An optimization method and system for fusing unmanned aerial vehicle reconnaissance strategy and rescue dispatch

Abstract

The application relates to the technical field of earthquake emergency dispatch, and particularly discloses an optimization method and system combining an unmanned aerial vehicle reconnaissance strategy and rescue dispatch, which comprises the following steps: taking the sum of the number of A-class wounded persons waiting for transfer at each disaster point and the number of B-class wounded persons disposed on site by a rescue team in a full period as a first optimization target; taking the total accumulated waiting time of A-class and B-class wounded persons surviving in the disaster point at the end of each period from the completion of reconnaissance to the arrival of the rescue team as a second optimization target; adopting a lexicographic layered comparison criterion to preferentially guarantee the first target; combining a double target and a constraint condition to construct a rescue coordination model, and forcibly executing the time sequence coupling of reconnaissance first and rescue later of the unmanned aerial vehicle through the constraint; after linearization conversion of a nonlinear constraint and a target function is completed through mathematical transformation, the optimal unmanned aerial vehicle reconnaissance path and the rescue team dispatch scheme are solved and output; and the application provides an optimal dispatch scheme for an earthquake initial scene with communication interruption and secondary disaster risks.

Description

Technical Field

[0001] This invention relates to the field of earthquake emergency dispatch technology, and in particular to an optimized method and system that integrates unmanned aerial vehicle (UAV) reconnaissance strategy with rescue and dispatch. Background Technology

[0002] In recent years, extreme natural disasters have occurred frequently around the world, with earthquakes posing a severe test to emergency response capabilities due to their suddenness and destructive power. During the "golden rescue time" after an earthquake, the efficiency of treating and transferring the injured directly determines the survival rate. However, the widespread paralysis of post-earthquake communication infrastructure prevents real-time access to disaster information, forcing rescue decisions to be made under severe information gaps. Simultaneously, secondary disasters induced by earthquakes (such as landslides and mudslides) continue to emerge at different times after the earthquake, and the affected areas continue to increase, making the dispatching problem both dynamic and uncertain. While the academic community has explored various approaches to this type of complex dispatching problem, existing research still has significant shortcomings in dealing with scenarios where communication disruptions and secondary disasters overlap. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides an optimization method that integrates UAV reconnaissance strategies with medical dispatch. This method uses a multi-objective mixed integer programming model with the goal of maximizing the benefits of casualty treatment and minimizing service waiting time. By embedding the dynamic evolution of casualties and the reconnaissance-rescue time sequence logic through systematic linearization techniques, it provides an optimal dispatch scheme for the initial earthquake scenario where communication is interrupted and there is a risk of secondary disasters.

[0004] On the one hand, an optimized method integrating UAV reconnaissance strategy and medical dispatch is provided, including: Define the disaster site set, the rescue team set, and the drone set; The primary optimization objective is to maximize the sum of the number of Category A injured persons awaiting transfer at each disaster site and the number of Category B injured persons being treated on-site throughout the entire time period. The secondary optimization objective is to minimize the total cumulative waiting time of surviving Category A and Category B injured persons at the end of each time period from the completion of reconnaissance to the arrival of the rescue team. A hierarchical objective based on lexicographical order is used as the comparison criterion, prioritizing the primary optimization objective. A rescue coordination model is constructed based on the primary and secondary optimization objectives and constraints. Constraints ensure the temporal coupling of using drones for reconnaissance first, followed by treatment. The nonlinear constraints and the first and second optimization objective functions in the rescue coordination model are linearized using mathematical transformations. Solve the linearized rescue coordination model, output the optimal UAV reconnaissance path and rescue team dispatch plan, and evaluate them.

[0005] Preferably, the disaster point set includes: an initial set of communicable points, an initial set of non-communicable points, and a set of secondary disaster-prone points.

[0006] Preferably, the primary optimization objective is to maximize the sum of the number of Category A injured persons awaiting transfer at each disaster site and the number of Category B injured persons being treated on-site throughout the entire time period; specifically:

[0007] in, Non-negative integers, time period Inside, the rescue team At the disaster site Number of Category A wounded personnel on board; Non-negative integers, time period Inside, the rescue team At the disaster site The first disposal Number of Class B casualties; For the set of categories of deterioration time limits for Class B wounded, index ; For time period sets, ,in , Step size; Assemble the rescue team, index is ; For the set of disaster points, the index is .

[0008] Preferably, the second optimization objective is to minimize the total cumulative waiting time from the completion of reconnaissance to the arrival of the rescue team for Category A and Category B casualties surviving in the reconnaissance sites at the end of each time period; specifically:

[0009] in, The time interval step; Non-negative integers, time period Finally, disaster site The actual number of surviving Category A wounded; Non-negative integers, time period Finally, disaster site The real survival of the first Number of Class B casualties; Binary, time period Finally, disaster site Whether it has been detected is indicated by 1 if yes, and 0 otherwise.

[0010] Preferably, the constraint conditions include: a first constraint condition, which defines an S-shaped survival probability function. and the corresponding survival rate for each time period , This serves as the basis for calculating the attenuation of the dynamic evolution of the wounded:

[0011]

[0012]

[0013] in, Identify the types of injured persons; For the first The asymptotic maximum of the survival probability of such casualties; This is the half-life parameter, which represents the survival time when the survival rate drops to 50%. The shape parameter controls the steepness of the survival curve; The time the injured person has survived since the moment of activation; Let be the survival probability function for Class A casualties; For the first Survival probability function for Class B casualties.

[0014] Preferably, the constraints further include: a third constraint, which enforces the reconnaissance pre-processing logic to ensure that no rescue operations are carried out when the initial non-communication point and secondary disaster point are not reconnoitered, while limiting the total amount of Class A loading at any disaster point in a single operation to not exceed the rated transfer capacity.

[0015] Preferably, the rescue coordination model after linearization is solved to output and evaluate the optimal UAV reconnaissance path and rescue team scheduling scheme. Specifically, a two-stage randomized greedy search algorithm is used to solve the rescue coordination model. Through the outer loop, a UAV reconnaissance plan is generated. Then, through the inner loop, a rescue team scheduling scheme is generated and evaluated. Loop control and post-processing patching operations are performed to finally output and evaluate the optimal UAV reconnaissance path and rescue team scheduling scheme.

[0016] On the other hand, an optimized system integrating UAV reconnaissance strategy and medical dispatch is provided, including: The parameter definition module defines the disaster site set, the rescue team set, and the drone set. The rescue coordination model construction module has the following objectives: First, to maximize the sum of the number of Category A casualties awaiting transfer and the number of Category B casualties being treated on-site at each disaster site throughout the entire time period; second, to minimize the total cumulative waiting time from the completion of reconnaissance to the arrival of the rescue team for surviving Category A and Category B casualties at the end of each time period. A hierarchical objective based on lexicographical order is used as the comparison criterion, prioritizing the first objective. Based on the first and second objectives and constraints, a rescue coordination model is constructed. Constraints ensure the temporal coupling of using drones for reconnaissance first, followed by treatment. The transformation module uses mathematical transformations to linearize the nonlinear constraints and the first and second optimization objective functions in the rescue coordination model. The solution module solves the linearized rescue coordination model, outputs the optimal UAV reconnaissance path and rescue team dispatch plan, and evaluates them.

[0017] Furthermore, an electronic device is also provided, including: Memory, used for non-transitory storage of computer-readable instructions; and Processor, for executing the computer-readable instructions, When the computer-readable instructions are executed by the processor, they perform the method described in the first aspect above.

