A park safety hazard detection system based on large model

Abstract

The present invention discloses a park safety hazard inspection system based on a large model, which relates to the field of data processing technology. The system includes: an input module, which captures user input data and establishes an input data set, and the input module includes a voice interaction unit and a text interaction unit; a search attention setting module, which is used to read the user's account data and obtain the scene information of the input data set to generate search preferences; a knowledge base module, which is used to store knowledge data; an enhancement module, which is used to establish basic hazard features based on the input data set, and call the knowledge base module to perform enhanced search and establish enhanced search results; a hazard description generation module, which is used to generate hazard descriptions according to enhanced search results, and establish a prescribed mapping between hazard descriptions and prescribed knowledge data. Thus, the technical effects of reducing the difficulty of data management, optimizing response delays, and comprehensively identifying hazards are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a park safety hazard troubleshooting system based on a large model. Background Art

[0002] As industrial parks and production facilities continue to expand in size and become more complex, safety hazard detection has become an important part of ensuring production safety and efficiency. Traditional detection systems mainly rely on manual inspections and empirical judgments. Faced with an increasingly complex safety hazard environment, especially in high-risk areas such as oil leakage, it is difficult to provide timely and accurate warnings and treatments. There are technical problems such as incomplete hazard identification, difficult data management, and slow response speed. Summary of the invention

[0003] The present invention provides a park safety hazard inspection system based on a large model to solve the technical problems of incomplete hazard identification, difficult data management, and slow response speed in the prior art, and achieves the technical effect of reducing data management difficulty, optimizing response delay, and comprehensively identifying hazards.

[0004] The present invention provides a large-scale model-based park safety hazard detection system, comprising:

[0005] An input module includes a voice interaction unit and a text interaction unit. The input module is used to capture user input data and establish an input data set.

[0006] The search attention setting module is used to read the user's account data and obtain the scene information of the input data set, and generate search preferences based on the account data and scene information.

[0007] The knowledge base module is used to store knowledge data, which includes industry knowledge data, execution standard knowledge data, and regulation knowledge data.

[0008] An enhancement module is used to perform hidden danger identification based on the input data set to establish basic hidden danger features, and use the basic hidden danger features as first-level features, the additional data set as second-level features, and the first-level features and second-level features as matching features, and call the knowledge base module through the search preference to perform enhanced search and establish enhanced search results, wherein the additional data set is an associated data set, including location association and equipment association.

[0009] The hidden danger description generating module is used to generate hidden danger description according to the enhanced search results and establish a prescribed mapping between the hidden danger description and prescribed knowledge data.

[0010] In a feasible implementation, the step of using the first level feature and the second level feature as matching features, calling the knowledge base module through the search preference to perform enhanced search, and establishing enhanced search results further includes:

[0011] A third level feature is established, where the third level feature is a dynamically adjusted feature, including real-time environmental features and personnel flow features.

[0012] The multi-objective optimization network is called to perform adaptive attention adjustment of the first-level features, the second-level features, and the third-level features, and an adaptive attention identifier is established.

[0013] The first level features and the second level features are used as matching features, and the third level features are used as dynamic matching features. Based on the search preference, the knowledge base module is called to perform enhanced searches of the corresponding level features.

[0014] The independent enhanced search results are retrospectively aggregated, and the enhanced aggregation results are established based on the retrospective aggregation results and the adaptive attention identifiers.

[0015] In a feasible implementation, the step of performing retrospective aggregation on the independent enhanced search results further includes:

[0016] In the enhanced search process, the matching features and dynamic matching features are used as central matching features, and the knowledge base module is called based on the search preference to perform initial searches of corresponding level features respectively to establish initial search results.

[0017] The initial search results are used to perform search screening to establish a first search result.

[0018] The first search result is used as the central matching feature, and the knowledge base module is called based on the search preference to continue the search, and a second search result is established. Search iterations are performed until a preset number of search iterations is met or the search result does not meet a preset search threshold. The iteration ends and all search results are used as independent search results corresponding to the first-level features, the second-level features, and the third-level features.

[0019] The search results are aggregated from the farthest search results of the independent search results, and the aggregated results are used as upward merging features to perform cluster merging. When all search results in the independent search results are merged, the backtracking aggregation is completed.

[0020] In a feasible implementation, the cluster merging step further includes:

[0021] Establish enhanced communication between independent search results.

[0022] When cluster merging is performed, enhanced compensation of the cluster merging result is performed through the enhanced communication, discrete elimination of the cluster merging result is performed according to the enhanced compensation result, and the cluster merging result is updated based on the discrete elimination result.

