Emergency rescue method and system based on drone and urban three-dimensional platform

Abstract

The invention relates to the technical field of digital simulation, and discloses an emergency rescue method and system based on unmanned aerial vehicles and a three-dimensional urban platform. The invention retrieves an emergency rescue regional model through the three-dimensional urban platform, drives a formation of unmanned aerial vehicles to the emergency region to obtain real-time regional information, utilizes a pre-trained accident situation image recognition algorithm, identifies accident situation characteristics such as fireworks, firefighting passage blockage, safety exit blockage, urban waterlogging, dangerous slopes, dangerous building detection, illegal construction on building roofs, river floating object identification, and high-voltage tower tilt monitoring, substitutes the accident situation characteristics into a regional three-dimensional model to generate an accident simulation model, provides real-time accident situation feedback through the accident simulation model, assists rescue personnel in making scientific rescue decisions, improves the detection efficiency of unmanned aerial vehicles through comprehensive analysis and precise simulation of the emergency rescue region, enhances the timeliness and precision of rescue, and solves the problem of low efficiency in manually collecting accident status in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of digital simulation, and in particular to an emergency rescue method and system based on an unmanned aerial vehicle and a three-dimensional urban platform. Background Art

[0002] With the development of big data, artificial intelligence and sensing technology, urban three-dimensional platforms have gradually become core tools for emergency rescue and disaster management. Urban three-dimensional platforms integrate multi-dimensional data such as geographic information, building models, and transportation networks, and can provide detailed spatial information support for emergency command systems. These platforms can achieve panoramic monitoring of disaster sites and help rescue personnel quickly grasp the situation on site.

[0003] In the existing technology, emergency rescue work can be guided by the urban three-dimensional platform, and the accuracy and efficiency of emergency rescue work can be improved by combining and analyzing the information fed back by the urban three-dimensional platform and the real-time conditions obtained on site. At present, the collection and combination of real-time conditions on site rely on manual analysis, which leads to low efficiency. Summary of the invention

[0004] The purpose of the present invention is to provide an emergency rescue method and system based on an unmanned aerial vehicle and a three-dimensional urban platform, aiming to solve the problem of low efficiency in manually collecting the current status of an accident in the prior art.

[0005] The present invention is implemented in this way. In a first aspect, the present invention provides an emergency rescue method based on a drone and a three-dimensional urban platform, comprising:

[0006] Acquire the location information of the emergency rescue area in the urban three-dimensional platform, and retrieve the regional model and analyze the detection plan of the urban three-dimensional platform according to the location information to obtain the regional three-dimensional model and drone detection plan of the emergency rescue area;

[0007] According to the drone detection scheme, drive the drone formation to the emergency rescue area to perform regional detection work to obtain real-time regional information of the emergency rescue area;

[0008] Performing image recognition of the accident situation on the real-time regional information through a pre-trained accident situation image recognition algorithm to obtain accident situation characteristics corresponding to the real-time regional information; wherein the accident situation characteristics include fireworks, fire blocking conditions, emergency exit blocking conditions, urban waterlogging conditions, dangerous slope conditions, dangerous building conditions, illegal construction conditions on the roof of a building, floating objects in a river, and tilted high-voltage tower conditions;

[0009] Substitute the accident condition characteristics into the regional three-dimensional model, and let the regional three-dimensional model simulate the accident condition according to the accident condition characteristics, so as to obtain an accident simulation model for digital feedback of the accident condition in the emergency rescue area; wherein, the accident simulation model is used to provide real-time simulation feedback of the accident condition in the emergency rescue area, so as to assist rescue personnel in making rescue decisions based on the accident condition.

[0010] In a second aspect, the present invention provides an emergency rescue system based on a drone and a three-dimensional urban platform, which is used to implement an emergency rescue method based on a drone and a three-dimensional urban platform as described in any one of the first aspects.

[0011] The present invention provides an emergency rescue method based on a drone and a three-dimensional urban platform, which has the following beneficial effects:

[0012] The present invention retrieves the emergency rescue area model through the urban three-dimensional platform, analyzes the detection plan, drives the drone formation to the emergency area to obtain real-time area information according to the detection plan, obtains real-time area information and uses the pre-trained accident situation image recognition algorithm to identify accident situation characteristics such as fireworks, fire, fire passage blockage, safety exit blockage, urban waterlogging, dangerous slopes, dangerous building detection, illegal construction on the roof of a building, river floating object identification, high-voltage tower inclination monitoring, etc., substitutes the accident situation characteristics into the regional three-dimensional model, performs simulation processing, generates an accident simulation model, provides real-time accident situation feedback through the accident simulation model, assists rescue personnel in making scientific rescue decisions, improves drone detection efficiency through comprehensive analysis and accurate simulation of the emergency rescue area, enhances the timeliness and accuracy of rescue, and solves the problem of low efficiency of manual collection of accident status in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a schematic diagram of the steps of an emergency rescue method based on a drone and a three-dimensional urban platform provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0015] The implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0016] Reference Figure 1 As shown, a preferred embodiment of the present invention is provided.

[0017] In a first aspect, the present invention provides an emergency rescue method based on a drone and a three-dimensional urban platform, comprising:

[0018] S1: Acquire the location information of the emergency rescue area in the urban three-dimensional platform, and retrieve the regional model and analyze the detection plan of the urban three-dimensional platform according to the location information to obtain the regional three-dimensional model and drone detection plan of the emergency rescue area;

[0019] S2: driving the drone formation to the emergency rescue area to perform regional detection according to the drone detection scheme to obtain real-time regional information of the emergency rescue area;

[0020] S3: performing image recognition of the accident condition on the real-time regional information through a pre-trained accident condition image recognition algorithm to obtain accident condition features corresponding to the real-time regional information; wherein the accident condition features include fireworks conditions, fire blocking conditions, emergency exit blocking conditions, urban waterlogging conditions, dangerous slopes, dangerous building detection, illegal construction on the roof of a building, river floating object recognition, and high-voltage tower tilt monitoring conditions;

[0021] S4: Substitute the accident condition characteristics into the regional three-dimensional model, and let the regional three-dimensional model simulate the accident condition according to the accident condition characteristics, so as to obtain an accident simulation model for digital feedback of the accident condition in the emergency rescue area; wherein, the accident simulation model is used to perform real-time simulation feedback of the accident condition in the emergency rescue area, so as to assist rescue personnel in making rescue decisions based on the accident condition.

[0022] Specifically, in step S1 of the embodiment provided by the present invention, the geographical location of the emergency rescue area is obtained from the urban three-dimensional platform. The location data usually comes from the GIS (geographic information system) platform, remote sensing technology (satellite images, drone data), and the positioning information of IoT (Internet of Things) devices. Precise positioning technology (such as GPS, RTK) is used to confirm the specific coordinates of the emergency area and ensure the accuracy and real-time nature of the information.

[0023] More specifically, the spatial scope of the emergency area is obtained and calibrated, factors that affect rescue (such as traffic congestion areas, dangerous buildings or fire areas, etc.) are identified, the accurate positioning of the emergency area is ensured, the accuracy of subsequent data processing and model analysis is improved, and real-time geographic information is provided to help decision makers quickly assess the scope of the accident.

[0024] More specifically, based on the positioning information, a detailed three-dimensional model of the target area is extracted from the urban three-dimensional platform, which contains detailed three-dimensional data of buildings, roads, underground facilities, green spaces, water bodies, etc. If the three-dimensional model is missing, the existing remote sensing data (such as lidar data) and images taken by drones are combined with automated modeling tools to build or improve the three-dimensional model.

[0025] More specifically, in combination with emergency rescue needs, a reasonable detection plan is formulated, which includes: selection and path planning of drone formations: planning the flight route and task allocation of drone formations based on the size, danger and air traffic restrictions of the rescue area; configuration of detection equipment: selecting appropriate drone-mounted equipment (such as thermal imaging cameras, lidar, gas sensors, high-definition cameras, etc.) according to the type of emergency (such as fire, toxic gas leak, etc.); allocation of detection plans: taking into account the terrain, building density and rescue priority in the area, dividing the detection area to ensure that the drone's detection mission can cover all key areas.

[0026] It is understandable that through high-precision three-dimensional modeling and data extraction, it is ensured that the three-dimensional regional model can truly reflect the actual situation of the emergency area, avoid misjudgment due to incomplete or inaccurate data, optimize the path planning and equipment configuration of the drone according to geographical and environmental conditions, improve the accuracy and coverage of detection, and ensure that the drone detection mission can cover key areas to the greatest extent by simulating and analyzing the detection plan in advance, reducing the waste of resources caused by aerial obstacles or complex terrain.

[0027] Specifically, in step S2 of the embodiment provided by the present invention, based on the preliminary detection plan analysis, the UAV formation task scheduling system will reasonably divide the UAV team according to factors such as the scale and complexity of the emergency area and the priority of the flight mission. The formation task system will assign the UAVs to different mission areas, and each UAV will be assigned a specific task based on its load capacity, flight range and sensor configuration.

[0028] More specifically, through the formation collaboration algorithm, drone teams maintain a safe flying distance and coordination to ensure the optimization of flight paths, avoid aerial conflicts, and achieve full coverage of the target area. The flight paths of drones are planned according to the actual conditions of the target area (such as building density, airspace restrictions, obstacles, etc.). These paths will be dynamically adjusted based on the pre-collected three-dimensional urban model data, meteorological data (such as wind speed, wind direction, temperature, etc.) and real-time traffic or other flight restrictions.

[0029] More specifically, after receiving instructions from the command center, the drone takes off through the automation system, starts the autonomous navigation mode, and flies along the preset path. Through the drone formation task scheduling system and collaborative algorithm, the task execution efficiency of multiple drones is improved, air interference and task duplication are avoided, and environmental factors are fully considered to ensure that drones can fly accurately in complex urban environments and avoid flight interruptions or collisions.

[0030] More specifically, during the flight, the drone begins to conduct real-time detection of the emergency area. According to the preset mission requirements, it is equipped with appropriate sensors (such as infrared cameras, high-definition cameras, lidar, gas sensors, etc.) to collect data. The collected data includes: Image data: high-definition images and videos, used to monitor fires, building damage, personnel activities, etc. Thermal imaging data: used to detect fire hotspots, abnormal temperature areas, etc. Gas sensor data: detect leakage of harmful gases (such as carbon monoxide, ammonia, etc.) LiDAR data: used to measure building structure, road conditions and terrain features.

[0031] More specifically, the drone will transmit the collected real-time data back to the command center through the wireless communication network in real time. The data will be transmitted through an encrypted channel to ensure the security and reliability of the data. After receiving the data, the command center will conduct preliminary analysis and visualization, and adjust the detection strategy (such as adjusting the flight path, re-planning task allocation, etc.) according to the feedback information. The drone uses different types of sensors to achieve multi-dimensional and all-round data collection, which improves the accuracy and breadth of disaster monitoring. The data collected by the drone can be transmitted back to the command center in real time, ensuring that decision makers can quickly obtain information and respond, thereby improving the speed of emergency response.

