121 results about "Neural network analysis" patented technology

The invention relates to the field of network security, and particularly discloses a network security defense strategy optimization method based on machine learning, and the method comprises the steps: aggregating isolated security events into attack activity clusters through causal association analysis, constructing a dynamically evolved global attack graph, and improving a defense perspective from a discrete event to a full combat view. In order to realize foresight, a graph neural network is utilized to analyze a graph to identify attack tactics and predict the next intention. A hierarchical reinforcement learning framework is innovatively introduced in the decision-making stage; an upper-layer strategic agent formulates a macroscopic defense target based on a global situation; and the lower-layer tactical agent focuses on the related attack sub-graph under the strategic guidance, and selects and executes the specific tactical action which can reach the target most. The strategy and tactical separated decision-making mode ensures that each defense action serves a long-distance target, so that strategic passivity caused by only taking care of previous threats is avoided.

The invention relates to the technical field of robot control, in particular to a cross-modal perceptual driving body robot compliance control system which comprises the steps that a sensor is adopted to synchronously collect environment information, multi-source data bias is eliminated through a space-time alignment algorithm, a radial basis function neural network is adopted to analyze multi-modal fusion features, and a multi-modal model is obtained; human operation intention probability distribution is extracted to decompose a task into a path planning layer and a motion control layer, a collision-free trajectory is generated through an RRT algorithm, a high-fidelity physical engine is adopted to construct a virtual interaction scene, and robot learning results are shared through federal learning. According to the method, the problems of inaccurate perception, incoordination between intention recognition and interaction control, difficulty in control strategy verification, slow new task adaptation and difficulty in multi-robot learning result sharing caused by multi-source data deviation and large multi-modal semantic difference of the body robot in a complex environment are solved.

The invention discloses a boiler operation state intelligent monitoring method and system, and the method comprises the steps: collecting multi-source heterogeneous data in a boiler operation process, carrying out the cleaning and normalization processing of the multi-source heterogeneous data, extracting key features through wavelet transform, and obtaining standard multi-source heterogeneous data; modeling time series data in the standard multi-source heterogeneous data on the basis of an LSTM (Long Short-Term Memory) network, capturing a dynamic change trend of boiler operation to obtain time series characteristics, analyzing a hearth flame image and an infrared thermal image by using a CNN (Convolutional Neural Network), and extracting combustion state characteristics and thermal distribution characteristics; and based on the comprehensive state vector, a confidence interval of each feature dimension is calculated through a GMM Gaussian mixture model, an operation state deviation degree is analyzed and evaluated in combination with an entropy value, and when the deviation degree exceeds a preset threshold value, graded early warning is triggered. And the accuracy of boiler operation state intelligent monitoring is improved.

The invention relates to the technical field of the Internet of Things, and discloses a biological material component analysis method and system based on the combination of the Internet of Things and a neural network, and the method comprises the steps: obtaining material sample data of a to-be-detected biological material and environment state data of an environment where the to-be-detected biological material is located, and extracting an intrinsic attribute index of the to-be-detected biological material from the material sample data, calculating the basic substance content of the biological material to be detected; calculating the characteristic absorption rate of the to-be-detected biological material so as to analyze the characteristic absorption intensity of the to-be-detected biological material; constructing a spectral line graph corresponding to the to-be-detected biological material, and re-extracting fingerprint spectral information of the to-be-detected biological material from the spectral line graph; and inputting the fingerprint spectrum information into a biological detection unit of a preset Internet of Things, outputting a corresponding induction signal sequence by using the biological detection unit so as to identify molecular configuration information of the to-be-detected biological material, and analyzing the material composition of the to-be-detected biological material by using a trained neural network. The accuracy of biological material component analysis can be improved.

