An intelligent query method and system for emergency command and dispatch based on large-scale models and cognitive memory

Abstract

This invention discloses an intelligent query method and system for emergency command and dispatch based on a large model combined with cognitive memory. The method includes: receiving user query requests, initializing the session, processing multi-turn dialogue content using natural language processing technology and saving complete session records; acquiring complete session records and analyzing the session content, extracting key emergency command information as long-term memory points and performing vector transformation and storage; establishing corresponding indexes for memory points and storing them in an emergency command information vector database, and updating or merging existing memories according to preset rules; performing semantic analysis on user input questions and matching memory point vectors, extracting memory fragments and sending them to a large model; continuously collecting user operation data and generating user profiles for each user. This invention combines cognitive memory with a large model-based emergency command and dispatch system, reducing computational resources, improving long-term memory capacity, interpretability, and model adaptability and generalization ability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of emergency command and dispatch management, and more particularly to an emergency command and dispatch intelligent query method and system based on a large model combined with thinking memory. BACKGROUND

[0002] With the continuous advancement of emergency management system construction, and the frequent occurrence of natural disasters, accidents and other emergencies, emergency command and dispatch, as the core disposal link, is facing increasingly complex challenges. Traditional emergency command and dispatch systems show obvious limitations in dealing with complex constraints and dynamic changes in the environment, and cannot adapt to changes in disaster conditions, task adjustments and sudden dangers in real time, resulting in low efficiency and serious waste of emergency resources.

[0003] In emergency rescue, public safety, urban emergency support and other scenarios, the complexity of emergency command and dispatch management is further improved. These scenarios usually involve multi-rescue force, multi-task, multi-constraint condition collaborative scheduling problems, and need to consider multiple factors such as rescue vehicle type, equipment load limit, rescue time window, route safety optimization, resource cost control, etc.

[0004] In addition, with the development of intelligent emergency system, emergency command and dispatch management is gradually evolving towards intelligence and automation; however, existing intelligent dispatch systems are mostly based on single models or algorithms, lacking comprehensive understanding and flexible response capabilities for complex emergency scenarios, especially in the face of escalating emergencies such as disaster expansion, secondary disasters, and rescue route interruptions, traditional systems often cannot make effective dispatch adjustments in time, resulting in decreased rescue efficiency and increased loss of life and property.

[0005] However, existing emergency command and dispatch systems based on large models still have the following limitations: large consumption of computing resources: the inference process of large models requires a large amount of computing resources and time, making it difficult to meet the needs of real-time emergency command and dispatch; lack of long-term memory ability: large models usually do not have memory function, and cannot remember historical rescue dispatch experience and user (emergency command personnel) preferences, resulting in the need to start from scratch every time; lack of explainability: the decision-making process of large models is difficult to explain, which is an important problem in emergency command enterprise-level applications that require auditing and compliance; limited adaptability and generalization ability: existing systems are usually designed for specific emergency scenarios, making it difficult to adapt to the diverse needs of different regions and different types of emergencies.

[0006] For example, Chinese patent CN120494567A proposes an emergency command and dispatch method and platform for industrial parks based on a large model. The aim is to enhance the response speed and decision-making accuracy of emergency response in industrial parks through the empowerment of large models. The core objective is to optimize the overall efficiency of hazard handling, personnel evacuation, and on-site command. However, this technical solution is designed for closed scenarios in industrial parks and does not consider the dispatching needs of open, large-scale emergency scenarios such as natural disasters and cross-regional accidents, resulting in insufficient generalization ability. It lacks a memory module, making it impossible to retain historical emergency response experience, commanders' decision preferences, and handling standards for similar accidents. Each emergency response requires restarting model reasoning, reducing decision-making efficiency. It also lacks a natural language interaction interface for commanders, failing to support their query needs for historical handling data and real-time dispatch parameters. Furthermore, it does not consider collaborative dispatching scenarios involving multiple rescue forces and cross-departmental resources, only adapting to emergency response within industrial parks and failing to meet the needs of multi-entity collaborative response in complex emergencies.