[0023] In a feasible implementation, the system further includes an evaluation module, which is used to:

[0024] The input data set is received, a multidimensional feature evaluation of the input data set is performed, and a multidimensional feature evaluation result is established, wherein the multidimensional feature evaluation includes:

[0025] Perform data richness evaluation on the input data set and establish a first evaluation result.

[0026] Perform a data trust evaluation on the input data set to establish a second evaluation result.

[0027] A context dependency evaluation is performed on the input data set to establish a third evaluation result.

[0028] An expansion compensation is generated according to the first evaluation result, the second evaluation result, and the third evaluation result, and the expansion compensation is fed back to the search preference to perform search preference optimization.

[0029] In a feasible implementation, the system further includes:

[0030] The early warning module is used to call the early warning database, perform early warning matching of hidden danger description and specified mapping, and perform early warning output, wherein the early warning database contains the associated mapping of early warning strategies and data, and the early warning strategies include color early warning strategies, flashing early warning strategies, and sound early warning strategies.

[0031] In a feasible implementation, the search preference is generated according to the account data and the scenario information, and the execution step further includes:

[0032] The user's behavior history is called according to the account data, and basic search preferences are generated after feature extraction of the behavior history.

[0033] The scene information is used as matching data to perform optimization of the basic search preference and establish the search preference.

[0034] In a feasible implementation, the system further includes:

[0035] The real-time feedback module is used to receive real-time feedback from users, generate dynamic compensation of feedback preferences, and perform real-time dynamic optimization of search preferences based on the dynamic compensation.

[0036] The present invention discloses a park safety hazard inspection system based on a large model, including: an input module, including a voice interaction unit and a text interaction unit, capturing user input data and establishing an input data set; a search attention setting module, reading user account data, combining the scene information of the input data set, generating search preferences; a knowledge base module, storing industry knowledge data, execution standard knowledge data and prescribed knowledge data; an enhancement module, identifying hidden dangers based on the input data set, establishing basic hidden danger features, using additional data sets as second-level features, matching first-level and second-level features, calling the knowledge base module through search preferences for enhanced search, generating enhanced search results, wherein the additional data sets are associated data sets related to locations and equipment; a hidden danger description generation module, generating hidden danger descriptions based on enhanced search results, and establishing a mapping between hidden danger descriptions and prescribed knowledge data. The park safety hazard inspection system based on a large model disclosed by the present invention solves the technical problems of incomplete hidden danger identification, difficult data management and slow response speed, and achieves the technical effects of reducing data management difficulty, optimizing response delays and comprehensively identifying hidden dangers. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a structural schematic diagram of a park safety hazard troubleshooting system based on a large model of the present invention;

[0038] Figure 2 The present invention is a flow chart of establishing enhanced search results in a large-model-based park safety hazard investigation system.

[0039] Explanation of the accompanying drawings: input module 11, search attention setting module 12, knowledge base module 13, enhancement module 14, hidden danger description generation module 15. DETAILED DESCRIPTION

[0040] The technical solution provided in the embodiments of the present invention is to solve the technical problems of incomplete hidden danger identification, difficult data management, and slow response speed in the prior art. The overall idea adopted is as follows:

[0041] First, an input module including a voice interaction unit and a text interaction unit is used to capture user input data and establish an input data set; then, the search attention setting module reads the user's account data, obtains the scenario information of the input data set, and generates search preferences based on the account data and scenario information; then, the knowledge base module stores knowledge data, which covers industry knowledge data, execution standard knowledge data and prescribed knowledge data; then, the enhancement module identifies hidden dangers based on the input data set, establishes basic hidden danger features, and uses them as first-level features, and uses additional data sets (including location associations, equipment associations) as second-level features, and uses these two types of features as matching features. The knowledge base module is called through search preferences to perform enhanced search and generate enhanced search results; then, the hidden danger description generation module generates hidden danger descriptions based on the enhanced search results, and establishes prescribed mappings between hidden danger descriptions and prescribed knowledge data.

[0042] The above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods of the specification to better understand the above technical solution. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments of the present invention. It should be understood that the present invention is not limited to the example embodiments used only to explain the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, it should be noted that, for the convenience of description, only the parts related to the present invention are shown in the drawings, rather than all of them.

[0043] Embodiment 1

[0044] Figure 1 The present invention is a flow chart of a park safety hazard troubleshooting system based on a large model, wherein the system includes:

[0045] The input module 11 includes a voice interaction unit and a text interaction unit. The input module 11 is used to capture user input data and establish an input data set.

[0046] Specifically, first, the user input data is captured and obtained through the input module and the voice interaction unit and text interaction unit configured in the module to obtain basic data for troubleshooting hidden dangers. Exemplarily, the obtained user input data refers to hidden danger items that need to be manually entered by the target user. These hidden danger items that need to be entered are often colloquial, unstructured, and simplified descriptions, lacking clear and professional descriptions of hidden danger items, and prone to memory errors of park safety regulations or safety knowledge, which is not convenient for subsequent understanding and processing.