The invention discloses an engineering construction digital intelligent supervision management and control system, and relates to the technical field of engineering construction supervision. The system comprises an information acquisition module which is used for acquiring equipment state, image and environment data; the data fusion layer module is used for carrying out cleaning, alignment and Kalman filtering fusion on the multi-source data to generate a multi-modal construction data flow; a progress calculation and prediction unit, a deviation judgment unit and a digital modeling unit are arranged in the analysis monitoring module, and the analysis monitoring module is used for updating a progress state based on an XGBoost model, analyzing a node dependency relationship by using a graph neural network, calculating a deviation propagation risk coefficient and generating a hierarchical deviation signal in combination with three-level judgment conditions; and the management and control layer module executes hierarchical response according to the level of the deviation signal, and feeds back execution effect data to the front-end module to realize model optimization and closed-loop iteration. According to the invention, intelligent perception, risk prediction and adaptive management and control of the construction process are realized, and the real-time performance, accuracy and autonomy of project supervision are improved.

The invention relates to the technical field of personnel state monitoring, in particular to a non-contact intelligent monitoring method for a fatigue state of an operator, which comprises the following steps of: a, acquiring a face heat distribution image and a dynamic behavior video stream through a non-contact sensor array; b, extracting facial thermal features based on a multispectral fusion algorithm; c, analyzing behavior characteristics through a space-time attention neural network; d, fusing physiological and behavior characteristics, and outputting a fatigue level; and e, triggering a multi-level linkage early warning mechanism according to the fatigue level. A common direct contact type sensor brings discomfort to operators and affects normal work, manual observation has subjectivity, misjudgment is prone to occurring, and real-time and comprehensive monitoring cannot be achieved. Compared with the prior art, the non-contact physiological feature sensor and the non-contact behavior feature sensor can be used for detecting a worker together, direct contact with the worker is not needed, interference to the working state of the worker is reduced, long-time continuous monitoring can be achieved, and misjudgment is avoided.

The invention belongs to the technical field of power distribution station operation state monitoring, and discloses a station building degradation track prediction method fusing physical prior and spatio-temporal knowledge embedding. Comprising the steps of collecting multi-source heterogeneous data; constructing a physical constraint equation of key physical quantities in the station building equipment; constructing a space-time knowledge graph, adopting a graph attention network as a graph embedding model, embedding attention weights of nodes to neighbor nodes, and obtaining a node embedding vector set; performing feature extraction on the multi-source heterogeneous data to obtain mixed features; a degradation track prediction result is obtained based on physical information neural network analysis, health degree evaluation is carried out on the degradation track prediction result, a health degree score is obtained, and corresponding early warning is carried out according to the health degree score; according to the method, high-precision prediction of the degradation track of the station building equipment in a fault-free data scene is realized, and meanwhile, the physical rationality and operation and maintenance guidance of a prediction result are guaranteed.

The present application relates to the technical field of clinical nursing, in particular to a clinical nursing system for interventional surgery, in the present application, through the combination of preoperative medical history, intraoperative real-time physiological data and operation type, depth neural network analysis can accurately identify and predict potential complications, and the nursing plan can be adjusted according to the individual differences of patients, the nursing plan can be adjusted according to the individual differences of patients, the depth neural network can process complex physiological data and identify key physiological data, the long short-term memory network can dynamically adjust the scheme in the nursing process, the change trend of intraoperative data is deeply mined, the physiological index fluctuation is tracked, the abnormality is identified in time and the potential risk is predicted, the nursing decision can be updated synchronously with the physiological state of the patient, the working efficiency of the nursing staff is optimized, and the postoperative recovery effect and comfort of the patient are improved.

The invention discloses a fan mechanism modeling and optimizing method and system based on a neural network, and the method comprises the steps: building a model of a wind energy utilization coefficient and a thrust coefficient of a fan, and carrying out the steady-state simulation, so as to obtain a two-dimensional relation table model of the wind energy utilization coefficient and the thrust coefficient, establishing a fan aerodynamic model based on the two-dimensional relation table model; respectively establishing a fan transmission chain system model and a fan blade and tower coupling model based on the nonlinear pneumatic torque Tr and the nonlinear air thrust Ft in the fan aerodynamic model to obtain a complete fan mechanism model; and collecting actual operation data of the fan, analyzing the deviation between the fan mechanism model and the actual operation data of the fan based on the neural network, and correcting the deviation through the neural network to obtain an optimized fan mechanism model. According to the invention, the precision of fan modeling can be improved.

