Emergency treatment medical resource scheduling method and device based on multi-objective optimization

Abstract

The application relates to an emergency treatment medical resource scheduling method and device based on multi-objective optimization. The method comprises the following steps: acquiring current emergency treatment basic information; calling a constructed multi-objective optimization model; adopting a random generation combined with a heuristic generation to generate an initial population; an individual in the initial population is obtained through double-section mixed integer coding on casualty information, medical institution information and emergency vehicle information; iterative optimization is performed on the initial population, the population individuals are non-dominantly sorted and elite individuals are reserved in the iterative process, constraint repair is performed on the individuals violating the constraint condition, and when a pre-set iteration stop condition is met, a Pareto optimal solution set of a first non-dominant layer is output; the Pareto optimal solution set is decoded, and a current emergency medical resource scheduling scheme is output. The method can improve the overall treatment efficiency and the treatment quality of severe casualties.

Description

Technical Field

[0001] This application relates to the field of resource scheduling technology, and in particular to a method and apparatus for scheduling emergency medical resources based on multi-objective optimization. Background Technology

[0002] Emergency scenarios such as sudden disasters, major traffic accidents, and public health emergencies are typically characterized by their suddenness, concentrated outbreaks of injuries, limited medical resources, and stringent treatment time windows. In these scenarios, the treatment process involves not only on-site first aid but also multiple stages, including patient triage, dispatch of emergency vehicles, selection of transport routes, reception at medical institutions, and subsequent tiered treatment. The three-tiered medical treatment system, consisting of primary emergency stations, secondary emergency centers, and tertiary general hospitals, is the core architecture of current emergency medical support. Delays in any stage can lead to worsening of injuries or even death. Therefore, achieving rapid, accurate, and efficient allocation of medical resources within a limited timeframe is a crucial issue in the emergency medical support system.

[0003] However, traditional emergency medical resource allocation methods often employ proximity-based allocation, first-come-first-served, random dispatch, or human experience-based decision-making, all based on simple, single-dimensional rules. These methods treat all medical institutions as homogeneous nodes with equal capabilities and all injured persons as indiscriminate dispatch objects, using only the shortest transport distance or first-come-first-served as the sole dispatch criterion. They fail to address the complex coupling relationships between multiple factors. While feasible in small-scale, resource-sufficient scenarios, in large-scale emergencies, this inevitably leads to the rapid overload of medical institutions closest to the accident site while other institutions remain idle. A large number of injured persons with low treatment needs occupy advanced medical resources capable of treating severe cases, while critically injured persons miss the golden window for treatment due to the inability to access matching medical resources in a timely manner. Simultaneously, high-value rapid resources such as helicopters are either abused by low-priority tasks, causing a surge in dispatch costs, or underutilized due to a lack of reasonable activation mechanisms. This fails to effectively coordinate the contradiction between limited medical resources and dynamic patient needs, making it difficult to guarantee overall treatment efficiency and the quality of care for critically injured persons. Summary of the Invention

[0004] Therefore, it is necessary to provide a method and apparatus for scheduling emergency medical resources based on multi-objective optimization to address the aforementioned technical problems.

[0005] A method for scheduling emergency medical resources based on multi-objective optimization, the method comprising: Obtain current basic emergency medical information; the basic emergency medical information includes information on the injured, medical institutions, and emergency vehicles. The pre-constructed multi-objective optimization model is invoked; the objective functions of the multi-objective optimization model include minimizing the weighted response time of the wounded, minimizing the risk of delayed treatment for critically ill patients, minimizing the imbalance of medical resource load, and minimizing the total cost of vehicle scheduling; the constraints of the multi-objective optimization model include unique allocation of wounded patients, capacity constraints of medical institutions, capacity constraints of vehicles, matching constraints of tiered treatment capabilities, constraints of the golden treatment time window, and scheduling threshold constraints for various emergency vehicles. An initial population is generated using a combination of random and heuristic generation methods; individuals in the initial population are obtained by performing two-segment mixed integer encoding on information about the wounded, medical institutions, and emergency vehicles. Iterative optimization is performed on the initial population. During the iteration, the population individuals are sorted in a non-dominated manner and elite individuals are retained. Constraint repair is performed on individuals that violate the constraints. When the pre-set iteration stopping condition is met, the iteration stops and the Pareto optimal solution set of the first non-dominated layer is output. The Pareto optimal solution set is decoded to output the current emergency medical resource scheduling scheme; the emergency medical resource scheduling scheme is used for dispatching emergency vehicles and allocating medical institutions for the wounded.

[0006] An emergency medical resource scheduling device based on multi-objective optimization, the device comprising: The information acquisition module is used to acquire current emergency medical treatment basic information; the emergency medical treatment basic information includes information on the injured, medical institutions, and emergency vehicles. The model scheduling module is used to call the constructed multi-objective optimization model. The objective functions of the multi-objective optimization model include minimizing the weighted response time of the wounded, minimizing the risk of delayed treatment for critically ill patients, minimizing the imbalance of medical resource load, and minimizing the total cost of vehicle scheduling. The constraints of the multi-objective optimization model include unique allocation of wounded patients, capacity constraints of medical institutions, capacity constraints of vehicles, matching constraints of tiered medical treatment capabilities, golden treatment time window constraints, and scheduling threshold constraints for various emergency vehicles. The population initialization module is used to generate an initial population by combining random generation and heuristic generation; the individuals in the initial population are obtained by performing two-segment mixed integer encoding on the information of the wounded, medical institutions, and emergency vehicles. The population optimization module is used to perform iterative optimization on the initial population. During the iteration process, the population individuals are sorted in a non-dominated manner and elite individuals are retained. Constraint repair is performed on individuals that violate the constraints. When the pre-set iteration stopping condition is met, the iteration stops and the Pareto optimal solution set of the first non-dominated layer is output. The result output module is used to decode the Pareto optimal solution set and output the current emergency medical resource scheduling plan; the emergency medical resource scheduling plan is used to dispatch emergency vehicles and allocate medical institutions for the wounded.

[0007] The aforementioned emergency medical resource scheduling method and apparatus based on multi-objective optimization, by invoking a multi-objective optimization model with four optimization objectives and multiple constraints, can balance treatment timeliness, critical illness risk, resource load, and scheduling cost, reasonably delineate the compliance boundaries of medical resource scheduling, encode three types of basic information in a two-segment mixed integer manner, and construct an initial population using a combination of random and heuristic generation. This balances the diversity of the algorithm's global search and the rationality of the initial solution, improving the optimization convergence ability. Iterative optimization of the initial population, combined with non-dominated sorting, elite individual retention, and constraint repair mechanisms, can obtain a Pareto optimal solution set with multi-objective equilibrium through hierarchical screening. By decoding the Pareto optimal solution set to generate a scheduling scheme, the precise allocation of emergency vehicles and medical institutions for the wounded can be intuitively achieved. This invention can balance the load on medical resources, reduce the risk of delayed critical illness treatment, and improve the efficiency and quality of emergency medical care. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating an emergency medical resource scheduling method based on multi-objective optimization in one embodiment; Figure 2 This is a schematic diagram of the convergence curve of the NSGA-II algorithm in one embodiment; Figure 3 This is a schematic diagram of the HV convergence curves of various algorithms in one embodiment, wherein, Figure 3 (a) HV convergence curve for improving the NSGA-II algorithm Figure 3 (b) shows the HV convergence curve of the MOEA / D algorithm. Figure 3 (c) shows the HV convergence curve of the PSO algorithm. Figure 3 (d) shows the HV convergence curve of the GA algorithm. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0010] In one embodiment, such as Figure 1 As shown, a multi-objective optimization-based emergency medical resource scheduling method is provided, including the following steps: Step 102: Obtain current basic emergency medical information. This information includes casualty information, medical institution information, and emergency vehicle information.