[0007] Therefore, how to reduce computational resources, long-term memory capacity, and interpretability in emergency command and dispatch based on large models, while improving adaptability and generalization ability to meet the diverse needs of different regions and different types of emergencies, is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0008] In view of the above problems, the present invention aims to provide an intelligent query method and system for emergency command and dispatch based on a large model combined with cognitive memory to overcome or at least partially solve the above problems.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] An intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory includes: S1. Receive user query requests, initialize the session, process multi-turn dialogue content using natural language processing technology, and save the complete session record; S2. Obtain the complete conversation record and analyze the conversation content, extract key emergency command information as long-term memory points, and perform vector transformation and storage; S3. Establish corresponding indexes for memory points, store memory points and corresponding indexes in the emergency command information vector database, and update or merge existing memories according to preset rules; S4. Perform semantic analysis on the user input question and match the memory point vector. Extract memory fragments based on the semantic matching results and send them to the large model. S5. Continuously collect user operation data and use data analysis algorithms to generate user profiles for each user based on the collected data.

[0011] Preferably, step S1 includes the following: S11. When a user starts a new session with the emergency command and dispatch intelligent data agent, create a new session record space; S12. In each round of interaction in the conversation, the user's input questions and the agent's answers are recorded in real time. At the same time, the attention mechanism is used to model the context, and natural language processing technology is used to analyze and understand the referential relationships in the dialogue. S13. When the session ends, the short memory module saves a complete record of the session, providing a data basis for memory retrieval.

[0012] Preferably, step S2 includes the following: S21. After the session ends, retrieve the saved complete session record, conduct a comprehensive analysis of the session content, and identify the key information points; S22. From the analyzed conversation content, extract key emergency command information that needs to be included in long-term memory as memory points. Memory points are information that is closely related to emergency command and dispatch and has long-term reference value. S23. Convert the extracted memory points into vector form and store them in the corresponding emergency command information vector database.

[0013] Preferably, the specific content of step S4 is as follows: S41. Each time a user initiates a new dialogue request, semantic analysis is performed on the question entered by the user, the question is converted into a semantic vector, and similarity calculation is performed with the memory point vector in the emergency command information vector database; S42. Based on the semantic matching results, extract memory fragments with similarity higher than a preset threshold from the vector database. The memory fragments contain historical experience and knowledge related to the problem. S43. Pass the extracted memory fragments to the large model to assist the large model in generating answers.

[0014] Preferably, the intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory further includes: the large model receiving and transmitting relevant memory fragments, generated user profiles, and user-input questions, performing in-depth analysis and reasoning, generating answers, and presenting them to the user.

[0015] An intelligent query system for emergency command and dispatch based on a large model combined with cognitive memory, and based on the aforementioned intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory, includes: a short memory module based on a conversation mechanism, a memory extraction module, a long memory module, a memory reading module, an interactive interface module, a user profile module, and a large model core module; The interactive interface module is used to receive user query requests and present the results processed by the core module of the large model to the user. The short memory module is used to initialize the session based on the user's query request, process the content of multiple rounds of dialogue using natural language processing technology, and save the complete session record; The memory extraction module is used to acquire complete conversation records and analyze conversation content, extract key emergency command information as long-term memory points, and perform vector transformation and storage. The long memory module is used to create corresponding indexes for memory points, store memory points and corresponding indexes in the emergency command information vector database, and update or merge existing memories according to preset rules. The memory retrieval module is used to perform semantic analysis on the user input question and match the memory point vector. Based on the semantic matching results, it extracts memory fragments and sends them to the core module of the large model. The user profiling module is used to continuously collect user operation data and use data analysis algorithms to generate a user profile for each user based on the collected data.

[0016] Preferably, an independent long memory module is built for each user based on the user account, and the memory information of the corresponding emergency command information vector database is dynamically updated through continuous interaction with the user.

[0017] Preferably, the emergency command and dispatch intelligent query system based on a large model combined with cognitive memory further includes a large model core module, which is used to receive and transmit relevant memory fragments, generate user profiles and user-input questions, perform in-depth analysis and reasoning, generate answers and present them to the user.

[0018] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory.

[0019] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the aforementioned intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory.