[0047] Optionally, the voice interaction unit includes a voice recognition module and a voice processing module. The voice recognition module is responsible for extracting voice data from the user's voice input and converting it into text or coded form; the voice processing module is responsible for analyzing voice features, judging the user's intention, and forming structured data output. The text interaction unit is used to capture text data input by the user through input devices such as keyboards and touch screens, and use natural language processing (NLP) technology to perform semantic analysis on the text data, including keyword extraction, syntactic analysis, etc. to capture key content in the text.

[0048] For example, if the user enters the hidden danger items by voice, the voice interaction unit of the system converts these items into text through voice recognition technology. Since the hidden danger items are usually colloquial and unstructured descriptions, the system will perform preliminary grammatical analysis and semantic understanding on the entered data to help standardize it and make the sentences fluent and clear.

[0049] Exemplarily, when the user chooses to input hidden danger entries in text form, the system's text interaction unit further structures these entries through the text analysis function. Since users usually tend to describe hidden dangers in a simple and direct manner, it is necessary to make preliminary simple corrections to the input content through rules and models, such as automatic completion, error correction and optimization, to make the description clearer.

[0050] Through intelligent analysis and processing, the system can convert the colloquial, unstructured hazard descriptions entered by users into structured, standardized entries, thereby improving the readability and accuracy of the entries and making it easier for companies to understand and take action.

[0051] The search attention setting module 12 is used to read the user's account data and obtain the scene information of the input data set, and generate the search preference according to the account data and the scene information.

[0052] Specifically, reading the user's account data and obtaining the scenario information of the input data set is to ensure that the user's personalized scenario is accurately located during the hidden danger investigation process, thereby providing a more accurate range reference for the subsequent knowledge base call and retrieval, which helps to improve the subsequent retrieval and matching efficiency.

[0053] Specifically, the user's account data includes identity information, authority level, department, and operation history. The purpose of reading account data is to obtain customized hidden danger information and input data sets related to the user based on the user's identity and authority. For example, the scope of authority and responsibilities involved in hidden danger investigation for security administrators and ordinary employees are different, so the system can determine the scope of knowledge involved for different users based on account data. Exemplarily, after verifying the user's identity through the user's login credentials (such as user name, password, or biometrics), the account data is extracted from the user database.

[0054] Specifically, scene information refers to the specific environment and background data of hidden danger inspection, such as the inspection area where the user is located, the type of current inspection task, the status of equipment and facilities on site, etc. The purpose of obtaining scene information is to help the system provide users with more accurate suggestions and feedback during the hidden danger inspection process. The scene information is combined with the user's account data to help accurately locate personalized knowledge data that meets the user's characteristics. Exemplary, it includes obtaining the user's actual location through GPS or internal positioning system, and determining the park area or equipment that the user is inspecting based on pre-arranged inspection tasks and schedules.

[0055] In some embodiments, the search preference is generated according to the account data and the scenario information, and the execution step further includes:

[0056] The user's behavior history is called according to the account data, and basic search preferences are generated after feature extraction of the behavior history; the basic search preferences are optimized using the scenario information as matching data to establish search preferences.

[0057] Specifically, the basic search preferences extracted based on the user's historical behavior are the park scene feature data that the user has processed. However, in actual applications, the user's behavior history may only cover certain specific areas or scenes in the park, rather than all possible scenes. That is, the basic search preferences are a subset of the park scene feature set involved by the actual user, and cannot cover other scene features that have not yet been operated. Therefore, it is necessary to optimize the basic search preferences based on scene information to ensure the completeness of the obtained search preferences.

[0058] Specifically, the user's behavior history includes the user's hidden danger inspection records, search records, click preferences, etc. in previous operations. By analyzing these historical data, the system can extract the user's operating habits, preferences, and common scenarios. Exemplarily, basic search preferences include frequently searched keywords, hidden danger types that tend to be concerned about, commonly used inspection methods, etc. Feature extraction methods include data clustering, keyword analysis, and pattern recognition. By analyzing the user's behavior history, the basic search preferences generated reflect the user's common habits and characteristics in previous operations, providing a preliminary framework for subsequent search optimization.

[0059] Specifically, based on the scenario information, the previously generated basic search preferences are optimized. Exemplarily, the system first traverses the basic search preferences for correlation matching according to the current device state or specific scenario of the user, such as extracting the feature data of the corresponding scenario features and scenario information in the basic search preferences; then, according to the preference degree of the basic search preferences (such as the frequency or frequency of occurrence in the behavior history), the weight and priority of the search results are adjusted to ensure that the feature data related to the scenario and with high user preference is displayed first.

