An intelligent emergency method and system based on an AI large model

Abstract

This invention discloses an intelligent emergency response method and system based on an AI-powered big data model, specifically relating to the field of emergency management technology. It addresses existing problems in emergency management such as insufficient intelligent emergency decision-making, fragmented multi-source data, inefficient emergency material management, separation of functions between normal and emergency situations, and poor communication reliability. The method reduces reliance on manual decision-making through an AI-powered big data model, achieving unified data management and real-time synchronization, eliminating blind spots in collaboration. Furthermore, it utilizes AI-powered big data model video recognition to achieve intelligent inventory and distribution traceability of emergency materials. The integrated dual-mode architecture for both normal and emergency situations enables second-level mode switching, allowing for rapid, verification-free distribution of materials and coordinated emergency resources in emergencies, while providing intelligent management of materials and equipment during normal times. It employs full-duplex network socket communication and a hierarchical messaging mechanism to ensure priority transmission of emergency commands, and supports real-time collaboration across multiple terminals. This addresses the issues of insufficient intelligent emergency decision-making, fragmented multi-source data, inefficient emergency material management, separation of functions between normal and emergency situations, and poor communication reliability.

Description

Technical Field

[0001] This invention relates to the field of emergency management technology, and more specifically, to an intelligent emergency response method and system based on an AI large-scale model. Background Technology

[0002] In the emergency management, public safety, Internet of Things, smart government, and industrial internet industries, emergency management systems undertake the core missions of risk prevention and control, emergency response, collaborative linkage, and efficiency improvement. They adapt to the needs of different industry scenarios, play a differentiated and indispensable key role, and become an important support for the safe and stable operation of various industries. However, current emergency management systems have core deficiencies in areas such as intelligent decision-making, multi-terminal collaboration, material management, and emergency response, failing to meet the core requirements of modern emergency response: "rapid, accurate, and collaborative." Specifically: The lack of intelligent emergency decision-making relies on human experience: Traditional emergency systems only have data recording and simple query functions, lacking AI-driven intelligent analysis capabilities. Emergency event classification, material matching, and rescue route planning all depend on human decision-making, resulting in delayed response and a high risk of decision-making errors. Furthermore, the lack of integrated professional AI models makes it impossible to process multimodal emergency data such as video and text, and it has a weak ability to identify non-standard emergency materials. Fragmented multi-source data leads to low collaboration efficiency: Emergency supplies inventory, emergency team locations, sensor monitoring data, and GIS geographic information are stored in different systems without a unified data fusion platform. Data synchronization delay is ≥10 seconds, which easily leads to collaboration blind spots such as "unclear material locations and unclear team status". The lack of standardized data interfaces makes cross-departmental and cross-regional data sharing difficult.

[0003] Emergency supplies management is crude and traceability is weak: emergency supplies rely heavily on manual inventory and paper records, inventory updates are not timely, and problems such as expired and short-term supplies occur frequently; existing identification solutions only support RFID barcode scanning for standard products and cannot cover non-standard products without labels; after emergency distribution, there is a lack of full-chain traceability, the flow of supplies is unclear, and it is difficult to review.

[0004] The system suffers from a disconnect between routine and emergency functions, resulting in a delayed emergency response: routine material management and emergency dispatch are separate systems with no dual-mode architecture design. In emergencies, manual system switching and data entry are required, with mode switching taking ≥30 minutes. The emergency triggering method is limited, lacking redundant designs such as automatic sensor monitoring and one-click hardware triggering, making it prone to failure in extreme scenarios. The system is paralyzed after a network outage, making it impossible to distribute basic materials and record events.

[0005] Poor communication reliability and instruction transmission delay: Using HTTP polling or simple MQTT communication, the delay in issuing emergency instructions is ≥3 seconds, and instructions are easily lost due to network congestion; there is no hierarchical message processing mechanism, and ordinary business and emergency instructions compete for resources; weak multi-terminal collaboration capability, and information is not synchronized between the command center, emergency terminals and rescue teams, which affects rescue efficiency. Summary of the Invention

[0006] In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides an intelligent emergency response method and system based on AI large model to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: An intelligent emergency response method based on a large AI model includes the following steps: S1, the multi-source data fusion layer collects multi-source emergency data through standardized interfaces, and performs fusion, cleaning and standardization processing on the collected data to provide high-quality and standardized data input support for the core layer of the AI ​​big model; The S2 AI big model core layer analyzes and processes standardized data through the emergency big model, identifies emergency event types and determines event levels, and provides accurate event judgment basis for subsequent emergency dispatch and handling. The core layer of the AI ​​big model combines geographic information and real-time traffic data to plan the optimal rescue route and match it with the nearest emergency teams, material reserve points and medical institutions, providing scientific route and resource matching support for the coordinated dispatch of emergency resources. The core layer of the AI ​​big model analyzes the event scenario, the scope of impact, and the number of people affected, calculates the types and quantities of emergency supplies needed, generates allocation lists and dispatch instructions, and provides clear instruction support for the distribution of supplies by smart emergency cabinets and the handling of rescue personnel. S3, the real-time network communication layer, prioritizes sending dispatch instructions to intelligent emergency cabinets and rescue terminals through a full-duplex real-time communication network. At the same time, it coordinates with the normal and emergency business layers to switch to emergency mode, providing communication and mode support for the timeliness and coordination of emergency response. S4. The intelligent emergency cabinet receives dispatch instructions and performs on-demand distribution of materials or one-click full opening. At the same time, it records the distribution of materials through visual recognition, providing real-time execution data support for system inventory updates and material traceability. S5, the IoT terminal layer continuously collects on-site situation, equipment status and material inventory data through various terminal devices, and transmits them synchronously to the command center, providing dynamic data support for commanders to grasp the on-site situation in real time and adjust the response plan. The S6 AI big model core layer analyzes data from the entire emergency response process, conducts debriefing analysis, and generates debriefing reports and optimization plans, providing decision support for subsequent emergency response process iterations and resource allocation optimization.

[0008] Furthermore, the multi-source data fusion layer includes a data management module, which is configured to: Perform interface adaptation, cleaning, transformation, and quality verification on multi-source heterogeneous data to achieve data standardization; Perform spatial, temporal, and semantic correlations on emergency data to complete multi-dimensional data fusion analysis; Data is distributed on demand based on user roles and emergency scenarios, and secure access is achieved through hierarchical permission control.

[0009] Furthermore, the core layer of the AI ​​large model is also configured as follows: Based on multi-source event input and unified analysis, combined with sensor time series data and trend prediction, early warning of abnormal events is achieved, and warning information is pushed out. The system uses video recognition to intelligently identify and inventory emergency supplies, automatically matching supply needs and generating supply warnings based on emergency scenarios. The rescue routes are dynamically updated by combining geographic information and traffic conditions, and standardized response procedures are pushed out based on the emergency response knowledge base. The entire emergency response process is analyzed to identify weaknesses and generate suggestions for strategy optimization.

[0010] Furthermore, the real-time network communication layer includes a communication scheduling module, which is configured to: Messages are prioritized according to their emergency level and scheduled using priority queues. Automatic connection establishment, disconnection reconnection, and heartbeat detection for terminal connections; Offline messages are cached and resent according to priority after the terminal reconnects; Authentication is performed on terminal access, and data transmission is encrypted for protection.

