Multi-modal disaster relief information analysis and task intelligent planning method and device

Abstract

The invention discloses a multi-modal disaster relief information analysis and task intelligent planning method and device, and relates to the technical field of artificial intelligence, and the method comprises the steps: carrying out the structural processing of multi-modal disaster information after a natural disaster, and generating multi-modal structural data; the preset large language model generates a current disaster relief task planning strategy based on a preset task prompt content generation algorithm and the multi-modal structured data; wherein the preset task prompt content generation algorithm comprises any one of a preset task prompt word template and a preset dynamic evolution prompt graph model composed of multiple reasoning nodes, and the multiple reasoning nodes at least comprise a disaster situation recognition node, a rescue priority judgment node, a resource constraint node and a path planning node; the reasoning nodes are connected through a preset dependency relationship, a preset reasoning sequence constraint and a preset conflict resolution rule to form a preset dynamic evolution prompt graph model. The scheme solves the problem that in the prior art, it is difficult to dynamically cope with constantly changing disaster relief sites.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a multimodal disaster relief information analysis and intelligent task planning method. Background Technology

[0002] After natural disasters (such as earthquakes, floods, and forest fires), the following challenges are common in disaster areas: complex scenarios and ever-changing tasks; damaged communication facilities and delayed information transmission; difficulty for manual dispatch to quickly adapt to changes on site; limited decision-making basis, slow response speed, and low resource allocation efficiency. These factors have become important bottlenecks hindering efficient rescue operations.

[0003] Current mainstream emergency response systems mostly rely on rule-driven static strategies and dispatching processes based on human experience. However, these methods are difficult to dynamically respond to the ever-changing disaster relief scene. Summary of the Invention

[0004] This application aims to at least solve the technical problems existing in the prior art. To this end, the first aspect of this application proposes a multimodal disaster relief information analysis and intelligent task planning method, which includes: The multimodal disaster information following natural disasters is processed in a structured manner to generate multimodal structured data; the multimodal disaster information includes disaster image data, disaster text information, and sensor perception data; The pre-defined large language model generates a current disaster relief task planning strategy based on a pre-defined task prompt content generation algorithm and multimodal structured data. The current disaster relief task planning strategy includes suggestions for personnel scheduling, resource allocation, and route planning. The pre-defined task prompt content generation algorithm includes either a pre-defined task prompt word template or a pre-defined dynamic evolution prompt graph model composed of multiple inference nodes. The multiple inference nodes include at least a disaster identification node, a rescue priority determination node, a resource constraint node, and a route planning node. The inference nodes are connected through pre-defined dependencies, pre-defined inference order constraints, and pre-defined conflict resolution rules to form the pre-defined dynamic evolution prompt graph model.

[0005] In one possible implementation, the preset task prompt content generation algorithm is a preset dynamic evolution prompt graph model, and the preset large language model generates the current disaster relief task planning strategy based on the preset task prompt content generation algorithm and multimodal structured data, including: Multimodal structured data is input into a preset dynamic evolution prompt graph model. Based on preset dependencies, preset inference order constraints, and preset conflict resolution rules, inference is performed sequentially through each inference node to generate the current disaster relief task planning strategy. Among them, preset dependencies are used to determine the execution dependency conditions of each inference node, preset inference order constraints are used to determine the execution order of each inference node, and preset conflict resolution rules are used to handle output conflicts between inference nodes.

[0006] In one possible implementation, multimodal structured data is input into a preset dynamic evolution prompt graph model. Based on preset dependencies, preset reasoning order constraints, and preset conflict resolution rules, reasoning is performed sequentially through each reasoning node to generate the current disaster relief task planning strategy, including: Multimodal structured data is input into a preset dynamic evolution prompt graph model. Based on preset dependencies, preset reasoning order constraints, and preset conflict resolution rules, reasoning is performed sequentially through each reasoning node to determine the rescue priority, resource allocation suggestions, and path planning suggestions for different disaster areas. Based on the information of the standby rescue teams, match the target rescue teams corresponding to each disaster area; Based on rescue priorities, resource allocation suggestions, route planning suggestions, and target rescue groups, a current disaster relief mission planning strategy is generated.

[0007] In one possible implementation, based on relevant information about the standby rescue teams, target rescue teams corresponding to each disaster area are matched, including: Obtain the location information of the standby rescue teams, their distance from each disaster area, and the mission requirements of each disaster area; Based on location information, distance information, and mission requirements, the corresponding type and number of target rescue personnel are determined to form corresponding target rescue teams.

[0008] In one possible implementation, the method further includes: The execution feedback information corresponding to each inference node is obtained in real time; the execution feedback information includes at least path unreachability information, resource usage conflict information, and task execution timeout or delay information. The execution feedback information is structured to generate failure feature description information; the failure feature description information is used to characterize infeasible decision elements or preset risk decision elements in the current disaster relief mission planning strategy. The preset dynamic evolution prompt graph model is updated based on the failure feature description information to generate a new dynamic evolution prompt graph model, and a new disaster relief task planning strategy is generated based on the new dynamic evolution prompt graph model.

[0009] In one possible implementation, the preset task prompt content generation algorithm is a preset task prompt word template, and the preset large language model generates the current disaster relief task planning strategy based on the preset task prompt content generation algorithm and multimodal structured data, including: Based on preset task prompt word templates and multimodal structured data, target prompt words are generated. The preset task prompt word templates include prompt structures corresponding to various types of disaster scenarios. The prompt structures include role setting fields, input image fields, input text fields, resource information fields, task target fields, task reasoning order fields, and output instruction fields. The input text fields include disaster description fields and weather and environment fields, and the resource information fields include available personnel fields, available vehicle fields, equipment list fields, and regional traffic data fields. Based on a pre-defined large language model, target prompts and pre-defined task instructions are processed to generate a current disaster relief task planning strategy.

[0010] In one possible implementation, the multimodal disaster information is disaster image data. The multimodal disaster information following a natural disaster undergoes structured processing, including: Object detection and recognition are performed on disaster image data to extract key target information; Optical character recognition is used to detect text regions in disaster image data to extract text information. Key target information and text information are aligned in a multimodal manner to obtain unified multimodal information; The unified multimodal information is encoded to obtain multimodal structured data.