[0020] As can be seen from the above technical solution, compared with the prior art, the present invention discloses an intelligent query method and system for emergency command and dispatch based on a large model combined with cognitive memory, which has the following beneficial effects: In terms of contextual semantic understanding, the short memory module based on the conversation mechanism can effectively handle the referential relationship in the dialogue, accurately eliminate ambiguity, and enable the intelligent questioning agent to always accurately understand the user's intention in the process of multiple rounds of interaction with the user, avoid wrong responses caused by semantic misunderstanding, and greatly improve the user interaction experience and emergency command and communication efficiency. The memory extraction module extracts key information from conversations and converts it into long-term memory, enriching the agent's emergency command knowledge base. When faced with new disaster response challenges, the memory retrieval module quickly extracts relevant memory fragments through semantic matching to assist the core modules of the large model. This not only improves the accuracy of answers but also allows the agent to provide more comprehensive and in-depth solutions based on past rescue experience, providing strong decision support for emergency command and dispatch. In an emergency command and dispatch environment, the number of users (emergency command units / personnel) changes dynamically. By building an independent memory module for each user account and continuously updating it, the system can provide customized services based on personalized memories, accurately adapt to the unique needs of different users, help improve user satisfaction, enhance user stickiness to the emergency command and dispatch platform, and promote the platform's continuous development and promotion. By updating the memory module in real time, the system can adapt to user changes in a more flexible and efficient manner, reducing operating costs and technical implementation difficulties. At the same time, the close collaboration among the various modules of the system optimizes the information processing flow, improves the system's response speed, and enables rapid response to user emergency dispatch requests, further enhancing the efficiency of emergency command and dispatch and saving valuable time for handling emergencies. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of an intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory, provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of an intelligent query system for emergency command and dispatch based on a large model combined with cognitive memory, provided in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1 This invention discloses an intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory, such as... Figure 1 ,include: S1. Receive user query requests, initialize the session, process multi-turn dialogue content using natural language processing technology, and save the complete session record; S2. Obtain the complete conversation record and analyze the conversation content, extract key emergency command information as long-term memory points, and perform vector transformation and storage; S3. Establish corresponding indexes for memory points, store memory points and corresponding indexes in the emergency command information vector database, and update or merge existing memories according to preset rules; S4. Perform semantic analysis on the user input question and match the memory point vector. Extract memory fragments based on the semantic matching results and send them to the large model. S5. Continuously collect user operation data and use data analysis algorithms to generate user profiles for each user based on the collected data.

[0025] To further implement the above technical solution, the specific content of step S1 includes: S11. When a user starts a new session with the emergency command and dispatch intelligent data agent, create a new session record space; S12. In each round of interaction in the conversation, the user's input questions and the agent's answers are recorded in real time. At the same time, the attention mechanism is used to model the context, and natural language processing technology is used to analyze and understand the referential relationships in the dialogue. In this embodiment, user queries are recorded. q i With system response a i Forming a dialogue sequence D = {(q 1 , a 1 ), (q 2 , a 2 ), …, (q n , a n )} To achieve referential resolution, an attention mechanism is used to model the context as follows:

[0026] For example, in a dialogue involving emergency command and dispatch, if a user asks, "How is the rescue team's handling of the situation today?", the agent replies, "The rescue team completed the transfer of 3 trapped people today, covering a distance of 80 kilometers," and records this round of interaction. Simultaneously, using natural language processing technology, the agent analyzes and understands the referential relationships in the dialogue. For instance, if a user later asks, "Does this handling efficiency meet the requirements?", based on the previously recorded dialogue, it clarifies that "this handling efficiency" refers to the handling efficiency of the rescue team mentioned in the previous round that day. In this way, the agent achieves accurate understanding of referential relationships in the dialogue and precisely eliminates ambiguity, ensuring that it accurately understands the user's intent throughout the entire conversation. S13. When the session ends, the short memory module saves a complete record of the session, providing a data basis for memory retrieval.

[0027] To further implement the above technical solution, step S2 includes the following: S21. After the session ends, retrieve the saved complete session record, conduct a comprehensive analysis of the session content, and identify the key information points; for example, in a session about the rescue team's handling of the situation, identify that the handling situation mentioned by the user should include key information such as the number of trapped people transferred, the mileage traveled, the mission completion time, and the consumption of emergency supplies. S22. Extract key emergency command information that needs to be included in long-term memory from the analyzed conversation content as memory points. Memory points are information that is closely related to emergency command and dispatch and has long-term reference value, such as multi-dimensional definitions of the handling of different types of rescue missions, and the correlation between specific disasters and the configuration of rescue forces. S23. Convert the extracted memory points into vector form and store them in the corresponding emergency command information vector database; In this embodiment, each memory point is converted into a vector representation through a pre-trained language model:

[0028] in, m j For memory points, BERT is a pre-trained language model.

