132 results about "Domain analysis" patented technology

The invention discloses a handwritten element automatic segmentation and extraction method for a complex layout, relates to the technical field of handwritten element automatic segmentation and extraction, and aims to solve the technical problem that the recognition and separation precision of handwritten contents in a mixed image-text layout is insufficient. S201, a dynamic threshold segmentation algorithm is carried out; s202, context sensing connected domain analysis is carried out; s203, judging whether the elements are handwritten elements or not; s204, if the judgment result is yes, the handwriting region candidate is reserved; and S205, if not, filtering and eliminating. According to the method, the dynamic threshold segmentation algorithm and the context sensing connected domain analysis technology are cooperated, the segmentation threshold can be adaptively adjusted according to the pixel mean value and the standard deviation of the image local window through dynamic threshold segmentation, and the context sensing connected domain analysis is combined with the context information of the document to perform semantic analysis on the connected domain. The problem that the recognition and separation precision of the handwritten content in the mixed image-text layout is insufficient is solved.

The invention relates to the technical field of image processing, and discloses a parallelization connected domain analysis method based on neighborhood labeling, computer equipment and a storage medium, through high parallelization design, the steps of father node, root node and area junction point searching, mapping establishing, re-numbering and the like are all designed into an independent parallel computing mode, and the parallel computing mode is designed into a parallel computing mode. According to the method, the acceleration capability of parallel computing hardware such as a GPU can be fully utilized, the processing speed is remarkably improved, the method is easy to expand, large-scale high-resolution images can be efficiently processed, the area of each connected domain is counted by adopting a Hash optimization method of grouping local reduction and global merging, atomic conflicts are effectively reduced, the statistical efficiency is improved, and the calculation efficiency is improved. The robustness of the algorithm in a parallel computing environment is further enhanced, so that the efficiency of the connected domain analysis algorithm is greatly improved, rapid processing of high-resolution images is realized, and powerful support is provided for real-time performance and large-scale application in the field of computer vision and image processing.

The invention discloses a semi-supervised medical image segmentation method based on momentum prototype alignment and adaptive uncertainty estimation, and aims to solve the problems of unstable feature alignment and false label noise accumulation caused by a batch effect in an existing semi-supervised learning method in a scene of scarcity of medical image annotation data. The method comprises the following steps: firstly, constructing a dual-network model based on a mean teacher architecture; secondly, designing a mixed strong and weak disturbance strategy, respectively applying weak data disturbance and strong data disturbance to the unlabeled image, respectively inputting the weak data disturbance and the strong data disturbance into a teacher and student segmentation network, and forcing the model to learn local and global robust features; secondly, a self-adaptive uncertainty estimation mechanism is introduced to dynamically screen false labels, and isolated noise is eliminated in combination with maximum connected domain analysis; meanwhile, the prototype of the current batch is calculated through uncertainty weighting, and a global category prototype library is updated and maintained through momentum. According to the method, the segmentation precision and the boundary integrity are remarkably improved under a small amount of annotated data.

According to the preparation method of the low-signal-loss copper-clad plate for the AI server, a microwave resonant cavity is integrated between heating plates of a laminating machine, on-line non-intrusive detection on dielectric response of a base material is achieved, the curing degree and stress distribution of resin of each layer are inversed in real time in combination with frequency domain analysis and material dielectric modeling, and the low-signal-loss copper-clad plate for the AI server is obtained. A pre-trained physical information neural network is used to map the three-dimensional pressure-deformation relation; according to the method, model prediction control is adopted to dynamically generate a pressure instruction, closed-loop feedback adjustment and optimization are carried out, technological parameter self-tuning and multi-objective optimization are matched, a knowledge base is established, and strategy autonomous recommendation is realized, so that the flexibility and precision of pressure distribution regulation and control and the consistency of finished products are improved, and continuous iterative optimization of technological experience is supported.