[0011] In one possible implementation, disaster image data includes drone aerial images, satellite remote sensing images, monitoring images, and spatial geographic information images of the disaster area; disaster text information includes text information in historical disaster databases, disaster briefings reported by relevant departments, as well as text information about the disaster in news reports, residents' self-reports, and social media platforms; sensor perception data includes post-disaster vital sign detection information and disaster dynamic identification information collected by sound sensors, infrared sensors, and thermal imaging sensors.

[0012] The second aspect of this application proposes a multimodal disaster relief information analysis and intelligent task planning device, which includes: The processing module is used to perform structured processing on multimodal disaster information after a natural disaster occurs, generating multimodal structured data; the multimodal disaster information includes disaster image data, disaster text information, and sensor perception data; The generation module is used to generate a current disaster relief task planning strategy based on a preset large language model, a preset task prompt content generation algorithm, and multimodal structured data. The current disaster relief task planning strategy includes suggestions for personnel scheduling, resource allocation, and route planning. The preset task prompt content generation algorithm includes any one of the following: a preset task prompt word template and a preset dynamic evolution prompt graph model composed of multiple inference nodes. The multiple inference nodes include at least a disaster identification node, a rescue priority determination node, a resource constraint node, and a route planning node. The inference nodes are connected through preset dependencies, preset inference order constraints, and preset conflict resolution rules to form the preset dynamic evolution prompt graph model.

[0013] In one possible implementation, the preset task prompt content generation algorithm is a preset dynamic evolution prompt graph model, and the aforementioned generation module is specifically used for: Multimodal structured data is input into a preset dynamic evolution prompt graph model. Based on preset dependencies, preset inference order constraints, and preset conflict resolution rules, inference is performed sequentially through each inference node to generate the current disaster relief task planning strategy. Among them, preset dependencies are used to determine the execution dependency conditions of each inference node, preset inference order constraints are used to determine the execution order of each inference node, and preset conflict resolution rules are used to handle output conflicts between inference nodes.

[0014] In one possible implementation, the above-described generation module is further configured to: Multimodal structured data is input into a preset dynamic evolution prompt graph model. Based on preset dependencies, preset reasoning order constraints, and preset conflict resolution rules, reasoning is performed sequentially through each reasoning node to determine the rescue priority, resource allocation suggestions, and path planning suggestions for different disaster areas. Based on the information of the standby rescue teams, match the target rescue teams corresponding to each disaster area; Based on rescue priorities, resource allocation suggestions, route planning suggestions, and target rescue groups, a current disaster relief mission planning strategy is generated.

[0015] In one possible implementation, the above-described generation module is further configured to: Obtain the location information of the standby rescue teams, their distance from each disaster area, and the mission requirements of each disaster area; Based on location information, distance information, and mission requirements, the corresponding type and number of target rescue personnel are determined to form corresponding target rescue teams.

[0016] In one possible implementation, the above-described generation module is further configured to: The execution feedback information corresponding to each inference node is obtained in real time; the execution feedback information includes at least path unreachability information, resource usage conflict information, and task execution timeout or delay information. The execution feedback information is structured to generate failure feature description information; the failure feature description information is used to characterize infeasible decision elements or preset risk decision elements in the current disaster relief mission planning strategy. The preset dynamic evolution prompt graph model is updated based on the failure feature description information to generate a new dynamic evolution prompt graph model, and a new disaster relief task planning strategy is generated based on the new dynamic evolution prompt graph model.

[0017] In one possible implementation, the preset task prompt content generation algorithm is a preset task prompt word template, and the preset large language model generates the current disaster relief task planning strategy based on the preset task prompt content generation algorithm and multimodal structured data, including: Based on preset task prompt word templates and multimodal structured data, target prompt words are generated. The preset task prompt word templates include prompt structures corresponding to various types of disaster scenarios. The prompt structures include role setting fields, input image fields, input text fields, resource information fields, task target fields, task reasoning order fields, and output instruction fields. The input text fields include disaster description fields and weather and environment fields, and the resource information fields include available personnel fields, available vehicle fields, equipment list fields, and regional traffic data fields. Based on a pre-defined large language model, target prompts and pre-defined task instructions are processed to generate a current disaster relief task planning strategy.

[0018] In one possible implementation, the above-mentioned processing module is specifically used for: Object detection and recognition are performed on disaster image data to extract key target information; Optical character recognition is used to detect text regions in disaster image data to extract text information. Key target information and text information are aligned in a multimodal manner to obtain unified multimodal information; The unified multimodal information is encoded to obtain multimodal structured data.

[0019] In one possible implementation, disaster image data includes drone aerial images, satellite remote sensing images, monitoring images, and spatial geographic information images of the disaster area; disaster text information includes text information in historical disaster databases, disaster briefings reported by relevant departments, as well as text information about the disaster in news reports, residents' self-reports, and social media platforms; sensor perception data includes post-disaster vital sign detection information and disaster dynamic identification information collected by sound sensors, infrared sensors, and thermal imaging sensors.

[0020] A third aspect of this application proposes an electronic device comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the multimodal disaster relief information parsing and intelligent task planning method as described in the first aspect.

[0021] The fourth aspect of this application proposes a computer-readable storage medium storing at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the multimodal disaster relief information parsing and intelligent task planning method as described in the first aspect.

