Multi-dimensional scene space construction method and device for major engineering project emergencies

Abstract

The invention provides a multi-dimensional scene space construction method and device for major engineering project emergencies, and the method comprises the steps: carrying out the hierarchical analysis of the emergencies from the four dimensions of a primary scene, a secondary scene, a scene object and a scene element through combining a preset major engineering project social stability risk vector database. The method comprises the following steps: generating primary scenes of different stages based on text information, refining step by step to form secondary scenes, identifying scene objects, extracting scene elements, and finally constructing a multi-dimensional scene space. According to the method, on the basis of a pre-constructed social stability risk vector database of the major engineering project, the emergencies which newly occur in the major engineering project are decomposed from four dimensions of'case-scene-object-element 'through a multi-dimensional scene space method, all factors of the emergencies and mutual relations of the factors are comprehensively understood and captured, and the safety of the emergencies is improved. Therefore, comprehensive and accurate assessment of the emergencies is realized, and objectivity and accuracy of analysis and assessment of the emergencies are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of emergency management for engineering projects, and particularly to a method for constructing a multi-dimensional scenario space for emergencies in major engineering projects, a device for constructing a multi-dimensional scenario space for emergencies in major engineering projects, an electronic device, and a computer-readable medium. Background Art

[0002] Traditional risk assessment methods for emergencies in major engineering projects mainly rely on experts' experience and historical data analysis, often being limited to the assessment of a certain specific dimension, and having strong subjectivity and limitations. These methods are not only inefficient but also lack a comprehensive understanding of the interactive effects of various complex factors, and it is difficult to adapt to the increasingly complex and changeable disaster scenarios. Summary of the Invention

[0003] In view of the above problems, the present invention is proposed to provide a method for constructing a multi-dimensional scenario space for emergencies in major engineering projects and a corresponding device for constructing a multi-dimensional scenario space for emergencies in major engineering projects, an electronic device, and a computer-readable medium that can overcome or at least partially solve the above problems.

[0004] The present invention discloses a method for constructing a multi-dimensional scenario space for emergencies in major engineering projects, the method comprising: Obtaining relevant text information of an emergency in a major engineering project, and performing a primary scenario analysis on the relevant text information of the emergency in the major engineering project in combination with a preset social stability risk vector database for major engineering projects to generate primary scenarios at different stages; Performing a secondary scenario analysis on the primary scenarios at different stages in combination with the preset social stability risk vector database for major engineering projects to generate secondary scenarios at different stages; Performing an object analysis based on the primary scenarios at different stages, the secondary scenarios at different stages, and the preset social stability risk vector database for major engineering projects to generate scenario objects at different stages respectively; Performing an element analysis based on the primary scenarios at different stages, the secondary scenarios at different stages, the scenario objects at different stages, and the preset social stability risk vector database for major engineering projects to generate scenario elements at different stages respectively; The multi-dimensional scenario space for emergencies in major engineering projects is composed of the primary scenarios at different stages, the secondary scenarios at different stages, the scenario objects at different stages, and the scenario elements at different stages.

[0005] Optionally, obtain the relevant text information of major engineering project emergencies, and combine it with the preset major engineering project social stability risk vector database to conduct a primary scenario analysis on the relevant text information of major engineering project emergencies, generating primary scenarios at different stages, including: Use the BERT model to convert the relevant text information of major engineering project emergencies into major engineering project emergency semantic vectors, and based on the major engineering project emergency semantic vectors, conduct knowledge search in the preset major engineering project social stability risk vector database to obtain major engineering project social stability risk vectors that match the major engineering project emergency semantic vectors; Based on the major engineering project social stability risk vectors that match the major engineering project emergency semantic vectors and the major engineering project emergency semantic vectors, generate primary scenario prompts, and use the large language model to generate primary scenarios at different stages based on the primary scenario prompts.

[0006] Optionally, combine with the preset major engineering project social stability risk vector database to conduct a secondary scenario analysis on the primary scenarios at different stages, generating secondary scenarios at different stages, including: Use the BERT model to convert the primary scenarios at different stages into primary scenario semantic vectors at different stages, conduct knowledge search in the preset major engineering project social stability risk vector database to obtain major engineering project social stability risk vectors that match the primary scenario semantic vectors at different stages; Based on the primary scenario semantic vectors at different stages and the major engineering project social stability risk vectors that match the primary scenario semantic vectors at different stages, generate secondary scenario prompts at different stages, and use the large language model to generate secondary scenarios at different stages based on the secondary scenario prompts at different stages.

[0007] Optionally, conduct object analysis based on the primary scenarios at different stages, the secondary scenarios at different stages, and the preset major engineering project social stability risk vector database, respectively generating scenario objects at different stages, including: Use the BERT-BiLSTM-CRF composite model to conduct object analysis based on the primary scenarios at different stages, the secondary scenarios at different stages, and the preset major engineering project social stability risk vector database, respectively generating scenario objects at different stages.