[0011] Furthermore, the emergency response method also includes a dual-mode switching mechanism for normal and emergency situations: In normal operation mode, the tasks include materials management, equipment inspection, and emergency drills. Emergency mode enables the reception of events, rapid distribution of supplies, resource coordination, and on-site situation feedback. It supports manual triggering, automatic triggering, and hardware triggering, enabling second-level mode switching and adaptive adjustment of business rules.

[0012] Furthermore, the emergency response method utilizes multiple IoT terminal devices in collaboration to achieve emergency data collection and command execution, specifically including: The intelligent emergency cabinet integrates a camera, electromagnetic lock, communication module, environmental sensor, one-button alarm and uninterruptible power supply, and can realize normal material management, emergency distribution, network outage caching and data synchronization, and supports wide temperature and low power consumption operation. Environmental monitoring data is collected in real time through independent IoT sensors, the collection frequency is adjusted according to the operating conditions, and early warning information is automatically reported when the data is abnormal. Through mobile terminal applications configured for command personnel, rescue personnel, and maintenance personnel respectively, emergency command, on-site handling, and equipment operation and maintenance management can be realized, forming a collaborative linkage with smart emergency cabinets and IoT sensors.

[0013] Furthermore, the emergency response method employs a hierarchical storage architecture, including: A relational database stores structured data such as emergency events, material information, access control data, dispatch plans, and debriefing reports, and is divided into tables by event identifier and region to improve query efficiency; Distributed caching units cache emergency instructions, real-time inventory, device status, and frequently accessed data, reducing database pressure through differentiated caching expiration times; A local embedded database on the terminal enables offline data storage and automatically synchronizes to the cloud after the network is restored; The object storage unit stores on-site audio, video, and document files, and supports file fragmentation and access control.

[0014] Furthermore, the intelligent emergency cabinet is also equipped with an AI visual recognition unit and a material lifecycle management unit for basic identification and inventory monitoring, specifically including: The system performs image acquisition, target detection, and feature recognition on the materials inside the cabinet to accurately identify the specifications, quantity, and condition of the materials. The identification results are compared with the system inventory data in real time to generate a real-time inventory ledger, which is then synchronized to the management platform. Identify materials in the cabinet that need to be replenished or replaced, generate replenishment reminders, and push them to relevant maintenance terminals; In the offline state, all identification data and inventory comparison results are cached locally and automatically synchronized to the management platform after the network is restored, ensuring the continuity of material status monitoring.

[0015] Furthermore, the AI ​​visual recognition unit and the material lifecycle management unit are also used for material validity period control and distribution priority configuration, specifically including: Automatically verify and cross-check the expiration dates of all materials in the cabinet to accurately identify near-expiration and expired materials; Near-expiry and expired materials are classified and marked, and early warning information is pushed to the management platform and operation and maintenance terminal in a hierarchical manner. At the same time, the storage cell corresponding to the expired materials is automatically locked to prevent them from being issued. Based on the type and level of the emergency and the needs of the on-site scenario, the priority of material distribution is dynamically configured to ensure the rapid retrieval of core emergency materials and adapt to the material dispatching needs under different emergency scenarios.

[0016] On the other hand, the present invention provides an intelligent emergency platform system based on an AI large model, comprising the following modules: The multimodal emergency intelligent engine module integrates four core capabilities: emergency event identification, intelligent decision-making, material identification, and solution generation, providing multi-source data processing support that is deeply adapted to emergency scenarios; The unified data fusion engine module integrates multi-source data through standardized interfaces, completing data cleaning, format standardization, and correlation analysis, providing high-quality data input for the multimodal emergency intelligent engine module; The communication module uses the Network Sockets Protocol to build a full-duplex real-time communication network. Combined with node service technology, it achieves high-concurrency message processing and priority scheduling. Through hierarchical message transmission, heartbeat detection, automatic reconnection, and offline message resending mechanisms, it ensures real-time collaboration between the terminal and the platform, realizes low-latency issuance of emergency commands and secure encrypted transmission of communications, and provides efficient communication support for stable and reliable linkage between the command center, intelligent emergency cabinet, and rescue terminal. The routine and emergency business module adopts a layered and modular architecture to build a routine and emergency dual-mode business processing logic, integrates the functions of routine material management and emergency dispatch, realizes intelligent mode switching and seamless business connection, covers the entire process of routine operation and maintenance, emergency response and post-event review, and provides efficient support for the routine and emergency business conversion and full-cycle closed-loop management. The IoT terminal layer module integrates various emergency IoT terminal devices to realize emergency material storage and management, environmental status monitoring and on-site interactive execution functions, and provides unified data collection and command execution support for the upper-layer platform; The data storage layer module adopts a hybrid storage architecture that combines relational databases, distributed caching, terminal-local embedded databases, and object storage.

[0017] Compared with existing technologies, this invention has the following beneficial effects: This emergency platform system, based on an AI large-scale model, not only reduces reliance on manual decision-making, thereby improving emergency response efficiency, but also achieves unified data management and real-time synchronization, eliminating blind spots in collaboration. Furthermore, it can achieve intelligent inventory and distribution traceability of emergency supplies through AI large-scale model video recognition, and performs local caching during network outages to ensure the continuity of material management. Simultaneously, its integrated peacetime and emergency dual-mode architecture enables second-level mode switching, allowing for rapid, verification-free distribution of supplies and linkage of emergency resources in emergencies, while intelligently managing materials and equipment during peacetime. Moreover, it employs full-duplex network socket communication and a hierarchical messaging mechanism to ensure priority transmission of emergency instructions, and supports real-time collaboration across multiple terminals. 1. The unified data fusion engine module automatically completes the collection, cleaning, and standardization of multi-source emergency data. The multimodal emergency intelligent engine module automatically identifies emergency events, determines their severity, plans rescue routes, matches resources, and generates dispatch instructions. It can uniformly process multimodal data such as video and text. The communication module enables automatic priority issuance of instructions and multi-terminal collaboration. The intelligent emergency cabinet and IoT terminal automatically execute material distribution and on-site data collection and feedback, replacing traditional manual data processing, analysis, dispatch, and operation. This significantly reduces reliance on human decision-making, reduces human error and response delay, and thus improves emergency response efficiency. 2. The unified data fusion engine module integrates, cleans, standardizes, and correlates multi-source emergency data. Standardized interfaces enable cross-departmental and cross-regional data exchange, achieving unified collection and processing of multi-source data. The communication module's full-duplex real-time communication, offline message caching and resending, and automatic terminal synchronization mechanisms enable two-way real-time data interaction between the command center, intelligent emergency cabinets, and rescue terminals. The hierarchical hybrid storage architecture categorizes, stores, and synchronizes structured data, cached data, offline terminal data, and file data, enabling unified management and real-time synchronization of emergency data, equipment status, material information, and on-site situation across the entire system. This eliminates data barriers and collaboration blind spots between terminals and links, thereby achieving efficient collaborative response. 3. The AI ​​visual recognition unit equipped in the smart emergency cabinet can intelligently identify and inventory the materials inside the cabinet. It can identify non-standard materials without labels and record relevant information on material distribution based on the material lifecycle management unit, so as to realize intelligent inventory and distribution traceability of emergency materials. At the same time, the identification data, inventory data and distribution data are cached locally by the local embedded database of the terminal in the absence of network. After the network is restored, they are automatically synchronized to the management platform, so that the identification, distribution and inventory status of materials are continuously controlled, thereby realizing intelligent inventory and traceable distribution of emergency materials and ensuring the continuity of material management. 4. By constructing a dual-mode integrated business processing logic for normal and emergency operations through the normal and emergency business modules, it supports multiple triggering methods, including manual, automatic, and hardware, to achieve second-level mode switching. Multiple triggering methods form redundancy protection. In emergency mode, it performs business such as event reception, rapid material distribution, resource linkage, and on-site situation feedback. In normal mode, it performs business such as material management, equipment inspection, and emergency drills. This enables the system to adaptively switch business rules, thereby achieving rapid material distribution without verification and linkage of emergency resources in emergency scenarios, as well as intelligent management of materials and equipment in normal situations, and integrated control of normal and emergency operations. 5. A full-duplex real-time communication network is constructed using the network socket protocol through the communication module. Combined with node service technology, high-concurrency message processing and priority scheduling are achieved. Simultaneously, the communication scheduling module prioritizes messages according to their emergency level and uses priority queue scheduling to prevent ordinary services from crowding out emergency command resources. Automatic connection establishment, reconnection after disconnection, and heartbeat detection are implemented for terminal connections. Offline messages are cached and resent according to priority after the terminal reconnects, ensuring that emergency commands are transmitted with priority. Stable data interaction is achieved between the command center, intelligent emergency cabinet, and rescue terminals, thereby ensuring the priority transmission of emergency commands and supporting real-time collaboration across multiple terminals. Attached Figure Description

