Construction method of vertical field large model

Abstract

The invention discloses a construction method of a vertical field large model, which comprises the following steps of: constructing a vertical field large model framework which takes a general large model as a base, takes data, knowledge and model multi-source driven knowledge enhancement engine as a core and is combined with technical application; constructing a data engine, a knowledge engine and a model engine in the vertical field large model framework; based on mapping association of a multi-risk scene and a business application scene in an urban lifeline, a vertical domain large model application system and a system platform are constructed through combined linkage of a general large model and a knowledge enhancement engine. Relates to the technical field of large models, and solves the problem that an existing general large model is difficult to realize reasonable analysis of professional data, accurate reply of problems, intelligent deduction of professional scenes and integrated application of professional businesses.

Description

Technical Field

[0001] The present invention belongs to the technical field of large models, and specifically is a method for constructing a large model in a vertical field. Background Art

[0002] As urban systems become increasingly complex, lifeline safety incidents are characterized by rapid evolution, coupled superposition, and chain-network transmission, placing multi-dimensional demands on risk prevention and control, including speed and precision. The emergence of artificial intelligence has brought revolutionary opportunities for urban lifeline safety, but current large-scale model applications still primarily rely on general-purpose large-scale models. Risk analysis and prevention in specialized verticals lacks a specialized knowledge base for urban infrastructure (such as water supply, power supply, and transportation), making it difficult to identify the vulnerabilities and coupling risks of complex systems. Limited understanding of industry standards, safety regulations, and policies and regulations can easily lead to analytical results that deviate from actual needs. Furthermore, it is difficult to accurately reason about and simulate cascading failures or chain reactions. Summary of the Invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a method for constructing a large model in a vertical field, which is used to solve the technical problems that the current general large model is difficult to identify the vulnerability and coupling risks of complex systems; the understanding of industry standards, safety specifications and policies and regulations is limited, which easily leads to the analysis results deviating from actual needs; and it is difficult to accurately infer cascading failures or chain reactions. Therefore, focusing on the above problems, the present invention establishes a large model application framework and platform for vertical fields by taking a general large model as the basis, with data, knowledge, and model multi-driven knowledge enhancement engine as the core, and combining the technical application of multiple risk scenarios in urban lifelines to form an overall "1+3+N" architecture that can solve the above problems.

[0004] To achieve the above objectives, the first aspect of the present invention provides a method for constructing a large vertical domain model, comprising:

[0005] Build a vertical field big model framework based on a general big model, with a knowledge enhancement engine driven by data, knowledge, and models from multiple sources as the core, and combined with technology applications;

[0006] Construct the data engine, knowledge engine and model engine in the vertical field large model knowledge enhancement engine;

[0007] Based on the mapping association between multiple risk scenarios and business application scenarios in urban lifelines, the business logic and business functions of the application scenarios are analyzed, and through the combination and linkage of general big models and knowledge enhancement engines, an application platform based on the vertical domain big model of urban lifelines is constructed.

[0008] Preferably, the data engine is used to collect, clean and fuse multi-source heterogeneous data, and build a data standard system through process-based data processing; as well as to perform spatiotemporal alignment, anomaly detection, trend analysis and hidden danger pattern mining on multi-source heterogeneous data; and to provide data input for the knowledge engine and model engine.

[0009] Preferably, the knowledge engine realizes the extraction, expression and dynamic update of domain knowledge based on a general large model and a knowledge graph; as well as indicator analysis, rule extraction, association analysis, response generation and similar case recommendation of domain knowledge, and forms a knowledge graph.

[0010] Preferably, the model engine is used to construct a model association representation of the risk transfer process; the risk transfer process refers to the transfer and propagation process of materials, energy, and information of an emergency; the model association representation refers to the loose coupling and collaborative reasoning between the physical model, data-driven model, and knowledge meta-rule model, that is, the coupling association between different mechanism models in actual scenarios.

[0011] Preferably, the data engine, the knowledge engine and the model engine can be jointly driven by a collaborative mechanism in addition to self-driving for functional application;

[0012] The collaborative mechanism includes: updating entity relationships in the knowledge graph through multi-source data, constraining physical models through causal relationships in the knowledge graph, guiding the construction of data standards through model prediction results, and performing analytical operations through the data engine. The data engine analytical operations refer to the calculation of the operation formulas in the model engine by strengthening the mathematical computing capabilities of the general large model, such as function solving.