[0060] The above steps, using scene information as matching data, help the system further optimize basic search preferences, thereby making the subsequent search results more compatible with the user's current task scenario, improving the efficiency and accuracy of troubleshooting. For example, when a user performs a hidden danger troubleshooting task in a specific area, the system will prioritize the extraction of hidden dangers and specifications related to the area or equipment, reducing interference from irrelevant information.

[0061] The knowledge base module 13 is used to store knowledge data, which includes industry knowledge data, execution standard knowledge data, and regulation knowledge data.

[0062] Specifically, industry knowledge data refers to professional knowledge and information related to a specific industry or field, providing users with all-round support when performing hidden danger inspections and ensuring the compliance and professionalism of operations.

[0063] Specifically, industry knowledge data contains information such as safety requirements, operating specifications, common hidden danger types, and risk prevention measures for multiple specific industries and multiple scenarios. When users perform safety inspections in the park, by calling industry knowledge data to provide specific knowledge information for the user's current industry or field, it helps to improve the accuracy and relevance of the inspection.

[0064] Specifically, implementation standard knowledge data refers to various standards related to hidden danger detection and safety management. These standards define the rules and operating procedures that must be followed in the safety management process, such as ISO standards, production safety standards, local standards, industry standards, etc. When users perform hidden danger detection, the system can automatically retrieve and display the implementation standards related to the current inspection item according to task requirements to ensure that the detection meets the specified requirements.

[0065] Specifically, the prescribed knowledge data refers to the regulations and operating specifications related to the park safety management and hidden danger investigation. For example, the park regulations and enterprise regulations, together with the above-mentioned implementation standard knowledge data, reflect the specific requirements for hidden danger investigation, rectification and treatment. The prescribed knowledge data can help enterprises avoid compliance risks and improve the legitimacy and effectiveness of hidden danger investigation.

[0066] The enhancement module 14 is used to perform hidden danger identification based on the input data set to establish basic hidden danger features, and use the basic hidden danger features as the first-level features, the additional data set as the second-level features, and the first-level features and the second-level features as matching features, and call the knowledge base module through the search preference to perform enhanced search to establish enhanced search results, wherein the additional data set is an associated data set, including location association and equipment association.

[0067] Specifically, first, the content of the data input by the user (such as hidden danger items reported by the user, scenario descriptions, etc.) is analyzed to preliminarily identify potential safety hazards and extract corresponding hidden danger features. For example, based on the keywords or professional terms in the user input data set, the hidden danger types and hidden danger scenarios of the basic hidden dangers are determined and output as basic hidden danger features.

[0068] Specifically, the additional data sets include location-related data and device-related data related to hidden dangers, which provide richer contextual information for hidden danger features to help the system identify and search for hidden dangers more accurately. Among them, location-related data refers to additional information related to the location where the hidden danger occurs. For example, the specific location where the hidden danger occurs is obtained through GPS or the campus positioning system, and the historical hidden danger records or risk conditions in the area are associated. Device-related data refers to additional data related to the equipment involved in the hidden danger. For example, based on the equipment hidden danger reported by the user, the system obtains the status data, historical fault records, performance parameters or maintenance status of the equipment.

[0069] Furthermore, the basic hidden danger features are taken as the first-level features, and the location association and equipment association in the additional data set are taken as the second-level features, and the two are combined through the feature matching algorithm to form a complete feature model. This feature model is the core basis for subsequent enhanced search.

[0070] In some embodiments, Figure 2 As shown, the first level feature and the second level feature are used as matching features, the knowledge base module is called through the search preference to perform enhanced search, and enhanced search results are established. The execution step also includes:

[0071] Establish third-level features, which are dynamically adjusted features, including real-time environmental features and personnel flow features; call the multi-objective optimization network to perform adaptive attention adjustment of first-level features, second-level features, and third-level features, and establish adaptive attention identification; use the first-level features and second-level features as matching features, use the third-level features as dynamic matching features, and call the knowledge base module based on search preferences to perform enhanced searches of corresponding level features respectively; perform retrospective aggregation of independent enhanced search results, and establish enhanced aggregation results based on the retrospective aggregation results and the adaptive attention identification.

[0072] Specifically, the third-level features refer to the features that the system dynamically adjusts when performing a search. The third-level features are generated as the environment or scene changes in real time, and are used to ensure that the search results are closely integrated with the on-site situation, thereby improving the accuracy and timeliness of hidden danger detection.

[0073] Specifically, the third-level features include real-time environmental features and personnel flow features. Among them, real-time environmental features are real-time data related to the current environment obtained through sensors or IoT devices. Such as temperature, humidity, air quality, noise level, etc. Personnel flow features characterize the activities of people in the park, such as the density of people or flow patterns in a certain area, which are used to determine the impact range or urgency of potential hidden dangers. In other words, the higher the density of people and the greater the flow of people, the greater the corresponding hidden dangers.