[0029] In step S3, corresponding indexes for memory points are established to ensure efficient storage and retrieval of memory points. For example, memory points are classified and stored according to their categories, such as rescue force configuration, disaster response, and emergency material management, and corresponding indexes are established to enable quick location and query. Nearest Neighbor (ANN) indexes are also established, such as using HNSW or Faiss index structures, to support efficient retrieval.

[0030] To further implement the above technical solution, the specific content of step S4 is as follows: S41. Each time a user initiates a new dialogue request, semantic analysis is performed on the user's input question, converting the question into a semantic vector, and similarity calculation is performed between the vector and the memory point vector in the emergency command information vector database:

[0031] in, q Query semantic vectors for users. v i This refers to the memory point vector in the emergency command information vector database; S42. Based on the semantic matching results, extract memory fragments with similarity higher than a preset threshold from the vector database. The memory fragments contain historical experience and knowledge related to the problem, such as discussions in the conversation about the statistical method of rescue mission handling time, reference standards for handling time under different disaster scenarios, etc. S43. Pass the extracted memory fragments to the large model to assist the large model in generating answers.

[0032] In step S5, the user's (emergency command personnel / emergency command unit's) operation data on the emergency command and dispatch platform includes, but is not limited to, question history, attention to disaster alarm information, and selection of rescue force dispatch strategies. For example, it records that user A frequently queries the alarm information of rescue force configuration for a certain type of disaster, while user B pays more attention to the impact of rescue route planning on the efficiency of response. The user profile generated for each user includes information such as the user's preferences for various types of disaster alarms, the frequency of use of emergency command functions, and the expected answer methods. Furthermore, as user operation data is continuously updated, the user profile is also adjusted in real time. For example, if user C recently begins to pay attention to the impact of emergency material reserves on continuous rescue, the user profile module will update the relevant information in a timely manner so that the intelligent question data agent can provide answers that are more in line with the user's needs based on the new profile.

[0033] To further implement the above technical solution, an intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory also includes: the large model receiving and transmitting relevant memory fragments, generating user profiles and user-input questions, performing in-depth analysis and reasoning, generating answers and presenting them to the user.

[0034] Example 2 An intelligent query system for emergency command and dispatch based on a large model combined with cognitive memory, such as Figure 2Based on an intelligent query method for emergency command and dispatch that combines a large model with cognitive memory, it includes: a short memory module based on a conversation mechanism, a memory extraction module, a long memory module, a memory reading module, an interactive interface module, a user profile module, and a core module of the large model; The interactive interface module is used to receive user query requests and present the results processed by the core module of the large model to the user. The short memory module is used to initialize the session based on the user's query request, process the content of multiple rounds of dialogue using natural language processing technology, and save the complete session record; The memory extraction module is used to acquire complete conversation records and analyze conversation content, extract key emergency command information as long-term memory points, and perform vector transformation and storage. The long memory module is used to create corresponding indexes for memory points, store memory points and corresponding indexes in the emergency command information vector database, and update or merge existing memories according to preset rules. The memory retrieval module is used to perform semantic analysis on the user input question and match the memory point vector. Based on the semantic matching results, it extracts memory fragments and sends them to the core module of the large model. The user profiling module is used to continuously collect user operation data and use data analysis algorithms to generate a user profile for each user based on the collected data.

[0035] To further implement the above technical solution, an independent long memory module is built for each user based on the user account, and the memory information of the corresponding emergency command information vector database is dynamically updated through continuous interaction with the user.

[0036] To further implement the above technical solution, an emergency command and dispatch intelligent query system based on a large model combined with cognitive memory also includes a large model core module, which is used to receive and transmit relevant memory fragments, generate user profiles and user-input questions, perform in-depth analysis and reasoning, generate answers and present them to the user.

