46 results about "Nonlinear dimension reduction" patented technology

The invention discloses a method for constructing an evaluation model for the influence of climate change on biodiversity, and particularly relates to the technical field of ecological modeling and climate response analysis. The method comprises the following steps: generating a high-resolution characteristic grid based on microclimate disturbance data and a remote sensing image, extracting ecological plaques and establishing a heterogeneity spatial index model, constructing a connectivity map and a coupling response path library, extracting a typical species response mode by using nonlinear dimension reduction and spectral clustering, and performing dynamic fitting in combination with biological survey data. And finally, generating a diversity attenuation trend prediction curve and a risk thermodynamic diagram, identifying a biodiversity collapse critical point and outputting an intervention priority, and the method can be widely applied to regional ecological early warning and protection planning.

The invention provides a backpack multi-sensor data fusion-based motion attitude anomaly real-time detection method, which comprises the following steps of: acquiring multi-channel original data through a backpack-type integrated triaxial accelerometer, a gyroscope, a magnetometer and a barometer, implementing synchronous sampling and time-space alignment, and constructing a standardized time sequence data stream through wavelet denoising and multi-dimensional normalization preprocessing; high-dimensional features representing motion differences are extracted, nonlinear dimensionality reduction is achieved through principal component and Laplacian feature mapping, and a motion mode prototype library is generated; according to the method, a new motion situation can be continuously and adaptively learned, and the motion recognition accuracy, the system robustness and the attitude anomaly detection capability in a complex scene are effectively improved.

The invention provides a medical insurance abnormal data detection method based on an adversarial auto-encoder and a bidirectional LSTM (Long Short Term Memory). A data preprocessing step: cleaning and normalizing the original medical insurance data; a feature engineering step: performing feature extraction and standardization on the preprocessed data to construct a feature set for detection; an unsupervised feature learning step: performing nonlinear dimensionality reduction on the feature set by using an adversarial auto-encoder to obtain low-dimensional potential feature representation; and a time sequence anomaly detection step: inputting the low-dimensional potential feature representation sequence into a bidirectional long-short-term memory network BiLSTM to model a time sequence dependency relationship of the low-dimensional potential feature representation sequence and output an anomaly detection result. AAE and BiLSTM are fused, dimension reduction feature extraction and time sequence modeling are combined, and an AAE-BiLSTM medical insurance anomaly detection model is constructed. The problem of coexistence of high-dimensional heterogeneity and strong time sequence of medical insurance data is effectively solved by fusing the unsupervised feature learning capability of an adversarial auto-encoder (AAE) and the strong time sequence modeling capability of a bidirectional long short-term memory network (BiLSTM).

The invention relates to the technical field of gene editing, and discloses an sgRNA activity prediction method and device based on deep features and automatic machine learning. The method comprises the following steps: acquiring sgRNA training data with editing efficiency labels and discretizing the sgRNA training data into classification labels; after the sgRNA and the PAM sequence are spliced, inputting the spliced sgRNA and PAM sequence into a pre-training depth sequence model to extract high-dimensional features; obtaining a low-dimensional feature vector through nonlinear dimensionality reduction; an automatic machine learning framework is combined with cross validation, and a prediction model is automatically trained and optimized based on low-dimensional features; and finally, carrying out activity prediction and sorting screening on candidate sgRNA by utilizing the model. According to the method, through fusion of deep semantic extraction, nonlinear structure maintenance and a full-automatic modeling process, the accuracy, robustness and development efficiency of sgRNA activity prediction are effectively improved, and an efficient pilot screening tool is provided for a gene editing experiment.