[0032] More specifically, the command center analyzes the images, thermal imaging, gas detection data, etc. sent back by the drone to understand the conditions in the emergency area in real time. For example, it can identify the fire spread area, the degree of damage to buildings, the location of pollution sources, etc., and use artificial intelligence, machine learning and other algorithms to process and analyze real-time data, automatically identify the severity of the disaster, changing trends, etc., and provide data support for the next emergency measures. Based on the analysis results, the command center can dynamically adjust the drone's mission. If new dangerous areas or details that need attention appear, the task scheduling system can immediately re-plan the drone's flight path and task assignment. If certain areas require more precise detection, the command center can assign specific drones to collect data with higher resolution.

[0033] More specifically, the real-time detection data collected by the drone (such as thermal imaging, images, gas sensor data, etc.) is combined with the original three-dimensional regional model of the city for data fusion. By matching the collected sensor data with the regional three-dimensional model, a dynamically updated three-dimensional environmental model is formed, which can more accurately reflect the disaster situation and regional changes, and generate real-time updated emergency scenarios on the three-dimensional platform, providing emergency command personnel with an intuitive three-dimensional view to help judge the development of the disaster, resource distribution, and rescue routes. The dynamic three-dimensional model can display the spread of fire and pollution, and help command personnel reasonably plan rescue routes.

[0034] It is understandable that through the real-time updated three-dimensional model, the command center can intuitively understand the changes in the emergency area, improve the accuracy and response speed of command, integrate drone detection data with the three-dimensional regional model, improve the comprehensiveness of the data and the depth of analysis, and provide efficient and accurate emergency decision-making support through multi-dimensional detection data and dynamic three-dimensional models, and ensure rapid response in the ever-changing disaster situation.

[0035] Specifically, in step S3 of the embodiment provided by the present invention, the accident area is monitored in real time using cameras, sensors, drones and other equipment. The image data source can be a static image or a video stream, and a deep learning model (such as a convolutional neural network CNN) or an integrated algorithm (such as YOLO, Faster R-CNN, etc.) is used to analyze the image and identify specific accident features, including:

[0036] Fire and smoke condition recognition: By identifying visual features such as flames and smoke, determine whether a fire has occurred and its scale. Fire passage blockage recognition: Use object detection algorithms to identify whether there are obstacles (such as vehicles, collapsed objects, etc.) in the fire passage. Emergency exit blockage recognition: Check whether the emergency exits in the area are blocked or blocked to ensure that the evacuation channels are unobstructed. Urban waterlogging recognition: Use water area detection and water flow direction analysis to identify the severity and scope of urban waterlogging. Dangerous slope recognition: Through the detection of slope morphology and changes in the image, identify whether there are signs of dangerous slope landslides or collapses.

[0037] Specifically, in step S4 of the embodiment provided by the present invention, the accident condition features identified from the image are extracted, including: fire area, smoke concentration, flame size and other smoke and fire condition features, the type and location of obstacles blocking the fire escape, whether the safety exit is blocked and the degree of blockage, the depth of waterlogging, the scope of water accumulation, the stability of dangerous slopes and the potential risk of landslides, and these features are digitized and recorded.

[0038] More specifically, a three-dimensional digital model of the emergency rescue area (such as three-dimensional spatial data of buildings, roads, fire passages, exits, etc.) is constructed. The model is usually generated using GIS data, CAD files or BIM models, and has accurate geographic and spatial structural information. The identified accident status characteristics are input as parameters into the three-dimensional regional model. For example: the fireworks status is parameterized as the location of the fire source, the speed of flame spread, etc., the congestion status is input as the specific location, size, type, etc. of the blocking object, the waterlogging status is input as the water depth, flow direction, etc., and the inclination angle of the dangerous slope, the landslide area and other information are also input into the three-dimensional model.

[0039] More specifically, based on the input accident condition characteristics, the three-dimensional model will simulate the accident. The simulation process includes: Fire spread simulation: According to the location of the fire source and the law of fire spread, simulate the spread of fire and the impact range of smoke, Evacuation and rescue simulation: Simulate the evacuation routes, rescue paths, obstruction handling, etc. of personnel under different accident situations, Flood simulation: Simulate the changes in water flow caused by waterlogging, the impact of water depth and flow rate on the region, determine the flooded area and safe passage, Slope landslide simulation: By simulating the stability of the slope, predict the potential location and impact range of the landslide, the results of the accident simulation process will be fed back in real time in the three-dimensional model, providing intuitive simulation images and data analysis, showing the trend of accident development and possible affected areas.

[0040] It is understandable that the accident simulation results are combined with real-time data to form a dynamic decision support system. Rescuers can use this system to understand the accident situation, personnel evacuation status, and the smoothness of fire passages in real time. The system will optimize the emergency rescue plan based on the simulation feedback data and indicate the best action routes and strategies for fire brigades, medical personnel, and rescuers.

[0041] The present invention provides an emergency rescue method based on a drone and a three-dimensional urban platform, which has the following beneficial effects:

[0042] The present invention retrieves the emergency rescue area model through the urban three-dimensional platform, analyzes the detection plan, drives the drone formation to the emergency area to obtain real-time area information according to the detection plan, obtains real-time area information and uses the pre-trained accident situation image recognition algorithm to identify accident situation characteristics such as fireworks, fire, fire passage blockage, safety exit blockage, urban waterlogging, dangerous slopes, dangerous building detection, illegal construction on the roof of a building, river floating object identification, high-voltage tower inclination monitoring, etc., substitutes the accident situation characteristics into the regional three-dimensional model, performs simulation processing, generates an accident simulation model, provides real-time accident situation feedback through the accident simulation model, assists rescue personnel in making scientific rescue decisions, improves drone detection efficiency through comprehensive analysis and accurate simulation of the emergency rescue area, enhances the timeliness and accuracy of rescue, and solves the problem of low efficiency of manual collection of accident status in the prior art.

[0043] Preferably, the steps of obtaining the positioning information of the emergency rescue area in the urban three-dimensional platform, and retrieving the regional model and analyzing the detection scheme of the urban three-dimensional platform according to the positioning information to obtain the regional three-dimensional model of the emergency rescue area and the drone detection scheme include:

[0044] S11: Acquire the location information of the emergency rescue area in the urban three-dimensional platform;

[0045] S12: positioning the city 3D platform in a designated area according to the positioning information, so as to mark a core positioning area corresponding to the emergency rescue area on the city 3D platform, and scheduling a regional model of the city 3D platform according to the core positioning area, so as to obtain a core 3D model corresponding to the core positioning area;

[0046] S13: Based on the core positioning area, a multi-dimensional correlation hierarchical analysis is performed on the urban three-dimensional platform to obtain the multi-dimensional correlation characteristics of the core positioning area on the urban three-dimensional platform, and model characteristics of the urban three-dimensional platform are extracted and scheduled according to the multi-dimensional correlation characteristics to obtain an additional three-dimensional model, and the additional three-dimensional model is attached to the core three-dimensional model to obtain the regional three-dimensional model; wherein the multi-dimensional correlation hierarchy includes a key building distribution hierarchy, a surrounding traffic road hierarchy, and an underground pipeline network hierarchy;

[0047] S14: Obtaining formation structure information and current position information of the UAV formation;

[0048] S15: Analyze and process the detection nodes of the regional three-dimensional model to obtain the detection node distribution characteristics, and perform node allocation processing on the detection node distribution characteristics according to the formation structure information to obtain a detection operation trajectory set corresponding to the UAV formation; wherein the detection operation trajectory set includes a plurality of detection operation trajectories, and the detection operation trajectory is used to describe the operation trajectory of the UAVs in the UAV formation for detecting multiple detection nodes;

[0049] S16: Calculating the flight path of the detection trajectory set to reach the initial node according to the current position information of the UAV formation, so as to obtain the initial flight path set of the UAV formation;

[0050] S17: Combining the detection operation trajectory set and the initial flight path set of the drone formation to obtain a drone detection plan for the drone formation.

[0051] Specifically, through the positioning system of the urban three-dimensional platform (such as GPS, base station positioning or Internet of Things data, etc.), the geographic coordinates or regional boundary information of the emergency rescue area can be obtained in real time, and high-precision positioning technology can be used to ensure that the obtained regional positioning data has high accuracy and timeliness. Through high-precision positioning technology, the precise location of the emergency rescue area is ensured, accurate data support is provided, and the location information of the emergency rescue area is updated in real time, which facilitates timely adjustment of the emergency response plan under dynamic changes.

[0052] More specifically, based on the positioning information obtained, the geographic information system (GIS) function of the urban three-dimensional platform is used to accurately locate the designated area, mark the core positioning area corresponding to the emergency rescue area in the urban three-dimensional model, and clearly mark the boundaries and important nodes of the area. Based on the information of the core positioning area, the three-dimensional model of the relevant area is retrieved from the urban three-dimensional platform database to ensure that the three-dimensional information of the core area, such as buildings, roads, and underground facilities, is accurately presented. Through precise positioning and area marking, the key areas for emergency rescue are identified, and relevant models are retrieved in the three-dimensional platform to achieve digital and three-dimensional display of the core area, supporting subsequent analysis and decision-making.

[0053] More specifically, a multi-dimensional analysis is conducted on the core positioning area, including the following levels: key building distribution level: analysis of the location and structure of important buildings and facilities in the core area to determine the key targets of rescue, surrounding traffic road level: analysis of the surrounding road network and its carrying capacity, assessment of traffic smoothness, auxiliary evacuation and resource allocation, underground pipeline network level: analysis of the distribution of underground pipelines (such as gas, water, electricity, etc.) and possible accident risks to avoid secondary disasters.

[0054] More specifically, various features such as buildings, transportation, and pipelines are extracted based on the analysis results, and model feature scheduling is performed to enhance the details and diversity of the three-dimensional model. Additional three-dimensional models (such as transportation facilities, underground pipelines, etc.) are added to the core three-dimensional model to construct a complete regional three-dimensional model. Through multi-dimensional analysis, more comprehensive spatial layout information is provided, which helps to evaluate the feasibility of different rescue plans. The additional three-dimensional models provide more regional details to help formulate more accurate emergency rescue plans.

[0055] More specifically, the structural information (such as quantity, configuration, flight altitude, etc.) and current location information of the drone formation are obtained to ensure that the detection tasks can be reasonably allocated according to the formation capabilities, and the detection nodes, that is, the key locations that the drones need to cover, are determined in the regional three-dimensional model. The detection nodes may include high-risk areas, key buildings, traffic arteries, environmental monitoring points, etc. The distribution characteristics of the detection nodes, such as the distance between nodes, priority, coverage, etc., are analyzed to ensure the rationality of task allocation.

[0056] More specifically, nodes are allocated based on the structural information of the UAV formation and the distribution characteristics of the detection nodes to ensure that each node can be effectively covered by the UAV. A set of detection operation trajectories is generated, including the flight path and detection sequence of each UAV, to ensure the efficient completion of the detection task.

[0057] More specifically, based on the current position information of the UAV formation, the flight path starting from the initial node is calculated, and factors such as flight time, path efficiency, and obstacle avoidance requirements are taken into account. The flight path of the formation is optimized by combining the detection trajectory and the initial path calculation results, and the final UAV detection plan is generated to ensure that the UAV covers all key detection nodes and avoids missing important locations. Through the path optimization algorithm, the detection efficiency of the UAV is improved, and unnecessary flight distance and time waste are reduced. The reasonable combination of formation flight and node allocation ensures the collaborative work of multiple UAVs and improves the emergency response speed.