The invention relates to the technical field of e-commerce recommendation systems, and discloses a commodity recommendation system and method based on deep learning. The method comprises the steps of obtaining an interaction behavior sequence of a target user, performing framing and slicing to generate a behavior time period data frame set, and extracting multi-dimensional interaction features; entities and relationships are extracted from the commodity knowledge graph, the features are associated and matched with the entities, and a dynamic preference sub-graph is constructed; analyzing the sub-graph by using a multi-modal fusion neural network, and outputting an implicit intention vector; executing multi-hop reasoning in the knowledge graph based on the vector, and retrieving to generate a candidate commodity pool; and calculating potential association strength of the commodities and the users through cross-domain transfer learning, and performing sorting to generate a final recommendation list. According to the method, the short-term preference dynamic state of the user can be described in a fine-grained manner, and the implicit intention of the user is deeply mined through dynamic knowledge association and fusion analysis, so that the accuracy and individuation degree of a recommendation result are effectively improved.

The invention relates to the technical field of enterprise climate risk prevention and control, and provides an enterprise climate risk prevention and control system applying big data and artificial intelligence, and the system comprises a multi-source climate data collection module which integrates satellite, ground station and enterprise Internet of Things data through federal learning; the enterprise asset digital twin module integrates BIM / GIS to construct a dynamic asset model; the AI risk prediction engine analyzes climate-asset relevance by adopting a space-time convolutional neural network, and quantifies a physical / transformation risk in combination with physical simulation and a generative adversarial network; the dynamic risk assessment module is used for generating a risk thermodynamic diagram and calculating a supply chain interruption probability; and the adaptive response decision module outputs graded early warning and is linked with the insurance platform to trigger automatic claim settlement. The system verifies the credibility of data through a block chain, realizes a local disaster recovery decision when a network is interrupted by using an edge computing node, forms a'monitoring-prediction-evaluation-response-optimization 'closed-loop management and control mechanism, and solves the problems of data island, response delay and unknown risk prediction of a traditional scheme.

The invention relates to the technical field of numerically-controlled machine tool monitoring, in particular to a numerically-controlled machine tool monitoring method and system, terminal equipment and a storage medium, and the numerically-controlled machine tool monitoring method comprises the following steps: based on key parts of a numerically-controlled machine tool, monitoring tiny wear or deformation by adopting a time domain reflection method and a frequency domain reflection method; and the collected ultrasonic signals are analyzed in combination with a convolutional neural network, phase changes and abnormal features in the signals are identified, data fusion processing is carried out, and wear identification indexes are generated. According to the method, the advanced monitoring and analysis technology is adopted, comprehensive optimization of the performance of the numerical control machine tool is achieved, the time domain reflection method and the frequency domain reflection method are combined with convolutional neural network analysis, the detection precision of tiny abrasion or deformation is improved, dynamic optimization is achieved on clamping force adjustment through fuzzy logic control, the method adapts to different machining conditions, and the machining precision of the numerical control machine tool is improved. Workpiece damage is reduced, the machining quality is improved, and the machining precision is improved through deep diagnosis and dynamic compensation strategies of multi-axis synchronous errors.

The invention provides a power distribution automatic monitoring and early warning method and system based on voltage measurement, and belongs to the field of power distribution monitoring. Performing wavelet decomposition on a voltage signal of each node in the time-space matrix of the topological structure of the power distribution network to obtain a frequency band component of each layer and calculating an energy entropy of the frequency band component; regarding each layer of frequency band component as an independent individual, and integrating the lowermost layer of energy entropy from bottom to top to form an individual feature; analyzing correlation among individuals based on mutual information, screening individual pairs as independent states of nodes, and integrating all states to obtain node features; on the basis of node characteristics, the correlation between any two nodes is calculated by using an improved dynamic time warping distance to serve as a voltage fluctuation correlation coefficient; and constructing a dynamic fault propagation graph by taking the voltage fluctuation correlation coefficient as an edge, analyzing node features by using a pre-trained multi-head attention neural network to obtain a node risk coefficient, and calculating a total risk coefficient of a connected path to perform early warning, thereby realizing automatic monitoring and early warning of the power distribution network.