The invention provides a space dimension reduction method for dynamic trajectory analysis, and belongs to the field of data analysis and machine learning. The method takes dynamic trajectory data as Riemannian function type data to carry out nonlinear full dimension reduction, and comprises the following steps of: 1, converting the Riemannian function type dynamic trajectory data into Euclidean space function type dynamic trajectory data; 2, constructing a Gram matrix of time data and a distance matrix of measurement space data for the Euclidean space function type dynamic trajectory data; and step 3, constructing a dimension reduction function to carry out nonlinear dimension reduction on the Euclidean space function type dynamic trajectory data. According to the method, dynamic trajectory data with a complex geometric structure can be effectively processed.

The invention discloses a multi-working-condition division method and system for nuclear power data based on a TSNE-improved HBBT model and a storage medium, and belongs to the technical field of intelligent operation and maintenance of nuclear power stations. The method comprises the following steps: firstly, acquiring historical operation data of a nuclear power system through a simulation platform, and extracting a mean value, a variance, a slope, an autocorrelation coefficient and an energy characteristic after normalization processing; dividing the feature data into a training set and a test set, and endowing working condition labels; carrying out nonlinear dimension reduction on the training set by adopting T-SNE; and finally, training and classifying by using an improved histogram gradient lifting tree model. The improved HBBT innovatively adopts an adaptive feature binning technology and a sample difficulty and cluster consistency weighting strategy, so that the model precision and robustness are effectively improved. The method solves the problems that an existing unsupervised method is high in randomness and inaccurate in classification, can accurately recognize the steady-state and transition working conditions of the nuclear power system, and provides reliable support for intelligent operation and maintenance.

The invention provides a docking method, device and equipment for a protein surface structure and a drug target, and the method comprises the steps: obtaining the three-dimensional structure data of a protein, generating a surface mesh model based on the three-dimensional structure data, and extracting the multi-modal surface features of the protein; based on the surface mesh model and the surface features of the protein, constructing a multi-scale 3D graph convolutional network, and generating feature codes of the protein; performing nonlinear dimensionality reduction on the feature code of the protein based on an adversarial auto-encoder to generate a low-dimensional feature of the protein; and performing iterative reinforcement learning docking based on the low-dimensional features and ligand positions of the proteins to obtain an optimal docking result. Through the method and the device, the surface features of the protein are extracted more comprehensively and accurately, the calculation amount is greatly reduced, and the problem of poor docking accuracy in the prior art is solved.

The invention provides a brain dynamic mode classification method based on a graph auto-encoder and soft and hard clustering. The brain dynamic mode classification method comprises the steps that S1, data preprocessing and blood oxygen level dependence signal extraction are carried out; s2, constructing a dynamic brain network; s3, nonlinear dimensionality reduction of the graph auto-encoder is carried out; s4, soft and hard clustering conjoint analysis and dynamic mode feature extraction; s5, performing feature screening and validity verification; and S6, classifier training verification and classification result output. According to the method, rs-fMRI data is taken as core input, accurate classification of brain dynamic modes is realized through a whole-process design of data preprocessing, brain network construction, nonlinear dimension reduction, clustering analysis and classification verification, and the problem of low classification accuracy of a traditional magnetic resonance image data classification method is solved.

The application relates to the technical field of network analysis, and discloses a network community discovery system and method through matrix analysis, which comprises the following modules: a network module, which converts a static network topology into an information propagation model, and establishes a node state time sequence dynamic model by defining a node information processing rule and a multi-round propagation mechanism; a phase space reconstruction module, which maps high-dimensional time sequence data to low-dimensional phase space through a nonlinear dimension reduction method to form node trajectory distribution data; a community feature module, which analyzes the convergence, oscillation mode and attractor feature of the node trajectory, and generates community structure feature data through trajectory similarity; and a community division module, which identifies a community boundary through density clustering, and constructs a hierarchical community organization through multi-scale analysis. The application can deeply mine the internal community structure of a network from the perspective of dynamic information propagation, and overcomes the limitation that traditional methods only consider static topology.