[0058] More specifically, according to the optimized UAV detection plan, the UAV's flight mission is executed, real-time data transmission and feedback are carried out, the UAV's flight status and detection results are monitored in real time through the urban three-dimensional platform, the flight path and task allocation are adjusted, and the UAV mission is monitored and adjusted in real time to ensure that emergency response in complex environments is effective. The real-time detection of UAVs can quickly provide data support for key areas and provide timely information for decision makers.

[0059] Preferably, the step of driving a drone formation to the emergency rescue area to perform regional detection according to the drone detection scheme to obtain real-time regional information of the emergency rescue area includes:

[0060] S21: driving the drone formation to fly to the position of the corresponding initial detection node in the emergency rescue area according to the initial flight path set in the drone detection scheme, so that the drones in the drone formation collect information from the detection node through the sensor group carried by them, so as to obtain the node detection information of the detection node; wherein the sensor group includes a camera sensor, an infrared sensor and a laser sensor, and the node detection information includes camera information, thermal imaging information and laser information;

[0061] S22: driving the drone formation to fly to each of the detection nodes in sequence according to the detection operation trajectory set in the drone detection scheme, and collecting information from the detection nodes through the onboard sensor group when at each of the detection nodes, so as to obtain node detection information of each of the detection nodes;

[0062] S23: The node detection information of each detection node together constitutes the real-time area information of the emergency rescue area;

[0063] S24: After completing a round of node information detection corresponding to the detection trajectory set, the detection trajectory set is subjected to trajectory correction processing according to the acquired real-time area information to obtain a subsequent detection trajectory set, and the UAV formation is driven to perform cyclic node detection on the emergency rescue area according to the subsequent detection trajectory set to update the real-time area information in real time.

[0064] Specifically, according to the initial flight path set in the drone detection plan, the drones in the formation are driven to the initial detection nodes in the emergency rescue area. The initial flight path set contains the flight paths from the starting point to each detection node, ensuring that the drone formation can effectively cover the predetermined detection positions.

[0065] More specifically, once the drones arrive at the designated initial detection node, the drones in the formation will use the onboard sensor group (camera sensor, infrared sensor, laser sensor) to start node information collection to obtain node detection information. The node detection information includes: camera information: high-quality images acquired by camera sensors for target recognition and visual analysis, thermal imaging information: thermal imaging data captured by infrared sensors for detecting temperature differences and identifying abnormal heat sources or living things, laser information: laser radar (LiDAR) data collected by laser sensors for precise terrain scanning and obstacle detection.

[0066] It can be understood that the initial flight path set ensures that the UAV can fly to the designated node accurately in a predetermined order and obtain target data. Through the combination of multiple sensors (camera, infrared, laser), it can comprehensively obtain the environmental information of the detection node, improve the multidimensionality and accuracy of the data, and enhance the perception of the area.

[0067] More specifically, according to the set of detection operation trajectories in the drone detection plan, the drone formation flies to the position of each detection node in turn. Every time it arrives at a new detection node, the drones in the formation will collect information again through the sensor group to obtain the node detection information of the node. The node detection information includes camera, thermal imaging and laser information, providing real-time regional information for the entire emergency rescue area. At each detection node, after the drone completes information collection, it transmits the data back to the ground command system for subsequent analysis and decision-making. The drone formation efficiently covers all key detection nodes according to the preset trajectory, ensuring that all important locations in the area can be monitored in real time. The detection information of each node is uploaded to the command system in real time, providing immediate support for decision-making and enhancing emergency response capabilities.

[0068] More specifically, after completing a round of detection, the real-time regional information of the entire emergency rescue area is obtained by analyzing the node detection information collected by each detection node. The real-time regional information may include environmental changes in the area, the presence of targets, potential dangerous areas, etc. According to the real-time regional information, the detection operation trajectory set is trajectory corrected so that subsequent detection tasks can more accurately cover the changed areas or incompletely covered parts. The corrected trajectory set is used to adjust the flight path of the UAV to ensure the comprehensiveness and timeliness of regional detection. After the trajectory correction, the UAV formation will perform cyclic node detection according to the new detection operation trajectory set, and continuously update the real-time regional information to ensure the real-time and accuracy of the data. After each cyclic detection, the real-time regional information is updated and transmitted back to the command system to facilitate emergency response and decision-making.

[0069] It is understandable that through real-time feedback of regional information, the flight trajectory can be dynamically corrected to ensure that the UAV can respond to environmental changes and supplement detection blind spots. Through cyclic detection and real-time updates, it is ensured that the information in the emergency rescue area is always up to date, enhancing the effectiveness of emergency response. Data-driven decision support: Each detection and data update provides an accurate reference for emergency decision-making, helping the command system to adjust strategies and deploy resources in a timely manner.

[0070] More specifically, drone formations work together to assign tasks and coordinate flight paths to ensure that multiple drones can perform tasks on different detection nodes at the same time. The sensor data fusion of multiple drones can improve the efficiency and accuracy of detection and ensure coverage of a wider area. The AI ​​algorithm analyzes the data collected in real time, automatically identifies important changes or potential risks in the area, and optimizes subsequent detection paths and task allocation. The system can also adjust the drone's flight altitude, speed and other parameters based on real-time feedback to optimize the detection effect.

[0071] It is understandable that formation drones work together to improve area coverage, multiple sensor data fusion enhances the accuracy and reliability of information, intelligent algorithms optimize detection solutions, and real-time feedback and decision support enhance the flexibility and timeliness of emergency response.

[0072] Preferably, the step of pre-training the accident situation image recognition algorithm includes:

[0073] S31: Construct a convolutional neural network model consisting of an input layer, a convolutional layer, a pooling layer, several fully connected layers, and an output layer;

[0074] S32: preparing training data for model training of the convolutional neural network model; wherein the training data includes drone detection data and accident condition characteristics of various emergency rescue accidents, the drone detection data is image data detected by the drone formation of the emergency rescue area, and the accident condition characteristics are accident conditions corresponding to the drone detection data, and the accident conditions include fireworks conditions, fire blocking conditions, emergency exit blocking conditions, urban waterlogging conditions, dangerous slopes, dangerous building detection, illegal construction on the roof of a building, river floating object identification, and high-voltage tower tilt monitoring conditions;

[0075] S33: Substituting the training data into the input layer;

[0076] S34: the input layer receives the training data, and transmits the training data to the convolution layer, and the convolution layer is used to collect spatiotemporal correlation features of the training data to obtain the spatiotemporal correlation features of the training data;

[0077] S35: the pooling layer is used to perform a pooling operation on the spatiotemporal correlation features of the training data extracted by the convolution layer to reduce the size of the spatiotemporal correlation features of the training data and retain key features of the spatiotemporal correlation features of the training data;

[0078] S36: Each of the fully connected layers is used to perform continuous vector flattening processing on the spatiotemporal correlation features of the training data processed by the pooling layer, so as to flatten the spatiotemporal correlation features of the training data into one-dimensional vector features; wherein the one-dimensional vector features are used to perform basic graphical expression on the spatiotemporal correlation features of the training data;

[0079] S37: the output layer receives and outputs the one-dimensional vector feature of the spatiotemporal correlation feature, and uses the one-dimensional vector feature of the spatiotemporal correlation feature as an accident status identification feature;

[0080] S38: Generate a mapping algorithm unit based on the accident status identification feature, and perform algorithm object and algorithm target deduction processing on the mapping algorithm unit according to the urban three-dimensional platform to obtain an algorithm object unit and an algorithm target unit of the mapping algorithm unit; wherein the algorithm object unit is used to substitute real-time regional information, and the algorithm target unit is used to output accident status features;

[0081] S39: Combining the mapping algorithm unit, the algorithm object unit and the algorithm target unit to obtain an accident condition image recognition algorithm.

[0082] Specifically, the convolutional neural network model consists of an input layer, a convolution layer, a pooling layer, several fully connected layers, and an output layer. The input layer receives panoramic data and drone detection data from the emergency rescue area. The input data is multi-dimensional and contains different types of sensor information, video frames, thermal imaging images, etc. The convolution layer is used to extract spatiotemporal correlation features from the input data. Each layer of convolution operation can extract features from the image or video data layer by layer to capture the spatial and temporal changes. The pooling operation of the pooling layer downsamples the features extracted by the convolution layer, thereby reducing the dimension of the data and retaining important features. This helps to reduce computational complexity and avoid overfitting. The fully connected layer flattens the pooled feature map and converts it into a one-dimensional vector. Multiple fully connected layers perform continuous vector flattening, and finally convert the spatiotemporal correlation features into one-dimensional vector features. The output layer outputs the final accident status image recognition algorithm based on the one-dimensional vector features. These features are used for subsequent accident expansion and simulation.

[0083] It is understandable that convolutional neural networks can automatically extract spatiotemporal correlation features in training data through multi-level convolution and pooling operations, avoiding the difficulty of manually designing features. Pooling operations reduce the dimension of the data and retain the key features of the data, which helps to improve the training efficiency of the model and reduce the computational burden.

[0084] More specifically, the training data is collected and preprocessed. The drone detection data is the image data matched by the drone's detection of the emergency rescue area, that is, the real-time area information waiting for image analysis in actual operation.

[0085] More specifically, the accident condition feature is the annotation information of the accident condition of the drone detection data, that is, the accident condition occurring in the emergency rescue area obtained based on the drone detection data. The accident conditions include fireworks conditions, fire blocking conditions, safety exit blocking conditions, urban waterlogging conditions, dangerous slopes, dangerous building detection, illegal construction on the roof of a building, river floating object identification, and high-voltage tower tilt monitoring conditions.

[0086] More specifically, in the convolution layer, the network extracts spatial features through convolution operations, and the pooling layer further reduces the dimension of the data. Each layer gradually learns higher-level features to model the accident situation in the spatiotemporal dimension. The features extracted after pooling are processed by the fully connected layer, and the spatiotemporal information is further flattened and vectorized, thereby providing a basis for the subsequent generation of accident condition image recognition features. Through the spatiotemporal features extracted by the convolution layer and the pooling layer, the model can identify and process the temporal and spatial changes in the accident process.

[0087] More specifically, based on the accident condition image recognition features output by the convolutional neural network, a mapping algorithm unit is generated. This mapping algorithm is used to map the spatiotemporal features into a broader accident simulation model, helping to identify the association between the image information in the drone detection data and the feedback accident conditions.

[0088] More specifically, an algorithm object unit and an algorithm target unit are generated based on a mapping algorithm unit. The algorithm object unit is used to substitute real-time area information, and the mapping algorithm unit is used to perform algorithm model execution on the algorithm object unit, and output accident status characteristics through the algorithm target unit.

[0089] Preferably, the step of substituting the accident condition characteristics into the regional three-dimensional model, and causing the regional three-dimensional model to simulate the accident condition according to the accident condition characteristics to obtain an accident simulation model for digitally feeding back the accident condition in the emergency rescue area comprises:

[0090] S31: Substituting the accident condition characteristics into the regional three-dimensional model;

[0091] S32: The three-dimensional model of the area is instructed to digitally simulate the fireworks condition, fire blocking condition, emergency exit blocking condition, urban waterlogging condition and dangerous slope condition of the emergency rescue area according to the characteristics of the accident condition, so as to obtain an accident simulation model for digital feedback of the accident condition in the emergency rescue area.

