Emergency treatment resource scheduling method based on multi-stage dynamic game

Abstract

The invention discloses an emergency treatment resource scheduling method based on a multi-stage dynamic game, and relates to the technical field of medical emergency management, and the method comprises the steps: according to a three-party dynamic game model corresponding to a pre-hospital first-aid team, a hospital department and a transfer unit, and the medical multi-modal data of each patient in a current treatment scene, carrying out the multi-stage dynamic game; determining a pre-hospital first-aid team treatment strategy, an in-hospital department treatment strategy and a transfer unit treatment strategy; according to the three-party treatment strategy, robust scheduling based on meta-learning is carried out on current existing emergency treatment resources, and a robust scheduling strategy is determined; iCU bed scheduling, operating room scheduling and ambulance scheduling are executed according to the robust scheduling strategy. According to the method, emergency treatment resource allocation balance under the dynamic game of multiple participants is realized, meanwhile, sudden disturbance can be quickly adapted through the meta-learning framework, the purpose of effectively allocating emergency treatment resources in response to emergencies is also realized, and the adaptability of a robust scheduling strategy to the emergencies is improved; and the resource scheduling strategy has interpretability and stability.

Description

Technical Field

[0001] This disclosure generally relates to the field of medical emergency management technology, and in particular to an emergency resource scheduling method based on multi-stage dynamic game theory. Background Technology

[0002] When public health emergencies or natural disasters occur, emergency resources are usually needed to treat a large number of patients. However, the resource scheduling model used by traditional emergency resource scheduling methods assumes that the three parties involved—the pre-hospital emergency team, the hospital department, and the transport unit—are completely rational and have symmetrical information. Furthermore, the traditional reinforcement learning algorithm used relies on a fixed resource state space, which makes it difficult to dynamically and evenly allocate emergency resources and cope with the effective allocation of emergency resources under emergencies. Summary of the Invention

[0003] In view of the above-mentioned defects or deficiencies in the existing technology, it is desirable to provide an emergency resource scheduling method based on multi-stage dynamic game theory. By using a multi-party dynamic game model, the limitations of static game theory can be solved, and the equilibrium of emergency resource allocation under multi-participant dynamic game theory can be achieved. At the same time, through the meta-learning framework, it can quickly adapt to sudden disturbances and generate robust scheduling strategies, solving the problem of insufficient real-time performance. This achieves the goal of effectively allocating emergency resources in response to emergencies, thereby significantly improving the adaptability of the robust scheduling strategy to emergencies and making the resource scheduling strategy interpretable and stable.

[0004] Firstly, this application provides an emergency resource scheduling method based on multi-stage dynamic game theory. The method includes: Based on the three-party dynamic game model corresponding to the pre-hospital emergency team, in-hospital departments and transport units, and the medical multimodal data of each patient in the current treatment scenario, the pre-hospital emergency team treatment strategy, in-hospital department treatment strategy and transport unit treatment strategy that represent the balance of interests of all parties are determined. Based on the pre-hospital emergency team treatment strategy, the in-hospital department treatment strategy, and the transport unit treatment strategy, robust scheduling of existing emergency resources based on meta-learning is performed to determine a robust scheduling strategy; the robust scheduling strategy includes the ICU beds, operating rooms, and ambulances required to complete the treatment of all the patients within a preset short time period. ICU bed scheduling, operating room scheduling, and ambulance scheduling are performed according to the robust scheduling strategy described above.

[0005] In conjunction with the first aspect, in one possible implementation, the determination of the pre-hospital emergency team treatment strategy, in-hospital department treatment strategy, and transport unit treatment strategy, representing the balance of interests among the parties, based on the three-party dynamic game model corresponding to the pre-hospital emergency team, in-hospital department, and transport unit, and the medical multimodal data of each patient in the current treatment scenario, includes: Based on the medical multimodal data of each patient, determine the current patient state vector and the current treatment priority score for each patient; Each current patient state vector and each current treatment priority score, as well as the current environmental disturbance parameters and the current emergency resource tension index, are input into the three-party dynamic game model to solve the three-party Nash equilibrium of the pre-hospital emergency team, the in-hospital department, and the transport unit, thereby obtaining the treatment strategy of the pre-hospital emergency team, the treatment strategy of the in-hospital department, and the treatment strategy of the transport unit. The three-party Nash equilibrium is used to characterize the strategies chosen by each party in a dynamic game, such that when the strategies of other parties remain unchanged, the party cannot increase its benefits by unilaterally changing its strategy.

[0006] In conjunction with the first aspect, in one possible implementation, determining the current patient state vector and current treatment priority score for each patient based on their medical multimodal data includes: Based on the physiological signals, medical history text, and image data of each patient, determine the current patient state vector for each patient; Based on each patient's SOFA score, vital sign variability, number of underlying diseases, and age group, a current treatment priority score is determined for each patient.

[0007] In conjunction with the first aspect, in one possible implementation, determining the current environmental disturbance parameters and the current emergency resource strain index includes: The current environmental disturbance parameters are determined based on the current traffic congestion index, the current weather disaster level, the current disease transmission rate, and the current distance between hospitals. The current emergency resource shortage index is determined based on parameters such as the number of available ICU beds, the current ventilator occupancy rate, the current ambulance response time, and the current operating room availability.

[0008] In conjunction with the first aspect, in one possible implementation, the method further includes: Based on the severity of each patient's condition and clinical indicators, the resource requirements for each patient are determined: The resource requirements of each patient are quantified to obtain the current number of available ICU beds, the current ventilator occupancy rate, the current ambulance response time, and the current operating room vacancy status parameters.

[0009] In conjunction with the first aspect, in one possible implementation, determining the severity of each patient's condition includes: Based on each patient's SOFA score, vital sign variability, number of underlying diseases, age group, and clinical indicators, disease severity is predicted and determined for each patient.

[0010] In conjunction with the first aspect, in one possible implementation, the robust scheduling strategy for existing emergency resources based on meta-learning is determined according to the pre-hospital emergency team's treatment strategy, the in-hospital department's treatment strategy, and the transport unit's treatment strategy. This robust scheduling strategy includes: Based on the game strategy update frequency, data fusion latency, and meta-learning inner loop steps, the system load rate used to determine the meta-learning update frequency is determined. The system load rate, the existing emergency resources, the pre-hospital emergency team treatment strategy, the in-hospital department treatment strategy, and the transport unit treatment strategy are input into the meta-learning optimization model for robust scheduling, and the robust scheduling strategy output by the meta-learning optimization model is obtained.

[0011] In conjunction with the first aspect, in one possible implementation, the method further includes: The patient arrival status and emergency resource occupancy status will be determined based on the execution results; Based on the patient arrival status and the emergency resource occupancy status, the robust scheduling strategy is dynamically adjusted to obtain an adjusted robust scheduling strategy, which is then used to execute ICU bed scheduling, operating room scheduling, and ambulance scheduling.

[0012] Secondly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the emergency resource scheduling method based on multi-stage dynamic game as described in the first aspect.

[0013] Thirdly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the emergency resource scheduling method based on multi-stage dynamic game as described in the first aspect.

[0014] The emergency resource scheduling method based on multi-stage dynamic game theory provided in this application first determines the pre-hospital emergency team treatment strategy, in-hospital department treatment strategy, and transport unit treatment strategy, representing the balance of interests among the parties, through a three-party dynamic game model. Then, it further combines these treatment strategies to perform robust scheduling of existing emergency resources based on meta-learning, determining a robust scheduling strategy. Finally, it executes ICU bed scheduling, operating room scheduling, and ambulance scheduling according to the robust scheduling strategy. In this way, the multi-party dynamic game model overcomes the limitations of static game theory, achieving an equilibrium in emergency resource allocation under multi-participant dynamic game theory. Simultaneously, the meta-learning framework enables rapid adaptation to sudden disturbances, generating robust scheduling strategies, solving the problem of insufficient real-time performance, and achieving the goal of effective allocation of emergency resources in response to emergencies. This significantly improves the adaptability of the robust scheduling strategy to emergencies, making the resource scheduling strategy interpretable and stable. Attached Figure Description

[0015] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is one of the flowcharts of an emergency resource scheduling method based on multi-stage dynamic game theory in one embodiment; Figure 2 This is the second schematic diagram of an emergency resource scheduling method based on multi-stage dynamic game theory in one embodiment; Figure 3 This is the third schematic diagram of an emergency resource scheduling method based on multi-stage dynamic game theory in one embodiment; Figure 4 This is the fourth schematic diagram of an emergency resource scheduling method based on multi-stage dynamic game theory in one embodiment; Figure 5 This is the fifth schematic diagram of an emergency resource scheduling method based on multi-stage dynamic game theory in one embodiment; Figure 6 This is the sixth schematic diagram of an emergency resource scheduling method based on multi-stage dynamic game theory in one embodiment; Figure 7 This is the seventh schematic diagram of an emergency resource scheduling method based on multi-stage dynamic game theory in one embodiment; Figure 8 This is a schematic diagram of the model architecture of an emergency resource intelligent scheduling and emergency system collaborative optimization method based on multi-stage dynamic game theory in one embodiment. Figure 9 This is a block diagram of an emergency resource scheduling device based on multi-stage dynamic game theory in one embodiment. Figure 10 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0016] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present application will now be described in detail with reference to the accompanying drawings and embodiments. Furthermore, the term "and / or" in this document 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, or B existing alone. The terms "first" and "second," etc., in the specification and claims of the embodiments of this application are used to distinguish different objects, not to describe a specific order of objects.

