellipsometry test on a to-be-tested magnetic multilayer film sample to obtain ellipsometry parameters of the sample; constructing an optical model according to the structure of the magnetic multilayer film sample to be tested, and performing fitting calculation by combining an optical theory and the ellipsometric parameters obtained by testing to obtain optical property parameters of each layer of film in the sample; the method comprises the following steps: carrying out a magneto-optical ellipsometry test on a to-be-tested magnetic multilayer film sample to obtain a reflection coefficient of the sample in an unmagnetized state and a magneto-reflection coefficient caused by a Kerr effect, deriving a total propagation matrix of light in the magnetic multilayer film sample, and calculating to obtain a Jones matrix of the sample; and calculating the magneto-optical coupling coefficient of each layer of film according to the Jones matrix, the reflection coefficient, the magneto-reflection coefficient and the optical property parameter of each layer of film. According to the invention, accurate characterization of magneto-optical parameters of magnetic multilayer films with various thicknesses, various materials and various layer numbers can be realized.

The invention discloses a wafer graph defect knowledge graph evolution reasoning method based on structural causal intervention, and belongs to the technical field of semiconductor manufacturing defect analysis and knowledge reasoning. Constructing a three-layer directed weighted knowledge graph comprising preparation process nodes, crystal grain defect nodes and wafer graph defect nodes; calculating conditional probability edge weights according to historical statistical data to form a causal propagation matrix; a structural causal model is constructed, causal intervention is performed on the preparation process nodes, and a causal purification propagation matrix is generated; constructing a graph convolution evolution reasoning model based on the purification propagation matrix, and obtaining node embedding representation; introducing a time variable to carry out incremental updating on the propagation matrix to realize knowledge graph evolution; and finally outputting the prior probability distribution vector of the defect type of the wafer graph. According to the method, wafer graph defect prior distribution can be generated under the condition that new products are put into production or few samples exist, and knowledge constraints are provided for defect prediction and detection model generation.

The application provides a subway vibration Green function fast calculation method and system, relates to the subway vibration evaluation technical field, and includes the following steps: a near-field finite element model of a subway tunnel is constructed, and a plurality of infinite domain virtual vibration sources are arranged outside the near-field finite element model; a coupling equation between the near-field finite element model and the infinite domain virtual vibration source array is established, and the coupling equation is solved for each frequency point to obtain an infinite domain virtual vibration source amplitude value vector irrelevant to a layered site; based on physical parameters of each layer of a target layered site, a layered site propagation matrix from a virtual vibration source position to an arbitrary site position is constructed; the infinite domain virtual vibration source amplitude value vector is input into the layered site propagation matrix, vibration response of the target layered site at a corresponding frequency is calculated, and subway vibration Green function is obtained; and the application overcomes the dual imbalance of the two dimensions of "precision-efficiency" and "stability-practicality" by coupling an infinite domain basic solution with a layered site.

The application discloses a kind of suitable for deep brackish aquifer seismic reflection wave response high fidelity simulation calculation method, the method includes S10: according to brackish aquifer development characteristics, establish brackish aquifer pore fluid rock physics model, calculate the viscosity, density, longitudinal wave velocity and bulk modulus of brackish aquifer pore fluid;S20: according to brackish aquifer development characteristics, establish brackish aquifer multi-scale rock physics model, obtain brackish aquifer equivalent stiffness matrix;S30: establish geology-seismic forward modeling;S40: utilize propagation matrix method to calculate brackish aquifer reflection and transmission coefficient matrix;S50: calculate brackish aquifer reflection wave seismic response.The application combines seismic rock physics modeling and seismic forward simulation technology, considers the viscoelasticity characteristics and thickness of brackish aquifer, is more in line with actual geological conditions, low in cost, high in accuracy, can quickly and accurately calculate deep brackish aquifer seismic reflection wave response, provides theoretical support for the evaluation and optimization of carbon storage space.

The application discloses a horizontal array matching positioning method based on matrix filtering. A high-order parabolic equation model is adopted to calculate sound field propagation matrices at different steps, and a system propagation matrix in a sound source-receiving range is obtained through multiplication operation of the different step matrices; according to the depth of the sound source and the positions of the horizontal array elements, matrix coefficients are extracted from the system propagation matrices at different distances to form a new sparse system propagation matrix; subsequently, the sparse system propagation matrix is processed; a matrix filter set is formed; finally, the horizontal array receiving signal is processed by using the matrix filter set to form a distance-depth ambiguity diagram, and a target positioning result is obtained. The application can effectively improve the energy ratio of the main lobe and the side lobe in the depth dimension, reduce the number of high noise points in the whole two-dimensional ambiguity diagram, improve the depth and distance resolution of target positioning, and has important value for performance improvement of target positioning in an actual environment.

The invention relates to the technical field of digital twinning and disaster early warning, in particular to an engineering geological area disaster early warning method and system based on digital twinning. The method comprises the following steps: dividing structural units with homogeneous mechanical response by constructing a structural evolution type digital twinborn body, establishing a directional stress propagation matrix and a coupling dynamics evolution equation, and simulating the mechanical state and evolution process of a regional geologic body; a disturbance trigger chain propagation matrix is constructed, the trigger chain strength and evolution rate of each unit are calculated, a potential chain type instability path is identified, and grading early warning is realized in combination with a dynamic threshold model; constructing a structure safety redundancy dynamic attenuation model, identifying an evolution acceleration region, and establishing a structure instability evolution index to carry out multi-physical quantity fusion evaluation; virtual micro-disturbance is actively applied, an irreversible evolution stage is judged by analyzing nonlinear amplification and residual offset of redundant response, and a closed-loop dynamic early warning mechanism is formed. According to the invention, high-reliability disaster early warning from trend prediction to active verification is realized.