[0013] Preferably, the data engine includes data identification and extraction, model operation data protocol, data processing and modeling tools, and builds a standardized data processing system through multi-source heterogeneous data "data collection-data processing-data storage" to form a city lifeline thematic database.

[0014] Preferably, the construction method of the knowledge engine includes:

[0015] A knowledge base is built based on standard specifications, accident cases, and historical maintenance records. A general large model and dynamic knowledge graph are used to carry out knowledge meta-description and association, and a knowledge association network is constructed through semantic understanding, relationship mining, causal reasoning and feature analysis of risk events.

[0016] It should be noted that the knowledge engine is structured by an associated semantic network composed of several knowledge elements.

[0017] Preferably, the model engine is constructed by:

[0018] Based on the mining of risk event correlation relationships, the sorting of infrastructure's "disaster-causing-disaster-response" attributes and accident feature analysis, combined with abstract modeling methods, the "soft reasoning + hard reasoning" method is used to realize single / cross-system risk identification and analysis, transfer logic relationships and implicit association expressions, form a model rule library, and obtain a sequential, quantified and visualized urban lifeline single / cross-system model engine.

[0019] The soft reasoning is a process of making empirical and speculative judgments on the evolution of emergency risk based on knowledge base accident cases, news reports, and standard specifications.

[0020] The hard reasoning is based on the analysis of the "disaster-causing-disaster-bearing" mechanism of infrastructure and is a process of reasoning about the evolution process and path of emergencies based on the mechanism model.

[0021] Compared with the existing technology, the beneficial effects of the present invention are: based on a general large model, with data, knowledge, and model multi-driven knowledge enhancement engine as the core, and combined with the technical application of multiple risk scenarios in urban lifelines to establish a vertical field large model application platform, forming an overall framework of "1+3+N". Among them, the data engine based on the knowledge enhancement engine can realize intelligent perception and deep understanding of multi-domain and multi-dimensional data, realize lifeline risk status analysis, evolution trend prediction, potential feature extraction, fault precursor prediction, etc.; the knowledge engine based on the knowledge enhancement engine can realize risk event relationship mining and cognitive reasoning, realize risk feature extraction, risk association reasoning and disposal strategy optimization, etc.; the model engine based on the knowledge enhancement engine can realize rapid analysis and integrated linkage of models under abnormal events, realize modeling, representation and deduction of the evolution path of single system / cross-system risk events of lifelines; based on the knowledge enhancement engine "data-knowledge-model" multi-source mutual drive, it can realize dynamic coupling of multi-source engines, realize cross-system risk assessment of urban lifelines, multi-source collaborative perception, accurate decision-making and judgment, targeted warning push, and emergency collaborative linkage. The results can provide support for the construction of system platforms in professional fields, further promote the development of urban lifeline safety towards risk controllability, intelligent prediction, efficient response and integrated protection, and achieve a paradigm shift from "passive emergency control" to "active prediction and defense". BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 A schematic diagram of the process of constructing a large vertical field model of the present invention;

[0024] Figure 2 This is the structural block diagram of the vertical field large model of the present invention. DETAILED DESCRIPTION

[0025] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0026] See also Figure 1-Figure 2 The first embodiment of the present invention provides a method for constructing a large vertical model, including:

[0027] Build a vertical field big model framework based on a general big model, with a knowledge enhancement engine driven by data, knowledge, and models from multiple sources as the core, and combined with technology applications;

[0028] Build data engines, knowledge engines, and model engines in the vertical field big model framework;

[0029] Based on the mapping association between multiple risk scenarios and business application scenarios in the urban lifeline, and analyzing the business logic and business functions of the business application scenarios; and through the combined linkage of model base, data engine, knowledge engine and model engine, a vertical field large model based on the urban lifeline is obtained; and further formed based on the "1+3+N" urban lifeline vertical field large model application platform, to achieve multi-source collaborative perception of urban lifeline risks, single / cross-system risk assessment, accurate decision-making and analysis, targeted intelligent early warning, accurate prevention and control and emergency coordinated linkage.