[0018] Figure 1 This is a flowchart of the emergency response method of the present invention; Figure 2 This is a flowchart of the emergency event identification and intelligent decision-making process of the present invention; Figure 3 This is a flowchart of the normal / urgent mode switching process of the present invention; Figure 4 This is a flowchart of the modules of the emergency platform system of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1: Figures 1-3 This invention presents an intelligent emergency response method based on a large AI model, comprising the following steps: S1, the multi-source data fusion layer collects multi-source emergency data through standardized interfaces, and performs fusion, cleaning and standardization processing on the collected data to provide high-quality and standardized data input support for the core layer of the AI ​​big model; The multi-source data fusion layer includes a data management module, which is configured to: perform interface adaptation, cleaning, transformation and quality verification on multi-source heterogeneous data to achieve data standardization; perform spatial correlation, temporal correlation and semantic correlation on emergency data to complete multi-dimensional data fusion analysis; distribute data on demand according to user roles and emergency scenarios; and achieve secure access through hierarchical permission control. The multi-source heterogeneous data interface adaptation includes: providing multiple access methods such as descriptive state transition application programming interfaces, network socket communication, and message queue telemetry transmission, adapting to the data formats of sensors, smart cabinets, and third-party systems, and achieving one-time access and sharing across the entire platform; cleaning and transformation includes: automatically filtering abnormal data and converting heterogeneous data from different sources into a unified system format; quality verification includes: periodically verifying the integrity, accuracy, and timeliness of data, triggering a re-collection mechanism for missing data, and marking expired data as invalid to ensure data reliability; Automatic filtering of abnormal data is achieved through a dual mechanism of real-time filtering based on a rule engine and intelligent recognition based on an AI model. For sensor false alarms, the system has a built-in configurable rule engine that verifies the physical threshold, numerical mutation rate, and state constancy of the data in real time. Data that triggers any rule will be marked as suspicious and blocked. For data with incorrect format, the system presets a strict data pattern and verifies the number, type, length, and enumeration values ​​of fields during parsing. Any data stream that does not conform to the preset structure will be immediately identified as a format error and transferred to the error handling channel. At the same time, for complex and implicit abnormal patterns, the system deploys an AI anomaly detection model trained on historical normal data to calculate the anomaly score of new data in real time, intelligently identify and isolate data that exceeds the threshold, and continuously optimize the filtering accuracy by combining manual review feedback. The standardized transformation of heterogeneous data is achieved through a combination of pluggable converter mode and visual configuration of the transformation pipeline. The system develops a dedicated data converter plugin for each type of heterogeneous data source. This plugin explicitly includes parsing logic and mapping logic. When data is accessed, the system automatically calls the corresponding converter according to its source identifier and outputs data objects that fully conform to the system's unified data format specifications. For non-technical users, the system provides a visual interface that supports flexible construction and execution of complex data cleaning and transformation processes through drag-and-drop component declarative configuration, thereby achieving one-time access and sharing across the entire platform. Data quality verification and assurance are achieved through automated multi-dimensional verification, closed-loop processing, and visual monitoring; integrity verification uses a heartbeat packet and time window counting mechanism to monitor whether the data stream is interrupted or sparse; accuracy verification uses cross-validation and business rule verification to identify outliers by comparing multi-source data or verifying business logic that cannot be violated; timeliness verification determines whether the data is expired by calculating data latency and comparing it with preset thresholds; and it also addresses missing data. To mitigate the impact of quality differences between different data sources on subsequent event classification and scheduling results, the multi-source data fusion layer calculates the credibility of each type of data source:

[0021] in, Indicates the first Class data source at time Credibility; Indicates the first Data integrity scoring for data sources; Indicates the first Data timeliness rating for data sources of the same type; Indicates the first Data consistency score for data sources of the same type; , , These are the weighting coefficients, and .

[0022] The value is obtained from the ratio of the number of valid fields already uploaded to the total number of fields that should be uploaded from the data source; It is calculated from the delay between the data source upload time and the platform reception time; It is determined by the degree of consistency between the correlation results of this data source and other data sources for the same event; , , Pre-set parameters for the system to increase the weight of data timeliness scores in fire scenarios; Based on credibility , obtained the Normalized fusion weights of similar data sources : in, Indicates the first Class data source at time Normalized fusion weights; This represents the sum of the credibility of all data sources involved in the integration; Based on the normalized fusion weights corresponding to each data source, similar features are fused to obtain standardized feature values ​​after fusion. : in, This represents the standardized eigenvalues ​​after fusion; Indicates the first Class data source at time The corresponding standardized eigenvalues; Indicates the first Class data source at time Normalized fusion weights; After the above processing, the multi-source data fusion layer outputs standardized event data packets, which are then used as input data for the core layer of the AI ​​large model.

[0023] The system automatically triggers a data collection replenishment mechanism: missing events are pushed into the task queue, and the scheduler automatically executes command resending, switches to backup data sources, or uses historical data for imputation; for expired / invalid data, the system automatically tags expired data in its metadata and performs downgraded use, automatic archiving, or safe disposal according to preset policies; all verification results and handling operations are recorded and visualized through a global data quality monitoring dashboard, reflecting the health of each data source and SLA achievement in real time, forming a complete reliability assurance closed loop from monitoring, alarming to handling.

[0024] Spatial correlation refers to linking emergency material reserve points, emergency teams, and incident locations through GIS coordinates to achieve spatial matching of events, materials, and teams, and quickly locate the nearest resources; temporal correlation refers to analyzing the temporal change trends of sensor data and the temporal patterns of material consumption to provide data support for AI big model early warning and prediction; semantic correlation refers to using the semantic understanding capabilities of big models to link emergency event descriptions and handling knowledge bases and material attributes to improve decision-making accuracy. This system constructs a multi-dimensional, intelligent data fusion and decision support framework. In terms of spatial correlation, it standardizes the address information of emergency resources and event locations into GIS coordinates through a unified geographic information platform. Utilizing spatial indexing and path analysis algorithms, it calculates and recommends optimal resource scheduling paths in real time, achieving precise spatial matching and visualized command of events, materials, and teams. In terms of temporal correlation, the system aggregates dynamic data streams from sensor monitoring and material consumption based on unified timestamps. Through sliding window analysis, trend decomposition, and pattern recognition algorithms, it deeply mines the inherent temporal patterns, periodic characteristics, and abnormal fluctuations, forming structured temporal features. This provides crucial data input for AI-powered large-scale models to conduct risk warnings, situation predictions, and resource demand calculations. In terms of semantic correlation, the system relies on a domain-knowledge-enhanced large-scale model to perform entity recognition, relationship extraction, and intent understanding on unstructured emergency event descriptions. It then performs vectorized semantic matching with the emergency response knowledge base and material attribute database, automatically associating historical similar cases, standard handling procedures, and suitable material resources, thereby improving the accuracy and intelligence of emergency decision-making. On-demand data distribution refers to pushing personalized data based on user roles and emergency scenarios; hierarchical access control is based on role-based permission levels to restrict the scope of data access and ensure data security.