[0018] In another aspect, a storage medium is also provided for non-transitory storage of computer-readable instructions, wherein when the non-transitory computer-readable instructions are executed by a computer, the method described in the first aspect is performed.

[0019] The above technical solution has the following advantages or beneficial effects: To address the problem of the disconnect between reconnaissance and rescue due to unknown disaster information, the model proposed in this invention divides disaster area nodes into three categories: those that can communicate, those that cannot communicate, and those prone to secondary disasters. It also stipulates that all rescue operations can only be carried out after the nodes have completed UAV reconnaissance, thus achieving the temporal coupling of "reconnaissance first, rescue later". To address the lack of physiological time-dependent evidence for the decline in the survival rate of wounded soldiers, the model proposed in this invention introduces an S-shaped survival function to characterize the change in the survival rate of wounded soldiers over time, and drives the dynamic evolution of the number of wounded soldiers with discrete time periods. The actual initial number of wounded soldiers at the time of reconnaissance is completed is converted into the current stock after the survival time has been adjusted, so that the treatment decision strictly follows the process of the deterioration of the injury. To address the issues of a lack of unified trade-offs between dual objectives and insufficient coupling of heterogeneous tasks, this invention proposes a model that constructs a dual-objective framework to maximize the total number of medical personnel treated and minimize waiting time. It also comprehensively considers multiple constraints, including rescue team transfer capacity, medical resource consumption, and movement time, as well as limitations on UAV endurance and reconnaissance time, thus fully reflecting the resource competition and collaborative relationships among heterogeneous forces during mission execution. This paper aims to provide emergency command departments with an optimal decision-making solution integrating information reconnaissance, dynamic monitoring of casualty survival, and rescue dispatch in extreme post-earthquake scenarios characterized by communication disruptions and a continuous surge in secondary disasters. Attached Figure Description

[0020] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0021] Figure 1 This is a flowchart of the method in Example 1; Figure 2This is a vehicle dispatch diagram for Example 1 when the secondary disaster has not occurred; Figure 3 This is a vehicle dispatch diagram for Example 1 when a secondary disaster has occurred; Figure 4 This is a diagram showing the distribution of nodes in a 26-node test road network according to the method in Example 1. Figure 5 The algorithm convergence curve of the method in Example 1 is shown; where (a) is the optimal number of patients to be treated, and (b) is the corresponding waiting time. Figure 6 The spatiotemporal diagram of the reconnaissance paths of UAVs 0-2 according to the method of Example 1; Figure 7 This is a task path diagram for Rescue Team 0 based on the method in Example 1; Figure 8 This is a mission path diagram for Rescue Team No. 4, based on the method in Example 1. Detailed Implementation

[0022] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0023] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the invention. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0024] In this embodiment of the invention, "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, in the description of this invention, "multiple" refers to two or more.

[0025] Furthermore, to facilitate a clear description of the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" are not necessarily different.

[0026] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0027] Example 1 To address the challenges of emergency resource allocation in the face of secondary earthquake disasters, some scholars have conducted research from the perspectives of material distribution and facility site selection. For example, some existing technologies address the often overlooked issues of multiple resource constraints and potential secondary disasters in emergency resource allocation by constructing an integer programming model and designing a heuristic algorithm for solving it. Simulations have verified that this method is effective, efficient, and applicable to large-scale scenarios. Other existing technologies focus on the site selection, inventory, and allocation of emergency materials and facilities under primary and secondary disasters, constructing a three-stage distributed robust chance constraint model based on Wasserstein fuzzy sets, which is then transformed into a MILP model for efficient solution. Simulations verify that it outperforms comparative models in terms of demand satisfaction and generalization performance, and provides insights for integrated disaster relief management. Still other existing technologies address the problem of temporary medical center site selection and casualty transfer under uncertain primary and secondary disasters by constructing a two-stage robust optimization model based on the minimum-maximum regret criterion, solved using a row generation algorithm. Case studies have demonstrated that it can significantly improve treatment efficiency. Other existing technologies address the emergency supply delivery problem under secondary disasters, considering road damage and dynamic changes in demand. They have constructed a dual-objective optimization model of truck-drone hybrid formation and designed a Q-learning-improved NSGA-II algorithm for solving the problem. Simulations and case studies have verified its effectiveness and superiority. While current technologies have made breakthroughs in modeling secondary disaster scenarios, they generally assume that disaster information is known and do not consider the real-world constraints of information blind spots at key nodes caused by post-earthquake communication disruptions.

[0028] Meanwhile, with the rapid development of drone technology, air-ground collaborative delivery models have been introduced into the field of disaster relief, aiming to overcome the limitations of road network damage and communication disruptions on ground transportation and information transmission. Some existing technologies focus on on-demand delivery problems under uncertain demand, constructing a two-stage Wasserstein distributed robust optimization model to jointly optimize the configuration and scheduling of drones and human delivery personnel. Experiments show that mixed formations are superior to single methods, with drones playing a dominant role, and provide management insights for key parameters. Other existing technologies address service delays caused by uncertain transportation times, proposing a time-bound stochastic and robust truck-drone routing problem model, designing a Benders decomposition algorithm to solve it, and minimizing the risk of service delays. Still other existing technologies address high-level accessibility constraints and air-ground collaborative scheduling challenges in port-warehouse-customer terminal transportation, constructing a truck-van-drone collaborative transfer model, and using a reinforcement learning-enhanced cascaded adaptive large neighborhood search algorithm to verify its superiority in cost and efficiency. Some existing technologies address the problem of low delivery timeliness in remote areas by constructing truck-drone collaborative network optimization models to meet timeliness constraints at the lowest cost, verifying improvements in timeliness and revealing the impact of key factors such as range. Other existing technologies address the challenge of material delivery under post-disaster road network damage by constructing truck-drone collaborative delivery models to minimize total waiting time at disaster sites, using simulated annealing algorithms to solve the problem, and simulations verifying that collaborative delivery significantly outperforms independent delivery models. These studies all focus on optimizing transportation routes for materials or personnel, but do not address the "reconnaissance-treatment" linkage in the process of treating the injured, i.e., how drone reconnaissance results drive the on-site response decisions of rescue teams in real time.

[0029] At the level of rescue team dispatch, existing work mainly focuses on the static or semi-dynamic problems of team layout optimization and task matching. Some existing technologies address the lack of a systematic network model and unclear robustness in the layout of mine emergency rescue teams by constructing single-layer coverage and two-layer collaborative network models. These models reveal a hub-led, group-coordinated structure with good resilience to random attacks but vulnerability to targeted attacks. Based on this, a three-level optimization strategy is proposed, providing a quantitative modeling method for rescue system layout. Other existing technologies address the challenge of collaborative operational planning for heterogeneous rescue teams by constructing a MILP model with the objective of minimizing total waiting time and designing the I-ALNS algorithm for efficient solution. Experiments verify the algorithm's performance and the impact of disaster parameters on collaborative rescue effectiveness. Still other existing technologies address the uncertainty of volunteer-task matching information in disaster relief by proposing a bilateral matching method based on evidence-based reasoning, taking into account both task requirements and volunteer willingness. The rationality and effectiveness of this method are verified through a coastal storm surge case study. However, the above scheduling models all assume that the rescue teams have known the exact needs of the wounded at each disaster site before carrying out the mission, and do not include the reconnaissance status of the disaster site as a decision variable in the constraints. Therefore, they are difficult to apply directly to dynamic scenarios where communication is paralyzed and new disaster sites continue to emerge due to secondary disasters.