[0092] Preferably, the accident condition features obtained by the image recognition algorithm and the sensor are extracted. These features include: fire condition: location of the fire source, spread of the fire, concentration and direction of the smoke, etc.; fire passage blockage condition: type, location, size and degree of blockage; emergency exit blockage condition: whether the exit is blocked and the condition of the blockage; urban waterlogging condition: waterlogging area, water depth, flow direction, etc.; dangerous slope condition: slope stability and landslide risk in dangerous areas, etc.

[0093] More specifically, the extracted accident condition features are substituted into the three-dimensional regional model. Specific operations include: marking the fire area, converting the fire source, fire spread, smoke and other information into the fire object in the model, and setting the fire spread path; inputting information such as fire passage blockage and exit blockage into the model, indicating which areas are blocked and whether the passage is unobstructed; inputting information such as the water depth and flow direction of urban waterlogging, indicating which areas are flooded, the flow direction of the accumulated water and its impact on regional traffic; marking the risks of dangerous slopes, simulating the possibility and impact range of slope landslides or collapses.

[0094] More specifically, fire models (such as simulations based on fire spread algorithms) are used to calculate the speed of fire spread, flame expansion, smoke flow, and the range of heat influence. Fire models are usually simulated based on physical models such as heat conduction, fluid mechanics, and smoke diffusion. This information can help predict the path of fire spread and provide fire dynamics information to rescuers in a timely manner.

[0095] More specifically, by simulating the congestion of the fire passage, the impact of obstructions on the passage's passability is calculated. For example, using an object collision detection algorithm, it is evaluated whether the passage can remain unobstructed, or what measures need to be taken to clear the passage. The simulation results can provide data support for rescuers to plan the optimal passage.

[0096] More specifically, it simulates whether the emergency exit is blocked and the degree of blockage. This can be done by calculating the flow model, exit capacity and evacuation model. For example, during the simulation of personnel evacuation, which exits are unobstructed and which exits are unusable due to blockage, so as to provide rescuers with evacuation route options.

[0097] More specifically, flood simulation usually simulates water accumulation, flow direction, water depth, etc. based on water flow equations, rainfall, and terrain undulations. Through this simulation, it is possible to determine which areas are prone to flooding and how water flow affects traffic, evacuation routes, and rescue activities within the area.

[0098] More specifically, the digital simulation of slope landslide risk can be based on the slope stability analysis model, taking into account factors such as slope, soil quality, rainfall, etc., to assess the landslide risk of the slope and the possible damage it may cause. This model can help identify areas where landslides may occur and provide rescue workers with recommendations for safe haven areas.

[0099] More specifically, through the above simulations, the digital simulation results of all accident conditions are integrated into a complete accident simulation model. This model can reflect the accident conditions in the emergency rescue area in real time, including the dynamic changes of fire, waterlogging, evacuation blockage, dangerous slopes and other situations.

[0100] More specifically, the simulation results of the model can be fed back in real time, such as being displayed to rescuers through a control center, a rescue command system, or through technologies such as virtual reality (VR). This digital feedback can help rescuers better understand the current accident situation, optimize rescue decisions, and ensure timely and effective emergency response.

[0101] It is understandable that by obtaining the accident status in real time and conducting digital simulation, the latest accident dynamic information can be provided for emergency rescue. Accident simulation can accurately reflect the spread of fire, waterlogging, channel blockage, etc., to avoid human misjudgment. The simulation results can assist rescue personnel in formulating reasonable rescue plans, selecting the most effective channels, arranging the best evacuation routes, predicting dangerous areas, etc. By simulating dangerous areas, possible dangers can be identified and avoided in advance to ensure the safety of rescuers and rescued persons.

[0102] Preferably, the method further includes substituting the real-time regional information into the regional three-dimensional model, so that the regional three-dimensional model simulates accident conditions at more angles according to the real-time regional information to supplement the accident simulation model, wherein the steps include:

[0103] S51: The regional three-dimensional model is used to digitally simulate the building damage and building abnormality according to the real-time regional information, so as to obtain the building damage and abnormality distribution characteristics based on the regional three-dimensional model.

[0104] S52: enabling the regional three-dimensional model to perform digital simulation on the regional population distribution status according to the real-time regional information, so as to obtain the population distribution characteristics based on the regional three-dimensional model;

[0105] S53: enabling the regional three-dimensional model to perform digital simulation of the meteorological environment conditions of the emergency rescue area according to the real-time regional information, so as to obtain meteorological environment characteristics based on the regional three-dimensional model;

[0106] S54: performing three-dimensional modeling correction and model block annotation on the three-dimensional model of the area according to the building damage and abnormal distribution characteristics, the crowd distribution characteristics and the meteorological environment characteristics, so as to obtain an accident simulation model for digital feedback of the accident status of the emergency rescue area.

[0107] Specifically, the real-time regional information (such as camera information, thermal imaging information, and laser information) collected by the drone is input into the existing regional three-dimensional model. This stage ensures that the regional three-dimensional model can obtain the latest real-time data, providing a basis for simulation processing. The data from different sensors are formatted into a format that can be read by the regional three-dimensional model to ensure the compatibility and accuracy of the information. Through data fusion technology, the visual information, thermal imaging data, and lidar scanning results in the sensor data are combined to establish a more accurate three-dimensional model. By substituting real-time data, it is ensured that the regional three-dimensional model always reflects the latest environmental and accident information, improving the real-time and accuracy of the model. Integrating different types of data into the model helps to build a comprehensive and highly operational three-dimensional simulation model.

[0108] More specifically, the regional three-dimensional model simulates the damage and abnormal conditions of buildings in the emergency rescue area based on real-time regional information, and combines thermal imaging data (reflecting temperature changes), lidar data (precisely identifying terrain and building structures) and other information to estimate the extent of damage to buildings, affected structural areas, and potential danger areas.

[0109] More specifically, the building damage and abnormal conditions are visualized, and the spatial distribution characteristics of the building damage are analyzed, such as the scale and shape of the damaged area, and potential collapse areas. By fusing sensor data, building damage can be simulated with high precision, helping emergency rescue personnel understand the extent of damage and key target areas, providing detailed analysis of buildings in the accident area, and helping to quickly assess on-site conditions and repair needs.

[0110] More specifically, based on the thermal imaging data and image data obtained from the real-time regional information, the crowd distribution in the emergency rescue area is simulated. The thermal imaging data can help identify the gathering areas of the crowd and deduce the density and distribution of the crowd. Based on the crowd distribution information, possible personnel behaviors (for example, escape routes, personnel concentration areas, etc.) are further simulated to provide a basis for subsequent rescue operations. Through real-time crowd simulation, the distribution of personnel and potential danger areas in the area can be quickly grasped, providing decision-making support for rescue personnel. Combined with real-time regional information, the crowd distribution can be continuously updated to improve crowd management and rescue efficiency.

[0111] More specifically, real-time meteorological data (such as wind speed, temperature, humidity, etc.) is input into the three-dimensional model to perform digital simulation of meteorological environmental conditions. The meteorological model is used to model weather conditions to obtain meteorological characteristics that affect rescue work. Based on the three-dimensional model, visualization results of meteorological environmental characteristics are generated, including wind speed maps, temperature distribution maps, etc., to help evaluate the impact of weather on the accident site (for example, fire spread, wind impact, etc.). Through meteorological simulation, the impact of meteorological changes on the accident area can be analyzed to help decision makers prepare for emergency response. The model can update weather conditions in real time and provide real-time environmental data support for emergency rescue.

[0112] More specifically, the regional three-dimensional model is modified according to the characteristics of building damage, abnormal distribution, crowd distribution and meteorological environment to ensure that the simulation results reflect the actual situation as accurately as possible. The three-dimensional model is marked with blocks, and different areas are marked according to indicators such as the degree of damage, meteorological conditions and crowd density, so as to more clearly identify areas that need priority treatment. The simulation results of building damage, abnormal conditions, crowd distribution and meteorological environment are integrated to generate an accident simulation model for digital feedback of accident conditions. This model can dynamically display the real-time situation of the accident area and provide the decision-making information required by the emergency response team.

[0113] It is understandable that the revised three-dimensional model and simulation feedback can provide accurate disaster assessment for emergency rescue and help formulate more effective rescue strategies. Through the analysis of simulation results, resource allocation and rescue routes can be effectively planned to optimize emergency response.

[0114] Preferably, it also includes: performing an expansion analysis on the accident simulation model through a pre-trained urban accident expansion analysis algorithm to obtain a regional accident expansion model for performing expansion analysis and digital feedback on locations in the emergency rescue area that cannot be directly detected by the drone formation; performing trend prediction on the accident simulation model through a pre-trained urban accident change analysis algorithm to obtain a regional accident trend model for performing trend analysis and digital feedback on accident conditions in the emergency rescue area; and performing a model combination analysis on the accident simulation model, the regional accident expansion model and the regional accident trend model to obtain an emergency rescue strategy for the emergency rescue area.

[0115] Specifically, in step S4 of the embodiment provided by the present invention, the real-time data collected by the drone (such as images, thermal imaging, gas sensors, lidar and other data) is integrated with the original three-dimensional urban area model, including mapping the fire hot spots, pollution source locations, building damage information, etc. collected by the drone to the corresponding areas in the three-dimensional model. The real-time regional information includes the fire spread speed, gas leakage, population density, etc. These data need to be updated in real time through an automated data processing system and mapped to the three-dimensional model.

[0116] More specifically, different levels of information in the three-dimensional model (such as buildings, roads, green spaces, etc.) are updated based on real-time detection data to ensure that the model always reflects the current actual situation. For example, the height of buildings, damage, and the scope of the fire area in the three-dimensional model must be adjusted as real-time information changes.

[0117] More specifically, on the basis of the three-dimensional model, the state of the area where the accident or disaster occurred is simulated according to the real-time data collected. The state of different stages after the accident is presented through the data model in the virtual environment. Through accurate data fusion, the real-time regional information is combined with the three-dimensional regional model, so that the model can reflect the latest emergency accident status, ensure the dynamic and real-time nature of the model, and enable the three-dimensional model to always reflect the current accident scene, thereby improving the authenticity and operability of the simulation.

[0118] More specifically, based on the updated three-dimensional model, digital simulation technology is used to simulate the accident situation. Common simulation methods include: fire spread simulation: use the fire dynamic model to simulate the spread of fire in buildings and areas, considering factors such as wind speed, temperature, and building materials; pollution diffusion simulation: according to the gas sensor data, simulate the diffusion process of harmful gases and predict the changing trend of the pollution area; personnel evacuation simulation: combine the structural information of the building and the location of the personnel to simulate the evacuation process of the personnel, evaluate the utilization of the safety exits, the congestion of the evacuation routes, etc.

[0119] More specifically, during the simulation process, simulation algorithms (such as fluid dynamics models, meteorological models, structural mechanics models, etc.) are combined to accurately simulate the different development stages of the accident, and the simulation results are dynamically updated. Accident simulation feedback is generated through the simulation results. These feedbacks include: risk areas and time predictions for fire or pollution spread, structural safety assessments of damaged buildings, blockage of evacuation passages and optimization of rescue routes. The digital feedback will be displayed through a graphical interface, including heat maps, path maps, accident development trends, etc.