The invention discloses a road freight channel identification method and device, computer equipment and a readable storage medium, and relates to the technical field of traffic transportation, and the method comprises the steps: obtaining the original OD travel data of a truck based on high-precision positioning data, extracting the related OD data, and carrying out the cleaning and feature construction; obtaining characteristic variables including district and county-level OD latitude and longitude, travel distance, travel quantity and azimuth angle; a multi-layer perceptron (MLP) is adopted to carry out nonlinear dimension reduction processing on the feature variables, and an optimal low-dimensional embedded vector is obtained; carrying out clustering analysis on the low-dimensional embedded vector by adopting a segmented clustering algorithm and filtering noise; and performing path fitting processing on the filtered line segments to generate a continuous transportation channel, and finally outputting a freight transportation channel identification result after matching with a road network. According to the method, through combination of deep learning dimension reduction and segmented clustering, the spatial distribution characteristics of the freight transport channel are effectively extracted, and the freight transport channel identification precision and efficiency are improved.

The application discloses an IGBT residual life prediction method based on multi-feature fusion and KPCA optimization. The method first collects IGBT collector current and voltage and other multi-source signals, extracts time domain, frequency domain, time-frequency domain and derived statistical domain features after pretreatment; then adopts a two-stage strategy of comprehensive evaluation index preliminary screening and mutual information regression fine screening to eliminate redundant features and retain high correlation features. On this basis, nonlinear dimension reduction is carried out by using kernel principal component analysis to construct a high-robustness health index, and the exponential weighted moving average and adaptive gradient detection are combined to accurately divide the degradation into three stages. Finally, the CNN-BiLSTM model is used to deeply mine the time sequence degradation features, and the MC Dropout algorithm is introduced to realize the accurate prediction and uncertainty quantification of the residual life. The application effectively solves the one-sidedness of single-source signal representation and the feature redundancy interference problem, and significantly improves the prediction accuracy and reliability.

The invention belongs to the field of image super-resolution reconstruction, and particularly discloses an image super-resolution reconstruction method and system based on nonlinear dimensionality reduction of hidden features of a state space model, and the method comprises the steps: achieving the super-resolution reconstruction of a two-dimensional image through a trained super-resolution reconstruction model; the super-resolution reconstruction model comprises a shallow feature extraction module, a deep feature extraction module and a reconstruction module. The shallow feature extraction module extracts shallow features. The deep feature extraction module comprises a plurality of attention enhancement modules which are connected in series, and the large kernel convolution fusion multi-head attention module enhances image detail features by extracting multi-scale spatial features and multi-channel features; the state space model hidden feature nonlinear dimension reduction module establishes a nonlinear mapping relation between compressed hidden features and a feature map, and corrects a linear mapping result through a channel attention mechanism; the reconstruction module obtains a reconstructed image based on the deep features. According to the invention, efficient and accurate two-dimensional image super-resolution reconstruction can be realized.

The application relates to the technical field of battery management systems, in particular to a battery system multi-performance index joint estimation method based on CNN-BiLSTM-SAM. The method collects time sequence data in the battery charging and discharging process, calculates the health state of the battery system, extracts health characteristic parameters, introduces kernel principal component analysis for nonlinear dimension reduction of the health characteristics, constructs a CNN-BiLSTM-SAM hybrid deep learning model of the battery system multi-performance, introduces a multilevel regularization system containing L2 regularization, a Dropout mechanism, an early stopping strategy and data enhancement in model training, effectively inhibits overfitting, uses a genetic algorithm to globally optimize hyperparameters, and finally realizes high-precision joint prediction of the battery system SOC, SOE and SOH. Through nonlinear dimension reduction, a multilevel regularization mechanism and genetic algorithm optimization, the application realizes high-precision joint estimation of the battery system multi-performance index, and has the advantages of superior feature expression capability, high prediction precision and strong generalization capability.