[0030] In summary, existing research still has three key shortcomings in the field of earthquake emergency dispatch: (1) Most models assume that disaster information (number of injured and location) is completely known or can be described by prior distribution, ignoring the information black hole formed by the initial non-communication disaster area and the activation point of secondary disaster due to communication interruption after the earthquake. This results in a time-series dependency between UAV reconnaissance and rescue team operations, but they are not coupled in the model. (2) For the physiological characteristic that the survival probability of the wounded decreases with waiting time, existing models mostly use fixed time windows or simple linear penalties, lacking time-by-time dynamic evolution constraints based on the real survival curve (S-shaped function), which makes the optimization of the timing of treatment lack a scientific basis. (3) Regarding decision-making objectives, existing studies either maximize the number of patients treated or minimize the waiting time, and rarely incorporate the two into a unified framework for weighing. Furthermore, when rescue teams simultaneously undertake two heterogeneous tasks—transferring Category A patients back to the hospital and treating Category B patients on-site—the constraints of resource capacity and travel time coupling have not been fully modeled.

[0031] To address the aforementioned issues, this invention proposes an optimized method that integrates UAV reconnaissance strategies with rescue and dispatch, specifically for dynamic rescue coordination scenarios where communication is disrupted in the early stages of an earthquake and there is a risk of secondary disasters. This method addresses the disconnect between reconnaissance and treatment due to unknown disaster information. The model categorizes disaster area nodes into three types: those that can communicate, those that cannot, and those prone to secondary disasters. It also stipulates that all rescue operations can only be carried out after drone reconnaissance is completed at the nodes, achieving a "reconnaissance first, treatment later" time-series coupling. To address the lack of physiological time-dependent data on casualty survival decline, the model introduces an S-shaped survival function to characterize the change in casualty survival rate over time. It drives the dynamic evolution of the number of casualties using discrete time periods, incorporating the actual initial number of casualties immediately after reconnaissance into the current stock after adjusting for survival time, ensuring that treatment decisions strictly follow the progression of injury deterioration. Furthermore, to address the lack of a unified trade-off between the two objectives and insufficient coupling of heterogeneous tasks, the model constructs a dual-objective framework that maximizes the total number of treatments and minimizes waiting time. It also coordinates multiple constraints such as rescue team transfer capacity, medical resource consumption, and movement time, as well as drone endurance and reconnaissance time limitations, comprehensively reflecting the resource competition and collaborative relationships among heterogeneous forces during mission execution.

[0032] This application provides an optimized method that integrates UAV reconnaissance strategy and medical dispatch. This method can be executed by an electronic device, which can be a server or a mobile terminal device. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services. The mobile terminal device can be a laptop, a desktop computer, etc., but is not limited to these.

[0033] like Figure 1 As shown, an optimized method integrating UAV reconnaissance strategy and medical dispatch is provided, including: Following the earthquake, communication infrastructure was severely damaged. While some disaster areas were able to establish contact with the command center immediately, others remained without communication, and the risk of secondary disasters could easily create new affected areas. To ensure effective treatment of the injured within the critical "golden 48 hours" after the disaster, a rescue team capable of carrying medical resources and a group of similar drones departed from multiple hospital locations to coordinate rescue missions. Based on the urgency of treatment, the injured were divided into two categories: Category A, requiring immediate transfer to hospitals, and Category B, which could be treated on-site. The rescue team was responsible for transporting Category A patients and treating Category B patients on-site, while the drones were responsible for aerial reconnaissance of the communication-disrupted disaster areas to ascertain the true number of injured. For communication-disrupted disaster areas, any rescue operation could only be carried out after reconnaissance of the corresponding disaster site was completed, and the drones' endurance was limited by both flight time and the reconnaissance area. As time progressed, the number of surviving injured in each disaster area dynamically changed due to natural attrition and the length of time they had survived, and the activation of secondary disaster points also had a random element. The key decision lies in coordinating the rescue team's route planning, the number of casualties loaded and treated, and the reconnaissance routes and timing of drones, in order to maximize the total number of casualties successfully treated and minimize the waiting time for all casualties to receive services, given limited rescue resources and incomplete information. A detailed flowchart is shown below. Figure 2 and Figure 3 As shown: To facilitate the solution of this problem, we make the following assumptions: (1) The geographical locations and shortest path distances between all hospitals and disaster sites (initially communicable disaster sites, initially non-communicable disaster sites, and secondary disaster-prone sites) are known information; (2) The initial true number of casualties at each disaster area is determined when the system is activated, but the true number of initial non-communicable disaster areas and secondary disaster-prone areas is unknown before reconnaissance and can only be obtained through drone reconnaissance. (3) The survival probability of the wounded follows a survival probability function with survival time as the independent variable. This function is related to the type of wounded, and the natural decay process of the wounded at each disaster site is independent of each other at each time period.

[0034] Step S1: Define the disaster site set, rescue team set, drone set, and hospital set.

[0035] Define the hospital set as ,index .

[0036] All disaster sites are aggregated as follows ,index , The disaster point set includes the initial set of communicable points. Initial set of non-communicable points Collection of Secondary Disaster Prone Areas .

[0037] The rescue team assembled ,index drone collection ,index .

[0038] Step S2 sets the first optimization objective as maximizing the sum of the number of Category A injured persons awaiting transfer at each disaster site and the number of Category B injured persons being treated on-site throughout the entire time period; the second optimization objective is to minimize the total cumulative waiting time from the completion of reconnaissance to the arrival of the rescue team for surviving Category A and Category B injured persons at the end of each time period within the reconnaissance sites; the hierarchical objectives based on lexicographical order are used as the comparison criterion, prioritizing the first optimization objective; a rescue coordination model is constructed based on the first optimization objective, the second optimization objective, and the constraints; the constraints ensure the temporal coupling of using drones for reconnaissance first, followed by treatment.

[0039] The primary optimization objective is to maximize the sum of the number of Category A injured persons awaiting transfer at various disaster sites and the number of Category B injured persons being treated on-site throughout the entire time period; specifically:

[0040] in, Non-negative integers, time period Inside, the rescue team At the disaster site Number of Category A wounded personnel on board; Non-negative integers, time period Inside, the rescue team At the disaster site The first disposal Number of Class B casualties; For the set of categories of deterioration time limits for Class B wounded, index ; For time period sets, ,in , Step size; Assemble the rescue team, index is ; For the set of disaster points, the index is .

[0041] The first optimization objective function aims to maximize the sum of the number of Category A injured persons awaiting transfer at each disaster site and the number of Category B injured persons being treated on-site throughout the entire time period.

[0042] The second optimization objective is to minimize the total cumulative waiting time from the completion of reconnaissance to the arrival of rescue teams for Category A and Category B casualties surviving in the reconnaissance sites at the end of each time period; specifically:

[0043] in, The time interval step; Non-negative integers, time period Finally, disaster site The actual number of surviving Category A wounded; Non-negative integers, time period Finally, disaster site The real survival of the first Number of Class B casualties; Binary, time period Finally, disaster site Whether it has been detected is indicated by 1 if yes, and 0 otherwise.

[0044] The second optimization objective function aims to calculate the cumulative product of the number of surviving wounded and the reconnaissance status at the end of each time period using a step size. By combining the first and second optimization objective functions into a dual-objective optimization framework, a systematic trade-off is established between the number of patients treated and the timeliness of services.

[0045] Using lexicographical order as the hierarchical objective as the comparison criterion, priority is given to ensuring the first optimization objective.