[0022] The embodiments of this application have the following beneficial effects: The multimodal disaster relief information parsing and intelligent task planning method provided in this application includes: structuring multimodal disaster information after a natural disaster to generate multimodal structured data, wherein the multimodal disaster information includes disaster image data, disaster text information, and sensor perception data; generating a current disaster relief task planning strategy based on a preset large language model and a preset task prompt content generation algorithm and the multimodal structured data, wherein the current disaster relief task planning strategy includes suggestions for personnel scheduling, resource allocation, and path planning; the preset task prompt content generation algorithm includes any one of a preset task prompt word template and a preset dynamic evolution prompt graph model composed of multiple inference nodes, wherein the multiple inference nodes include at least a disaster identification node, a rescue priority determination node, a resource constraint node, and a path planning node, and the preset dynamic evolution prompt graph model is formed by connecting the inference nodes through preset dependencies, preset inference order constraints, and preset conflict resolution rules. This solution integrates large language models and prompt word technology with multimodal data processing capabilities to realize an intelligent decision support system and method for disaster relief mission path planning, personnel scheduling, and equipment resource allocation in disaster emergency scenarios. It is flexibly applicable to intelligent auxiliary decision-making scenarios in various sudden natural disasters. Furthermore, through joint analysis of multimodal disaster information, a pre-set large language model is driven to generate disaster relief mission planning strategies. The output disaster relief mission planning strategies can be used as input to the command and dispatch system to achieve subsequent mission execution and evaluation feedback, forming a closed-loop process of "disaster understanding—mission planning—execution support," improving rescue efficiency and the scientific nature of decision-making. Simultaneously, due to the nature of disaster relief missions… The system employs a "combat-like" chain structure of "intelligence-decision-execution-evaluation," thereby introducing "combat-like" strategy scheduling concepts into disaster relief scenarios. Combined with the multimodal understanding and prompt word control capabilities of a large language model, it provides a novel, efficient, and universal solution for disaster task generation. Furthermore, by adopting a pre-defined dynamic evolution prompt graph model, the disaster relief task planning process is elevated from text-level prompts to a structured reasoning process. This enables explicit modeling of the internal logic of complex disaster relief tasks, making the reasoning process of the large language model more controllable and interpretable. It also allows for dynamic planning under multiple disaster areas and constraints, significantly improving the stability and practicality of disaster relief task planning. Attached Figure Description

[0023] Figure 1 A block diagram of a computer device provided in an embodiment of this application; Figure 2 A flowchart illustrating the steps of a multimodal disaster relief information analysis and intelligent task planning method provided in this application embodiment; Figure 3 A flowchart illustrating the steps for obtaining multimodal structured data is provided in this application embodiment. Figure 4A flowchart illustrating the steps for generating a current disaster relief mission planning strategy is provided in this embodiment of the application. Figure 5 A flowchart illustrating the steps for updating a preset dynamic evolution prompt graph model, as provided in this application embodiment; Figure 6 This is a structural block diagram of the multimodal disaster relief information analysis and intelligent task planning device provided in the embodiments of this application. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0025] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of embodiments of this disclosure, unless otherwise stated, "a plurality of" means two or more. Furthermore, the use of "based on" or "according to" implies openness and inclusiveness, because processes, steps, calculations, or other actions "based on" or "according to" one or more of the stated conditions or values ​​may in practice be based on additional conditions or beyond the stated values.

[0026] The multimodal disaster relief information analysis and intelligent task planning method provided in this application can be applied to computer equipment (electronic devices). The computer equipment can be a server or a terminal. The server can be a single server or a server cluster composed of multiple servers. This application does not specifically limit this. The terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets and portable wearable devices.

[0027] Taking a computer device as an example, Figure 1 A block diagram of a server is shown, such as Figure 1 As shown, the server may include a processor and memory connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. When the computer program is executed by the processor, it implements a multimodal disaster relief information analysis and intelligent task planning method.

[0028] Those skilled in the art will understand that Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the server to which the present application is applied. Optionally, the server may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0029] It should be noted that the execution subject of the embodiments of this application can be a computer device or a multimodal disaster relief information analysis and task intelligent planning device. The following method embodiments will be described with computer device as the execution subject.

[0030] Figure 2 This document presents a flowchart illustrating the steps of a multimodal disaster relief information analysis and intelligent task planning method, as provided in an embodiment of this application. Figure 2 As shown, the method includes the following steps: Step 202: Perform structured processing on the multimodal disaster information after the natural disaster occurs to generate multimodal structured data.

[0031] The multimodal disaster information includes disaster image data, disaster text information, and sensor perception data. Optionally, disaster image data includes drone aerial images, satellite remote sensing images, monitoring images, and spatial geographic information images of the disaster area; disaster text information includes text information in historical disaster databases, disaster briefings reported by relevant departments, as well as text information about the disaster in news reports, residents' self-reports, and social media platforms; sensor perception data includes post-disaster vital sign detection information and disaster dynamic identification information collected by sound sensors, infrared sensors, and thermal imaging sensors.

[0032] The aforementioned disaster image data is used to provide visual information such as the geographical environment and building damage in the disaster area. Spatial geographic information images include spatial information such as geographic coordinates, administrative divisions, transportation routes, and water conservancy distribution, used to assist in disaster location and route planning. The historical disaster database includes cases of similar historical disasters, loss assessments, and records of rescue strategies, used for large-scale model inference and strategy reference.

[0033] After obtaining multimodal disaster information after a natural disaster, the information can be input into a pre-defined large language model to perform structured processing.

[0034] The data processing involves structuring multimodal disaster information to provide accurate and clearly structured input data for the subsequent prompt word construction module and large model scheduling module.

[0035] In some optional embodiments, when the multimodal disaster information is disaster image data, such as Figure 3 As shown, Figure 3 A flowchart illustrating the steps for obtaining multimodal structured data, provided in this application embodiment, includes: Step 302: Perform object detection and recognition on the disaster image data to extract key target information.

[0036] Step 304: Detect optical character recognition in the text region of the disaster image data to extract the text information.

[0037] Step 306: Perform multimodal alignment processing on the key target information and text information to obtain unified multimodal information.

[0038] Step 308: Encode the unified multimodal information to obtain multimodal structured data.

[0039] Among these methods, image recognition processing technology can be used to detect and identify objects in disaster image data, and the key target information extracted can include information such as collapsed houses, road interruptions, and gatherings of people.

[0040] Next, optical character recognition (OCR) technology can be used to detect and recognize text regions in the image, extracting text information such as disaster slogans, signs, and map annotations.

[0041] After multimodal alignment of key target information and textual information, unified multimodal information is obtained, thus achieving cross-modal context unification. Finally, semantic encoding processing is performed on the unified multimodal information to obtain multimodal structured data, providing a semantic foundation for subsequent prompt word construction and reasoning in large language models.