The invention discloses a crack prediction method and device, electronic equipment and a medium. The method can comprise the following steps: optimizing imaging of a complex structure with a breaking joint; crack post-stack attribute description is carried out for different types of breaking joints; fracture post-stack attributes are fused through an LLE nonlinear dimension reduction algorithm, and cracks are comprehensively represented. According to the method, a data basis is tamped through a pre-processing related technology, different parts of the breaking joint reservoir structure are subjected to hierarchical prediction through different attributes, and finally, the crack is subjected to comprehensive characterization through an attribute fusion technology, so that the crack description precision is greatly improved, and a good support is provided for high-quality reservoir prediction.

The invention discloses a multi-source remote sensing forest biomass estimation method and system for high-precision inversion, and relates to the technical field of forest biomass estimation, and the method comprises the steps: obtaining multi-source remote sensing data of a target artificial forest region; performing pretreatment; constructing a multi-dimensional feature vector based on the preprocessed multi-source remote sensing data, wherein the multi-dimensional feature vector comprises a spectral feature, a texture feature, a radar scattering feature and a forest three-dimensional structure feature; inputting the multi-dimensional feature vectors into an auto-encoder model, and performing nonlinear dimension reduction on high-dimensional features by an encoding layer to obtain low-dimensional deep feature representation; constructing an optimization model taking the above-ground biomass inversion error as a target function, and performing collaborative global optimization by adopting a simulated annealing algorithm to obtain an optimal feature combination and an optimal model parameter; an XGBoost above-ground biomass inversion model is constructed, and the inversion model is trained by using actually measured sample above-ground biomass data; and spatial continuous estimation of the above-ground biomass is realized. According to the method, the biomass inversion precision under the complex artificial forest condition is remarkably improved.

This application discloses a method, device, and medium for comprehensive credit profiling of highway user groups, relating to the field of credit scoring technology. The method includes: calculating the original credit index vector of a user's multidimensional credit status within a time window based on multi-source traffic data; constructing and training a multilayer perceptron autoencoder, inputting the original credit index vector into the trained multilayer perceptron autoencoder for nonlinear dimensionality reduction, and outputting the user's embedded feature vector in the low-dimensional space; calculating the modulus of the embedded feature vector as the user's individual credit index within the time window; modeling group credit characteristics based on the individual credit index to generate a group credit vector; and weighting and summing the features in each dimension of the group credit vector to generate the group credit index of the user group within the time window. This application achieves a full-process credit profiling process from dynamic characterization of individual credit, modeling of group credit structure, to analysis of evolutionary trends through the above method.

The invention discloses a multivariate quality detection method and system based on wavelet packet decomposition and kernel principal component analysis, social security and a medium, and the method comprises the steps: setting a collection frequency, and obtaining a rolling bearing vibration signal data set through collecting vibration data of different positions in real time; the method comprises the following steps: carrying out multilayer wavelet packet decomposition on a rolling bearing vibration signal, and mapping nonlinear data to a high-dimensional space by using a Gaussian kernel function on high-frequency and low-frequency signals decomposed from the vibration signal and sub-bands after wavelet packet decomposition; performing nonlinear dimensionality reduction modeling on the sub-band signals one by one, extracting principal components, calculating corresponding principal component scores, and calculating corresponding Tstatistics; determining the control limit of each sub-band signal by calculating the historical data distribution of the Tstatistics; and comparing the Tstatistic calculated in real time with the control limit, judging whether the process is abnormal or not, and if the Tstatistic exceeds the control limit, judging that the process is abnormal. According to the invention, weak anomalies under complex working conditions can be effectively identified.