[0025] A four-dimensional mapping model of user-role-scenario-data is constructed to achieve intelligent on-demand distribution and security management. At the on-demand distribution level, the system first defines and maintains detailed user role profiles and emergency scenario templates. When an event occurs, the system automatically matches a preset data distribution rule engine based on the logged-in user's role and the currently activated emergency scenario. According to the rules, this engine filters, assembles, and pushes personalized data views needed only for that role in a specific scenario from the merged data pool in real time. For example, it pushes global situation and decision-making reports to the commander, and pushes on-site videos and personnel locations to rescuers, achieving a personalized experience for each user. Information support; at the hierarchical access control level, the system implements a role-based access control model; through a unified access management center, data access permissions are precisely assigned to each role, with permission granularity refined to the data table, field, and even data row level, following the principle of minimum necessity; the system intercepts and authenticates all data requests in real time, verifying the requester's role credentials and permission scope, and strictly controlling the data content and functional modules that can be accessed and operated; at the same time, audit logs record all data access behaviors, ensuring that any unauthorized operations are traceable, thereby building a robust data security defense while ensuring efficient data distribution; The S2 AI big model core layer analyzes and processes standardized data through the emergency big model, identifies emergency event types and determines event levels, and provides accurate event judgment basis for subsequent emergency dispatch and handling. Multimodal event acquisition supports multi-source event inputs such as text, video, sensor data, and voice, and is uniformly analyzed through a large model. The large model automatically determines the event level based on dimensions such as event type, scope of impact, and casualty prediction, with accurate and efficient judgment and rapid response. Through time series analysis of sensor data and trend prediction of the large model, potential emergency risks can be warned in advance, and warning information can be pushed to the command center and relevant personnel. The emergency event identification and intelligent decision-making process is as follows: Step 1: The intelligent engine module obtains event data from the event acquisition module, providing a reliable data source and feature support for subsequent event level determination; Step 2: The data preprocessing module preprocesses the event data obtained from the event acquisition module to provide cleaned and normalized high-quality data for subsequent event level determination, and removes redundant and abnormal information to improve the accuracy and response speed of model recognition and classification. Step 3: The data feature extraction module extracts features from the preprocessed event data, providing key feature vectors and discrimination criteria for subsequent event level determination, enabling the grading model to quickly and accurately complete event grading and risk assessment. Step 4: The feature normalization module normalizes the extracted features, mapping each feature value to a standardized range of 0 to 1, eliminating the differences in the dimensions and magnitudes of different features, providing a uniform feature input for subsequent risk scoring calculations, and ensuring the accuracy and reliability of event level determination. Step 5: The risk scoring calculation module calculates the risk score based on the unified characteristics, providing a quantitative basis for subsequent event level determination and ensuring the scientific nature and consistency of the classification results. Risk scoring algorithm:

[0026] Symbol definition: : Overall risk value for the event, ranging from 0 to 1; : No. Each feature is a normalized value (0 to 1); Manually enter the number Each feature contextual weight ; Correction factor: 0 to 0.1 (add 0.1 if there are casualties / major losses); Feature weights: Normalized values ​​of monitored data: =0.30; Normalized value of regional population density: =0.25; Emergency resource distance normalized value: =0.15; Regional risk baseline: =0.10; Special population markers: =0.08; On-site damage characteristic values: =0.07; Historical event matching degree: =0.05; Event severity determination rules: 0≤ <0.3→Level 1 (General); 0.3≤ <0.6→Level 2 (relatively large); 0.6≤ <0.9→Level 3 (Serious); 0.9≤ ≤1→Level 4 (Extremely Serious); Step Six: The event level determination module classifies the event level according to the numerical range of the risk score, based on preset criteria. The range outputs emergency event levels 1 to 4, providing clear hierarchical instructions for subsequent structured outputs and business execution; Step 7: The result structured output module encapsulates the event level determination results and generates standardized structured data containing the event's unique identifier, event level, confidence level, and handling instructions, providing a unified format for subsequent network status judgment and result execution. Step 8: The network status detection module makes a real-time judgment on the current network connectivity. Depending on whether the network is normal or abnormal, it triggers either cloud-based fusion processing or local result direct use logic to ensure that the graded results are output normally in the case of network outage. Local architecture (edge / smart cabinet): Local preprocessing module: low-light enhancement, dehazing, deblurring; Local feature extraction module: input 12-dimensional emergency features; Local hierarchical engine: execution. Calculation; Local caching module: saves events and classification results when the network is disconnected; Cloud-based architecture: Feature enhancement: Deep semantic parsing of local features; Hierarchical calibration: Confidence calibration of local levels; Suggestion supplementation: Output processing suggestions and resource scheduling suggestions; Integrated decision-making structure: Final grade = Local grade (weight 0.7) + Cloud calibration (weight 0.3); Local is the primary method to ensure emergency reliability; Cloud is secondary to improve the accuracy of grade classification; API timeout / failure → automatic rollback to pure local mode; Standard Input / Processing / Output Structure: Input structure: Sensor data; Geographic location; Regional information; on-site text / video frame features; historical grading data of similar events; Processing structure: data cleaning and standardization; multi-dimensional feature extraction; weighted risk scoring; level range matching; local + cloud decision fusion; Output structure: Event ID: Unique event identifier; Classification result: Event level 1-4; Confidence level: Confidence probability of identification; Risk score: Risk assessment score; Mode switching command: Normal / emergency mode switching command; Alarm triggered: Whether an alarm is triggered; Unlock type: Smart emergency cabinet unlock type; Step 9: The result push execution module pushes the final graded results to the business layer and terminal layer, driving the execution of business processes for emergency dispatch and material matching; The core layer of the AI ​​big model combines geographic information and real-time traffic data to plan the optimal rescue route and match it with the nearest emergency teams, material reserve points and medical institutions, providing scientific route and resource matching support for the coordinated dispatch of emergency resources. The core layer of the AI ​​big data model constructs a dynamic panoramic view of the emergency situation by integrating multi-source data such as geographic information, traffic conditions, emergency resource distribution, and event details in real time. Based on this, the rescue planning is modeled as a multi-constraint optimization problem with the core objectives of minimizing overall response time, maximizing resource utilization, and maximizing route reliability. Graph optimization and resource scheduling algorithms are used for efficient solution, generating an integrated intelligent scheduling plan that includes the optimal route considering real-time traffic conditions, a precise resource matching scheme adapted to disaster needs, and a resource coordination timeline. Before the plan is issued, digital twin technology can be used for rapid simulation and evaluation to optimize it. During execution, the actual situation is continuously monitored, and replanning is dynamically triggered in case of significant deviations. Ultimately, a closed-loop intelligent scheduling system of perception-planning-execution-evaluation-adjustment is formed, providing continuously optimized and scientifically reliable decision support for the coordinated scheduling of emergency resources. The core layer of the AI ​​big data model is also configured to: achieve early warning of abnormal events based on multi-source event input and unified analysis, combined with sensor time-series data and trend prediction, and push warning information; intelligently identify and inventory emergency supplies through video recognition, automatically match supply needs according to emergency scenarios and generate supply warnings; dynamically update rescue routes based on geographic information and traffic conditions, and push standardized handling procedures based on emergency response knowledge base; analyze the entire emergency response process, identify weak links and generate strategy optimization suggestions; material video recognition: collect video streams through smart emergency cabinet cameras, and use the big data model to accurately identify various standard and non-standard emergency supplies, enabling effective identification of unlabeled materials; automatic demand matching: based on event type, impact range and disaster scale, the big data model automatically analyzes the types, quantities and allocation priorities of required emergency supplies, generates corresponding material allocation lists, and adapts to the logic of prioritizing the distribution of core materials; intelligent inventory counting: supports regular automatic inventory counting and manual triggered inventory counting, obtains actual inventory information through video recognition, and compares the big data model with system data to issue warnings for material shortages and expiration, so as to replenish and update in a timely manner; The core layer of the AI ​​big data model analyzes the event scenario, impact range, and number of affected people, calculates the types and quantities of emergency supplies needed, and generates allocation lists and dispatch instructions, providing clear instruction support for the distribution of supplies to smart emergency cabinets and the handling of rescue personnel. Rescue route planning: Combining GIS geographic information, real-time traffic data, and road damage conditions, the big data model plans the optimal rescue route, avoiding congested and dangerous areas, and dynamically updates the route plan. Emergency resource coordination: Automatically matches the nearest emergency teams, material reserve points, and medical institutions, generates resource dispatch instructions, clarifies the order of arrival and task allocation, and supports one-click issuance. Disposal plan recommendation: Based on an emergency response knowledge base trained by the big data model, it recommends standardized disposal procedures for different event types to assist on-site personnel.