[0018] When public health emergencies or natural disasters occur, emergency resources are usually needed to treat a large number of patients. However, the resource scheduling model used by traditional emergency resource scheduling methods assumes that the three parties involved—the pre-hospital emergency team, the hospital department, and the transport unit—are completely rational and have symmetrical information. Furthermore, the traditional reinforcement learning algorithm used relies on a fixed resource state space, which makes it difficult to dynamically and evenly allocate emergency resources and cope with the effective allocation of emergency resources under emergencies.

[0019] For example, traditional emergency resource scheduling methods have the following three technical bottlenecks: Technical Bottleneck 1: Limitations of Static Game Theory: Traditional resource scheduling models (such as integer programming) assume that the three participants—pre-hospital emergency teams, in-hospital departments, and transport units—are completely rational and have symmetrical information, without considering the dynamic conflicts of interest and strategic game among them.

[0020] Technical bottleneck 2: High-dimensional data coupling: In emergency scenarios, patient conditions (physiological indicators, imaging data), emergency resource status (beds, equipment), and environmental variables (traffic congestion, weather) form high-dimensional heterogeneous data, which existing models cannot effectively integrate.

[0021] Technical bottleneck 3: Traditional reinforcement learning methods rely on a fixed resource state space, which makes it difficult to cope with sudden disturbances in emergency scenarios (such as a large influx of injured people or equipment failure).

[0022] To address the aforementioned technical issues, this application provides an emergency resource scheduling method based on multi-stage dynamic game theory.

[0023] The following is combined Figures 1 to 10This application describes an emergency resource scheduling method based on multi-stage dynamic game theory. The executing entity of this method can be a personal computer, server, embedded system, or other computer device. This application does not specifically limit its application in this regard. Furthermore, the emergency resource scheduling method based on multi-stage dynamic game theory can also be applied to an emergency resource scheduling device based on multi-stage dynamic game theory installed in a computer device. This device can be implemented through software, hardware, or a combination of both. The following description uses a computer device as an example to illustrate this method.

[0024] To facilitate understanding of the emergency resource scheduling method based on multi-stage dynamic game theory provided in this application, the following exemplary embodiments will provide a detailed description of the method. It is understood that these exemplary embodiments can be combined with each other, and similar concepts or processes may not be repeated in some embodiments.

[0025] refer to Figure 1 This is a schematic diagram of the emergency resource scheduling method based on multi-stage dynamic game theory provided in the embodiments of this application, such as... Figure 1 As shown, the emergency resource scheduling method based on multi-stage dynamic game includes the following steps 101 to 103.

[0026] Step 101: Based on the three-party dynamic game model corresponding to the pre-hospital emergency team, in-hospital departments and transport units, and the medical multimodal data of each patient in the current treatment scenario, determine the pre-hospital emergency team treatment strategy, in-hospital department treatment strategy and transport unit treatment strategy that represent the balance of interests among the parties.

[0027] It should be noted that the three-party dynamic game model in step 101 can be constructed based on the first historical dataset. Each first historical data in the first historical dataset includes multiple core variables of multiple different variable categories, as well as the pre-hospital emergency team treatment strategy, in-hospital department treatment strategy, and transport unit treatment strategy under the corresponding historical treatment scenarios.

[0028] For example, when the variable categories are patient characteristic variables, resource status variables, environmental variables, system parameter variables, dynamically derived variables, game strategy variables, and scheduling outcome variables, there can be 24 core variables. These 24 core variables are used to aggregate four types of data—physiological, resource, environmental, and system—in real time, and can output a unified variable set, which serves to provide panoramic situational awareness. The classification of these 24 core variables is shown in Table 1.

[0029] Table 1

[0030] Using the aforementioned first historical dataset and considering the pre-hospital emergency team, in-hospital departments, and transport units as three participants, the weight coefficients of the profit functions of the pre-hospital emergency team, in-hospital departments, and transport units are solved to obtain the weight coefficients in the profit functions of each of the three parties. Then, based on the profit functions of each party whose weight coefficients are known, a three-party dynamic game model is constructed.

[0031] For example, the revenue function of the pre-hospital emergency team is defined as follows: It can represent the immediate gains of the pre-hospital emergency team in a multi-stage game, and is used to calculate the Nash equilibrium and guide actions such as ambulance routes and patient priorities, as shown in Equation (1).

[0032] (1) In equation (1), This represents the strategy vector that the pre-hospital emergency care team can choose in a dynamic game. This represents the strategy vector that departments within the hospital can choose in a dynamic game. This represents the strategy vector that a transport unit can choose in a dynamic game; This is the weighting coefficient for transit time, representing the impact of transit time on revenue; For penalty items The weighting coefficient represents the negative impact of operational errors or delays on revenue; For synergistic benefits The weighting coefficient represents the positive impact of collaborating with other participants on the returns. Indicates transit time.

[0033] For policy vector Policy vector and policy vector The specific details are shown in Table 2.

[0034] Table 2

[0035] Transit time The calculation formula is shown in equation (2).

[0036] (2) In equation (2), This represents the transfer distance, which can be understood as the real-time distance between the ambulance's current location and the target hospital's location during this mission. It changes dynamically depending on the vehicle's location, road selection, and detours due to congestion. Mathematically, it is the length of the following path, measured in km. This represents the baseline vehicle speed, with the congestion index ∈ [0, 1]. This indicates the actual vehicle speed.

[0037] The congestion index is a continuous coefficient at the road segment or path level, ranging from 0 to 1. It is calculated by the navigation API or vehicle GPS based on the traffic congestion index V9 and real-time vehicle speed, and is used to calculate the single-trip transfer time. The conversion relationship involved can be referred to formula (3).

[0038] Road segment congestion index = V9 / 10 or route congestion index = 1 - / (3) In equation (3), the traffic congestion index V9 is a macro-level, regional-level integer or floating-point indicator, ranging from 0 to 10, provided by the urban traffic management department, and used for global decision-making such as cross-departmental scheduling and route planning.

[0039] Penalty items The derivation process is shown in equation (4).

[0040] (4) In equation (4), This is a penalty coefficient, representing the penalty value for other risks caused by operational errors or delays. The specific data can be adjusted according to the actual situation. For example, γ = 10 (unit: time unit or cost unit) for operational errors and γ = 5 (unit: time unit or cost unit) for delays. Operational errors specifically include, but are not limited to, improper use of emergency equipment and incorrect execution of emergency measures. Delays specifically include, but are not limited to, failure to complete the emergency task within the stipulated time and delayed emergency response. Other risks specifically include, but are not limited to, emergency personnel not wearing protective equipment as required and poor management of the emergency scene. If the emergency team does not make any of the above errors or delays during the operation, the penalty item... =0.

[0041] Synergistic benefits The derivation process is shown in equation (5).

[0042] (5) In equation (5), and This was derived from statistics obtained using the aforementioned first historical dataset; This is used to measure whether the resources of hospital departments and transport units match the needs of the pre-hospital emergency care team. It is used to measure the degree of time coordination between pre-hospital emergency teams, in-hospital departments, and transport units.

[0043] Understandably, the inter-hospital distance V12 is the static Euclidean distance (or road network distance) between two fixed hospitals, used to measure the macroscopic distance of inter-hospital transfers, measured in km, with a value range of 1-50 km. It is an environmental variable. Therefore, if this task involves inter-hospital transfers, then the transfer distance... The value range fluctuates around the inter-hospital distance V12 (due to detours, congestion, and GPS errors); if the current task is pre-hospital emergency care → this hospital, then the transfer distance... It is unrelated to the distance V12 between hospitals.

[0044] Thus, in practical applications, the aforementioned three-party dynamic game model can be used to determine the pre-hospital emergency team treatment strategy, in-hospital department treatment strategy, and transport unit treatment strategy, which represent the balance of interests among the parties in the current treatment scenario. This means determining the pre-hospital emergency team treatment strategy vector, in-hospital department treatment strategy vector, and transport unit treatment strategy vector. Specifically, these vectors are defined as pre-hospital emergency team treatment strategy parameter V21, in-hospital department treatment strategy parameter V22, and transport unit treatment strategy parameter V23. This resolves conflicts of interest among the parties and ensures fair and efficient dispatching.

[0045] Step 102: Based on the pre-hospital emergency team treatment strategy, in-hospital department treatment strategy, and transport unit treatment strategy, robust scheduling of existing emergency resources based on meta-learning is performed to determine the robust scheduling strategy. The robust scheduling strategy includes the ICU beds, operating rooms, and ambulances required to complete the treatment of all patients within a preset short time period.

[0046] It should be noted that, in order to ensure the system remains robust under other disturbances such as a surge of patients or equipment failure, robust scheduling strategies can be obtained by performing meta-learning-based scheduling on existing emergency resources, based on the pre-hospital emergency team's treatment strategies, the in-hospital department's treatment strategies, and the transport unit's treatment strategies.

