A method and system for real-time personnel positioning and rescue path planning in emergency situations

Abstract

This invention discloses a method and system for real-time personnel location and rescue route planning in emergency situations, belonging to the field of emergency response technology. The method utilizes multi-source heterogeneous data from satellite remote sensing, UAVs, and ground sensors, employing deep data fusion and an improved multimodal Transformer algorithm to achieve precise personnel location and behavioral status recognition. Dynamic patch analysis and an adaptive hierarchical network are used to dynamically model and update hazardous areas in real time. Combining intent recognition and priority decision-making models, graph neural networks and reinforcement learning are used to dynamically evaluate the priority of rescue targets and tasks. Based on this, a multi-objective hybrid ant colony algorithm and the A* algorithm are used to plan the optimal rescue route, and rescue teams are dynamically dispatched according to task and resource status, significantly improving the efficiency and safety of emergency rescue.

Description

Technical Field

[0001] This invention belongs to the field of emergency response technology, and more specifically relates to a method and system for real-time personnel positioning and rescue route planning in emergency situations. Background Technology

[0002] In recent years, with the continuous expansion of urban construction and the increase in population density, emergencies such as natural disasters, accidents, or the threat of war have placed higher demands on civil defense engineering and national defense mobilization systems. In emergency situations, timely and accurate personnel location and rescue route planning play a crucial role in ensuring personnel safety and improving rescue efficiency. Traditional civil defense engineering and national defense mobilization often rely on static maps or single data sources for personnel distribution analysis and route planning, which is difficult to adapt to complex and ever-changing real-world scenarios. Furthermore, in emergency environments, issues such as information silos, data delays, and dynamic changes in dangerous areas can severely impact decision-making and response speed.

[0003] With the development of multi-source heterogeneous data acquisition technologies such as satellite remote sensing, drone aerial photography, and ground sensors, and the widespread application of artificial intelligence in spatial modeling, data fusion, and intelligent decision-making, how to effectively apply these new technologies to emergency response in civil defense engineering and to build a real-time personnel positioning and intelligent rescue route planning system has become a crucial issue that urgently needs to be addressed in the fields of civil defense engineering and national defense mobilization. Current technologies lack a complete system capable of integrating multi-source data, dynamically identifying personnel status, intelligently modeling hazardous areas, and optimizing rescue routes in real time, thus limiting the efficiency and safety of emergency response. Therefore, developing a real-time personnel positioning and rescue route planning system for emergencies has significant practical importance and application value for improving the intelligence level and emergency rescue capabilities of civil defense engineering and national defense mobilization. Summary of the Invention

[0004] This invention addresses the challenges of timely and accurate acquisition and identification of the dynamic location and status of personnel during emergencies. It also highlights the real-time changes in hazardous areas as events unfold, making it difficult for traditional path planning methods to adapt to complex and dynamic environments and to flexibly optimize task priorities and rescue routes based on actual rescue needs. This results in low rescue efficiency and high safety risks. To overcome these difficulties, this invention proposes a method that integrates multi-source heterogeneous data, dynamically models hazardous areas, prioritizes tasks based on task intent recognition, and adaptively plans and schedules rescue routes. This method aims to improve the efficiency and safety of emergency rescue in complex environments.

[0005] To achieve the above objectives, the present invention employs the following technical solution: the method comprises:

[0006] Data fusion and dynamic personnel identification, through real-time collection of multi-source heterogeneous data, the collected raw data is processed by deep data fusion algorithms to make up for the limitations of a single data source;

[0007] Dynamic modeling and real-time updating of hazardous areas: Using dynamic patch analysis and adaptive hierarchical network construction algorithms, a spatial model of hazardous areas under emergencies is generated. An event-driven spatiotemporal network adjustment mechanism is introduced to automatically optimize the boundary of hazardous areas based on real-time sensor feedback and potential hazard source diffusion models, and effective avoidance zones and access obstacle areas are marked in real time.

[0008] The rescue priority ranking based on intent recognition incorporates the mission intent, injury status, and importance level information of the national defense mobilization personnel into the calculation through a priority decision model. An algorithm combining an improved graph neural network (GNN) and reinforcement learning is used to evaluate the priority order of each rescue target in real time.

[0009] Adaptive rescue path planning and scheduling employs a path planning method combining multi-objective hybrid ant colony and A* algorithm to achieve adaptive rescue path optimization based on dynamic danger zones, personnel distribution, and priorities.

[0010] In one approach, the data fusion and dynamic personnel identification include: real-time acquisition of multi-source heterogeneous data using satellite remote sensing, UAV aerial photography, and ground-based sensors and special equipment for national defense mobilization; and spatial-temporal preprocessing, feature extraction, and multi-modal feature fusion of the raw information from different data sources using a deep data fusion algorithm based on an improved multimodal Transformer architecture. At the same time, a dynamic weight allocation mechanism is adopted to improve the accuracy of personnel positioning and behavioral status identification.