[0037] In this embodiment, the system is tested using an emergency command and dispatch query scenario for urban flooding disasters as an example. A southern city is experiencing continuous heavy rainfall that causes flooding. The emergency command center needs to dispatch multiple rescue teams in real time to carry out personnel transfer and material distribution. Commanders frequently query questions such as the progress of a rescue team's response, the priority standards for rescue in different waterlogged areas, and historical experience in handling similar flooding situations. The system is required to respond quickly, provide accurate answers, and understand the referential relationships in multi-turn dialogues. Test configuration: Ten emergency command personnel were selected as test users, each initiating 20 rounds of related queries, including 10 rounds of multi-turn dialogue involving referential references; the test group used the method of this invention, while the control group used the existing large-scale emergency query system without memory mechanism; the core test indicators included: multi-turn dialogue referential resolution accuracy, query response time and answer accuracy.

[0038] Testing process: The test group activated the short memory module, long memory module, and user profile module, and initialized general knowledge of emergency response to urban flooding disasters, such as the correspondence between water depth and rescue priority. Ten users initiated queries in turn, and the system recorded the dialogue content in real time and extracted key memory points, such as the rescue priority of waterlogging in old urban areas and the dispatching and allocation standards of assault boats, which were stored in the long memory module. The control group relied only on real-time reasoning of the large model and did not retain historical dialogue and user preference information.

[0039] Results data and algorithm details: The test group achieved a 96.8% accuracy rate in resolving multi-turn dialogue referencing, compared to 72.3% in the control group; the average query response time was 0.8 seconds in the test group, compared to 2.1 seconds in the control group; the answer accuracy rate (matching emergency response standards and historical experience) was 94.5% in the test group, compared to 81.7% in the control group; the core algorithm process data is as follows: In the referential resolution stage: the short memory module uses an attention mechanism to model the dialogue context. For referential queries such as "Does this handling efficiency meet the requirements?", the attention weight of the target referential (the handling efficiency of the previous rescue team) is calculated to be 0.92, which is significantly higher than that of other irrelevant entities (the weights are all below 0.15), ensuring accurate resolution of ambiguity. Memory matching stage: After the user query "differences in the allocation of rescue forces between the old city and the new city in similar historical flooding situations" is encoded into a semantic vector by the Encoder, it is compared with the memory point vectors in the long memory module's vector database using cosine similarity calculation. Three highly similar memory fragments are selected, with similarities of 0.91, 0.88, and 0.85, respectively, providing effective support for the large model's inference. Ultimately, the accuracy of this query reached 98.2%. In contrast, the control group, lacking a memory mechanism, relied solely on real-time inference from the large model, resulting in a lack of access to similar historical experience and an accuracy of only 65.1%. Vector conversion efficiency: The memory extraction module uses the BERT model to convert memory points such as "priority standards for flood rescue in the old city" into 768-dimensional vectors. The conversion time for a single memory point is only 0.03 seconds, ensuring the real-time update of the long memory module.

[0040] In this embodiment, the system is implemented using a cross-regional emergency command and dispatch query scenario for earthquake disasters in mountainous areas as an example. A 4.8-magnitude earthquake occurs in a mountainous area, triggering a landslide. Rescue forces need to be dispatched across three administrative regions, including fire, medical, and engineering rescue teams. Emergency commanders need to frequently query cross-scenario and multi-dimensional issues such as "cross-regional rescue team assembly time standards," "safety assessment basis for landslide area rescue routes," and "priority of transfer for different types of injured persons." Moreover, different commanders have different focuses. For example, medical commanders focus on the transfer of injured persons, while engineering commanders focus on route clearing.

[0041] Test configuration: Fifteen emergency command personnel from different professional backgrounds were selected: 5 medical personnel, 5 engineering personnel, and 5 comprehensive dispatch personnel. Each person initiated 25 rounds of queries, including 15 rounds of personalized request queries. The test group used the system of this invention to build a personalized profile for each user and update it dynamically. The control group used an existing general emergency query system without personalized adaptation function. The core test indicators included: personalized request matching rate, cross-scenario query accuracy rate, and user satisfaction score (1-10 points).

[0042] Testing process: The test group initialized general knowledge of cross-regional dispatching for earthquake disasters and simultaneously enabled the user profile data collection function to record the query preferences of command personnel of different specialties. For example, medical personnel repeatedly searched for "precautions for the transfer of spinal injury patients", while engineering personnel focused on "landslide stability assessment indicators". The system extracted personalized needs and key standards for cross-regional dispatching, such as the assembly time limits of rescue teams in different regions, into memory points and stored them in the long memory module through the memory extraction module. The control group used a uniform response template without distinguishing user specialties and preferences.