[0042] Similarly, disaster-related textual information and sensor-sensed data can also be structured using appropriate techniques to obtain multimodal structured data. For example, natural language processing techniques can be used to structure disaster-related textual information, while data cleaning, standardization, feature extraction, and encoding techniques can be used to structure sensor-sensed data.

[0043] In this embodiment, since the inference performance of the preset large language model is highly dependent on the cleanliness and structure of the input data, and the original multimodal disaster information usually contains a lot of noise, such as low-quality scanned text and image interference elements, directly inputting it into the preset large language model may lead to comprehension bias. Data processing uses image recognition, OCR, and other methods to clean, parse, and structure the original multimodal disaster information, ensuring a high-quality input foundation for subsequent task prompt generation and model inference. In addition, in practical applications, disaster tasks often require the identification of key information such as location, personnel, and event type. Data processing can generate intermediate results such as key target information and marked text information in advance, which serve as an important part of prompt construction. This not only improves the inference efficiency of the model but also makes the system output more directional, controllable, and interpretable. Furthermore, separating the data processing process from the preset large language model is beneficial for modular system design. At the same time, data processing can also screen and de-identify sensitive information before the data enters the preset large language model, preventing the leakage of user privacy data and helping the system meet data compliance requirements.

[0044] Step 204: Based on the preset task prompt content generation algorithm and multimodal structured data, the preset large language model generates the current disaster relief task planning strategy.

[0045] The current disaster relief mission planning strategy includes suggestions on personnel scheduling, resource allocation, and route planning. The algorithm for generating the aforementioned pre-defined mission prompts includes either a pre-defined mission prompt word template or a pre-defined dynamic evolution prompt graph model composed of multiple inference nodes. Each inference node includes at least a disaster identification node, a rescue priority determination node, a resource constraint node, and a route planning node. These inference nodes are connected through pre-defined dependencies, pre-defined inference order constraints, and pre-defined conflict resolution rules to form the pre-defined dynamic evolution prompt graph model.

[0046] Specifically, the disaster identification node is used to perform semantic recognition of the disaster area status, disaster type, and severity based on multimodal structured data. The rescue priority determination node is used to determine the priority of different disaster areas or tasks by combining the disaster identification results with personnel risk information. The resource constraint node is used to perform constraint modeling on available personnel, vehicles, equipment, and time windows. The route planning node is used to generate candidate rescue route schemes under the premise of satisfying resource constraints.

[0047] In some optional embodiments, the preset task prompt content generation algorithm is a preset dynamic evolution prompt graph model, and the preset large language model generates the current disaster relief task planning strategy based on the preset task prompt content generation algorithm and multimodal structured data, including: Multimodal structured data is input into a preset dynamic evolution prompt graph model. Based on preset dependencies, preset reasoning order constraints, and preset conflict resolution rules, reasoning is performed sequentially through each reasoning node to generate the current disaster relief task planning strategy.

[0048] The system includes several key components: Preset dependencies, which define the execution dependencies of each inference node; and preset inference order constraints, which determine the execution order of each inference node, with the inference result of a preceding node serving as the input constraint for subsequent nodes. Preset conflict resolution rules are used to handle output conflicts between inference nodes; when multiple inference nodes produce conflicting outputs, the conflicting outputs are filtered or reordered based on these rules.

[0049] During execution, the pre-defined large language model no longer directly processes complete text prompts. Instead, it executes the reasoning process node by node according to the prompt graph structure corresponding to the pre-defined dynamic evolution prompt graph model, completing disaster understanding, task decomposition, and plan generation. By adopting the pre-defined dynamic evolution prompt graph model, the disaster relief task planning process is elevated from text-level prompts to a structured reasoning process. This achieves explicit modeling of the internal logic of complex disaster relief tasks, making the reasoning process of the large language model more controllable and interpretable. It can also adapt to the dynamic planning needs under multiple disaster areas and multiple constraints, significantly improving the stability and practicality of disaster relief task planning.

[0050] In some alternative embodiments, such as Figure 4 As shown, Figure 4 A flowchart illustrating the steps for generating a current disaster relief mission planning strategy, provided in this application embodiment, includes: Step 402: Input the multimodal structured data into the preset dynamic evolution prompt graph model. Based on the preset dependency relationship, preset reasoning order constraint and preset conflict resolution rule, perform reasoning processing through each reasoning node in sequence to determine the rescue priority, resource allocation suggestion and path planning suggestion for different disaster areas.

[0051] Step 404: Based on the relevant information of the standby rescue team, match the target rescue team corresponding to each disaster area.

[0052] Step 406: Based on rescue priorities, resource allocation suggestions, route planning suggestions, and target rescue groups, generate the current disaster relief mission planning strategy.

[0053] Specifically, for different disaster areas, multimodal structured data can be input into a pre-defined dynamic evolution prompt graph model. Based on pre-defined dependencies, pre-defined reasoning order constraints, and pre-defined conflict resolution rules, reasoning is performed sequentially through each reasoning node to determine the rescue priority, resource allocation suggestions, and route planning suggestions for different disaster areas. Furthermore, target rescue teams corresponding to each disaster area can be matched based on the relevant information of the standby rescue teams. Optionally, the relevant information of the standby rescue teams may include, but is not limited to, the location information of the standby rescue teams and their distance information from each disaster area.

[0054] Furthermore, in actual disaster relief operations, rapid changes in the disaster area environment, road blockages, or updated resource status can lead to failures in disaster relief mission planning strategies generated based on the initial disaster situation, such as unreachable paths, resource conflicts, or mission delays. Existing technologies often rely on manual re-analysis or regeneration of the overall plan, making it difficult to achieve rapid adaptive correction of the planning strategy. To address this issue, a failure-driven disaster relief mission planning correction and rollback mechanism is proposed, enabling dynamic adjustment of the planning strategy without requiring online training of a pre-set large language model.

[0055] Optionally, such as Figure 5 As shown, Figure 5 A flowchart illustrating the steps for updating a preset dynamic evolution hint graph model, as provided in this application embodiment, includes: Step 502: Obtain the execution feedback information corresponding to each inference node in real time.

[0056] The execution feedback information includes at least path unreachability information, resource usage conflict information, and task execution timeout or delay information.

[0057] Step 504: Perform structured processing on the execution feedback information to generate failure feature description information.

