546 results about "Singular value decomposition" patented technology

The invention relates to the technical field of image data processing, and discloses a few-sample scene-oriented industrial product surface defect image analysis method, which comprises the following steps of: obtaining surface gray level image data of a product to be analyzed, calculating a structure tensor matrix and generating an anisotropy degree graph; searching similar blocks in a preset search neighborhood, and constructing a local texture data matrix; performing singular value decomposition on the local texture data matrix to extract a main subspace; constructing a projection operator and utilizing the projection operator to carry out orthogonal projection reconstruction on the local image block to generate a reconstructed background image block; according to the method, through an orthogonal subspace projection mechanism, good product textures and defect signals are separated, random noise is removed, meanwhile, high-frequency structural features are completely reserved, and the method is high in robustness, high in robustness and high in robustness. And the defect detection precision of a complex texture surface in a few-sample scene is improved.

The invention discloses a large language model progressive field fine tuning and knowledge fusion method for shield engineering. The method comprises the following steps: constructing a layered shield training course containing a wide-area academic theory and a proprietary enterprise construction method; parallelly training a plurality of physically isolated parameter efficient adapters based on the frozen base; performing singular value decomposition on the adapter, extracting a geometric feature subspace representing knowledge distribution, and calculating a conflict correlation degree; based on this, a uniform adaptation mechanism of resource awareness is constructed. The mechanism not only can generate a static fusion model for conflict removal, but also can dynamically activate a specific rank slice of an adapter through a routing network based on real-time hardware resource budget (video memory / FLOPs) and geometry-resource signature. According to the method, multi-source knowledge is reserved, and adaptive dynamic scheduling of edge hardware resources by model reasoning is realized.

The invention relates to the technical field of numerical control programming, in particular to a five-axis machining path planning method and system based on data driving, and the method comprises the following steps: obtaining real-time coordinates of each axis of a machine tool, calculating linear velocity and angular velocity components to construct a Jacobian matrix, executing singular value decomposition, and calculating a conditional number ratio by using maximum and minimum singular values; and inputting a nonlinear mapping function to calculate a dynamic penalty factor, generating a rotating shaft weighted item in combination with a rotating shaft identifier, constructing a weighted damping least square objective function, calculating a five-axis motion increment, and accumulating the five-axis motion increment with a real-time coordinate to generate a target absolute position coordinate. According to the method, the pose singularity degree is quantified by monitoring the machine tool pose condition number ratio and converted into the dynamic penalty factor to apply the self-adaptive constraint to the rotating shaft, the severe sudden change of the rotating shaft in the singularity area is inhibited, the tool nose track following error is minimized, and meanwhile smooth distribution of the motion increment is achieved; and the dynamic stability and the surface quality of five-axis linkage machining are improved.

The invention provides a photovoltaic inverter operation state diagnosis method, which belongs to the technical field of photovoltaic inverters, and comprises the following steps: collecting multi-source operation signals of a photovoltaic inverter, carrying out wavelet packet decomposition on the signals to construct a time-frequency characteristic dense matrix, generating a fault characteristic super-sparse representation vector through singular value decomposition and sparse processing, and carrying out fault characteristic super-sparse representation on the fault characteristic super-sparse representation vector. Performing envelope demodulation on sensitive mode components obtained by complete set empirical mode decomposition to extract approaching periodic feature vectors, and inputting three types of complementary features into a weak fault recognition model with a circulation attention mechanism to perform fusion diagnosis. And whether preventive maintenance early warning is triggered or not is judged according to the output determinant characteristic value of the convergence state matrix, and an operation strategy is adjusted. The technical problem that early weak fault characteristics of the photovoltaic inverter are difficult to be accurately identified and early warned in time in a strong noise background is solved.