[0047] For example, robust scheduling based on meta-learning can be accomplished through a pre-trained meta-learning optimization model. This meta-learning optimization model can be obtained by training an initial meta-learning network model based on a second task sample set, which includes multiple task samples. The initial meta-learning network model predefines a meta-learning objective function, which means a MAML outer loop loss that quickly adapts to new emergency scenarios, ensuring that the system remains robust under disturbances such as a large influx of patients or equipment failure. This meta-learning objective function can quickly adapt to new emergency scenarios through the MAML outer loop loss, ensuring that the system remains robust under disturbances such as a large influx of patients or equipment failure.

[0048] At this point, the meta-learning objective function can be learned through inner and outer loop updates, thereby completing the training process for the initial meta-learning network model and obtaining the meta-learning optimized model.

[0049] For example, the meta-learning objective function can be shown in equation (6).

[0050] (6) In equation (6), , Indicates the learning rate of the inner loop; Indicates in the task The above pair of functions The loss, that is, in the first i Task Above, on the updated model Calculate the loss function, model It is the model after being updated via the inner loop; Indicates the first i One task sample; This represents a task sample set, containing multiple task samples. The parameters represent the initial meta-learning model. This represents the meta-learning parameters, used to update the initial model parameters. θ To quickly adapt to new tasks.

[0051] In the derivation of equation (6), the inner loop update process can be referred to as equation (7), and the outer loop update method can be referred to as equation (8).

[0052] (7) (8) In equations (7) and (8), Indicates the first j One task sample, N This represents the number of task samples required during the inner loop update process. This represents the number of task samples required during the outer loop update process. Indicates the first i Task Above, the parameters used are θ model f In the j Data samples Loss function on; loss function Used to measure the difference between model predictions and actual results; This represents the meta-learning objective function, used to optimize the meta-learning model during training so that the model can quickly adapt to new tasks; Indicates the first i For each task, the model parameters are updated after the inner loop. This indicates the gradient operation, used to calculate the gradient of the loss function with respect to the model parameters, in order to guide the updating of the model parameters.

[0053] Step 103: Execute ICU bed scheduling, operating room scheduling, and ambulance scheduling according to the robust scheduling strategy.

[0054] Understandably, for a given robust scheduling strategy, the resource scheduler can perform ICU bed scheduling, operating room scheduling, and ambulance scheduling. Specifically, the resource scheduler allocates V5 ICU beds, V8 operating rooms, and V7 ambulances in milliseconds according to the robust scheduling strategy and immediately locks the resources. The output can be a specific bed number, operating room number, and vehicle number.

[0055] For example, if the current treatment scenario is for a large-scale infectious disease outbreak, the resource scheduler can precisely issue instructions to reserve ICU beds, ensuring 10 beds are available every hour; simultaneously, it can allocate 30 beds per hour to general wards for mild cases to rationally triage patients; for the laboratory, the resource scheduler also issues expedited testing tasks, requiring 200 samples to be tested per hour, which is crucial for accurately assessing patients' conditions and formulating subsequent treatment plans. The resource scheduler will also send specific triage instructions to the fever clinic, clearly informing which severe patients should be sent to the ICU and which mild patients can be arranged in general wards, and the fever clinic will implement triage treatment for patients based on these instructions.

[0056] If the current treatment scenario involves a major traffic accident, the resource dispatcher systematically issues instructions to the top-tier hospital, requesting the reservation of 4 operating rooms and 15 ICU beds to meet the emergency treatment needs of the seriously injured; simultaneously, it reserves 10 orthopedic beds for the specialized hospital, considering the potential for numerous fractures and other orthopedic injuries. For the blood bank, the resource dispatcher also issues instructions to allocate blood products, requesting 20 units of red blood cells to address potential massive blood loss. The resource dispatcher then sends batch transfer instructions to the ambulance fleet, clearly specifying which injured should be sent to which hospital. Following these instructions, the ambulance fleet transfers 20 seriously injured patients to the top-tier hospital and 10 lightly injured patients to the specialized hospital.

[0057] The emergency resource scheduling method based on multi-stage dynamic game theory provided in this application first determines the pre-hospital emergency team treatment strategy, in-hospital department treatment strategy, and transport unit treatment strategy, representing the balance of interests among the parties, through a three-party dynamic game model. Then, it further combines these treatment strategies to perform robust scheduling of existing emergency resources based on meta-learning, determining a robust scheduling strategy. Finally, it executes ICU bed scheduling, operating room scheduling, and ambulance scheduling according to the robust scheduling strategy. In this way, the multi-party dynamic game model overcomes the limitations of static game theory, achieving an equilibrium in emergency resource allocation under multi-participant dynamic game theory. Simultaneously, the meta-learning framework enables rapid adaptation to sudden disturbances, generating robust scheduling strategies, solving the problem of insufficient real-time performance, and achieving the goal of effective allocation of emergency resources in response to emergencies. This significantly improves the adaptability of the robust scheduling strategy to emergencies, making the resource scheduling strategy interpretable and stable.

[0058] Based on the above Figure 1 In one example embodiment of the method shown, step 101 involves determining the pre-hospital emergency team treatment strategy, in-hospital department treatment strategy, and transport unit treatment strategy—representing the balance of interests among the parties—based on a three-way dynamic game model involving the pre-hospital emergency team, in-hospital departments, and transport unit, and the medical multimodal data of each patient in the current treatment scenario. The specific process in this embodiment can be achieved through… Figure 2 Steps 201 and 202 shown are implemented.

[0059] Step 201: Based on the medical multimodal data of each patient, determine the current patient state vector and the current treatment priority score for each patient.

[0060] Step 202: Input each current patient state vector, each current treatment priority score, as well as the current environmental disturbance parameters and the current emergency resource tension index into the three-party dynamic game model to solve the three-party Nash equilibrium of the pre-hospital emergency team, the in-hospital department, and the transport unit, and obtain the treatment strategies of the pre-hospital emergency team, the in-hospital department, and the transport unit.

[0061] Among them, the three-party Nash equilibrium is used to characterize the strategies chosen by each party in a dynamic game, such that when the strategies of other parties remain unchanged, the party cannot increase its benefits by unilaterally changing its strategy.

[0062] It is understandable that medical multimodal data refers to multi-dimensional medical information that includes at least physiological signals, medical history texts, and imaging data. Specifically, it can be collected using electrocardiogram monitors, electronic medical record systems, and medical imaging equipment to comprehensively reflect the patient's health status.

[0063] The current patient state vector is a vectorized representation obtained by compressing the corresponding patient's physiological signals, medical history text, and image data through multimodal fusion, which is used to quantify the current physiological state of the corresponding patient.

[0064] The current treatment priority score is a numerical score calculated based on indicators such as the severity of the illness and the fluctuation rate of vital signs. It is used to objectively assess the urgency of treatment for the corresponding patients and to determine the allocation order of emergency resources. For example, the patient priority score V17 in Table 1 above is a continuous value with a range of 0 to 1.

[0065] The current environmental disturbance parameter refers to a quantitative indicator that reflects external environmental factors such as traffic congestion and weather disasters. It is used to assess the external interference faced by resource scheduling and its impact on scheduling robustness. For example, the environmental disturbance coefficient V19 in Table 1 above is a continuous value and its value ranges from 0 to 1.

[0066] The current emergency resource shortage index is a comprehensive indicator that represents the supply and demand status of key resources such as ICU beds and ventilators. It can be realized through real-time statistical calculation by the resource monitoring system. It is used to reflect the resource load of medical institutions and to trigger adjustments to scheduling strategies. For example, the resource shortage index V18 in Table 1 above is a continuous value with a range of 0 to 1.

[0067] Specifically, by integrating multi-dimensional medical data for each patient, a current patient state vector and a current treatment priority score are generated, providing an objective decision-making basis for the three-party dynamic game model. Current environmental disturbance parameters and current resource scarcity index are used as input parameters for the three-party dynamic game model, enabling it to perceive changes in the external environment and the state of internal resources. During the three-party dynamic game, the pre-hospital emergency team, in-hospital departments, and transport units engage in strategy competition based on real-time parameter combinations. By solving for the Nash equilibrium, an equilibrium solution is obtained where none of the parties can gain additional benefits by unilaterally changing their strategies. The treatment strategy corresponding to this equilibrium solution considers both the individual medical characteristics of each patient and environmental disturbance factors and resource constraints, thus achieving a dynamic equilibrium state in the multi-party interest game.

[0068] For example, the three-party Nash equilibrium can be solved by defining a three-party treatment strategy. The Nash equilibrium condition is derived, and the specific process is shown in equation (9).

[0069] (9) In equation (9), This indicates that, given the strategies of other participants... If the parties remain unchanged, i Choose strategy σi *The profit at that time; Indicates the participating parties i The optimal strategy, and Including the participating parties i The set of strategies chosen by other participants; Indicates that the strategy of other participants is In the case of the participating parties i Choose any strategy Equation (9) is used to calculate the benefit of the participating parties when the strategies of other participating parties are determined. i The benefits that can be obtained by changing one's own strategy; the meaning of the Nash equilibrium condition is that no one in the three parties (pre-hospital-in-hospital-transfer) can increase their benefits by unilaterally changing their strategy. Its role is to ensure that the scheduling strategy is at a stable and fair equilibrium point.

