26results about How to "Solve the imbalance" patented technology

The invention discloses a bearing fault diagnosis method integrated with a support vector dynamic neural network, and the method comprises the following steps: obtaining vibration data of a key position of a bearing of a rotating machine, carrying out the data preprocessing of the vibration data, and obtaining a preprocessed vibration signal data set; constructing a support vector machine model with a quadratic programming problem form based on the vibration signal data set, and designing a corresponding penalty function and a Lagrange function to simplify the support vector machine model; designing a variable parameter dynamics solving equation, and establishing a final integrated support vector dynamic neural network bearing fault diagnosis model to solve the support vector machine model; and simultaneously training a plurality of support vector dynamic neural network bearing fault diagnosis models based on a bagging classifier method, carrying out bearing state diagnosis on the vibration signal data set, and outputting a diagnosis result. According to the invention, the accuracy of the machine learning method, the convergence proof of the neural network solver and the interpretability of the intelligent fault diagnosis model are improved.

The invention relates to the technical field of network security detection systems, in particular to a multi-modal fusion network abnormal behavior detection system and method. The multi-modal feature extraction module is connected with the multi-modal data collection module; the multi-modal fusion module is connected with the multi-modal feature extraction module and used for fusing the feature vectors Vi; the behavior recognition module is connected with the multi-modal fusion module and is used for analyzing the fusion vector F, judging whether the host to be tested has an abnormal behavior or not, and judging the category and confidence of the abnormal behavior; the active learning annotation screening module is used for screening and annotating unannotated data so as to optimize the model performance; multi-modal data collection provides rich original information for feature extraction, deep feature extraction provides high-quality feature representation for multi-modal fusion, a fusion method of an attention mechanism makes full use of the features, and an active learning module continuously optimizes the performance of the whole system.

The invention discloses a cross-regional load prediction method, system and device based on a shared encoder, and a storage medium. The method comprises the following steps: constructing a common covariable pre-training data set and a specific region data set; constructing a time sequence prediction backbone network, wherein the time sequence prediction backbone network comprises an input coding layer and a shared encoder based on a self-attention mechanism; designing a multi-task learning framework which comprises a multi-task learning network architecture and a dynamic weighting loss function; pre-training the encoder by using the common covariable pre-training data set; performing fine adjustment on the encoder in a specific region by utilizing the specific region data set of each region and combining with a multi-task learning framework; performing time sequence prediction on large-scale and cross-regional load data by using an encoder; and evaluating the performance of the encoder according to the time sequence prediction result. According to the method, the problems of data sparsity, regional difference, long-term dependency relationship capture, data imbalance and the like in cross-regional load prediction are effectively solved, and the prediction precision and the model generalization ability are remarkably improved.

The invention provides a management method and system for automatically generating standardized OT data based on a large model. The method comprises the steps that S1, an OT metadata ontology library is constructed into an OT term vector space according to standard regularized metadata meeting preset requirements; s2, mapping a natural language instruction to an OT term vector space through a large model to obtain standard structured OT data; and S3, carrying out quality verification and enhancement processing on the standard structured OT data. According to the method, the data consistency can be improved, the naming conflict in the integration process is solved through the unified cloud standard specification, and the data interoperability is remarkably enhanced; the efficiency revolution can be implemented, the data cleaning workload is effectively reduced, the qualitative change from a project system to productization is realized from traditional manpower integration, and the basic support of high-quality training data is provided for large-scale industrial deployment.

The invention discloses a training data balance processing method and device, equipment, a medium and a product. The invention relates to the technical field of data processing. The method comprises the following steps: when a user in a user set is associated with an object in an object set, generating associated data according to the associated user and object, and adding the associated data into an associated data set; acquiring other users of the same category as the target user corresponding to the to-be-supplemented data of the associated data set; obtaining a target object associated with the target user in the associated data set; according to each target object, determining candidate objects of the target user in other objects associated with the other users in the associated data set; and generating associated data according to the target user and each candidate object, and adding the associated data into the associated data set. The embodiment of the invention can balance the training data of the model.