[0008] Optionally, use the BERT-BiLSTM-CRF composite model to conduct object analysis based on the primary scenarios at different stages, the secondary scenarios at different stages, and the preset major engineering project social stability risk vector database, respectively generating scenario objects at different stages, including: The BERT model is used to convert the primary scenario and secondary scenario into scenario semantic vectors, and knowledge search is performed in the preset social stability risk vector database of major engineering projects according to the scenario semantic vectors at different stages to obtain the social stability risk vectors of major engineering projects that match the scenario semantic vectors; The BiLSTM model is used to extract the temporal dependence relationship between the scenario semantic vectors and the social stability risk vectors of major engineering projects that match the scenario semantic vectors, output the labeled probability sequence, and the CRF model is used for decoding to generate the label sequence; Based on the label sequence, scenario object prompts at different stages are generated, and the large language model is used to generate scenario objects at different stages based on the scenario object prompts at different stages respectively.

[0009] Optionally, according to the primary scenarios at different stages, secondary scenarios at different stages, scenario objects at different stages, and the preset social stability risk vector database of major engineering projects, element analysis is performed to generate scenario elements at different stages, including: The BERT model is used to convert the primary scenario, secondary scenario, and scenario object into scenario-object semantic vectors, and knowledge search is performed in the preset social stability risk vector database of major engineering projects to obtain the social stability risk vectors of major engineering projects that match the scenario-object semantic vectors; According to the scenario-object semantic vectors and the social stability risk vectors of major engineering projects that match the scenario-object semantic vectors, scenario element prompts at different stages are generated, and the large language model is used to generate scenario elements at different stages based on the scenario element prompts at different stages.

[0010] Optionally, the method further includes: Collect multi-source text data related to the social stability risk of major engineering projects, including social stability risk assessment reports, historical major engineering project emergencies, policy and regulation data, news and social data; Perform data cleaning and preprocessing on the multi-source text data related to the social stability risk of major engineering projects to obtain the social stability risk data set of major engineering projects; Based on the time sequence, perform multi-dimensional scenario space decomposition on each historical major engineering project emergency to obtain multiple scenarios of each event and the information of each scenario, and convert the multiple scenarios of each event and the information of each scenario into semantic vectors and store them in the social stability risk vector database of major engineering projects; the information of each scenario includes scenario objects and scenario elements; Convert the social stability risk assessment report, policy and regulation data, news and social data into semantic vectors and store them in the social stability risk vector database of major engineering projects.

[0011] The present invention also discloses a device for constructing a multi-dimensional scenario space for emergencies in major engineering projects, and the device includes: A primary scenario generation module, configured to obtain relevant text information of emergencies in major engineering projects, and perform primary scenario analysis on the relevant text information of emergencies in major engineering projects by combining a preset social stability risk vector database for major engineering projects, so as to generate primary scenarios at different stages; A secondary scenario generation module, configured to perform secondary scenario analysis on the primary scenarios at different stages by combining a preset social stability risk vector database for major engineering projects, so as to generate secondary scenarios at different stages; A scenario object generation module, configured to perform object analysis according to the primary scenarios at different stages, the secondary scenarios at different stages, and a preset social stability risk vector database for major engineering projects, so as to generate scenario objects at different stages respectively; A scenario element generation module, configured to perform element analysis according to the primary scenarios at different stages, the secondary scenarios at different stages, the scenario objects at different stages, and a preset social stability risk vector database for major engineering projects, so as to generate scenario elements at different stages respectively; A multi-dimensional scenario space composition module, configured to constitute a multi-dimensional scenario space for emergencies in major engineering projects by the primary scenarios at different stages, the secondary scenarios at different stages, the scenario objects at different stages, and the scenario elements at different stages.

[0012] The present invention also discloses an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory complete communication with each other through the communication bus; The memory is used for storing a computer program; When the processor is used to execute the program stored on the memory, it implements the method for constructing a multi-dimensional scenario space for emergencies in major engineering projects as described in the present invention.

[0013] The present invention also discloses one or more computer-readable media, on which instructions are stored, and when executed by one or more processors, cause the processors to execute the method for constructing a multi-dimensional scenario space for emergencies in major engineering projects as described in the present invention.

[0014] The present invention has the following advantages: The method for constructing a multi-dimensional scenario space for emergencies in major engineering projects of the present invention analyzes emergencies hierarchically from four dimensions: primary scenarios, secondary scenarios, scenario objects, and scenario elements by combining a preset database of social stability risk vectors for major engineering projects. First, primary scenarios at different stages are generated based on text information, then gradually refined to form secondary scenarios, then scenario objects are identified and scenario elements are extracted, and finally a multi-dimensional scenario space is constructed. Based on the pre-constructed database of social stability risk vectors for major engineering projects, for newly-occurring emergencies in major engineering projects, the multi-dimensional scenario space method is used to decompose the emergencies from four dimensions of "case - scenario - object - element", comprehensively understand and capture all factors of the emergencies and their interrelationships, realize a comprehensive and accurate assessment of the emergencies, and improve the objectivity and accuracy of the analysis and assessment of emergencies. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a flowchart of the steps of a method for constructing a multi-dimensional scenario space for emergencies in major engineering projects provided by an embodiment of the present invention; Figure 2 is an architecture diagram of constructing a multi-dimensional scenario space by a BERT - BiLSTM - CRF - large language model provided by an embodiment of the present invention; Figure 3 is a structural block diagram of a device for constructing a multi-dimensional scenario space for emergencies in major engineering projects provided by an embodiment of the present invention; Figure 4 is a block diagram of an electronic device provided by an embodiment of the present invention; Figure 5 is a schematic diagram of a computer-readable medium provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] Refer to Figure 1 , which shows a flowchart of the steps of a method for constructing a multi-dimensional scenario space for emergencies in major engineering projects provided by an embodiment of the present invention, and specifically may include the following steps: Step 101, obtain relevant text information of an emergency in a major engineering project, and perform primary scenario analysis on the relevant text information of the emergency in the major engineering project by combining a preset database of social stability risk vectors for major engineering projects to generate primary scenarios at different stages; Step 102, combine the preset database of social stability risk vectors for major engineering projects to perform secondary scenario analysis on the primary scenarios at different stages to generate secondary scenarios at different stages; Step 103, performing object analysis according to the primary scenarios at different stages, the secondary scenarios at different stages, and the preset social stability risk vector database of major engineering projects, and generating scenario objects at different stages respectively; Step 104, performing factor analysis according to the primary scenarios at different stages, the secondary scenarios at different stages, the scenario objects at different stages, and the preset social stability risk vector database of major engineering projects, and generating scenario factors at different stages respectively; Step 105, a multi-dimensional scenario space of major engineering project emergencies is formed by primary scenarios at different stages, secondary scenarios at different stages, scenario objects at different stages, and scenario elements at different stages.