[0011] Emergency medical care basic information is a set of essential data elements necessary for the coordinated scheduling of medical resources in emergency medical care scenarios such as sudden disasters and major accidents. In a specific embodiment of this invention, the casualty information includes data representing the individualized treatment needs of the casualties, such as the severity of injury, location coordinates, time of injury, golden treatment window, treatment priority weight, and resource occupancy capacity. The medical institution information covers the treatment capacity level, remaining capacity, resource load status, geographical location, and attribute data suitable for the types of casualties admitted by primary emergency points, secondary emergency centers, and tertiary general hospitals. The emergency vehicle information includes relevant parameter data such as equipment type, speed, single-trip carrying capacity, operating cost, real-time availability status, and scheduling activation threshold of transport vehicles such as ambulances and rescue helicopters.

[0012] Step 104: Invoke the constructed multi-objective optimization model. The objective functions of the multi-objective optimization model include minimizing the weighted response time for wounded patients, minimizing the risk of delayed treatment for critically ill patients, minimizing the imbalance of medical resource load, and minimizing the total cost of vehicle scheduling. The constraints of the multi-objective optimization model include unique allocation constraints for wounded patients, capacity constraints for medical institutions, capacity constraints for vehicles, matching constraints for tiered medical treatment capabilities, constraints for the golden treatment time window, and threshold constraints for scheduling various emergency vehicles.

[0013] The multi-objective optimization model is a mixed-integer programming model adapted to the construction of a multi-level emergency medical treatment system. It can transform the complex scheduling relationships of casualty treatment, institutional admission, and vehicle transfer into a standardized mathematical solution. The objective function is an evaluation index that measures the overall merits of medical resource scheduling schemes from a quantitative perspective. It is used to quantitatively characterize the weighted response time of casualty treatment, the risk of delayed treatment for critically ill patients, the degree of overall resource load balance of medical institutions, and the economic cost of emergency vehicle scheduling. The constraints are the business rules and physical resource boundaries that emergency medical resource scheduling must follow. The compliance scope of scheduling is defined from dimensions such as the uniqueness of casualty allocation, the upper limit of institutional and vehicle capacity, the matching of injury and treatment capacity, the golden treatment time window, and vehicle scheduling thresholds.

[0014] Understandably, this step can integrate fragmented emergency medical treatment and dispatch requests into a quantifiable and solvable multi-objective optimization problem, simultaneously taking into account multiple mutually restraining dispatch optimization objectives. By relying on multiple constraints to limit the reasonable dispatch boundary, it can effectively avoid unreasonable dispatch behaviors such as cross-level treatment of the wounded, overloading of medical institutions, and abuse of emergency vehicles.

[0015] Step 106: An initial population is generated using a combination of random and heuristic generation. Individuals in the initial population are obtained by performing two-segment mixed integer encoding on information about the wounded, medical institutions, and emergency vehicles.

[0016] The initial population is the initial solution set composed of several coded individuals representing potential medical scheduling schemes when solving the scheduling model using intelligent optimization algorithms. The two-segment hybrid integer coding is a chromosome coding structure specifically designed for emergency medical scheduling, splitting the patient scheduling order segment and the resource allocation segment, with integer permutations and integer assignments corresponding to patient scheduling priority, medical institution matching relationships, and emergency vehicle allocation relationships, respectively. The random generation method generates a population of coded individuals by randomly arranging the order of the injured and randomly matching medical institutions and emergency vehicles, thus broadening the algorithm's search space. The heuristic generation method generates a population of coded individuals that conform to the scenario rules, based on actual medical treatment logic such as patient treatment priority, institution distance and load, and vehicle activation rules.

[0017] It is understandable that this step constructs the initial population by combining random generation with heuristic generation, which can simultaneously take into account the quality of the initial solution and the diversity of the population. Relying on heuristic rules, it can quickly generate high-quality initial solutions that fit the business logic of rescue and treatment, effectively improving the convergence speed of the algorithm. At the same time, by using random generation to retain the global search breadth, it is beneficial to reduce the probability of the algorithm getting trapped in local optima too early.

[0018] Step 108: Perform iterative optimization on the initial population. During the iteration, perform non-dominated sorting on the population individuals and retain elite individuals. Perform constraint repair on individuals that violate the constraints. When the pre-set iteration stopping condition is met, stop the iteration and output the Pareto optimal solution set of the first non-dominated layer.

[0019] Iterative optimization is a global optimization process that iteratively updates and optimizes potential medical scheduling solutions by repeatedly performing evolutionary, selection, and correction operations on the initial population. Non-dominated ranking is a rule for classifying population individuals into performance levels based on the objective function values ​​of a multi-objective optimization model. If an individual's objective function values ​​are all no weaker than another individual's, and at least one objective function value is better, then the former is considered to dominate the latter. Elite individuals are those with superior overall performance selected after non-dominated ranking, representing the optimal medical scheduling candidate solution in the current iteration. The Pareto optimal solution set is the set of optimal solutions composed of all individuals in the first non-dominated layer. No scheduling solution within this set can optimize a single objective without worsening other optimization objectives, thus adapting to different emergency scheduling decision preferences.

[0020] Understandably, this step enables continuous global iterative optimization and hierarchical selection of medical resource allocation schemes. By retaining the high-quality allocation schemes from each iteration through the elite individual retention mechanism, the optimal solution is prevented from being lost during the evolution process. Constraint repair includes two parts: hard constraint correction and soft constraint adaptation. Through constraint repair, it ensures that all output schemes comply with the hard rules of emergency treatment, and can stably output the optimal allocation solution set under multi-objective equilibrium, adapting to different decision-making tendencies in emergency command.

[0021] Step 110: Decode the Pareto optimal solution set and output the current emergency medical resource scheduling plan. The emergency medical resource scheduling plan is used for dispatching emergency vehicles and allocating casualties to medical facilities.

[0022] The decoding process can extract the order of casualty dispatch, matching relationships between medical institutions, and allocation relationships of emergency medical vehicles. The emergency medical resource dispatch plan is a structured decision-making scheme formed after decoding, clearly defining the corresponding transport vehicle, receiving medical institution, and dispatch execution order for each casualty. Emergency vehicle dispatch involves allocating vehicle deployment tasks, transport recipients, and operational arrangements according to the dispatch plan. Casualty medical institution allocation accurately matches casualties of different injury levels to medical institutions at corresponding levels of treatment capabilities based on the dispatch plan.

[0023] Understandably, this step can transform the abstract coding optimal solution at the algorithm level into medical resource dispatch instructions that can be directly implemented at the emergency rescue site, clearly standardize the business process of dispatching emergency vehicles and allocating injured persons to medical institutions, provide emergency command departments with intuitive and implementable dispatch decision-making basis, and improve the execution efficiency of emergency medical resource dispatch.

[0024] The aforementioned multi-objective optimization-based emergency medical resource scheduling method, by invoking a multi-objective optimization model with four optimization objectives and multiple constraints, can balance treatment timeliness, critical illness risk, resource load, and scheduling cost, reasonably delineating the compliance boundaries of medical resource scheduling. It uses a two-segment mixed integer encoding for three types of basic information and constructs an initial population using a combination of random and heuristic generation, balancing the diversity of the algorithm's global search with the rationality of the initial solution, improving optimization convergence. Iterative optimization of the initial population, combined with non-dominated sorting, elite individual retention, and constraint repair mechanisms, allows for hierarchical screening to obtain a Pareto optimal solution set that achieves multi-objective equilibrium. By decoding the Pareto optimal solution set to generate a scheduling scheme, it can intuitively and accurately allocate emergency vehicle dispatch and casualty medical institutions. This embodiment of the invention can balance medical resource load, reduce the risk of delayed critical illness treatment, and improve the efficiency and quality of emergency medical care.