The invention discloses a line laser anti-welding noise center point extraction and weld joint point identification method, which is applied to the field of welding quality detection and machine vision measurement. The problems of low extraction precision, poor robustness, difficulty in welding seam point positioning and the like are solved for welding spatter, light reflection and background noise influences. According to the technical scheme, the method comprises the following steps: carrying out Otsu adaptive threshold segmentation, connected domain analysis, morphological opening operation and Gaussian filtering denoising; positioning an interference abnormal line and extracting a preliminary central point by a normal centroid method, and further smoothly extracting a detailed central point in an abnormal region by fitting straight lines of 10 front and back central points to assist judgment; and by calculating a second-order difference value, fitting front and back 50 points at the maximum difference value to identify welding seam points, and optimizing a track. The method can be applied to welding seam positioning and contour detection, the accuracy and stability in a complex noise environment are improved, the welding seam point recognition error is smaller than or equal to 0.2 mm, and the actual requirement is met.

The application relates to a complex texture textile defect recognition method and system based on feature decoupling, and relates to the field of computer technology, which comprises the following steps: obtaining a multi-scale feature set of an original textile image to obtain a complex texture fusion feature; projecting the complex texture fusion feature to two mutually orthogonal hidden subspaces to obtain a reference texture component and an abnormal disturbance component; performing manifold alignment processing on the reference texture component based on a preset texture library to obtain a texture reconstruction feature; performing feature fusion on the texture reconstruction feature and the abnormal disturbance component to obtain a texture residual feature; performing adaptive threshold segmentation on the texture residual feature to obtain a defect feature map of the original textile; and performing connected domain analysis processing on the defect feature map to output a defect detection result. The application has the effects of improving the robustness of textile detection, meeting the needs of real-time detection in a pipeline, and reducing the computing resources and labor costs.

The invention discloses a broadband random harmonic compression analysis method based on LSTM and multi-feature fusion, and belongs to the technical field of electric digital data processing. Comprising the following steps: collecting a broadband random harmonic signal for time domain analysis, and extracting a period; judging whether each period signal can be compressed or retained or not by using LSTM (Long Short Term Memory); carrying out down-sampling and coding compression on the compressible periodic section signal; decoding and reconstructing the signal through an LSTM decoder; the decoded and reconstructed signals are preprocessed, multi-dimensional features of the signals are extracted, and evaluation indexes are normalized; calculating the compression ratio of the dynamic fusion index and the signal; and dynamically displaying the evaluation index and the dynamic fusion index. According to the method, self-adaptive decision of compression or not is realized, the compression ratio is improved, and meanwhile, the spectrum fidelity and phase consistency are also kept; and parts which do not need to be compressed are reserved as original samples, so that real-time performance and error control are both considered.

The invention discloses an AI-based Jetson equipment system state dynamic atlas diagnosis system and method, and relates to the technical field of Jetson equipment state analysis, and the analysis system comprises a system call collection module which carries out the structural processing of a system call track; the calling sequence extraction module is used for carrying out continuous calling identification on the execution process recording path and extracting a calling sequence fragment with a specific mode; the state atlas construction module is used for establishing a system state change atlas for the screened calling sequence fragments in a graph structure splicing manner; the AI intelligent diagnosis module is used for analyzing the system state change atlas by using a pre-trained artificial intelligence model; according to the AI-based Jetson equipment system state dynamic graph diagnosis system and method, the system state change graph is established through graph structure splicing, so that a dynamic behavior has a static structure expression capability, the running state of equipment can be analyzed more accurately, and the fault detection precision and reliability are improved.

The invention provides a method and system for calculating the topological connectivity of a two-dimensional discrete fracture network containing n-type nodes, and the method comprises the steps: carrying out the binarization processing of an original fracture image of a rock surface, and generating a fracture binary image; extracting each crack region from the crack binary image based on connected domain analysis; extracting an outer contour pixel point set and a central axis point set of any crack region; an intersection point cluster is extracted from the central axis point set, and an effective extension point, located on the central axis, of each intersection point is obtained; determining the node type of each intersection point according to the number of the effective extension points of each intersection point; and calculating connectivity parameters of any crack region according to the node type of each intersection point of any crack region. Through the method, the node types in the two-dimensional discrete fracture network can be automatically identified, accurate quantization of connectivity is realized in combination with graph theory parameters, and the analysis precision and efficiency of the complex fracture network are remarkably improved.