[0018] Based on the case scenario decomposition principle of the multi-dimensional scenario space method, the present invention uses the LangChain framework to realize intelligent decomposition (such as Figure 2 As shown in Figure 3, information is retrieved from a vector database through retrieval-augmented generation technology to enhance the generation capability of a large language model.

[0019] The automatic construction process of the multidimensional scenario space in the present invention can be divided into four main steps: scenario generation, object extraction, element generation and scenario optimization. First, the scenario recognition module uses a large language model to analyze the input text data and identify the key components of the scenario. Then, the object extraction module accurately identifies the core entities in the scenario, such as people involved in the disaster, through the BERT-BiLSTM-CRF model. Subsequently, the element extraction module generates element prompts through vector retrieval based on the scenario and object information, and generates key elements related to the current scenario. Finally, the scenario optimization module uses retrieval enhancement generation technology to further adjust and optimize the generated scenario model to ensure that the final scenario space can reflect the real scenario changes. In the present invention, major engineering project emergencies may include disasters, accidents, mass incidents, etc.

[0020] The BERT-BiLSTM-CRF-Large Language Model multi-dimensional scenario space construction model proposed in this invention, combined with advanced natural language processing (NLP) technology, can automatically decompose new cases into four main components: "scenario", "sub-scenario", "object" and "element". During the construction process, the model can fully mine the scenario information in the text data and automatically decompose scenarios and objects according to the needs of different stages. This method not only improves the efficiency and accuracy of scenario analysis, but also can handle more complex and changeable disaster scenarios, thereby providing more accurate support for emergency management and decision-making.

[0021] In an alternative embodiment of the present invention, relevant text information of major engineering project emergencies is obtained, and primary scenario analysis is performed on the relevant text information of major engineering project emergencies in combination with a preset social stability risk vector database of major engineering projects, generating primary scenarios at different stages, including: The BERT model is used to convert the relevant text information of major engineering project emergencies into semantic vectors of major engineering project emergencies, and knowledge search is performed in the preset social stability risk vector database of major engineering projects according to the semantic vectors of major engineering project emergencies, obtaining social stability risk vectors of major engineering projects that match the semantic vectors of major engineering project emergencies; According to the social stability risk vectors of major engineering projects that match the semantic vectors of major engineering project emergencies and the semantic vectors of major engineering project emergencies, a primary scenario prompt is generated, and a large language model is used to generate primary scenarios at different stages based on the primary scenario prompt.

[0022] In an alternative embodiment of the present invention, secondary scenario analysis is performed on the primary scenarios at different stages in combination with a preset social stability risk vector database of major engineering projects, generating secondary scenarios at different stages, including: The BERT model is used to convert the primary scenarios at different stages into semantic vectors of the primary scenarios at different stages, and knowledge search is performed in the preset social stability risk vector database of major engineering projects, obtaining social stability risk vectors of major engineering projects that match the semantic vectors of the primary scenarios at different stages; According to the semantic vectors of the primary scenarios at different stages and the social stability risk vectors of major engineering projects that match the semantic vectors of the primary scenarios at different stages, secondary scenario prompts at different stages are generated, and a large language model is used to generate secondary scenarios at different stages based on the secondary scenario prompts at different stages.

[0023] The generation process of primary scenarios and secondary scenarios at different stages is as follows: First, the first step of scenario decomposition is to convert text data into a vector representation that can be processed. In the present invention, the BERT model is adopted. By inputting a text sequence, its semantic features are automatically extracted and converted into a multi-dimensional vector representation. BERT processes context information through a Transformer structure and can take into account bidirectional information of the left and right contexts to generate context-aware word vectors. This feature makes BERT perform excellently in understanding semantic and implicit relationships in complex texts.

[0024] The multi-dimensional vectors generated by the BERT model can not only reflect various types of information in the text, but also capture the spatio-temporal dependencies in the scenario, providing a rich semantic basis for subsequent scenario analysis and reasoning. For dynamic fields such as disaster management, the scenario vectors generated by BERT can flexibly respond to scenario changes at different stages, thus providing accurate semantic support for scenario decomposition.