[0030] The construction of the vertical domain big model in this embodiment refers to forming a vertical domain big model in the form of a plug-in library based on a general big model through a joint knowledge enhancement engine. The above-mentioned risk scenarios include but are not limited to single-system scenarios, including natural disasters, geological disasters, man-made damage, facility-induced disasters and other scenarios, such as construction damage, pipeline abnormalities, underground cavities in roads, urban waterlogging, dynamic load and heavy load damage, ground subsidence, etc.; cross-system scenarios, such as scenarios where different systems influence each other, such as underground cavities causing pipeline abnormalities, and further affecting surrounding buildings and facilities, leading to fires and explosions, etc.

[0031] Based on a universal large-scale model, with a knowledge enhancement engine driven by data, knowledge, and models at its core, and integrating technical applications for multiple risk scenarios in urban lifelines, a vertical large-scale model application platform is established, forming a "1+3+N" overall framework. "1+3+N" refers to an overall platform architecture based on a single model foundation and three knowledge enhancement engine drivers: the data engine, the knowledge engine, and the model engine, enabling N scenarios of application.

[0032] The above-mentioned multi-source collaborative perception of urban lifeline risks refers to the construction of an intelligent entity for analyzing hidden dangers, faults and risks of urban lifelines based on data engines and knowledge engines, and the realization of predictions of hidden dangers, faults and risks of urban lifelines based on data mining and correlation analysis. Furthermore, regarding the prediction of hidden dangers, faults and risks, data mining can be carried out for specific regions, cities, counties, provinces and the whole country to clarify the high incidence trends of urban infrastructure diseases in different regions, development stages, climate characteristics, etc., and in-depth analysis of the multi-dimensional, identifiable and quantifiable signs of urban infrastructure diseases based on data engines and knowledge engines is carried out to construct a mathematical model reflecting the critical transition from "pre-illness" to failure and a "pre-illness" analysis and judgment system, so as to realize the critical transition characterization of the service status of urban lifelines driven by multi-source data.

[0033] The aforementioned urban lifeline single and cross-system risk assessment involves analyzing the performance degradation patterns and risk evolution characteristics of urban lifeline facilities through data, knowledge, and model engines. This analysis then constructs a cross-system urban lifeline risk evolution network that integrates data aggregation and understanding, knowledge reasoning, and model linkage. This network is driven by the correlation analysis, mining, and integration of multimodal sensory data and risk spatiotemporal big data to identify and deduce the overall risk landscape of urban lifeline systems. Furthermore, based on the combined deduction of infrastructure risk landscapes, with infrastructure risk correlation and coupling analysis as the primary focus, the comprehensive risk of urban lifeline regions within single and cross-systems is assessed using parameters such as coupling probability and coupling-induced disaster consequences.

[0034] The above-mentioned precise decision-making and analysis refers to the association and fusion expression of multi-domain and full-time risk data of urban lifelines based on risk networks, driving the cross-system risk causal reasoning and dynamic reconstruction of graphs of urban lifelines, combined with model call analysis, to achieve dynamic analysis of urban lifeline risk situations and output decision-making recommendations.

[0035] The above-mentioned targeted intelligent early warning refers to the realization of advanced risk perception and targeted early warning based on the single / cross-system risk evolution network of the city lifeline through analysis of the type, scope and degree of disaster impact.

[0036] The above-mentioned precise prevention and control refers to the key issues to be addressed by forming reasonable emergency response and coordinated disposal plans based on data engines, knowledge engines, and model engines, on the basis of risk evolution networks, using complex network algorithms, risk level rankings, occurrence probability rankings and other analysis methods, through the analysis of cross-cutting, weak, and controllable key nodes of risk evolution, and enhancing the scientificity, foresight, and agility of urban lifeline risk management.

[0037] The above-mentioned emergency coordination and linkage refers to the cross-departmental coordinated handling of risk events based on the single / cross-system risk evolution network of the city lifeline, combined with the knowledge engine risk scenario response measure extraction and intelligent matching.