[0120] It is understandable that through simulations from multiple angles such as fire, pollution, and personnel evacuation, the impact of the accident on the region is comprehensively assessed, and accurate simulation feedback data is generated. Based on accurate physical models and real-time data, a refined simulation of the development of the accident is carried out to ensure the high accuracy of the simulation results. The accident situation is visualized through digital feedback, helping emergency command personnel to quickly understand changes in the situation and the effectiveness of the emergency response.

[0121] More specifically, the accident simulation results will provide emergency commanders with detailed information on fire spread, pollution diffusion, evacuation progress, etc. By analyzing these simulation results, commanders can obtain: the expected spread range and time of fire or harmful gases. Safety assessment of each area, especially the dangerous situation of key areas (such as hospitals, schools, office buildings, etc.). The load and congestion of evacuation channels. Quantitative analysis of this information can obtain real-time prediction of accident development trends to ensure that emergency response can be adjusted in time according to the simulation results.

[0122] More specifically, as the drone detection mission progresses and new data is collected, the three-dimensional model and accident simulation will be continuously updated, and new real-time data (such as new fire points, gas leakage sources, rescue progress, etc.) will be reflected in the accident simulation in real time. By updating and optimizing the accident simulation model in real time, it is ensured that in a rapidly changing emergency environment, the simulation results always reflect the most realistic accident conditions. According to the latest accident simulation results, the emergency response plan is dynamically adjusted. For example, when a new fire point appears, the drone's detection path is readjusted, or the evacuation route of personnel is adjusted according to the evacuation situation, the deployment of rescue resources is optimized, and it is ensured that manpower and materials can reach the areas where they are most needed in a timely manner.

[0123] Specifically, in step S4 of the embodiment provided by the present invention, historical urban accident data is collected, covering different types of disasters (such as fire, chemical leakage, flood, etc.) and corresponding emergency rescue information. These data include the spatial distribution, time evolution, meteorological conditions, infrastructure conditions, etc. of the accidents, and obtain geographic spatial data (such as city maps, building distribution, transportation networks, etc.), infrastructure data of urban areas (such as water supply, power supply, communication facilities, etc.), and all historical case data related to urban emergency response.

[0124] More specifically, based on historical data, an accident expansion analysis model is established using machine learning, deep learning and other algorithms. Commonly used algorithms include: regression analysis: predicting the spread and impact area of ​​the accident, especially areas that drones cannot reach; cluster analysis: grouping urban areas according to potential risks, building density, road network and other factors to predict the possibility of accident expansion; spatiotemporal data analysis: analyzing the evolution trend of accidents at different time and space scales through spatiotemporal data models.

[0125] More specifically, historical data is used to train the model, verify the accuracy of the extended analysis, and through methods such as cross-validation, ensure that the model can accurately predict the development and evolution of accidents in unknown areas, and adjust model parameters to improve the model's predictive ability for invisible or difficult-to-access areas in complex urban environments.

[0126] It is understandable that through machine learning and data mining technology, potential patterns can be extracted from massive historical data, the accuracy of accident prediction and extended analysis can be improved, and adaptive adjustments can be made according to different types of urban accidents (fires, chemical leaks, traffic accidents, etc.), and potential risk areas under various accident modes can be predicted.

[0127] More specifically, the pre-trained urban accident extension analysis algorithm is integrated with the current accident simulation model. The integration process includes applying the extension analysis model to areas that cannot be detected by drones to estimate and expand the scope of potential accidents. For areas that cannot be detected by drones in real time (such as underground facilities, inside high-rise buildings, areas affected by weather, etc.), the extension analysis model is used to infer possible accident development trends and generate corresponding accident extension predictions.

[0128] More specifically, based on the original accident simulation model, an extended analysis algorithm is used to predict the scope of the accident. For example, by analyzing the current fire spread speed, meteorological data, building layout, underground facilities, etc., the potential spread area of ​​​​the fire or harmful gas leakage is inferred. For areas far away from the detection equipment or the flight path of the drone, corresponding extended models are generated. The accident status of these areas may be supplemented by remote sensing equipment (such as ground sensors, satellite images, etc.) for analysis.

[0129] More specifically, based on the results of the expansion analysis, a new regional accident expansion model is formed. This model can provide digital feedback on areas that cannot be covered by drones, display potential danger areas, possible accident expansion paths, and estimated rescue time windows. It uses a virtual simulation platform to display regional accident expansion situations, such as fire spread, gas diffusion, trapped personnel, etc., to help command personnel prepare for emergency response in advance.

[0130] Specifically, relevant urban accident data are collected, including the time, location, scale, environmental factors (such as weather, traffic conditions, infrastructure, etc.) of different types of disasters (such as fire, earthquake, flood, chemical leak, etc.), including real-time monitoring data (drones, sensors, satellite images, etc.), and historical accident data. These data provide a basis for the prediction of accident changes. Based on the collected data, features closely related to the development trend of accidents are extracted, such as the speed of spread after the accident, the impact of climate change on the accident, the response time of emergency rescue resources, the status of urban infrastructure, etc., and relevant features of spatial and temporal factors are extracted, such as the density of accident locations, the smoothness of the transportation network, the distribution of emergency rescue resources, etc.

[0131] More specifically, machine learning or deep learning algorithms suitable for processing time series data are used, such as: regression analysis: used to predict numerical trends in accident development (such as fire spread area, pollutant concentration, etc.), time series analysis: such as ARIMA, LSTM and other models, which can capture the time series patterns in accident development and predict the evolution trend of accidents in the future, deep neural networks (DNN): can mine complex nonlinear relationships, perform trend prediction and pattern recognition, graph neural networks (GNN): if complex interactive relationships involving urban infrastructure are involved, graph neural networks can effectively capture these complex associations.

[0132] More specifically, historical data is used to train the prediction model to ensure that the model can accurately capture the potential trends in the development of accidents. The prediction effect of the model is verified through cross-validation, validation sets and other methods. The model parameters are adjusted and the prediction ability of the model is optimized. The trained algorithm can extract key patterns from complex accident data to achieve accurate accident trend prediction. The spatiotemporal characteristics are used to predict the changing trends of accidents in different locations and times, thereby improving the accuracy of the prediction.

[0133] More specifically, the pre-trained urban accident change analysis algorithm is combined with the accident simulation model, and the accident trend is predicted in combination with real-time data, such as the spread trend of fire, the speed of pollutant diffusion, etc. Based on the accident simulation model, the state of the accident at different time nodes is simulated to generate trend forecasts for a period of time in the future, especially for areas that cannot be directly observed or accident situations that are difficult to detect.

[0134] More specifically, based on the output of the accident change analysis algorithm, an accident trend model for a specific area is generated. The model can provide decision makers with possible paths for the future development of the accident, including the expansion of the accident scope, the demand for rescue resources, changes in the scope of impact, etc. For example, when predicting a chemical leak accident, the model will analyze meteorological data, the size of the leak source, wind speed and other factors to predict the spread trend of the accident pollution and provide a corresponding emergency response timetable.

[0135] More specifically, the trend prediction results are converted into digital feedback to display the development trend graph of the accident, risk assessment graph, possible future impact areas, etc. Through charts and visualization methods, decision makers can quickly understand the possible development trend of the accident, and provide a basis for dynamic adjustment for the emergency command center, such as whether additional rescue personnel need to be dispatched, whether the evacuation route needs to be readjusted, etc. The regional accident trend model provides the ability to analyze the development of accidents from different dimensions (time, space, impact range, etc.), helping decision makers to fully grasp the development dynamics of the accident, and provide real-time digital trend feedback for the emergency rescue command system to support rapid response and decision-making.

[0136] More specifically, during an accident, real-time monitoring data (such as drone, sensor data, meteorological data, etc.) will be input into the trend prediction model to adjust the prediction results. For example, as meteorological conditions change, the speed or direction of fire spread may change, and the model needs to be updated in real time. By integrating real-time data, the trend prediction model can dynamically adjust and correct accident trends in the future to ensure more accurate predictions. The prediction results and real-time feedback are combined to automatically generate emergency response strategies through intelligent decision-making systems. For example, when it is predicted that the fire will spread to a certain area, the system can automatically prompt changes in evacuation routes and deploy additional fire brigades or emergency medical resources.

[0137] It is understandable that through the input of real-time data and the dynamic adjustment of the trend prediction model, the prediction results can be continuously updated during the development of the accident, and accurate emergency response strategies can be provided. Through real-time digital feedback, decision makers can quickly adjust emergency response plans based on the latest trend prediction results, optimize resource allocation, and improve response speed.

[0138] More specifically, the accident simulation model is a digital model that provides digital feedback on the accident conditions in the emergency rescue area based on the real-time regional information collected by the drone formation. The regional accident expansion model is a digital model that expands and simulates the locations that cannot be directly detected by the drone formation based on the accident simulation model to obtain their accident conditions. The regional accident trend model is a digital model that simulates and predicts the future accident conditions in the emergency rescue area based on the accident simulation model.

[0139] More specifically, a combined model analysis is performed based on the accident simulation model, the regional accident expansion model and the regional accident trend model. The emergency rescue strategy is analyzed by analyzing the accident conditions that are directly detected and indirectly inferred in the emergency rescue area. At the same time, the feasibility of various emergency rescue strategies is judged by predicting the development trend of future accidents. Ultimately, the emergency rescue strategy implemented by the rescue personnel is derived.

[0140] More specifically, the emergency rescue strategy includes rescue workers rescuing trapped people and using drones to provide voice guidance to the emergency rescue area.

[0141] It is understandable that through the combined analysis of the accident simulation model, the regional accident expansion model and the regional accident trend model, it is possible to comprehensively assess the accident risk, optimize resource allocation and rescue routes, and formulate efficient and accurate emergency rescue strategies. The technical effects include comprehensive decision support, real-time dynamic adjustment, efficient resource allocation and rescue path optimization, which not only improves the efficiency of emergency response, but also continuously optimizes emergency response capabilities through feedback mechanisms to ensure timely and effective disaster response.

[0142] Preferably, the step of pre-training the urban accident expansion analysis algorithm includes:

[0143] Construct a convolutional neural network model consisting of an input layer, a convolutional layer, a pooling layer, several fully connected layers, and an output layer;

[0144] Preparing training data for model training of the convolutional neural network model; wherein the training data includes regional panoramic data and drone detection data of various emergency rescue accidents, the regional panoramic data is the status data of all positions in the emergency rescue area, and the drone detection data is the status data detected by the drone formation on the emergency rescue area;

[0145] Substituting the training data into the input layer;

[0146] The input layer receives the training data and transmits the training data to the convolution layer, and the convolution layer is used to collect spatiotemporal correlation features of the training data to obtain the spatiotemporal correlation features of the training data;

[0147] The pooling layer is used to perform a pooling operation on the spatiotemporal correlation features of the training data extracted by the convolution layer to reduce the size of the spatiotemporal correlation features of the training data and retain key features of the spatiotemporal correlation features of the training data;

[0148] Each of the fully connected layers is used to perform continuous vector flattening processing on the spatiotemporal correlation features of the training data processed by the pooling layer, so as to flatten the spatiotemporal correlation features of the training data into one-dimensional vector features; wherein the one-dimensional vector features are used to perform basic graphical expression on the spatiotemporal correlation features of the training data;

[0149] The output layer receives and outputs the one-dimensional vector feature of the spatiotemporal correlation feature, and uses the one-dimensional vector feature of the spatiotemporal correlation feature as the urban accident expansion correlation feature;

[0150] A mapping algorithm unit is generated based on the urban accident expansion association characteristics, and the algorithm object and algorithm target of the mapping algorithm unit are deduced according to the urban three-dimensional platform to obtain an algorithm object unit and an algorithm target unit of the mapping algorithm unit; wherein the algorithm object unit is used to substitute into the accident simulation model, and the algorithm target unit is used to output the regional accident expansion model;

[0151] The mapping algorithm unit, the algorithm object unit and the algorithm target unit are combined to obtain an urban accident expansion analysis algorithm.