The invention discloses a multi-coal-source coal blending data processing method based on a WSMOTE algorithm, and belongs to the technical field of coal processing data processing, and the method comprises the steps: data collection and classification: collecting multi-dimensional data of a raw coal floating and sinking test, a coal blending test and the like, carrying out the preprocessing, dividing the data into a missing data set and a complete data set, carrying out the interpolation optimization of a missing value, and carrying out the calculation of the complete data set. And carrying out adaptive processing according to data distribution types, or carrying out weighted combination normalization according to feature importance, analyzing the number and distribution characteristics of minority class samples, dynamically adjusting the neighbor sample size and the new sample synthesis amount, and completing data enhancement and standardization. According to the method, standardized parameters are dynamically adjusted to adapt to data distribution changes, abnormal values are accurately processed, the combined interpolation model is subjected to weight optimization, errors are reduced, and data integrity is improved. WSMOTE parameters are adaptively combined with data distribution to synthesize samples, data are balanced, and the over-fitting risk is reduced. In practical application, the quality of coal blending model training data can be improved, and optimization of a multi-coal-source coal blending scheme is assisted.

The invention discloses a lightweight network traffic intrusion detection method fusing TrafficFormer-knowledge distillation, and aims to solve the problem of model lightweight demand and network traffic data imbalance in a resource limited scene. The method comprises the following steps: firstly, segmenting and filtering pcap-format original traffic data, dividing a data set according to a preset proportion, and then carrying out data balance on a training set by adopting an RIFA method; performing effective data extraction and byte-level processing on all the data sets to generate a standardized token sequence; then, a teacher model is built through fine tuning on the basis of a TrafficFormer pre-training model, and the model is pre-trained through MBM and SODF double-self-supervision tasks and has the excellent flow feature representation capacity; then, a four-layer Transform encoder is adopted as a lightweight student model, and teacher model knowledge is migrated to the student model through a double-loss function of KL divergence distillation loss and Focal supervision loss; and finally evaluating the performance of the obtained student model.

The invention, which relates to the photoelectric display field, discloses a double-current-gate regulation light-emitting device comprising a first surface gate, a first electron transport layer, a first surface electrode layer, a second electron transport layer, a light-emitting composite layer, a second hole transport layer, a second surface electrode layer, a first hole transport layer, and a second surface gate. A first bias power supply is applied to the first surface grid and the first surface electrode layer, and a second bias power supply is applied to the second surface grid and the second surface electrode layer; a driving signal is applied between the first surface electrode layer and the second surface electrode layer and is used for driving hole-electrons to be compounded on the light-emitting composite layer and emitting light; the first configuration power supply and the second configuration power supply can regulate and control the electron hole migration amount in the light-emitting composite layer. According to the invention, the current gate structures are additionally arranged on two sides of a traditional light-emitting device, so that the problem of imbalance of electron carrier mobility and hole carrier mobility in the light-emitting composite layer can be effectively solved, the electron-hole light-emitting composite energy efficiency is effectively improved, and meanwhile, the light-emitting brightness and the light-emitting efficiency can be optimized at the same effect.

This invention discloses a method, system, device, and medium for automatic extraction of black soil layer interfaces using ground-penetrating radar (GPR) based on an improved U-Net, belonging to the field of GPR data processing and intelligent interpretation technology. The method includes acquiring and preprocessing GPR B-scan image data, constructing a training dataset, building a layer extraction model based on an improved U-Net neural network model, constructing a hybrid loss function including at least two combinations of weighted binary cross-entropy loss, Dice loss, structural similarity loss, and smoothness loss, training the layer extraction model based on the training dataset and the hybrid loss function, and inputting the GPR B-scan image data to be detected into the trained layer extraction model to achieve automatic extraction of black soil layer interfaces using GPR. This invention effectively overcomes signal attenuation and noise interference in GPR images, achieving high-precision and continuous soil layer interface extraction.