[0027] S3, the real-time network communication layer, prioritizes sending dispatch instructions to intelligent emergency cabinets and rescue terminals through a full-duplex real-time communication network. At the same time, it coordinates with the normal and emergency business layers to switch to emergency mode, providing communication and mode support for the timeliness and coordination of emergency response. The real-time network communication layer includes a communication scheduling module, which is configured to: prioritize messages according to their emergency level and use priority queue scheduling; automatically establish, reconnect, and detect heartbeats for terminal connections; cache offline messages and resend them according to priority after the terminal reconnects; authenticate terminal access and encrypt transmitted data. Message priority Calculate using the following formula: in, Indicates the message priority at time t; Indicates the final event level; Indicates the instruction type level value; Indicates whether it belongs to offline resending; , , These are the weighting coefficients; Derived from the final event level determination result in S2; Values ​​are assigned based on instruction type, with material distribution instructions, cabinet door unlocking instructions, and route scheduling instructions having higher type level values ​​than ordinary query instructions; Determined by the terminal's online status and message cache status, when the message is a resend message. ,otherwise ; , , Pre-set parameters for the system to improve performance in emergency scenarios. and The weights; Network real-time communication layer according to Messages are written to the priority queue from high to low priority, and high-priority messages are sent first. After receiving a message, the terminal sends an acknowledgment status back to the platform. If the platform does not receive an acknowledgment status within a preset time window, the message is written to the offline resend queue. For messages in the offline resend queue, the network real-time communication layer resends them according to message priority after the terminal reconnects.

[0028] The routine and emergency service layer receives the final event level. Afterwards, it determines whether the conditions for switching to emergency mode are met. In emergency mode, the system cancels the normal requisition verification process and allows the smart emergency cabinet to perform one-click full opening or on-demand distribution. In normal mode, the system maintains the identity verification requisition, inventory count, equipment inspection and routine material management processes.

[0029] The network real-time communication layer constructs a full-duplex real-time communication network based on the network socket communication protocol. Combined with node service technology, it realizes high-concurrency message processing, supports hierarchical transmission of emergency instructions, heartbeat detection, automatic reconnection, and offline message resending, ensuring real-time collaboration between command centers, smart terminals, and emergency teams. It replaces the polling of traditional hypertext transmission protocols and improves communication efficiency and reliability. Hierarchical message processing prioritizes messages as follows: emergency commands have a priority of 10, data synchronization messages have a priority of 8, and ordinary query messages have a priority of 5. Priority queue scheduling: The network socket communication server uses a priority queue to process messages, with high-priority messages being executed in the queue to ensure no delay in the transmission of emergency commands. Long connection management includes automatic connection and reconnection: after the terminal starts, it automatically establishes a long connection with the server. If the connection is broken, reconnection is triggered within 1 second; if reconnection fails, a local alarm is pushed. Heartbeat detection mechanism: with a heartbeat interval of 30 seconds, the server periodically checks the terminal's online status. If no heartbeat is received for more than 90 seconds, the terminal is marked as offline, and an offline warning is pushed to the command center. Offline message retransmission and terminal offline caching: during the terminal's offline period, the server caches all sent messages and sorts them by priority. Network recovery retransmission: after the terminal reconnects, the server automatically retransmits offline messages to ensure that the terminal data is consistent with the platform and that no messages are lost. Communication security and authentication: when a terminal accesses the network, it is authenticated using a unique device identifier and a key to prevent unauthorized access. Data encryption: Message transmission uses Advanced Encryption Standard-256 encryption to prevent data tampering and theft; Flow control: Limits the message sending frequency of a single terminal to prevent malicious attacks from causing service congestion; The real-time network communication layer constructs a full-duplex real-time communication network based on the network socket protocol, combined with a node service architecture to achieve high-concurrency message processing. The communication scheduling module first performs strict identity authentication. When a terminal connects, it needs to verify the device's unique identification code and key to ensure connection security. The transmitted data is encrypted throughout the process using Advanced Encryption Standard-256. In terms of network connection management, after the terminal starts up, it automatically establishes a long connection with the server and starts a heartbeat detection mechanism, sending a heartbeat packet every 30 seconds. If the server does not receive a heartbeat within 90 seconds, it marks the terminal as offline and issues an alarm. If the connection is abnormally disconnected, a reconnection process is automatically triggered within 1 second. In terms of message processing, the system defines messages with three levels of priority: emergency instructions are set to the highest priority of 10, data synchronization messages are set to priority 8, and ordinary queries are set to priority 5. The server uses a priority queue to schedule messages. High-priority messages can be sent in real time, and all messages sent to offline terminals can be cached. After the terminal recovers the connection, they are automatically resent in priority order to ensure that critical instructions arrive with zero delay and no message loss. At the same time, the system implements a flow control strategy to limit the sending frequency of a single terminal and defend against malicious attacks. Overall, this communication layer replaces the traditional inefficient HTTP polling with the above-mentioned full-link security control, intelligent connection management and hierarchical message scheduling mechanism, providing highly available and secure communication support for the real-time reliable transmission of emergency commands, stable access of smart terminals and seamless switching between normal and emergency business modes.