[0025] In one embodiment, the method further includes: obtaining scheduling rule parameters and configuring a multi-objective optimization model based on the scheduling rule parameters; the scheduling rule parameters include graded treatment matching parameters, patient priority weights, golden treatment time window thresholds, and scheduling thresholds for various emergency vehicles.

[0026] In this embodiment, the problem of emergency medical resource allocation arises from highly uncertain scenarios such as sudden disasters, major traffic accidents, and public health emergencies. In these scenarios, injured individuals typically appear in a concentrated manner within a short period, with significant differences in the severity of their injuries. Simultaneously, the resources available for transport and treatment, such as ambulances, helicopters, medical personnel, beds, operating rooms, and ICUs, are limited in number, making it difficult to fully meet the needs of all injured individuals in a short time. Therefore, the core task of emergency medical allocation is to rationally match injured individuals, vehicles, and medical institutions under limited medical resources, enabling injured individuals to reach medical institutions with appropriate treatment capabilities in the shortest possible time.

[0027] From an operations research perspective, this problem can be viewed as a typical resource allocation and path optimization problem within a multi-tiered healthcare system. Its nodes mainly include casualty points, primary emergency response points, secondary emergency centers, tertiary general hospitals, and vehicle depots. Connectivity between nodes represents the passage relationships between ambulances or helicopters. Casualty demand manifests as the demand for medical services at different locations, with varying degrees of injury and different treatment time windows. Medical resource supply is reflected in the capacity, treatment capabilities, and vehicle availability of medical institutions at each level. Dispatch decisions include casualty allocation, vehicle dispatch, selection of transport routes, and hospital reception arrangements. In this system, primary emergency response points are typically located close to the accident site and have a fast response time, but their treatment capabilities are limited. Secondary emergency centers possess certain surgical capabilities and strong capabilities for treating severe injuries. Tertiary general hospitals have the most comprehensive resources and can receive critically injured patients, but their numbers are small, their locations may be far, and they are prone to congestion during large-scale emergencies. Therefore, the specific scheduling problem in this scenario can be summarized as follows: given information such as the location of the injured, the severity of the injury, the time of injury, the location and capacity of medical institutions, and the number and speed of ambulances and helicopters, determine which vehicle to use to transport each injured person, which type of medical institution to send them to, and which route to take to complete the transfer, so that the system achieves the overall optimal balance between treatment timeliness, resource utilization, critical care risk control, and scheduling costs.

[0028] Table 1 Functional Division of the Three-Tier Emergency Medical Treatment System

[0029] Traditional dispatching methods commonly employ strategies such as proximity allocation, first-come-first-served, and random dispatch. While these methods are computationally simple, they can easily lead to problems such as overload of local medical facilities, excessively long waiting times for critically injured patients, and underutilization or misuse of high-value resources like helicopters when there is a large number of casualties, strained medical resources, or complex road conditions.

[0030] To ensure the model's computability and to reflect the actual characteristics of emergency medical treatment scenarios as much as possible, this invention makes the following basic assumptions.

[0031] (1) Hierarchical treatment hypothesis: Medical institutions are divided into three levels according to their treatment capabilities.

[0032] Critically injured patients should, in principle, be sent to tertiary general hospitals. In special circumstances, they may be sent to secondary emergency centers for treatment to stabilize their vital signs, but they must not be sent to primary emergency points as their final treatment location.

[0033] (2) Assumption of identifiable casualty information: At each dispatch time, the emergency command center can obtain basic information on the casualties that have been found, including the casualty's location, time of injury, time of discovery, severity of injury, and initial treatment needs. Casualties that have not yet been found are not included in the current dispatch cycle, but can be added to the model in subsequent rolling dispatches.

[0034] (3) Assumptions on the conversion of medical resource capacity: The resources of medical institutions, such as beds, medical staff, emergency equipment, and drug reserves, are uniformly converted into the comprehensive treatment capacity. The amount of resources occupied by patients with different injury levels is different, with critically injured patients occupying the most resources, followed by seriously injured patients, and lightly injured patients occupying the least resources.

[0035] (4) Assumptions on vehicle operation: Ambulances travel along the road network, and their travel time is affected by distance, average speed and road conditions; helicopters fly approximately in a straight line, but there is a take-off preparation time, and they can only land at medical facilities with take-off and landing conditions.

[0036] (5) Time window assumption: Patients with different injury levels have different golden treatment time windows. The time window is the shortest for critically injured patients, followed by those with serious injuries, and relatively lenient for those with minor injuries. Patients will not be directly removed after the time window has expired, but there will be a penalty for exceeding the time limit.

[0037] (6) Assumption of task continuity: In principle, transfer tasks that have been executed should not be interrupted or redistributed arbitrarily; for tasks that have not yet been executed, they can be re-optimized based on the latest information in subsequent rolling scheduling cycles.

[0038] (7) Scheduling cycle assumption: The emergency command center performs rolling optimization at fixed time intervals, updating the status of the wounded, the status of vehicles, the remaining capacity of hospitals, and the road traffic conditions. The scheduling cycle of this invention can be set to 5 minutes.

[0039] Based on the above fundamental assumptions that define the overall boundaries and preconditions for emergency medical dispatch, and further adapting to the actual on-site rescue process, corresponding medical resource dispatch rules are designed. Specifically, these include: The time frame for treating the injured. Emergency medical treatment is highly time-sensitive. The "golden hour" concept in severe trauma treatment states that if critically injured patients receive effective treatment within a short period, their survival rate and prognosis will significantly improve; if the critical treatment time is exceeded, the risk of death and disability will rise rapidly. Existing research has shown that the golden hour concept has important guiding significance in the treatment of severe trauma. However, in actual dispatch, the injured person needs to go through multiple stages from injury to receiving effective treatment, including: the injured person is found and reported; the dispatch center allocates ambulances or helicopters; the vehicle travels from the station to the injured person's location; on-site triage, loading, and initial treatment; the vehicle transfers the injured person from the location to the medical institution; and the hospital hands over the patient and begins effective treatment. Let's assume the injured person... The time of injury was The time of arrival at the medical institution is The maximum permissible treatment time window is Ideally, the following should be satisfied:

[0040] Considering the unavoidable delays that may occur under resource constraints, a delay variable can be introduced:

[0041] in The larger the value, the more severe the delay beyond the golden treatment time. For critically injured patients, this delay is penalized more severely. There is a clear conflict between time sensitivity and dispatch efficiency. On the one hand, to shorten the waiting time for critically injured patients, vehicles should be dispatched as quickly as possible and transported directly to high-level hospitals; on the other hand, to improve ambulance utilization, minorly injured patients and some seriously injured patients can be transported together, but waiting for carpooling may increase the risk to the injured. Therefore, the model needs to strike a balance between rapid response and efficient loading.