The invention relates to a document tampering detection method based on dynamic kernel fusion and adaptive gradient modulation, and belongs to the technical field of digital forensics and computer vision. The method comprises the following steps: obtaining an RGB image of a to-be-detected document, constructing a feature encoder containing a double-branch feature extraction and feature updating module, taking ConvNeXt V2-Base as a backbone, combining visual and frequency domain features with a cross-domain feature calibration module, and obtaining optimized multi-scale features through learnable wavelet decomposition; a frequency domain adaptive feature decoder is constructed, dynamic kernel fusion is completed through Fourier frequency band grouping weight, kernel element precise modulation and space-frequency band dynamic matching, and a tampering prediction map is output; and during training, joint optimization of adaptive gradient cosine loss, cross entropy loss and LoWitz loss is adopted. According to the method, the frequency abnormity and tiny traces of document tampering can be accurately captured, the problem of class imbalance is effectively relieved, and the method has high detection precision and robustness in different compression scenes.

The application relates to an energy consumption anomaly monitoring method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring first data, which is the energy consumption data of a building within a first time. The first time is segmented to obtain a plurality of time periods, and each time period in the plurality of time periods has the same time interval. A first matrix is obtained based on the plurality of time periods, and the first matrix is an energy consumption data matrix arranged based on the plurality of time periods. The first matrix is supplemented to obtain a second matrix, and the second matrix is an energy consumption data matrix meeting the training requirements of a convolutional neural network. The second matrix is subjected to self-encoding training through the convolutional neural network, and the trained second matrix is clustered through a K-means algorithm to obtain a corresponding convolutional neural network model. The convolutional neural network model is called to classify newly input building energy consumption data to determine whether the newly input building energy consumption data is abnormal.

The present application relates to the field of Internet of Things technology, and discloses a method for controlling the dynamic balance of a climbing frame, including: step S1: connecting a balance control system to a hydraulic press driving the climbing frame; step S2: controlling the hydraulic press to start through the balance control system; step S3: setting the distance A and direction that the climbing frame is to run through the balance control system; step S4: setting the deviation B of the allowed running result of the climbing frame; step S5: sending a running instruction to the hydraulic press through the balance control system; step S6: reading the displacement output value of each hydraulic press; step S7: controlling the start and stop of each hydraulic press based on the calculation result in step S6; step S8: judging whether the actual running distance P of the climbing frame satisfies P≥A±B, if yes, stop running, otherwise return to step S5. The present application solves the imbalance problem in the lifting process of the existing climbing frame, and improves the safety and construction efficiency of the climbing frame.

The invention relates to the technical field of network security, in particular to a network situation awareness method and system based on user behavior analysis, and the method comprises the following steps: collecting a network data packet in real time, preprocessing the collected network data packet, and extracting flow data and log data; extracting behavior characteristics from the flow data and the log data, and constructing an organization diagram; inputting the organization diagram into a pre-trained diagram convolutional network model, and outputting to obtain a network security situation awareness result; the graph convolutional network model is integrated with a first generative adversarial network. By implementing the security situation awareness technology based on user behavior analysis, the capabilities of identifying, analyzing and responding to complex security threats in a large-scale network environment are effectively improved. The graph convolutional network model can self-learn and adapt to a new threat mode, the detection strategy is automatically adjusted, the threat detection accuracy and recognition capability are improved, and the detection efficiency is also improved through automatic detection.

