Information scheduling system and method based on large language model

Abstract

The invention provides an information scheduling system and method based on a large language model. The system comprises a camera, a multi-mode large model, a scheduling terminal, a language recognition model, a cue word manager, the large language model, an API interface, a database, a POI library and an equipment information library. Compared with the prior art, the system has the natural language understanding ability, and a user can interact with the system through a daily language. Moreover, the system can carry out rapid analysis processing based on voice information, text information and road condition monitoring, and carries out cross-library joint query and information integration through an API interface on the basis of responding to a complex user intention, thereby obtaining a more accurate result in time.

Description

Technical Field

[0001] The present invention belongs to the technical field of traffic information processing, and in particular relates to an information dispatching system and method based on a large language model. Background Art

[0002] The advancement of the digital transformation of urban transportation requires relevant departments to accumulate a large amount of traffic information. This traffic information covers multiple dimensions such as real-time traffic flow, public transportation system operation status, and road condition monitoring images. Due to the high heterogeneity and complexity of this information, the technology of the relevant information processing platform to query the corresponding matching information in the database based on the actual information obtained has not yet been perfected. It often requires a lot of manpower to analyze and make decisions on traffic information. This process is slow, the correlation judgment is not in-depth enough, the decision-making support is insufficient, and to a certain extent, it limits the efficiency of traffic information management and the emergency response rate.

[0003] In order to overcome the above problems, the prior art proposes an intelligent traffic management system, which mainly applies POI positioning, information query and equipment monitoring and scheduling technologies. These technologies usually rely on geographic information systems (GIS), database management systems and manual information analysis. Specifically, in terms of POI positioning, the prior art uses the spatial analysis capabilities of GIS, combined with various geographic markers and user behavior information, to determine specific points of interest. In terms of information query, the prior art is usually based on keyword search engines to query relevant information by matching user queries and records in the database. In terms of monitoring camera information scheduling, it mainly relies on preset monitoring rules, and obtains and controls the angle and focal length of the camera through manual operation to monitor specific areas.

[0004] The above-mentioned prior art has the following technical defects:

[0005] 1) Existing technologies usually rely on static information and predefined queries, and lack the ability to respond to complex information queries and dynamic information in real time.

[0006] 2) There are many manual steps involved, which is inefficient and prone to errors.

[0007] 3) Traditional information query methods are usually unable to accurately understand the user's true intentions, resulting in the returned results may not fully match the user's needs. Summary of the invention

[0008] The present invention provides an information scheduling system and method based on a large language model, which are used to solve the problems existing in the prior art.

[0009] In a first aspect, to achieve the above-mentioned purpose, the present invention provides an information scheduling system based on a large language model, which comprises:

[0010] A camera, used to capture images of target road conditions;

[0011] The multimodal large model is used to parse the target road condition image according to the preset label prompt words, obtain the camera label data, and write the camera label data into the device information library;

[0012] The dispatch terminal is used to obtain the voice information and the first text data converted from the text query information, and to display the query answer;

[0013] A language recognition model, used for converting the speech information into second text data;

[0014] A prompt word manager, used for extracting text prompt words according to the text in the first text data and the second text data;

[0015] A large language model is used to train based on text prompts and output corresponding query instructions; the large language model is also used to perform deep training on query results to generate query answers; and

[0016] The API interface has at least three interface channels;

[0017] Among them, the API interface has three interface channels, one of which is connected to the database, another is connected to the POI (Point of Interest) library, and another is connected to the device information library. The API interface is used to perform information query in the database, POI library and device information library according to the query instructions to obtain the query results.

[0018] Through the above technical solution, the system of the present invention has the ability to understand natural language, and users can interact with the system through daily language. In addition, the system can also quickly parse and process based on voice information, text information and road condition monitoring, and perform cross-database joint query and information integration through API interfaces based on complex user intentions, so as to obtain more accurate results in a timely manner. In addition, the system of the present invention requires very few manual operation steps, has a high degree of automation, and is more efficient.

[0019] Preferably, the prompt word manager includes at least a related information query module, a POI scheduling module and a camera precision scheduling module;

[0020] Preferably, the API interface is provided with an associated information query interface channel matching the associated information query module, a POI scheduling interface channel matching the POI scheduling module, and a camera precision scheduling interface channel matching the camera precision scheduling module; the associated information query interface channel is communicatively connected to the database; the POI scheduling interface channel is communicatively connected to the POI library; and the camera precision scheduling interface channel is communicatively connected to the device information library.