[0038] In addition, examples of knowledge enhancement engine functions and application scenarios under the linkage of data, knowledge, and model engines individually and in combination are shown in the following table:

[0039]

[0040]

[0041]

[0042]

[0043] It should be noted that the general large model mentioned above can adopt open source models such as chat GPT and deepseek R1; the above-mentioned knowledge enhancement engine includes data engine, knowledge engine and model engine; the above-mentioned urban lifeline refers to the key infrastructure system that maintains the basic operation of the city and guarantees residents' lives and production activities, including water supply, drainage, gas, heating, electricity, transportation, communications, bridges, tunnels, integrated pipelines, etc.

[0044] For the mapping association between the above-mentioned multiple risk scenarios and business application scenarios, the following is an example:

[0045]

[0046] In the gas-electricity coupling risk scenario, the model output of "probability of exceeding the limit of methane concentration in cable pipelines" can be directly mapped to the gas company's inspection frequency adjustment, emergency event classification warning and disposal, and pipeline environment maintenance of the power department.

[0047] As a preferred embodiment, the business logic and business functions of the business application scenario are analyzed, including analysis of business logic and analysis of business functions;

[0048] It's important to note that business logic is a chain reaction of "conditions-actions." Its goal is to clarify how a vertical domain big model responds to inputs and produces outputs in specific scenarios. Business functions, on the other hand, provide specific operational capabilities for vertical domain big models, directly serving user needs. Their goal is to clarify "what operations need to be implemented" to support business logic.

[0049] Exemplarily, the analysis of business logic includes the following steps:

[0050] Use swimlane diagrams to clearly identify the nodes involved in different businesses, use decision tables to formalize the mapping relationship between conditions and actions, and then define the response mechanism for data conflicts and system failures;

[0051] Gas Leak Emergency Response: If the gas concentration in the valve well exceeds 5% and there is heavy traffic in the surrounding area, trigger automatic valve closure and alarm, trigger a manual review process, and use the gas diffusion range mechanism model, explosion consequence prediction model, and leak point tracing model to conduct situation analysis. A graded emergency warning is issued, and corresponding emergency response and disposal (e.g., venting, leak point maintenance, etc.) is carried out based on the warning level. If sensor data is abnormal (drift value ±20%), trigger a manual review process.

[0052] The analysis of specific business functions includes the following steps:

[0053] Use use case diagrams to divide functional boundaries, define functional requirements from the user's perspective, and evaluate functional requirement priorities and technical dependencies.

[0054] Similarly, take the emergency response to gas leak as an example:

[0055]

[0056]

[0057] As another preferred embodiment, the data engine is used to collect, clean and fuse multi-source heterogeneous data, and summarize and summarize them to form data standards; as well as perform spatiotemporal alignment, anomaly detection, trend analysis and hidden danger pattern mining on multi-source heterogeneous data.

[0058] It should be noted that the cleaning of multi-source heterogeneous data can be done through distributed stream processing frameworks such as Apache Flink (low latency) and Spark Structured Streaming (high throughput or micro-batch processing); and the method of fusing multi-source heterogeneous data is to disambiguate entities through knowledge graphs and use deep learning models to semantically align text reports with structured data to generate structured output that meets the standards of urban lifeline data.

[0059] Furthermore, spatiotemporal alignment involves statistical analysis of multivariate heterogeneous data in terms of timestamps and spatial coordinates. Anomaly detection primarily detects null values, values ​​that do not conform to a manually set range, and values ​​that differ from previous data by a threshold. Standard deviation and quartile methods can be used for implementation. Trend analysis, on the other hand, involves predicting future trends in data based on historical data analysis. Time series analysis, regression analysis, and machine learning training methods can be used for implementation. Hidden danger pattern mining involves identifying patterns in hidden danger data based on changes in certain data related to the failure status of a facility in historical data. This can be achieved through machine learning algorithms.

[0060] As another preferred embodiment, the knowledge engine is used to extract, express and dynamically update domain knowledge; as well as to perform indicator analysis, rule extraction, association analysis, response generation and similar case recommendation on domain knowledge, and form a knowledge graph to obtain the correlation relationship between risk events.