[0043] Results data and algorithm details: The personalized demand matching rate of the test group reached 95.3%, while that of the control group was 68.4%; the cross-scenario query accuracy rate of the test group reached 93.8%, while that of the control group was 76.9%; the average user satisfaction score of the test group was 9.2 points, while that of the control group was 6.5 points; the core algorithm process and data are as follows: Personalized profiling and memory fusion: For personalized queries by medical command personnel regarding "special casualty transport route planning," the user profiling module outputs a user preference vector, emphasizing dimensions such as spinal injury transport and safety of mountain routes. This vector is then weighted and fused with relevant memory vectors extracted by the memory retrieval module, including historical mountain casualty transport cases and landslide area route assessment standards. The fusion formula is f=α. q+β u+γ m, where α=0.3 is the query weight, β=0.4 is the user profile weight, and γ=0.3 is the memory weight. After fusion, the semantic focus of the vector on special wounded persons and mountain routes is improved by 40%. Memory retrieval efficiency: The long memory module adopts the HNSW index structure, and the retrieval time for memory points related to "cross-regional rescue team assembly time standard" is only 0.05 seconds, which is 82% more efficient than traditional brute-force retrieval, and the retrieval recall rate reaches 97%, ensuring that no key memory information is missed; Cross-scenario adaptation algorithm: For scheduling queries across three administrative regions, the system adjusts the semantic vector generation weights through a domain-adaptive coding strategy, increasing the weight of cross-regional related feature dimensions to 0.65, effectively adapting to multi-regional collaborative scenarios, and ultimately achieving a cross-scenario query accuracy of 93.8%. In contrast, the control group, without an adaptation algorithm and using a unified coding strategy, had a cross-scenario feature weight of only 0.32 and an accuracy of only 76.9%. Among these, for personalized queries of "special casualty transfer route planning" by medical command personnel, the test group combined fused vectors and historical transfer experience to achieve a matching rate of 97.6%, while the control group could only provide a general transfer process, failing to adapt to mountainous scenarios and the special needs of casualties, with a matching rate of only 58.3%.

[0044] Example 3 In the second embodiment of the intelligent query system for emergency command and dispatch based on a large model and combined with cognitive memory, the memory module is integrated and optimized. The short memory module and the memory extraction module are constructed into a conversation memory unit, while retaining the core functions of the long memory module and the memory reading module. The conversation memory unit completes the dialogue record and preliminary extraction of key information simultaneously during the conversation, without waiting for the end of the conversation to start the extraction process separately. After each round of input by the user, the conversation memory unit automatically identifies the key emergency command information, such as vehicle number, task type, mileage threshold and other key vehicle management data, generates temporary memory points in real time and stores them temporarily. After the conversation ends, only the temporary memory points need to be verified and vectorized before they can be stored in the long memory module. Furthermore, by integrating the user profile module into the core module of the large model in the second embodiment system, the large model can dynamically capture users' query habits during the dialogue process, such as preferred vehicle data dimensions and commonly used query phrases. This eliminates the need to maintain a separate profile database, reduces the frequency of data transmission between modules, and can shorten the session response time by about 30% in vehicle dispatching scenarios for small and medium-sized enterprises and institutions, while also reducing the hardware resource requirements for system deployment.

[0045] Example 4 In the second embodiment of the intelligent query system for emergency command and dispatch based on a large model and combined with cognitive memory, the memory extraction module is replaced with a knowledge graph construction unit, and a new graph retrieval module is added to work in collaboration with the long memory module. After the session ends, the knowledge graph construction unit constructs the extracted key emergency command and dispatch information into triplets, such as "Rescue Team A - Transferred 3 people today", based on the preset ontology in the field of emergency command and dispatch, such as rescue team entities, disaster task entities, attribute relationships, etc., and stores them in the knowledge graph library.

[0046] When processing new user queries, the memory retrieval module first performs entity association queries in the knowledge graph through the graph retrieval module to obtain structured knowledge directly related to the question. For example, when a user asks "Is the response efficiency of rescue team A normal?", the historical response efficiency standard value of the team can be directly retrieved. This structured knowledge is then used as context input into the core module of the large model. This solution is suitable for large emergency command centers with complex emergency data associations and can improve the accuracy of cross-session information association queries. Especially when dealing with issues related to the coordinated dispatch of multiple rescue forces, the response accuracy can be improved by about 25% compared to the original solution.