This invention discloses a method and system for detecting diverse passenger demands in subway systems based on the BERTopic model. First, the original passenger text is collected and segmented, then transformed into high-dimensional semantic vectors using an embedding model. Next, nonlinear dimensionality reduction is performed using the UMAP model, followed by density clustering of the low-dimensional vectors using the HDBSCAN algorithm to form topic clusters. Then, the original text undergoes Chinese preprocessing, including cleaning and word segmentation, to obtain standardized text. Based on the clustering results and standardized text, topic keywords are extracted for each cluster using CountVectorizer and c-TF-IDF, and the MMR method is used to optimize the keywords to improve their representativeness and diversity. Finally, a visual report is generated to output the analysis results. This invention organically combines deep semantic modeling with the characteristics of Chinese text processing. Through modular modeling and system integration of various functional areas, it can automatically and efficiently identify semantically consistent and interpretable diverse topics from massive amounts of passenger demands, providing accurate data support for subway operation decisions.

The invention discloses a feature selection and dimension reduction method and system for multi-omics data fusion, and the system comprises a multi-omics data collection and heterogeneous preprocessing module which is used for carrying out the omics specific preprocessing of genome, transcriptome, proteome and metabolome data; the cross-omics heterogeneous graph neural network fusion module is used for constructing a heterogeneous graph and carrying out information fusion by adopting a heterogeneous graph attention network; the feature selection module based on multi-task sparse learning is used for selecting a key feature subset from the fused features; a non-linear dimension reduction module based on an adversarial variational auto-encoder and used for performing dimension reduction on the key feature subset to a low-dimensional submerged space; and a result visualization and biological interpretation module. According to the method, efficient fusion and dimension reduction of heterogeneous and high-dimensional multi-omics data are realized, the accuracy of feature selection and the interpretability of results are improved, and an effective calculation tool is provided for disease typing and biomarker discovery.

The invention provides a network attack pattern recognition method based on machine learning, which belongs to the technical field of network security, and comprises the following steps of: extracting an initial feature set by collecting a network flow data packet, and calculating Hausdorff dimensions under a plurality of time window scales by using a box counting method to construct a fractal dimension spectrum sequence; combining the initial feature set with the fractal dimension spectrum sequence, performing nonlinear dimensionality reduction and feature selection to form a final feature vector, performing balance processing on an unbalanced training data set by using a boundary synthesis minority class oversampling technology, and training a traffic behavior recognition model by using a sparse connection learning framework based on dynamic topology reconstruction. In the real-time detection stage, the dynamic threshold is adaptively adjusted according to the deviation degree of the fractal dimension spectrum to judge the attack traffic, and the technical problem that the real-time performance and the accuracy of attack detection in the high-speed encryption network traffic environment are difficult to consider at the same time is solved.

The invention relates to the technical field of power equipment state monitoring, and discloses a power equipment remote operation, maintenance and diagnosis method, which comprises the following steps of: firstly, off-line mapping a full-spectrum feature vector of health equipment under active disturbance to a low-dimensional hidden space through a nonlinear dimensionality reduction model to construct a health manifold reference model; state deviation is judged by calculating the statistical distance between the current state point and the health manifold, and when the state deviation occurs, a causal inference process is activated: a fault migration vector representing a fault evolution path is calculated; aiming at the candidate fault subsystem, actively injecting a differential disturbance signal to obtain a causal response vector of the candidate fault subsystem; and finally, quantitatively comparing the similarity between the fault migration vector and the causal response vector, and positioning the subsystem with the maximum causal contribution degree as a fault source. According to the method, state monitoring and active causal inference are combined, the crossing from fault detection to physical root positioning is realized, and the diagnosis sensitivity and accuracy are remarkably improved.

The application provides a student performance prediction method based on a woa (whale optimization algorithm), a T-SNE (t-distribution stochastic neighbor embedding) algorithm and an LSTM model, and comprises the following steps: A, obtaining a student data set and cleaning abnormal data; B, using the woa algorithm for data weighting; C, using the T-SNE algorithm for nonlinear dimension reduction processing of the data; and D, predicting the processed data through an optimized Adam-LSTM model to obtain student predicted performance. The application provides a student performance prediction method based on a deep learning model LSTM and a data set processing algorithm WOA and T-SNE. In the experiment, an Adam optimizer is fused with an LSTM model, the weight of the model can be more efficiently updated, and the stability and prediction efficiency of the LSTM model are improved. Meanwhile, in order to make the model pay more attention to important samples in the training process and improve the performance of the model, the WOA algorithm is applied to the weighted processing of the data set. In this paper, the T-SNE dimension reduction technology is also used to transform the original data by generating a new attribute set, which not only significantly reduces the number of attributes, but also ensures that most of the effective information in the original data is retained, further optimizes the LSTM model and obtains more accurate student ability prediction results.