[0011] In one approach, the dynamic modeling and real-time updating of the hazardous area includes: using dynamic patch analysis and adaptive hierarchical network construction algorithms to model the spatial state map of personnel coordinates and hazard signals obtained from multi-source data fusion and dynamic personnel identification; and dynamically optimizing the boundary of the hazardous area and marking the avoidance zone and access obstacles through spatial adaptive clustering, event-driven spatiotemporal network adjustment mechanism and risk feedback and boundary correction mechanism.

[0012] In one approach, the rescue priority ranking based on intent recognition includes: inputting multimodal information such as the personnel's task intent, injury status, and importance level through a priority decision model, and then integrating task instructions, health status, and job attributes through an intent recognition module;

[0013] A heterogeneous dynamic graph structure is constructed, and a graph neural network and reinforcement learning algorithm are combined. The attention mechanism dynamically evaluates the priority of each rescue target, and automatically optimizes the allocation and response sequence of rescue teams based on personnel attributes, cooperation relationships, scene changes and distribution of rescue resources, so as to achieve optimal allocation and response ranking of multiple objectives of rescue missions.

[0014] In one scheme, the adaptive rescue path planning and scheduling includes: based on a dynamic hierarchical network and the latest personnel distribution, danger zone and priority information, a path planning method combining multi-objective hybrid ant colony and A* algorithm is adopted to perform weighted directed graph modeling on the discrete grid of the spatial environment, and to optimize the safety path and task allocation of the rescue team in real time by comprehensively considering dynamic edge weights such as distance, danger level, and terrain obstacles.

[0015] By combining task priority, resource status, and scenario feedback, rescue teams can be dynamically dispatched, and the best routes, hazard warnings, and dispatch suggestions can be pushed to the terminal in a visual manner, enabling efficient, rapid, and safe rescue operations in complex environments.

[0016] In one approach, the dynamic modeling and real-time updating of hazardous areas are implemented, while an adaptive hierarchical network is used to achieve multi-level environmental modeling. This supports automatic adjustment of regional states and physical field simulation, enabling precise dynamic grading of hazardous areas and real-time labeling of safety passages, thus providing a reliable spatial reference for rescue route planning.

[0017] On the other hand, a real-time personnel location and rescue route planning system for emergencies is provided. The system is applicable to the method described above. The system includes: a data fusion and dynamic personnel identification module, a dynamic modeling and real-time updating module for dangerous areas, a rescue priority ranking module based on intent recognition, and an adaptive rescue route planning and scheduling module.

[0018] The data fusion and dynamic personnel identification module achieves personnel positioning and behavioral status identification through real-time acquisition of multi-source heterogeneous data and deep data fusion algorithms;

[0019] The hazardous area dynamic modeling and real-time update module uses dynamic patch analysis and adaptive hierarchical network construction algorithms to generate a spatial model of the hazardous area, and automatically optimizes the boundary of the hazardous area and marks the avoidance zone and the access obstacle area in real time through an event-driven spatiotemporal network adjustment mechanism and risk feedback.

[0020] The rescue priority ranking module based on intent recognition uses a priority decision model to input information such as personnel's task intent, injury status, and importance level. It employs a graph neural network and reinforcement learning algorithm to evaluate the priority of rescue targets in real time and automatically optimize the allocation of rescue teams and response sequences.

[0021] The adaptive rescue route planning and scheduling module optimizes the route and task allocation of rescue teams in real time based on personnel distribution, dangerous areas and priority information, using a route planning method that combines multi-objective hybrid ant colony and A* algorithm. It also achieves efficient and safe rescue operations through dynamic scheduling and visual push notifications.

[0022] Beneficial effects of this invention:

[0023] This invention, by fusing multi-source heterogeneous data and dynamically modeling hazardous areas, combined with task intent recognition and adaptive path planning, effectively improves the emergency response efficiency and rescue safety in civil defense engineering and national defense mobilization during emergencies. This invention can acquire and analyze the dynamic location and status information of on-site personnel in real time and accurately, promptly identify and warn of changes in hazardous areas, and achieve intelligent prioritization of rescue tasks and scientific and flexible path optimization. This significantly reduces rescue blind spots and delays, lowers personnel risks during rescue operations, and improves the rationality of resource allocation.

[0024] This system not only adapts to complex and ever-changing emergency environments, but also has good scalability and compatibility. It helps to promote the intelligent upgrading of emergency response systems in the fields of civil air defense engineering and national defense mobilization, enhances their overall disaster prevention and rescue capabilities, and has good application and promotion prospects and social benefits. Attached Figure Description

[0025] Figure 1 This is a flowchart of the method of the present invention;

[0026] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0027] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Typical embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0028] Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. To facilitate understanding, the invention will now be described more fully with reference to the accompanying drawings. Typical embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to make the disclosure of the invention more thorough and complete.

[0029] like Figure 1 As shown, a method for real-time personnel location and rescue route planning in emergency situations includes:

[0030] Step 1: Data Fusion and Dynamic Personnel Identification

[0031] First, real-time collection of multi-source heterogeneous data, including satellite remote sensing, UAV aerial photography, and ground sensors (RFID, infrared, life detectors), was used to collect personnel location data. The collected raw data was then processed by a deep data fusion algorithm—based on an improved multimodal Transformer architecture—which overcomes the limitations of single data sources and efficiently identifies personnel location and behavioral status. This algorithm can also automatically adapt to environmental changes during emergencies, improving the accuracy of personnel identification and laying the foundation for subsequent route planning.