[0074] Specifically, the multi-objective optimization network is called to adaptively adjust the focus of the first-level features (basic hidden danger features), second-level features (additional data sets), and third-level features (dynamically adjusted features). That is, according to the feature weights of different scenarios, the priority of each level of features is dynamically adjusted to ensure that the system can focus on the most critical features.

[0075] For example, the weights of features at each level are calculated through multi-objective optimization (such as genetic algorithm or Pareto optimal solution), and the focus of features is dynamically adjusted according to the needs of different scenarios. For example, in areas with high population density, higher weights are given to features of personnel flow; and in scenarios where equipment is running at high load or the downstream of equipment is of high importance, the priority of equipment-related features is increased.

[0076] Furthermore, based on the first-level features (basic hidden danger features) and the second-level features (additional data sets such as location association and equipment association), the third-level features are used as dynamic auxiliary features to perform a search in the knowledge base module called based on the search preference. Specifically, the system dynamically adjusts the search results based on the real-time environmental characteristics and personnel flow characteristics. For example, if the environmental characteristics show an abnormally high temperature, the system will give priority to displaying the equipment troubleshooting knowledge or nouns related to the situation; if the population is dense, emergency evacuation plans or handling standards for hidden dangers in high-risk areas are recommended. Dynamic matching helps to generate more timely search results based on the current real-time situation and helps to accurately obtain knowledge data related to the current hidden danger scenario.

[0077] Specifically, the enhanced aggregation results include retrospective aggregation results and corresponding adaptive attention marks. The retrospective aggregation results represent the hidden danger reporting information matched in the hidden danger investigation, and the adaptive attention marks represent the contribution of different hidden danger reporting information to the hidden danger, that is, the user's attention priority for different hidden danger reporting information in the current scenario.

[0078] In some implementations, the step of performing retrospective aggregation on the independent enhanced search results further includes:

[0079] In the enhanced search process, the matching features and dynamic matching features are used as central matching features, and the knowledge base module is called based on the search preference to perform initial searches of the corresponding level features respectively to establish an initial search result; the initial search result is used for search screening to establish a first search result; the first search result is used as the central matching feature, and the knowledge base module is called based on the search preference to continue searching to establish a second search result, and search iterations are performed until a preset number of search iterations is met or the search result does not meet a preset search threshold, and the iteration ends, and all search results are used as independent search results corresponding to the first level feature, the second level feature, and the third level feature; the search results are aggregated from the farthest search result of the independent search result, and the aggregated result is used as an upward merge feature to perform cluster merging, and when all search results within the independent search result are merged, the backtracking aggregation is completed.

[0080] Specifically, first, the first-level features, second-level features, and third-level features are used as central matching features to start the initial search, retrieve the hidden danger information (such as hidden danger index maps), regulations, standards, etc. related to these features, and generate initial search results. Then, the initial search results are screened to filter out search results that are irrelevant to hidden dangers or have low relevance. Optionally, the screening criteria are determined based on the feature weights, priorities, information quantity constraints, and other feasible preset rules set by the search preferences. Exemplarily, based on the preset information quantity constraints, a portion of the initial search results that have the highest degree of matching with the central matching features are screened and retained, and output as the first search result.

[0081] Furthermore, based on the initial screening, the key information in the first search result is used as the new central matching feature, and the knowledge base module is called again based on the search preference to continue the search iteration. The search iteration is used to dig deeper into the first search result, expand the search scope, and generate deeper related search results. When the number of iterations reaches the upper limit, or the similarity of the search results is no longer significantly improved, the system will end the iteration and aggregate all search results to form independent search results corresponding to the first-level features, second-level features, and third-level features. Among them, the search results of the first-level features, second-level features, and third-level features are stored separately for subsequent processing. Through search iteration and result aggregation, it helps to ensure that all relevant information is included in the search results, providing a high-quality data foundation for subsequent retrospective aggregation.

[0082] Furthermore, starting from the farthest search result in the independent search results, the aggregation operation is performed upward in sequence. The process of retrospective aggregation includes merging standards, regulations, and knowledge information of different levels or categories, retaining the largest range, the most stringent, and the best information, and achieving effective integration. Among them, the farthest search result represents the deepest and most relevant information in the iterative process. Through the bottom-up, local-to-global aggregation method, a more comprehensive search result can be generated, and the enhanced aggregation result is consistent with all the information in the knowledge base module.

[0083] In some implementations, performing cluster merging, the performing step further includes:

[0084] Establish enhanced communication between independent search results; when cluster merging is performed, enhance compensation of the cluster merging results is performed through the enhanced communication, discrete elimination of the cluster merging results is performed according to the enhanced compensation results, and the cluster merging results are updated based on the discrete elimination results.