The invention discloses a flow field prediction-oriented full-process automatic joint reduced-order modeling method, which comprises the following steps of: specifying a target physical field parameter space, and randomly generating a sample space according to a Latin hypercube sampling method; constructing a full-process automatic simulation tool chain, driving target physical field numerical calculation and generating a training data set; carrying out singular value decomposition-based intrinsic orthogonal decomposition on the output physical field data, and only retaining first r main feature components to construct a reduced-order data set; constructing a multi-input multi-output full-connection feedforward neural network, and modeling and training a nonlinear mapping relation between input parameters and reduced-order features; new working condition parameters are input, reduced-order features are predicted through the trained neural network, distribution of a target physical field is reconstructed according to a singular value decomposition reduction matrix, and more flexible and reliable technical support is provided for reducing the training cost of a reduced-order model and improving simulation efficiency.

The invention discloses a robot hand-eye calibration method, control equipment and a robot system, and relates to the technical field of robots. The robot hand-eye calibration method comprises the following steps: uniformly correcting a plurality of pixel coordinates obtained by dynamic shooting to a reference pixel coordinate system which is consistent with a camera coordinate system at a reference position in direction; and meanwhile, the world coordinates of the end effector corresponding to each sampling position are localized relative to the world coordinates of the end effector at the reference position, so that interference caused by global coordinate offset and camera orientation difference is eliminated, and the application range of the hand-eye calibration method is expanded. Besides, a singular value decomposition algorithm is adopted to solve a mapping relation between pixel coordinates and local world coordinates under a reference pixel coordinate system, so that a solved rotation matrix can meet orthogonality constraints, the hand-eye transformation process strictly conforms to a robot motion model with translation and rotation motion, and the calibration precision is improved. The calibration process can be automatically calculated and completed, and the calibration efficiency is high.

The invention discloses a partial discharge signal denoising method based on STFT-SVD and Bayesian kurtosis threshold adaptive optimization, and the method comprises the steps: collecting an analog signal outputted by a high-frequency current transformer, carrying out the analog-to-digital conversion, obtaining a one-dimensional time domain signal sequence, carrying out the DC component removal and amplitude normalization of the signal, and obtaining a preprocessing time domain signal; performing short-time Fourier transform on the preprocessed time-domain signal to obtain a time-frequency spectrum, suppressing low-amplitude noise by adopting a soft mask method, and retaining main characteristics of partial discharge pulses; performing singular value decomposition on the time-frequency spectrum after soft masking, automatically selecting a principal component number according to a principal component, and adaptively reserving a main signal component to obtain a principal component spectrum; and performing inverse short-time Fourier transform on the principal component atlas to reconstruct a time domain signal, adaptively selecting a kurtosis threshold in combination with a Bayesian optimization algorithm, and outputting a denoised time domain signal.

The invention belongs to the technical field of wireless communication, and discloses a CNN-Transform direction estimation method based on covariance-unitary matrix input, and the method comprises the following steps: generating received signal sample data of different incident angles; calculating a covariance matrix for a received signal sample, performing singular value decomposition, extracting real parts and imaginary parts of the covariance matrix and a unitary matrix, and constructing a four-channel two-dimensional real number tensor as input; extracting local spatial features and coherence structures through CNN; serializing the feature map and adding a position code; a Transform encoder module is input, and global spatial dependence is modeled; carrying out average pooling and full-connection classification on the output to realize direction angle prediction; and training is carried out by adopting cross entropy loss, an AdamW optimizer and a cosine annealing learning rate scheduler. According to the method, local and global features are fused, and the method has high precision, strong robustness and excellent generalization ability in complex environments of low signal-to-noise ratio, multipath interference and the like.

The invention relates to the technical field of quantum computing, and provides a linear complexity quantum state preparation method based on tensor decomposition and a quantum circuit construction system.The high-dimensional tensor is decomposed into a series of low-rank core tensors through continuous singular value decomposition, each core tensor in a core tensor sequence is expanded into a unitary matrix, and the unitary matrix is used as a quantum circuit; the unitary matrix sequence is mapped to quantum lines coupled using adjacent qubits, and the quantum lines are run to prepare a target quantum state. Based on this, the line generated by the method can approximately or accurately prepare a target quantum state only by coupling adjacent quantum bits, and is perfectly adaptive to quantum chips of linear or grid topologies such as superconducting and semiconductor quantum dots and the like.