[0061] It should be noted that domain knowledge extraction primarily utilizes a model combining BERT and BiLSTM-CRF to identify risk entities. Furthermore, rule-based approaches and deep learning models are combined to further extract complex semantic information, such as causal relationships and association rules. Domain knowledge is expressed by constructing an ontology model for the urban lifeline domain, utilizing a Neo4j graph database to store entities and their relationships. Expert experience is formalized as SWRL rules to enhance knowledge reasonability. The dynamic update mechanism for domain knowledge utilizes Kafka to receive new incident reports in real time, triggering the automatic expansion and revision of the knowledge base, thereby enabling the continuous evolution of the knowledge system.

[0062] For indicator analysis and rule extraction from domain knowledge, we can take a gas pipeline network operation and maintenance standard as an example to illustrate how to extract specific operation and maintenance indicators from it. This process can be achieved through semantic parsing and information extraction technology of standard texts, and key performance indicators related to gas pipeline network operation (such as pressure thresholds, inspection cycles, etc.) can be structured and extracted. In terms of rule extraction, risk-related decision rules can be automatically identified and generated from standard texts and accident reports. For example, using a Seq2Seq-based generative model (such as T5 or BART) combined with rule templates, unstructured text can be converted into risk management rules in natural language form, thereby assisting in the construction of a highly interpretable and executable business rule library.

[0063] The association analysis of domain knowledge aims to construct a causal network between risk factors. Graph neural networks (such as GAT) can be used to embed and learn entities and their relationships, thereby exploring potential complex dependencies. Response generation is based on a rule mapping mechanism preset by experts in the field or obtained through knowledge extraction from large models. It matches the identified risk types with the corresponding disposal measures to form preliminary response recommendations. The similar case recommendation module is based on a historical accident case library and uses the BERT model to calculate the semantic similarity between case texts, screening out historical events that are most similar to the current situation, and assisting decision makers in formulating response strategies based on past experience.

[0064] As another preferred embodiment, the model engine is used to construct a model association representation of the risk transfer process; the model association representation refers to the loose coupling and collaborative reasoning between the physical model, the data-driven model and the knowledge element rule model. For example, the scope and depth of urban waterlogging are calculated through the rainstorm and flood analysis model, and the analysis results are used as input parameters of the gas pipeline leakage and diffusion and accident emergency rescue analysis model for model association and quantitative representation of the accident deduction process; the data-driven model refers to a non-mechanistic, semi-quantitative analysis model derived from statistical data and data trend analysis, which is used to fill the gaps in the current physical model / mechanistic model in some areas; the knowledge element rule model refers to the response rules for specific scenarios and parameters based on standards and specifications, emergency plans, management requirements, and work regulations, such as the handling process under different accident conditions and the control measures when the pipeline parameters exceed the threshold.

[0065] For example, loose coupling allows for a gas pipeline jet fire model and a building fire analysis model to be linked by simply determining the jet fire's high-temperature impact range and the distance between buildings. When the jet fire model detects that the high-temperature impact range exceeds a threshold, the thermal radiation data output by the jet fire model is used as input to the building fire model, automatically triggering analysis of the building fire model to analyze a scenario where a gas pipeline jet fire ignites a bridge. Similarly, collaborative reasoning allows for a pipeline leak to produce both jet fire and explosion. By combining these two models, a comprehensive analysis of the consequences of a gas pipeline disaster can be conducted.

[0066] As another preferred embodiment, the data engine, knowledge engine and model engine are jointly driven through a collaborative mechanism.

[0067] It should be noted that the collaborative mechanism includes: updating entity relationships in the knowledge graph through multi-source data, constraining the physical model through causal relationships in the knowledge graph, guiding the construction of data standards through model prediction results, and performing analytical operations through the data engine. Specifically, constraining the physical model can be achieved by adding a knowledge verification layer before reasoning, while guiding the construction of data standards through model prediction results is based on the model's input and output, combined with data processing.

[0068] As another preferred embodiment, the data engine includes multimodal data recognition and extraction, model operation data protocol, data processing and modeling tools. Through multi-source heterogeneous "data collection-data processing-data storage", a standardized data processing system is constructed to form a city lifeline thematic database.