The invention provides a method for identifying an abnormal mode of a thermal energy storage system, and belongs to the technical field of thermal energy storage systems.The method comprises the steps that temperature distribution data, pressure fluctuation data and heat flux density data in the operation process of the thermal energy storage system are collected, and multi-scale wavelet decomposition is conducted on the temperature data to extract high-frequency components and low-frequency components; performing time-domain and frequency-domain analysis on the pressure data to extract peak frequency and amplitude attenuation coefficient, performing gradient calculation and topological analysis on the heat flow data to extract singular point distribution characteristics, and fusing and normalizing the three types of characteristics to establish a system operation state characteristic library; and applying the trained model to a system operation state feature library to extract deep abnormal features and output an abnormal mode classification result and a confidence score, and when the confidence score exceeds a preset threshold, determining that the state is an abnormal state, thereby solving the technical problem of insufficient recognition accuracy of the abnormal mode of the thermal energy storage system.

The invention relates to the technical field of signal feature fusion, and discloses a vibration signal multi-domain feature fusion extraction method and system, and the method comprises the steps: obtaining vibration signal data of to-be-detected equipment, uploading the vibration signal data to a preset platform, and carrying out the wavelet transformation processing, and obtaining the vibration component data, containing the vibration amplitude of each time point, of the to-be-detected equipment under a plurality of scales; respectively carrying out time domain analysis, frequency domain analysis and scale weight extraction based on the vibration component data containing the vibration amplitude of each time point under a plurality of scales of the to-be-detected equipment, and correspondingly obtaining a time domain excitation characteristic value, a frequency domain working condition response characteristic value and a scale weight characteristic value of each scale; and fusing the time domain excitation characteristic value, the frequency domain working condition response characteristic value and the scale weight characteristic value of each scale, analyzing to obtain a working condition collaborative mapping characteristic value of the to-be-detected equipment, and completing fusion extraction of vibration signal multi-domain characteristics. The technical problem of inaccurate working condition recognition caused by difficulty in realizing multi-domain feature adaptive fusion of vibration signals in the prior art is solved.

This invention relates to a method and system for online dynamic stability assessment of a substation control system, belonging to the field of substations. The method includes: collecting raw data from the substation control system and constructing a joint state vector set; performing local neighborhood analysis based on the joint state vector set to obtain the local covariance matrix and metric matrix corresponding to each state point; estimating the tangent space curvature based on the local covariance matrix to obtain the curvature value corresponding to each state point; simulating geodesic deviation based on the metric matrix to obtain the local geodesic deviation index; and performing dynamic stability assessment based on the curvature value and the local geodesic deviation index, with online rolling updates and early warnings. This invention is a dynamic stability assessment method that integrates system command and response information, possesses online assessment capabilities, and can provide early warnings of instability trends.

This invention discloses a method for identifying geological hazard risk zones based on a local search strategy, belonging to the field of geological hazard monitoring and early warning technology. The method includes the following steps: locating seed points within a target area; finding similar developmental condition r grid cells within the local search window Ω of the seed point s, forming a similar developmental condition region; then, using the raster area where the seed point s is located as the initial region, iteratively performing morphological dilation, expanding the current region by a certain number of pixels during each iteration and intersecting it with the similar developmental condition region, updating the current region with the intersecting region; after iteration terminates, obtaining the similar neighborhood of the seed point; converting the similar neighborhood region raster of the seed point into vector polygons, and optimizing the boundary smoothing to obtain the risk zone identification range. This invention focuses on core area calculation, avoiding the low efficiency problem of full-domain analysis and identification, and is suitable for rapid screening and dynamic updating of large-scale risk zones.