[0070] The derivation of equation (9) can be achieved through policy gradient update and advantage function calculation, specifically referring to the policy gradient update of equation (10) and the advantage function calculation of equation (11).

[0071] (10) (11) In equations (10) and (11), Representation Strategy The performance metric is typically the expected return or cumulative reward of the strategy, used to measure its performance in a given state. s Next, according to the strategy Select Action The expected value of the long-term returns that can be obtained; Representation Strategy The parameters define the probability distribution for choosing a specific action in a given state; Indicates the state s and actions a The expected value is calculated based on the probability distribution of states and actions under the policy. Representation Strategy Regarding its parameters The logarithmic gradient describes how changing policy parameters affects the action selection given a state. The probability of.

[0072] Therefore, it is understandable that, compared with existing technologies, traditional methods assume that the participants are perfectly rational and have symmetrical information, failing to consider the impact of environmental disturbances and resource constraints on decision-making, resulting in strategies that are difficult to adapt to complex real-world scenarios. This application addresses the problem of strategy imbalance caused by information asymmetry in traditional methods by introducing multimodal medical data to construct a patient status assessment system and establishing a three-party dynamic game model by combining environmental parameters and resource indices. Furthermore, the Nash equilibrium solution mechanism replaces traditional subjective experience-based judgments, making the generation process of treatment strategies by all parties mathematically verifiable.

[0073] Through the above technical solutions, this application achieves a dynamic balance of interests among the three parties involved, avoids resource allocation conflicts caused by information asymmetry, significantly improves the adaptability of the treatment strategy to emergencies by integrating real-time environmental parameters and resource status data, effectively enhances the accuracy of treatment priority determination based on objective and quantitative patient assessment indicators, ensures that critically ill patients receive timely treatment, and makes the resource scheduling strategy interpretable and stable by using game theory mathematical tools.

[0074] Based on the above Figure 2 In one example embodiment of the method shown, step 201 determines the current patient state vector and current treatment priority score for each patient based on their medical multimodal data. The specific process in this embodiment can be achieved through… Figure 3 Steps 301 and 302 shown are implemented.

[0075] Step 301: Determine the current patient state vector for each patient based on their physiological signals, medical history text, and imaging data.

[0076] Step 302: Determine the current treatment priority score for each patient based on their SOFA score, vital sign variability, number of underlying diseases, and age group.

[0077] It is understood that steps 301 and 302 can be executed simultaneously or sequentially. This application does not impose specific limitations on this.

[0078] It should be noted that in the embodiments of this application, the current patient state vector of each patient can be defined as the fusion of the corresponding patient's physiological signals, medical history text and image data, and the specific fusion process can be referred to Equation (12).

[0079] (12) In equation (12), Indicates the first i The current patient state vector of each patient maps three heterogeneous data types—physiological signals, imaging data, and medical history text—into a unified patient state vector, which can serve as a common input for game-theoretic strategy networks and disease prediction networks; GNN_physio(·) represents the physiological signal graph neural network encoder, the th... i Physiological signals of a patient =( V physio , E physio ); V physio The characteristics of nodes in the graph typically include various vital signs data, such as heart rate (HR), blood pressure (BP), and blood oxygen saturation (SpO2). E physio The edge features in the graph typically characterize the temporal correlations or relationships between other physiological signals between nodes; (·) represents the image graph neural network encoder, the first... i Imaging data of one patient =( V image , E image ); V image The edge features in the graph typically characterize the temporal correlations or relationships between other physiological signals between nodes; E imag The edge features in the graph typically represent the spatial adjacency between image segmentation regions or the relationships between other image features. (·) represents the text graph neural network encoder, the first... i Medical history text of each patient =( V text , E text ); V text The node features in the diagram typically include medical terms from the medical history text, such as "dyspnea" and "shock". E text The edge features in the graph typically represent semantic relationships in text, such as the strength of association between medical terms. , and All are learnable fusion weights, and all can be linear mapping matrices. The bias vector is represented. During the training process, the three fused weights and the bias vector are used together for end-to-end learning. The training constraints in end-to-end learning specifically include weight regularization and the loss function in end-to-end training, as shown in equations (13) and (14).

[0080] Weighted regularization: (δ=0.01)(13) Loss function: (14) In equations (13) and (14), This represents the loss function associated with the patient priority scoring model. This loss function measures the difference between the patient priority score predicted by the model and the actual priority score, and is used to guide the model learning to more accurately predict the patient's treatment priority. The weight coefficients of the regularization term are used to balance the learning ability and generalization ability of the model. Regularization is a technique to prevent overfitting by adding an extra term to the loss function to penalize the complexity of the model parameters (such as the size of the parameter values). Controlling the relative importance of this penalty term, a larger value means a stronger regularization effect.

[0081] When the i When the current patient state vector of a patient is determined through a multimodal fusion network MM-CNN, the first patient's current patient state vector can be determined. i Physiological signals, imaging data, and medical history texts of each patient are input into a pre-trained multimodal fusion network MM-CNN. The graph neural network fuses them into the current patient state vector of the corresponding patient, thereby achieving the goal of converting heterogeneous high-dimensional data into computable features.

[0082] For example, when the physiological signal for each patient is specifically: HR sequence 30 points → GNN output (ℝ represents the set of real numbers, i.e., all numbers without an imaginary part; the dimension notation ℝⁿ represents an n-dimensional real vector space); the specific image data for each patient is: CT lesion in 8 regions → GNN output. Each patient's medical history text consists of: chief complaint + 12 past medical history nodes → GNN output. Based on this, the first... i Current patient state vector for each patient: The results obtained can then be directly fed into the subsequent three-party dynamic game model to complete real-time resource allocation.

[0083] For the derivation of equation (12), the physiological signals of each patient In the equation, nodes represent vital signs (such as heart rate and blood pressure), and edges represent the temporal correlation matrix, which can be referred to in equation (15).

[0084] (15) In equation (15), Indicates the first i Each time point typically corresponds to a timestamp of a sampling point in a physiological signal sequence; Indicates the first j Each time point corresponds to the timestamp of another sampling point in the physiological signal sequence. minute.

[0085] Imaging data for each patient In the diagram, nodes represent image segmentation regions (such as lesion areas in lung CT scans), and edges represent spatial adjacency matrices. Adjacency relationship.

[0086] Each patient's medical history text In the diagram, nodes represent medical terms (such as dyspnea and shock), and edges represent semantic association matrices. .

[0087] 3 fusion weights , and Through end-to-end training and learning, it meets the requirements. , Represents the identity matrix.

[0088] In addition, it should be noted that a patient priority scoring model can be pre-constructed in this application embodiment. The meaning is to compress the 4-dimensional information such as disease condition, age and comorbidity into a priority score of 0-1, which is used to determine the order in which patients enter resources such as ICU and operating room. In this patient priority scoring model, the patient priority score V17 is defined as a function of SOFA score, vital sign fluctuation, number of underlying diseases and age group, and the specific formula is shown in (16).

[0089] (16) In equation (16), SOFA score Calculated using physiological indicators , This represents the score for each physiological indicator (such as PaO2 / FiO2, platelet count, etc.). This represents the weighting coefficient corresponding to the score of each physiological indicator; vital sign variability. The number of underlying diseases is calculated using equation (17). This was obtained by directly counting the number of types of past medical history for each patient (1-5); age grouping. : Sigmoid mapping: converts linear combinations Mapped to the interval [0,1], satisfying .

[0090] (17) In equation (17), This represents the heart rate sequence for each patient, that is, each patient within a sliding time window. (Unit: minutes) within the first t The instantaneous heart rate measurement values ​​at each sampling point are measured in beats per minute (bpm). The data sources include pre-hospital monitors, ambulance multi-parameter modules, wearable sensors, or hospital bedside monitors. This represents the baseline value for each patient, which is the average heart rate of the same patient in a stable state (without strenuous exercise or drug intervention in the first 30 minutes). If there are missing values ​​in the preprocessing, linear interpolation is used. If there are abnormal values ​​of [40, 180] bpm in the preprocessing, they are clipped according to the clinical threshold and marked as abnormal. This represents the blood oxygen saturation sequence for each patient, that is, each patient within the same time window. (Unit: minutes) within the first t Instantaneous blood oxygen saturation measurements at each sampling point, in percentage (%), collected from fingertip pulse oximeters, ear clip probes, or disposable sensors; The baseline value represents the average oxygen saturation of the same patient under stable conditions. If there are missing values ​​in the pretreatment, linear interpolation is used. If there are abnormal values ​​(<70% or >100%) in the pretreatment, they are processed according to the clinical threshold.

[0091] Vital sign fluctuations Calculation example (time window) T = 5 min, sampling frequency 1 Hz, a total of 300 sampling points) as shown in Table 3.

[0092] Table 3

[0093] but .

[0094] For the volatility of vital signs It is important to note the data flow aspect. and Synchronization with timestamps is required; anomalies are marked when time drift exceeds 2 seconds. Baseline updates: The baseline is automatically recalculated 5 minutes after each patient's condition stabilizes to prevent long-term drift. Abnormal alarm conditions: If the fluctuation rate of vital signs... If three consecutive time windows are all greater than 0.5, an emergency response signal needs to be triggered to drive the three-party dynamic game model to upgrade resource priority for emergency treatment.