[0046] The hierarchical objective based on lexicographical order serves as the comparison criterion, following a strict priority order of "number of people treated first, waiting time second." After obtaining the number of people treated and the waiting time for the current UAV reconnaissance plan and corresponding rescue scheduling scheme in each outer iteration, the number of people treated is compared first: if the number of people treated in the current scheme is greater than the global optimum, it directly replaces the global optimum; if it is less than the global optimum, the scheme is discarded and no further comparison is performed. Only when the number of people treated in the current scheme is equal to the global optimum (considering floating-point precision, a difference less than 0.01 is allowed) does the process proceed to the second level of comparing waiting time. If the waiting time of the current scheme is smaller, the waiting time of the global optimum is updated and the optimal solution is replaced; otherwise, the original scheme is retained. This criterion applies the above hierarchical judgment to each feasible scheme throughout the search process, and simultaneously updates the global optimum archive and convergence curve record, prioritizing the first optimization objective.

[0047] Furthermore, based on the first optimization objective, the second optimization objective, and the constraints, a rescue coordination model is constructed; the constraints ensure the temporal coupling of using drones for reconnaissance first, followed by rescue.

[0048] The constraints include: The first constraint defines the S-shaped survival probability function. and the corresponding survival rate for each time period , This serves as the basis for calculating the attenuation of the dynamic evolution of the wounded:

[0049]

[0050]

[0051] in, Identify the types of injured persons; For the first The asymptotic maximum of the survival probability of such casualties (usually taken as 1); This is the half-life parameter, which represents the survival time when the survival rate drops to 50%. The shape parameter controls the steepness of the survival curve (i.e., the rate of change of the decay rate). The number of times the injured person has survived since the moment of activation (unit and time step). Consistent); Let be the survival probability function for Class A casualties; For the first Survival probability function for Class B casualties.

[0052] The second constraint describes the dynamic evolution of the stock of Category A and various types of B casualties at each disaster site. It is jointly determined by the natural decay of survivors from the previous period, the amount of rescue and removal in the current period, the conversion of Category B to Category A, and the number of survivors after the initial surge in casualties immediately after reconnaissance, calculated based on their survival time.

[0053]

[0054]

[0055] ,and

[0056] in, For time period Finally, disaster site The actual number of surviving Category A wounded; For time period The end, that is, the previous period, the disaster site The actual number of surviving Category A wounded; The survival rate for each time period; The actual number of Category A casualties at the time of activation. ; Disaster point The activation period (initial disaster point is 0, secondary disaster point is ≥1). , ; Binary, disaster point Reconnaissance during the period If it has just been completed, the value is 1; otherwise, the value is 0. For time period Finally, disaster site The real survival of the first Number of Class B casualties; For time period The end, that is, the previous period, the disaster site The real survival of the first Number of Class B casualties; For time period Inside, the rescue team At the disaster site The first disposal Number of Class B casualties; For the true first time of activation Number of Class B casualties, , ; For time period Finally, disaster site Whether it has been detected: 1 if yes, 0 otherwise; For time period The end, that is, the previous period, the disaster site Whether it has been detected: 1 if yes, 0 otherwise; Disaster point Reconnaissance during the period If it has just been completed, the value is 1; otherwise, the value is 0.

[0057] The third constraint enforces the pre-reconnaissance logic, ensuring that no rescue operations are carried out when initial communication is impossible or secondary disaster sites are not reconnoitered, while limiting the total amount of Category A loading to any single disaster site to no more than the rated transfer capacity:

[0058]

[0059] in, This represents the maximum number of Category A injured persons that the rescue team can transfer in a single operation.

[0060] The fourth constraint is to track the spatiotemporal location updates of the rescue team between the disaster site and the hospital, ensuring consistency between the timing of departure and arrival of the mission:

[0061]

[0062]

[0063]

[0064] in, For the initial time ( Rescue Team Is it located in a disaster area? , is always 0; For time period Finally, the rescue team Is it located in a disaster area? If yes, it is 1; otherwise, it is 0. For the initial rescue team Is it located in a hospital? , by parameters Decide; For time period Finally, the rescue team Is it located in a hospital? If yes, it is 1; otherwise, it is 0. For time period Inside, the rescue team From the disaster area Depart for the node Performing Category B casualty treatment tasks ( (If it can be a disaster site or a hospital), it is 1; otherwise, it is 0. For the rescue team Is it in an earlier time period ( From node Depart for the disaster area Execute Class B tasks, after The time period is after the time period arrive If yes, it is 1; otherwise, it is 0. For time period Inside, the rescue team From the hospital Depart for the disaster area If a task of type B is executed, the value is 1; otherwise, the value is 0. For time period Inside, the rescue team From the hospital Depart for the disaster area If a Category A transshipment task is executed, the value is 1; otherwise, the value is 0. For time period Finally, the rescue team Is it located in a hospital? If yes, it is 1; otherwise, it is 0. For the rescue team Is it during the time period? From the disaster site Depart for the hospital Executing a type A task, the total round-trip time is ( (After the time period) Arrival at the hospital is counted as 1, otherwise as 0; For the rescue team Is it during the time period? From the disaster site Depart for the hospital Execute Class B tasks, after The time period is after the time period Arrival at the hospital is counted as 1, otherwise as 0; For time period Inside, the rescue team From the hospital Depart for the disaster area If a task of type A is executed, the value is 1; otherwise, the value is 0. For time period Inside, the rescue team From the hospital Depart for the disaster area If a task of type B is executed, the value is 1; otherwise, the value is 0. For the initial rescue team Located in the hospital The value is 1 if it is above, and 0 otherwise. .

[0065] The fifth constraint stipulates that the total amount of medical resources consumed by the rescue team in performing all Category B response missions shall not exceed its carrying capacity: ，

[0066] in, For the rescue team Is it during the departure time? From node Heading to the disaster site The value is 1 if task B is executed, and 0 otherwise. ; For time period Inside, the rescue team At the disaster site The first disposal Number of Class B casualties; For the disposal unit Number of medical resource units required for Class B casualties. ; This represents the maximum number of medical resources a rescue team can carry. For the rescue team from the node arrive The movement requires time. .

[0067] The sixth constraint maintains the continuity of the rescue team's path by balancing the number of times they enter and exit each disaster site for Category B missions:

[0068] in, Binary, time period Inside, the rescue team From node Depart for the node Rescuing Category B casualties is rated as 1, otherwise it is rated as 0. For time period Inside, the rescue team From the disaster area Depart for the node ( If it can perform Category B tasks for other disaster sites or hospitals, the value is 1; otherwise, it is 0.

[0069] The seventh constraint stipulates that drones can only fly to non-communication disaster sites that have not yet been reconnoitered, in order to avoid wasting resources due to repeated reconnaissance:

[0070] in, Binary, time period Inside, drones From node fly to node If yes, it is 1; otherwise, it is 0. For time period The end, that is, the previous period, the disaster site Whether it has been detected is indicated by 1 if yes, and 0 otherwise.