The invention discloses a model processing method and device, equipment, a storage medium and a program product, and relates to the technical field of artificial intelligence, in the method, singular value decomposition is carried out on a network weight matrix of a hybrid expert model, and in an obtained right singular vector matrix, columns with singular values smaller than a threshold value are screened to serve as a null-space basis matrix. In this way, the null-space basis matrix can represent the activation mode direction which contributes little to the activation of the expert network. Furthermore, a sub-network weight matrix is constructed based on each column of the null-space basis matrix and the network weight matrix. Therefore, according to the maximum singular value and the minimum singular value of the sub-network weight matrix, the closeness degree of linear correlation between each column of the network weight matrix and the null-space base matrix can be determined, then the invalid feature sensitivity of each expert network can be determined, and according to the invalid feature sensitivity, the invalid feature sensitivity of each expert network can be determined. And cutting network parameters of the hybrid expert model. In this way, the model parameter quantity can be reduced.

The invention discloses a lanyard data line interaction control method based on artificial intelligence, and the method comprises the following steps: S1, collecting interaction data of a lanyard data line, and carrying out the preprocessing of the interaction data; s2, segmenting according to a set window, performing singular value decomposition on each segment, and constructing a main feature set; s3, performing hypersphere embedding on the principal component vector, performing linear projection on the residual vector, and generating a characteristic spectrum by adopting Laplacian mapping; s4, performing tensor decomposition on the characteristic spectrum, and constructing a multi-order structure; s5, executing bidirectional loop iteration on the tensor interaction sequence and controlling information transmission; s6, analyzing the prediction action, matching the prediction action with an instruction mapping table, generating a control instruction, and sending the control instruction to the terminal equipment; and S7, counting execution feedback, updating a tensor interaction sequence weight, and optimizing an interaction control strategy. According to the invention, high-precision identification and stable adaptive control of the lanyard data line are realized, and the interaction precision, the response speed and the use convenience are effectively improved.

The invention provides a TOF (Time of Flight) estimation method of a matrix bundle based on partitioning and random SVD (Singular Value Decomposition). The method comprises the following steps: firstly, acquiring channel state information (CSI) by using a commercial WiFi device, and constructing a CSI matrix; and secondly, performing block overlapping processing on the CSI matrix, constructing an augmented matrix for each CSI data block, and performing SVD decomposition on the augmented matrix, thereby reducing the calculation complexity of SVD decomposition. And then, the time of flight (TOF) of each CSI data block is estimated by using a matrix pencil algorithm, and the minimum TOF is direct path estimation. According to the method designed by the invention, the robustness of the system and the estimation precision of the TOF are improved under the condition that the algorithm complexity is greatly reduced.

The invention relates to the field of array signal processing, and particularly discloses a DOA joint estimation method based on a quaternion polarization sensitive array. Electromagnetic wave dual polarization components are captured through the orthogonal dipole and the loop antenna, and a quaternion observation matrix is constructed; calculating a quaternion covariance matrix by adopting a sliding window mechanism, and separating a signal / noise subspace by adopting quaternion singular value decomposition; and constructing a spatial spectrum function in combination with a quaternion steering vector, and realizing joint estimation of an azimuth angle and a pitch angle through two-dimensional search. According to the method, the unified characterization capability of quaternions on polarization-airspace information is fully utilized, and the DOA estimation precision and the anti-interference performance of the multi-polarization signal are remarkably improved.

The invention discloses a Non-IID federated learning backdoor attack defense method and system and a medium, and relates to the field of federated learning and network security. In order to solve the problems that an existing method depends on model parameters and is easy to avoid by a malicious client, and adaptive Non-IID scenes are poor, truncated singular value decomposition is executed through the client to extract first p left singular vectors, and the first p left singular vectors are uploaded to a server; the server constructs a matrix based on left singular vector cosine similarity and performs hierarchical clustering, and calculates a client similarity score and an aggregation weight by using a zoom dot product attention mechanism in combination with a left singular vector of the clean reference data set; the server distributes a global model, the client uploads parameters after local training, and the server weights and aggregates the model parameters in the cluster according to the weight and iteratively optimizes the model parameters. According to the method, the malicious client is identified from the data essential features, the malicious proportion does not need to be preset, the good client contribution and the data privacy are guaranteed while the backdoor attack is inhibited, the method is suitable for a Non-IID scene, and the model robustness and the main task performance are improved.