The invention provides an intelligent monitoring and regulation and control method and system based on a distributed power supply access unit, and relates to a distributed power supply intelligent regulation and control technology. The method comprises the following steps: collecting and processing power grid load and power output data to obtain a load fluctuation trend and an output characteristic deviation; establishing a prediction model by using a support vector machine, and identifying dynamic mismatching points; when the mismatching points exceed the limit, parameter variation is analyzed through a neural network, and an adjustment coefficient is calculated; generating a regulation and control instruction by adopting a particle swarm optimization algorithm, and calibrating the output characteristics of the power supply; key synchronization points are extracted for matching judgment, and stable access configuration is obtained; performing risk assessment through a feedback mechanism, and updating the regulation and control model according to the risk assessment; and finally realizing balanced distribution of the power grid pressure. The system correspondingly comprises a data acquisition module, a dynamic analysis module, an instruction generation module, a configuration and evaluation module, an optimization balancing module and the like. According to the invention, accurate monitoring and adaptive regulation and control of distributed power supply access are realized, and the stability and efficiency of power grid operation are improved.

The invention belongs to the field of engineering geology, particularly relates to an engineering geology soil layer name judgment system, method and equipment carried on a static sounding instrument, and aims to solve the problem that real-time dynamic soil layer name presumption cannot be carried out in an existing geotechnical engineering investigation technology. The system comprises a spectrum acquisition module which comprises a spontaneous light source and a multispectral sensor and is integrated between static sounding probe rods; the data processing module is used for analyzing the spectral data through a neural network and processing end resistance / side resistance / friction resistance ratio data of static sounding through a regression algorithm; and the data output module compares the two types of judgment results: outputting a single soil layer name when the two types of judgment results are consistent, and outputting double results and converting the double results into standard results for display when the two types of judgment results are inconsistent. According to the method, the soil layer name is presumed by adopting a real-time dynamic method, the detected sample does not need to be remodeled, the recognition accuracy and efficiency are improved, the model is fed back and optimized according to a new test set, the prediction precision is kept, and the detection speed is improved.

The invention relates to the field of seismology, signal processing and artificial intelligence, namely to automated methods and systems for early detection of earthquakes based on the analysis of seismic data, and can be used to recognize and classify longitudinal waves (P-waves) preceding the main seismic shocks, using deep learning and computer vision methods. The aim of the present invention is to create an automated early earthquake detection system using the classification of seismic signal spectrograms generated by the complex Morlet wave CWT using the YOLO deep neural network architecture. The technical result is an increase in the accuracy and speed of early earthquake detection by analyzing the time-frequency characteristics of seismic signals and automatically localizing P-wave signatures using a neural network model. The device includes a seismic sensor, an analog-to-digital converter, and a microprocessor implementing time-frequency analysis and neural network detection algorithms. The seismic signal is recorded in real time, digitized, segmented into time intervals, and subjected to preliminary digital processing, including noise filtering and amplitude normalization. Each time interval is converted into a time-frequency representation using a continuous wavelet transform, generating a two-dimensional distribution of signal energy over time and frequency. Based on the obtained data, a spectrogram is generated and fed to the YOLO neural network detection model, which is capable of automatically detecting and localizing longitudinal P-wave signatures. Based on the neural network analysis, a determination is made regarding the presence of a P-wave and its arrival time is determined. If a predetermined threshold is exceeded, an early warning signal is generated. The system provides for data accumulation and the possibility of subsequent retraining of the neural network model.

The invention relates to the technical field of running intelligent testing, in particular to a running intelligent testing method, system and equipment based on multi-camera cooperation. According to the invention, identity verification is carried out through deep learning face recognition, and basic information is obtained; establishing an adaptive rule set including a test distance and a speed setting range by combining real-time environmental parameters; multi-camera cooperative acquisition is utilized, and space mapping is established by calibrating common coordinate points to realize track splicing; a neural network is adopted to analyze motion characteristics such as speed trend, acceleration distribution and attitude change, and track effectiveness is evaluated; and finally, generating a report containing test data and training suggestions. According to the invention, through combination of multi-camera cooperation and an adaptive rule, omnibearing monitoring is realized, the standard can be flexibly adjusted according to the environment and individual characteristics, and support is provided for scientific training.