The application discloses a remote sensing image water body extraction method based on physical statistical distribution guidance and large model space constraint, comprising the following steps: acquiring a remote sensing image and performing pretreatment, calculating a physical statistical feature map and performing probability density distribution analysis, automatically determining a scene category to which the remote sensing image belongs, adaptively matching a hierarchical threshold calculation strategy, and acquiring a seed threshold and a growth threshold; performing region growing processing based on the seed threshold and the growth threshold to generate a physical prior coarse mask; performing connected domain analysis and adaptive outer expansion processing on the physical prior coarse mask to form a prompt box; inputting the pretreated remote sensing image and the prompt box into a visual basic model to output a fine segmentation prediction map; constructing a physical constraint search area based on the physical prior coarse mask, performing cascade fusion on the fine segmentation prediction map and the physical constraint search area, and obtaining a high-precision water body binary mask. Based on this, the application realizes automatic water body extraction without training samples, with strong noise resistance and fine edges.

The invention relates to the field of information security, and discloses a random number entropy source quality evaluation system, which comprises a data acquisition module, a time domain analysis module, a frequency domain analysis module, a feature fusion and prediction module and an entropy source degradation detection module. According to the random number entropy source quality detection method, system and device, millisecond-level real-time monitoring is realized through cooperative work of a data acquisition module and a feature fusion and prediction module, and the data acquisition module acquires a random number bit stream in a continuous stream mode and divides the random number bit stream into fixed-length sequence blocks; and the feature fusion and prediction module splices features extracted by the time domain analysis module and the frequency domain analysis module and outputs a prediction probability through a dimension reduction network, and the design utilizes a mode of combining hardware acceleration and deep learning, so that the detection delay of an entropy source degradation event is greatly shortened, the real-time response capability of the system is ensured, and the detection efficiency is improved. And a security vulnerability window period is effectively avoided.

The invention relates to the technical field of data processing, in particular to a power grid multi-dimensional information base construction method and system based on data mapping. According to the method, data source change events are monitored in real time, and incremental data records are extracted; constructing a data association graph containing a dynamic association relationship, calculating the influence degree of incremental data records on other data items based on graph path propagation or space-time neighborhood analysis, and obtaining a quantized influence degree factor; updating tasks of other data items are generated according to the influence degree factors, priority ranking is carried out, and a multi-version concurrency control mechanism is adopted to carry out atomic updating on the data items according to the priority. The real-time accurate updating of the multi-source heterogeneous data of the power grid is realized by capturing the increment change in real time and quantifying the dynamic association influence, the problems of data lag and weak association caused by a traditional batch processing mode are solved, and the timeliness of power grid state sensing and scheduling is effectively improved.

The invention provides a multi-domain electrocardio intelligent analysis method of a mixed Fourier and wavelet convolutional neural network, belongs to the technical field of artificial intelligence and medical health crossing, and solves the technical problems that global spectrum features and local transient features are difficult to consider and task generalization is poor in traditional single-domain ECG analysis. According to the technical scheme, the method comprises the following steps: S1, preprocessing an ECG signal, filtering, segmenting, normalizing and enhancing data; s2, constructing a three-branch model; s3, fusing the attention mechanism with multi-domain features; s4, setting a task classification head; s5, using Adam optimization and an early stop strategy to prevent overfitting; and S6, inputting data, and outputting arrhythmia classification, biological recognition and sleep apnea detection results. According to the method, generalization and accuracy are improved through multi-domain fusion, multiple tasks are supported, and clinical and biological recognition scenes are adapted.