The invention discloses an automatic history fitting method based on multiple data assimilation of an improved set smoother, and relates to the technical field of petroleum engineering. The method comprises the following steps: firstly, constructing a plurality of oil reservoir models, setting a prior model, a real model and an initial expansion factor, obtaining the yield of each oil reservoir model by utilizing oil reservoir model numerical simulation, and calculating a residual error; based on corrected data covariance matrix singular value decomposition, production observation data disturbance enhancement, adaptive learning rate matrix scaling and geological boundary constraint, improving a set smoother, updating the permeability of each oil reservoir model, performing numerical simulation again, updating an expansion factor, judging whether the expansion factor meets a preset condition or not, continuing iteration if the expansion factor meets the preset condition, and if the expansion factor does not meet the preset condition, continuing iteration until the expansion factor meets the preset condition; and if not, updating the expansion factor of the iteration, ending the iteration, and outputting the permeability field of each updated oil reservoir model, thereby solving the problems of parameter overshoot, covariance statistical deviation and low calculation efficiency in oil reservoir history fitting, and facilitating history fitting of complex oil reservoir production parameters.

The invention relates to a metal additive manufacturing three-dimensional temperature field Gaussian process prediction method, belongs to the technical field of material science, and particularly relates to a metal additive manufacturing three-dimensional temperature field prediction method. Logarithmic transformation is adopted to preprocess a temperature field, and prediction difficulty caused by extreme gradient near a molten pool is avoided; dividing the overall computational domain into a plurality of sub-domains by adopting a domain decomposition strategy to reduce the problem dimension; for each sub-domain, further combining singular value decomposition to extract a temperature field reduced-order base; establishing a local Gaussian process regression model based on a Maren kernel function and carrying out parallel training so as to establish rapid mapping from process parameters to reduced-order output; during online prediction, efficient and accurate prediction of a complete temperature field is realized through parallel calculation and full-field assembly of each local model.

The invention discloses a medical knowledge graph construction method and device for multi-source heterogeneous data fusion and incremental updating, and relates to the technical field of medical knowledge graphs. The method comprises the steps that multi-source medical data such as outpatient records, hospitalization medical records, inspection reports and image data of a patient in the whole life cycle are collected; performing time sequence alignment by adopting a space-time alignment algorithm to obtain standardized medical data, calculating a data fusion weight and constructing an incidence matrix; performing singular value decomposition on the incidence matrix to extract a feature vector, and obtaining a dynamic update coefficient through standardization, cosine similarity calculation, convolutional neural network processing and exponential weighted moving average; adjusting the reference map according to the dynamic updating coefficient increment to generate an updated medical knowledge map, and outputting diagnosis and treatment decision support information; and multi-dimensional medical data space monitoring abnormity can be constructed, a joint optimization model is established to generate a personalized diagnosis and treatment scheme, and edge deployment is realized through a knowledge distillation compression map.

The invention discloses a power distribution system harmonic source positioning method based on subsystem division, and aims to improve the accuracy and anti-interference capability of harmonic source identification in a power distribution system. According to the method, firstly, an obtained voltage signal is preprocessed, frequency domain feature information is extracted, a subspace projection operator is constructed through singular value decomposition, and effective extraction of each harmonic component in the signal is achieved. Then, carrying out subsystem division on the power distribution system according to the position of the tail end measurement point, for each subsystem, respectively injecting simulated harmonic current into each candidate node by adopting a virtual harmonic injection method, and calculating the response of the node to the tail end voltage; and comparing a simulation result with a measured value, constructing an error vector, screening out a node with the minimum error as a backup harmonic source node, and finally positioning the harmonic source node through zero setting of a connection node and a correction strategy of the most downstream node. The method is suitable for a multi-node and multi-source coexistence large-scale power distribution network harmonic suppression scene.