[0071] The eighth constraint tracks the spatiotemporal location updates of the drone between the disaster site and the hospital to ensure the consistency of the timing between the departure and arrival of the mission:

[0072]

[0073]

[0074]

[0075] in, For the initial moment of the drone Is it located in a disaster area? , always 0 (initially all located in the hospital); For time period Finally, drones Is it located in a disaster area? If yes, it is 1; otherwise, it is 0. For the initial moment of the drone Is it located in a hospital? , by parameters Decide; For time period Finally, drones Is it located in a hospital? If yes, it is 1; otherwise, it is 0. For time period Finally, drones Is it located in a disaster area? If yes, it is 1; otherwise, it is 0. For time period Inside, drones From the disaster site fly to node ( (If it can be a disaster site or a hospital), it is 1; otherwise, it is 0. For drones Was it in an earlier period? From node Flying to the disaster area ,flight After a period of time If the destination is reached, the value is 1; otherwise, the value is 0. For time period Inside, drones From the hospital Flying to the disaster area If yes, it is 1; otherwise, it is 0. For time period Finally, drones Is it located in a hospital? If yes, it is 1; otherwise, it is 0. For time period Inside, drones From the hospital Flying to the disaster area If yes, it is 1; otherwise, it is 0. Is it a drone? During the period From the disaster site Fly to the hospital ,flight After a period of time Arrival at the hospital is counted as 1, otherwise as 0; For the initial moment of the drone Located in the hospital The value is 1 if it is above, and 0 otherwise. .

[0076] The ninth constraint incorporates the UAV flight time and the area-based reconnaissance time for each disaster site into the endurance consumption metric, ensuring that the total time does not exceed the maximum endurance period:

[0077]

[0078]

[0079] in, Indicates drone Is it during the departure time? From node Flying to the disaster area The value is 1 if it is true, and 0 otherwise. ; Indicates drone Is it during the departure time? From node Fly to the hospital The value is 1 if it is true, and 0 otherwise. ; For drones from nodes To the hospital The number of flight hours required; For the rescue team from the node To the hospital The number of time periods required for the movement; For drones from nodes arrive Flights require time to complete. ; The reconnaissance area of ​​the disaster site where communication is impossible. ; This refers to the reconnaissance speed of a drone per unit time and unit area. This represents the total flight time of the drone. ; For the rescue team from the node arrive The movement requires time. ; Hospital collection, index .

[0080] The tenth constraint explicitly defines the type and range of values ​​for each decision variable, ensuring the mathematical completeness and solution feasibility of the model:

[0081]

[0082] in, This represents the maximum number of Category A injured persons that the rescue team can transfer in a single operation. It is the set of nonnegative real numbers; It is the set of non-negative integers.

[0083] The above constraints are combined to realize the temporal coupling logic of reconnaissance first and then treatment, and comprehensively reflect the resource competition and cooperation relationship between heterogeneous rescue forces in the process of performing heterogeneous tasks.

[0084] Step S3: Use mathematical transformations to linearize the nonlinear constraints and the first and second optimization objective functions in the rescue coordination model.

[0085] By linearizing the nonlinear constraints and objective function terms in the model through mathematical transformations, the original model is transformed into a mixed-integer linear programming problem that can be solved efficiently.

[0086] In S3.1, in the second constraint condition, In It is a non-linear survival function, and its parameters depend on the time period. Disaster point activation time Therefore, a discretization prediction algorithm is used to linearize it: (1) First, for each Calculate: Number of survival periods constant coefficients ; For nodes arrive The shortest path distance.

[0087] (2) The constant coefficients When the original nonlinear term is input into the model as a constant parameter, it becomes...

[0088] The linearized expression is:

[0089] Similarly, the evolutionary constraints for Class B casualties undergo the same linearization transformation process, using a discretization prediction algorithm for linearization transformation: (1) For each and each category of wounded Pre-calculation: constant coefficients ; (2) When the original nonlinear term is input into the model as a constant parameter, it becomes...

[0090] The linearized expression is: .

[0091] In S3.2, under the third constraint, and For the nonlinear part, the Big M method is used, and auxiliary variables are introduced. and After performing a linear transformation, the linearized expression is:

[0092]

[0093]

[0094]

[0095]

[0096] Where M takes the value of .

[0097]

[0098]

[0099]

[0100]

[0101]

[0102] Where M takes the value of .

[0103] S3.3 Regarding the fifth constraint condition For the nonlinear part, the Big M method is used, and an auxiliary continuous variable is introduced. Then the linearized expression is:

[0104]

[0105]

[0106]

[0107]

[0108] S3.4 linearizes the second optimization objective function into:

[0109] Through the above linear transformation process, the rescue coordination model after linear transformation is obtained.

[0110] Step S4: Solve the linearized rescue coordination model, output the optimal UAV reconnaissance path and rescue team dispatch plan, and evaluate them.

[0111] The rescue coordination model constructed in this invention has a decision space encompassing UAV reconnaissance sequence arrangement, rescue team task type selection and path planning, and multi-time window coupling constraints, which constitutes a large-scale combinatorial optimization problem. Using enumeration or directly calling a mixed-integer programming solver would result in severe dimensionality curse and computational time bottlenecks.

[0112] To address this, this invention designs a two-stage randomized greedy search algorithm (TS-RGS). This algorithm strictly follows a sequential logic of outer-layer enumeration of reconnaissance schemes, inner-layer generation of rescue scheduling, step-by-step optimal archiving, and post-event patching enhancement, advancing each computational step accordingly. Specifically: Step S4.1: Initialize parameters and set global archive to empty.

[0113] Set the maximum number of iterations for the outer layer. Inner layer single plan repetition count Unit time period Total planning duration Initialize the global optimal number of patients to be treated. Optimal waiting time Global optimal solution archive Leave blank.

[0114] Step S4.2: Generate the UAV reconnaissance plan through the outer loop, specifically as follows: (1) Maintain state variables for each UAV. The state variables include: the current time. ,Location Cumulative time per flight Number of consecutive non-return flights Initially, they were all located in their respective hospitals.

[0115] (2) Collect information on disaster sites that need to be investigated The data is randomly shuffled using the Fisher-Yates algorithm and processed sequentially. For each disaster site, drones are traversed, and two modes are evaluated: continuous reconnaissance (currently, for non-hospital areas, the configuration cost of a single drone is considered). ，d Fly directly from the current location to the target, with a total time not exceeding the range. ) and departure after returning (return to the hospital first and then depart, arriving no earlier than the activation time); Select the drone with the shortest reconnaissance completion time to execute the mission. If no feasible candidate is found, randomly select one and force it to return to base before restarting. The total time exceeds [a certain limit]. Forced return to home and reset at any time. This represents the total flight time of the drone. d .

[0116] The shuffling algorithm is an algorithm that generates random permutations of a finite set with a time complexity of O(n), guaranteeing that each permutation occurs with equal probability. It randomizes the processing order of UAV reconnaissance missions, avoiding systematic bias towards certain problematic areas, enhancing the diversity of reconnaissance plans, and expanding the solution space coverage of the outer search layer.

[0117] (3) Summarize the earliest reconnaissance completion time for each disaster site. ( Disaster Default ).

[0118] Step S4.3: Feasibility Verification Inspect all disaster sites that require reconnaissance ( The complete set and middle Are all points covered? If any If no record is found, discard the plan and return to step S4.2; if all records are covered, proceed to the inner scheduling layer.

[0119] Step S4.4: Generate and evaluate the rescue team dispatch plan through the inner loop, specifically as follows: (1) Build a min-heap and sort by available time. All teams are initially located in their respective hospitals and have 0 available time. Pop the earliest available rescue team. The A / B category of the mission is determined based on the hospital, the current time, and the percentage of remaining wounded.

[0120] (2) Screening candidate disaster sites (meeting the following conditions) (Disaster site) (Earliest possible rescue time) And the number of remaining wounded is greater than 0.001, of which, ,like ,otherwise Randomly select one of three strategies: waiting time priority, quantity priority, and mixed weighted sorting, and randomly select a target from the top 5.