[0152] Specifically, the convolutional neural network model consists of an input layer, a convolution layer, a pooling layer, several fully connected layers, and an output layer. The input layer receives panoramic data and drone detection data from the emergency rescue area. The input data is multi-dimensional and contains different types of sensor information, video frames, thermal imaging images, etc. The convolution layer is used to extract spatiotemporal correlation features from the input data. Each layer of convolution operation can extract features from the image or video data layer by layer to capture the spatial and temporal changes. The pooling operation of the pooling layer downsamples the features extracted by the convolution layer, thereby reducing the dimension of the data and retaining important features. This helps to reduce computational complexity and avoid overfitting. The fully connected layer flattens the pooled feature map and converts it into a one-dimensional vector. Multiple fully connected layers perform continuous vector flattening, and finally convert the spatiotemporal correlation features into one-dimensional vector features. The output layer outputs the final urban accident expansion correlation features based on the one-dimensional vector features. These features are used for subsequent accident expansion and simulation.

[0153] It is understandable that convolutional neural networks can automatically extract spatiotemporal correlation features in training data through multi-level convolution and pooling operations, avoiding the difficulty of manually designing features. Pooling operations reduce the dimension of the data and retain the key features of the data, which helps to improve the training efficiency of the model and reduce the computational burden.

[0154] More specifically, the training data is collected and preprocessed. The regional panoramic data represents the overall situation of the emergency rescue area, including various conditions in the accident area (such as building damage, fire smoke, road blockage, etc.). The drone detection data is the regional data detected by the drone formation, including high-precision images, thermal imaging data, and ground video streams. That is, the regional panoramic data is the accident status data of all locations in the emergency rescue area obtained through data analysis and real-scene collection, and the drone detection data is the data that can be directly detected by the drone formation in actual operation. It can be seen that the regional panoramic data includes the part that cannot be directly detected in the drone detection data. Therefore, by analyzing the correlation between the regional panoramic data and the drone detection data, the correlation between the drone detection data and the location that the drone cannot directly detect can be obtained, and then the accident status of the undetected location can be inferred based on the drone detection data.

[0155] More specifically, various accident features in the labeled dataset are used, such as building damage, casualties, road blockage, etc. Data augmentation techniques (such as image rotation, scaling, and cropping) are used to expand the dataset and enhance the generalization ability of the model. Diversified training data (including panoramic data and drone data) ensures that the convolutional neural network can learn features in more scenarios. Data augmentation improves the adaptability of the model in unknown environments, helps avoid overfitting, and improves the generalization ability of the model.

[0156] More specifically, in the convolution layer, the network extracts spatial features through convolution operations, and the pooling layer further reduces the dimension of the data. Each layer gradually learns higher-level features, thereby modeling the accident situation in the spatiotemporal dimension. The features extracted after pooling are processed by the fully connected layer, and the spatiotemporal information is further flattened and vectorized, thereby providing a basis for the subsequent generation of urban accident extension-related features. Through the spatiotemporal features extracted by the convolution layer and the pooling layer, the model can identify and process the temporal and spatial change patterns during the accident process, such as the temporal development of building damage, changes in the meteorological environment, etc. After converting the spatiotemporal features into one-dimensional vectors, it can be more easily linked with other models to generate more effective accident extension-related features.

[0157] More specifically, a mapping algorithm unit is generated based on the urban accident expansion correlation features output by the convolutional neural network. This mapping algorithm is used to map the spatiotemporal features into a broader accident simulation model to help predict and expand the evolution of accidents. The algorithm objects (i.e., disaster simulation areas, buildings, roads, facilities, etc.) are deduced based on the urban three-dimensional platform and simulation data. Furthermore, the target unit of the algorithm is determined through target deduction, and the accident expansion model of the emergency rescue area is output.

[0158] It is understandable that the mapping algorithm unit can automatically adjust parameters according to the training data and deduce different accident expansion models, which enables the model to adapt to different types of accident scenarios and has strong adaptability. Through the deduction and mapping of the simulation model, it is possible to predict possible disaster accidents in the city and provide early warning and decision support for emergency rescue.

[0159] More specifically, the convolutional neural network is trained using pre-prepared training data through optimization methods such as the gradient descent algorithm, and the network weights are adjusted to make the output urban accident expansion correlation features as accurate as possible. According to the feedback of the verification data during the training process, the model is optimized, such as adjusting the structure of the convolution layer and pooling layer, or modifying the depth of the fully connected layer, etc., to improve the accuracy and stability of the model. The model is continuously optimized through training and can learn effective spatiotemporal features from massive data, thereby improving the accuracy of accident expansion analysis. The optimized model can provide more detailed accident expansion predictions and help decision makers take more appropriate emergency measures when disasters occur.

[0160] Preferably, the step of pre-training the urban accident change analysis algorithm includes:

[0161] Construct a long short-term memory network model consisting of an input layer, an LSTM layer, several fully connected layers, and an output layer;

[0162] Preparing training data for model training of the long short-term memory network model; wherein the training data includes drone detection data of emergency rescue areas arranged in chronological order corresponding to various types of emergency rescue accidents;

[0163] Substituting the training data into the input layer;

[0164] The input layer receives the training data and transmits the training data to the LSTM layer, and the LSTM layer is used to collect the characteristics of the long-short dependency relationship in the time sequence of the training data to obtain the spatiotemporal development characteristics of the training data;

[0165] Each of the fully connected layers is used to perform continuous vector flattening processing on the spatiotemporal development features extracted by the LSTM layer, so as to flatten the spatiotemporal development features into one-dimensional vector features; wherein the one-dimensional vector features are used to perform basic graphical expression on the spatiotemporal development features;

[0166] The output layer receives and outputs the one-dimensional vector feature of the spatiotemporal development feature, and uses the one-dimensional vector feature of the spatiotemporal development feature as the urban accident trend feature;

[0167] A mapping algorithm unit is generated based on the urban accident trend characteristics, and the algorithm object and algorithm target of the mapping algorithm unit are deduced according to the urban three-dimensional platform to obtain an algorithm object unit and an algorithm target unit of the mapping algorithm unit; wherein the algorithm object unit is used to substitute into the accident simulation model, and the algorithm target unit is used to output a regional accident trend model;

[0168] The mapping algorithm unit, the algorithm object unit and the algorithm target unit are combined to obtain an urban accident change analysis algorithm.

[0169] Specifically, a long short-term memory network model consisting of an input layer, an LSTM layer, several fully connected layers and an output layer is constructed.

[0170] More specifically, the input layer receives training data, which is drone detection data arranged in chronological order, representing regional data of various emergency rescue accidents. Each frame of data includes the regional status at that time point (such as building damage, fire, traffic conditions, etc.).

[0171] More specifically, the LSTM layer is responsible for capturing the time series features in the training data, especially the long- and short-term dependencies. LSTM effectively stores and transmits time information through its special gating mechanism (forget gate, input gate, output gate), solving the gradient vanishing and gradient exploding problems of traditional RNN when processing long sequences.

[0172] More specifically, the fully connected layer flattens the spatiotemporal development features extracted by the LSTM layer and converts them into one-dimensional vector features. Multiple fully connected layers gradually refine the feature vectors and ultimately obtain an expression of the development trend of urban accidents.

[0173] More specifically, the output layer receives one-dimensional vector features and outputs urban accident trend features, which describe the possible development paths of accidents and provide support for further analysis.

[0174] It can be understood that LSTM effectively captures long-term dependencies in time series data and is able to model complex accident development processes, thereby providing more accurate trend forecasts. LSTM is particularly suitable for processing time series data, such as emergency rescue data detected by drones, and can better predict the future development of accidents.

[0175] More specifically, drone detection data at different time points are collected. These data include various dynamic information in the emergency rescue area, such as real-time images, thermal imaging, sensor data, etc. The data are arranged in chronological order so that the LSTM model can learn the time dependency and annotate the various accident features in the data, such as fire spread, traffic accidents, building structure collapse, etc. Data enhancement operations (such as time offset and data perturbation) can expand the training set and improve the robustness of the model.

[0176] It can be understood that by arranging the data in chronological order, LSTM is able to learn the potential temporal correlation in the data and help predict the future development of the accident. The labeling and enhancement techniques ensure the high quality and diversity of the training data and improve the predictive ability and generalization performance of the model.

[0177] More specifically, the LSTM layer extracts dynamic changes in time through its gating mechanism and combines the spatial characteristics of the regional state to form spatiotemporal development characteristics, which can describe the entire process from the occurrence to the expansion of the accident and reflect the accident trends at different time points.

[0178] More specifically, the spatiotemporal features are flattened through the fully connected layer and converted into one-dimensional vector features, forming a basis for graphical expression and further modeling. Through the LSTM network, the spatiotemporal dynamic changes of the accident can be accurately captured, helping to analyze the future development trend of the accident. Flattening the features through the fully connected layer simplifies the complex spatiotemporal data, facilitating subsequent algorithm processing and model optimization.

[0179] More specifically, based on the extracted urban accident trend characteristics, a mapping algorithm unit is generated to deduce the possible development of the accident. On the urban three-dimensional platform, the mapping algorithm unit is deduced to obtain an algorithm object unit and an algorithm target unit. The algorithm object unit represents a simulated accident scenario, while the algorithm target unit is used to output a regional accident trend model. By combining the mapping algorithm unit, the algorithm object unit and the algorithm target unit, an urban accident change analysis algorithm is finally constructed. Through the mapping algorithm unit, the evolution process of the accident can be deduced in real time to provide decision support for emergency management. Combined with the urban three-dimensional platform, the accident trend model can be visualized, allowing emergency rescue personnel to understand the development dynamics of the accident more intuitively.

[0180] More specifically, this algorithm can effectively predict and analyze the changing trends of urban accidents, especially in the process of emergency rescue and post-disaster recovery. According to the accident trend characteristics output by the model, the emergency response plan can be optimized, and the rescue efficiency and resource allocation can be improved. Through these steps, the urban accident change analysis algorithm can accurately capture the spatiotemporal characteristics and development trends of accidents, and provide strong data support and decision-making basis.

