Search

CN-122020054-A - Multi-model large-model emergency monitoring method

CN122020054ACN 122020054 ACN122020054 ACN 122020054ACN-122020054-A

Abstract

The application discloses a multi-model large-model emergency monitoring method, and relates to the technical field of emergency management. The emergency scene multi-source monitoring method comprises the steps of collecting emergency scene multi-source monitoring data, establishing a tracing map of 'abnormal characteristics-data sources-associated dimensions', generating a standardized emergency monitoring data set with the tracing map, generating an emergency abnormal tracing monitoring tensor based on the standardized data set, inquiring an emergency treatment knowledge base to match a basic emergency treatment strategy, generating and issuing an emergency treatment dynamic adaptation instruction set, generating a model plan collaborative optimization parameter tensor by utilizing feedback data after emergency treatment, and adjusting the accuracy of abnormal tracing and emergency response. By means of multi-source data fusion, multi-model cascade analysis, dynamic plan adaptation and closed-loop optimization mechanisms, the problems of data island, difficult tracing, stiff response, optimization deficiency and the like of a traditional emergency monitoring system are solved, and the comprehensiveness, tracing accuracy and response efficiency of emergency monitoring are improved.

Inventors

  • WANG JIAYING
  • CHEN HONGYU
  • LI HUAWEI
  • Zhang Ninglu
  • FAN ZHICHAO

Assignees

  • 北京甲板智慧科技有限公司

Dates

Publication Date
20260512
Application Date
20260203

Claims (10)