The invention discloses a GNSS (Global Navigation Satellite System) signal sparse compression capture method for a low-correlation measurement matrix, which comprises the following steps of: firstly, constructing a code phase matrix and a frequency offset matrix, and carrying out sparse transformation on a downsampled GNSS signal to obtain a correlation matrix; the method comprises the following steps of: firstly, carrying out iterative optimization on a measurement matrix by taking a Gramb matrix as a matrix, then taking a Frobenius norm of a difference between the Gramb matrix and an equiangular tight frame matrix as a target function, alternately using singular value decomposition (SVD) and a shrinkage operator to carry out iterative optimization on the measurement matrix, and carrying out compression measurement on a correlation matrix through the optimized measurement matrix. Secondly, a correlation matrix of compression measurement is quickly and accurately reconstructed by using a least square operator and an orthogonal matching pursuit algorithm; and finally, when the peak value ratio of the primary peak to the secondary peak of the reconstructed correlation matrix is greater than a threshold value, outputting code phase time delay and Doppler frequency offset corresponding to the primary peak. And if the ratio of the primary peak value to the secondary peak value of the reconstructed correlation matrix is smaller than the threshold value, capturing the GNSS signal again. According to the method, the maximum correlation coefficient and the average correlation coefficient of the measurement matrix can be reduced, the GNSS signal capturing sensitivity is improved, and hardware resources in the signal capturing process in a GNSS receiver can be reduced.

The invention relates to a matrix decomposition device and method based on a memristor cross array, which are suitable for efficient hardware implementation of singular value decomposition (SVD), the memristor cross array receives a conductance value mapped by a Grubrum matrix constructed by a to-be-decomposed matrix and a bit line input voltage converted by a bit line column vector, and outputs a corresponding current; converting the corresponding current into a voltage vector; performing normalization processing on the voltage vector to obtain a normalized vector; judging whether convergence occurs or not; performing normalization processing on the steady-state voltage vector to obtain a final output vector; and according to the final output vector, main characteristic values are extracted based on a first normalization circuit, and main singular values and corresponding singular matrixes are calculated. Compared with the prior art, the high parallel computing characteristic of the memristor cross array is utilized, efficient operation and hardware acceleration of matrix decomposition are achieved, the method has the advantages of being low in power consumption, high in speed and good in expansibility, and the method is suitable for application scenes such as artificial intelligence, signal processing and large-scale matrix operation.

The invention discloses a cross-sensor long-time-sequence InSAR deformation monitoring method and device, a medium and a product, and relates to the field of synthetic aperture radar interferometry, the method comprises the following steps: obtaining image data sets of a plurality of SAR sensors in a monitoring area, and sequentially carrying out differential interference and time sequence deformation calculation to obtain an LOS direction deformation result; geocoding all results to unify a coordinate system; screening high-coherence point targets and constructing an observation geometry unified analysis model; time information is obtained by combining the images, and a long-time-sequence earth surface deformation resolving function model after observation geometric correction is established; and further introducing a truncated singular value decomposition operator, constructing a cross-sensor long-time-sequence deformation resolving analysis model, and finally resolving to obtain a high-precision and long-time-span earth surface deformation product in the monitoring area. According to the method, the problems of rank deficiency and solution failure caused by time shortage and observation geometric difference in multi-platform InSAR data joint solution are effectively solved, and the precision and applicability of long-time-sequence deformation monitoring are remarkably improved.

The invention relates to the technical field of defect detection, in particular to a defect detection method and system based on a microwave frequency crystal resonator, and the method comprises the steps: forming a microwave probe through a microwave frequency crystal resonator array, placing a to-be-detected material below the microwave probe, and setting a scanning area and scanning step lengths in two directions on the surface of the material; controlling the probe to perform point-by-point scanning according to step length, and collecting a plurality of reflection coefficients of each sampling point to form an echo signal matrix; performing singular value decomposition on the matrix, applying an adjustable coefficient to a first singular value and a second singular value, and adaptively selecting an optimal coefficient according to imaging quality to reconstruct an echo signal; performing two-dimensional Fourier transform on the reconstructed matrix, constructing a spatial filter in combination with probe spacing and defect depth to perform phase compensation, performing two-dimensional Fourier transform to obtain a defect image, and performing threshold segmentation on the defect image to obtain a binary image only containing defects; and analyzing the connected regions to obtain the gravity center position of each defect and the defect area determined by the pixel number and the single pixel area. The problems of serious clutter interference, insufficient defect imaging contrast ratio, difficulty in quantitative evaluation of defect size and the like in the existing microwave detection can be solved.