This invention discloses an upward-rotating and downward-rotating spray gun, belonging to the field of agricultural irrigation technology. Based on the original double-rotating spray gun, this invention is optimized for high-sediment-content water environments. It includes: an outer cylinder (1), a central water supply pipe (2), a drive rotating shaft (3), a planetary reduction mechanism (4), a high-speed recoil drive arm (5), and a low-speed spraying arm (6). The improvements are as follows: the base (7) is provided with first and second sealing rings (701, 702) to isolate sand and dust; the upper end face of the base and the central groove are provided with sand discharge holes (703); the lower end of the first water distribution chamber (501) is provided with a double sealing ring layer (504), and the high-speed recoil drive arm at the corresponding position is provided with a debris discharge hole (505); the top of the high-speed recoil drive arm is provided with a sealing thread sleeve (506) to cooperate with the upper sealing ring (507); the inner wall of the central water supply pipe (2) is provided with ribs (202) that have both reinforcement and flow guiding functions; the low-speed spray arm (6) consists of two bifurcated arms of different heights, with nozzles of different sizes installed at angles of 30-38° and 5-20° respectively to balance the reaction force. While maintaining the advantages of the original double swirl field, the present invention significantly improves the anti-clogging ability and operational balance, and is especially suitable for surface water irrigation with high sand content.

The invention discloses a somatosensory enhanced muscle carving handle and a T-shaped rod handle for strength training, the somatosensory enhanced muscle carving handle for strength training comprises a handle, the peripheral wall of the handle is suitable for a palm and fingers to surround and hold; the protruding part is arranged at the end of the handle, and after the palm center and the fingers surround and hold the handle, the thumb can tilt backwards to lift the jaw and abut against the outer wall of the protruding part to exert force in the strength training process. Due to the unique design, when a user pulls, the hand can be naturally in an outward inclined state, namely, the center line of the conventional natural holding of the hand is inclined relative to the center line of the handle, so that the muscles on the outer side of the back are better activated, and a stronger muscle engraving effect is provided; in addition, due to the design of the protruding parts, the fatigue feeling of the hands in long-time training is reduced, the durability and safety of training are improved, and the user can obtain better training experience and effect during strength training.

The invention discloses a multi-source attack detection method for computing power cluster security protection, and belongs to the technical field of host security attack detection. The method comprises the following steps: inputting an operating system auditing log of a computing node in a computing power cluster to be identified into a pre-trained time sequence anomaly detection model to calculate an anomaly probability score of the operating system auditing log, when the anomaly probability score exceeds a threshold value, judging that the operating system auditing log is attacked, and otherwise, judging that the operating system auditing log is safe; the pre-trained time sequence anomaly detection model is obtained through the following steps: collecting an operating system audit log of each computing node, analyzing an event in the log into a standardized quintuple event, and obtaining a global cause and effect graph based on the event; active entity nodes in the global causal graph are screened, and multi-dimensional feature coding is carried out to generate a feature sequence; and inputting the feature sequence into a pre-constructed time sequence anomaly detection model to obtain a trained time sequence anomaly detection model. According to the invention, the problem that the existing detection technology is difficult to restore a cross-host complete attack link is solved.

The invention discloses an intelligent voltage regulating transformer fault diagnosis method and system based on an improved CNN-BiLSTM model. The method comprises the following steps: firstly, collecting historical data of a voltage regulating transformer for preprocessing; then training is carried out based on historical data of the voltage regulating transformer to obtain a transformer fault diagnosis model, the transformer fault diagnosis model is an improved CNN-BiLSTM model, and an ECA-TAM attention mechanism is introduced into the CNN-BiLSTM model; and finally, inputting the current operation data of the voltage regulating transformer into the transformer fault diagnosis model to obtain a fault diagnosis result of the voltage regulating transformer. According to the method, an ECA-TAM attention mechanism is introduced into the CNN-BiLSTM network model for voltage regulating transformer fault diagnosis, purer and more representative feature input can be provided, interference of redundant information on time sequence modeling is reduced, the key dynamic learning ability is improved, efficient and accurate diagnosis of voltage regulating transformer fault types is achieved, and the fault diagnosis efficiency is improved. The accuracy and timeliness of transformer fault prediction are improved, and stable and safe operation of an electric power system is guaranteed.