[0042] The matching of tiered medical care capabilities with the needs of the injured is crucial. Emergency medical care is not a homogeneous service; patients with different injury levels require different levels of medical institutions. Critically injured patients typically require ICU, complex surgeries, and advanced life support equipment; seriously injured patients require routine surgery and continuous monitoring; and lightly injured patients mainly require basic treatment and short-term observation. Without considering the differences in the capabilities of medical institutions, two types of problems may arise: lower-level institutions may receive high-needs patients, such as sending critically injured patients to Level 1 emergency stations. Although the transport distance may be short, Level 1 emergency stations lack effective treatment capabilities, leading to ineffective treatment and secondary transfers; higher-level institutions may be occupied by low-needs patients. If a large number of lightly injured patients are sent to tertiary general hospitals, it will occupy beds, emergency resources, and medical staff, preventing the timely reception of critically injured patients who truly require advanced treatment. Therefore, the emergency dispatch model must reflect the matching relationship between medical service capabilities and the treatment needs of the injured. Indicates medical institutions Does the facility have the capability to treat injuries of a certain severity? The ability, if the wounded The severity level of the injury is Then it must satisfy:

[0043] Typically, the following matching rules can be set: Table 2. Matching Relationship between Medical Institution Level and Injury Type

[0044] This matching relationship reflects the principle of tiered treatment: critically injured patients are given priority to be sent to tertiary general hospitals; seriously injured patients are given priority to be sent to secondary or tertiary medical institutions; and slightly injured patients are given priority to be treated in primary or secondary institutions, so as to avoid excessive occupation of advanced medical resources.

[0045] Ambulance transfer rules. Ambulances and helicopters are key transportation resources connecting casualty locations with medical facilities. Due to the limited number of vehicles, their dispatch efficiency directly affects the overall treatment outcome. Ordinary ambulances can typically transport multiple lightly injured patients or some seriously injured patients simultaneously, while critically injured patients often require separate transport or occupy more capacity due to the need for stretchers, monitoring equipment, and accompanying medical personnel. From a resource utilization perspective, if an ambulance transports only one lightly injured patient at a time, it may result in wasted transport capacity; however, from a treatment timeliness perspective, delaying departure to wait for full capacity may lead to seriously injured patients waiting too long. Therefore, there is a contradiction between ambulance dispatch efficiency and the risk of patient waiting. An ambulance departure threshold mechanism can be introduced. Let ambulances... The maximum capacity is The current collection of wounded is The wounded Usage capacity is Then the ambulance loading rate is:

[0046] The ambulance dispatch rules can be designed as follows: an ambulance must depart immediately when the load factor reaches a threshold; if the threshold is not reached, but the waiting time for any injured person on board reaches the maximum permissible waiting time, the ambulance must still depart. This can be represented as:

[0047] in, For ambulance departure time, The time when the vehicle is fully loaded. For the wounded Boarding time, For the wounded The maximum allowed waiting time inside the vehicle. This mechanism makes the model more consistent with actual rescue processes and expands the dispatching problem from a simple vehicle allocation problem to a vehicle dispatching problem that considers departure strategies.

[0048] Helicopter Transfer Rules. Helicopters offer advantages in emergency rescue, including high speed, minimal impact from ground roads, and suitability for long-distance critical care transfers, especially in situations where roads are blocked, the accident site is far from a high-level hospital, or the injured have limited remaining golden treatment time. However, helicopter resources are scarce, takeoff preparation time is long, operating costs are high, and they are subject to weather, takeoff and landing point, and airspace conditions. Therefore, helicopters cannot be used indiscriminately as ordinary vehicles but should be used as a high-value, rapid resource with restricted dispatch. To avoid misuse of helicopter resources, helicopter activation thresholds can be set. Definition of Injured Persons. The helicopter's operational rating is:

[0049] in, For indicator functions; Indicates the wounded Distance to the nearest tertiary hospital The maximum permissible treatment time window is the time required to reach a medical facility. .when Helicopters may be used in certain situations. However, helicopters should only be prioritized when the injured are critically ill, the distance is long, ground ambulances cannot reach them in time, or the remaining treatment time is short. This allows for a basic balance between the effectiveness of critical care and the use of high-cost resources.

[0050] In one embodiment, the basic emergency medical information also includes road access information, which is used to calculate the patient transfer time and medical response time during the iterative optimization process.

[0051] In this embodiment, the patient transfer time and treatment response time are used to calculate the objective function of the weighted treatment response time in the multi-objective optimization model, and at the same time provide basic data for the quantitative assessment of the risk of delayed treatment for critically ill patients. Emergency medical resource allocation needs to simultaneously consider treatment efficiency, the risk of delayed treatment for critically ill patients, resource balance, and allocation costs. Therefore, this invention sets four optimization objectives, and the model variables are defined as follows: : Wounded soldiers collection. All newly discovered wounded soldiers will be added to the collection.

[0052] : wounded The time of injury.

[0053] : wounded The location of.

[0054] : wounded The severity level of the injury.

[0055] : wounded The window of opportunity for treatment.

[0056] : wounded Vehicle capacity occupied.

[0057] Assembly point for Level 1 emergency medical services.

[0058] Assemble at the Level 2 Emergency Center.

[0059] A collection of tertiary general hospitals.

[0060] Medical institutions The maximum capacity for admitting patients.

[0061] Medical institutions The remaining capacity at time t.

[0062] Medical institutions Is it possible to treat the level of [illness]? The wounded.

[0063] Medical institutions Resource utilization rate.

[0064] Ambulances assemble.

[0065] Helicopter assembly.

[0066] Vehicle Maximum load capacity.

[0067] Vehicle The capacity already carried.

[0068] Vehicle The average speed.

[0069] Vehicle Is it available?

[0070] Vehicle Operating cost per unit time.

[0071] Vehicle Preparation time.

[0072] :node To the node The distance between them.

[0073] :node To the node The road correction factor between them.

[0074] :node To the node Travel time between them.

[0075] : wounded Boarding time.

[0076] : wounded Arrival time.

[0077] : Variables assigned to medical institutions.

[0078] : Variables assigned to medical institutions.

[0079]

[0080]

[0081] (1) Minimize the weighted total response time for the injured: Response time is an important indicator for measuring the efficiency of emergency medical dispatch. The shorter the time from injury to arrival at a medical institution with treatment capabilities, the better the treatment effect is usually. Considering the difference in importance of different injury levels, priority weights for different injury levels are introduced. Construct a weighted total response time target:

[0082] (2) Minimize the penalty for delayed treatment of critically injured patients: Severe trauma treatment has a distinct time window. If a patient fails to reach a suitable medical institution beyond the golden treatment window, their risk of death or worsening of condition will significantly increase. Establish a penalty target for delayed treatment:

[0083] in, For the first i The treatment of the injured has exceeded the time limit.

[0084] (3) Minimizing the imbalance of medical resource load: In emergency scenarios, if all the injured are sent to the nearest hospital, it may cause some hospitals to quickly reach full capacity, while other hospitals still have spare capacity. This imbalance in resource allocation will reduce the overall treatment capacity of the system. Therefore, this invention introduces the concept of medical institution resource utilization rate:

[0085] (4) Minimize the total cost of vehicle dispatch: There are significant differences in the usage costs of ambulances and helicopters. Helicopters are fast, but scarce, expensive, and require long preparation times. Therefore, it is necessary to avoid using helicopters for low-return missions. Total vehicle runtime:

[0086] The model constraints include: Each wounded soldier can only be assigned to one medical facility:

[0087] Each wounded person can only be transported by one type of vehicle:

[0088] The injured can only be sent to medical institutions with the appropriate treatment capabilities:

[0089] No medical institution may use more resources than its capacity to receive wounded patients.

[0090] Each ambulance or helicopter must not exceed its maximum capacity for carrying casualties during a single mission.

[0091] Wounded personnel should arrive at medical facilities within the golden window for treatment. Delays are permissible under extreme resource constraints, but these delays will be penalized in the objective function.

[0092] There is a trade-off between ambulance capacity utilization and the risk of patients waiting. For minor injuries, waiting time can be extended for combined transport; for serious injuries, waiting time should be limited; for critically injured patients, ambulances should be dispatched immediately.