The invention relates to the technical field of state monitoring and fault diagnosis of ship power devices, and provides a noise data processing method based on cavitation flow characteristics of a propeller. According to the method, observable dynamic frequencies such as leaf frequency and cavitation bubble falling main frequency and statistical non-stationary quantities such as a frequency spectrum broadening factor, self-correlation time, broadband energy and kurtosis are unified in the same analysis framework, a mapping model of a cavitation state and signal processing parameters is constructed, analysis parameters such as FFT, STFT and CWT are adaptively configured, and the analysis parameters of the cavitation state and the signal processing parameters are analyzed. And multi-scale feature extraction of non-stationary pressure pulsation signals under different cavitation types and cavitation intensities is realized. According to the method, the cavitation state and type are identified and evaluated on the basis of the extracted multi-domain features, the threshold and mapping parameters are reversely updated according to the identification result, a closed loop of state judgment, parameter mapping, multi-domain analysis, feature output, cavitation identification and parameter updating is formed, and the precision and engineering applicability of propeller cavitation monitoring and diagnosis are improved.

The invention relates to an industrial sensor data risk identification method based on time-frequency cross-domain analysis. The method comprises the following steps: carrying out dimension independent decomposition on multi-dimensional time sequence data collected by an industrial sensor to obtain a plurality of single-dimensional time sequences; performing instance normalization processing on the data, respectively inputting the normalized data into a time-frequency double-branch architecture, training a double-branch architecture encoder based on a cross-domain loss function, and respectively obtaining time-domain and frequency-domain reconstruction data; calculating a reconstruction error of the time domain and the frequency domain, and generating an abnormal score of each time point through weighted aggregation; and determining a dynamic threshold value by adopting the quantile of the abnormal score of the verification set, and realizing abnormality judgment by comparing the abnormal score with the threshold value. According to the method, time-frequency characteristic internal association is deeply learned under the condition of no mark, risk identification of a multi-dimensional time sequence is realized, the detection accuracy is high, and the method is suitable for health monitoring and risk early warning of industrial equipment.

The invention discloses a power distribution network voltage sag sag domain analysis method considering distributed power supply access, and relates to the technical field of power system electric energy quality control. According to the method, a power distribution network asymmetric short circuit three-sequence equivalent model containing a distributed power supply and sensitive equipment is constructed, a looped network is split based on a voltage lowest node when a looped network structure exists, and a voltage sag sag domain general equation and a characteristic equation under any line fault condition are established; and further through sag subsystem division and Kron order reduction processing, an order reduction node impedance matrix about fault position parameters is formed, and the fault position parameters enabling the residual voltage of the sensitive nodes to reach a preset threshold value are determined by adopting an optimization iteration mode, so that the voltage sag sag domain boundary is accurately obtained. The method can give consideration to the calculation precision and efficiency, and is suitable for the voltage sag analysis of the power distribution network containing the distributed power supply and the complex topological structure.

This application discloses a method, apparatus, and related products for reverse analysis of CAN signals. The method calculates the entropy value of each bit in the original CAN message data, the covariance matrix between every two bits, performs connected component analysis, and calculates the temporal variance and range of each bit, resulting in a multi-dimensional feature matrix. This multi-dimensional approach more accurately and completely identifies signal boundaries, reducing the inaccurate signal segmentation caused by segmenting signals using a single feature. The feature matrix is ​​input into a prediction model to obtain candidate parsing results represented in a multi-dimensional form of six-tuples, solving the problem of missing parameter dimensions. LLM (Logical Learning Model) is introduced for logical verification, effectively filtering out erroneous parsings that violate common sense, and obtaining a logical score for each candidate parsing result, further improving accuracy. Furthermore, the method requires no manual intervention, improving reverse analysis efficiency, and is transparent throughout the process, enhancing the credibility and interpretability of the parsing results.