[0025] After the scenario vectors are generated, these scenario vectors are used as input retrieval conditions. Through vector retrieval technology, similar scenarios or relevant knowledge can be quickly found from massive data, and relevant scenario data can be automatically retrieved. This retrieval method based on the vector space is more efficient and accurate than the traditional text matching-based method.

[0026] The present invention also combines retrieval-augmented generation technology to further enhance the generation ability of the large language model. The retrieval-augmented generation technology first combines the retrieval results in the vector database previously with the generation ability of the large language model, and infers and optimizes the output through the generated scenario prompts. This process not only accelerates the speed of scenario inference, but also improves the accuracy of scenario analysis. In disaster scenario modeling, the retrieval-augmented generation technology can also generate secondary scenarios S' according to historical data and the current scenario S, and generate more accurate scenario speculations for each stage.

[0027] After completing scenario vectorization and knowledge retrieval, the large language model (such as GLM4) optimizes the generation of scenarios and secondary scenarios at different time stages based on the retrieved data to generate scenarios and secondary scenarios at different stages. The scenarios at different time stages not only reflect the evolution of the disaster event in time, but also can effectively display the key decision points required in the disaster management process.

[0028] In an optional embodiment of the present invention, object analysis is performed according to the primary scenarios at different stages, the secondary scenarios at different stages, and the preset social stability risk vector database of major engineering projects, and scenario objects at different stages are respectively generated, including: The BERT-BiLSTM-CRF composite model is used to perform object analysis according to the primary scenarios at different stages, the secondary scenarios at different stages, and the preset social stability risk vector database of major engineering projects, and scenario objects at different stages are respectively generated.

[0029] In an optional embodiment of the present invention, the BERT-BiLSTM-CRF composite model is used to perform object analysis according to the primary scenarios at different stages, the secondary scenarios at different stages, and the preset social stability risk vector database of major engineering projects, and scenario objects at different stages are respectively generated, including: The BERT model is used to convert the primary scenario and secondary scenario into scenario semantic vectors, and based on the scenario semantic vectors at different stages, knowledge search is performed in the preset social stability risk vector database of major engineering projects to obtain the social stability risk vectors of major engineering projects that match the scenario semantic vectors; The BiLSTM model is used to extract the temporal dependence relationship between the scenario semantic vectors and the social stability risk vectors of major engineering projects that match the scenario semantic vectors, output the labeled probability sequence, and the CRF model is used for decoding to generate the label sequence; Based on the label sequence, scenario object prompts at different stages are generated, and the large language model is used to generate scenario objects at different stages based on the scenario object prompts at different stages respectively.

[0030] On the basis of scenario decomposition, another important task is object extraction. Object extraction aims to identify the core entities related to the scenario from complex texts, such as disaster participants, etc. To improve the extraction accuracy, the present invention adopts a BERT-BiLSTM-CRF composite model, which combines the semantic understanding ability of the BERT model, the bidirectional long-distance dependence learning ability of BiLSTM, and the label dependence modeling ability of CRF, and can effectively identify and extract key information in the text.

[0031] The process of scenario object extraction is as follows: (1) The BERT model extracts semantic features: As a powerful pre-trained language model, the BERT model plays a crucial role in scenario object extraction. The innovation of the BERT model lies in its ability to generate more accurate word vectors by considering both left and right context information through a bidirectional Transformer architecture. In the object extraction task, BERT first tokenizes the input text and extracts the context information of each word through a deep neural network, and then generates its semantic representation.

[0032] Through the BERT model, phrase-level semantics, syntactic structures, and high-level semantic information in the text can be captured, which provides very rich features for subsequent object extraction. Taking disaster management as an example, BERT can accurately identify the core entities in disaster events, such as "Environmental Protection Bureau", "media", "rescue team", etc., providing an important basis for subsequent object recognition and relationship analysis.

[0033] (2) Combining BiLSTM and CRF to extract key information labels: Although BERT can generate high-quality word vectors, there are still certain limitations in solely relying on the output of BERT when dealing with complex annotation tasks. Especially when dealing with long sequence data, BERT may not be able to effectively capture the dependencies between labels. Therefore, the present invention further introduces BiLSTM and CRF modules to enhance the model's context information processing ability. The BiLSTM model is an improved LSTM model that processes the input sequence bidirectionally, taking into account both the forward and backward information of the sequence, thereby improving the modeling ability for long-distance dependencies. In the object extraction task, BiLSTM can effectively capture the context information in the sequence. Especially when dealing with the dependencies between adjacent entities, BiLSTM shows better performance. To further optimize the dependencies between labels, the present invention introduces the Conditional Random Field (CRF) model. CRF can learn the optimal label sequence by modeling the transition relationships between adjacent labels. Based on BiLSTM, the CRF module models the dependencies between labels to ensure that the model can select the most contextually appropriate label sequence during the annotation process, thereby improving the accuracy of object extraction.