The invention discloses a large model parameter protection method based on singular value decomposition in a trusted execution environment. Positioning a key layer of the large model through gradient sensitivity and interlayer correlation measurement in a large model fine tuning training process, and decomposing a weight matrix of the key layer by using singular value decomposition to obtain sensitive parameters and non-sensitive parameters; dividing a large model operation environment into a rich execution environment and a trusted execution environment, and generating a reversible key matrix in the trusted execution environment; when input data pass through the key layer for the first time, the rich execution environment uses non-sensitive parameter reasoning and then transmits the input data to the trusted execution environment for sensitive parameter reasoning, then the data are encrypted by using the key matrix, the encrypted data are transmitted back to the rich execution environment for reasoning, and when the input data pass through the key layer again, the rich execution environment conducts reasoning and then transmits the input data to the trusted execution environment. And the final data is output after decryption and reasoning. According to the method, the side channel attack risk is reduced, the data transmission security and the calculation efficiency are improved, and the model reasoning delay is reduced.

The invention relates to the technical field of wind power plant power control, in particular to a wind power plant dynamic power distribution optimization method based on projection dimensionality reduction. The method comprises the following steps: collecting a high-dimensional original operation data stream of a wind power plant, and preprocessing to output a high-dimensional state vector; extracting a dominant mode matrix through an intrinsic orthogonal decomposition algorithm, and projecting a high-dimensional state vector to a low-dimensional space; future modal coefficient evolution is predicted through an autoregression prediction model, and a low-dimensional rolling optimization problem is constructed and solved through a sequential quadratic programming algorithm; reconstructing the optimal low-dimensional modal coefficient vector into a target power instruction of each fan, and issuing and executing the target power instruction; and updating the dominant mode matrix through an incremental singular value decomposition algorithm and correcting parameters of the autoregressive prediction model. According to the method, projection dimensionality reduction from high-dimensional data to a low-dimensional space is realized through intrinsic orthogonal decomposition, and the problem of low optimization solution efficiency caused by high data dimensionality of a traditional method is solved.

The invention relates to the technical field of defect detection, in particular to a copper wire internal defect detection method and system combined with eddy current. The method comprises the following steps: applying composite excitation formed by superposing multi-frequency harmonic signals, collecting eddy current response signals, and obtaining an initial time-frequency spectrum matrix through short-time Fourier transform; singular value decomposition is carried out on the matrix, and after main singular value components related to the lift-off effect are removed, a lift-off correction spectrum matrix is reconstructed; calculating the maximum Lyapunov exponent and the spectrum kurtosis value of the correction matrix as a first defect characteristic quantity and a second defect characteristic quantity respectively; and constructing a two-dimensional feature plane based on the two feature quantities, calculating an angle correction quantity by utilizing energy of a rejected component so as to adjust a defect judgment area, and finally judging the defect by combining a Manhattan distance and a corrected angle interval. According to the scheme, the lift-off effect interference can be effectively inhibited, the dynamic criterion is constructed, and high-precision and high-reliability detection of the weak defects in the copper wire is realized.

The invention discloses a natural gas pipeline leakage detection method based on novel wavelet basis transformation and singular value decomposition in a two-dimensional convolutional neural network. Firstly, a sound signal collected by a sound wave sensor is converted into a digital signal; secondly, in the data preprocessing stage, singular value decomposition is carried out on the digital signals to effectively eliminate background noise interference, and then batch normalization is carried out on the processed data; then, converting the one-dimensional time sequence signal into a two-dimensional time-frequency image by adopting a self-defined Morlet wavelet basis function; and finally, based on the time-frequency images, constructing and training a 2D-CNN model for fault classification, and presenting a diagnosis result through a confusion matrix and a comparison graph. According to the method, 97.55% of fault recognition accuracy is obtained in a public data set, and compared with other competitive methods, the method shows more excellent noise robustness and classification performance, and has higher accuracy and wide application prospects in pipeline leakage diagnosis in a complex noise environment.