The invention relates to a double-layer identification and filtering method based on large and small model collaboration, and belongs to the technical field of artificial intelligence application. The method comprises the following steps: identifying original data of an application scene by using a small model; carrying out multi-modal integration on the identification result of the small model by the large model; the large model performs confidence coefficient and data integrity verification on the processing result of the large model, triggers the small model to perform secondary identification, integrates the two identification results, re-verifies the confidence coefficient and the data integrity, and outputs a final result; automatically adjusting calling priorities of the small models by utilizing a configurable application packaging platform, and elastically allocating computing resources of the large models; the large model regularly performs cross-domain analysis on the recognition result of the small model to extract cross-scene universal features; the small model integrates the general features and the specific features to form the complete recognition capability of the small model. According to the method, the processing precision of complex scenes can be improved, the efficiency and expansibility are balanced, and the application packaging process is simplified.

The invention provides an airborne software structure incremental coverage rate analysis system and analysis method. The airborne software structure incremental coverage rate analysis system comprises a self-adaptive baseline management module, a change influence domain analysis module, an incremental coverage rate calculation module and an automatic report generation module. The method comprises the following steps: establishing a tracing relation among requirements, codes and use cases, capturing code version change to construct a change influence domain, carrying out change difference analysis in combination with historical version coverage rate data, identifying and running use cases which need to be additionally and independently executed in a current version, and obtaining a newest code structure coverage rate. And a coverage rate analysis report can be automatically updated and generated. By applying the analysis system and the analysis method, the technical bottlenecks in the aspects of large coverage analysis difficulty of an airborne software verification structure, redundant analysis work and the like can be solved.

The invention discloses a visual guidance and geometric deambiguity-based high-precision detection method for a weld defect of a rocket wallboard. The method comprises the following steps: acquiring a 2D image and a 3D point cloud of a weld of the rocket wallboard; a 2D global reconnaissance network is constructed, and a composite loss function is adopted for training; deploying a trained network to perform reasoning on the 2D image, and generating a guidance signal through connected domain analysis and noise filtering; based on the guidance signals and 3D point cloud quantitative analysis, macroscopic geometric indexes of the whole weld joint and microscopic geometric feature descriptors of the suspicious defect areas are analyzed; performing macroscopic-microcosmic hybrid decision on the defects, and outputting a weld defect decision list; and packaging the judgment list to obtain a structured digital quality archive report, and outputting the structured digital quality archive report to a downstream quality tracing system in a JSON format. While optical artifacts and geometric noise interference are processed, high efficiency and high robustness are kept, key defects can be captured through a structured report, and the welding seam quality can be accurately monitored.

The invention discloses a multi-domain parameter adaptive adjustment method for a digital oscilloscope matched with signal frequency characteristics, and the method comprises the steps: firstly sampling an unknown input signal through a high-speed data collection system, obtaining a digital parallel sampling signal, and transmitting the digital parallel sampling signal to a multi-information-domain data processing module in an FPGA (Field Programmable Gate Array); in the multi-information-domain data processing module, time-frequency graph data acquisition of unknown signals is completed based on accurate control, fast Fourier transform and data delay feedback of time-domain data required by each frame of frequency spectrum, and the time-frequency graph data is transmitted to an industrial personal computer through a high-speed interface after being acquired. The method comprises the following steps: performing data analysis, obtaining a maximum value of spectrum energy and calculating an effective frequency index value to obtain a maximum effective frequency index value and a minimum effective frequency index value, then obtaining a multi-domain analysis parameter center frequency and an analysis bandwidth through calculation, and setting, feeding back and adjusting a multi-domain analysis processing module on a user interface of a digital oscilloscope.

The application belongs to the technical field of weather forecast, and discloses a storm intensity spatial distribution prediction and falling area demarcation method based on a graph neural network, which comprises the following steps: dividing a target area into multiple spatial units, taking the geometric center of each spatial unit as a node, collecting multi-source meteorological data and mapping the multi-source meteorological data to each node to construct a spatial node set; calculating the edge weight value between any two nodes to construct an adaptive adjacency matrix; generating a node feature vector of each node to form a node feature matrix; inputting the adaptive adjacency matrix and the node feature matrix into a pre-constructed graph convolutional neural network model to output a storm intensity prediction matrix and identify storm nodes; performing connected domain analysis on the storm nodes to extract the peripheral boundary contour of each connected domain and generate a storm falling area set; and the application can realize intelligent processing of the whole process from storm prediction to emergency response, improve the accuracy of storm early warning and the fine degree of falling area demarcation.