[0095] Compared to existing technologies, traditional methods rely solely on single vital sign parameters or subjective disease grading, failing to effectively integrate multimodal medical data. Furthermore, existing emergency triage systems typically rely on simple scoring based on nurses' on-site observations, lacking quantitative analysis of imaging results and neglecting the impact of medical history data on treatment decisions. In contrast, the embodiments of this application construct a multidimensional feature fusion model, achieving a comprehensive assessment of the patient's physiological state, medical history risk, and degree of organ damage.

[0096] Through the above technical solutions, this application addresses the problem of inaccurate assessment of individualized patient needs during emergency resource allocation. By integrating real-time monitoring data with historical medical records, it accurately quantifies the patient's organ function status; by dynamically calculating the fluctuation rate of vital signs, it promptly captures trends of disease deterioration; by statistically analyzing the number of underlying diseases, it predicts the risk of multi-system complications; and by matching differentiated treatment plans through age grouping, it ultimately forms a scientific and objective priority score, providing reliable data support for subsequent resource scheduling.

[0097] Based on the above Figure 2 In one example embodiment of the method shown, the determination process of the current environmental disturbance parameters and the current emergency resource strain index mentioned in step 202 can be achieved through... Figure 4 Steps 401 and 402 shown are implemented.

[0098] Step 401: Determine the current environmental disturbance parameters based on the current traffic congestion index, current weather disaster level, current disease transmission rate, and current distance between hospitals.

[0099] Step 402: Determine the current emergency resource shortage index based on the current number of available ICU beds, current ventilator occupancy rate, current ambulance response time, and current operating room availability.

[0100] It is understood that steps 401 and 402 can be executed simultaneously, or they can be executed sequentially. This application does not impose specific limitations on this.

[0101] The current traffic congestion index is a quantitative indicator that reflects the real-time capacity of roads. Specifically, it can be calculated using average vehicle speed and road saturation data collected by urban traffic monitoring systems. It is used to assess the potential delay risk in ambulance transfer efficiency.

[0102] The current weather disaster level refers to the disaster warning level data issued by the meteorological department. Specifically, it can be obtained through real-time warning information accessed by the meteorological monitoring platform, and is used to predict the degree of physical obstruction of emergency transport routes by extreme weather.

[0103] The current disease transmission rate refers to the statistical indicator of new cases per unit of time. It can be calculated in real time based on the infectious disease reporting system of the public health data platform and is used to quantify the spread trend of public health emergencies.

[0104] The distance between hospitals refers to the geographical distance between medical institutions in the emergency network. Specifically, it can be measured by a geographic information system to determine the optimal path distance between the emergency center and the receiving hospital, and is used to assess the time cost of multi-institutional collaborative dispatch.

[0105] The number of currently available ICU beds refers to the number of beds available in the intensive care unit in real time. Specifically, it can be obtained by collecting bed occupancy data from various departments through the hospital resource management system, which is used to reflect the supply and demand gap of critical care resources.

[0106] The current ventilator occupancy rate refers to the proportion of ventilators currently in use out of the total number of ventilators in the hospital. Specifically, it can be calculated based on real-time operational status data transmitted by IoT sensors of medical devices, and is used to assess the shortage of equipment for the treatment of critically ill patients.

[0107] The current ambulance response time refers to the average time from receiving the alarm to arriving at the scene. It can be calculated by statistically analyzing the vehicle's GPS trajectory data recorded by the emergency dispatch system, and is used to measure the operational efficiency of the pre-hospital emergency care system.

[0108] The current operating room availability parameter refers to a dynamic indicator of the real-time availability of the operating room. Specifically, it can be calculated by combining data from the surgical scheduling system and the staff duty roster to assess the real-time load of surgical treatment capacity.

[0109] It should be noted that in this embodiment, the current environmental disturbance parameters can be calculated from the current traffic congestion index, the current weather disaster level, the current disease transmission rate, and the current distance between hospitals. The values ​​are continuous and range from 0 to 1, which improves the robustness of image scheduling.

[0110] For example, the current environmental disturbance parameters can specifically refer to the environmental disturbance parameters in equation (18). V 19. The calculation formula is used to obtain the result.

[0111] (18) In equation (18), w 9. w 10. w 11 and w12 are all weighting coefficients used to adjust the degree of influence of each variable on the environmental disturbance parameters. These weights can be adjusted according to the actual situation to reflect the actual impact of different environmental factors on scheduling robustness. V 19 is the environmental disturbance parameter, which represents the degree of influence of environmental factors on scheduling robustness, and its value ranges from 0 to 1; V 9 represents the traffic congestion index, indicating the impact of current traffic conditions on ambulance speed; V A score of 10 represents the level of weather disaster, indicating the impact of current weather conditions on resource allocation. V 11 represents the disease transmission rate. R 0 indicates the impact of the current rate of epidemic spread on the demand for isolation wards; V 12 represents the distance between hospitals, indicating the impact of the distance between different hospitals on inter-hospital transfer time; w 9. w 10. w 11 and w 12 are all weighting coefficients used to adjust the degree of influence of each environmental factor on the environmental disturbance coefficient. This style (18) can assess the impact of the current environmental conditions on emergency resource scheduling, thereby improving the adaptability and robustness of the scheduling strategy.

[0112] In addition, in this embodiment, a resource tension index model can be pre-constructed, which means that the overall tension of ICU beds, ventilators, ambulances and operating rooms is quantified in real time to trigger the emergency allocation threshold and drive the game strategy update; in this resource tension index model, the resource tension index V18 is defined as a function of ICU beds, ventilators, ambulances and operating rooms, and the specific formula is shown in (19).

[0113] (19) In equation (19), This indicates the number of currently available ICU beds. Indicates the ICU bed shortage. Indicates the total number of ICU beds; This indicates the current ventilator occupancy rate; real-time monitoring values ​​(0~1) can be used directly. Indicates the current ambulance response time. For standardized processing, minute; This parameter represents the current operating room availability status; it is a binary variable (0 = occupied, 1 = idle). Indicates the percentage occupied; weighting coefficient: , , , .

[0114] Compared to existing technologies, traditional emergency resource scheduling methods typically estimate resource demand based solely on historical data, without establishing quantitative models of dynamic environmental parameters. For example, existing technologies only consider hospital distance in ambulance dispatching, while this application's embodiment additionally introduces traffic congestion index and weather disaster level parameters, enabling automatic calculation of alternative route travel time when roads are closed. Existing technologies rely on fixed schedules for operating room resource allocation, while this application's embodiment uses operating room vacancy status parameters to perceive changes in surgical treatment capacity in real time, quickly activating backup operating rooms when a sudden influx of casualties arrives. Existing technologies only count the total number of beds to assess resource scarcity, while this application's embodiment combines ventilator occupancy rate and ambulance response time to more accurately identify bottlenecks in the critical care chain.

[0115] Through the above technical solutions, the embodiments of this application effectively solve the problem of imbalanced allocation of emergency resources caused by dynamic environmental disturbances and the lack of real-time monitoring of resource status. In complex scenarios with severe road congestion and insufficient ICU beds, the system can automatically activate the inter-hospital transfer mechanism based on environmental disturbance parameters, and dynamically adjust the triage priority of critically ill patients according to ventilator occupancy rates. When an infectious disease outbreak leads to a surge in disease transmission, a zoned admission strategy can be implemented by combining inter-hospital distance parameters to avoid the risk of cross-infection while optimizing the utilization rate of medical resources. This dynamic quantitative evaluation mechanism enables emergency resource scheduling to respond in real time to changes in the internal and external environment, improving the balance and timeliness of medical resource allocation during emergencies.

[0116] Based on the above Figure 4 In one example embodiment of the method shown, the parameters mentioned in step 402—currently available ICU bed numbers, current ventilator occupancy rate, current ambulance response time, and current operating room availability—are specifically determined in this embodiment by means of… Figure 5 Steps 501 and 502 shown are implemented.

[0117] Step 501: Determine the resource needs of each patient based on the severity of their condition and clinical indicators.

[0118] Step 502: Quantify the resource needs of each patient to obtain parameters such as the number of available ICU beds, the current ventilator occupancy rate, the current ambulance response time, and the current operating room availability.

[0119] It should be noted that in this embodiment, a resource demand mapping model is pre-constructed, which means that the disease score and clinical characteristics are regressed into specific demand quantities such as ICU, operating room, blood, etc., to directly give "how many beds and how many ventilators are needed"; the regression of resource demand is defined in the resource demand mapping model, and its regression process is shown by equation (20).

[0120] (20) In equation (20), This indicates the severity of each patient's condition. Indicates clinical indicators; and All represent regression coefficients, which can be estimated using the least squares method. This indicates the resource needs of each patient.

[0121] By using formula (21), clinical indicators and the severity of each patient's condition are reduced to specific resource quantities, so that "how severe" directly corresponds to "how many beds and how many ventilators are needed".

[0122] (twenty one) In equation (21), Represents resource demand mapping, This indicates the severity of each patient's condition. This represents the clinical indicators for each patient. This indicates the resource needs of each patient.