[0030] S4. The intelligent emergency cabinet receives dispatch instructions and performs on-demand distribution of materials or one-click full opening. At the same time, it records the distribution of materials through visual recognition, providing real-time execution data support for system inventory updates and material traceability. Emergency response methods utilize multiple IoT terminal devices to collaboratively collect emergency data and execute commands. Specifically, these include: a smart emergency cabinet integrating a camera, electromagnetic lock, communication module, environmental sensors, a one-button alarm, and an uninterruptible power supply; enabling routine material management, emergency distribution, network outage caching and data synchronization, and supporting wide-temperature, low-power operation; real-time environmental monitoring data collection via independent IoT sensors, adjusting the collection frequency based on operating conditions, and automatically reporting early warning information when data anomalies occur; and mobile terminal applications configured for command personnel, rescue personnel, and maintenance personnel to achieve emergency command, on-site handling, and equipment operation and maintenance management, forming a collaborative linkage with the smart emergency cabinet and IoT sensors. After receiving the dispatch instruction message, the main control board of the intelligent emergency cabinet analyzes the category, compartment location and quantity of the materials to be distributed based on the target equipment identification and material list.

[0031] When the cabinet opening mode is on-demand distribution mode, the main control board only controls the electromagnetic lock of the corresponding material compartment to open; when the cabinet opening mode is one-click full opening mode, the main control board controls the opening of the preset emergency medical supplies compartments.

[0032] Before the supplies are distributed, the AI ​​visual recognition unit uses a camera inside the cabinet to capture images of the items before distribution and identifies the items. Quantity of such supplies before distribution After the item is taken and the cabinet door is closed, the AI ​​visual recognition unit re-captures the image after distribution and identifies the item. Quantity of such supplies after distribution ; Based on the results of the two identifications, the first Actual quantity of such supplies distributed Calculate using the following formula: in, Indicates the first Such supplies at all times The actual number of items distributed; Indicates the first Quantity of such supplies identified before distribution; Indicates the first The quantity of such supplies is identified after distribution; This is to prevent the actual number of distributions from becoming negative due to fluctuations in identification; The system will distribute the actual number of [items / funds]. The number of instructions issued in the scheduling instructions By comparison, the distribution deviation of the j-th type of materials is obtained. : in, Indicates the first Such supplies at all times Distribution deviation; The first one represents the requirement of the scheduling instruction. Quantity of various types of supplies distributed; This indicates the actual quantity of Category j materials distributed, as confirmed by the AI ​​visual recognition unit.

[0033] when When the value is 0, the system determines that the distribution of supplies is normal. when At that time, the system generates a record to be reviewed; when When an abnormal distribution occurs, the system generates an abnormal distribution record and pushes the abnormal distribution record to the operation and maintenance terminal and the command center terminal; Emergency measures also include a dual-mode switching mechanism for peacetime and emergency situations: Normal mode performs material management, equipment inspection and emergency drills; emergency mode performs event reception, rapid material distribution, resource coordination and on-site situation feedback; supports manual triggering, automatic triggering and hardware triggering, realizes second-level mode switching and adaptively adjusts business rules. The normal and emergency business layer is built on a server-side development framework to construct a dual-mode business logic for both normal and emergency operations. This logic integrates routine material management and emergency dispatch functions, enabling intelligent mode switching and seamless business integration, covering the entire process of routine operation and maintenance, emergency response, and post-event review. Normal mode business: Intelligent material management: Supports material inbound, outbound, inventory warning, and transfer scheduling, and automatically generates inventory reports; Equipment inspection and maintenance: Intelligent emergency cabinets perform regular self-inspections, generate inspection reports, automatically report faults and push maintenance reminders; Emergency drill management: Supports emergency drill planning, drill process recording, and drill effect evaluation, and optimizes emergency processes by analyzing drill data through large-scale models. Emergency Mode Services: Emergency Event Reception: Supports multi-channel event reporting, automatically synchronizing to the AI ​​big data model for classification; Rapid Material Distribution: After emergency mode is activated, identity verification is canceled, supporting one-click full opening of smart emergency cabinets or on-demand distribution, with AI identifying distributed materials and recording their flow; Emergency Resource Coordination: Automatically dispatches emergency teams, medical institutions, and material reserve points, pushing dispatch instructions and real-time data; Real-time On-site Situation Feedback: Real-time situation is obtained through on-site personnel's App and smart cabinet cameras, synchronized to the command center, supporting remote command; Normal / urgent mode switching: Multiple triggering methods: supports manual triggering, automatic triggering, and hardware triggering; Second-level mode switching: switching response time ≤ 5 seconds, and business rules are automatically adjusted after switching; The process of switching between normal and emergency modes is as follows: Step 1: The normal operation module maintains the system's daily operation and maintenance status, executes routine business, and provides advance preparation and operational foundation for emergency response; Step 2: The emergency event detection module monitors multi-source emergency data in real time and determines the event status, providing a basis for emergency response triggering and mode switching; Step 3: When an emergency event is detected, the emergency response trigger module initiates the entire emergency response process, providing triggering conditions for subsequent event classification and mode switching. Step 4: The AI ​​large-scale model event classification module performs multimodal data analysis and risk assessment on the triggered emergency events, and completes the automatic determination of the event level, providing a classification basis for system mode switching and dispatch instruction generation; Step 5: The emergency mode switching module switches the system to emergency mode based on the event classification results, adjusts business rules to adapt to emergency needs, and provides mode support for the rapid distribution of emergency supplies. Step Six: The scheduling instruction generation and issuance module generates scheduling instructions based on the event level and issues them with one click, providing a clear basis for the smart cabinet to perform operations; Step 7: The smart cabinet instruction execution module receives and executes the scheduling instructions, completes the distribution of materials, and updates the inventory status synchronously.