The application discloses a virtual anchor intelligent idle chat system based on a dynamic knowledge graph, relates to the technical field of artificial intelligence and virtual anchors, and comprises data preprocessing, cross-mode fusion, intention distribution, response generation and knowledge updating modules. Current and historical dialogue data of a user is acquired first; dialogue text features and entity relationship features of a knowledge graph are extracted and fused; the correlation degree of user intention and knowledge entities is analyzed by using a graph neural network; idle chat responses fused with knowledge are generated accordingly; and finally, knowledge graph updating strategies are generated according to response feedback to drive the dynamic evolution of the knowledge base. The system realizes the deep dynamic fusion of idle chat dialogue and structured knowledge, and can autonomously optimize knowledge according to interactive feedback, thereby improving the knowledge, accuracy and long-term adaptability of responses of the virtual anchor.

Near Infrared Spectroscopy is employed to non-invasively detect blood glucose concentrations, in a multi-sensing detection device. A multi-layered artificial neural network is used to assess these relationships of non-linear interference from human tissue, as well as differences among individuals, and accurately estimate blood glucose levels. Diffuse reflectance spectrum from the palm at six different wavelengths analyzed with a neural network, results in a correlation coefficient as high as 0.9216 when compared to a standard electrochemical glucose analysis test.

The invention relates to the technical field of digital twinning, and particularly discloses a chemical plant water resource leakage early warning system and method based on digital twinning. Comprising a digital twinborn model construction module, a water pipeline change abnormity monitoring module, a water resource leakage risk prediction module, a leakage risk prediction result correction module, a water resource leakage coordinate positioning module and a water resource leakage type identification module. According to the method, the water resource system digital twinborn model is constructed, and the wall thickness change deviation rate and the polarization resistance sudden drop rate are monitored to identify the pipeline abnormity; a fluorescent tracing technology is adopted to accurately position a leakage point; analyzing the ultrasonic voiceprint recognition leakage type by using a convolutional neural network; and performing model correction based on the actual leakage rate and the predicted value deviation to form a closed-loop optimization mechanism. According to the invention, prediction, positioning and type identification of water resource leakage are realized, the early warning accuracy is improved, and resource waste and potential safety hazards are reduced.

A video processing apparatus according to an embodiment is disclosed, which includes at least one processor, wherein the at least one processor is configured to generate a plurality of feature information for each time and frequency by analyzing a video signal including a plurality of images based on a first deep neural network (DNN), extract a first height component and a first plane component corresponding to a motion of an object in the video from the video signal based on a second DNN, extract a second plane component corresponding to a motion of a sound source in a first audio signal not having a height component by using a third DNN, generate a second height component from the first height component, the first plane component, and the second plane component, output a second audio signal including the second height component based on the feature information, and synchronize the second audio signal with the video signal and output the signal.

The invention discloses a real-time positioning method for a tumor target region and a gold label implant, and belongs to the technical field of tumor localization, and the method specifically comprises the following steps: implanting a gold label in a tumor region, constructing a gold label-tumor dynamic model through CT, and obtaining a respiratory phase feature vector; realizing space matching of the two-dimensional projection and the three-dimensional model by using an improved hybrid registration algorithm, and obtaining a tumor compound displacement vector; establishing a respiration-displacement correlation model through time sequence alignment, generating a tumor movement map containing a real-time position and a predicted trajectory, and updating the reference coordinate system; establishing a rigid safety boundary based on a gold label, analyzing soft tissue texture and a prediction trajectory in combination with a convolutional neural network to generate an elastic treatment boundary, and dynamically adjusting a dose weight of a non-label area; and finally, submillimeter-level dynamic tracking is executed through a multi-leaf collimator. The problems of tumor motion uncertainty and individual deformation difference are solved, and the target area coverage precision and the dose distribution reasonability are improved.