[0093]

[0094] Helicopters are fast, but their numbers are limited, their costs are high, and they require takeoff preparation time, therefore they cannot be used unconditionally. This invention establishes a helicopter activation score to determine whether a wounded person is suitable for helicopter transport:

[0095] when Helicopters can be used at this time, but they can only land at general hospitals.

[0096] In one embodiment, the method further includes: dynamically updating basic emergency medical information according to a preset scheduling cycle.

[0097] In this embodiment, since information about the injured, road conditions, medical facility capacity, and vehicle status may all change over time during emergency rescue, a single static optimization is insufficient to adapt to real-world scenarios. Therefore, this invention introduces a rolling time-domain scheduling mechanism. Let the scheduling period be: At each dispatch point, the system updates the following information: newly discovered casualties; casualties whose treatment and transfer have not yet been completed; remaining capacity of each medical institution; current location and availability of ambulances and helicopters; road conditions; and status of casualties who have been loaded but not yet delivered.

[0098] Within each scheduling cycle, the model only optimizes tasks in the set that have not yet been executed or are adjustable. For tasks that have already been executed and cannot be interrupted, the original plan remains unchanged. The rolling scheduling process is as follows: (1) At time 1 (2) Collect the latest information on wounded and resources; (3) Construct the current set to be scheduled. (3) Update the remaining capacity of medical institutions (4) Update vehicle availability status (5) Run the multi-objective optimization model; (6) Output the vehicle dispatch and casualty allocation plan for the current period; (7) Advance the time to Then, proceed to the next round of optimization. Through rolling optimization, the model can adapt to the realities of continuous casualties and dynamic changes in resource status.

[0099] It is worth noting that the number of wounded is set as follows: The number of medical institutions is The number of vehicles is If we only consider that each wounded person chooses one medical facility and one vehicle, the theoretical number of combinations can reach [number missing]. Therefore, it is evident that as the problem size increases, solving it using exact optimization methods will face a severe combinatorial explosion problem.

[0100] Furthermore, the model of this invention contains several conflicting optimization objectives, including: minimizing the weighted total response time for the injured; minimizing the risk of delays for critically injured and severely injured patients; minimizing the imbalance of resource load on medical institutions; and minimizing the total cost of dispatching emergency vehicles. These objectives are clearly conflicting. For example, if the shortest response time is the sole objective, the dispatch system tends to send the injured to the nearest hospital, which may lead to overload of local hospitals; if resource balance is excessively pursued, the transfer distance for some injured patients may increase; if dispatch costs are strictly controlled, helicopter usage may be reduced, thereby affecting the golden time satisfaction rate for critically injured patients. Therefore, single-objective optimization is difficult to fully reflect the actual needs of emergency medical dispatch, and it is also difficult to propose appropriate and precise weight values ​​to balance the various objectives in a short time during the actual decision-making process.

[0101] Therefore, this invention selects the non-dominated sorting genetic algorithm NSGA-II as the basic solution framework, and maintains the convergence and diversity of the solution set through non-dominated sorting and crowding distance, making it suitable for solving the multi-objective combinatorial optimization model constructed in this invention. NSGA-II can handle multiple optimization objectives simultaneously, but it is difficult to directly adapt to the characteristics of multi-objective coupling, multi-constraint intertwining, and complex scenario-specific scheduling rules in the emergency medical resource scheduling scenario. When applied, it is prone to problems such as inconsistent initial solution quality, insufficient adaptability to complex constraints, difficulty in integrating scenario scheduling rules into the algorithm iteration process, and the generation of a large number of infeasible solutions that do not conform to the actual treatment logic during the optimization process. To address these issues, this invention, based on the specific business characteristics of emergency medical scheduling, makes overall adaptation improvements to the standard NSGA-II algorithm. It integrates optimization design from multiple dimensions, including encoding structure, population initialization, constraint governance, objective function penalty, genetic operator design, and decoding rule embedding, enabling the improved algorithm to adapt to the complex constraints and scenario-specific scheduling rules of this model, reducing the proportion of infeasible solutions, and improving optimization convergence efficiency and scheduling scheme feasibility.

[0102] In one embodiment, the two-segment mixed integer encoding of casualty information, medical institution information, and emergency vehicle information includes: calculating a casualty dispatch sequence segment based on the injury level, treatment priority, remaining treatment time, and location attributes in the casualty information; generating a medical institution allocation segment based on the treatment capacity level, remaining admission capacity, and load status in the medical institution information; generating a vehicle allocation segment based on the type, availability status, and activation threshold conditions of the emergency vehicle information; and concatenating the casualty dispatch sequence segment, medical institution allocation segment, and vehicle allocation segment to obtain an individual encoded using the two-segment mixed integer encoding.

[0103] In this embodiment, chromosome encoding is crucial for solving combinatorial optimization problems using genetic algorithms. The problem addressed in this invention requires not only determining the service order of the wounded but also identifying the type of vehicle used to transport them and the medical facility to which they are taken. Therefore, a two-segment hybrid integer encoding structure is employed.

[0104] An individual with one chromosome can be represented as .in The sequence of services for the wounded. This is the resource allocation segment.

[0105] The sequence of services for the wounded is coded using permutation: This section indicates the order in which the wounded are prioritized for treatment by the dispatch system.

[0106] The resource allocation section indicates the medical facility and vehicle selection for each wounded soldier: ,in Indicates the wounded The medical facility to which they were sent. in Indicates the wounded By vehicle transport.

[0107] For example, encoded as ,in , , The injured were dispatched in the order of 3, 1, 5, 2, 4, with each injured person having a corresponding hospital and vehicle allocation plan.

[0108] In one embodiment, generating an initial population using a combination of random and heuristic generation includes: generating heuristic individuals based on the priority of casualty classification, pre-set matching rules for medical institution treatment capabilities, and rules for activating emergency vehicles; obtaining random individuals by randomly generating coding sequences and initially repairing codes that clearly violate hard constraints; and constructing an initial population based on a preset ratio of heuristic individuals and random individuals.

[0109] In this embodiment, the quality of population initialization directly affects the convergence speed and search performance of the genetic algorithm. If completely random initialization is used, it may generate a large number of infeasible or poor-quality solutions, such as critically injured patients being sent to primary emergency stations, minorly injured patients occupying a large amount of tertiary hospital resources, and distant casualties not being adequately treated by helicopter. Therefore, a hybrid initialization strategy of heuristic initialization + random perturbation is adopted. Random initialization accounts for 50% of the population, and heuristic initialization accounts for the other 50%.

[0110] The specific steps of heuristic initialization include: Based on the encoding principles mentioned above, and corresponding to the two-segment encoding, three encoding initialization schemes are designed.

[0111] (1) Order of care for the wounded Based on the priority weight of the injured Remaining treatment time and get to the nearest available treatment point Construct a comprehensive priority index:

[0112] in , , All are weighting constants. Arranging the wounded from largest to smallest number provides a heuristic service order. This ensures that critically injured patients, those with short remaining time, and those requiring long-distance transport are prioritized for dispatch.

[0113] (2) Allocation of medical institutions From the set of feasible institutions, a comprehensive score combining distance and capacity can be used to select medical institutions:

[0114] in , All values ​​are weighted constants. Selecting medical institutions with lower scores as target institutions prioritizes those that are closer and have a lighter workload.

[0115] (3) Vehicle allocation For critically injured patients, first determine if they meet the helicopter activation threshold. If they do, prioritize assigning a helicopter; otherwise, assign a critical care ambulance or the nearest available ambulance. For seriously injured and slightly injured patients, prioritize assigning ambulances that are closer and available earlier.