[0047] Example 5 A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory.

[0048] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements an intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory.

[0049] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0050] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory, characterized in that, include: S1. Receive user query requests, initialize the session, process multi-turn dialogue content using natural language processing technology, and save the complete session record; S2. Obtain the complete conversation record and analyze the conversation content, extract key emergency command information as long-term memory points, and perform vector transformation and storage; S3. Establish corresponding indexes for memory points, store memory points and corresponding indexes in the emergency command information vector database, and update or merge existing memories according to preset rules; S4. Perform semantic analysis on the user input question and match the memory point vector. Extract memory fragments based on the semantic matching results and send them to the large model. S5. Continuously collect user operation data and use data analysis algorithms to generate user profiles for each user based on the collected data.

2. The intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory as described in claim 1, characterized in that, The specific content of step S1 includes: S11. When a user starts a new session with the emergency command and dispatch intelligent data agent, create a new session record space; S12. In each round of interaction in the conversation, the user's input questions and the agent's answers are recorded in real time. At the same time, the attention mechanism is used to model the context, and natural language processing technology is used to analyze and understand the referential relationships in the dialogue. S13. When the session ends, the short memory module saves a complete record of the session, providing a data basis for memory retrieval.

3. The intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory as described in claim 1, characterized in that, The specific content of step S2 includes: S21. After the session ends, retrieve the saved complete session record, conduct a comprehensive analysis of the session content, and identify the key information points; S22. From the analyzed conversation content, extract key emergency command information that needs to be included in long-term memory as memory points. Memory points are information that is closely related to emergency command and dispatch and has long-term reference value. S23. Convert the extracted memory points into vector form and store them in the corresponding emergency command information vector database.

4. The intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory as described in claim 1, characterized in that, The specific content of step S4 is as follows: S41. Each time a user initiates a new dialogue request, semantic analysis is performed on the question entered by the user, the question is converted into a semantic vector, and similarity calculation is performed with the memory point vector in the emergency command information vector database; S42. Based on the semantic matching results, extract memory fragments with similarity higher than a preset threshold from the vector database. The memory fragments contain historical experience and knowledge related to the problem. S43. Pass the extracted memory fragments to the large model to assist the large model in generating answers.

5. The intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory as described in claim 1, characterized in that, Also includes: The large model receives relevant memory fragments, generated user profiles, and user-input questions, performs in-depth analysis and reasoning, generates answers, and presents them to the user.

6. An intelligent query system for emergency command and dispatch based on a large model combined with cognitive memory, characterized in that, An intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory, as described in any one of claims 1-5, includes: a short memory module based on a conversation mechanism, a memory extraction module, a long memory module, a memory reading module, an interactive interface module, a user profile module, and a large model core module. The interactive interface module is used to receive user query requests and present the results processed by the core module of the large model to the user. The short memory module is used to initialize the session based on the user's query request, process the content of multiple rounds of dialogue using natural language processing technology, and save the complete session record; The memory extraction module is used to acquire complete conversation records and analyze conversation content, extract key emergency command information as long-term memory points, and perform vector transformation and storage. The long memory module is used to create corresponding indexes for memory points, store memory points and corresponding indexes in the emergency command information vector database, and update or merge existing memories according to preset rules. The memory retrieval module is used to perform semantic analysis on the user input question and match the memory point vector. Based on the semantic matching results, it extracts memory fragments and sends them to the core module of the large model. The user profiling module is used to continuously collect user operation data and use data analysis algorithms to generate a user profile for each user based on the collected data.

7. The emergency command and dispatch intelligent query system based on a large model combined with cognitive memory as described in claim 6, characterized in that, Each user is assigned an independent long-memory module based on their user account. The memory information of the corresponding emergency command information vector database is dynamically updated through continuous interaction with the user.

8. The emergency command and dispatch intelligent query system based on a large model combined with cognitive memory as described in claim 6, characterized in that, It also includes a large model core module, which receives and transmits relevant memory fragments, generated user profiles, and user-input questions, performs in-depth analysis and reasoning, generates answers, and presents them to the user.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements an intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory, as described in any one of claims 1 to 5.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements an intelligent query method for emergency command and dispatch based on a large model combined with cognitive memory, as described in any one of claims 1 to 5.

Images