[0085] Specifically, enhanced communication refers to information exchange and coordination between independent search results during cluster merging, ensuring that the search results can complement, correct and improve each other when merged. It also ensures that search results from different sources (such as results with different feature levels) remain consistent and complete during the merging process.

[0086] For example, in the process of cluster merging, an information exchange strategy between independent search results (under what circumstances information exchange is performed, the types of information allowed to be exchanged, etc.) is established so that search results of different levels of features can exchange information with each other. For example, the search results of basic hidden danger features can supplement the missing information in the real-time environment features through the enhanced communication mechanism.

[0087] Furthermore, on the basis of establishing enhanced communication, independent search results are clustered and merged, and similar or related search results are classified and integrated to reduce redundancy and improve the readability and effectiveness of the results. For example, first, clustering algorithms (such as K-means clustering, hierarchical clustering, etc.) are used to group independent search results according to feature similarity. For example, search results with the same hidden danger type but from different levels of features (such as basic hidden danger features and equipment-related features) will be grouped into the same group. Then, in the cluster merging process, enhanced compensation is performed through the enhanced communication mechanism, that is, if some key information is missing in a cluster merging result, this missing information is supplemented by other search results. Next, the cluster merging results are optimized through the discrete elimination mechanism to remove irrelevant or redundant information. For example, knowledge data far from the cluster center is regarded as irrelevant items that do not meet expectations or affect the quality of search results, and is eliminated. Finally, according to the results of discrete elimination, the cluster merging results are dynamically updated to ensure that the final search results obtained by the user are streamlined and optimized, which helps to improve the effectiveness and quality of search results.

[0088] The hidden danger description generating module 15 is used to generate hidden danger description according to the enhanced search results, and establish a prescribed mapping between the hidden danger description and prescribed knowledge data.

[0089] Optionally, the system converts the enhanced search results into corresponding hazard descriptions based on a preset hazard description format (such as form template, table template, database format, etc.). The hazard description is a structured and standardized representation of safety hazard entries, and the description of safety hazards is more comprehensive and rich than the input data set, with better readability and explanation. This helps users intuitively understand the hazard problem and its corresponding knowledge data.

[0090] Specifically, through natural language generation technology, the system converts structured hidden danger features into hidden danger descriptions that meet the requirements of the target park or target scene. For example, if the enhanced search results include "short circuit of wires, fire hazard", the system will automatically generate a hidden danger description of "there is a short circuit of wires in this area, and there is a potential risk of causing a fire". Through natural language generation technology, the system can automatically present complex hidden danger information in a clear and easy-to-understand way, helping users quickly understand the nature and severity of the hidden dangers.

[0091] Specifically, the generation of hidden danger descriptions also involves standardization of hidden danger descriptions. The non-standard terms in the user input are matched with the standard expressions in the terminology library through the term replacement algorithm, and the replacement is automatically completed to ensure the comprehensiveness and accuracy of the description. For example, "a thread" or "a line" is standardized into a specific unit of quantity, such as "0.1 mm"; if the user describes the gas leakage as "one unit", it can be converted into "1 cubic meter / hour". The standardization of hidden danger descriptions ensures the consistency and good matching of the description content in the hidden danger investigation process.

[0092] In some embodiments, the execution steps of the system further include:

[0093] Receive the input data set, perform multidimensional feature evaluation on the input data set, and establish a multidimensional feature evaluation result, wherein the multidimensional feature evaluation includes: performing data richness evaluation on the input data set to establish a first evaluation result; performing data trust evaluation on the input data set to establish a second evaluation result; performing context dependency evaluation on the input data set to establish a third evaluation result; generating expansion compensation based on the first evaluation result, the second evaluation result, and the third evaluation result, feeding the expansion compensation back to the search preference, and performing search preference optimization.

[0094] Specifically, data richness refers to whether the content of the input data set is comprehensive and detailed, and whether it covers all the key information required for hidden danger investigation. This includes content coverage analysis (i.e., whether the information types are complete) and information integrity detection (i.e., whether the information dimensions are complete).

[0095] Specifically, data trust evaluation is used to evaluate whether the information in the input data set is reliable and trustworthy. In particular, when the data comes from multiple sources or different users, the system needs to determine whether the data is trustworthy to avoid erroneous information affecting the results of hidden danger detection. Optionally, data trust evaluation includes data source verification, data consistency check, transmission integrity check, etc. The second evaluation result generated by the data trust evaluation helps the system identify and filter possible unreliable information to ensure that hidden danger detection relies on trustworthy data.