[0069] It should be noted that multimodal data includes but is not limited to existing data access, tables, text, audio, image recognition data, and natural language input data; for example, the BERT model is used to extract key risk words from accident reports, and the CNN model is used to identify potential dangers in camera images (such as pipe cracks, gas leakage flames, etc.). In order to achieve standardized interaction between models, this embodiment uses JSON-LD or Protocol Buffers to define the data format and communication protocol of model input and output. In terms of data processing, KNNImputer is used to fill in missing values ​​in structured data; for outliers in numerical fields, the IQR method is used to filter. The modeling process is orchestrated and managed by Apache Airflow, and the core prediction model is built based on TensorFlow or PyTorch to ensure the efficiency and scalability of the system.

[0070] As another preferred embodiment, the knowledge engine is constructed by building a knowledge base based on standards, accident cases, expert experience, and historical maintenance records. Using a general large model and dynamic knowledge graph technology, it extracts, describes, and associates knowledge elements. Furthermore, through semantic understanding, relationship mining, causal reasoning, and feature analysis of risk events, it constructs a knowledge association network to achieve knowledge element association, risk transmission reasoning, and disposal strategy optimization. Risk transmission reasoning refers to the process of theoretically reasoning about the risk transmission of emergencies based on the knowledge base knowledge graph. Disposal strategy optimization refers to the ability to analyze, match, and optimize corresponding measures for risk events.

[0071] It should be noted that standard specifications refer to the guidance documents, national standards, industry standards, etc. in the technical field being analyzed, and the knowledge element description includes but is not limited to name, definition, attributes, related cases, and response measures; the knowledge element relationship includes but is not limited to causal relationship, hierarchical relationship, dependency relationship, similarity relationship, time-space relationship, and whole-part relationship.

[0072] As another preferred embodiment, the model engine is constructed as follows:

[0073] Based on the mining of risk event correlation relationships, the sorting of infrastructure's "disaster-causing-disaster-response" attributes and accident feature analysis, combined with abstract modeling methods, the "soft reasoning + hard reasoning" method is used to realize single / cross-system risk identification and analysis, transfer logic relationships and implicit association expressions, form a model rule library, and obtain a sequential, quantified and visualized urban lifeline single / cross-system model engine.

[0074] It should be noted that the risk scenarios corresponding to the sorting out of risk events, infrastructure "disaster-causing-disaster-response" attributes, and accident feature analysis include but are not limited to natural disasters, geological disasters, man-made damage, and disasters caused by the facility itself; the abstract modeling method refers to the process of abstracting the characteristics of sudden incidents and accidents and the "disaster-causing-disaster-response" attributes of infrastructure in the risk transfer process, and expressing the key attributes of infrastructure and accident features to cover various types of infrastructure and accident types; the abstraction of infrastructure includes but is not limited to structural and functional classification, such as abstracting buried gas pipelines into gas materials, buried laying methods, and pipeline shapes; the abstraction of accident features includes but is not limited to material, energy, and information classification, such as abstracting fires into flame damage and thermal radiation damage.

[0075] Exemplary:

[0076]

[0077]

[0078] In this embodiment, soft reasoning refers to the process of making empirical and speculative judgments about the evolution of emergency risk based on a knowledge engine, accident cases, expert experience, news reports, and standards. Hard reasoning refers to the process of reasoning about the evolution and path of emergencies based on the analysis of the "disaster-causing and disaster-bearing" mechanism of infrastructure and the mechanism model. The model rule library constructs a model library based on abstract modeling, and model types include mechanism models and empirical models. Mechanism models can be divided into common (general) mechanism models and non-common (non-general) mechanism models based on the abstraction of infrastructure and accident characteristics. Empirical models refer to qualitative, semi-quantitative, and quantitative analysis models used when mechanism model methods are lacking.

[0079] Urban lifeline single / cross-system risk networks are risk networks with infrastructure / events as nodes and risk transfer rules as edges. When the risk transfer rules are met, the event risk transfers from the previous node to the next node. Single-system risk networks refer to emergencies caused by direct external or internal influences on the system. The consequences of the emergency are directly catastrophic to the system, and the risk is limited to the system itself, without analyzing the risks of adjacent or related systems. Examples include gas leaks and jet fires, water pipeline leaks, and methane accumulation and explosions in drainage pipes. Cross-system risk refers to the interconnectedness and dependencies between different systems. When an emergency occurs in one system, the incident spreads from one system to related systems, triggering risks in these systems. For example, a methane explosion in an underground drainage pipe causes vibration damage to surrounding buried gas pipelines, and a gas pipeline leak spreads to adjacent power pipelines and buildings, further triggering power outages and building fires.