[0032] In the field of national defense mobilization, the accuracy and real-time nature of personnel location in emergencies are crucial. However, traditional methods based on a single data source are insufficient to meet the demands of complex environments and dynamic scenarios. Therefore, this method adopts multi-source heterogeneous data fusion and dynamic personnel identification as its first innovative step. It systematically utilizes satellite remote sensing, UAV aerial photography, and ground sensors (including RFID, infrared life detectors, etc.) for spatiotemporal coordinated data collection to maximize the ability to capture personnel targets in different environments.

[0033] First, for the visual data from satellites and drones, spatial and temporal synchronized preprocessing is performed. Coordinate correction and data cleaning algorithms are then used to eliminate errors caused by differences in resolution, viewing angle, or acquisition time between different sensors. Assume the data source set is... The data collected from each data source can be denoted as... After synchronous processing, normalized spatial-temporal points are generated. .

[0034] The next step is data fusion. A multimodal Transformer architecture is constructed for national defense mobilization scenarios. Assuming there are N different modalities (e.g., radar, vision, infrared), each modal input is processed by a feature extractor to generate a feature vector. Traditional single-modal aggregation often loses cross-modal correlation information. Therefore, we define a multi-modal feature fusion mechanism: first, a linear transformation and an encoding layer are used to map each type of feature, specifically... ,in , It is learned through training. Simultaneously, to fully utilize the information interaction between modalities, a multi-path self-attention mechanism is introduced. Let the total feature concatenation be... The self-attention formula is used:

[0035]

[0036] Among them Obtained from the mapping of F, The feature dimension is defined as follows. Multimodal attention is computed not only within a modality but also between modalities, thus capturing the complex relationships influencing human behavior through multiple source signals. To enhance the model's adaptability to changes in data modality validity under sudden environmental conditions, this method introduces a dynamic weight allocation mechanism. Let the real-time reliability of each modality be denoted as... Credibility is assessed by combining historical error, time-varying noise, and current signal quality, and is defined as follows:

[0037]

[0038] Among them Current modal signal quality, The mean of other modalities in the same batch is... Adjust the parameters. Weight the fused feature vectors as a whole:

[0039]

[0040] By dynamically adjusting the weights of different data sources, the system can still adaptively allocate signal resources and maintain accurate and stable positioning even if some sensors temporarily fail.

[0041] Personnel localization and identification uses multimodal fusion features as input and employs a scalable spatial-behavior joint recognition network. The network output not only provides the spatial location of the personnel, but also... It also predicts behavioral states simultaneously. The location output uses a regression method, and the behavior recognition uses a classification method. The overall loss function is:

[0042]

[0043] in For real location and behavior, For the predicted results, The weight parameters are used. During model training, a small amount of manually labeled data is adaptively combined with a large amount of unsupervised data. The model simulates emergency scenarios (such as nighttime, smoke, and cover) using the actual data collection environment in national defense mobilization scenarios, which significantly enhances the model's generalization ability to special environments.

[0044] In practical applications, the scene-adaptive inference module allows the model to automatically adjust its area perception strategy based on real-time input. For example, when satellite remote sensing is limited, priority is given to activating drone aerial photography and ground sensors; when signal interference is severe in densely populated areas, the weight of RFID, infrared, and other sensor data is increased. Furthermore, a rapid adaptation mechanism based on meta-learning is introduced, setting up small-batch burst sample sets. Rapidly fine-tuning model parameters enables The model can achieve rapid adaptation loss by minimizing the loss:

[0045]

[0046] Enables real-time decision-making in emergency situations.

[0047] Step 2: Dynamic Modeling and Real-time Updates of Hazardous Areas

[0048] After obtaining current personnel distribution data, a spatial model of hazardous areas under emergencies is generated using dynamic patch analysis and an adaptive hierarchical network construction algorithm. This algorithm overcomes the limitations of traditional static maps by introducing an event-driven spatiotemporal network adjustment mechanism. It can automatically optimize hazardous area boundaries based on real-time sensor feedback and potential hazard diffusion models (such as chemical leaks and explosion ripples), and instantly mark effective refuge zones and areas with access obstacles. This dynamic modeling provides real-time and effective spatial references for rescue routes.

[0049] Following multi-source heterogeneous data fusion and dynamic personnel identification, effective rescue in disaster environments depends not only on the accurate location of trapped personnel but also on the dynamic modeling and real-time updating of hazardous areas. Therefore, this method employs dynamic patch analysis combined with an adaptive hierarchical network construction algorithm to achieve refined spatiotemporal modeling of hazardous areas. First, the coordinate set of trapped personnel output by the aforementioned positioning system is overlaid with spatial data from various hazard signals to form a spatiotemporal state map. The vertex set V represents a discrete spatial grid or important regions, the edge set E reflects the connectivity or barrier relationships between regions, and T is the time axis of event occurrence. The spatial grid is defined by a coordinate system. Discrete, each grid point records aggregated sensor data (such as temperature, gas concentration, radiation level, etc.), denoted as .