[0096] Specifically, context dependency evaluation is to determine whether the information in the input data set is relevant to the current scene or task, among which data with strong context dependency is more helpful for hidden danger detection and risk assessment. Optionally, context dependency evaluation involves scene matching analysis and timeliness analysis. The third evaluation result generated by context dependency evaluation helps the system determine whether the input data is closely related to the current scene and whether the input data has good timeliness, thereby improving the accuracy of hidden danger detection.

[0097] Specifically, supplementary optimization is performed based on the first evaluation results (data richness), the second evaluation results (data trust), and the third evaluation results (context dependency), including supplementary data source calls, which automatically supplement the information in the input data set by calling the knowledge base, device data, or other external data sources. Trust adjustment and marking, that is, automatically marking untrustworthy data, and reducing its weight or excluding untrustworthy data in subsequent searches. According to the results of the context dependency evaluation, the search preference is adjusted, and the weight or order in the search preference is temporarily adjusted to prioritize the search for information related to the current task.

[0098] The above steps, by evaluating the multi-dimensional features of the input data set, can determine the integrity, credibility and contextual relevance of the data, thus providing a basis for subsequent search and compensation.

[0099] Furthermore, the system also includes an early warning module, which is used to call the early warning database, perform early warning matching of hidden danger descriptions and specified mappings, and perform early warning outputs, wherein the early warning database contains an associated mapping of early warning strategies and data, and the early warning strategies include color warning strategies, flashing warning strategies, and sound warning strategies.

[0100] Specifically, the early warning database includes various early warning strategies and the associated mapping between early warning strategies and specific hidden danger data. When the system detects a potential hidden danger, it will extract the relevant early warning strategies and data from the early warning database to execute the early warning.

[0101] Specifically, after the early warning database is called, the system will match the early warning according to the hidden danger description and regulations, that is, select the appropriate early warning strategy according to the specific situation of the hidden danger. Among them, the hidden danger description is a detailed description of the detected potential safety hidden danger, exemplarily including the hidden danger type, location, severity and other information.

[0102] Specifically, the early warning strategies include color early warning strategy, flashing early warning strategy and sound early warning strategy. Among them, the color early warning strategy uses signal lights or display screens of different colors to indicate different levels of early warning. For example, green indicates safety, yellow indicates warning, and red indicates danger. The flashing early warning strategy uses flashing lights or display screens to attract attention, and the flashing frequency and pattern are adjusted according to the severity and type of hidden dangers. The sound early warning strategy uses sound signals (such as alarm sounds, voice prompts) to remind relevant personnel. Similarly, the type and volume of the sound can be adjusted according to the severity and type of hidden dangers.

[0103] The above-mentioned early warning module selects the appropriate early warning strategy according to the severity of the hidden danger, so as to optimize the allocation of resources. At the same time, through multi-path early warning strategies (color, flashing, sound), it can quickly attract the attention of relevant personnel, thereby improving the early warning effect and hidden danger response efficiency.

[0104] Furthermore, the system also includes a real-time feedback module for receiving real-time feedback from users, generating dynamic compensation of feedback preferences, and performing real-time dynamic optimization of search preferences based on the dynamic compensation.

[0105] Specifically, real-time user feedback is the immediate feedback provided by users regarding the system’s warnings or other functions during the use of the system. This feedback can be opinions, suggestions or complaints submitted through the user interface, or user behavior data detected by sensors.

[0106] Specifically, after receiving real-time user feedback, the real-time feedback module generates dynamic compensation for feedback preferences. The dynamic compensation characterizes the components of errors, errors or deviations in the operation of the system, and is used to compensate and correct search preferences. For example, the dynamic compensation content includes erroneous or missing user scenarios, user industries and other information. By generating corresponding dynamic compensation based on user feedback preferences, and adjusting and optimizing them, we can better meet the needs and preferences of users, and continuously improve the accuracy of the hidden danger elimination process.

[0107] In summary, the park safety hazard investigation system based on a large model provided by the present invention has the following technical effects:

[0108] Through the input module, including the voice interaction unit and the text interaction unit, the user's input data is captured and the input data set is established; the search attention setting module reads the user account data, combines the scene information of the input data set, and generates the search preference; the knowledge base module stores the industry knowledge data, the execution standard knowledge data, and the prescribed knowledge data; the enhancement module identifies hidden dangers based on the input data set, establishes basic hidden danger features, takes the additional data set as the second-level feature, matches the first-level and second-level features, calls the knowledge base module through the search preference for enhanced search, and generates enhanced search results, where the additional data set is an associated data set related to the location and equipment; the hidden danger description generation module generates hidden danger descriptions based on the enhanced search results, and establishes a mapping between the hidden danger descriptions and the prescribed knowledge data. This achieves the technical effect of reducing the difficulty of data management, optimizing response delays, and comprehensively identifying hidden dangers.