[0080] Single / cross-system risk networks can dynamically update the network based on the import of regional basic data and monitoring data; further, they can realize backward scenario deduction of event consequences and forward tracing of event causes of event nodes, fully releasing the value of monitoring data.

[0081] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for constructing a large vertical model, characterized in that: include: Build a vertical field big model framework based on a general big model, with a knowledge enhancement engine driven by data, knowledge, and models from multiple sources as the core, and combined with technology applications; Construct the data engine, knowledge engine and model engine in the vertical field large model knowledge enhancement engine; Based on the mapping association between multiple risk scenarios and business application scenarios in urban lifelines, the business logic and business functions of the application scenarios are analyzed, and through the combination and linkage of general big models and knowledge enhancement engines, an application platform based on the vertical domain big model of urban lifelines is constructed.

2. The method for constructing a large vertical domain model according to claim 1, characterized in that: The data engine is used to collect, clean and fuse multi-source heterogeneous data, and build a data standard system through process-based data processing; as well as perform spatiotemporal alignment, anomaly detection, trend analysis and hidden danger pattern mining on multi-source heterogeneous data.

3. The method for constructing a large vertical domain model according to claim 1, characterized in that: The knowledge engine is used to extract, express and dynamically update domain knowledge; as well as perform indicator analysis, rule extraction, association analysis, response generation and similar case recommendation on domain knowledge, and form a knowledge graph.

4. The method for constructing a large vertical domain model according to claim 1, characterized in that: The model engine is used to construct a model association representation of the risk transfer process; the model association representation refers to the loose coupling and collaborative reasoning between the physical model, the data-driven model and the knowledge meta-rule model.

5. The method for constructing a large vertical domain model according to claim 1, characterized in that: In addition to self-driving functional applications, the data engine, the knowledge engine, and the model engine can also be jointly driven through a collaborative mechanism; The collaborative mechanism includes: updating entity relationships in the knowledge graph through multi-source data, constraining physical models through causal relationships in the knowledge graph, guiding the construction of data standards in the data engine through model prediction results, and analyzing operations through the data engine.

6. The method for constructing a large vertical domain model according to claim 1, characterized in that: The data engine includes data identification and extraction, model operation data protocol, data processing and modeling tools. Through the "data collection-data processing-data storage" of multi-source heterogeneous data, a standardized data processing system is constructed to form a city lifeline thematic database.

7. The method for constructing a large vertical domain model according to claim 1, characterized in that: The construction method of the knowledge engine includes: A knowledge base is built based on standard specifications, accident cases, and historical maintenance records. A general large model and dynamic knowledge graph technology are used to carry out knowledge meta-description and association, and a knowledge association network is constructed through semantic understanding, relationship mining, causal reasoning and feature analysis of risk events.

8. The method for constructing a large vertical domain model according to claim 1, characterized in that: The model engine is constructed in the following manner: Based on the mining of risk event correlation relationships, the analysis of infrastructure's "disaster-causing-disaster-response" attributes and accident characteristics, combined with abstract modeling methods, single / cross-system risk identification and analysis, transfer logic relationships and implicit association expressions are realized in a reasoning manner, forming a model rule library, and obtaining a sequential, quantified, and visualized urban lifeline single / cross-system model engine. Among them, reasoning includes soft reasoning and hard reasoning.

9. The method for constructing a large vertical domain model according to claim 8, characterized in that: The soft reasoning is a process of making empirical and speculative judgments on the evolution of emergency risk based on knowledge base accident cases, news reports, and standard specifications.

10. The method for constructing a large vertical domain model according to claim 8, characterized in that: The hard reasoning is the process of reasoning about the evolution process and path of emergencies based on the analysis of the "disaster-causing-disaster-bearing" mechanism of infrastructure.

Images