[0034] (3) Mathematical model representation: This process can be expressed using a mathematical model as follows: The model first passes the input sequence to BiLSTM for processing, obtaining the output score matrix of BiLSTM , The size of is , where is the number of words, represents the -th word's -th label score. Therefore, for each input sequence , the score function for the predicted sequence is:

[0035] Among them, is the transition score matrix, is the score for the label transitioning to the label . The size of , from which the conditional probability generated by the predicted sequence is expressed as:

[0036] represents the true sequence, Denote all possible sequences. Take the logarithm of both sides of the above formula to obtain the likelihood function of the predicted sequence:

[0037] The output sequence with the maximum score after decoding is the optimal predicted sequence:

[0038] In an alternative embodiment of the present invention, according to the primary scenarios at different stages, secondary scenarios at different stages, scenario objects at different stages, and a preset vector database of social stability risks for major engineering projects, factor analysis is performed to generate scenario factors at different stages, including: Use the BERT model to convert the primary scenario, secondary scenario, and scenario object into scenario-object semantic vectors, and perform knowledge search in the preset vector database of social stability risks for major engineering projects to obtain the social stability risk vectors of major engineering projects that match the scenario-object semantic vectors; According to the scenario-object semantic vectors and the social stability risk vectors of major engineering projects that match the scenario-object semantic vectors, generate scenario factor prompts at different stages, and use a large language model to generate scenario factors at different stages based on the scenario factor prompts at different stages.

[0039] The generation process of the scenario factors is as follows; The last step in scenario decomposition is scenario factor extraction, whose purpose is to further extract key factors from the scenario model and combine them with the relevant knowledge base to ensure that the extracted factors can accurately reflect the internal logic and requirements of the scenario. In the present invention, scenario factor extraction mainly relies on the previous scenario modeling results, including the scenarios and secondary scenarios generated by the large language model, and the object O extracted by the BERT-BiLSTM-CRF composite model. On this basis, the present invention performs vector retrieval through the vector database and combines the retrieved relevant knowledge to further generate factor prompts so that the subsequent model can generate more accurate scenario factors F.

[0040] In the process of scenario factor extraction, it is first necessary to use vector retrieval technology to search for knowledge related to the current scenario in the vector database. By converting the scenario data (including scenario S, secondary scenario S', and object O) into a multi-dimensional vector representation, similar scenario patterns and relevant knowledge can be accurately matched in the vector space. During the retrieval process, not only similar scenarios are searched for, but also specific factors closely related to the target scenario are focused on. This process is completed by calculating the similarity between vectors (such as using cosine similarity), so that valuable information related to the current scenario can be accurately extracted.

[0041] As the retrieval results are obtained, the model will automatically generate relevant element prompts. These element prompts include key factors that may affect the scenario development. For example, in the disaster management scenario, they may include "disaster time", "affected area", "post-disaster recovery ability", etc. By integrating these element prompts into the subsequent large language model generation process, it can ensure that the generated element F is more in line with the real scenario under different scenarios and highly matches the characteristics of scenario S.

[0042] In an alternative embodiment of the present invention, the method further includes: Collect multi-source text data related to the social stability risks of major engineering projects, including social stability risk assessment reports, historical major engineering project emergencies, policy and regulation data, news and social data; Perform data cleaning and preprocessing on the multi-source text data related to the social stability risks of major engineering projects to obtain a social stability risk dataset for major engineering projects; Perform multi-dimensional scenario space decomposition on each historical major engineering project emergency based on the time sequence to obtain multiple scenarios and the information of each scenario for each event, and convert the multiple scenarios and the information of each scenario for each event into semantic vectors and store them in the social stability risk vector database for major engineering projects; the information of each scenario includes scenario objects and scenario elements; Convert the social stability risk assessment reports, policy and regulation data, news and social data into semantic vectors and store them in the social stability risk vector database for major engineering projects.

[0043] Construct a scenario case library, that is, collect historical major engineering project emergencies, take major engineering projects as the research object, and focus on the mass incidents caused by the under-construction and completed engineering projects in the "National Catalogue of Major Engineering Projects" issued by the state. For this purpose, the present invention has collected reports and research papers on mass incidents caused by major engineering projects published on multiple news websites, newspapers and academic journals from 2007 to 2024. After screening and removing duplicate and less relevant invalid data, about 2000 pieces of data have been finally collected and sorted out, covering multiple fields and different types of cases.

[0044] To ensure the representativeness of the cases, the selection of cases follows the following criteria: First, the event must occur in the country and be a mass incident caused by a major engineering project; Second, the case information should be sufficient to fully display the occurrence and development process of the event; Third, select those cases with greater social impact and wide public attention. This screening process has undergone strict review to ensure the high quality and diversity of the cases.

[0045] Finally, the present invention constructs a scenario case library covering more than 200 major engineering project mass incidents. The composition of this case library includes the following main parts: ① Case name: Named according to the location where the incident occurred and the name of the relevant engineering project, such as "Incident of PX Project in City M"; ② Case occurrence time: Refers to the specific time when the mass incident was triggered during the construction and operation of the engineering project; ③ Case development process: The key processes and important information of each incident are sorted out in detail according to the time sequence; ④ Number of participants: According to the social stability risk assessment level standard, the number of participants in the social stability incidents involved in the case is divided into three categories: "less than 20 people", "20 - 200 people", and "more than 200 people".

[0046] In order to improve the generation ability of the large language model in the disaster multi - dimensional scenario space model, based on the scenario case library, the present invention widely collects multi - dimensional data sets such as social stability risk assessment reports, historical accident reports, and relevant policy texts as the original materials for learning data. These data sets not only contain case information about major engineering project mass incidents, but also cover aspects such as policy environment, historical accidents, and social stability assessment, providing diversified background information for the large language model and further enriching the model's learning ability.