[0123] At this point, by quantifying the resource needs of each patient, parameters such as the number of available ICU beds, the current ventilator occupancy rate, the current ambulance response time, and the current operating room availability status are obtained.

[0124] Specifically, by combining individualized patient physiological characteristics with clinical treatment standards, dynamically changing resource demand parameters are generated. For example, when a patient's condition severity reaches a preset threshold, an ICU bed demand flag is automatically triggered, and a predicted ventilator occupancy time is generated based on the respiratory failure index, a clinically validated indicator. Subsequently, based on the total resource demand of all patients, the number of available ICU beds and ventilator occupancy rates are updated in real time. Simultaneously, response time is calculated based on real-time ambulance location data, and combined with the current operating room usage status, a multi-dimensional resource status parameter set is formed. This dynamic mapping mechanism enables the resource scheduling system to continuously adjust resource allocation benchmarks according to the actual treatment needs of patients, avoiding the supply-demand imbalance problem caused by traditional methods relying on fixed resource capacity data.

[0125] Compared to existing technologies, traditional emergency resource scheduling methods typically allocate resources based on static resource capacity tables, such as using a fixed total number of ICU beds and ventilators as the scheduling basis, which cannot dynamically adjust demand assessments according to patients' conditions. However, the embodiments of this application establish a real-time correlation model between patient condition characteristics and resource needs, enabling resource status parameters to be automatically updated as patients' individualized treatment needs change. For example, when a sudden mass casualty incident leads to a large influx of critically ill patients, the system can dynamically adjust the ICU bed allocation priority based on the real-time calculated severity of the condition, while simultaneously updating ambulance response time prediction data, achieving collaborative optimization of cross-departmental resource allocation.

[0126] Through the above technical solutions, the embodiments of this application solve the problem of lack of dynamic evaluation caused by the reliance on fixed resource state space in traditional methods, and realize the real-time and accurate calculation of critical medical resource needs. Specifically, it manifests as follows: generating dynamic resource demand parameters based on individualized patient condition characteristics, enabling ICU bed and ventilator allocation to match actual treatment priorities; transforming abstract condition data into quantifiable resource occupancy parameters through clinical validation indicators, improving ventilator utilization efficiency; and integrating ambulance response time and operating room status data to ensure the spatiotemporal matching of transport resources and in-hospital treatment resources, thereby maintaining a dynamic balance in the allocation of emergency resources during emergencies.

[0127] Based on the above Figure 5 In one example embodiment of the method shown, the specific determination process for the severity of each patient's condition in step 501 can be achieved through the following steps.

[0128] Disease severity is predicted based on each patient's SOFA score, vital sign variability, number of underlying diseases, age group, and clinical indicators.

[0129] Specifically, age grouping is used to screen high-risk patient groups. For example, elderly patients may face a higher risk of complications due to weakened immunity, requiring priority allocation of intensive care resources. Gender differences are used to optimize treatment strategies for specific conditions. For instance, male patients have a higher probability of sudden cardiovascular events, necessitating adjustments to emergency response priorities. The current patient status vector dynamically reflects the trend of disease progression by integrating real-time physiological data and historical medical records. For example, abnormal fluctuations in respiratory rate combined with lung lesions shown on imaging can trigger an early warning of disease escalation. Combined with clinical indicators such as gender and age, these three factors work together to construct a multi-dimensional assessment system, avoiding misjudgments caused by relying on a single vital sign indicator, thereby improving the clinical suitability of resource allocation decisions.

[0130] In this embodiment, a disease prediction model can be pre-constructed, which means that MLP is used to map patient characteristics to disease severity, providing continuous scores for resource demand estimation and priority ranking; the disease prediction model predicts disease severity based on SOFA score, vital sign variability, number of underlying diseases and number of age groups and other clinical indicators (such as age and gender), and the specific process can be obtained through equation (22).

[0131] (twenty two) In equation (22), This indicates the severity of each patient's condition; The input characteristics for each patient include, but are not limited to, the patient's SOFA score, vital sign variability, number of underlying diseases, age group, and other clinical indicators (such as age, gender, etc.). This represents a neural network, specifically a multilayer perceptron (MLP) structure, that outputs the severity of the illness. ; This represents the noise term, used to model and predict uncertainty. ; The noise term indicates that it follows a pattern with a mean of 0 and a variance of . It follows a normal distribution.

[0132] Compared with existing technologies, traditional methods usually determine the condition based solely on vital signs or subjective experience scores, such as classifying the severity level based only on heart rate or blood pressure thresholds. However, the embodiments of this application establish a dynamic quantitative assessment model by integrating demographic characteristics and multimodal data, which not only covers individual patient differences but also enhances the real-time capture capability of disease evolution, thus solving the problem of the one-sidedness of single indicator assessment.

[0133] Through the above technical solutions, the embodiments of this application achieve accurate stratification of patients' conditions, providing an objective basis for the dynamic scheduling of emergency resources. For example, in the scenario of mass casualty treatment, high-risk patients who need to be transferred to the intensive care unit can be quickly identified, avoiding resource misallocation caused by assessment bias, thereby improving the efficiency of emergency treatment.

[0134] Based on the above Figure 1 In one example embodiment of the method shown, step 102 involves robustly scheduling existing emergency resources based on meta-learning according to the pre-hospital emergency team's treatment strategy, the in-hospital department's treatment strategy, and the transport unit's treatment strategy, thereby determining a robust scheduling strategy. The specific process in this embodiment can be achieved through… Figure 6 Steps 601 and 602 shown are implemented.

[0135] Step 601: Determine the system load rate used to determine the meta-learning update frequency based on the game strategy update frequency, data fusion delay time, and meta-learning inner loop steps.

[0136] Step 602: Input the system load rate, existing emergency resources, pre-hospital emergency team treatment strategies, in-hospital department treatment strategies, and transport unit treatment strategies into the meta-learning optimization model for robust scheduling, and obtain the robust scheduling strategy output by the meta-learning optimization model.

[0137] The system load rate is a parameter dynamically calculated based on the game strategy update frequency, data fusion latency, and the number of steps in the meta-learning inner loop. Specifically, a weighted fusion algorithm can be used to map the values ​​of these three factors to a normalized value between 0 and 1, which represents the real-time computational pressure of the current system. This parameter is used to dynamically adjust the update rhythm of the meta-learning model to avoid policy update delays due to excessive load.

[0138] For example, the system load rate can be calculated according to equation (23). V20 The calculation formula is used to obtain the result.

[0139] V 20= w CPU CPU utilization + w 内存 ⋅Memory usage+ w 网络 ⋅Network bandwidth utilization (23) In equation (23), w CPU , w 内存 and w 网络 These are all weighting coefficients, used to reflect their relative importance to the system load.

[0140] Meta-learning optimization models refer to robust scheduling models built on meta-learning frameworks. Specifically, they can employ model-independent meta-learning algorithms, using an inner loop to quickly adapt to dynamic environmental changes and an outer loop to optimize initial model parameters. This enables the model to rapidly generate stable scheduling strategies under varying load conditions. The model achieves dynamic strategy optimization by integrating multi-party game strategies and real-time resource states.

[0141] Specifically, in dynamic game theory, the frequency of strategy updates reflects the urgency of policy adjustments by each party, the data fusion latency reflects the real-time differences in data transmission between the parties, and the number of loop steps in meta-learning determines the computational intensity of model training. By integrating these three factors into the system load rate, the impact of environmental disturbances on the scheduling strategy can be perceived in real time. When the system load rate is high, the meta-learning optimization model automatically reduces the number of loop steps to lower computational overhead and shortens the policy update interval to ensure the real-time performance of the scheduling strategy; when the system load rate is low, the number of loop steps is increased to improve policy accuracy. Thus, the model can adaptively adjust the parameter update step size and exploration range according to load fluctuations, balancing computational efficiency and policy stability during resource scheduling.

[0142] For example, by inputting the pre-hospital emergency team treatment strategy parameters V21, the in-hospital department treatment strategy parameters V22, and the transport unit treatment strategy parameters V23, as well as the system load rate and the current existing emergency resources, into a pre-trained meta-learning optimization model, the system can quickly adapt to disturbances such as outbreaks of illness and equipment failures, and perform robust scheduling to output a robust scheduling strategy; thereby ensuring that the more disturbances occur, the more intelligent the system becomes.

[0143] Compared to existing technologies, traditional methods typically employ reinforcement learning models with a fixed update frequency, which cannot dynamically adjust training intensity based on system load. This leads to policy updates lagging under high load or resource waste under low load. In contrast, the embodiments of this application dynamically calculate the system load rate and couple environmental disturbances, data transmission latency, and model training intensity for analysis. This enables the meta-learning optimization model to adaptively adjust the parameter update mechanism according to real-time load conditions, thereby maintaining the accuracy and timeliness of the scheduling strategy under sudden high load scenarios.

[0144] Through the above technical solution, the embodiments of this application can perceive system load fluctuations in real time during emergency resource scheduling, dynamically adjust the training intensity and update frequency of the meta-learning model, and avoid strategy failure caused by insufficient computing resources or data transmission delays. Therefore, the emergency resource scheduling strategy can quickly adapt to the dynamic changes in public health emergencies, ensuring that the allocation schemes for ICU beds, operating rooms, and ambulances maintain high responsiveness and robustness even under high system load.