The application discloses a three-dimensional city space intelligent planning management and control method and system and a storage medium, and the method comprises the following steps: based on a pre-constructed space rule knowledge base, RAG query enhancement and rule digital translation are performed on to-be-queried geographic rule information, and a standardized JSON Schema is generated; a machine executable task chain is generated through a finite state machine model according to the standardized JSON Schema; based on the machine executable task chain, a three-dimensional space scene is constructed, space region analysis and space conflict identification are performed through a multi-modal Agent intelligent agent; and a space planning analysis report is generated according to the constructed three-dimensional space scene graph, the space region analysis result and the space conflict identification result, and geographic information is planned and controlled. The application realizes city intelligent planning management and control by intelligently translating unstructured planning management and control rules into machine executable instructions and autonomously driving a three-dimensional space analysis tool by using an intelligent agent.

The application relates to the technical field of value accounting, and discloses an ecosystem gross domestic product accounting method based on GIS spatial analysis, which comprises the following steps: accounting for regional basic data and satellite image data, constructing a two-way dynamic semantic mapping library corresponding to land change survey block classification-ecosystem type; obtaining Internet of Things monitoring data and statistical report data, performing spatial interpolation-boundary correction and sliding window smoothing space-time fusion to obtain a standardized data set; constructing a double-level accounting index system of urban areas and typical areas, and synchronously calculating physical quantity and value quantity; based on GIS, a model correction factor is called to quantize the index to a grid unit, neighborhood analysis is performed to obtain a result, and after the parameters are calibrated, a multi-year ecological value trend graph is generated in combination with GIS; customized results are output according to scenes such as ecological compensation, and after three-level quality control verification, the final results are output. The application can meet the needs of accounting accuracy and practicability of ecological compensation, EOD projects and the like.

The invention discloses a dynamic retrieval letter text analysis method, device and equipment and a storage medium, and relates to the technical field of text retrieval, and the analysis method comprises the following steps: obtaining a sentence vector of a letter text and a document text sequence thereof; obtaining a problem vector, carrying out cluster analysis, and selecting out a cluster representative problem; analyzing the representative problem of the cluster and the document text sequence, and constructing a context sensing door to realize preliminary retrieval; based on the sentence vector and the problem vector after preliminary retrieval, deep retrieval is realized; and constructing a thinking chain demonstration, and performing analysis to obtain an answer. According to the method, the letter text is preliminarily retrieved by introducing the context sensing gate, so that the correlation of the letter text is improved; the context information in the text vector is dynamically adjusted, valuable information can be accurately selected from the source, the risk that a large language model generates wrong answers due to misleading information is greatly reduced, and the accuracy, reliability and efficiency of final answers are remarkably improved.

The application relates to the technical field of structural health monitoring and engineering safety, in particular to a dam automatic multi-element data fusion monitoring system and method based on the Internet of Things; the method comprises the following steps: an initial state reference without historical sample dependence is established by deploying checked heterogeneous sensing units on site and constructing a dynamic baseline mechanism; then, cross-scale mapping matrices and feature-dependent correction coefficients are used to perform multi-domain analysis and scale self-adaptive adjustment on time series data, so that the difference capturing capability is enhanced; a dynamic credibility adjustment and multi-source weighted fusion method is adopted at the edge node, preliminary event discrimination is performed in combination with physical connectivity and neighborhood cooperation; finally, abnormal diffusion trend analysis and dynamic early warning level determination are realized based on a space-time evolution matrix and a probability prediction model, and a response strategy and system parameter optimization are automatically generated through a closed-loop feedback mechanism. The application significantly improves the accuracy, self-adaptive capability and risk control level of dam monitoring.