The invention relates to the technical field of target recognition, discloses an unmanned aerial vehicle recognition model training method and an unmanned aerial vehicle recognition method, and aims to solve the problems of high cost and low efficiency of an existing unmanned aerial vehicle recognition scheme, and the scheme mainly comprises the following steps: segmenting radar data of an unmanned aerial vehicle track to obtain a plurality of first unmanned aerial vehicle samples; correspondingly adding the feature data of each trace point on any two first unmanned aerial vehicle samples according to the same random weight, and then adding random perturbation to obtain a plurality of second unmanned aerial vehicle samples; segmenting the radar data of the non-unmanned aerial vehicle flight path to obtain a plurality of non-unmanned aerial vehicle samples; randomly selecting a second unmanned aerial vehicle sample as a third unmanned aerial vehicle sample; and constructing an identification sample set based on the first unmanned aerial vehicle sample, the third unmanned aerial vehicle sample and the non-unmanned aerial vehicle sample, and training an unmanned aerial vehicle identification model based on the identification sample set. The training efficiency and accuracy of the unmanned aerial vehicle identification model are improved, the cost is reduced, and the method is particularly suitable for radar detection.

The application provides a midpoint potential balance control method based on a notch filter and a compensator, and belongs to the technical field of converters, and comprises the following steps: determining a difference signal of a positive midpoint bus capacitor voltage signal and a negative midpoint bus capacitor voltage signal of a three-level converter DC side, and inputting the difference signal into a notch filter for processing; inputting the processed difference signal as a negative feedback signal into a compensator together with a zero reference signal; dividing a compensator output signal by a specific proportional coefficient as a zero sequence modulation signal, and superimposing the zero sequence modulation signal into three-phase modulation signals of a current converter, and generating a PWM signal according to the three-phase modulation signals superimposed with the zero sequence modulation signal, so as to control the midpoint potential balance of the three-level converter DC side; the application generates a zero sequence modulation signal by using a notch filter and a compensator, and superimposes the zero sequence modulation signal on three-phase alternating current modulation signals, so as to control the current on the neutral line, and the problem of midpoint potential imbalance of the three-level converter DC side capacitor can be solved.

The invention discloses an ejection structure for a mobile phone support product, and relates to the technical field of mobile phone support product ejection, the ejection structure comprises a mold base and a first assembly mounted on the mold base, the cross section of the mold base is U-shaped, and the first assembly comprises an ejection transmission assembly, a mold forming assembly and a guide positioning assembly; the ejection transmission assembly comprises an ejector pin bottom plate and an ejector pin panel, the ejector pin bottom plate is in transmission connection with the ejector pin panel through a guide column, and the ejector pin panel is provided with a positioning hole used for installing the ejection execution assembly; through cooperative operation of a sliding rod, a spiral groove, a limiting body and the like in the fourth assembly, linear ejection force restored by a spring is converted into stable rotary lifting motion, the ejection speed of the inclined ejection block is further reduced through the design of a large spiral lead, the lifting stability is improved, uniform power transmission is achieved in cooperation with composite ejection motion of the inclined ejection block in the 45-degree direction, and the service life of the inclined ejection block is prolonged. And the problem of unbalanced ejection of the small bracket is thoroughly solved, and the product deformation rate is reduced to an extremely low level.

The invention discloses a micro-expression recognition method and device based on AU guidance. The micro-expression recognition method comprises: preprocessing a micro-expression sample video to obtain a micro-expression sequence; generating a facial muscle movement text description for each micro-expression sample through an AU coding label of a public data set and a facial action coding system; sending the processed micro-expression sequence to a visual encoder to obtain a coded visual motion feature; sending the motion text description to a text encoder to obtain a text motion feature; by generating AU-guided fine-grained micro-expression facial muscle movement text description, and using CLIP contrast loss to perform cross-modal semantic alignment, the model's perception and capture of weak local micro-expression movements can be improved, significant spatiotemporal micro-expression movement features can be learned, the movement features are effectively converted into emotional features, and the Focal Loss classification loss is used to solve the problem of micro-expression data class imbalance, thereby greatly improving the recognition performance of the model.