[0116] Random initialization is used to ensure population diversity and prevent the algorithm from getting trapped in local optima too early. The specific method is as follows: (1) Randomly generate a permutation of the wounded soldier numbers 1 to n, as the service sequence segment; (2) Randomly select a medical institution for each wounded soldier; randomly select a vehicle for each wounded soldier; (3) Perform preliminary repairs on genes that clearly violate hard constraints, such as sending the wounded to medical institutions that do not have the capacity to treat them.

[0117] Random initialization can broaden the search scope, giving the algorithm a chance to discover potential high-quality solutions that heuristic rules cannot generate.

[0118] In one embodiment, iterative optimization is performed on the initial population. During the iteration, the individuals in the population are sorted in a non-dominated manner, and elite individuals are retained. Constraint repair is performed on individuals that violate the constraints. When the pre-set iteration stopping condition is met, the iteration stops, and the Pareto optimal solution set of the first non-dominated layer is output. This includes: performing selection, crossover, and mutation operations on the current population individuals in sequence to update the offspring population; checking the compliance of the offspring population individuals; performing constraint repair on individuals that violate the constraints; comparing the performance of individuals according to the objective function and dividing the dominance level according to the Pareto dominance rule; selecting the better-performing elite individuals from each level to participate in the next round of iteration; stopping the iteration when the pre-set iteration stopping condition is met; and obtaining the Pareto optimal solution set based on the current individuals of the first non-dominated layer.

[0119] In this embodiment, genetic operators directly affect the algorithm's search capability. Because the chromosomes in this invention employ a two-segment hybrid encoding—the wounded soldier service sequence segment uses permutation encoding, and the resource allocation segment uses integer encoding—crossover and mutation operations need to be designed separately.

[0120] Selection operator: A binary tournament selection method is used. Two individuals are randomly selected each time. and The winning individual will be determined according to the following rules: like Then choose .

[0121] If the two individuals have the same level of non-dominance, then the individual with the larger crowding distance should be selected.

[0122] in, Represents an individual The non-dominated level. The smaller the non-dominated level, the better the individual; the larger the crowding distance, the sparser the individual is in the target space, which helps to maintain the diversity of Pareto solutions.

[0123] Crossover operator: This invention uses different crossover methods for the two parts of the chromosome.

[0124] (1) Service sequence segment crossover: The service sequence segment is the complete permutation of the casualty numbers. After crossover, it must be ensured that there are no duplicates or omissions. Therefore, this invention adopts partial matching crossover, i.e., PMX crossover. The steps are as follows: Randomly select two intersection positions and ; Swap two parent generations in the interval Gene fragments within; Repair duplicate genes outside the interval based on the mapping relationship; Obtain a valid offspring arrangement.

[0125] (2) Resource allocation segment crossover: The resource allocation segment is an integer code, including hospital allocation and vehicle allocation. This invention adopts a uniform crossover method. For each wounded soldier, the medical institution and vehicle allocation genes corresponding to the two parents are swapped with a probability of 0.5.

[0126] Mutation Operators: To enhance the algorithm's ability to escape local optima, this invention designs a variety of mutation operations corresponding to two-stage encoding.

[0127] (1) Service order swap mutation: Randomly select two positions in the service order segment of the wounded and swap their genes. This operation changes the service priority of the wounded and helps to search for different scheduling and ordering schemes.

[0128] (2) Medical institution redistribution variation: a wounded person was randomly selected. In its available medical facilities The target institution is then reselected. This operation helps improve hospital capacity distribution and resource load balancing.

[0129] (3) Randomly select one wounded person Based on the available vehicle set, a vehicle is reselected. For critically injured patients, if they meet the helicopter activation threshold, a helicopter can be assigned; otherwise, an ambulance is assigned first.

[0130] Based on the above design, the overall flow of the improved NSGA-II algorithm of this invention is shown in Table 3 below.

[0131] Table 3 Improved NSGA-II Algorithm Flowchart

[0132] Instead of using a fully random strategy during the initialization phase, the algorithm generates an initial population through a heuristic injection mechanism. This mechanism pre-assigns some critically injured patients to tertiary general hospitals based on the principle of matching injury priority with medical level, thereby providing a high-quality search starting point for population evolution and accelerating algorithm convergence.

[0133] An individual's gene sequence encodes the allocation scheme of wounded personnel, hospitals, and vehicles. During decoding, discrete genes must be mapped to specific scheduling sequences. Given the rigid requirements of emergency medical care on resource levels and vehicle environments, the algorithm introduces a forced repair operator. If the decoding result violates the hospital capacity limit or the principle of tiered medical care, the repair operator will dynamically adjust gene positions based on the principle of cost minimization, ensuring that each generation of individuals falls within the feasible solution domain and avoiding unnecessary computation.

[0134] The algorithm simultaneously calculates four core objective functions. For time window constraints such as the "golden hour," this invention employs an external penalty function method. When a scheduling scheme causes delays in treating the wounded, the algorithm adds a penalty term exponentially related to the delay duration to the original objective function. By widening the fitness gap between different schemes, this forces the population to evolve.

[0135] In the iterative loop, the algorithm generates offspring through binary tournament selection, multi-point discrete crossover, and adaptive mutation operators, and performs fast non-dominated sorting after merging parent and offspring. By calculating crowding distance, the algorithm retains the highest-level elite individuals while ensuring a uniform distribution of the population on the Pareto front. This mechanism ensures that the algorithm can escape local optima and exhibits a certain degree of robustness. Finally, it outputs a set of non-dominated Pareto optimal solutions, representing the optimal balance state under different weight preferences.

[0136] In one embodiment, the division of dominance levels according to the Pareto dominance rule based on the comparison of individual performance using objective functions includes: if all objective function values ​​of an individual are not weaker than those of another individual, and at least one objective function value is better, then the former is determined to dominate the latter; individuals with no other individuals to dominate are assigned to the first non-dominated level; and dominated individuals are assigned to the next lower dominance level in sequence.

[0137] In this embodiment, based on the fast non-dominated sorting mechanism of the NSGA-II algorithm, the individual performance of the population can be automatically classified and stratified according to the value of the multi-objective function. The optimal level individuals with no dominant individuals can be quickly screened out, providing a regular hierarchical basis for subsequent population optimization iterations, and ensuring the orderliness of the multi-objective optimization process and the accuracy of hierarchical screening.

[0138] In one embodiment, performing constraint repair on an individual who violates constraints includes: if the current individual violates a hard constraint, adjusting the encoded gene loci based on the principle of minimum cost to make the individual meet the constraint requirements; hard constraints include medical institution capacity constraints, vehicle capacity constraints, and tiered treatment capacity matching constraints; if the current individual violates a soft constraint, correcting the objective function calculation result using a penalty method; soft constraints include the golden treatment time window constraint.

[0139] In this embodiment, by differentiating between hard constraint coding gene position adjustment and soft constraint objective function penalty, the proportion of infeasible individuals that do not meet the rules of medical institution capacity, vehicle capacity and golden treatment time window during the iteration process can be effectively reduced, thereby improving the overall quality of the population and the stability of iterative optimization.

[0140] In one embodiment, decoding the Pareto optimal solution set to output the current emergency medical resource scheduling scheme includes: parsing the patient scheduling sequence segment, medical institution allocation segment, and vehicle allocation segment of the encoded individuals in the Pareto optimal solution set; establishing the corresponding relationship between injury level, medical institution treatment capacity, and emergency vehicles based on the parsed encoded information; and generating an emergency medical resource scheduling scheme based on the relationship.

[0141] The effectiveness of the method of this invention was verified through experiments. To comprehensively evaluate the scheduling scheme from the aspects of timeliness, security, balance, and economy, the following evaluation indicators were established in this invention.