[0050] The main algorithm is based on the concept of dynamic patch segmentation, performing spatial adaptive clustering on the raw sensor data stream and environmental field changes. Let the attribute vector of the region of interest at any given time be... By applying a Gaussian Mixture Model (GMM) clustering method to the global grid, several spatial "patches" were obtained. Each patch possesses a highly homogeneous distribution of hazard characteristics. The GMM (Gross Matrix) approach to patch classification can be expressed as maximizing the likelihood function:

[0051]

[0052] Where K is the number of patches, Normal distribution These are the model parameters. The distribution boundary is automatically estimated by the fitted Gaussian kernel diffusion radius, making the region dynamically blurred and adaptable to the diffusion of different types of disasters (such as gas, explosion, flame, collapse, etc.).

[0053] Building upon the modeling foundation, an event-driven spatiotemporal network adjustment mechanism and an adaptive hierarchical network structure are introduced. First, the dynamic evolution of the danger zone boundary over time is defined as follows:

[0054]

[0055] in The velocity field of hazard source diffusion is jointly determined by sensor feedback, physical diffusion theory, and disaster simulation models (such as CFD, electromagnetic leakage, second-order reaction-diffusion equations, etc.); the system... Real-time adaptive. Specifically, for regions with directional or complex terrain influences, a weighted directed graph is used to construct regional accessibility. Node edge weights. Given a combination of factors such as safety level, distance, and obstacles, it can be expressed as:

[0056]

[0057] in For distance, for Risk level gradient between two points The terrain / obstacle influence coefficient is used, and the weights of the three factors are dynamically adjusted according to the scenario.

[0058] Meanwhile, to prevent blind spots in hazard area detection caused by information distortion, clustering errors, or sensor failure, the method designs a risk feedback and boundary correction mechanism. For each patch boundary, Bayesian spatiotemporal filtering is combined to achieve closed-loop modeling based on "prior hazard field – sensor update – posterior correction". Assuming the patch... The probability of a dangerous label at time t is, New sensor observations Then the recursive formula for the posterior probability is:

[0059]

[0060] This probabilistic self-updating mechanism ensures that the boundaries of dangerous patches can quickly adapt to new changes and sensory inputs, improving sensitivity to sudden anomalies.

[0061] In the adaptive hierarchical network part, the system models the environmental space as a multi-level implicit graph. The bottom layer reflects various raw extreme signals, the next layer reveals clustered danger patches, and the middle and upper layers represent refuge zones, obstacles, and accessible rescue subnets. An automatic gating coupling mechanism is used between layers to achieve vertical information flow: a sudden change in the attributes of any lower-level patch affects the refuge zone labels in the upper layers, controlled by a state transition probability matrix. The inter-layer states are denoted as... The state transition probability is (Trained from historical disaster evolution data, or self-adjusted in real time during simulation exercises), the action space reflects the splitting, merging, or mobility switching of blocks.

[0062]

[0063] in, For different node states.

[0064] Once a sudden event (such as a secondary explosion, structural collapse, or a sudden change in wind direction) is detected by the system, dynamic adjustments will be triggered in the relevant high-risk area indicators of the layered network. The inference module calls the diffusion physics simulation function in real time to solve the reaction-diffusion partial differential equation.

[0065]

[0066] Where C is the concentration of the hazardous substance. Let be the diffusion coefficient, and be the diffusion coefficient. Source term, automatically corrected by combining sensor feedback parameters. And push the latest results in real time to ensure that the boundaries are adjusted quickly according to the actual evolution.

[0067] For rescue missions, as the final output, the system will automatically annotate the following information on the spatial model: 1) abruptly hazardous areas and their confidence levels; 2) currently known efficient avoidance zones and safe passageways; 3) obstacles, buffer zones, and progressively dangerous zones generated in real time due to regional evolution. The specific output is expressed in a graph structure and dynamically colored according to real-time feedback. The rescue route planning layer can directly provide a highly reliable spatiotemporal reference for emergency routes based on this multi-level safety network.

[0068] Step 3: Prioritizing Rescue Efforts Based on Intent Recognition

[0069] The priority decision-making model incorporates information such as the mission intent, injury status, and importance level of national defense mobilization personnel into its calculations. It employs an algorithm combining an improved graph neural network (GNN) and reinforcement learning to evaluate the priority of each rescue target in real time. Based on dynamic mission objectives and diverse data inputs, the model automatically adjusts the allocation and response sequence of rescue teams, improving overall rescue efficiency and ensuring optimal resource allocation.

[0070] Based on dynamic modeling and real-time updates of dangerous areas, the priority ranking of rescue in national defense mobilization scenarios cannot rely solely on traditional criteria such as "injury level" or "distance," but must comprehensively consider personnel's mission intentions, vital signs, job importance, and even dynamic changes in the scenario.