The invention discloses a method and system for identifying a dynamic degradation mechanism of a wide bandgap semiconductor based on high-frequency electrical fingerprints and deep learning, and the method comprises the steps: collecting high-frequency electrical fingerprints, i.e., synchronously collecting the transient high-dimensional waveform data of a power device switch on line; a deep feature vector extraction step: performing nonlinear dimension reduction processing on the high-dimensional waveform data by using a one-dimensional convolution auto-encoder model, and outputting a low-dimensional deep feature vector; and a degradation mechanism classification and identification step: fusing the deep feature vector and the real-time temperature data, inputting the fused data into a classifier model for reasoning, and outputting the current main degradation mechanism category and confidence of the device. The system comprises a high-frequency electrical fingerprint acquisition module and an embedded processing module which are used for realizing the method. The method can overcome the defects that online diagnosis cannot be achieved and diagnosis information is fuzzy in the prior art, and has high sensitivity, high robustness and real-time online diagnosis capacity.

The application belongs to the field of magnetic resonance sounding signal noise filtering, and is a kind of magnetic resonance sounding signal extraction method based on intelligent optimization manifold learning, the parameter group of the local linear embedding manifold learning method is initialized, the genetic algorithm in the intelligent optimization algorithm is used, the signal-to-noise ratio is taken as the fitness function, and the parameter group in the local linear embedding manifold learning method is optimized, the local linear embedding manifold learning method uses the optimized parameter group to sequentially perform first processing and second processing on the magnetic resonance sounding signal, removes random noise, and obtains the final denoised magnetic resonance sounding signal. The application effectively retains the signal characteristics through the nonlinear dimension reduction of manifold learning, avoids the information loss caused by frequency band selection, maintains the local relationship between data points in the dimension reduction process, ensures that adjacent points in m-dimensional space remain adjacent in d-dimensional space, only compresses and filters noise, and realizes signal extraction.

A multi-model fusion method for photovoltaic power prediction based on dimensionality reduction and clustering is proposed. This method sequentially performs four steps: feature selection and data preprocessing, clustering and prediction model construction, hyperparameter and weight co-optimization, and model training and prediction. First, key meteorological features are selected using Spearman correlation coefficient. Then, UMAP is used to perform nonlinear dimensionality reduction while preserving the local and global structure of the data. GMM clustering combined with the BIC criterion is used to perform soft clustering on the dimensionality-reduced data to divide the data into clusters with similar power output characteristics, and an LSTM-Attention prediction model is built for each cluster. Bayesian optimization (BO) and quantum particle swarm optimization (QPSO) are introduced to achieve global optimization of model hyperparameters and fine-tuning of network weights, respectively. Finally, the BO-QPSO-LSTM-Attention model is trained, and the photovoltaic power prediction value is obtained by using mean squared error as the loss function and Adam as the optimizer.