[0142] (1) Weighted average response time: The average response time is used to measure the average time required for all injured persons to reach a medical facility from the time they are injured or found, and is defined as:

[0143] The smaller this indicator is, the higher the overall treatment response efficiency.

[0144] (2) Critical Care Golden Time Fulfillment Rate: The critical care golden time fulfillment rate measures the proportion of critically ill and severely injured patients who arrive at appropriate medical institutions within the prescribed treatment time window, and is defined as:

[0145] in, This indicates the total number of critically injured and seriously injured patients. This indicates the number of critically injured and severely injured patients who reach appropriate medical facilities within the golden time window. The higher this indicator, the stronger the dispatch plan's ability to protect critically injured patients.

[0146] (3) Variance of medical resource load: Variance of medical resource load is used to measure the degree of balance in resource utilization among medical institutions.

[0147]

[0148] (4) Total Dispatch Cost: The total dispatch cost consists of ambulance operating costs and helicopter flight costs, and is defined as follows:

[0149] The improved multi-objective optimization algorithm proposed in this invention was compared with the traditional nearest-neighbor allocation + first-come-first-served greedy strategy and random allocation scheme. To eliminate random errors, the algorithm was run independently 20 times and the average value was taken. The experimental results are shown in Table 4.

[0150] Table 4 Comparison of different scheduling strategies in low-density scenarios

[0151] As shown in Table 4, in low-density scenarios, both the proximity allocation strategy and the algorithm of this invention can ensure that critically injured patients receive treatment within the golden time window, with a 100% satisfaction rate for the golden time for critically injured patients. This is because medical resources are relatively abundant in this scenario, the number of injured patients is relatively small, and there is no significant resource competition during the scheduling process.

[0152] Regarding average response time, the nearest allocation strategy achieves 28.33 minutes, while the algorithm of this invention achieves 31.69 minutes, a relatively small difference. The reason for the slight advantage of nearest allocation is that this strategy directly selects the nearest medical institution, enabling shorter transportation times when resources are sufficient. In contrast, the algorithm of this invention considers not only response time but also medical resource load balancing and scheduling costs. Therefore, it may allocate some lightly injured patients to slightly more distant but less overloaded medical institutions, resulting in a slight increase in average response time.

[0153] Regarding resource load variance, the algorithm of this invention has a variance of 181.34, which is significantly lower than the 435.22 of the nearest allocation strategy, indicating that the algorithm of this invention can better balance the receiving pressure of various medical institutions. Although the effect of resource balancing is not as obvious in low-density scenarios as in high-density scenarios, this result shows that the algorithm of this invention will not cause unnecessary load concentration in local medical institutions in pursuit of the shortest time.

[0154] The random allocation strategy is significantly inferior to the other two strategies in terms of average response time and the rate of meeting the golden time for critical care, indicating that disordered scheduling is difficult to meet the time requirements of emergency treatment. Therefore, in the subsequent analysis of medium and high density scenarios, the nearest allocation strategy and the algorithm of this invention will be mainly compared.

[0155] Overall, in small-scale scenarios with sufficient resources, the algorithm of this invention can maintain a response efficiency similar to that of the traditional proximity allocation strategy, while performing better in terms of resource load balancing, indicating that the proposed model has good stability and adaptability in simple scenarios.

[0156] To further verify the solution performance of the improved NSGA-II algorithm, this invention observes the changes in the objective function during the algorithm's iteration process. Taking a medium-density scenario as an example, the average response time, severe overdue penalty, and resource load variance changes of the algorithm at different iteration numbers are recorded. Figure 2The diagram showing the convergence curve of the NSGA-II algorithm illustrates that, from the perspective of the algorithm's iteration process, the convergence process of the improved NSGA-II can be roughly divided into three stages. The first stage is the rapid descent stage. In the early stages of iteration, the population contains many random and infeasible solutions, resulting in poor objective function values. After non-dominated sorting, selection, crossover, and mutation operations, inferior individuals are quickly eliminated, and the average response time and overdue penalty decrease significantly. The second stage is the local adjustment stage. As iterations progress, most individuals in the population satisfy the basic constraints, and the algorithm begins to carefully weigh response time, resource balance, and scheduling costs. At this point, the rate of descent of the objective function slows down, but the quality of the Pareto solution set gradually improves. The third stage is the stable convergence stage. When the number of iterations reaches a certain scale, the number of non-dominated solutions in the population tends to stabilize, the variation amplitude of each objective function decreases, and the algorithm gradually converges. Due to the adoption of the crowding distance preservation mechanism, the final solution set has a good distribution in the objective space, providing decision-makers with multiple optional scheduling schemes. Therefore, the improved NSGA-II can better adapt to the solution requirements of the multi-objective, multi-constraint, and high-dimensional combinatorial optimization problem of this invention.

[0157] Based on the multi-objective evaluation system of operations research, this invention selects hypervolume (HV) and generation distance (IGD) as core evaluation indicators. Hypervolume refers to the volume of the hypercube enclosed in the target space between the non-dominated solution set obtained by the algorithm and the reference point. A larger HV value indicates better convergence and diversity of the solution set. Generation distance measures the average of the shortest distances from points on the true Pareto front to the non-dominated front obtained by the algorithm. A smaller IGD value indicates that the solution set obtained by the algorithm approximates the true optimal front more closely.

[0158] Under the same medium-density sudden conflict scenario, the three algorithms were run independently 10 times, with a population size of 100 and a maximum number of iterations of 1000 generations. The performance statistics are shown in Table 5.

[0159] Table 5 Comparison of performance evaluation metrics for each algorithm

[0160] like Figure 3 The diagrams shown illustrate the HV convergence curves of each algorithm. Figure 3 (a) HV convergence curve for improving the NSGA-II algorithm Figure 3 (b) shows the HV convergence curve of the MOEA / D algorithm. Figure 3 (c) shows the HV convergence curve of the PSO algorithm. Figure 3(d) shows the HV convergence curve of the GA algorithm. The initial HV value of the improved NSGA-II is significantly higher than that of other algorithms. This is because the present invention introduces a heuristic + random hybrid initialization strategy. In operations research, pure random search is prone to generating a large number of infeasible solutions in highly constrained feasible regions; while heuristic rules are equivalent to providing the algorithm with a high-quality initial feasible basis solution, enabling the population to quickly lock onto the high-quality sub-regions of the target space in the early stages of iteration, and to stably output feasible solutions that satisfy the constraints in the first 5 generations. The GA algorithm has the shortest running time, but because it uses linear weighted summation, it cannot search for the optimal solution on the concave surface for non-convex Pareto fronts in multi-objective optimization, and is prone to getting trapped in local optima. Although the improved NSGA-II has a slightly longer running time, this is because it incorporates feasibility repair and dynamic penalty function calculation, exchanging a certain computational cost for a significant leap in the quality of the solution set.

[0161] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated in this invention, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Furthermore, Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0162] In one embodiment, an emergency medical resource scheduling device based on multi-objective optimization is provided, comprising: The information acquisition module is used to acquire current basic emergency medical information, including information on the injured, medical institutions, and emergency vehicles. The model scheduling module is used to call the constructed multi-objective optimization model. The objective functions of the multi-objective optimization model include minimizing the weighted response time of the wounded, minimizing the risk of delayed treatment for critically ill patients, minimizing the imbalance of medical resource load, and minimizing the total cost of vehicle scheduling. The constraints of the multi-objective optimization model include unique allocation of wounded patients, capacity constraints of medical institutions, capacity constraints of vehicles, matching constraints of tiered medical treatment capabilities, golden treatment time window constraints, and scheduling threshold constraints for various emergency vehicles. The population initialization module is used to generate the initial population by combining random generation and heuristic generation. Individuals in the initial population are obtained by performing two-segment mixed integer encoding on the information of the wounded, medical institutions, and emergency vehicles. The population optimization module is used to perform iterative optimization on the initial population. During the iteration process, the population individuals are sorted in a non-dominated manner and elite individuals are retained. Constraint repair is performed on individuals that violate the constraints. When the pre-set iteration stopping condition is met, the iteration stops and the Pareto optimal solution set of the first non-dominated layer is output. The output module is used to decode the Pareto optimal solution set and output the current emergency medical resource scheduling plan; the emergency medical resource scheduling plan is used to dispatch emergency vehicles and allocate medical institutions for the wounded.