[0047] For these multi - dimensional data sets, first, according to the principle of the multi - dimensional scenario space, the present invention splits the cases in the scenario database part of the multi - dimensional data set into multiple phased scenarios in chronological order to ensure that all key nodes and dynamic changes in the event development can be captured. This step, through the division of the time dimension, helps to comprehensively understand the hierarchical relationship and evolution process between scenarios and provides more detailed scenario information. Next, the present invention further splits multiple different phased scenarios of the case, and finally forms a multi - level structure. The optimization of this structure lies in that different levels of multiple different phased scenarios can show different degrees of detail of information. For example, the first - level scenario level can display the macro - framework of the scenario case, while the second - level and third - level levels can delve into more specific scenario objects and detail elements. This hierarchical structure makes the description of the case scenario data no longer static, but a data containing dynamic and hierarchical connections.

[0048] On this basis, the present invention uses the BERT embedding model to process each case scenario data text and convert it into a high - dimensional semantic vector. This vector can not only accurately represent the deep semantic information of the text, but also capture the subtle differences in the text, greatly enhancing the effect of semantic similarity calculation and text retrieval. Specifically, each detail in the scenario is quantified through this embedding method, thus providing a more accurate basis for the large language model when generating secondary scenarios, objects, and elements.

[0049] For the processing of other parts of the dataset, the present invention adopts the same embedding technology to directly convert the remaining text data into semantic vectors. This approach not only improves the depth of understanding of texts such as social stability risk assessment reports, historical accident reports, and relevant policies, but also enhances the accuracy of generating secondary scenarios, objects, and elements through semantic vector representation. The generated semantic vectors will be stored in a vector database and constructed into a vector dataset. Through this vector database, the large language model can efficiently search and query similar vectors, greatly improving the efficiency of semantic retrieval. Additionally, the vector database can not only handle large-scale data but also has good real-time response capabilities and scalability, enabling the present invention to flexibly respond to the update and processing requirements of new case scenarios. This design provides strong technical support for the continuous improvement of the disaster scenario model and can be dynamically adjusted at any time according to new scenarios or cases.

[0050] In addition, during the further fine-tuning process of the large language model, the present invention introduces a method based on semantic similarity matching. Based on the learning of a large number of social stability assessment reports, the model's risk assessment and scoring capabilities for new cases are optimized. This process enables the model to more accurately give assessment results through similarity reasoning and improves its generation capabilities in complex scenarios.

[0051] As an example of the present invention, a case analysis of the "PX project incident in City M": To verify the effectiveness of the model, the present invention selects the "PX project incident in City M" as a case and conducts an experiment on the automatic construction of a multi-dimensional scenario space, using GLM4 as the large language model. Specifically, in this experiment, the relevant text data of this incident is input into the model, and multiple components including the scenario S, secondary scenario S', object O, and element F at different stages are automatically disassembled. During the construction process, the model can identify the main scenario of this incident, such as "environmental pollution accident", and extract relevant secondary scenarios from it, such as "government response measures" or "public opinion guidance", etc.

[0052] Through vector retrieval and generating element prompts, the model further optimizes the key elements in the event scenario. For example, for the theme of the "PX project incident in City M", the model automatically identifies elements such as "project location" and "project progress", and generates corresponding scenario descriptions for each element. These automatically generated scenario elements not only help to more comprehensively understand the internal relationships of this incident but also can provide valuable information for subsequent decision-making support.

[0053] By analyzing the automatic construction results of the "PX project incident in M City", it can be found that the multi-dimensional scenario space construction method proposed by the present invention demonstrates significant advantages when dealing with complex scenarios and multi-stage events. First, the model can automatically extract each dimension in the scenario, ensuring the comprehensiveness and accuracy of the scenario. Second, by introducing the BERT-BiLSTM-CRF model, the model can effectively handle various label dependencies in long texts, thereby improving the accuracy of scenario object extraction. Finally, by combining the large language model and the vector data set, the model can automatically generate relevant scenario elements according to the changes in the scenario, ensuring that the generated scenario space can conform to the actual situation and adapt to different scenario requirements.

[0054] The present invention has the following advantages: 1. The present invention uses the multi-dimensional scenario space method to conduct case analysis and risk identification of emergency events, and by combining natural language processing technologies such as BERT, BiLSTM, and CRF and the large language model, it improves the efficiency of emergency event analysis and assessment, and also reduces the possibility of human intervention, thereby improving the objectivity and accuracy of disaster event analysis and assessment; 2. Through the combination of the large language model and natural language processing technologies, it can accurately learn the multi-dimensional information characteristics of different disaster scenarios based on historical cases, providing a more scientific and reliable basis for disaster identification and analysis; 3. The present invention can update the scenario case library and vector database based on newly occurred emergency events, and incorporate the new disaster event scenarios and knowledge into the disaster multi-dimensional scenario space construction model in a timely manner, enabling the present invention to flexibly respond and maintain high-efficiency disaster event analysis capabilities when facing the constantly changing disaster event environment.

[0055] It should be noted that for the method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the embodiments of the present invention are not limited by the described action sequences, because according to the embodiments of the present invention, certain steps can be carried out in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily essential for the embodiments of the present invention.