[0034] S5, the IoT terminal layer continuously collects on-site situation, equipment status and material inventory data through various terminal devices, and transmits them synchronously to the command center, providing dynamic data support for commanders to grasp the on-site situation in real time and adjust the response plan. IoT terminal layer: Integrates IoT devices such as smart emergency cabinets, IoT sensors, mobile terminals (rescue personnel App, maintenance App), and GIS terminals to realize functions such as emergency material storage, environmental monitoring, and on-site interaction, and provides the platform with data collection and command execution capabilities; Intelligent Emergency Cabinet; Hardware Integration: Integrates 4-channel MIPI cameras, 8-channel GPIO electromagnetic locks, 4G / 5G communication module, multiple types of sensors, one-button alarm, and UPS power supply; Core Functions: Supports intelligent material requisition and inventory during normal times, supports one-button full opening / on-demand distribution in emergencies, locally caches operation data during network outages, and synchronizes to the platform after network recovery, operates in a wide temperature range of -20℃ to 60℃, and consumes ≤8W; The intelligent emergency cabinet is also equipped with an AI visual recognition unit and a material lifecycle management unit for basic identification and inventory monitoring, specifically including: image acquisition, target detection, and feature recognition of materials in the cabinet to accurately identify the specifications, quantity, and condition of materials; real-time comparison of the recognition results with system inventory data to generate a real-time inventory ledger and synchronize it to the management platform; identification of materials in the cabinet that need to be replenished or replaced. The system generates replenishment reminders for materials and pushes them to relevant operation and maintenance terminals. In the event of a network outage, all identification data and inventory comparison results are cached locally and automatically synchronized to the management platform after network recovery, ensuring continuous monitoring of material status. The AI ​​visual recognition unit and the material lifecycle management unit are also used for material expiration date control and distribution priority configuration. Specifically, this includes: automatically verifying and cross-checking the expiration dates of all materials in the cabinet to accurately identify near-expiration and expired materials; classifying and marking near-expiration and expired materials, pushing tiered warning information to the management platform and operation and maintenance terminals, and automatically locking the storage cell corresponding to expired materials to prohibit distribution; dynamically configuring material distribution priorities based on the type and level of the emergency event and the needs of the on-site scenario, prioritizing the rapid retrieval of core emergency materials, and adapting to the material dispatching needs under different emergency scenarios. Internet of Things (IoT) sensors: Types covered: flame sensor, smoke sensor, vibration sensor, water level sensor, temperature and humidity sensor; Functions: Real-time collection of environmental data, automatic reporting to the platform in case of anomalies, triggering emergency warnings, data collection frequency is configurable; Mobile terminal application adopts a cross-platform development model: Command personnel terminal: View emergency event status, dispatch resources, issue instructions, and view debriefing reports; Rescue personnel terminal: Receive disposal plans, report on-site status, navigate rescue routes, and query material locations; Maintenance personnel terminal: Equipment inspection, material replenishment, fault repair, and inventory check. The emergency response approach employs a layered storage architecture, including: a relational database storing structured data such as emergency events, material information, access control data, scheduling plans, and debriefing reports, with tables partitioned by event identifier and region to improve query efficiency; a distributed caching unit caching emergency instructions, real-time inventory, equipment status, and frequently accessed data, reducing database pressure through differentiated caching durations; a terminal-embedded database for offline data storage, automatically synchronizing to the cloud after network recovery; an object storage unit storing on-site audio, video, and document files, supporting file fragmentation and access control; and a data storage layer employing a hybrid storage architecture combining structured data storage, caching and high-concurrency data, and terminal offline storage with file storage, balancing data persistence, high-concurrency access, offline caching, and file storage to ensure data reliability and access efficiency. Structured data storage (MySQL): Storage content: Structured data such as emergency event records, material inventory information, user permission data, dispatch plans, and debriefing reports; Optimized design: Tables are partitioned by "event ID + region" to improve query efficiency, support transaction operations, and ensure data consistency; Caching and High-Concurrency Data (Redis): Storage content: emergency instructions, real-time inventory, equipment online status, user tokens, and frequently accessed material and team information; Optimized design: setting differentiated cache expiration times to reduce MySQL pressure and improve response speed; Terminal offline storage (SQLite): Storage content: Operation logs, material distribution records, and sensor data of the smart emergency cabinet when it is offline; Optimized design: Automatic synchronization to MySQL after network recovery to ensure no loss of offline data; File storage (MinIO): Storage content: Emergency scene videos, pictures, and debriefing report files; Optimized design: Supports file chunk upload and download, adapts to large file storage, and provides access control.

[0035] After the supplies are distributed, the IoT terminal layer continuously collects the distribution records, compartment status, door lock status and local inventory status of the smart emergency cabinet, collects smoke concentration and temperature change data uploaded by environmental sensors, collects on-site situation feedback uploaded by rescuers' mobile terminals, and receives supplementary dispatch instructions issued by the command center terminal. For the Category of materials, updated inventory quantity Calculate using the following formula: in, Indicates the first Updated inventory quantity of this type of material; Indicates the first Inventory quantity of the same type of material before the update; Indicates the first The actual quantity of these types of supplies distributed; Indicates the first The quantity of such supplies returned, in the current fire emergency distribution scenario. .

[0036] Sourced from the local inventory ledger of the smart emergency cabinet; Derived from the visual recognition results in S4; Sourced from subsequent return or recycling records; S6, the core layer of the AI ​​big model, analyzes data from the entire emergency response process, conducts debriefing analysis, and generates debriefing reports and optimization plans, providing decision support for subsequent emergency response process iterations and resource allocation optimization; Event debriefing and summary: The big data model automatically analyzes data from the entire emergency response process, generates a debriefing report, and identifies weaknesses; Optimization plan generation: Based on the debriefing results, the big data model proposes targeted optimization suggestions and continuously iterates emergency management strategies.

[0037] The core layer of the AI ​​large model calculates and evaluates the results based on the aforementioned full-process data. : in, This indicates the post-mortem evaluation value; This indicates the response timeliness evaluation value; This represents the evaluation value for the accuracy of resource matching; This indicates the message delivery success rate evaluation value; This indicates the accuracy evaluation value for the distribution of supplies; , , , This represents the weight coefficient of each evaluation item.

[0038] It is obtained from the difference between the event trigger time and the time when the first scheduling instruction is issued; It is determined by the degree of matching between the recommended resources and the resources actually used in the end; The value is obtained from the ratio of the number of successfully confirmed messages to the total number of messages sent. This was obtained from the statistical results of the distribution deviation; , , , Preset coefficients for the system; The core layer of the AI ​​large model is based on the post-evaluation values. The system generates a review report for each of its components and outputs optimization suggestions. When the review results show that the inventory of smoke masks in a smart emergency cabinet in a certain park is insufficient for a long period of time, the optimization suggestions include increasing the baseline configuration of smoke masks at that location; when there is a delay in message delivery, the optimization suggestions include adjusting the message priority weight; when the passage efficiency of a certain rescue route is lower than the preset requirements, the optimization suggestions include adjusting the candidate route strategy in the subsequent emergency plan for that area. Through the above S1 to S6, this embodiment realizes a complete closed loop from multi-source data access, standardized integration, event identification and classification, resource matching, route planning, instruction generation, priority issuance, intelligent emergency cabinet execution, visual recognition verification, inventory traceability, network outage reissue, reconciliation and merging to review and optimization.

[0039] Example 2: Figure 4 A schematic diagram of the structure of an intelligent emergency platform system based on an AI large-scale model is provided. This intelligent emergency platform system based on an AI large-scale model is used to implement the intelligent emergency response method based on an AI large-scale model according to any one of claims 1-9, characterized in that it includes the following modules: The multimodal emergency intelligent engine module integrates four core capabilities: emergency event identification, intelligent decision-making, material identification, and solution generation, providing multi-source data processing support that is deeply adapted to emergency scenarios; The unified data fusion engine module integrates multi-source data through standardized interfaces, completing data cleaning, format standardization, and correlation analysis, providing high-quality data input for the multimodal emergency intelligent engine module; The communication module uses the Network Sockets Protocol to build a full-duplex real-time communication network. Combined with node service technology, it achieves high-concurrency message processing and priority scheduling. Through hierarchical message transmission, heartbeat detection, automatic reconnection, and offline message resending mechanisms, it ensures real-time collaboration between the terminal and the platform, realizes low-latency issuance of emergency commands and secure encrypted transmission of communications, and provides efficient communication support for stable and reliable linkage between the command center, intelligent emergency cabinet, and rescue terminal. The routine and emergency business module adopts a layered and modular architecture to build a routine and emergency dual-mode business processing logic, integrates the functions of routine material management and emergency dispatch, realizes intelligent mode switching and seamless business connection, covers the entire process of routine operation and maintenance, emergency response and post-event review, and provides efficient support for the routine and emergency business conversion and full-cycle closed-loop management. The IoT terminal layer module integrates various emergency IoT terminal devices to realize emergency material storage and management, environmental status monitoring and on-site interactive execution functions, and provides unified data collection and command execution support for the upper-layer platform; The data storage layer module adopts a hybrid storage architecture that combines relational databases, distributed caching, terminal-local embedded databases, and object storage.