[0181] Preferably, the step of performing model combination analysis on the accident simulation model, the regional accident expansion model and the regional accident trend model to obtain the emergency rescue strategy of the emergency rescue area includes:

[0182] Combining the regional accident expansion model with the accident simulation model to obtain a current accident panoramic simulation model;

[0183] Performing future expected deduction processing on the current accident panoramic simulation model according to the regional accident trend model to obtain a predicted accident panoramic simulation model;

[0184] Analyze and process the refuge location, the location to be evacuated and the corresponding evacuation route of the current accident panoramic simulation model to obtain several refuge routes in the emergency rescue area;

[0185] Conducting feasibility analysis on each evacuation route according to the predicted accident panoramic simulation model to obtain a feasibility index of each evacuation route, and selecting each evacuation route according to the feasibility index of each evacuation route to obtain a number of execution routes;

[0186] A feasibility analysis of emergency rescue methods is performed on each of the execution routes to obtain an emergency rescue method corresponding to each of the execution routes. Each of the execution routes and the corresponding emergency rescue method together constitute the emergency rescue strategy.

[0187] Specifically, the regional accident expansion model is used to describe the expansion of accidents in different regions, including the evolution path of accidents, the scope of impact, the possibility of accidents, etc. The model is based on the initial conditions of the accident, such as the starting point of the fire, the location of the explosion source, etc. The accident simulation model simulates the process and specific impact of the accident in detail, including building damage, personal injury, resource requirements, etc. By combining these two models, a panoramic simulation model of the current accident is obtained. This model shows the overall evolution of the accident from the beginning to the current moment, providing detailed scenes and a global perspective of the accident. The panoramic simulation model provides decision makers with an overall view, enabling them to understand the specific manifestations of the accident in various regions, thereby ensuring the globality and accuracy of emergency decision-making. The regional accident expansion and accident simulation models are combined to help track the changing process of the accident in real time and make flexible adjustments.

[0188] More specifically, the current accident panoramic simulation model is deduced for future expectations based on the regional accident trend model. The regional accident trend model is based on historical data, real-time information and prediction algorithms, and can predict the future development trend of accidents. Taking into account factors such as environmental changes, resource allocation, and personnel evacuation, the regional accident trend model is used to deduce the current accident panoramic simulation model to obtain a predicted accident panoramic simulation model. The model shows the possible evolution of the accident in the future and provides a forecast of the future situation. By deducing the future development of accidents, it helps emergency management personnel identify possible risks in advance, provides a scientific basis for emergency response, and can dynamically adjust emergency strategies and resource allocation based on the deduction results to avoid misjudgment or omission of important factors.

[0189] More specifically, the current accident panorama simulation model is further analyzed to identify possible refuge locations and evacuation locations, which can be safe areas of buildings, evacuation passages, medical rescue points, etc. The refuge locations and evacuation locations are connected in the model to plan a safe evacuation route. The analysis takes into account factors such as road accessibility, traffic congestion, obstacles, and accident types. The analysis results provide clear refuge locations and evacuation routes to ensure the safe evacuation of personnel and reduce the risk of casualties. By analyzing the evacuation route, the optimal resource allocation plan can be planned to avoid the waste of rescue resources.

[0190] More specifically, a feasibility analysis of the evacuation routes is conducted based on the predictive accident panoramic simulation model. For each planned evacuation route, a feasibility analysis is conducted based on the predictive accident panoramic simulation model. The analysis factors include: traffic conditions: assessing road conditions, traffic flow and possible traffic jams; obstacles and dangerous areas: analyzing possible obstacles, fires, explosions and other dangerous factors on the evacuation route; time limit: considering the evolution speed of the accident to ensure that the evacuation route can complete the evacuation of personnel within the time limit.

[0191] More specifically, each evacuation route is calculated with a feasibility index based on the results of its feasibility analysis. The index reflects the safety and effectiveness of the route. Through feasibility analysis and calculation of the feasibility index, the safest and most effective evacuation route is selected to reduce possible risks during the evacuation process, optimize the evacuation route, improve the evacuation speed and efficiency, and ensure that the safe evacuation of personnel is completed in the shortest time.

[0192] More specifically, a detailed analysis of emergency rescue methods is conducted for each evacuation route, including: Medical rescue: setting up temporary medical stations at the end or midway of the evacuation route to ensure that the injured can receive timely treatment, Material supply: evaluating the materials and support required for each route, such as food, water, rescue tools, etc., based on actual conditions, Firefighting and rescue: evaluating the fire or explosion situations that may be encountered on each route to determine whether additional firefighting measures are needed.

[0193] More specifically, based on the analysis results of execution routes and rescue methods, combined with the feasibility of each route, several execution routes are selected, and corresponding emergency rescue methods are determined for each route. Finally, a complete emergency rescue strategy is formed to provide a special emergency rescue method for each evacuation route to ensure that different types of rescue needs are met. The emergency rescue strategy integrates emergency responses in multiple aspects such as evacuation, medical care, and fire fighting, and can provide all-round support in actual rescue.

[0194] Preferably, the step of analyzing and processing the refuge position, the position to be evacuated and the corresponding evacuation route of the current accident panoramic simulation model to obtain a plurality of refuge routes in the emergency rescue area includes:

[0195] Analyze the safe position in the area and the evacuation position at the area boundary of the current accident panoramic simulation model to obtain the safe position in the area and the evacuation position at the area boundary of the current accident panoramic simulation model, and use the safe position in the area and the evacuation position at the area boundary as the refuge position;

[0196] Analyze the crowd concentration area and the individual location area of ​​the current accident panoramic simulation model to obtain the evacuation position of the current accident panoramic simulation model;

[0197] Analyzing the evacuation routes of each of the refuge locations and each of the locations to be evacuated based on the current accident panoramic simulation model to obtain evacuation routes corresponding to each of the refuge locations;

[0198] Each of the locations to be evacuated is combined with each corresponding evacuation route to obtain a plurality of refuge routes.

[0199] Specifically, the current panoramic accident simulation model is analyzed to identify safe locations within the area (such as places far away from the accident source and with better refuge effects) and evacuation locations at the boundaries of the area (such as exits and roads leading to safe areas, etc.). These locations should be able to provide sufficient safety and reduce the dangers that may be faced when an accident occurs.

[0200] More specifically, safe locations include refuge areas inside buildings, temporary shelters, areas that are relatively unaffected by accidents, etc. Evacuation locations include nearby safe exits, passages, or roads, etc., and safe locations within the area and evacuation locations at the area boundary are obtained, and these locations are merged into refuge locations as key nodes for crowd gathering and evacuation.

[0201] More specifically, based on the current panoramic simulation model of the accident, the areas where people gather (such as work areas, residential areas, public areas) and areas where individuals are located (such as personal residences, separate work areas, etc.) within the accident area are identified, and emergency assessments are conducted on these areas to determine which areas are most in need of evacuation, as well as the order and priority of evacuation, and the locations to be evacuated, that is, the areas or gathering points for personnel that need emergency evacuation, are obtained. By comprehensively analyzing the current panoramic simulation model of the accident, the safe points and emergency evacuation points in the area can be clearly delineated to ensure that the crowd can be evacuated quickly and efficiently in an emergency. By analyzing the areas where people gather and the areas where individuals are located, the order and focus of evacuation can be optimized to reduce confusion during the evacuation process.

[0202] More specifically, a path analysis is performed on each refuge location and location to be evacuated to identify feasible evacuation routes. This analysis needs to consider the smoothness of the road, the impact of accidents on the route (such as fire, structural collapse, traffic jam, etc.), and the safety of the route. A simulation algorithm (such as a graph algorithm, a shortest path algorithm, an obstacle avoidance algorithm, etc.) is used to calculate the optimal evacuation route from each refuge location to each location to be evacuated, and the evacuation route corresponding to each refuge location is obtained. A reasonable evacuation channel is designed for each refuge point and evacuation point, and each location to be evacuated and the evacuation route are combined to ensure that each area to be evacuated has multiple feasible evacuation routes. This can be achieved by calculating the combination of each evacuation location and all available routes to ensure that emergencies (such as a route being blocked) can be dealt with during the evacuation process, and several refuge routes can be obtained, providing multiple alternative routes for the evacuation of personnel in the accident. Through path analysis, the optimal evacuation route can be determined according to the accident scenario, avoiding traffic congestion and route blockages, and improving evacuation efficiency. Multiple alternative evacuation routes ensure that in the event of a specific route being blocked, there are still alternative routes for personnel to evacuate, reducing the risks caused by the inaccessibility of a single route. The simulation algorithm is used to analyze and optimize the route, so that the emergency management system can dynamically adjust the evacuation route according to the real-time situation, thereby improving rescue efficiency and emergency response capabilities.

[0203] More specifically, based on the preliminary analysis, the various evacuation routes are finally integrated and optimized to ensure the feasibility, timeliness and safety of each route. The implementation details of each evacuation route and possible emergency methods (such as medical rescue, evacuation team support, etc.) are combined to eventually form a complete emergency rescue strategy. Through precise path analysis and evacuation route design, a comprehensive rescue strategy covering multiple scenarios, locations and personnel can be formed to ensure that rescue plans in various accident situations have reliable support. When an accident occurs, accurate evacuation routes and feasible emergency strategies can effectively reduce rescue time and casualties, and improve the overall emergency response speed.

[0204] Preferably, the steps of performing feasibility analysis of emergency rescue methods on each of the execution routes to obtain emergency rescue methods corresponding to each of the execution routes include:

[0205] Analyzing the execution efficiency and the execution result of the execution route according to several preset methods to obtain the execution efficiency parameters and the execution result parameters of the execution route corresponding to the several preset methods;

[0206] A weighted analysis is performed on the execution efficiency parameters and execution result parameters of several preset methods corresponding to the execution route to obtain the execution feasibility characteristics of various preset methods, and the emergency rescue method of the execution route is obtained by matching and evaluating the execution feasibility characteristics of various preset methods.

[0207] Specifically, the execution efficiency of each evacuation route is evaluated according to several preset methods (such as the number of rescue personnel, time window, resource allocation, etc.). The execution efficiency can be measured from multiple aspects, such as the time required for each route, the number of resources that can be mobilized, the evacuation speed of personnel, etc. The execution results of each route are evaluated according to the preset method, which can include the achievement of goals, such as the success rate of safe evacuation of personnel, the number of people successfully rescued on each route, the expected rescue tasks, etc., to obtain execution efficiency parameters and execution result parameters, that is, the efficiency and effect of each execution route under different preset methods. Through quantitative analysis of the preset methods, the potential efficiency and results of each execution route can be accurately evaluated, providing a scientific basis for the subsequent selection of the optimal execution method. It not only evaluates from the perspective of time efficiency, but also comprehensively considers multi-dimensional factors such as resource allocation and personnel scheduling, making the analysis more comprehensive and accurate.

[0208] More specifically, a weighted analysis is performed on each execution route based on the execution efficiency parameters and execution result parameters of different preset methods. The weighted analysis can assign different weights to different parameters based on the actual needs of emergency rescue. For example, in some emergency rescue situations, time efficiency is more important than resource utilization, or the success rate of safe evacuation of personnel is more critical than the evacuation speed. The comprehensive evaluation score of each execution route is calculated in combination with the weighted parameters, and the comprehensive effect of the execution route under different methods is determined. The execution feasibility characteristics of each execution route under each preset method are obtained. These characteristics can be used to determine which execution methods are suitable for different evacuation routes. Through weighted analysis, the performance of each route under different rescue methods can be accurately quantified to ensure the selection of the optimal execution plan. According to the results of the weighted analysis, the most appropriate rescue method can be selected for different emergency situations to ensure the optimal configuration and maximum effect of rescue resources.