[0021] In a second aspect, in order to achieve the above-mentioned purpose, the present invention further provides an information scheduling method based on a large language model, which is used in conjunction with the above-mentioned information scheduling system based on a large language model. The information scheduling method of the present invention comprises:

[0022] S10: Acquire voice information, first text data, and camera tag data obtained by parsing based on a multimodal large model;

[0023] S20: using a speech recognition model to convert the speech information into second text data;

[0024] S30: extracting text prompt words according to the text in the first text data and the second text data using a prompt word manager;

[0025] S40: placing the text prompt words into a large language model for deep training to obtain a query instruction;

[0026] Among them, the query instruction is obtained by converting after understanding the text;

[0027] S50: After the camera tag data is written into the device information database, based on the query instruction, the database, POI database and device information database are queried through the API interface respectively to obtain the database query result, POI query result and device information query result respectively;

[0028] S60: reversely placing the database query results, POI query results, and device information query results into a large language model for deep training to generate query answers;

[0029] S70: Forward the query answer to the dispatch terminal.

[0030] Preferably, the camera tag data obtained by parsing based on the multimodal large model in step S10 includes:

[0031] Use the camera to obtain and intercept the target road condition image through the Rtsp stream;

[0032] Set the label prompt word of the target road condition image;

[0033] The target road condition image and label prompt words are placed into the multimodal large model for analysis to obtain the camera label data;

[0034] Write the camera tag data to the device information library.

[0035] Preferably, step S50 includes:

[0036] Based on the query instruction, the corresponding context-related data or information in the database is queried through the API interface, and the context-related data or information is used as the database query result.

[0037] Preferably, the context-related data or information includes at least one or more of environmental information, population information, enterprise information, land use data, BIM data, water and electricity information, and building distribution information.

[0038] Preferably, step S50 includes:

[0039] Based on the query instruction, the corresponding POI address information and map information in the POI library are queried through the API interface, and the POI address information and map information are used as the POI query result.

[0040] Preferably, step S50 includes:

[0041] Based on the query instruction, the corresponding camera angle information in the device information library is queried through the API interface, and the camera angle information is used as the device information query result.

[0042] Preferably, the instruction type of the query instruction is a JSON instruction.

[0043] Preferably, the large language model can adopt GPT-4, Wenxinyiyan, Zhipu GLM, Kimi, Qwen or llama model.

[0044] In a third aspect, to achieve the above-mentioned purpose, the present invention also provides a computer device, comprising an internal memory, a processor, a non-volatile storage medium, and a computer program stored in the non-volatile storage medium and executable on the processor, wherein the processor implements the above-mentioned information scheduling method based on a large language model when executing the computer program.

[0045] Compared with the prior art, the present invention has at least the following advantages:

[0046] 1) Compared with the prior art, the system of the present invention has the ability to understand natural language, and users can interact with the system through daily language. In addition, the system can also quickly parse and process based on voice information, text information and road condition monitoring, and perform cross-database joint query and information integration through API interfaces based on complex user intentions, so as to obtain more accurate results in a timely manner. In addition, the system of the present invention requires very few manual operation steps, has a high degree of automation, and is more efficient.

[0047] 2) Compared with the prior art, the method of the present invention belongs to an information scheduling algorithm based on a large language model, which can quickly provide key information required for decision-making and support more accurate decision-making through advanced data analysis. This not only improves the efficiency of decision-making, but also improves the quality of decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 It is a structural diagram of an information scheduling system based on a large language model in Embodiment 1 of the present invention.

[0050] Figure 2 This is a flow chart of an information scheduling method based on a large language model in Embodiment 2 of the present invention.

[0051] Figure 3 It is a schematic diagram of the structure of the computer device in the third embodiment of the present invention. DETAILED DESCRIPTION

[0052] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] In the description of the present invention, it should be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside", etc. indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they cannot be understood as limitations on the present invention.

[0054] Embodiment 1

[0055] like Figure 1 As shown, this embodiment discloses an information scheduling system based on a large language model, which includes a camera, a multimodal large model, a scheduling terminal, a language recognition model, a prompt word manager, a large language model, a database, a POI library, a device information library and an API interface with three interface channels.

[0056] In this embodiment, the dispatch terminal is a large command screen, which is an interactive all-in-one machine with a liquid crystal display function.