[0058] Among them, the failure feature description information is used to characterize the infeasible decision elements or preset risk decision elements in the current disaster relief mission planning strategy.

[0059] Step 506: Update the preset dynamic evolution prompt graph model based on the failure feature description information to generate a new dynamic evolution prompt graph model, and generate a new disaster relief task planning strategy based on the new dynamic evolution prompt graph model.

[0060] When updating the preset dynamic evolution prompt graph model, it may include, but is not limited to, adjusting the weight of the corresponding inference node; disabling or reducing the planning candidates corresponding to the failed path; and increasing the priority of alternative rescue paths or alternative resource solutions.

[0061] This allows the large language model to re-execute the inference process based on the new dynamic evolutionary cue graph model after the cue graph constraint update is completed, in order to generate new disaster relief task planning strategies.

[0062] This embodiment requires no online training or parameter updates; it only adjusts and updates the prompt graph structure and node weights, forming a closed loop of "planning generation → execution feedback → failure modeling → planning correction." By introducing a failure-driven planning correction mechanism, the disaster relief mission planning system possesses the ability to adaptively adjust based on execution feedback. Under dynamically changing disaster conditions, it can quickly avoid infeasible paths and resource conflicts, improving the robustness and response efficiency of the overall disaster relief mission planning.

[0063] In some alternative embodiments, the preset task prompt content generation algorithm is a preset task prompt word template. The preset large language model generates the current disaster relief task planning strategy based on the preset task prompt content generation algorithm and multimodal structured data, including: generating target prompt words based on the preset task prompt word template and multimodal structured data; wherein, the preset task prompt word template includes prompt structures corresponding to various types of disaster scenarios; the prompt structure includes role setting fields, input image fields, input text fields, resource information fields, task target fields, task reasoning order fields, and output instruction fields, the input text fields include disaster description fields, weather and environment fields, and the resource information fields include available personnel fields, available vehicle fields, equipment list fields, and regional traffic data fields; and the preset large language model processes the target prompt words and preset task instructions to generate the current disaster relief task planning strategy.

[0064] The role setting field allows you to define the background identity of a pre-defined large language model, guiding the model to generate answers from a professional perspective. For example, it could be: "You are a national-level emergency rescue dispatch expert, possessing disaster analysis and task deployment capabilities."

[0065] The input image field is used to input disaster image data, including coordinates and environmental details, to identify the affected area. For example, it can be: an aerial image of an earthquake-stricken area, including information such as building damage and road blockages.

[0066] The disaster description field can provide details about the specific disaster situation in the area reflected in the disaster image data. For example, it could read: "Severe building collapses occurred in the eastern part of a city after the earthquake, and it is preliminarily determined that many people are trapped and major traffic roads are blocked."

[0067] The weather and environment fields can be used to supplement information describing weather conditions that affect rescue efficiency. For example, this could be: continuous light rain with gusts of wind, low visibility, limiting airborne supply delivery capabilities.

[0068] The "Available Personnel" field contains structured personnel data for task allocation reference. For example, it could state: a total of 120 emergency personnel are available for deployment, stationed at two assembly points in the north and southeast.

[0069] The Available Vehicles field contains information on currently available vehicles and equipment. For example, it could include: 8 ambulances, 4 loaders, and 2 communications command vehicles, stationed in area X and area Y respectively.

[0070] The equipment list field is a list of auxiliary tools and specialized equipment matched by disaster type. For example, it could include: 5 life detectors, 3 sets of hydraulic demolition equipment, and 8 night vision devices.

[0071] The regional traffic data field represents the current traffic situation (road accessibility or obstruction). For example, it could be: the southwest main road is open, the eastern railway is interrupted, and the southern bridge is structurally damaged and impassable.

[0072] The task objective field clarifies the reasoning and generation tasks that the model will perform in this round. For example, it could be: determining the optimal rescue route for the current disaster area, rationally allocating personnel teams, and planning a list of equipment requirements.

[0073] The task reasoning order field is used to refine the reasoning logic and clarify the order of decisions within the model. For example, it can be: first, identify the disaster-stricken area in the image, then construct the task scenario by combining it with the text, match adjustable resources, and output rescue suggestions.

[0074] The output instruction field is used to limit the output content and format, facilitating the standardization of results for integration into subsequent modules. For example, it could include: rescue route planning (coordinates and Chinese description), personnel dispatch suggestions (dividing personnel into task groups by region, specifying the location, responsibilities, and personnel list for each group), and equipment requirement suggestions (automatically matching the required equipment and quantity based on the disaster type and current task).

[0075] Therefore, the generated multimodal structured data can be input into the preset task prompt word template. The preset task prompt word template can extract the corresponding information from the multimodal structured data according to the required field requirements, thereby generating the corresponding target prompt word.

[0076] In some optional embodiments, when matching target rescue teams corresponding to each disaster area based on the relevant information of the standby rescue teams, the location information of the standby rescue teams and their distance information from each disaster area can be obtained, as well as the task requirements information of each disaster area. Then, based on the location information, distance information, and task requirements information, the corresponding type and number of target rescue personnel are determined to form the corresponding target rescue teams.

[0077] For example, in 2025, a magnitude 7.1 earthquake occurred in a certain area, causing widespread building collapses, power outages, and road damage in parts of the city. The preset task instruction is that the first response team wants to quickly obtain suggestions on routes, resources, and personnel scheduling.

[0078] This allows for image input, including aerial photos of city blocks taken by drones and post-disaster maps from satellite remote sensing (containing streets, water bodies, and traffic nodes); it also allows for text input, including disaster briefings and information on available resources. The disaster briefing could be something like, "Four hours after the earthquake, buildings in the xx block in the northern part of the city were severely damaged. It is estimated that about 150 people are trapped, and there is currently no information on how to get around them." The available resource information could be something like, "A total of 4 off-road vehicles and 30 personnel can be mobilized, carrying 5 life detectors and 20 emergency lights." Then, data processing can be performed to obtain multimodal structured data.