[0209] More specifically, by evaluating the feasibility characteristics of different preset methods and matching the needs of emergency rescue methods, the feasibility characteristics may include route accessibility, path obstruction, resource requirements, staffing, time constraints and other aspects. By comparing the characteristics of different execution methods, the emergency rescue method that best matches each evacuation route is selected to ensure that the adaptability of the rescue method and the route is maximized, and the best emergency rescue method for each evacuation route is obtained, that is, the execution plan that best suits the characteristics of the route. Through matching evaluation, the most suitable rescue method can be quickly selected for each execution route, thereby improving emergency rescue efficiency and reducing time waste. Each evacuation route is tailored to the most suitable rescue method according to its characteristics to ensure the efficiency and accuracy of emergency rescue.

[0210] More specifically, based on the results of all the previous analyses, an emergency rescue method is selected for each evacuation route, and the method with the best execution efficiency and results and in line with feasibility characteristics is selected. The selected emergency rescue method is further optimized. According to the specific disaster scenario, resource availability and real-time conditions, resource allocation, personnel scheduling and route arrangement are optimized to ensure that emergency rescue is smoother in actual operation and to obtain a complete and optimized emergency rescue plan.

[0211] It is understandable that through the final selection and optimization, the rescue efficiency can be maximized and the losses can be minimized. Taking into account the possible changes in actual emergency situations, the optimization plan can be dynamically adjusted to ensure the resilience and flexibility of emergency rescue.

[0212] In a second aspect, the present invention provides an emergency rescue system based on a drone and a three-dimensional urban platform, which is used to implement an emergency rescue method based on a drone and a three-dimensional urban platform as described in any one of the first aspects.

[0213] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An emergency rescue method based on drones and urban three-dimensional platforms, characterized in that: include: Acquire the location information of the emergency rescue area in the urban three-dimensional platform, and retrieve the regional model and analyze the detection plan of the urban three-dimensional platform according to the location information to obtain the regional three-dimensional model and drone detection plan of the emergency rescue area; According to the drone detection scheme, drive the drone formation to the emergency rescue area to perform regional detection work to obtain real-time regional information of the emergency rescue area; Performing image recognition of the accident situation on the real-time regional information through a pre-trained accident situation image recognition algorithm to obtain accident situation characteristics corresponding to the real-time regional information; wherein the accident situation characteristics include fireworks, fire blocking conditions, emergency exit blocking conditions, urban waterlogging conditions, dangerous slope conditions, dangerous building conditions, illegal construction conditions on the roof of a building, floating objects in a river, and tilted high-voltage tower conditions; Substituting the accident condition characteristics into the regional three-dimensional model, allowing the regional three-dimensional model to simulate the accident condition according to the accident condition characteristics, so as to obtain an accident simulation model for digitally feeding back the accident condition in the emergency rescue area; wherein the accident simulation model is used to provide real-time simulation feedback on the accident condition in the emergency rescue area, so as to assist rescue personnel in making rescue decisions according to the accident condition; The steps of obtaining the location information of the emergency rescue area in the urban three-dimensional platform and retrieving the regional model of the urban three-dimensional platform according to the location information to obtain the regional three-dimensional model of the emergency rescue area include: Obtain the location information of the emergency rescue area in the city's three-dimensional platform; Positioning the designated area of ​​the urban three-dimensional platform according to the positioning information, so as to mark a core positioning area corresponding to the emergency rescue area on the urban three-dimensional platform, and scheduling a regional model of the urban three-dimensional platform according to the core positioning area, so as to obtain a core three-dimensional model corresponding to the core positioning area; Based on the core positioning area, a multi-dimensional correlation hierarchical analysis is performed on the urban three-dimensional platform to obtain the multi-dimensional correlation characteristics of the core positioning area on the urban three-dimensional platform, and model characteristics of the urban three-dimensional platform are extracted and scheduled according to the multi-dimensional correlation characteristics to obtain an additional three-dimensional model, and the additional three-dimensional model is attached to the core three-dimensional model to obtain the regional three-dimensional model; wherein the multi-dimensional correlation hierarchy includes a key building distribution hierarchy, a surrounding traffic road hierarchy and an underground pipeline network hierarchy.

2. The emergency rescue method based on unmanned aerial vehicle and urban three-dimensional platform as claimed in claim 1, characterized in that: The steps of analyzing the detection scheme of the urban three-dimensional platform to obtain the drone detection scheme include: Get the formation structure information and current position information of the UAV formation; Performing analysis and processing of detection nodes on the regional three-dimensional model to obtain the detection node distribution characteristics, and performing node allocation processing on the detection node distribution characteristics according to the formation structure information to obtain a detection operation trajectory set corresponding to the UAV formation; wherein the detection operation trajectory set includes a plurality of detection operation trajectories, and the detection operation trajectory is used to describe the operation trajectory of the UAVs in the UAV formation for detecting multiple detection nodes; Calculating the flight path of the detection trajectory set to reach the initial node according to the current position information of the UAV formation, so as to obtain the initial flight path set of the UAV formation; The detection operation trajectory set and the initial flight path set of the drone formation are combined and processed to obtain a drone detection plan for the drone formation.

3. The emergency rescue method based on unmanned aerial vehicle and urban three-dimensional platform as claimed in claim 2, characterized in that: The steps of driving the drone formation to the emergency rescue area to perform regional detection according to the drone detection scheme to obtain real-time regional information of the emergency rescue area include: According to the initial flight path set in the drone detection scheme, the drone formation is driven to fly to the position of the corresponding initial detection node in the emergency rescue area, so that the drones in the drone formation collect information from the detection node through the sensor group carried by them to obtain the node detection information of the detection node; wherein the sensor group includes a camera sensor, an infrared sensor and a laser sensor, and the node detection information includes camera information, thermal imaging information and laser information; According to the detection operation trajectory set in the drone detection scheme, the drone formation is driven to fly to each of the detection nodes in sequence, and when at each of the detection nodes, information is collected from the detection nodes through the sensor group carried by the drone to obtain node detection information of each of the detection nodes; The node detection information of each detection node together constitutes the real-time area information of the emergency rescue area; After completing a round of node information detection corresponding to the detection trajectory set, the detection trajectory set is subjected to trajectory correction processing according to the acquired real-time area information to obtain a subsequent detection trajectory set, and the drone formation is driven to perform cyclic node detection on the emergency rescue area according to the subsequent detection trajectory set to update the real-time area information in real time.

4. The emergency rescue method based on unmanned aerial vehicle and urban three-dimensional platform as claimed in claim 1, characterized in that: The steps for pre-training the accident situation image recognition algorithm include: Construct a convolutional neural network model consisting of an input layer, a convolutional layer, a pooling layer, several fully connected layers, and an output layer; Prepare training data for model training of the convolutional neural network model; wherein the training data includes drone detection data and accident condition characteristics of various emergency rescue accidents, the drone detection data is image data detected by the drone formation of the emergency rescue area, and the accident condition characteristics are accident conditions corresponding to the drone detection data, and the accident conditions include fireworks conditions, fire blocking conditions, emergency exit blocking conditions, urban waterlogging conditions, dangerous slope conditions, dangerous building conditions, illegal construction conditions on the roof of buildings, floating objects in rivers, and tilted high-voltage towers; Substituting the training data into the input layer; The input layer receives the training data and transmits the training data to the convolution layer, and the convolution layer is used to collect spatiotemporal correlation features of the training data to obtain the spatiotemporal correlation features of the training data; The pooling layer is used to perform a pooling operation on the spatiotemporal correlation features of the training data extracted by the convolution layer to reduce the size of the spatiotemporal correlation features of the training data and retain key features of the spatiotemporal correlation features of the training data; Each of the fully connected layers is used to perform continuous vector flattening processing on the spatiotemporal correlation features of the training data processed by the pooling layer, so as to flatten the spatiotemporal correlation features of the training data into one-dimensional vector features; wherein the one-dimensional vector features are used to perform basic graphical expression on the spatiotemporal correlation features of the training data; The output layer receives and outputs the one-dimensional vector feature of the spatiotemporal correlation feature, and uses the one-dimensional vector feature of the spatiotemporal correlation feature as an accident status identification feature; A mapping algorithm unit is generated based on the accident condition identification feature, and the algorithm object and algorithm target of the mapping algorithm unit are deduced according to the urban three-dimensional platform to obtain an algorithm object unit and an algorithm target unit of the mapping algorithm unit; wherein the algorithm object unit is used to substitute real-time regional information, and the algorithm target unit is used to output accident condition features; The mapping algorithm unit, the algorithm object unit and the algorithm target unit are combined to obtain an accident condition image recognition algorithm.

5. The emergency rescue method based on unmanned aerial vehicle and urban three-dimensional platform as claimed in claim 1, characterized in that: The steps of substituting the accident condition characteristics into the regional three-dimensional model, and causing the regional three-dimensional model to simulate the accident condition according to the accident condition characteristics to obtain an accident simulation model for digitally feeding back the accident condition in the emergency rescue area include: Substituting the accident condition characteristics into the regional three-dimensional model; The three-dimensional model of the area is required to digitally simulate the smoke and fire conditions, fire blocking conditions, safety exit blocking conditions, urban waterlogging conditions and dangerous slope conditions of the emergency rescue area according to the characteristics of the accident conditions, so as to obtain an accident simulation model for digital feedback of the accident conditions in the emergency rescue area.

6. The emergency rescue method based on unmanned aerial vehicle and urban three-dimensional platform as claimed in claim 1, characterized in that: The method further includes substituting the real-time regional information into the regional three-dimensional model, so that the regional three-dimensional model simulates accident conditions at more angles according to the real-time regional information to supplement the accident simulation model, wherein the steps include: The regional three-dimensional model is used to digitally simulate the building damage and building abnormality conditions according to the real-time regional information to obtain building damage and abnormality distribution characteristics based on the regional three-dimensional model; The regional three-dimensional model is used to digitally simulate the distribution of the population in the region according to the real-time regional information, so as to obtain the distribution characteristics of the population based on the regional three-dimensional model; The regional three-dimensional model performs digital simulation of the meteorological environment conditions of the emergency rescue area according to the real-time regional information to obtain meteorological environment characteristics based on the regional three-dimensional model; According to the building damage and abnormal distribution characteristics, the crowd distribution characteristics and the meteorological environment characteristics, the three-dimensional model of the area is corrected and the model blocks are marked to supplement the accident simulation model.

7. The emergency rescue method based on unmanned aerial vehicle and urban three-dimensional platform as claimed in claim 1, characterized in that: It also includes performing an expansion analysis on the accident simulation model through a pre-trained urban accident expansion analysis algorithm to obtain a regional accident expansion model for performing expansion analysis and digital feedback on locations in the emergency rescue area that cannot be directly detected by the drone formation; performing trend prediction on the accident simulation model through a pre-trained urban accident change analysis algorithm to obtain a regional accident trend model for performing trend analysis and digital feedback on accident conditions in the emergency rescue area; The accident simulation model, the regional accident expansion model and the regional accident trend model are subjected to model combination analysis to obtain an emergency rescue strategy for the emergency rescue area.

8. An emergency rescue system based on drones and urban three-dimensional platform, characterized in that: Used to implement an emergency rescue method based on a drone and a three-dimensional urban platform as described in any one of claims 1-7.

Images