The invention provides a soil layer identification method based on a random forest and a synthetic minority class oversampling technology, and is suitable for the field of soil layer identification. Comprising the steps of collecting data based on the underwater deep cement mixing pile, processing the data, obtaining a balanced sample data set, preparing the data, obtaining optimal random forest algorithm parameters, constructing a classification model, testing and optimizing the classification model, and applying the model. The soil layer identification method based on the random forest and the synthetic minority class oversampling technology has high classification precision and wide applicability, and can be widely applied to the field of soil layer identification.

The invention provides an oversampling multi-cluster fusion data prediction method and device, equipment and a medium, and relates to the field of data prediction. The method comprises the following steps: obtaining an original data set, preprocessing the original data set, and dividing the original data set into minority data and majority data; clustering the minority class data by adopting different clustering algorithms, and adding the generated clustering center data into the minority class data; carrying out IF-SNNDPC clustering on the minority class data after the clustering center is added, and calculating the oversampling number of each sub-cluster; in each sub-cluster, calculating a centroid between a point of two randomly selected non-clustering centers and the clustering center, and adding the centroid as a new sample into the minority class data until the sample quantity required by the corresponding sub-cluster is met, so as to obtain final minority class data; and inputting the final minority class data and the final majority class data into a pre-trained XGBoost model to obtain predicted data. The number of samples of a few categories can be increased, and the problem that the categories of sample data are unbalanced is solved.

The application discloses a panoramic service intelligent sensing and accurate classification method, comprising the following steps: a sensing probe collects characteristic data of different service types of service end information, and divides the characteristic data into labeled data and unlabeled data; a feature subset of the labeled data is constructed based on a mixed feature selection method, and a global optimal feature subset is obtained; based on a semi-supervised learning model of an improved K-means clustering algorithm, service end infection data is output; the sensing probe collects, stores and transmits service request information of a user end, and the service request information is used as prior information; a trained convolutional neural network model is used to output high-level features of the prior information and the service end infection data; and a trained random forest service sensing and identification model is used to classify the high-level features. The panoramic service intelligent sensing and accurate classification method can improve the accuracy of mixed service identification.

A kind of multi-flow chain type perception enhanced multimodal aspect-level sentiment analysis method is referred to as MCPE model, it is related to multimodal sentiment analysis technical field.The present application solves the problem of low accuracy of fine-grained sentiment analysis caused by the difference of information density within modal and the imbalance of information between modal in the prior art.The present application comprises: obtaining a multi-modal text-image pair to be analyzed, inputting the text-image pair into the trained MCPE model to obtain aspect words and their sentiment polarity;The MCPE model comprises: a feature extraction module, a chain enhancement module, a multi-flow interaction module and a classifier;Text features are extracted by BART and image features are extracted by Faster R-CNN in the feature extraction;The chain enhancement module suppresses image and text noise and enhances fine-grained semantics through IFE-TFE double-chain architecture;The multi-flow interaction module realizes the dynamic complementary fusion of text reasoning and visual evidence by using bidirectional cross-modal attention;The classifier outputs the analysis result based on the fusion features, and does not require external tools.The present application is suitable for sentiment analysis of social media comments.

The invention discloses a hydrogen production mode identification method and system based on a tree classifier, and belongs to the technical field of hydrogen analysis, the hydrogen production mode identification method based on the tree classifier comprises the following steps: S1, extracting the original component content of a hydrogen sample; s2, constructing engineering features based on the original component content, wherein the engineering features comprise a component proportion, a composition entropy, logarithm transformation, a component interaction item and a principal component; and S3, hydrogen production mode identification: taking the engineering characteristics as input, synthesizing minority class oversampling for relieving class imbalance, and performing hydrogen production mode judgment through a tree classifier. By adopting the technical scheme, the hydrogen production mode can be quickly analyzed by utilizing the tree classifier based on the original component content and engineering characteristics of the hydrogen sample.