[0145] Based on the above Figure 1 In one example embodiment, the method shown can further optimize robust scheduling measurements after step 103 by analyzing the execution results. The specific process in this embodiment can be achieved through… Figure 7 Steps 701 and 702 shown are implemented.

[0146] Step 701: Determine the patient arrival status and emergency resource occupancy status based on the execution results.

[0147] Step 702: Based on the patient arrival status and emergency resource occupancy, dynamically adjust the robust scheduling strategy to obtain the adjusted robust scheduling strategy, and execute ICU bed scheduling, operating room scheduling and ambulance scheduling according to the adjusted robust scheduling strategy.

[0148] Understandably, for a given robust scheduling strategy, the resource scheduler can perform ICU bed scheduling, operating room scheduling, and ambulance scheduling. Specifically, the resource scheduler allocates ICU beds, ventilators, operating rooms, and ambulances in milliseconds according to the robust scheduling strategy and immediately locks the resources. The output can be a specific bed number, operating room number, and vehicle number. After the preset scheduling time is executed, the resource occupancy status can be fed back to available ICU beds (V5), ventilator occupancy rate (V6), ambulance response time (V7), and operating room idle status (V8) for the next round of KTI calculation.

[0149] The robust scheduling strategy is continuously optimized through a closed-loop feedback method, which includes a multi-stage MDP and a reward function. The multi-stage MDP uses Poisson + Bernoulli lotus to measure patient arrival and resource occupancy. The reward function quantifies waiting time, utilization rate, and critical illness weight into immediate rewards, driving the meta-learning optimization model to continuously update the robust scheduling strategy. The state transition is shown in equation (24), the reward function is shown in equation (25), the resource transition probability is shown in equation (26), and the immediate reward is shown in equation (27).

[0150] (twenty four) (25) (26) (27) In equations (24) to (27), Indicates the state Take action below Post-resources r The probability of resource occupancy is typically used to simulate or predict resource occupancy under specific decisions, such as in emergency resource scheduling. This might represent the probability of resource occupancy at a given time. t Afterwards, the probability of ICU beds or ambulances being occupied; Indicates the state Take action below The rate of new patient arrivals can be used to simulate or predict the rate of new patient arrivals under specific decisions, such as in emergency resource scheduling. This can represent the rate of arrival at a given time. t After that, the rate at which new patients arrive at the emergency room; , and Each represents a different weighting coefficient in the reward function, used to adjust the impact of different factors on the total reward. For example, in emergency resource scheduling, these weighting coefficients can correspond to the relative importance of reducing waiting time, improving resource utilization, and prioritizing the treatment of critically ill patients, respectively. Indicates the first i A binary variable indicating whether a patient was successfully treated; this is an indicator variable. i If successfully treated, then =1; otherwise =0. In the reward function, this binary variable reflects the treatment outcome, thus affecting the learning and optimization of the scheduling strategy; resource transfer probability. The values ​​are derived from available ICU beds (V5), ventilator occupancy rate (V6), and disease transmission rate (V11). This represents the stochastic evolution of resource allocation and patient arrival using Bernoulli+Poisson, employed to construct a multi-stage MDP to support policy gradient learning; immediate reward. The score is obtained through patient treatment priority score V17, resource tension index V18, and comprehensive scheduling efficiency score V24. Its meaning is single-step scheduling reward, that is, reducing waiting time and improving utilization rate. Prioritizing critical care is used to provide immediate feedback for reinforcement learning and drive strategy updates.

[0151] The following are specific implementation methods for three emergency scenarios. The entire process can be implemented through... Figure 8 The model structure diagram shown illustrates the implementation of the model, which combines the optimization features and parameter variables of the hospital's resource allocation, and includes data simulation and result analysis. Scenario 1: Emergency resource allocation during a large-scale infectious disease outbreak, the implementation methods of which include: Data Collection and Disease Prediction: Patient vital signs (temperature, blood oxygen, respiratory rate) are uploaded in real-time through the pre-hospital emergency system, and the SOFA score is calculated by combining this data with electronic medical record data. ) and disease fluctuation rate ( The dynamic game engine uses disease severity scores ( ). Patients are classified into severe cases. ), moderate disease ( ), mild cases ( ).

[0152] Dynamic resource allocation: 1. ICU beds: Priority will be given to critically ill patients, and the resource allocation parameters for each bed are as follows: .

[0153] 2. General wards: Patients with mild symptoms are assigned in batches, with 1 doctor and 2 nurses assigned to every 10 beds.

[0154] 3. Laboratory Department: Dynamically adjust testing priorities based on patient severity scores; testing is delayed for severely ill patients. minute.

[0155] Optimization goal: 1. Maximize the survival rate of critically ill patients (objective function: , This indicates that the critically ill patient has been successfully admitted.

[0156] 2. Minimize the resource conflict index ( ).

[0157] The data simulation and results are shown in Table 4.

[0158] Table 4

[0159] Results Analysis: Through a three-way dynamic game model, the average waiting time for critically ill patients decreased from 40 minutes to 25 minutes, and the resource conflict index decreased by 15%. Laboratory testing delays were controlled within 25 minutes, meeting the requirements. Minute goal.

[0160] Scenario 2: Multi-agency collaborative rescue in major traffic accidents, the implementation methods include: Data fusion and injury assessment: Ambulances upload vital signs (blood pressure, heart rate) and accident type (impact / fall) of injured persons in real time, and the proportion of serious injuries is predicted by the MM-GNN model. The dynamic game engine calculates the three-party strategy: the pre-hospital team prioritizes the transfer of seriously injured patients. Operating rooms reserved in hospital departments ( The transfer unit coordinates cross-hospital resources ( ).

[0161] Dynamic resource allocation: 1. Operating room: Seriously injured patients ( (Priority will be given to each operating room, with one chief surgeon and two nurses assigned to each room.)

[0162] 2. Blood products: Based on the predicted amount of blood loss from the injured (blood loss amount) Dynamic allocation of red blood cells and plasma 3. Transfer routes: based on real-time traffic index ( Optimize ambulance routes and reduce speed on congested roads. .

[0163] Optimization goal: 1. Minimize the total time for seriously injured patients to travel from the accident scene to the operating room ( ).

[0164] 2. Maximize resource utilization ( ).

[0165] The data simulation and results are shown in Table 5.

[0166] Table 5

[0167] Results analysis: The start time for surgery of seriously injured patients was reduced from 60 minutes to 45 minutes, and resource utilization increased to 88%. The dynamic game strategy reduced ambulance route conflicts by 40%, and the success rate of inter-hospital transfers remained at 95%.

[0168] Scenario 3: The emergency recovery phase after a natural disaster, the implementation methods of which include: Resource gap assessment: Calculate the medical resource gap in the disaster area through the data fusion center. Priority should be given to ensuring the supply of emergency medicines. ) and mobile ICU ( ).

[0169] Dynamic resource allocation: 1. Expert Team: Five surgeons and two critical care specialists were transferred from the regional central hospital to provide support to the disaster area in batches.

[0170] 2. Equipment Repair: The equipment repair team prioritizes repairing ventilators and monitors (failure rate...). ).

[0171] 3. Medication Allocation: Based on the predicted injury severity (fracture / trauma ratio), allocate trauma kits as needed. ) and antibiotics.

[0172] Optimization goal: 1. Minimize the recovery time of medical resources in disaster areas ( ).

[0173] 2. Maximize the survival rate of wounded soldiers (objective function: ).

[0174] The data simulation and results are shown in Table 6.

[0175] Table 6

[0176] Results analysis: By optimizing resource allocation through a dynamic game theory model, the drug shortage was reduced by 80%, and the number of mobile ICUs doubled. The survival rate of the wounded increased from 85% to 90%, and the equipment failure rate was significantly reduced.

[0177] The implementation of all three scenarios is based on the fusion of a three-party dynamic game model and real-time data, and efficient resource allocation is achieved by optimizing the following feature values ​​and parameter variables: 1. Severity of the condition : Dynamically adjust patient priority.

[0178] 2. Resource Tension Index Real-time monitoring of bed and equipment occupancy rates.

[0179] 3. Environmental disturbance coefficient : To cope with sudden disturbances such as traffic congestion and weather.

[0180] 4. System load rate Balancing computing resources with scheduling efficiency.

[0181] Through the above four configurations, the core objectives of improving the critical care rate, reducing resource conflicts, and increasing patient survival rate can be achieved.

[0182] In addition, the emergency resource allocation task covers the entire process from patient arrival at the emergency department to resource allocation, specifically including: Phase division: 1) Initial response phase (patient call received by the pre-hospital system); 2) Pre-hospital treatment phase (continuous monitoring during ambulance transfer); 3) Dynamic decision-making phase (multi-stage game calculation); 4) Resource scheduling phase (in-hospital resource coordination); 5) Implementation phase (patient transfer and admission).

[0183] Key interactive elements: 1) The data fusion center acts as an information hub, integrating multi-source data and calculating disease scores; 2) The three-party dynamic game model achieves a balance of interests among the three parties; 3) The resource scheduler queries and allocates resources in real time.

[0184] The cyclical process includes: 1) continuous monitoring and scoring updates of vital signs during ambulance transport; and 2) real-time adjustment of dynamic game strategies.