Optical coherence tomography (OCT) data can be analyzed with neural networks trained on OCT data and known clinical outcomes to make more accurate predictions about the development and progression of retinal diseases, central nervous system disorders, and other conditions. The methods take 2D or 3D OCT data derived from different light source configurations and analyze it with neural networks that are trained on OCT images correlated with known clinical outcomes to identify intensity distributions or patterns indicative of different retina conditions. The methods have greater predictive power than traditional OCT analysis because the invention recognizes that subclinical physical changes affect how light interacts with the tissue matter of the retina, and these intensity changes in the image can be distinguishable by a neural network that has been trained on imaging data of retinas.

Near Infrared Spectroscopy is employed to non-invasively detect blood glucose concentrations, in a multi-sensing detection device. A multi-layered artificial neural network is used to assess these relationships of non-linear interference from human tissue, as well as differences among individuals, and accurately estimate blood glucose levels. Diffuse reflectance spectrum from the palm at six different wavelengths analyzed with a neural network, results in a correlation coefficient as high as 0.9216 when compared to a standard electrochemical glucose analysis test.

The invention relates to the technical field of digital casting, in particular to an inorganic mineral casting processing parameter optimization method based on real-time data analysis, which comprises the following steps: detecting melt components by a spectrometer, inputting the melt components into a neural network for analysis, generating component characteristic analysis results, extracting pouring temperature and heat preservation time, and monitoring viscosity and filling time. The method comprises the steps of generating a pouring parameter configuration information table, monitoring the temperature and the cooling rate in real time, screening cooling parameters through particle swarm optimization, adjusting the strength of the cooling stage to judge the demolding time, comparing the parameters to adjust the demolding temperature and outputting an inorganic mineral casting machining parameter optimization scheme. The intelligent algorithm automatically analyzes pouring parameters, dynamically analyzes temperature and rheological characteristics, autonomously adjusts multi-stage temperature and cooling strength, feeds back and optimizes surface and internal quality, improves parameter consistency, reduces subjective errors, balances performance indexes, and enhances production stability and resource efficiency.

Approaches are disclosed for enhancing frame rate and visual smoothness in real-time video streams through multi-frame interpolation. A classification neural network analyzes two sequential frames, outputting confidence scores that indicate the reliability of motion data for each pixel. These scores determine whether a pixel's motion is accurately described by motion vectors or should be treated as static. The classification results are reused to generate intermediate frames by warping the original frames based on the motion characteristics. Blending weights are calculated by combining warped motion vector confidence values with static values, and a second neural network refines the alignment and blending of candidate frames. This second network predicts intermediate flows and generates new blending weights, which are used to warp and blend the candidate frames, ultimately producing a final interpolated frame that enhances visual smoothness and consistency in the video stream.

This method for converting the radioactivity of each nuclide using an artificial neural network comprises the steps of: analyzing a spectrum of a radionuclide by using an artificial neural network and outputting an output layer of the artificial neural network as the radioactivity ratio of each nuclide; and converting the radioactivity ratio of each nuclide into the radioactivity of each nuclide.

The present invention relates to a computer-implemented method for training a neural network for classifying images of objects suspended in a substance, preferably crystals of that substance, and to a computer-implemented method that uses this trained neural network for analyzing objects suspended in said substance, particularly crystals suspended in said substance, especially for real-time monitoring of the progress of a crystallization process of that substance. Figure 7 to be published

The invention provides a method, device and equipment for intelligently detecting gateway memory leak and a medium, and the method comprises the steps: deploying a lightweight AI prediction model at a main gateway, and carrying out the online learning of a memory leak mode and real-time reasoning; the slave gateway collects memory data, preprocesses the memory data, generates a memory change trend vector by adopting a dynamic sliding window, and transmits the memory change trend vector to the master gateway; the main gateway analyzes the memory change trend vector transmitted by the slave device in real time through an AI prediction model, identifies a potential leakage mode of the memory change trend vector, and performs graded early warning on an identification result; constructing a lightweight graph neural network to analyze a memory dependency relationship, evaluating a repair measure influence range, and dynamically selecting an optimal repair strategy for equipment needing to be repaired according to a historical repair effect in combination with a reinforcement learning model; and regularly carrying out data statistics and generating a whole-network visual predictive report. According to the invention, memory leak detection, classified statistics and intelligent alarm of the master / slave gateway can be realized, and the network stability is improved.