[0121] A two-way backoff mechanism is adopted, with the actual departure time... ,in For the current available start time for the rescue team, To enable the rescue team to move from its current location to the target disaster site Required travel time. Category A service, 1 time period, loading... Type B generates continuous service sequences (maximum 4 points), with each point having a service time period of [number missing]. Accumulated resources do not exceed The end time (including the return trip) shall not exceed .

[0122] (3) If the task is successful, update and re-enter the heap; otherwise, try another type or advance the time. Repeat until all wounded are healed or time runs out, then perform conflict resolution.

[0123] The evolution of casualties is advanced according to discrete time periods: the surviving casualties at the identified disaster sites are accumulated up to the waiting time, activation and rescue events are processed, and the survival probability factor is multiplied at the end of each time period for decay. The output is the number of people treated and the waiting time.

[0124] Step S4.5: Compare and update the global best archive at each level.

[0125] like or and Then store the current solution. ,renew , Otherwise, retain the original save file.

[0126] Step S4.6: Perform loop control. Inner loop insufficient. If the condition is met, return to step S4.4; if the condition is met, exit the inner loop and record the convergence data. (Outer loop insufficient.) If the condition is met, return to step S4.2; if the condition is met, proceed to post-processing.

[0127] Step S4.7: Perform post-processing patching.

[0128] Group rescue teams and determine the end time of their last mission and their current location. Within the remaining time, repeatedly select disaster sites with the most remaining Category A disaster sites. If the total round-trip service time is within [a certain range]... If the task is completed, add it; otherwise, mark it as infeasible. Re-simulate after patching, updating the number of patients treated and the waiting time.

[0129] Step S4.8: Perform convergence determination. Extract the optimal number of patients treated and the waiting time series from the last 200 rounds, and calculate the standard deviation. and .like and If convergence is confirmed, then convergence is confirmed; otherwise, convergence is not complete.

[0130] Step S4.9: Output the optimal UAV reconnaissance route and rescue team dispatch plan and evaluate them.

[0131] The system prints global statistics of the optimal solution to the console, including the initial total number of injured at each disaster site, the number of deaths and survivors, the reconnaissance completion time at each disaster site, and treatment details. It also outputs the complete mission sequence for each rescue team and each drone. Simultaneously, it generates PNG charts: algorithm convergence curve, node spatial distribution, drone group path diagram, independent mission path diagram for each rescue team, and convergence status determination based on the standard deviation of the last 200 rounds, providing a comprehensive evaluation for decision-making.

[0132] This invention employs a two-stage solution framework that combines outer-layer random enumeration with inner-layer greedy scheduling. The outer layer ensures solution space coverage through multi-sample generation of reconnaissance plans, while the inner layer drives rescue scheduling with a priority queue, achieving a balance between broad exploration and deep development. A bidirectional backoff scheduling mechanism is designed, allowing rescue teams to depart in advance based on the earliest available rescue time at the disaster site and wait en route, effectively eliminating idle time. Simultaneously, the task type is dynamically decided based on the rescue team status and the proportion of remaining wounded in the global pool, taking into account resource balance. A hierarchical target comparison criterion based on lexicographical order is adopted, eliminating the need for weighting coefficients and strictly adhering to the emergency decision-making logic of prioritizing the number of people to be treated and secondarily waiting time. A post-processing patching strategy is also provided to further explore rescue potential without disrupting the original scheduling. A convergence monitoring module is integrated, which quantifies the search convergence status by calculating the standard deviation of the optimal value sequence and outputs the UAV flight path, rescue team task sequence, and convergence curve, providing an intuitive basis for scheme evaluation and parameter optimization.

[0133] This invention achieves integrated decision-making for the random generation of reconnaissance plans and the coordinated optimization of rescue scheduling, balancing the exploration depth of the understanding space with the development efficiency of optimization. The bidirectional backoff scheduling mechanism effectively overcomes the scheduling rigidity caused by the coupling between disaster point activation time and reconnaissance completion time, significantly improving time utilization. The hierarchical target comparison criterion and post-processing patching strategy can obtain high-quality solutions. Simultaneously, the algorithm's built-in randomized enumeration framework and convergence criterion have good extensibility, providing a reference for other approximation methods (such as tabu search and genetic algorithms), and offering a rigorous and practical method for solving large-scale emergency logistics scheduling problems.

[0134] The following is the experimental procedure of this invention: This invention implements a two-stage randomized greedy search algorithm (TS-RGS) on the Python platform. The model and algorithm are experimentally analyzed using a test network containing 26 nodes, which are: 2 hospitals, 5 initial communicable points, 4 initial non-communicable points, and 15 secondary disaster-prone points.

[0135] The experimental environment consisted of an 11th Gen Intel(R) Core(TM) i5-1135G7 @ 2.40GHz processor and 16 GB of RAM. The algorithm parameters were set as follows: maximum number of outer layer iterations. Inner single plan repetition count Total planning duration The convergence threshold is set for each time period. and To determine the stability of the search process. The spatial distribution and topology of the nodes are as follows: Figure 4 As shown.

[0136] In this numerical experiment, there were 6 rescue teams, with a travel speed of 40 km / hour, a total rescue resource capacity of 250 units per team, and a maximum capacity of 15 Class A casualties per transport. There were also 6 drones, with a travel speed of 50 km / hour and a reconnaissance speed of 15 km / hour, and a maximum flight time of 24 hours per sortie. The number of casualties at each disaster site was based on historical post-disaster statistics and scaled proportionally. The survival probability of casualties was calculated using a logistic decay function, with the half-life parameter set according to the severity of the injury (48 hours for Class A, and 12, 24, and 36 hours for Class B). The planning period was 48 hours, and all parameter values ​​were taken with reference to provincial emergency material reserve standards and historical post-disaster dispatch data to ensure the typicality and reliability of the simulation.

[0137] To verify the convergence performance of the proposed two-stage randomized greedy search algorithm (TS-RGS), this invention records the current optimal number of people to be treated and the corresponding waiting time after each outer iteration, and calculates the standard deviation based on the data from the last 200 iterations to determine the stability of the search process. Figure 5 shows the evolution trajectory of the optimal number of people to be treated and the corresponding waiting time in 10,000 outer iterations, where (a) is the optimal number of people to be treated and (b) is the corresponding waiting time. From the convergence curve, it can be observed that the algorithm achieves a rapid increase in the number of people to be treated within the first 2,000 iterations, indicating that the diversified generation strategy of the randomized UAV reconnaissance plan can effectively cover high-potential spatial areas; thereafter, the optimal number of people to be treated stabilizes at 1,500 and does not increase further in subsequent iterations. Statistical analysis of the last 200 iterations shows that the standard deviation of the number of people to be treated is 0.000 and the standard deviation of the waiting time is 0.000, both of which are zero, indicating that the algorithm has fully converged to a stable optimal solution.

[0138] It should be noted that the waiting time curve still exhibits slight fluctuations after the number of people receiving treatment stabilizes. This is because the hierarchical optimization mechanism continuously searches for solutions with lower waiting times, assuming the same number of people receiving treatment. The final optimal solution corresponds to a waiting time of 8275.49 person-times, which reflects the cumulative burden of injured people waiting for treatment during post-disaster relief efforts. No numerical instability warnings appeared during the overall solution process, indicating that the model's numerical properties are good.