[0079] Based on preset task prompt word templates and multimodal structured data, the specific generation process for generating target prompt words may include: { "role": "National-level earthquake emergency dispatch expert", "objective": "Based on the marked standby rescue groups (A / B / C groups) in the image, the disaster area, road traffic status, and structured resource data, quickly formulate efficient earthquake rescue personnel dispatch, equipment allocation, and route planning schemes." "inputs": { Image description: { Content: "Aerial photos taken by drone, clearly marked: disaster areas A / B / C, command center, material warehouse, locations of rescue personnel on standby groups A / B / C, green unobstructed main roads and red blocked roads." "Require": [ "Identifying the spatial relationship between standby groups and disaster areas", "Scheduling and route analysis based on road traffic conditions" ] }, "Text information": { "Disaster Situation Report": { "Disaster Area B": "Houses have collapsed severely, an estimated 30 people are trapped, communications are interrupted, roads are collapsed, and rescue efforts are the highest priority." "Disaster Area A": "A small number of people were injured, and traffic is basically open." "Disaster Area C": "Communications are disrupted, building structures are unstable, and people may be trapped." }, Weather conditions: 12°C, light rain, wind speed 18km / h, low visibility, unsuitable for airdrops and helicopter operations. }, "Structured Resources": { "Personnel": { Total number of search and rescue personnel: 60 "Medical personnel": 30, "Communication guarantee": 10, "Logistics": 20 }, "Vehicles": { "Ambulance": 4, "Excavator": 3, "Supply transport trucks": 2, "Communication Command Vehicle": 1 }, "equipment": [ {"Name": "Life Detector", "Quantity": "6 units", "Function": "Locate trapped individuals in rubble"}, {"Name": "Hydraulic demolition tools", "Quantity": "4 sets", "Function": "Clearing obstacles / Demolition"}, {"Name": "Emergency Lights", "Quantity": "12 sets", "Function": "Nighttime Lighting"}, {"Name": "Stretcher", "Quantity": "10", "Purpose": "Transfer of the wounded"}, {"Name": "Medical Kit", "Quantity": "20 sets", "Purpose": "Emergency Treatment"} ] } }, "task_modules": [ { Module: Disaster Area Prioritization and Task Analysis "Explanation": "Based on the disaster description and traffic conditions, determine the rescue priority for disaster areas A / B / C, and determine the initial deployment areas based on the location of the standby teams." }, { Module: "Personnel Scheduling and Standby Group Matching" Output format: Structured JSON or table "Note": "Based on the location and distance of standby groups A / B / C, and the mission requirements in the disaster area, personnel and tasks should be allocated appropriately." }, { "Module": "Route Planning Suggestions", "Description": "Avoid red-blocked roads and generate the optimal route to the disaster area". Output format: Path coordinates + Chinese description }, { Module: "Equipment and Vehicle Deployment Plan" "Notes": "Equipment, vehicles, and deployment plan required for each group", Output format: Two-dimensional table (equipment name, dispatch team, area of ​​use, task description) } ], "output_requirements": { Language: Chinese Structure: Clear JSON / Table / Path Description "Key Points": [ "Rescue priorities are clear" "Reasonable scheduling" "Efficient path", "Optimal utilization of resources" ] } } The target prompts generated based on the aforementioned preset task prompt templates and multimodal structured data can be input into a preset large language model for processing, ultimately yielding a disaster relief task planning strategy. For example, priority ranking and task analysis can be performed first, including determining the priority of rescue efforts for three disaster areas (A, B, and C) based on the disaster description and traffic conditions. Disaster Area B: Houses have collapsed severely, an estimated 30 people are trapped, communications are down, and roads have collapsed. Rescue is the highest priority there.

[0080] Disaster Area C: Communication is disrupted, building structures are unstable, and people may be trapped. Rescue priority is secondary.

[0081] Disaster Area A: A small number of people were injured, and transportation is basically unimpeded. Rescue priority is lowest.

[0082] Next, target rescue teams can be matched, including rationally allocating personnel and tasks based on the location, distance, and disaster area requirements of standby rescue teams A, B, and C. For example, target rescue team A, located near disaster area A, is allocated 15 search and rescue personnel, 10 medical personnel, 3 communications support personnel, and 5 logistics personnel, with the corresponding task of conducting search and rescue and providing medical assistance in disaster area A. Target rescue team B, located near disaster area B, is allocated 25 search and rescue personnel, 10 medical personnel, 3 communications support personnel, and 5 logistics personnel, with the corresponding task of prioritizing emergency search and rescue and medical assistance in disaster area B. Target rescue team C, located near disaster area C, is allocated 20 search and rescue personnel, 10 medical personnel, 4 communications support personnel, and 10 logistics personnel, with the corresponding task of conducting search and rescue in disaster area C, with a focus on ensuring communication restoration.

[0083] The route planning suggestions include avoiding red-blocked roads and generating the optimal route to the disaster area. For example, the specific coordinates of the optimal route from the disaster relief command center to disaster area B are represented as [longitude 1, latitude 1], [longitude 2, latitude 2],... The suggestions also include a Chinese description of the optimal route: "From the disaster relief command center, travel south along the main road until you reach disaster area B." Route planning suggestions are also output for disaster areas A and C, but examples of these are not provided here.

[0084] The resource allocation plan for each target rescue team can include the corresponding relationships between different types of equipment, quantities, different target rescue teams, different disaster areas, and mission descriptions. Optionally, different types of equipment may include, but are not limited to, life detectors, hydraulic rescue tools, emergency lights, stretchers, and medical kits. The mission description for life detectors is to locate trapped individuals in rubble, locate potentially trapped individuals in rubble, and search for injured persons; the mission description for hydraulic rescue tools is to clear obstacles and demolish structures; the mission description for emergency lights is nighttime illumination; the mission description for stretchers is to transfer the wounded; and the mission description for medical kits is emergency treatment.

[0085] Taking life detectors as an example, the corresponding relationships can be: Life Detector - 4 units - Target Rescue Team B - Disaster Area B - Locating trapped people in the rubble; Life Detector - 1 unit - Target Rescue Team C - Disaster Area C - Locating potentially trapped people in the rubble; Life Detector - 1 unit - Target Rescue Team A - Disaster Area A - Searching for injured people.