[0057] Specifically, the camera is used to capture the target road condition image. The multimodal large model is used to parse the target road condition image and the preset label prompt word to obtain the camera label data, and write the camera label data into the device information library. The dispatch terminal is used to obtain voice information and the first text data converted from the text query information. The language recognition model is used to convert the voice information into the second text data. The first text data is generally converted from the text query information entered by the user on the command screen. The prompt word manager is used to extract the text prompt word according to the text in the first text data and the second text data. The large language model is used to train according to the text prompt word and output the corresponding query instruction to the API interface. The API interface has three interface channels, one of which is connected to the database, another of which is connected to the POI library, and another of which is connected to the device information library. The three channels are used to query relevant information or data from the database, POI library and device information library according to the above query instructions, and output the query results back to the large language model for deep training to generate query answers. Finally, the query answer is displayed from the dispatch terminal.

[0058] It should be noted that the above-mentioned multimodal large models can use closed source large models such as GPT-4o and Zhipu GLM, or locally deployable open source large models such as CogVLM and Qwen-VL. The specific selection is based on the business. If the data is not sensitive, a closed source model can be used, which has larger parameters and is smarter. If the data is sensitive, a locally deployed model is safer. The above-mentioned language recognition model can use language recognition models produced by iFlytek, Tencent, and Google. The above-mentioned large language model can use GPT-4, Wenxin Yiyan, Zhipu GLM, Kimi, Qwen, or llama models.

[0059] In this embodiment, the target road condition image, the first text data and the voice information are collected in parallel.

[0060] Further explanation is needed:

[0061] 1. Prompt Word Manager

[0062] The prompt word manager is an information scheduling algorithm model based on a large model. It mainly includes three modules: related information query, POI scheduling, and camera precision scheduling. These three functional modules are equipped with three sets of prompt words. The prompt word manager outputs text understanding to the large language model based on the first text data and the second text data. The large language model then trains the text understanding and outputs structured data to the prompt word manager. The prompt word manager then performs back-end parsing on the above structured data and outputs it to the API interface. For example: the user enters "Where is the Window of the World?" in the scheduling terminal, and the prompt word manager finally recognizes the user's intention as "POI scheduling" and outputs POI-related prompt words.

[0063] Specifically, the prompt word acquisition program processing method of the POI scheduling module includes the following three parts:

[0064] (1) Role Definition

[0065] Suppose you are an intelligent map assistant that understands user needs, converts them into built-in command outputs, then receives the command query results returned by the system and replies to the user based on the context.

[0066] (2) Built-in functions of POI scheduling module

[0067] The built-in functions of the system are (some examples):

[0068] 1) Search for POIs based on prompt words

[0069] {"intent":"findPoi","kv":"prompt word 1,prompt word 2"}

[0070] To implement fuzzy query, the prompt words need to be split appropriately. For example, ecological technology park needs to be split into ecology, technology, and park.

[0071] 2) Find POIs of a specific type near a POI

[0072] {"intent":"findNearByPoi","poi_id":"poi_ids","poi_type":"prompt word 1,prompt word 2","distance":200}

[0073] Among them, poi_id is the unique code of poi. If there are multiple, they are separated by commas. poi_type includes catering services, parking lots, cameras, etc., and distance is the distance.

[0074] 3) Reference examples

[0075] User: Where is Window of the World?

[0076] You: {"intent":"findPoi","kv":"World,Window"}

[0077] User: What food is nearby?

[0078] You: {"intent":"findNearByPoi","poi_id":"poi_id1,poi_id2","poi_type":"catering services"}

[0079] Based on the program logic and data structure processing in (1)-(2) above, the POI prompt words can be output. After the above process, structured data containing the query results can be obtained. The data is input into the large language model again. The large language model generates the corresponding answer based on its text generation ability and user input. At the same time, the system will locate the POI and zoom the map according to the queried POI longitude and latitude to realize user interaction and POI intelligent positioning functions.

[0080] Specifically, the program processing of the camera precision scheduling module is described as follows:

[0081] (1) Prompt word setting

[0082] The camera precise scheduling prompt consists of three parts:

[0083] a. Role Definition

[0084] Suppose you are an intelligent map assistant that understands user needs, converts them into built-in command outputs, then receives the command query results returned by the system and replies to the user based on the context.

[0085] b. System built-in functions

[0086] User: Get the camera that shoots a certain POI angle:

[0087] {"intent":"cameraLookAt","poi_type":"place 1,place 2","kv":"angle 1,angle 2"}

[0088] It should be noted that the camera angle may also need to be split, for example: shared bicycle parking should be split into shared bicycle, parking.