[0071] To this end, with an intent recognition-based priority decision-making model as the core, the system integrates multimodal data (personnel identity, location, health status, task instructions, etc.) and scene features to construct a real-time adaptive graph neural network (GNN) and reinforcement learning collaborative algorithm to improve the scientific nature of rescue response and the optimal allocation of resources.

[0072] First, personnel nodes and their attributes are abstracted into a heterogeneous dynamic graph structure in the system. Among them, the node set Represents all personnel and critical equipment requiring rescue; edge set E represents collaboration / dependency or geographical association between personnel (such as a small group, linked equipment, or spatial proximity); attribute set X contains health status. Injury level Importance of the position Task Intent The data includes dimensions such as on-site intelligence. Mission intent data is generated from real-time pushes from the national defense mobilization command system, processed using natural language understanding and normalization, and mapped into vectors. For example, the mission intent vector for an emergency communicator might emphasize "high importance and contactability," while for an injured soldier it might reflect "urgent physiological need for medical treatment."

[0073] The intent recognition module processes task instruction text, speech, or on-site alarm information using a multimodal Transformer or a pre-trained language model (such as BERT or GPT), fusing deep semantic features with personnel status into a comprehensive intent representation. The formula is:

[0074]

[0075] in It is a feature fusion neural network. All node attributes are embedded into a graph structure to form a node feature matrix. At the same time, the edge weights between nodes Dynamically calculated based on collaboration requirements, relative spatial locations, and dependencies:

[0076]

[0077] in Assess intent similarity. Let be the distance function, and be Importance weight of collaboration Adjust parameters for the scene.

[0078] Based on the above graph structure, the system employs an improved graph neural network (GNN) to encode the interaction attributes between personnel and the scene, and introduces an attention mechanism to capture the influence of key nodes. The GNN node update formula is:

[0079]

[0080] in Let i be the set of neighbors of node i. The attention weights are determined by the edge weights and attributes. For learnable parameters, Activation function. Attention weights are adaptively calculated as follows:

[0081]

[0082] After multiple rounds of GNN propagation, node features are deeply integrated with diverse information such as global and local intent, health status, and on-site correlation.

[0083] Prioritization is essentially a multi-objective decision optimization process. A "rescue value function" is defined for each personnel node. It integrates information such as the urgency and importance of the injury, the probability of mission execution, and the accessibility of evacuation. This can be formalized as:

[0084]

[0085] in To normalize the injury level, The importance of the position, The model identifies the urgency of the intent. Score the reachability from the node to the safe zone. The weights can be adaptively adjusted. The entire sorting is then determined by the GNN output vector. Mapped to priority scores via linear projection ,Right now

[0086]

[0087] The rescue order queue is generated by a sorting algorithm (such as Heap, Top-K or greedy algorithm).

[0088] To enhance the scientific rigor and adaptability of the ranking process, a reinforcement learning (RL) mechanism is introduced to optimize the overall team allocation based on the rescue performance of the current priority queue. The RL environment is set as a real-world rescue scenario, with the state space... This includes personnel priority scores, team resource distribution, and changes in hazardous areas; action space. For different rescue strategies (such as decentralized and centralized approaches, prioritizing high-risk individuals, and optimizing boundaries), the reward function *r* is dynamically defined based on rescue effectiveness, time, resource consumption, and post-rescue evaluation. For example, for real-time rescue missions, the reward can be set as follows:

[0089]

[0090] in For the number of successful rescues, Total task time Resource consumption, three factors are dynamically unified. RL uses a model to adaptively adjust the response sequence of the rescue team, and feeds back the priority ranking and resource allocation matrix after each round of execution, interacting and cooperating with the GNN output to achieve end-to-end closed-loop optimization.

[0091] The overall process is as follows: 1) Input multimodal data such as personnel attributes, scene changes, and command tasks from the global dynamic graph structure and patch modeling; 2) Map abstract semantic features through the intent recognition model; 3) Construct an adaptive graph containing node attributes and collaborative edge weights, and calculate the priority comprehensive score of each node through multiple rounds of GNN propagation and attention fusion; 4) Utilize a joint reinforcement learning strategy to optimize the allocation and sorting of queues online based on team resources and real-world feedback, and finally output the dynamic response sequence of the rescue team.

[0092] Step 4: Adaptive Rescue Route Planning and Scheduling

[0093] Building upon the achievements of the preceding steps, a path planning method combining multi-objective hybrid ant colony optimization and the A* algorithm is employed to achieve adaptive rescue path optimization based on dynamic danger zones, personnel distribution, and priorities. This algorithm can adjust the path in real time based on the latest map data and the status of the rescued individuals, avoiding the slow response of conventional algorithms to emergencies. The system pushes the optimal path and scheduling suggestions to the rescue personnel's terminals in real time, ensuring rapid and safe rescue operations.

[0094] This stage not only handles complex and dynamic environments but also integrates the latest personnel distribution, regional hazard information, and response priorities to transform mission objectives into efficient, safe, and real-time rescue execution plans. This method innovatively combines a multi-objective hybrid ant colony algorithm with the A* algorithm. It leverages the global search and multi-objective optimization capabilities of the ant colony algorithm while utilizing the heuristic and efficient local optimization of the A* algorithm, thereby greatly enhancing the adaptability and timeliness of path planning and ensuring optimal action for rescue teams in complex and dynamic environments.