[0163] Specific limitations regarding the multi-objective optimization-based emergency medical resource scheduling device can be found in the limitations of the multi-objective optimization-based emergency medical resource scheduling method described above, and will not be repeated here. Each module in the aforementioned multi-objective optimization-based emergency medical resource scheduling device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0164] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0165] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A multi-objective optimization-based emergency treatment medical resource scheduling method, characterized in that, The method includes: Obtain current basic emergency medical information; the basic emergency medical information includes information on the injured, medical institutions, and emergency vehicles. The pre-constructed multi-objective optimization model is invoked; the objective functions of the multi-objective optimization model include minimizing the weighted response time of the wounded, minimizing the risk of delayed treatment for critically ill patients, minimizing the imbalance of medical resource load, and minimizing the total cost of vehicle scheduling; the constraints of the multi-objective optimization model include unique allocation of wounded patients, capacity constraints of medical institutions, capacity constraints of vehicles, matching constraints of tiered treatment capabilities, constraints of the golden treatment time window, and scheduling threshold constraints for various emergency vehicles. An initial population is generated using a combination of random and heuristic generation methods; individuals in the initial population are obtained by performing two-segment mixed integer encoding on information about the wounded, medical institutions, and emergency vehicles. Iterative optimization is performed on the initial population. During the iteration, the population individuals are sorted in a non-dominated manner and elite individuals are retained. Constraint repair is performed on individuals that violate the constraints. When the pre-set iteration stopping condition is met, the iteration stops and the Pareto optimal solution set of the first non-dominated layer is output. The Pareto optimal solution set is decoded to output the current emergency medical resource scheduling scheme; the emergency medical resource scheduling scheme is used for dispatching emergency vehicles and allocating medical institutions for the wounded.

2. The method of claim 1, wherein, The two-segment mixed integer encoding of casualty information, medical institution information, and emergency vehicle information includes: The order of dispatching casualties is calculated based on the severity of injury, priority of treatment, remaining treatment time and location attributes in the casualty information. Based on the treatment capacity level, remaining admission capacity and load status of medical institutions, a medical institution allocation sub-segment is generated. Based on the type, availability, and activation threshold conditions of the emergency vehicle information, a vehicle allocation sub-segment is generated. By concatenating the wounded dispatch sequence segment, the medical institution allocation segment, and the vehicle allocation segment, an individual with a two-segment mixed integer encoding is obtained.

3. The method of claim 1, wherein, The initial population is generated using a combination of random and heuristic generation methods, including: Based on the priority of casualty classification, pre-set matching rules for medical institution treatment capacity, and rules for the activation of emergency vehicles, heuristic individuals are generated. Random individuals are obtained by randomly generating encoding sequences and initially repairing encodings that clearly violate hard constraints; An initial population is constructed based on a preset ratio of heuristic individuals and random individuals.

4. The method according to claim 1, characterized in that, Iterative optimization is performed on the initial population. During the iteration, the population individuals are non-dominated and elite individuals are retained. Constraint repair is performed on individuals that violate the constraints. When the pre-set iteration stopping condition is met, the iteration stops, and the Pareto optimal solution set of the first non-dominated layer is output, including: The selection, crossover, and mutation operations are sequentially performed on the current population individuals to update the offspring population. The compliance of the offspring population individuals is checked, and constraint repair is performed on individuals that violate the constraints. Based on the objective function, individual performance is compared, and domination levels are determined according to the Pareto domination rule. Select the best performing elite individuals from each level to participate in the next round of iteration. When the pre-set iteration stopping condition is met, stop the iteration and obtain the Pareto optimal solution set based on the individuals in the current first non-dominated layer.

5. The method according to claim 4, characterized in that, The step of comparing individual performance based on the objective function and dividing the dominance levels according to the Pareto dominance rule includes: If all objective function values ​​of an individual are no weaker than those of another individual, and at least one objective function value is better, then the former is determined to dominate the latter. Individuals that have no other individuals to control are classified into the first non-dominated layer; The dominated individuals are successively assigned to the next level of domination.

6. The method according to claim 1, characterized in that, The constraint repair performed on individuals who violate the constraints includes: If an individual violates the hard constraints, the coding gene positions are adjusted based on the principle of minimum cost to make the individual meet the constraints; the hard constraints include medical institution capacity constraints, vehicle capacity constraints, and tiered medical treatment capacity matching constraints. If an individual violates the soft constraints, the objective function calculation result will be corrected by a penalty method; the soft constraints include the golden treatment time window constraint.

7. The method according to claim 1, characterized in that, Decoding the Pareto optimal solution set and outputting the current emergency medical resource scheduling scheme includes: The Pareto optimal solution set is analyzed by encrypting the wounded dispatch sequence segment, medical institution allocation segment, and vehicle allocation segment of the encoded individuals. Based on the coded information obtained from the analysis, establish a corresponding relationship between injury level, medical institution's treatment capacity, and emergency vehicle; An emergency medical resource allocation plan is generated based on the aforementioned relationships.

8. The method according to claim 1, characterized in that, The method further includes: Obtain scheduling rule parameters and configure the multi-objective optimization model according to the scheduling rule parameters; the scheduling rule parameters include graded treatment matching parameters, patient priority weights, golden treatment time window thresholds, and scheduling thresholds for various emergency vehicles.

9. The method according to claim 1, characterized in that, The basic information for emergency medical treatment also includes road traffic information, which is used to calculate the patient transfer time and medical response time during the iterative optimization process.

10. An emergency medical resource scheduling device based on multi-objective optimization, characterized in that, The device includes: The information acquisition module is used to acquire current emergency medical treatment basic information; the emergency medical treatment basic information includes information on the injured, medical institutions, and emergency vehicles. The model scheduling module is used to call the constructed multi-objective optimization model. The objective functions of the multi-objective optimization model include minimizing the weighted response time of the wounded, minimizing the risk of delayed treatment for critically ill patients, minimizing the imbalance of medical resource load, and minimizing the total cost of vehicle scheduling. The constraints of the multi-objective optimization model include unique allocation of wounded patients, capacity constraints of medical institutions, capacity constraints of vehicles, matching constraints of tiered medical treatment capabilities, golden treatment time window constraints, and scheduling threshold constraints for various emergency vehicles. The population initialization module is used to generate an initial population by combining random generation and heuristic generation; the individuals in the initial population are obtained by performing two-segment mixed integer encoding on the information of the wounded, medical institutions, and emergency vehicles. The population optimization module is used to perform iterative optimization on the initial population. During the iteration process, the population individuals are sorted in a non-dominated manner and elite individuals are retained. Constraint repair is performed on individuals that violate the constraints. When the pre-set iteration stopping condition is met, the iteration stops and the Pareto optimal solution set of the first non-dominated layer is output. The result output module is used to decode the Pareto optimal solution set and output the current emergency medical resource scheduling plan; the emergency medical resource scheduling plan is used to dispatch emergency vehicles and allocate medical institutions for the wounded.

Images