[0089] c. Reference examples

[0090] User: Find the camera at the bus stop facing the Children's Hospital;

[0091] You: {"intent":"cameraLookAt","poi_type":"Children's Hospital","kv":"Bus Stop"}.

[0092] Specifically, the relevant procedures for related information query are as follows:

[0093] Assume that you are an intelligent information query assistant, focusing on understanding the user's complex information query requirements about specific POI points, converting them into built-in command outputs, and then receiving the query results returned by the system of the present invention and replying to the user in the most appropriate way based on the context.

[0094] Specifically, the built-in functions of the related information query module are (some examples):

[0095] 1. Query the related information of a specific POI point

[0096] {"intent":"queryAssociation","poi_id":"poi_id","association_type":"Keyword1,Keyword2"}

[0097] association_type is the type of associated information that the user wants to query, such as a list of engineering projects, shelters, surrounding facilities, scenic spots, buildings, land use information, population, etc.

[0098] 2. Query specific related information around a specific POI point

[0099] {"intent":"queryNearbyAssociation","poi_id":"poi_id","association_type":"keyword1,keyword2","distance":200};

[0100] distance is the distance range that the user wants to query, in meters.

[0101] (3) Reference examples

[0102] User: I am at this construction site and would like to know what projects are being built here?

[0103] You: {"intent":"queryAssociation","poi_id":"construction site poi_id","association_type":"project list"};

[0104] User: Are there any shelters in this area?

[0105] You: {"intent":"queryNearbyAssociation","poi_id":"areapoi_id","association_type":"evacuation place","distance":100"}.

[0106] The above content is the routine related to the prompt word manager designed by the system of the present invention.

[0107] 2. Large Language Model

[0108] The large language model has learned the grammar, semantics and contextual relationships of the language through large-scale data training, and has powerful natural language processing capabilities. Using deep learning technology and pre-training and fine-tuning mechanisms, the large language model can understand complex language patterns and accurately parse user intentions.

[0109] After receiving the user input and the output of the prompt word manager, the large language model understands the content of the prompt word and follows the prompt word instructions to output the corresponding JSON instructions. For example, if the large model receives "where is the window of the world" + POI prompt word, it will output {"intent":"findPoi","kv":"window of the world"}.

[0110] 3. API interface

[0111] At the data processing level, the API interface is used for the back-end to parse JSON instructions, identify the intent "findPoi", call the corresponding API interface based on this keyword, and pass "window of the world" as a parameter, so that the system of the present invention can query the POI library and accurately obtain the address and longitude and latitude of the target POI point.

[0112] Among them, the three interface channels of the API are the associated information query interface channel, the POI scheduling interface channel and the camera precision scheduling interface channel. The associated information query interface channel is connected to the database. The POI scheduling interface channel is connected to the POI library. The camera precision scheduling interface channel is connected to the device information library. These three interfaces are used to transmit the output data of the corresponding functional modules of the above-mentioned prompt word manager.

[0113] Compared with the prior art, the system of the present invention has the following advantages:

[0114] 1. Have the ability to understand natural language:

[0115] The large language model built into the system of the present invention is good at understanding and processing natural language, which enables users to interact with the system directly through daily language without having to learn specific query syntax or commands. This lowers the user's usage threshold, allowing non-technical personnel to easily use the system of the present invention.

[0116] 2. It has the ability to quickly parse and respond to complex requests:

[0117] Based on the design of the prompt words in this paper combined with the advanced semantic understanding and text generation capabilities of the large language model, the system of the present invention can accurately parse the user's complex and ambiguous queries, even unstructured requests. This improves the speed and accuracy of processing requests, allowing the system of the present invention to provide relevant query results more quickly.

[0118] 3. Efficient cross-database high-level joint query:

[0119] The system of the present invention can perform complex joint queries in multiple databases, including cross-database information association and integration, which not only improves the flexibility of the query, but also can obtain more comprehensive and in-depth analysis results.

[0120] 4. Reduce manual participation:

[0121] Since the system of the present invention can automatically parse natural language requests and execute complex queries, the need for human intervention is greatly reduced, which not only reduces the burden on professionals, but also reduces the possibility of operational errors.

[0122] 5. Improve the efficiency and accuracy of emergency decision-making:

[0123] In emergency decision-making scenarios, it is crucial to quickly obtain accurate information. The information dispatch system based on the large language model can quickly provide the key information required for decision-making and support more accurate decision-making through advanced data analysis. This not only improves the efficiency of decision-making, but also improves the quality of decision-making.