[0095] The system modeling is based on the aforementioned dynamic hierarchical network and hazard zone patches, with the spatial environment discretized into a weighted directed mesh graph. Node V represents a spatial grid point, edge set E represents adjacent feasible paths, and edge weight W dynamically reflects distance, hazard level, drivability, and time-varying environment. Specifically, the edge weight model is as follows:

[0096]

[0097] in The distance between the two nodes. Risk scores are assigned to hazardous areas along the route (updated in real time based on the aforementioned dynamic patch model). The terrain obstacle coefficient, weight Adaptive adjustments are made to achieve detailed modeling of the space and scene. The starting point of each rescued object and team is mapped to the starting and destination points in the diagram.

[0098] The path planning algorithm primarily employs Multi-Objective Hybrid AntColony Optimization (MO-HACO), embedding A* heuristic fine-tuning as a local optimum subprocess. The main process is as follows: First, the system assigns tasks (i.e., target nodes or target sets) to each rescue team based on the latest priority ranking. Then, each path task undergoes ant colony initialization on the spatial network. Each ant represents a candidate rescue path, starting at the team's current position and ending at the target rescue object. During movement, the ant's decision probability is based on a comprehensive heuristic function from the current node to its neighbors and the pheromone distribution, defined as:

[0099]

[0100] in Let be the probability that the k-th ant goes from i to j. For pheromone intensity, For multi-objective heuristic evaluation functions, For adjustable weights, This represents the current feasible neighbor set. A multi-objective heuristic function integrates distance and safety:

[0101]

[0102] in Updated in real time by animated GIFs. To prevent small quantities with a denominator of zero.

[0103] After each round of path search, the entire ant colony evaluates the candidate paths found based on the current set of candidate paths, combined with a multi-objective fitness function:

[0104]

[0105] in The total distance is... Accumulate hazard values ​​for the path. For time estimation, Weights are uniformly and adaptively adjusted to reflect real-time rescue needs. To consider priorities, on-site conditions, or team characteristics, reward items adjusting for the priority of the rescued individuals can be introduced. Pheromones updates employ a combination of global and local mechanisms.

[0106]

[0107]

[0108] in Here, Q is the pheromone evaporation coefficient, and Q is the total amount of pheromones. It is the multi-objective fitness of the k-th path.

[0109] In response to sudden changes in the dynamic environment (such as danger zone expansion, personnel movement, and obstacle emergence), the system can immediately trigger local reconstruction of relevant paths upon detecting map changes. During this process, the A* algorithm acts as an embedded "corrector," making real-time fine-tuning of path segments where significant changes occur in the affected area or dynamic boundaries. The A* heuristic path function is:

[0110]

[0111] in Let n be the true cost from the starting point to the current node n (e.g., time, danger-weighted distance). A* estimates the residual cost to the endpoint (with real-time access to the latest map and hazard rating). A* local interpolation can significantly shorten update response time and avoid delays caused by global recalculation.

[0112] In the computation phase, the parallelism of the hybrid algorithm is fully utilized. Each rescue team's path optimization task can be completed independently and in real-time in parallel on a distributed platform or multi-core system, ensuring high concurrency and low latency path output in large-scale scenarios. Simultaneously, the scheduling module combines the aforementioned priority scores, remaining energy of the rescue teams, their positions, and current path costs, employing a "queue-based dynamic scheduling" principle: continuously evaluating each team.

[0113]

[0114] Pick The system assigns tasks based on the maximum allocation and adapts team / task combinations in real time according to feedback and changes in the scenario.

[0115] The system outputs data visualizations, pushing the current optimal rescue route and dispatch suggestions to rescue personnel via smart terminals. The interface intuitively displays key information such as the shortest safe route, warning zones for potentially dangerous road sections, recommended alternative routes, and task switching reminders. If on-site operators deviate from the suggested route, the system can automatically "replan" and generate a corrected route based on GPS / IoT positioning and environmental perception data, providing dynamic prompts to ensure that the emergency rescue team always advances along the optimal strategy in dangerous and ever-changing situations.

[0116] This method overcomes the limitations of conventional Dijkstra's algorithm or static A* algorithm in handling dynamic, large-scale rescue scenarios by fusing multi-objective hybrid ant colony optimization with real-time A* data. This significantly improves the response capability to emergencies and the efficiency of resource allocation. Actual engineering experiments demonstrate that the system can continuously output high-quality, multi-constraint adaptable rescue paths even under conditions of frequent adjustments in personnel distribution, environmental information, and command intentions, and achieve rapid and intelligent troop dispatch. Ultimately, it provides cutting-edge and robust operational support for various national defense mobilization emergency rescue missions.

[0117] like Figure 2 As shown, a real-time personnel location and rescue route planning system for emergencies is presented.