[0056] Referring to Figure 3 , a structural block diagram of a multi-dimensional scenario space construction device for major engineering project emergency events provided in an embodiment of the present invention is shown, which may specifically include the following modules: The primary scenario generation module 301 is used to obtain the relevant text information of major engineering project emergency events, and perform primary scenario analysis on the relevant text information of major engineering project emergency events in combination with a preset social stability risk vector database of major engineering projects to generate primary scenarios at different stages; The secondary scenario generation module 302 is configured to perform secondary scenario analysis on the primary scenarios at different stages in combination with a preset social stability risk vector database for major engineering projects, and generate secondary scenarios at different stages; The scenario object generation module 303 is configured to perform object analysis based on the primary scenarios at different stages, the secondary scenarios at different stages, and the preset social stability risk vector database for major engineering projects, and respectively generate scenario objects at different stages; The scenario element generation module 304 is configured to perform element analysis based on the primary scenarios at different stages, the secondary scenarios at different stages, the scenario objects at different stages, and the preset social stability risk vector database for major engineering projects, and respectively generate scenario elements at different stages; The multi-dimensional scenario space composition module 305 is configured to constitute a multi-dimensional scenario space for emergencies of major engineering projects from the primary scenarios at different stages, the secondary scenarios at different stages, the scenario objects at different stages, and the scenario elements at different stages.

[0057] For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple. For related parts, please refer to the partial description of the method embodiment.

[0058] In addition, the embodiment of the present invention further provides an electronic device, as Figure 4 shown, including a processor 401, a communication interface 402, a memory 403, and a communication bus 404. Among them, the processor 401, the communication interface 402, and the memory 403 complete mutual communication through the communication bus 404. The memory 403 is used to store a computer program; The processor 401 is configured to implement the multi-dimensional scenario space construction method for emergencies of major engineering projects as described in the above embodiments when executing the program stored on the memory 403.

[0059] The communication bus mentioned in the above terminal may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of simplicity, only a thick line is shown in the figure, but it does not mean that there is only one bus or one type of bus.

[0060] The communication interface is used for communication between the above terminal and other devices.

[0061] The memory may include a Random Access Memory (RAM), or may also include a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located far from the aforementioned processor.

[0062] The aforementioned processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0063] As Figure 5 shown, in another embodiment provided by the present invention, there is also provided a computer-readable storage medium 501, in which instructions are stored. When it runs on a computer, it enables the computer to execute the method for constructing a multi-dimensional scenario space for emergencies in major engineering projects described in the above embodiments.

[0064] In another embodiment provided by the present invention, there is also provided a computer program product containing instructions. When it runs on a computer, it enables the computer to execute the method for constructing a multi-dimensional scenario space for emergencies in major engineering projects described in the above embodiments.

[0065] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (such as coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more integrated available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk (SSD)).

[0066] It should be noted that in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including", or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or device that includes a series of elements includes not only those elements but also other elements that are not explicitly listed, or also includes elements that are inherent to such process, method, article, or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article, or device that includes the element.

[0067] Each embodiment in this specification is described in a related manner. The same or similar parts among the embodiments can be referred to each other, and the differences between each embodiment and other embodiments are emphasized. In particular, for the device embodiments, since they are basically similar to the method embodiments, they are described relatively simply, and the relevant parts can be referred to the partial description of the method embodiments.

[0068] The above are only the preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention are all included within the protection scope of the present invention.

Claims

1. A method for constructing a multi-dimensional scenario space for emergencies in major engineering projects, characterized in that, The method includes: Obtain the relevant text information of major engineering project emergencies, and perform primary scenario analysis on the relevant text information of major engineering project emergencies in combination with the preset social stability risk vector database of major engineering projects to generate primary scenarios at different stages; Combined with the preset social stability risk vector database of major engineering projects, perform secondary scenario analysis on the primary scenarios at different stages to generate secondary scenarios at different stages; According to the primary scenarios at different stages, the secondary scenarios at different stages, and the preset social stability risk vector database of major engineering projects, perform object analysis to generate scenario objects at different stages respectively; According to the primary scenarios at different stages, the secondary scenarios at different stages, the scenario objects at different stages, and the preset social stability risk vector database of major engineering projects, perform element analysis to generate scenario elements at different stages respectively; The multi-dimensional scenario space of major engineering project emergencies is composed of the primary scenarios at different stages, the secondary scenarios at different stages, the scenario objects at different stages, and the scenario elements at different stages.

2. The method according to claim 1, wherein Obtain the relevant text information of major engineering project emergencies, and perform primary scenario analysis on the relevant text information of major engineering project emergencies in combination with the preset social stability risk vector database of major engineering projects to generate primary scenarios at different stages, including: Use the BERT model to convert the relevant text information of major engineering project emergencies into semantic vectors of major engineering project emergencies, and perform knowledge search in the preset social stability risk vector database of major engineering projects according to the semantic vectors of major engineering project emergencies to obtain social stability risk vectors of major engineering projects that match the semantic vectors of major engineering project emergencies; According to the social stability risk vectors of major engineering projects that match the semantic vectors of major engineering project emergencies and the semantic vectors of major engineering project emergencies, generate primary scenario prompts, and use a large language model to generate primary scenarios at different stages based on the primary scenario prompts.