The present application relates to the technical field of image segmentation, and particularly relates to a quantitative evaluation method of momordica grosvenori based on image segmentation. The method comprises the following steps: performing spectral decomposition on momordica grosvenori data image information by using a spatial-frequency feature component decoupling algorithm to generate decoupled light and shadow features; extracting cross-scale deep semantic information to construct a semantic probability distribution field, performing edge tensor correction and generating a pixel-level fruit body lesion semantic deconstruction set; performing contour topology mapping and spatial domain gravity field analysis to obtain a geometric shape vector represented by a geometric moment operator, a lesion distribution density and a surface texture loss weight value; and performing a preset trait residual energy level discriminant to perform nonlinear dimension reduction mapping and output an evaluation grade. Through deep coupling of signal decoupling, semantic field analysis and geometric topology measurement, the present application establishes a logical mapping link from pixel semantic features to physical energy levels, realizes image signal representation of momordica grosvenori irregular edges in a complex background, and realizes nonlinear coupling between pixels and physical trait residual energy level attributes.

The invention provides a classification method and system for detecting abnormal network traffic, and relates to the technical field of network security, and the method comprises the steps: obtaining and preprocessing an original network traffic data packet, and forming an initial traffic data set; constructing and training a stacked sparse denoising auto-encoder network, and performing nonlinear dimension reduction and robustness feature representation learning on the high-dimensional basic features to obtain potential spatial feature vectors; a reconstruction error of each network flow is calculated, and abnormal flow classification is realized according to a combined criterion of a reconstruction error threshold and dynamic clustering analysis; a dynamic abnormal flow prediction model is constructed and applied, future probability distribution is predicted, and an early warning signal is generated; the system comprises a data acquisition module and the like. Through deep feature compression and a three-level progressive classification mechanism, comprehensive optimization of high anomaly detection rate, low false alarm rate and low calculation overhead is realized, and the method is suitable for various network deployment environments.

The invention discloses an electricity market quantitative transaction method and system based on interpretable artificial intelligence and probability prediction, electronic equipment and a storage medium. The method comprises the following steps: firstly, acquiring high-dimensional time sequence data of supply and demand and electricity price of an electricity market, performing nonlinear dimension reduction by using a time sequence convolutional network auto-encoder, and extracting low-dimensional time sequence features; training an explainable spot electricity price point prediction function based on a symbol regression technology; through a Bootstrap resampling model residual error, point prediction is expanded into a probability density function containing uncertainty information; further constructing a buying and selling bidirectional optimization decision model fusing expectation, standard deviation and quantile constraints, and generating a to-be-audited transaction scheme; and finally, outputting a final transaction instruction through a multi-stage auditing, sorting and duplicate removal mechanism. The method has the advantages of high prediction precision, model transparency and risk quantification capability, and improves the reliability, intellectualization and prederability of quantitative transaction of the electricity market.

This invention discloses an epilepsy prediction system based on feature channel fusion and deep learning. The system employs a T-distributed random nearest neighbor embedding algorithm (t-SNE) with nonlinear dimensionality reduction to fuse feature channel information from epileptic EEG signals. Time-domain and frequency-domain information obtained through short-time Fourier transform is used as feature input to a deep residual shrinking neural network. By identifying the interictal and preictal phases of epilepsy, the system predicts seizures. This method improves feature dimensionality and classifier design, eliminating the need for manual feature extraction and improving the representation of feature information. This provides a new approach for the clinical application of epilepsy prediction.

The present application belongs to the technical field of mutual inductor measurement error identification, and particularly relates to a current transformer measurement error online identification method, device, equipment and medium, wherein the method comprises decomposing the three-phase current signal at the secondary side of the current transformer by VMD, and screening out the residual signal after decomposition; performing wavelet decomposition on the residual signal to obtain an approximate coefficient set; performing nonlinear dimension reduction processing on the approximate coefficient set by a t-distributed stochastic neighbor embedding algorithm to obtain low-dimensional feature data; estimating the probability density distribution of the low-dimensional feature data, and combining the preset confidence to determine the measurement error overrun threshold; comparing the low-dimensional feature data with the overrun threshold, if the low-dimensional feature data is less than the overrun threshold, it indicates that the current transformer is in a normal operating state at this time; otherwise, it is considered that the current transformer may be in an abnormal operating state at this time, solving the problem of inaccurate online identification of current transformer measurement error in the background technology.