[0086] The resource allocation recommendations for each target rescue team may include the vehicle deployment plan, which may include the correspondence between different types of vehicles, their uses, different target rescue teams, different disaster areas, optimal routes, personnel types, and equipment. Optionally, different types of vehicles may include, but are not limited to, ambulances, excavators, supply trucks, and command and control vehicles.

[0087] For example, the correspondence could be: ambulance - rapid transfer of the wounded - B target rescue team - disaster area B - direct access to main road - medical personnel and stretchers; another example is excavator - rubble clearing - B target rescue team - disaster area B - direct access to main road - search and rescue personnel and hydraulic demolition tools. Further examples of these correspondences will not be provided here.

[0088] In some optional embodiments, new multimodal disaster information can be acquired in real time; this information is then processed and input into a preset large language model for further processing. The dynamic prompt word update mechanism of the large language model is used to adjust the suggested content for personnel scheduling, resource allocation, and route planning, thereby generating a new disaster relief task planning strategy. The process of generating a new disaster relief task planning strategy from the new multimodal disaster information can refer to the implementation process of the various embodiments of the current disaster relief task planning strategy described above, and will not be repeated here.

[0089] In this embodiment, the method includes: structuring multimodal disaster information after a natural disaster to generate multimodal structured data, wherein the multimodal disaster information includes disaster image data, disaster text information, and sensor perception data; a preset large language model generates a current disaster relief task planning strategy based on a preset task prompt content generation algorithm and the multimodal structured data; wherein the current disaster relief task planning strategy includes suggestions for personnel scheduling, resource allocation, and path planning; the preset task prompt content generation algorithm includes any one of a preset task prompt word template and a preset dynamic evolution prompt graph model composed of multiple inference nodes, wherein the multiple inference nodes include at least a disaster identification node, a rescue priority determination node, a resource constraint node, and a path planning node, and each inference node is connected through preset dependency relationships, preset inference order constraints, and preset conflict resolution rules to form a preset dynamic evolution prompt graph model. This solution integrates large language models and prompt word technology with multimodal data processing capabilities to realize an intelligent decision support system and method for disaster relief mission path planning, personnel scheduling, and equipment resource allocation in disaster emergency scenarios. It is flexibly applicable to intelligent auxiliary decision-making scenarios in various sudden natural disasters. Furthermore, through joint analysis of multimodal disaster information, a pre-set large language model is driven to generate disaster relief mission planning strategies. The output disaster relief mission planning strategies can be used as input to the command and dispatch system to achieve subsequent mission execution and evaluation feedback, forming a closed-loop process of "disaster understanding—mission planning—execution support," improving rescue efficiency and the scientific nature of decision-making. Simultaneously, due to the nature of disaster relief missions… The system employs a "combat-like" chain structure of "intelligence-decision-execution-evaluation," thereby introducing "combat-like" strategy scheduling concepts into disaster relief scenarios. Combined with the multimodal understanding and prompt word control capabilities of a large language model, it provides a novel, efficient, and universal solution for disaster task generation. Furthermore, by adopting a pre-defined dynamic evolution prompt graph model, the disaster relief task planning process is elevated from text-level prompts to a structured reasoning process. This enables explicit modeling of the internal logic of complex disaster relief tasks, making the reasoning process of the large language model more controllable and interpretable. It also allows for dynamic planning under multiple disaster areas and constraints, significantly improving the stability and practicality of disaster relief task planning.

[0090] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0091] Figure 6 This is a structural block diagram of a multimodal disaster relief information analysis and intelligent task planning device provided in an embodiment of this application.

[0092] like Figure 6 As shown, the multimodal disaster relief information analysis and intelligent task planning device 600 includes: The processing module 602 is used to perform structured processing on multimodal disaster information after a natural disaster occurs, and generate multimodal structured data; wherein, the multimodal disaster information includes disaster image data, disaster text information and sensor perception data.

[0093] The generation module 604 is used to generate a current disaster relief task planning strategy based on a preset large language model, a preset task prompt content generation algorithm, and multimodal structured data. The current disaster relief task planning strategy includes suggestions for personnel scheduling, resource allocation, and route planning. The preset task prompt content generation algorithm includes any one of the following: a preset task prompt word template and a preset dynamic evolution prompt graph model composed of multiple inference nodes. The multiple inference nodes include at least a disaster identification node, a rescue priority determination node, a resource constraint node, and a route planning node. The inference nodes are connected through preset dependencies, preset inference order constraints, and preset conflict resolution rules to form the preset dynamic evolution prompt graph model.

[0094] Regarding the apparatus in the above embodiments, the specific methods by which each module performs its operations have been described in detail in the embodiments related to the method, and will not be elaborated upon here. Each module in the above-described multimodal disaster relief information analysis and intelligent task planning apparatus can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations of each module.

[0095] In one embodiment of this application, a computer device is provided, the computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps: The multimodal disaster information following natural disasters is processed in a structured manner to generate multimodal structured data; the multimodal disaster information includes disaster image data, disaster text information, and sensor perception data; Based on preset task prompt word templates and multimodal structured data, target prompt words are generated; Based on a pre-defined large language model, target prompts and pre-defined task instructions are processed to generate a current disaster relief task planning strategy. The current disaster relief task planning strategy includes suggestions on personnel scheduling, resource allocation, and route planning.

[0096] The computer device provided in this application embodiment has a similar implementation principle and technical effect to the above method embodiment, and will not be described again here.

[0097] In one embodiment of this application, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, it performs the following steps: The multimodal disaster information following natural disasters is processed in a structured manner to generate multimodal structured data; the multimodal disaster information includes disaster image data, disaster text information, and sensor perception data; Based on preset task prompt word templates and multimodal structured data, target prompt words are generated; Based on a pre-defined large language model, target prompts and pre-defined task instructions are processed to generate a current disaster relief task planning strategy. The current disaster relief task planning strategy includes suggestions on personnel scheduling, resource allocation, and route planning.

[0098] The computer-readable storage medium provided in this embodiment is similar in principle and technical effect to the method embodiment described above, and will not be repeated here.