[0124] Embodiment 2

[0125] like Figure 2 As shown, this embodiment discloses an information scheduling method based on a large language model, which is used in conjunction with the information scheduling system based on a large language model in Embodiment 1. The information scheduling method of the present invention is specifically as follows:

[0126] S10: Acquire voice information, first text data, and camera tag data obtained by parsing based on a multimodal large model.

[0127] It should be explained that the above-mentioned voice information, first text data and camera tag data can be collected and processed in parallel using a multi-threaded system kernel.

[0128] The specific method of obtaining the camera tag data based on the multimodal large model analysis in step S10 is:

[0129] Use the camera to obtain and intercept the target road condition image through the Rtsp stream;

[0130] Set the label prompt word of the target road condition image;

[0131] The target road condition image and label prompt words are placed into the multimodal large model for analysis to obtain the camera label data;

[0132] Write the camera tag data to the device information library.

[0133] Specifically, in order to realize the calibration of relevant facility features, the method of the present invention has multimodal information processing capabilities, intelligently converts the target road condition image into text label data (i.e., camera label data), and realizes intelligent understanding and classification of camera monitoring content. Among them, the above-mentioned label prompt words need to be set according to the camera monitoring scene requirements, such as facing the intersection, facing the hospital, facing the school, facing the bus stop, etc. The above-mentioned label prompt words can be added manually, and a camera can be marked with multiple label prompt words, and the label prompt words are separated by commas.

[0134] Through the above technical scheme, the method of the present invention applies a large multimodal model to learn the complex relationship between image features and semantic information by combining visual and language processing capabilities, using deep neural networks and a large amount of annotated graphic data, and through multi-level feature extraction and cross-modal alignment, the large multimodal model can accurately extract key information from the target road condition image and generate corresponding descriptions (i.e., camera label data).

[0135] It should be noted that the general multimodal large model may have certain deviations in its understanding of the target road condition image. If necessary, the camera label data can be manually corrected.

[0136] After the above step S10, the method of the present invention further comprises:

[0137] S20: Using a speech recognition model, converting the speech information into second text data.

[0138] S30: Using a prompt word manager, extracting text prompt words according to the text in the first text data and the second text data.

[0139] S40: placing the text prompt words into a large language model for deep training to obtain a query instruction.

[0140] It should be explained that the query instruction is obtained by converting the text after understanding it. The type of the query instruction is a JSON instruction, and its instruction structure can be seen in the routine demonstration provided in the first embodiment.

[0141] S50: After the camera tag data is written into the device information library, based on the query instruction, the database, POI library and device information library are queried through the API interface respectively to obtain the database query result, POI query result and device information query result respectively.

[0142] It should be explained that before creating a database, it is necessary to clarify the requirements, determine the data source, and clean the collected data to ensure data quality. When creating a database, it is necessary to design the relationship between various tables, such as POI table, population information table, enterprise information table, etc., and use foreign keys to establish associations. In order to improve the response speed, it is necessary to create indexes for commonly used fields, such as creating a geospatial index or an index for POI names.

[0143] Specifically, step S50 includes:

[0144] Based on the query instruction, the corresponding context-related data or information in the database is queried through the API interface, and the context-related data or information is used as the database query result.

[0145] Based on the query instruction, the corresponding POI address information and map information in the POI library are queried through the API interface, and the POI address information and map information are used as the POI query result.

[0146] Based on the query instruction, the corresponding camera angle information in the device information library is queried through the API interface, and the camera angle information is used as the device information query result.

[0147] The context-related data or information includes at least one or more of environmental information, population information, enterprise information, land use data, BIM data, water and electricity information, and building distribution information.

[0148] S60: The database query results, POI query results and device information query results are reversely placed into the large language model for deep training to generate query answers.

[0149] It should be explained that after the large language model receives the user's question + the structured information obtained from the query (i.e. the query results, POI query results and device information query results), it will conduct in-depth analysis and understanding based on the user's specific needs and the associated information obtained. Then the large model will accurately match the relevant information based on the context and content of the user's question, and generate highly natural and relevant query answers to ensure the accuracy of information transmission and the superiority of user experience.

[0150] Preferably, the large language model can adopt the Pangu large model.

[0151] S70: Forward the query answer to the dispatch terminal.

[0152] Compared with the prior art, the method of the present invention belongs to an information scheduling algorithm based on a large language model, which can quickly provide key information required for decision-making and support more accurate decision-making through advanced data analysis. This not only improves the efficiency of decision-making, but also improves the quality of decision-making.