  1. 1. A multi-model large-model emergency monitoring method, comprising: step 1, collecting multi-source monitoring data of an emergency scene to establish a traceability map of 'abnormal characteristics-data sources-associated dimensions', and marking original data and associated data of each abnormal characteristic to generate a standardized emergency monitoring data set with the traceability map; step 2, generating an emergency anomaly traceability monitoring tensor based on a standardized emergency monitoring data set with a traceability map; Step 3, searching an emergency treatment knowledge base based on the emergency anomaly traceability monitoring tensor, and matching a basic emergency treatment strategy to generate and issue an emergency treatment dynamic adaptation instruction set; And 4, generating a model plan collaborative optimization parameter tensor based on the feedback data after emergency treatment generated by responding to the emergency treatment dynamic adaptation instruction set, and adjusting the accuracy of anomaly tracing and emergency response based on the model plan collaborative optimization parameter tensor.
  2. 2. The multi-model large-model emergency monitoring method according to claim 1, wherein step 1 comprises: step 11, carrying out format analysis and dimension calibration on multi-source monitoring data of an emergency scene to generate an initial multi-source monitoring data tensor; step 12, inputting an initial multi-source monitoring data tensor into a multi-mode data fusion-anomaly tracing preprocessing operation framework to execute noise filtering and data alignment so as to generate a denoising alignment data tensor; And 13, carrying out abnormal feature extraction and traceability dimension association on the denoising alignment data tensor, establishing a traceability map of 'abnormal feature-data source-associated dimension', and fusing the denoising alignment data tensor and the traceability map to generate a standardized emergency monitoring data set with the traceability map, wherein the standardized emergency monitoring data set with the traceability map carries out tensor binding on the abnormal feature and the associated dimension of the traceability map.
  3. 3. The multi-model large-model emergency monitoring method according to claim 2, wherein step 11 comprises: step 111, performing frame analysis and visual feature calibration on video monitoring data in the multi-source monitoring data, performing numerical dimension calibration on sensing data, performing environment element dimension calibration on environment data, and performing space coordinate dimension calibration on positioning data to generate single-source monitoring data features; And 112, integrating various single-source monitoring data features and performing tensor dimension mapping to generate an initial multi-source monitoring data tensor, wherein the initial multi-source monitoring data tensor integrates all single-source data according to the monitoring dimension.
  4. 4. The multi-model large-model emergency monitoring method according to claim 1, wherein step 2 comprises: Step 21, constructing a multi-model cascading-abnormality traceability analysis framework, dividing an emergency special large model cluster into an image recognition operation unit, an abnormality detection operation unit and a risk assessment operation unit, and setting cascading operation logic; Step 22, inputting a standardized emergency monitoring data set with a traceability map into an image recognition operation unit to analyze visual anomalies and label position information, and inputting an anomaly detection operation unit to recognize numerical anomalies and label change trends so as to generate an anomaly basis monitoring tensor; And step 23, inputting the abnormal basic monitoring tensor, the tracing map and the scene risk weight into a risk assessment operation unit, calculating the risk level and the abnormal root, and fusing the abnormal basic monitoring tensor, the risk level and the abnormal root to generate an emergency abnormal tracing monitoring tensor, wherein the emergency abnormal tracing monitoring tensor binds the abnormal root and the associated dimension of the tracing map.
  5. 5. The multi-model large model emergency monitoring method according to claim 4, wherein step 21 comprises: Step 211, configuring a visual anomaly identification model of an image identification operation unit, configuring a numerical anomaly trend identification model of an anomaly detection operation unit, and configuring a risk level calculation model and an anomaly source positioning model of a risk assessment operation unit; Step 212, setting three-unit cascade operation logic, namely, the image recognition operation unit and the anomaly detection operation unit output results in parallel, and synchronously inputting the results to the risk assessment operation unit for fusion analysis so as to generate a multi-model cascade-anomaly traceability analysis architecture, wherein the architecture carries out protocol binding on the operation models of the three units according to the cascade logic.
  6. 6. The multi-model large-model emergency monitoring method according to claim 1, wherein step 3 comprises: Step 31, inputting an emergency response dynamic matching-optimizing operation unit into an emergency anomaly tracing monitoring tensor, and analyzing anomaly types and anomaly sources to generate an emergency response matching feature vector; Step 32, querying an emergency treatment knowledge base based on the emergency response matching feature vector, and matching a corresponding basic emergency treatment strategy to generate an initial emergency treatment tensor; step 33, adjusting the disposal flow of the initial emergency disposal tensor based on the abnormal source, generating and issuing an emergency disposal dynamic adaptation instruction set containing an early warning mode, a disposal step, a responsibility division and a source disposal strategy, and binding the source disposal strategy and the abnormal source of the emergency abnormality tracing monitoring tensor by the emergency disposal dynamic adaptation instruction set.
  7. 7. The multi-model large model emergency monitoring method according to claim 6, wherein step 31 comprises: Step 311, extracting features of the emergency anomaly tracing monitoring tensor, and extracting anomaly type features, anomaly location features, risk level features and anomaly root features to generate initial response features; step 312, performing dimension normalization and weight assignment on the initial response feature to generate an emergency response matching feature vector, wherein the dimension of the emergency response matching feature vector is consistent with the dimension of the strategy feature of the emergency treatment knowledge base.
  8. 8. The multi-model large-model emergency monitoring method according to claim 1, wherein step 4 comprises: Step 41, collecting feedback data after the dynamic adaptation instruction set of response emergency treatment is executed, and performing format normalization processing to generate a standardized emergency treatment feedback tensor; Step 42, inputting a standardized emergency treatment feedback tensor into a model-plan double-optimization adaptation mechanism, and adjusting the operation weight and the abnormal feature recognition threshold of the multi-model cascade-abnormal traceability analysis architecture to generate a model optimization parameter subset; And 43, updating an emergency treatment knowledge base based on a standardized emergency treatment feedback tensor, adjusting the association accuracy of a tracing map, fusing a model optimization parameter subset, updated knowledge base data and map adjustment parameters to generate a model plan collaborative optimization parameter tensor, and adjusting the accuracy of abnormal tracing and emergency response based on the parameter tensor.
  9. 9. The multi-model large model emergency monitoring method according to claim 8, wherein step 41 comprises: 411, performing acquisition cycle calibration and numerical standardization on the feedback data, labeling a release state label on the abnormal release type data, and labeling an adaptation score on the predetermined plan adaptation type data to generate single type feedback characteristics; Step 412, integrating various single-class feedback features and performing tensor dimension mapping to generate a standardized emergency treatment feedback tensor, wherein the tensor dimension is associated with the feature dimension of the standardized emergency monitoring dataset with the traceability map.
  10. 10. The multi-model large model emergency monitoring method according to claim 8, wherein step 43 comprises: step 431, updating a root cause treatment strategy library in an emergency treatment knowledge base based on the root cause elimination effect data of the standardized emergency treatment feedback tensor, and supplementing the mapping rule of the efficient treatment flow; Step 432, inputting a standardized emergency treatment feedback tensor into an adjustment module of a traceability map, optimizing the association weight of the abnormal characteristics, the data sources and the association dimension, and adjusting the map association accuracy; and 433, fusing the model optimization parameter subset, the updated knowledge base data and the map adjustment parameters to generate a model plan collaborative optimization parameter tensor, wherein the parameter tensor integrates various optimization parameters according to the application module.