[0118] It includes: a data fusion and dynamic personnel identification module, a hazardous area dynamic modeling and real-time update module, an intent recognition-based rescue priority ranking module, and an adaptive rescue route planning and scheduling module;

[0119] The data fusion and dynamic personnel identification module achieves personnel positioning and behavioral status identification through real-time acquisition of multi-source heterogeneous data and deep data fusion algorithms.

[0120] The hazardous area dynamic modeling and real-time update module uses dynamic patch analysis and adaptive hierarchical network construction algorithms to generate a spatial model of the hazardous area, and automatically optimizes the boundary of the hazardous area and marks the avoidance zone and the access obstacle area in real time through an event-driven spatiotemporal network adjustment mechanism and risk feedback.

[0121] The rescue priority ranking module based on intent recognition uses a priority decision model to input information such as personnel's task intent, injury status, and importance level. It then uses a graph neural network and reinforcement learning algorithm to evaluate the priority of rescue targets in real time and automatically optimize the allocation of rescue teams and response sequences.

[0122] The adaptive rescue route planning and scheduling module optimizes the route and task allocation of rescue teams in real time based on personnel distribution, dangerous areas and priority information, using a route planning method that combines multi-objective hybrid ant colony and A* algorithm. It also achieves efficient and safe rescue operations through dynamic scheduling and visual push notifications.

[0123] Example:

[0124] 1. Scene Description

[0125] A 7.2-magnitude earthquake struck a certain area, causing severe damage to ground structures. The site environment was complex, with risks of aftershocks and chemical leaks. The national defense mobilization system rapidly collected information about the disaster area using multi-source heterogeneous data, including satellite remote sensing, drone aerial photography, and ground sensors. It was necessary to locate trapped personnel and scientifically plan rescue routes and task priorities for rescue teams to achieve efficient and precise rescue operations.

[0126] 2. Data Acquisition and Fusion

[0127] Multiple data acquisition nodes were deployed in the disaster area: satellite remote sensing (SR), unmanned aerial vehicle (UAV), ground sensors (GS), and dedicated positioning equipment for national defense mobilization (DM). The following sample data were collected:

[0128]

[0129] The data is fused using an improved multimodal Transformer algorithm with dynamic weight allocation to obtain accurate information on the location and behavioral status of individuals. For example, person A (34.124, 117.679) is identified as being in a distress call state, with a location error of less than 1 meter.

[0130] 3. Dynamic modeling and real-time updating of hazardous areas

[0131] Spatial modeling of the location of trapped personnel and danger signals was performed, and dynamic patch analysis and adaptive hierarchical network algorithm were used to divide the disaster area into multiple risk zones.

[0132]

[0133] Spatial adaptive clustering automatically adjusts boundaries and marks danger zones and obstacle zones in real time, providing rescue teams with refined spatial references.

[0134] 4. Rescue Prioritization Based on Intent Recognition

[0135] Multimodal information about the rescue targets is collected, including mission intent, injury status, job attributes, and importance level, and then input into the decision-making model.

[0136]

[0137] By employing graph neural networks and reinforcement learning, the priority order of rescue targets is dynamically optimized, and rescue teams and response sequences are automatically allocated to ensure that resources are prioritized for rescuing high-value targets and high-risk individuals.

[0138] 5. Adaptive rescue route planning and scheduling

[0139] Based on the latest information on dangerous areas, personnel distribution, and priorities, a directed graph model of the spatial environment is constructed using a multi-target hybrid ant colony and the A* algorithm.

[0140] Assuming rescue team 1 (starting point 34.120, 117.675) needs to rescue A (priority 1), the path planning process is as follows:

[0141]

[0142] Taking into account distance, hazard level, and terrain obstacles, the algorithm automatically selects the safest and optimal path.

[0143] Starting point > N1 > N3 (avoiding the high-risk N2 area, total distance 230 meters)

[0144] Simultaneous dispatch of rescue team resources:

[0145]

[0146] Rescue routes and hazard warnings are pushed to operators in real time via data terminals, ensuring the safety and efficiency of rescue teams in carrying out their missions.

[0147] 6. Automated and Visual Feedback

[0148] Real-time environmental data, personnel location, rescue routes, and mission scheduling are displayed through the visualization terminal of the National Defense Mobilization Command Center:

[0149]

[0150] Intelligent algorithms dynamically update rescue missions, routes, and priorities, enabling efficient, rapid, and safe rescue operations in complex environments and with ever-changing tasks.

[0151] This embodiment fully demonstrates the practical application effect of the present invention in emergency situations within the field of national defense mobilization. The closed-loop system, which integrates multi-source data fusion, real-time modeling of hazardous areas, intent recognition and priority ranking, and adaptive path planning and scheduling, significantly improves rescue efficiency and personnel safety.