Claims

1. An intelligent emergency response method based on an AI large-scale model, characterized in that, The steps include the following: S1, the multi-source data fusion layer collects multi-source emergency data through standardized interfaces, and performs fusion, cleaning and standardization processing on the collected data; S2, the core layer of the AI ​​big model, uses the emergency big model to analyze and process standardized data, identify emergency event types and determine event levels; The core layer of the AI ​​big model combines geographic information and real-time traffic data to plan the optimal rescue route and match the nearest emergency teams, material reserve points and medical institutions. The core layer of the AI ​​big model analyzes the event scenario, the scope of impact, and the number of people affected, calculates the types and quantities of emergency supplies needed, and generates allocation lists and dispatch instructions. S3, the real-time network communication layer uses a full-duplex real-time communication network to prioritize sending dispatch instructions to the intelligent emergency cabinet and rescue terminal, while coordinating with the normal and emergency business layers to switch to emergency mode. S4. The intelligent emergency cabinet receives dispatch instructions and performs on-demand distribution of supplies or one-click full opening operation, while recording the distribution of supplies through visual recognition. S5, the IoT terminal layer continuously collects on-site situation, equipment status and material inventory data through various terminal devices and transmits them synchronously to the command center; The S6 AI big model core layer analyzes data from the entire emergency response process, conducts post-mortem analysis, and generates post-mortem reports and optimization plans.

2. The intelligent emergency response method based on an AI large model according to claim 1, characterized in that, The multi-source data fusion layer includes a data management module, which is configured as follows: Perform interface adaptation, cleaning, transformation, and quality verification on multi-source heterogeneous data to achieve data standardization; Perform spatial, temporal, and semantic correlations on emergency data to complete multi-dimensional data fusion analysis; Data is distributed on demand based on user roles and emergency scenarios, and secure access is achieved through hierarchical permission control.

3. The intelligent emergency response method based on an AI large model according to claim 1, characterized in that, The core layer of the AI ​​large model is also configured as follows: Based on multi-source event input and unified analysis, combined with sensor time series data and trend prediction, early warning of abnormal events is achieved, and warning information is pushed out. The system uses video recognition to intelligently identify and inventory emergency supplies, automatically matching supply needs and generating supply warnings based on emergency scenarios. The rescue routes are dynamically updated by combining geographic information and traffic conditions, and standardized response procedures are pushed out based on the emergency response knowledge base. The entire emergency response process is analyzed to identify weaknesses and generate suggestions for strategy optimization.

4. The intelligent emergency response method based on an AI large model according to claim 1, characterized in that, The network real-time communication layer includes a communication scheduling module, which is configured to: Messages are prioritized according to their emergency level and scheduled using priority queues. Automatic connection establishment, disconnection reconnection, and heartbeat detection for terminal connections; Offline messages are cached and resent according to priority after the terminal reconnects; Authentication is performed on terminal access, and data transmission is encrypted for protection.

5. The intelligent emergency response method based on an AI large model according to claim 1, characterized in that, The emergency response method also includes a dual-mode switching mechanism for both normal and emergency situations: In normal operation mode, the tasks include materials management, equipment inspection, and emergency drills. Emergency mode enables the reception of events, rapid distribution of supplies, resource coordination, and on-site situation feedback. It supports manual triggering, automatic triggering, and hardware triggering, enabling second-level mode switching and adaptive adjustment of business rules.

6. The intelligent emergency response method based on an AI large model according to claim 1, characterized in that, The emergency response method utilizes multiple IoT terminal devices in collaboration to achieve emergency data collection and command execution, specifically including: The intelligent emergency cabinet integrates a camera, electromagnetic lock, communication module, environmental sensor, one-button alarm and uninterruptible power supply, and can realize normal material management, emergency distribution, network outage caching and data synchronization, and supports wide temperature and low power consumption operation. Environmental monitoring data is collected in real time through independent IoT sensors, the collection frequency is adjusted according to the operating conditions, and early warning information is automatically reported when the data is abnormal. Through mobile terminal applications configured for command personnel, rescue personnel, and maintenance personnel respectively, emergency command, on-site handling, and equipment operation and maintenance management can be realized, forming a collaborative linkage with smart emergency cabinets and IoT sensors.

7. The intelligent emergency response method based on an AI large model according to claim 1, characterized in that, The emergency response method employs a tiered storage architecture, including: A relational database stores structured data such as emergency events, material information, access control data, dispatch plans, and debriefing reports, and is divided into tables by event identifier and region to improve query efficiency; Distributed caching units cache emergency instructions, real-time inventory, device status, and frequently accessed data, reducing database pressure through differentiated caching expiration times; A local embedded database on the terminal enables offline data storage and automatically synchronizes to the cloud after the network is restored; The object storage unit stores on-site audio, video, and document files, and supports file fragmentation and access control.

8. The intelligent emergency response method based on an AI large model according to claim 1, characterized in that, The intelligent emergency cabinet is also equipped with an AI visual recognition unit and a materials lifecycle management unit for basic identification and inventory monitoring, specifically including: The system performs image acquisition, target detection, and feature recognition on the materials inside the cabinet to accurately identify the specifications, quantity, and condition of the materials. The identification results are compared with the system inventory data in real time to generate a real-time inventory ledger, which is then synchronized to the management platform. Identify materials in the cabinet that need to be replenished or replaced, generate replenishment reminders, and push them to relevant maintenance terminals; In the offline state, all identification data and inventory comparison results are cached locally and automatically synchronized to the management platform after the network is restored, ensuring the continuity of material status monitoring.

9. The intelligent emergency response method based on an AI large model according to claim 8, characterized in that, The AI ​​visual recognition unit and the material lifecycle management unit are also used for material validity period control and distribution priority configuration, specifically including: Automatically verify and cross-check the expiration dates of all materials in the cabinet to accurately identify near-expiration and expired materials; Near-expiry and expired materials are classified and marked, and early warning information is pushed to the management platform and operation and maintenance terminal in a hierarchical manner. At the same time, the storage cell corresponding to the expired materials is automatically locked to prevent them from being issued. Based on the type and level of the emergency and the needs of the on-site scenario, the priority of material distribution is dynamically configured to ensure the rapid retrieval of core emergency materials and adapt to the material dispatching needs under different emergency scenarios.

10. An intelligent emergency response platform system based on an AI large-scale model, used to implement the intelligent emergency response method based on an AI large-scale model as described in any one of claims 1-9, characterized in that, Includes the following modules: The multimodal emergency intelligent engine module integrates four core capabilities: emergency event identification, intelligent decision-making, material identification, and solution generation, providing multi-source data processing support that is deeply adapted to emergency scenarios; The unified data fusion engine module integrates multi-source data through standardized interfaces, completing data cleaning, format standardization, and correlation analysis, providing high-quality data input for the multimodal emergency intelligent engine module; The communication module uses the network socket protocol to build a full-duplex real-time communication network. Combined with node service technology, it realizes high-concurrency message processing and priority scheduling. Through hierarchical message transmission, heartbeat detection, automatic reconnection and offline message resending mechanism, it ensures real-time collaboration between the terminal and the platform, realizes low-latency issuance of emergency commands and secure encrypted transmission of communication, and provides efficient communication support for stable and reliable linkage between the command center, intelligent emergency cabinet and rescue terminal. The routine and emergency business module adopts a layered and modular architecture to build a routine and emergency dual-mode business processing logic, integrates the functions of routine material management and emergency dispatch, realizes intelligent mode switching and seamless business connection, covers the entire process of routine operation and maintenance, emergency response and post-event review, and provides efficient support for the routine and emergency business conversion and full-cycle closed-loop management. The IoT terminal layer module integrates various emergency IoT terminal devices to realize emergency material storage and management, environmental status monitoring and on-site interactive execution functions, and provides unified data collection and command execution support for the upper-layer platform; The data storage layer module adopts a hybrid storage architecture that combines relational databases, distributed caching, terminal-local embedded databases, and object storage.

Images