Description

Multi-model large-model emergency monitoring method Technical Field The application relates to the technical field of emergency management, in particular to a multi-model large-model emergency monitoring method. Background Along with the acceleration of the urban process and the complexity of an industrial system, the occurrence risk of various emergencies is increased in the scenes of chemical industry parks, urban rail transit, large public places and the like, and serious challenges are formed for public safety and production safety, so that high-efficiency and comprehensive emergency monitoring technical support is needed to realize timely response and scientific treatment of the emergencies. According to the existing emergency monitoring scheme, part types of monitoring data such as video monitoring and sensors are collected, a single model is adopted to analyze and identify abnormality of the data, and then response actions are triggered according to a preset fixed emergency plan. The scheme is that monitoring data of limited types are collected through specific equipment, simple processing is carried out, a single model is input to carry out abnormality judgment, and finally a preset plan is matched and a response flow is executed. However, the prior art scheme has the obvious defects that the deep binding of the abnormality, the data source and the related factors is difficult to realize, the abnormality source cannot be traced accurately, the emergency response is lack of pertinence, the treatment efficiency is low, and the refined requirement of emergency monitoring in a complex scene is difficult to meet. Disclosure of Invention In order to solve the technical problems, the application provides a multi-model large-model emergency monitoring method for at least alleviating the technical problems. The technical scheme provided by the embodiment of the application is as follows: A multi-model large-model emergency monitoring method comprises the following steps of 1, collecting multi-source monitoring data of an emergency scene to establish a tracing map of abnormal characteristics, data sources and associated dimensions, marking original data and associated data of each abnormal characteristic to generate a standardized emergency monitoring data set with the tracing map, 2, generating an emergency abnormal tracing monitoring tensor based on the standardized emergency monitoring data set with the tracing map, 3, querying an emergency treatment knowledge base based on the emergency abnormal tracing monitoring tensor, matching a basic emergency treatment strategy to generate an emergency treatment dynamic adaptation instruction set and issuing the emergency treatment dynamic adaptation instruction set, 4, generating a model plan collaborative optimization parameter tensor based on the emergency treated feedback data generated by responding to the emergency treatment dynamic adaptation instruction set, and adjusting the accuracy of an abnormal tracing source and an emergency response based on the model plan collaborative optimization parameter tensor. The technical scheme provided by the application has the following technical advantages: Firstly, in the data acquisition and preprocessing stage, multi-source monitoring data of an emergency scene are acquired, a traceability map of 'abnormal characteristics-data sources-associated dimensions' is established, and a standardized emergency monitoring data set with the traceability map is generated. The traditional scheme only collects partial type data and does not establish effective association, the scheme integrates multi-source heterogeneous data, deep binding of abnormal features, original data and association factors is achieved through a tracing map, a structural basis is provided for subsequent tracing analysis, the problem that abnormality and data sources are disjointed in the traditional scheme is solved, and abnormal tracing is enabled to be according. And secondly, an emergency anomaly traceability monitoring tensor is generated based on a standardized emergency monitoring dataset with a traceability map, compared with a traditional single model, the emergency anomaly traceability monitoring tensor can only identify surface layer anomalies, the scheme can identify basic information such as anomaly types, positions and the like by means of a multi-model cascade architecture and synthesize multi-dimensional data and traceability map information, and can also mine anomaly sources and risk grades, so that an emergency monitoring result is more comprehensive and more explanatory, key support is provided for accurate response, and a technical short board that the traditional scheme cannot trace the anomaly sources is made up. And moreover, inquiring an emergency treatment knowledge base based on the emergency anomaly tracing monitoring tensor, matching a basic emergency treatment strategy and generating a dynamic adaptation instruction set for issui