A sensing device (20) includes a control unit (210) and a communication unit (220). The communication unit (220) receives a sensing signal from a sensing transmitter (10). The control unit (210) calculates a phase difference of a reference signal included in the sensing signal, demodulates data included in the sensing signal, estimates a phase difference of the demodulated data on the basis of the phase difference of the reference signal, generates a propagation matrix on the basis of the phase difference of the data, and detects a detection target on the basis of the propagation matrix.

The invention discloses an electromagnetic scattering characteristic modeling method based on generalized characteristic current optimization, and the method comprises the steps: constructing an equivalent closed curved surface which surrounds a scatterer, and setting the frequency of an electromagnetic scattering field to be calculated; performing triangular mesh generation on the equivalent closed curved surface, establishing a corresponding RWG primary function, and determining an impedance matrix corresponding to the frequency of the electromagnetic scattering field; calculating a propagation matrix from the RWG primary function to the far-field spherical surface; based on the impedance matrix and the propagation matrix, constructing a generalized characteristic equation and solving a corresponding generalized characteristic current and a generalized characteristic propagation matrix; setting a measurement point outside the equivalent closed curved surface, applying a plane wave based on the set electromagnetic scattered field frequency and incident angle in the microwave anechoic chamber environment, and determining scattered field measurement data at the measurement point; constructing a convex optimization model to optimize the generalized characteristic current coefficient, and obtaining the surface induction current of the scatterer under the irradiation of the plane wave; and determining a scattered field at any position of the space based on the surface induction current.

The application discloses a towed array passive positioning method based on wideband matrix filtering, calculates a sound field propagation matrix by using a two-dimensional parabolic equation model, obtains different distance sound source-receiver system propagation matrices through matrix multiplication operation, extracts matrix coefficients from different distance system propagation matrices according to sound source depth and horizontal array element positions, and forms a new sparse system propagation matrix. The sparse system propagation matrix is processed to form a wideband matrix filter set, finally, the horizontal array receiving signal is processed by using the wideband matrix filter set, energy normalization is performed on different distance processing results, a wideband distance-depth ambiguity diagram is formed, and target depth and distance estimation results are obtained. The application can effectively improve the target position resolution of the towed array wideband signal passive positioning, reduce the sidelobe intensity in the two-dimensional ambiguity diagram, and greatly reduce the energy outside the main lobe, and has important value for improving the target positioning performance in the actual environment.

The application discloses a multimodal preference driven graph convolution combined optimization learning path generation method, comprising the following steps: acquiring multimodal learning feature data, and performing multimodal feature data fusion processing to obtain a learning resource initial feature matrix; constructing a multi-relation adjacency matrix, and performing normalization processing to construct a normalized propagation matrix; updating a graph convolution feature, and obtaining a learning gain score; calculating a comprehensive utility based on the learning gain score; establishing an integer programming under a plurality of constraint conditions, and screening to obtain an optimized candidate resource; generating an optimal learning path in the optimized candidate resource by using an edge cost function and a learning path target function; in the optimal learning path generation process, dynamic re-planning is adopted; the dynamic re-planning comprises mastery updating, coverage demand decreasing and time length budget updating. By the above scheme, the application has the advantages of simple logic, reliable multimodal fusion and the like.

The invention relates to the technical field of human deep neural network model quantification, and discloses a hierarchical variance propagation and gradient cooperative modulation-based neural network uncertainty quantification method, which comprises the following steps: S1, parameter posteriori distribution extraction: using a Bayesian neural network trained by Gaussian prior with a parameter of the mean value to obtain posteriori distribution with the same parameter, and extracting the posteriori distribution; wherein the parameter mean value is a variance; s2, posterior distribution truncation: truncating posterior distribution by using a dynamic distribution truncation mechanism; s3, constructing a continuous belief function, carrying out Mobius inversion, and obtaining confidence quality distribution for uncertainty expression from the belief function; and S4, fitting Dirichlet distribution, and constructing a hierarchical variance propagation matrix. And S5, carrying out further quantification on the trained model by adopting gradient weighting, layer selection gradient and gradient perturbation integral methods of a specific category. According to the method, the reliability, stability and risk controllability of model prediction are remarkably improved.

The invention discloses a label propagation algorithm based on triple integration similarity. The method comprises the following steps of: 1, respectively representing an original data set as similarity graphs under three different visual angles based on three measures of distance, correlation coefficient and density, and representing all known label information in the data set as a label matrix by using one-hot coding; 2, respectively solving three corresponding local similarity graphs and three local kernel matrixes; 3, solving a final triple integration similarity graph through an iteration mode; 4, performing corresponding transformation on off-diagonal elements and diagonal elements in the triple integration similarity graph to obtain a standard propagation matrix; and 5, spreading label information from a labeled sample to a non-labeled sample in an iteration mode, wherein the maximum index corresponding to each sample in the label matrix is the category to which the sample is divided. The method has the advantages that the adaptability to the distribution condition of a data set is better, the adaptability to data distribution is higher, and the robustness is higher.