3. The method according to claim 1, characterized in that, Combined with the preset social stability risk vector database of major engineering projects, perform secondary scenario analysis on the primary scenarios at different stages to generate secondary scenarios at different stages, including: Use the BERT model to convert the primary scenarios at different stages into semantic vectors of the primary scenarios at different stages, and perform knowledge search in the preset social stability risk vector database of major engineering projects to obtain social stability risk vectors of major engineering projects that match the semantic vectors of the primary scenarios at different stages; According to the semantic vectors of the primary scenarios at different stages and the social stability risk vectors of major engineering projects that match the semantic vectors of the primary scenarios at different stages, generate secondary scenario prompts at different stages, and use a large language model to generate secondary scenarios at different stages based on the secondary scenario prompts at different stages.

4. The method according to claim 1, wherein According to the primary scenarios at different stages, the secondary scenarios at different stages, and the preset social stability risk vector database of major engineering projects, perform object analysis to generate scenario objects at different stages respectively, including: Using the BERT-BiLSTM-CRF composite model, object analysis is carried out according to the primary scenarios at different stages, the secondary scenarios at different stages, and the preset social stability risk vector database of major engineering projects, and scenario objects at different stages are generated respectively.

5. The method according to claim 4, wherein Using the BERT-BiLSTM-CRF composite model, object analysis is carried out according to the primary scenarios at different stages, the secondary scenarios at different stages, and the preset social stability risk vector database of major engineering projects, and scenario objects at different stages are generated respectively, including: Using the BERT model to convert the primary scenario and the secondary scenario into scenario semantic vectors, and according to the scenario semantic vectors at different stages, knowledge search is carried out in the preset social stability risk vector database of major engineering projects to obtain the social stability risk vectors of major engineering projects that match the scenario semantic vectors; Using the BiLSTM model to extract the temporal dependence relationship between the scenario semantic vectors and the social stability risk vectors of major engineering projects that match the scenario semantic vectors, output the labeled probability sequence, and use the CRF model for decoding to generate the label sequence; Generating scenario object prompts at different stages based on the label sequence, and using the large language model to generate scenario objects at different stages respectively based on the scenario object prompts at different stages.

6. The method according to claim 1, wherein According to the primary scenarios at different stages, the secondary scenarios at different stages, the scenario objects at different stages, and the preset social stability risk vector database of major engineering projects, element analysis is carried out, and scenario elements at different stages are generated respectively, including: Using the BERT model to convert the primary scenario, the secondary scenario, and the scenario object into scenario-object semantic vectors, and carrying out knowledge search in the preset social stability risk vector database of major engineering projects to obtain the social stability risk vectors of major engineering projects that match the scenario-object semantic vectors; Generating scenario element prompts at different stages according to the scenario-object semantic vectors and the social stability risk vectors of major engineering projects that match the scenario-object semantic vectors, and using the large language model to generate scenario elements at different stages based on the scenario element prompts at different stages.

7. The method according to claim 1, wherein The method further includes: Collecting multi-source text data related to the social stability risk of major engineering projects, including social stability risk assessment reports, historical major engineering project emergencies, policy and regulation data, news and social data; Performing data cleaning and preprocessing on the multi-source text data related to the social stability risk of major engineering projects to obtain a social stability risk dataset of major engineering projects; Performing multi-dimensional scenario space decomposition on each historical major engineering project emergency based on the time sequence to obtain multiple scenarios of each event and the information of each scenario, and converting the multiple scenarios of each event and the information of each scenario into semantic vectors and storing them in the social stability risk vector database of major engineering projects; the information of each scenario includes scenario objects and scenario elements; Converting the social stability risk assessment report, policy and regulation data, news and social data into semantic vectors and storing them in the social stability risk vector database of major engineering projects.

8. A device for constructing a multi-dimensional scenario space for emergencies in major engineering projects, characterized in that, The device includes: The primary scenario generation module is used to obtain the relevant text information of major engineering project emergencies, and perform primary scenario analysis on the relevant text information of major engineering project emergencies in combination with the preset social stability risk vector database of major engineering projects to generate primary scenarios at different stages; The secondary scenario generation module is used to perform secondary scenario analysis on the primary scenarios at different stages in combination with the preset social stability risk vector database of major engineering projects to generate secondary scenarios at different stages; The scenario object generation module is used to perform object analysis according to the primary scenarios at different stages, the secondary scenarios at different stages, and the preset social stability risk vector database of major engineering projects, and respectively generate scenario objects at different stages; The scenario element generation module is used to perform element analysis according to the primary scenarios at different stages, the secondary scenarios at different stages, the scenario objects at different stages, and the preset social stability risk vector database of major engineering projects, and respectively generate scenario elements at different stages; The multi-dimensional scenario space composition module is used to constitute the multi-dimensional scenario space of major engineering project emergencies by the primary scenarios at different stages, the secondary scenarios at different stages, the scenario objects at different stages, and the scenario elements at different stages.

9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete mutual communication through the communication bus; The memory is used to store computer programs; When the processor is used to execute the program stored on the memory, it realizes the multi-dimensional scenario space construction method for major engineering project emergencies as described in any one of claims 1-7.

10. One or more computer-readable media, characterized in that, Instructions are stored thereon, and when executed by one or more processors, the processor is caused to execute the multi-dimensional scenario space construction method for major engineering project emergencies as described in any one of claims 1-7.

Images