[0152] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0153] It should be understood that the above detailed description of the technical solutions of the present invention with reference to preferred embodiments is illustrative and not restrictive. Those skilled in the art can modify the technical solutions described in the embodiments or make equivalent substitutions for some of the technical features based on reading this specification; however, these modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for real-time personnel positioning and rescue path planning in emergency situations, characterized in that: The method includes: Data fusion and dynamic personnel identification, through real-time collection of multi-source heterogeneous data, the collected raw data is processed by deep data fusion algorithms to make up for the limitations of a single data source; Dynamic modeling and real-time updating of hazardous areas: Using dynamic patch analysis and adaptive hierarchical network construction algorithms, a spatial model of hazardous areas under emergencies is generated. An event-driven spatiotemporal network adjustment mechanism is introduced to automatically optimize the boundary of hazardous areas based on real-time sensor feedback and potential hazard source diffusion models, and effective avoidance zones and access obstacle areas are marked in real time. The rescue priority ranking based on intent recognition incorporates the mission intent, injury status, and importance level information of the national defense mobilization personnel into the calculation through a priority decision model. An algorithm combining an improved graph neural network (GNN) and reinforcement learning is used to evaluate the priority order of each rescue target in real time. Adaptive rescue path planning and scheduling employs a path planning method combining multi-objective hybrid ant colony and A* algorithm to achieve adaptive rescue path optimization based on dynamic danger zones, personnel distribution, and priorities.

2. The method according to claim 1, wherein: The aforementioned data fusion and dynamic personnel identification include: real-time acquisition of multi-source heterogeneous data using satellite remote sensing, UAV aerial photography, and ground sensor-based national defense mobilization equipment; and spatial-temporal preprocessing, feature extraction, and multi-modal feature fusion of raw information from different data sources using a deep data fusion algorithm based on an improved multimodal Transformer architecture. At the same time, a dynamic weight allocation mechanism is adopted to improve the accuracy of personnel positioning and behavioral status identification.

3. The method of claim 1, wherein: The dynamic modeling and real-time updating of the dangerous area includes: using dynamic patch analysis and adaptive hierarchical network construction algorithms to model the spatial state map of personnel coordinates and danger signals obtained from multi-source data fusion and dynamic personnel identification; and dynamically optimizing the boundary of the dangerous area and marking the avoidance zone and passage obstacles through spatial adaptive clustering, event-driven spatiotemporal network adjustment mechanism and risk feedback and boundary correction mechanism.

4. The method of claim 1, wherein: The aforementioned rescue priority ranking based on intent recognition includes: inputting multimodal information such as personnel's task intent, injury status, and importance level through a priority decision model, and then integrating task instructions, health status, and job attributes through an intent recognition module; A heterogeneous dynamic graph structure is constructed, and a graph neural network and reinforcement learning algorithm are combined. The attention mechanism dynamically evaluates the priority of each rescue target, and automatically optimizes the allocation and response sequence of rescue teams based on personnel attributes, cooperation relationships, scene changes and distribution of rescue resources, so as to achieve optimal allocation and response ranking of multiple objectives of rescue missions.

5. The method of claim 1, wherein: The adaptive rescue path planning and scheduling includes: based on a dynamic hierarchical network and the latest personnel distribution, danger zone and priority information, a path planning method combining multi-objective hybrid ant colony and A* algorithm is adopted to perform weighted directed graph modeling on the discrete grid of the spatial environment, and to optimize the safety path and task allocation of the rescue team in real time by comprehensively considering distance, danger level and dynamic edge weight of terrain obstacles. By combining task priority, resource status, and scenario feedback, rescue teams can be dynamically dispatched, and the best routes, hazard warnings, and dispatch suggestions can be pushed to the terminal in a visual manner, enabling efficient, rapid, and safe rescue operations in complex environments.

6. The method of real-time personnel positioning and rescue path planning in emergency situations according to claim 1, characterized in that: The aforementioned dynamic modeling and real-time updating of hazardous areas, along with the adoption of an adaptive hierarchical network to achieve multi-level environmental modeling, support automatic adjustment of regional states and physical field simulation, realize fine dynamic classification of hazardous areas and real-time labeling of safety passages, and provide reliable spatial reference for rescue route planning.

7. A real-time personnel location and rescue route planning system for emergencies, wherein the system is applicable to the method as described in any one of claims 1-6, characterized in that: The system includes: a data fusion and dynamic personnel identification module, a hazardous area dynamic modeling and real-time update module, an intent recognition-based rescue priority ranking module, and an adaptive rescue route planning and scheduling module. The data fusion and dynamic personnel identification module achieves personnel positioning and behavioral status identification through real-time acquisition of multi-source heterogeneous data and deep data fusion algorithms; The hazardous area dynamic modeling and real-time update module uses dynamic patch analysis and adaptive hierarchical network construction algorithms to generate a spatial model of the hazardous area, and automatically optimizes the boundary of the hazardous area and marks the avoidance zone and the access obstacle area in real time through an event-driven spatiotemporal network adjustment mechanism and risk feedback. The rescue priority ranking module based on intent recognition uses a priority decision model to input personnel's task intent, injury status, and importance level information. It then uses a graph neural network and reinforcement learning algorithm to evaluate the priority of rescue targets in real time and automatically optimize the allocation of rescue teams and response sequences. The adaptive rescue route planning and scheduling module optimizes the route and task allocation of rescue teams in real time based on personnel distribution, dangerous areas and priority information, using a route planning method that combines multi-objective hybrid ant colony and A* algorithm.

Images