The invention discloses a fault diagnosis system for an aviation turboshaft engine, and relates to the technical field of fault diagnosis. An acquisition module acquires horizontal, vertical and axial vibration signals; the processing module is used for generating real-time feature vectors by introducing dynamic short-time Fourier transform, variational mode decomposition and singular value decomposition of an unfixed-width Gaussian window; the incremental diagnosis module calculates the matching degree between a real-time feature vector or a prediction feature vector and a known class through a dynamic expansion network, selects to belong to the known class or triggers an expansion mechanism according to an adaptive resonance theory, constructs joint loss by using a Bayesian learning framework after expansion, and performs Gaussian process modeling; loss is determined through Bayesian optimization, and network parameters are updated; the prediction module generates a prediction feature vector when there is no fault; the maintenance module sends or stores a maintenance scheme according to needs, self-adaptive recognition of known classes and new fault classes is achieved, the state can be pre-judged in advance, the operation safety of the engine is effectively guaranteed, and the maintenance cost is reduced.

The invention discloses a large language model increment method and device and a storage medium, and belongs to the technical field of data processing. Comprising the following steps: acquiring an initial base model and a current model, calculating a model difference parameter between the current model and the initial base model, performing singular value decomposition on the model difference parameter to obtain a left singular vector matrix, a singular value matrix and a right singular vector matrix, selecting a target singular value, and calculating the target singular value. And based on the target singular value, the left singular vector matrix and the right singular vector matrix, constructing a low-rank approximate representation, generating a first low-rank weight assembly and a second low-rank weight assembly according to the low-rank approximate representation, and deploying the first low-rank weight assembly and the second low-rank weight assembly in the initial base model to form a large language model. According to the method, the low-rank approximate representation is constructed based on the difference value between the initial base model and the current training stage model, and the low-rank weight assembly which can be directly loaded on the initial base model is generated, so that the dependency relationship between modules in different stages is cut off, and the modular multiplexing capability of a training result is enhanced.

The application discloses a partial discharge signal denoising method, comprising the following steps: building a partial discharge signal detection platform and a partial signal collection platform; measuring and detecting characteristic signals in the partial discharge signals; screening and classifying the measured signals through a Gaussian test to obtain the characteristic signals; processing and denoising the signals through a fast S transform algorithm and a singular value decomposition SVD algorithm; testing and diagnosing the denoising effect by using two coefficients of noise suppression ratio and amplitude attenuation ratio; and displaying the diagnostic analysis result on a platform display interface. The application discloses a partial discharge signal diagnostic system based on a fast S transform and an adaptive singular value, and solves the problems that the existing technical methods are difficult to balance the denoising effect and extraction speed in the characteristic extraction of the partial discharge signals and are difficult to extract the characteristics.

The invention relates to an aero-engine fault diagnosis method and device based on tensor decomposition and a medium, and the method comprises the steps: collecting the gas path parameter time sequence data of a plurality of parts of an aero-engine, and carrying out the time alignment and working condition label labeling; for the multi-parameter data of each component, a kernel principal component analysis and automatic encoder fused feature extraction method is adopted, and parameters of different components are unified into fixed dimension features; the feature matrixes obtained after feature extraction of all the components are combined, and a time-component-feature third-order tensor is constructed; training a high-order singular value decomposition model based on the engine health monitoring data; fault diagnosis is carried out by calculating the reconstruction error of the test data and the monitoring data; and fault positioning is realized by combining core tensor difference analysis. According to the method, the problem of feature fusion of engine multi-source heterogeneous data under variable working conditions is solved, and the engine fault detection sensitivity and positioning precision are improved.