[0153] Embodiment 3

[0154] like Figure 3 As shown, this embodiment discloses a computer device, which includes an internal memory, a processor, a non-volatile storage medium, and a computer program stored in the non-volatile storage medium and executable on the processor. When the processor executes the computer program, the information scheduling method based on the large language model in Embodiment 2 is implemented.

[0155] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk, etc. Various media that can store program codes.

[0156] For those skilled in the art, various other corresponding changes and deformations can be made according to the technical solutions and concepts described above, and all of these changes and deformations should fall within the protection scope of the claims of the present invention.

Claims

1. An information scheduling system based on a large language model, characterized in that: include: A camera, used to capture images of target road conditions; A multimodal large model, used to parse the target road condition image according to preset label prompt words, obtain camera label data, and write the camera label data into a device information library; The dispatch terminal is used to obtain the voice information and the first text data converted from the text query information, and to display the query answer; A language recognition model, used for converting the voice information into second text data; A prompt word manager, used for extracting text prompt words according to the text in the first text data and the second text data; A large language model is used to perform training according to the text prompt words and output corresponding query instructions; the large language model is also used to perform deep training on the query results to generate the query answer; as well as The API interface has at least three interface channels; Among them, the three interface channels of the API interface, one of which is connected to the database, another of which is connected to the POI library, and another of which is connected to the device information library. The API interface is used to perform information query in the database, the POI library and the device information library respectively according to the query instruction to obtain the query results.

2. The information scheduling system based on a large language model according to claim 1, characterized in that: The prompt word manager at least includes a related information query module, a POI scheduling module and a camera precision scheduling module; and / or The API interface is provided with an associated information query interface channel matched with the associated information query module, a POI scheduling interface channel matched with the POI scheduling module, and a camera precision scheduling interface channel matched with the camera precision scheduling module; The associated information query interface channel is communicatively connected to the database; the POI scheduling interface channel is communicatively connected to the POI library; and the camera precision scheduling interface channel is communicatively connected to the device information library.

3. An information scheduling method based on a large language model, used in conjunction with the information scheduling system based on a large language model according to any one of claims 1-2, characterized in that: The information scheduling method comprises: S10: Acquire voice information, first text data, and camera tag data obtained by parsing based on a multimodal large model; S20: using a speech recognition model to convert the speech information into second text data; S30: using a prompt word manager to extract text prompt words according to the text in the first text data and the second text data; S40: placing the text prompt word into a large language model for deep training to obtain a query instruction; Wherein, the query instruction is obtained by converting after understanding the text; S50: After the camera tag data is written into the device information library, based on the query instruction, the database, POI library and device information library are queried through the API interface respectively to obtain database query results, POI query results and device information query results respectively; S60: Reversely placing the database query result, the POI query result and the device information query result into the large language model for deep training to generate a query answer; S70: Forward the query answer to the dispatch terminal.

4. The information scheduling method based on a large language model according to claim 3, characterized in that: The camera tag data obtained by parsing based on the multimodal large model in step S10 includes: Use the camera to obtain and intercept the target road condition image through the Rtsp stream; Setting a label prompt word for the target road condition image; Put the target road condition image and the label prompt word into the multimodal large model for parsing to obtain camera label data; The camera tag data is written into the device information library.

5. The information scheduling method based on a large language model according to claim 3, characterized in that: The step S50 comprises: Based on the query instruction, the corresponding context-related data or information in the database is queried through the API interface, and the context-related data or information is used as the database query result.

6. The information scheduling method based on a large language model according to claim 5, characterized in that: The context-related data or information includes at least one or more of environmental information, population information, enterprise information, land use data, BIM data, water and electricity information, and building distribution information.

7. The information scheduling method based on a large language model according to claim 3, characterized in that: The step S50 comprises: Based on the query instruction, the corresponding POI address information and map information in the POI database are queried through an API interface, and the POI address information and the map information are used as the POI query result.

8. The information scheduling method based on a large language model according to claim 3, characterized in that: The step S50 comprises: Based on the query instruction, the corresponding camera angle information in the device information library is queried through the API interface, and the camera angle information is used as the device information query result.

9. The information scheduling method based on a large language model according to claim 3, characterized in that: The command type of the query command is a JSON command.

10. A computer device, characterized in that: The invention comprises an internal memory, a processor, a non-volatile storage medium, and a computer program stored in the non-volatile storage medium and executable on the processor, wherein when the processor executes the computer program, the information scheduling method based on a large language model as described in any one of claims 3 to 9 is implemented.

Images