[0109] It should be understood that the embodiments disclosed in the present invention and the above description can enable those skilled in the art to use the present invention to implement the present invention. At the same time, the present invention is not limited to the above-mentioned embodiments. It should be understood that those skilled in the art can still modify the technical solutions recorded in the above-mentioned embodiments, or replace some of the technical features therein by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A park safety hazard investigation system based on a large model, characterized in that: The system comprises: An input module, the input module comprising a voice interaction unit and a text interaction unit, using the input module to capture user input data and establish an input data set; A search attention setting module, used to read the user's account data and obtain the scene information of the input data set, and generate the search preference according to the account data and the scene information; A knowledge base module, used to store knowledge data, including industry knowledge data, implementation standard knowledge data, and regulation knowledge data; an enhancement module, configured to perform hidden danger identification based on the input data set to establish basic hidden danger features, and use the basic hidden danger features as first-level features, the additional data set as second-level features, the first-level features and the second-level features as matching features, and call the knowledge base module through the search preference to perform enhanced search and establish enhanced search results, wherein the additional data set is an associated data set, including location association and equipment association; A hidden danger description generating module, used to generate hidden danger description according to the enhanced search results, and establish a prescribed mapping between hidden danger description and prescribed knowledge data; The step of using the first level feature and the second level feature as matching features, calling the knowledge base module through the search preference to perform enhanced search, and establishing enhanced search results further includes: Establishing third-level features, wherein the third-level features are dynamically adjusted features, including real-time environmental features and personnel flow features; Calling a multi-objective optimization network to perform adaptive attention adjustment of first-level features, second-level features, and third-level features, and establishing an adaptive attention identifier; The first level feature and the second level feature are used as matching features, and the third level feature is used as a dynamic matching feature, and the knowledge base module is called based on the search preference to perform enhanced searches of the corresponding level features respectively; Performing retrospective aggregation on the independent enhanced search results, and establishing enhanced aggregation results based on the retrospective aggregation results and the adaptive attention identifier; The step of performing backtracking aggregation on the independent enhanced search results also includes: In the enhanced search process, the matching features and the dynamic matching features are used as the central matching features, and the knowledge base module is called based on the search preference to perform initial searches of the corresponding level features respectively, and establish initial search results; Perform search screening based on the initial search results to establish a first search result; The first search result is used as the central matching feature, the knowledge base module is called based on the search preference to continue searching, a second search result is established, and a search iteration is performed until a preset number of search iterations is met or the search result does not meet a preset search threshold, the iteration ends, and all the search results are used as independent search results corresponding to the first level feature, the second level feature, and the third level feature; Aggregate the search results from the farthest search results of the independent search results, and use the aggregated results as upward merging features to perform cluster merging, and complete the backtracking aggregation when all search results in the independent search results are merged; The cluster merging step further includes: Establishing enhanced communication between independent search results; When cluster merging is performed, enhanced compensation of the cluster merging result is performed through the enhanced communication, discrete elimination of the cluster merging result is performed according to the enhanced compensation result, and the cluster merging result is updated based on the discrete elimination result.

2. A park safety hazard investigation system based on a large model as described in claim 1, characterized in that: The system further comprises an evaluation module, for: The input data set is received, a multidimensional feature evaluation of the input data set is performed, and a multidimensional feature evaluation result is established, wherein the multidimensional feature evaluation includes: Performing data richness evaluation on the input data set to establish a first evaluation result; Performing a data trust evaluation on the input data set to establish a second evaluation result; Performing a context dependency evaluation on the input data set to establish a third evaluation result; An expansion compensation is generated according to the first evaluation result, the second evaluation result, and the third evaluation result, and the expansion compensation is fed back to the search preference to perform search preference optimization.

3. A park safety hazard investigation system based on a large model as described in claim 1, characterized in that: The system further comprises: The early warning module is used to call the early warning database, perform early warning matching of hidden danger description and specified mapping, and perform early warning output, wherein the early warning database contains the associated mapping of early warning strategies and data, and the early warning strategies include color early warning strategies, flashing early warning strategies, and sound early warning strategies.

4. A large-model-based park safety hazard investigation system as claimed in claim 1, characterized in that: The step of generating the search preference according to the account data and the scenario information further includes: Retrieving the user's behavior history according to the account data, and generating basic search preferences after extracting features from the behavior history; The scene information is used as matching data to perform optimization of the basic search preference and establish the search preference.

5. A large-model-based park safety hazard investigation system as described in claim 2, characterized in that: The system further comprises: The real-time feedback module is used to receive real-time feedback from users, generate dynamic compensation of feedback preferences, and perform real-time dynamic optimization of search preferences based on the dynamic compensation.

Images