[0139] In the optimal solution, the UAVs completed reconnaissance missions at 17 disaster sites, covering all J2-type disaster sites (5 sites) and 12 J3-type disaster sites with an activation time not exceeding 48 time slots. The UAV reconnaissance completion times were distributed between the 3rd and 48th time slots, such as... Figure 6 As shown in Figure 1 and Table 1, reconnaissance of the initial non-communicable disaster site (J2) was concentrated in the early period (t = 3–7), while reconnaissance of secondary disaster-prone sites (J3) was arranged sequentially according to their activation time, reflecting the adaptive response of the UAV plan to the activation sequence of disaster sites. All reconnaissance missions were completed within 1–5 hours after the disaster site was activated, ensuring that the disaster information on which the rescue team relied for decision-making could be obtained in a timely manner.

[0140] Table 1. Spatiotemporal table of UAV reconnaissance paths 3–5

[0141] In terms of rescue team dispatch, the six rescue teams (numbered 0 to 5) carried out a total of 49 rescue missions, including 36 Category A transfer missions and 13 Category B on-site treatment missions, serving a total of 1,500 injured people (396, 434, 230, 153, 157, and 128 people served by each team respectively), with a waiting time of 8,275.49 person-times. From the perspective of resource utilization, there is significant heterogeneity in the division of labor among the rescue teams: the three rescue teams of Hospital 0 (numbers 0 to 2) undertook more Category B continuous treatment missions, and their mission sequences usually included chain visits to 3 to 4 disaster sites, with a single dispatch cycle of 16 to 37 time periods, reflecting the dispatch logic of "continuous treatment nearby" for Category B missions; the three rescue teams of Hospital 1 (numbers 3 to 5) mainly carried out Category A transfer missions, with a shorter round-trip time for each mission (2 to 5 time periods), forming a high-frequency short-distance transfer mode. The difference in task allocation between the two types of hospital rescue teams is essentially the result of the model's self-organized optimization based on the spatial distribution of disaster sites and the proportion of different types of casualties: Hospital 0 is adjacent to areas with a high concentration of Category B casualties, so it undertakes more on-site treatment tasks; Hospital 1 is adjacent to areas with a high concentration of Category A casualties, so it mainly focuses on transport tasks. In terms of time, Hospital 0's rescue teams 0 and 1 started their missions from time period 0, while rescue team 2 was delayed until time period 22 to wait for reconnaissance information from secondary disaster sites before being dispatched; Hospital 1's three rescue teams experienced a gap in their missions between time periods 12 and 22, which also reflects the decision-making logic of "reconnaissance first, rescue later".

[0142] Overall, the rescue teams' missions were spatially staggered and temporally phased, avoiding path conflicts and resource waste. The spatial distribution of the rescue team mission paths is shown in Figure 7. Figure 8As shown (taking rescue teams 0 and 4 as examples), the red path represents type A transfer tasks, and the blue path represents type B on-site treatment tasks. The shade of color indicates the chronological order of the tasks. It can be observed that the type B task path presents a chain structure with continuous visits to multiple nodes, while the type A task path is a round-trip pattern between the hospital and a single disaster site. The difference between the two types of path forms intuitively reflects the differentiated needs of different types of injured persons for rescue resource allocation modes.

[0143] In terms of treatment effectiveness, the optimal plan treated a total of 1,500 injured people within 48 time periods, accounting for 70.67% of the total demand of 2,123 injured people, as shown in Table 2. Among them, 443 people received Category A treatment and 1,057 people received Category B treatment. If no rescue was implemented, the number of survivors after 48 time periods would be far lower than the actual level, as the injuries of the injured continued to deteriorate over time (attenuation factor of 0.5 for Category A, and attenuation factors of 0.5, 0.333, and 0.25 for the three categories of Category B, respectively). Through reasonable reconnaissance timing and rescue route planning, the model successfully controlled the total number of deaths to 526, significantly reducing the loss of life compared to the scenario of no rescue or inefficient rescue. At the same time, the waiting time corresponding to the optimal plan was 8,275.49 person-times. If we consider the 1,500 injured people who were rescued, the average waiting time per person was about 331 minutes (about 5.5 hours). Within the critical 48-hour window for disaster relief, this waiting time is within an acceptable range, demonstrating that the model has achieved an effective balance between maximizing the number of people treated and minimizing the waiting time.

[0144] Table 2 Statistics on Treatment and Survival of the Wounded

[0145] It is important to note that this treatment outcome was achieved under strict pre-reconnaissance constraints. All disaster sites were reconnoitered before the arrival of rescue teams, with no instances of blind rescue. The model, through a hierarchical optimization mechanism (prioritizing the number of people treated, followed by waiting time), ensured information completeness while fully utilizing the operational capabilities of the six rescue teams. This concentrated limited resources on high-priority, high-survival-probability injured groups, achieving a balance between rescue efficiency and fairness. These results fully validate the practical value of the "reconnaissance first, rescue later" approach in post-disaster emergency scenarios: by using drones to obtain disaster information beforehand, each rescue team deployment had a clear objective, avoiding the detours and delays that can occur with traditional experience-based dispatching.

[0146] This invention addresses the challenge of rescue coordination during the initial stages of an earthquake, when communication is disrupted and secondary disasters are imminent. It constructs an integrated optimization model that combines UAV reconnaissance with rescue team scheduling, and designs a two-stage randomized greedy search algorithm (TS-RGS) to solve it. Simulation experiments based on a test network containing 26 nodes yield the following conclusions: (1) At the model level, the temporal coupling constraint of "reconnaissance first, treatment later" is used to make the disaster site reconnaissance status a prerequisite for rescue operations, which effectively solves the dilemma of missing disaster information caused by communication paralysis; the dynamic evolution constraint of the injured based on the S-shaped survival probability function makes the treatment decision strictly follow the physiological law that the injury deteriorates with waiting time, which is more scientifically based than the fixed time window method.

[0147] (2) At the objective level, a dual-objective framework of maximizing the number of people treated and minimizing the waiting time was constructed. Pareto selection was performed using lexicographical hierarchical comparison, avoiding the subjective dependence of the weighted summation method on the weight coefficients. Experiments showed that the optimal solution treated a total of 1,500 injured people (accounting for 70.67% of the total) within 48 time periods, with an average waiting time of about 331 minutes per person, verifying the effectiveness of the dual-objective trade-off.

[0148] (3) At the algorithm level, the two-stage randomized greedy search algorithm works in conjunction with the outer reconnaissance enumeration and the inner priority scheduling, and the bidirectional backoff mechanism effectively eliminates the idle time of the rescue team. The algorithm converges in only 2000 iterations, and the standard deviation of the number of people treated and the waiting time in the last 200 rounds are all zero, which verifies the search stability and convergence performance.

[0149] (4) At the dispatching level, the six rescue teams formed a self-organized division of labor based on the location of the hospitals: Hospital 0 is located near the densely populated area of ​​Class B casualties and mainly uses multi-node chain visits, with a single dispatch cycle of 16 to 37 time periods; Hospital 1 is located near the concentrated area of ​​Class A casualties and mainly uses high-frequency short-distance round trips, with a single trip taking 2 to 5 time periods. The parallel operation of the two modes achieved the optimal synergy between spatial coverage and time efficiency.

[0150] This invention provides a decision-making tool that integrates reconnaissance, dynamic tracking of the wounded, and rescue dispatch for extreme scenarios where communication is interrupted after an earthquake and there is a risk of secondary disasters.