[0099] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0100] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0101] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A multimodal disaster relief information analysis and intelligent task planning method, characterized in that, The method includes: Multimodal disaster information following natural disasters is processed in a structured manner to generate multimodal structured data; wherein, the multimodal disaster information includes disaster image data, disaster text information, and sensor perception data; A pre-defined large language model generates a current disaster relief task planning strategy based on a pre-defined task prompt content generation algorithm and the multimodal structured data. The current disaster relief task planning strategy includes suggested content for personnel scheduling, resource allocation, and route planning. The pre-defined task prompt content generation algorithm includes either a pre-defined task prompt word template or a pre-defined dynamic evolution prompt graph model composed of multiple inference nodes. Each inference node includes at least a disaster identification node, a rescue priority determination node, a resource constraint node, and a route planning node. These inference nodes are connected through pre-defined dependencies, pre-defined inference order constraints, and pre-defined conflict resolution rules to form the pre-defined dynamic evolution prompt graph model.

2. The method according to claim 1, characterized in that, The preset task prompt content generation algorithm is a preset dynamic evolution prompt graph model. The preset large language model generates the current disaster relief task planning strategy based on the preset task prompt content generation algorithm and the multimodal structured data, including: The multimodal structured data is input into the preset dynamic evolution prompt graph model. Based on the preset dependency relationships, preset inference order constraints, and preset conflict resolution rules, inference processing is performed sequentially through each of the inference nodes to generate the current disaster relief task planning strategy. The preset dependency relationships are used to determine the execution dependency conditions of each of the inference nodes, the preset inference order constraints are used to determine the execution order of each of the inference nodes, and the preset conflict resolution rules are used to handle output conflicts between the inference nodes.

3. The method according to claim 2, characterized in that, The process of inputting the multimodal structured data into the preset dynamic evolution prompt graph model, and generating the current disaster relief task planning strategy by sequentially performing reasoning processing through each of the preset dependency relationships, preset reasoning order constraints, and preset conflict resolution rules based on the preset dependency relationships, preset reasoning order constraints, and preset conflict resolution rules, includes: The multimodal structured data is input into the preset dynamic evolution prompt graph model. Based on the preset dependency relationship, preset reasoning order constraint and preset conflict resolution rule, reasoning is performed sequentially through each of the inference nodes to determine the rescue priority, resource allocation suggestions and path planning suggestions for different disaster areas. Based on the information of the standby rescue teams, match the target rescue teams corresponding to each disaster area; Based on the rescue priority, resource allocation suggestions, route planning suggestions, and the target rescue group, the current disaster relief mission planning strategy is generated.

4. The method according to claim 3, characterized in that, The process of matching target rescue teams with each disaster area based on relevant information from the standby rescue teams includes: Obtain the location information of the standby rescue team, the distance information between the team and each disaster area, and obtain the task requirements information of each disaster area; Based on the location information, the distance information, and the task requirement information, the corresponding type and number of target rescue personnel are determined to form the corresponding target rescue team.

5. The method according to any one of claims 2-4, characterized in that, The method further includes: The execution feedback information corresponding to each inference node is obtained in real time; wherein, the execution feedback information includes at least path unreachable information, resource occupation conflict information, and task execution timeout or delay information. The execution feedback information is structured to generate failure feature description information; wherein, the failure feature description information is used to characterize infeasible decision elements or preset risk decision elements in the current disaster relief task planning strategy; The preset dynamic evolution prompt graph model is updated based on the failure feature description information to generate a new dynamic evolution prompt graph model, and a new disaster relief task planning strategy is generated based on the new dynamic evolution prompt graph model.

6. The method according to claim 1, characterized in that, The preset task prompt content generation algorithm is a preset task prompt word template. The preset large language model generates the current disaster relief task planning strategy based on the preset task prompt content generation algorithm and the multimodal structured data, including: Based on the preset task prompt word template and the multimodal structured data, target prompt words are generated; wherein, the preset task prompt word template includes prompt structures corresponding to various types of disaster scenarios; the prompt structure includes a role setting field, an input image field, an input text field, a resource information field, a task target field, a task reasoning order field, and an output instruction field, the input text field includes a disaster description field and a weather and environment field, and the resource information field includes a mobilizeable personnel field, an available vehicle field, an equipment list field, and a regional traffic data field; Based on the preset large language model, the target prompt words and preset task instructions are processed to generate the current disaster relief task planning strategy.

7. The method according to any one of claims 1-4, characterized in that, The multimodal disaster information is disaster image data. The structuring processing of the multimodal disaster information after a natural disaster includes: Object detection and recognition are performed on the disaster image data to extract key target information; Optical character recognition is performed on the text regions in the disaster image data to extract the text information; The key target information and text information are subjected to multimodal alignment processing to obtain unified multimodal information; The unified multimodal information is encoded to obtain the multimodal structured data.

8. The method according to any one of claims 1-4, characterized in that, The disaster image data includes drone aerial images, satellite remote sensing images, monitoring images, and spatial geographic information images of the disaster area; the disaster text information includes text information in the historical disaster database, disaster briefings reported by relevant departments, as well as text information about the disaster in news reports, residents' self-reports, and social media platforms; the sensor perception data includes post-disaster vital sign detection information and disaster dynamic identification information collected by sound sensors, infrared sensors, and thermal imaging sensors.

9. A multimodal disaster relief information analysis and intelligent task planning device, characterized in that, The device includes: The processing module is used to perform structured processing on multimodal disaster information after a natural disaster occurs, generating multimodal structured data; wherein, the multimodal disaster information includes disaster image data, disaster text information, and sensor perception data; The generation module is used to generate a current disaster relief task planning strategy based on a preset large language model, a preset task prompt content generation algorithm, and the multimodal structured data. The current disaster relief task planning strategy includes suggested content for personnel scheduling, resource allocation, and path planning. The preset task prompt content generation algorithm includes either a preset task prompt word template or a preset dynamic evolution prompt graph model composed of multiple inference nodes. The multiple inference nodes include at least a disaster identification node, a rescue priority determination node, a resource constraint node, and a path planning node. Each inference node is connected through preset dependencies, preset inference order constraints, and preset conflict resolution rules to form the preset dynamic evolution prompt graph model.

10. An electronic device, characterized in that, The electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the multimodal disaster relief information parsing and intelligent task planning method as described in any one of claims 1-8.

Images