[0151] Example 2 This embodiment provides an integrated optimization system that combines reconnaissance strategies and medical dispatch, including: The parameter definition module defines the disaster site set, the rescue team set, and the drone set. The rescue coordination model construction module has the following objectives: First, to maximize the sum of the number of Category A casualties awaiting transfer and the number of Category B casualties being treated on-site at each disaster site throughout the entire time period; second, to minimize the total cumulative waiting time from the completion of reconnaissance to the arrival of the rescue team for surviving Category A and Category B casualties at the end of each time period. A hierarchical objective based on lexicographical order is used as the comparison criterion, prioritizing the first objective. Based on the first and second objectives and constraints, a rescue coordination model is constructed. Constraints ensure the temporal coupling of using drones for reconnaissance first, followed by treatment. The transformation module uses mathematical transformations to linearize the nonlinear constraints and the first and second optimization objective functions in the rescue coordination model. The solution module solves the linearized rescue coordination model, outputs the optimal UAV reconnaissance path and rescue team dispatch plan, and evaluates them.

[0152] Example 3 This embodiment also provides an electronic device, including: one or more processors, one or more memories, and one or more computer programs; wherein, the processor is connected to the memory, and the one or more computer programs are stored in the memory. When the electronic device is running, the processor executes the one or more computer programs stored in the memory to cause the electronic device to perform the method described in Embodiment 1.

[0153] Example 4 This embodiment also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the method described in Embodiment 1.

[0154] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An optimized method integrating UAV reconnaissance strategy and medical dispatch, characterized in that, include: Define the disaster site set, the rescue team set, and the drone set; Based on the urgency of the injured treatment, the injured were divided into two categories: Category A injured who need to be immediately transferred to the hospital for treatment, and Category B injured who can be treated on-site in the disaster area. The first optimization objective is to maximize the sum of the number of Category A injured persons awaiting transfer at each disaster site and the number of Category B injured persons being treated on-site throughout the entire time period; the second optimization objective is to minimize the total cumulative waiting time of Category A and Category B injured persons surviving in the disaster sites that have been reconnoitered at the end of each time period from the completion of reconnaissance to the arrival of the rescue team. Using lexicographical hierarchical objectives as the comparison criterion, the first optimization objective is prioritized; a rescue coordination model is constructed based on the first optimization objective, the second optimization objective, and the constraints. Constraints are used to ensure the temporal coupling of using drones for reconnaissance first, followed by medical treatment; The nonlinear constraints and the first and second optimization objective functions in the rescue coordination model are linearized using mathematical transformations. Solve the linearized rescue coordination model, output the optimal UAV reconnaissance path and rescue team dispatch plan, and evaluate them; The constraints include: a first constraint, which defines an S-shaped survival probability function. and the corresponding survival rate for each time period , This serves as the basis for calculating the attenuation of the dynamic evolution of the wounded: in, Identify the types of injured persons; For the first The asymptotic maximum of the survival probability of such casualties; This is the half-life parameter, which represents the survival time when the survival rate drops to 50%. The shape parameter controls the steepness of the survival curve; The time the injured person has survived since the moment of activation; Let be the survival probability function for Class A casualties; For the first Survival probability function for Class B casualties; For the set of categories of deterioration time limits for Class B wounded, index ; The time interval step; The second constraint describes the dynamic evolution of the stock of Category A and various types of B casualties at each disaster site. It is jointly determined by the natural decay of survivors from the previous period, the amount of rescue and removal in the current period, the conversion of Category B to Category A, and the number of survivors after the initial surge in casualties immediately after reconnaissance, calculated based on their survival time. ,and in, For time period Finally, disaster site The actual number of surviving Category A wounded; For time period The end, that is, the previous period, the disaster site The actual number of surviving Category A wounded; The survival rate for each time period; The actual number of Category A casualties at the time of activation. ; Disaster point The activation period (initial disaster point is 0, secondary disaster point is ≥1). , ; Binary, disaster point Reconnaissance during the period If it has just been completed, the value is 1; otherwise, the value is 0. For time period Finally, disaster site The real survival of the first Number of Class B casualties; For time period The end, that is, the previous period, the disaster site The real survival of the first Number of Class B casualties; For time period Inside, the rescue team At the disaster site The first disposal Number of Class B casualties; For the true first time of activation Number of Class B casualties, , ; For time period Finally, disaster site Whether it has been detected: 1 if yes, 0 otherwise; For time period The end, that is, the previous period, the disaster site Whether it has been detected: 1 if yes, 0 otherwise; Disaster point Reconnaissance during the period If it has just been completed, the value is 1; otherwise, the value is 0. Non-negative integers, time period Inside, the rescue team At the disaster site Number of Category A wounded personnel on board; For time period sets, ,in , Step size; Assemble the rescue team, index is ; For the set of disaster points, the index is ; The constraints also include: a third constraint, which enforces the reconnaissance pre-processing logic to ensure that no rescue operations are carried out when the initial non-communication point and secondary disaster point are not reconnoitered, while limiting the total amount of Category A loading at any disaster point in a single operation to not exceed the rated transfer capacity. in, This represents the maximum number of Category A injured persons that the rescue team can transfer in a single operation. For the initial set of non-communication points, This is a collection of locations prone to secondary disasters.

2. The optimization method for integrating UAV reconnaissance strategy and medical dispatch according to claim 1, characterized in that, The disaster point set includes: an initial set of communicable points, an initial set of non-communicable points, and a set of points prone to secondary disasters.

3. The optimization method for integrating UAV reconnaissance strategy and medical dispatch according to claim 1, characterized in that, The primary optimization objective is to maximize the sum of the number of Category A injured persons awaiting transfer at various disaster sites and the number of Category B injured persons being treated on-site throughout the entire time period; specifically: 。 4. The optimization method for integrating UAV reconnaissance strategy and medical dispatch according to claim 1, characterized in that, The second optimization objective is to minimize the total cumulative waiting time from the completion of reconnaissance to the arrival of rescue teams for Category A and Category B casualties surviving in the reconnaissance sites at the end of each time period; specifically: 。 5. The optimization method for integrating UAV reconnaissance strategy and medical dispatch according to claim 1, characterized in that, The linearized rescue coordination model is solved to output and evaluate the optimal drone reconnaissance path and rescue team scheduling scheme. Specifically, a two-stage randomized greedy search algorithm is used to solve the rescue coordination model. The outer loop generates a drone reconnaissance plan. Then, the inner loop generates and evaluates the rescue team scheduling scheme. Loop control and post-processing patching operations are performed to finally output and evaluate the optimal drone reconnaissance path and rescue team scheduling scheme.

6. An optimized system integrating UAV reconnaissance strategy and medical dispatch, characterized in that, An optimization method for integrating unmanned aerial vehicle (UAV) reconnaissance strategy and medical dispatch as described in any one of claims 1-5 includes: The parameter definition module defines the disaster site set, the rescue team set, and the drone set. The rescue coordination model construction module has the following objectives: First, to maximize the sum of the number of Category A casualties awaiting transfer and the number of Category B casualties being treated on-site at each disaster site throughout the entire time period; second, to minimize the total cumulative waiting time from the completion of reconnaissance to the arrival of the rescue team for surviving Category A and Category B casualties at the end of each time period. A hierarchical objective based on lexicographical order is used as the comparison criterion, prioritizing the first objective. Based on the first and second objectives and constraints, a rescue coordination model is constructed. Constraints ensure the temporal coupling of using drones for reconnaissance first, followed by treatment. The transformation module uses mathematical transformations to linearize the nonlinear constraints and the first and second optimization objective functions in the rescue coordination model. The solution module solves the linearized rescue coordination model, outputs the optimal UAV reconnaissance path and rescue team dispatch plan, and evaluates them.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the steps of the method described in any one of claims 1-5.

Images