[0185] Participant instructions: 1) Patient: The party requesting emergency services; 2) Ambulance: The executor of pre-hospital emergency care; 3) Pre-hospital emergency care system: The vehicle-mounted information system; 4) Data fusion center: The multimodal data processing center; 5) Tripartite dynamic game model: The core of resource allocation decision-making; 6) Resource scheduler: The resource coordinator; 7) Hospital resource system: The hospital resource database; 8) Emergency department: The final executor of treatment.

[0186] Scenario 1: Emergency resource allocation during a large-scale infectious disease outbreak Specific scenario: A large influx of patients necessitates rapid triage and treatment; priority must be given to ICU resources, and the ratio of mild to severe cases must be dynamically adjusted; laboratory testing capacity has become a critical bottleneck.

[0187] In the emergency situation of a large-scale infectious disease outbreak, this timeline diagram presents the entire process from the influx of patients to the dynamic allocation of resources. First, a large number of patients arrive at the fever clinic at a rate of 50 per hour, creating a huge initial impact on the healthcare system. The fever clinic, as the front-end receiving point, rapidly uploads batches of patients' vital sign data, including key indicators such as body temperature and blood oxygen saturation, to the data fusion center. The data fusion center bears a heavy burden; it not only needs to process this massive amount of data quickly but also uses specific algorithms to calculate the condition score (V17) for each patient in batches. This score, based on a comprehensive analysis of multi-dimensional indicators, is crucial for subsequent resource allocation. Next, the data fusion center sends the overall data, containing information on numerous patient statuses, to the dynamic game engine. The dynamic game engine, acting as the "brain" of the entire system, considers many factors, such as the quantity of various resources and the proportion of patients with different conditions, and formulates a resource allocation strategy through complex calculations, clearly prioritizing the treatment of critically ill patients. Subsequently, it transmits the tiered treatment plan to the resource scheduler. Based on this plan, the resource scheduler precisely issues instructions to reserve critical care beds in the ICU, ensuring 10 beds are available every hour; simultaneously, it allocates 30 beds per hour to general wards for mild cases to rationally triage patients. For the laboratory, the resource scheduler also issues expedited testing tasks, requiring 200 samples to be tested per hour, which is crucial for accurately assessing patient conditions and formulating subsequent treatment plans. The resource scheduler also sends specific triage instructions to the fever clinic, clearly indicating which critically ill patients should be sent to the ICU and which mildly ill patients can be arranged in general wards. The fever clinic implements triage and treatment based on these instructions. Throughout the process, the resource scheduler continuously reports the actual resource usage to the data fusion center. The data fusion center updates the resource status information based on this feedback, and the dynamic game engine recalculates based on the latest resource status, dynamically adjusting the resource allocation strategy to adapt to the ever-changing epidemic situation and patient needs, ensuring that medical resources are used most rationally and to treat patients to the greatest extent possible.

[0188] Scenario 2: Multi-agency collaborative rescue in major traffic accidents Scenario characteristics: multi-institutional collaboration (tertiary hospitals + specialized hospitals + blood banks); dynamic calculation of optimal transport routes and hospital allocation; operating rooms and blood products become key resources.

[0189] Following a major traffic accident, the entire rescue process requires close coordination among multiple agencies. This sequence diagram clearly illustrates this complex process. Once an emergency occurs at the accident scene, a distress signal is immediately sent to the ambulance fleet, clearly stating that 30 injured individuals require rescue. Upon receiving the distress signal, the ambulance fleet acts swiftly, simultaneously reporting the number of injured and their initial condition to the 120 emergency command center. The 120 command center, acting as an information relay hub, uploads batch data on the injured to the data fusion center. The data fusion center faces a significant challenge: it needs to rapidly assess the injuries within a short timeframe, predicting, using advanced algorithms and models, that the proportion of seriously injured patients is approximately 40%. This assessment result is crucial for subsequent resource allocation. The data fusion center sends the emergency rescue request, containing detailed information on the injured, to the dynamic game engine. The dynamic game engine undertakes the core decision-making task, considering numerous factors such as the specialty capabilities of different hospitals, the availability of beds and operating rooms, etc., and derives the optimal allocation plan through complex calculations: arranging for 20 seriously injured patients to go to a tertiary hospital and 10 lightly injured patients to a specialized hospital. Subsequently, the dynamic game engine sent the inter-hospital transfer strategy to the resource scheduler. Based on this strategy, the resource scheduler systematically issued instructions to the tertiary hospital, requesting the reservation of 4 operating rooms and 15 ICU beds to meet the emergency treatment needs of the seriously injured; simultaneously, it reserved 10 orthopedic beds for the specialized hospital, considering the potential for numerous fractures and other orthopedic injuries. For the blood bank, the resource scheduler also issued instructions to allocate blood products, requesting 20 units of red blood cells to address potential massive blood loss. The resource scheduler then sent batch transfer instructions to the ambulance fleet, clearly specifying which injured should be sent to which hospital. Following these instructions, the ambulance fleet transferred 20 seriously injured patients to the tertiary hospital and 10 lightly injured patients to the specialized hospital. After the injured arrived at the hospital, the tertiary hospital and the specialized hospital reported the reception status to the data fusion center. Based on this feedback, the data fusion center updated the resource status information in a timely manner so as to further coordinate resources and adjust rescue strategies, ensuring that the entire rescue process is carried out efficiently and in an orderly manner, and maximizing the protection of the injured's lives.

[0190] Scenario 3: Emergency Recovery Phase After a Natural Disaster Objectives: Disaster area resource gap assessment → external support → equipment repair → maximizing survival rate; Scenario characteristics: The disaster area is extremely short of resources and requires external support; helicopter transportation has become a critical route; and the repair of medicines and equipment has become a priority for recovery.

[0191] In the emergency situation of a large-scale infectious disease outbreak, this timeline diagram presents the entire process from the influx of patients to the dynamic allocation of resources. First, a large number of patients arrive at the fever clinic at a rate of 50 per hour, creating a huge initial impact on the healthcare system. The fever clinic, as the front-end receiving point, rapidly uploads batches of patients' vital sign data, including key indicators such as body temperature and blood oxygen saturation, to the data fusion center. The data fusion center bears a heavy burden; it not only needs to process this massive amount of data quickly but also uses specific algorithms to calculate the condition score (V17) for each patient in batches. This score, based on a comprehensive analysis of multi-dimensional indicators, is crucial for subsequent resource allocation. Next, the data fusion center sends the overall data, containing information on numerous patient statuses, to the dynamic game engine. The dynamic game engine, acting as the "brain" of the entire system, considers many factors, such as the quantity of various resources and the proportion of patients with different conditions, and formulates a resource allocation strategy through complex calculations, clearly prioritizing the treatment of critically ill patients. Subsequently, it transmits the tiered treatment plan to the resource scheduler. Based on this plan, the resource scheduler precisely issues instructions to reserve critical care beds in the ICU, ensuring 10 beds are available every hour; simultaneously, it allocates 30 beds per hour to general wards for mild cases to rationally triage patients. For the laboratory, the resource scheduler also issues expedited testing tasks, requiring 200 samples to be tested per hour, which is crucial for accurately assessing patient conditions and formulating subsequent treatment plans. The resource scheduler also sends specific triage instructions to the fever clinic, clearly indicating which critically ill patients should be sent to the ICU and which mildly ill patients can be arranged in general wards. The fever clinic implements triage and treatment based on these instructions. Throughout the process, the resource scheduler continuously reports the actual resource usage to the data fusion center. The data fusion center updates the resource status information based on this feedback, and the dynamic game engine recalculates based on the latest resource status, dynamically adjusting the resource allocation strategy to adapt to the ever-changing epidemic situation and patient needs, ensuring that medical resources are used most rationally and to treat patients to the greatest extent possible.

[0192] It should be noted that although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flowchart can be performed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0193] In one embodiment, the present invention also provides an emergency resource scheduling device based on multi-stage dynamic game theory, such as... Figure 9As shown, the emergency resource scheduling device based on multi-stage dynamic game includes: a three-party strategy determination unit 901, a robust scheduling determination unit 902, and an emergency resource scheduling unit 903.

[0194] The tripartite strategy determination unit 901 is used to determine the pre-hospital emergency team treatment strategy, in-hospital department treatment strategy, and transport unit treatment strategy, which represent the balance of interests among the parties, based on the tripartite dynamic game model corresponding to the pre-hospital emergency team, in-hospital department, and transport unit, as well as the medical multimodal data of each patient in the current treatment scenario.

[0195] The robust scheduling determination unit 902 is used to perform robust scheduling of existing emergency resources based on meta-learning according to the pre-hospital emergency team treatment strategy, in-hospital department treatment strategy and transport unit treatment strategy, and to determine the robust scheduling strategy; the robust scheduling strategy includes the ICU beds, operating rooms and ambulances that need to be scheduled to complete the treatment of all patients within a preset short time period; Emergency resource scheduling unit 903 is used to perform ICU bed scheduling, operating room scheduling and ambulance scheduling according to a robust scheduling strategy.

[0196] Optionally, the three-party strategy determination unit 901 is used to determine the current patient state vector and current treatment priority score of each patient based on the medical multimodal data of each patient; each current patient state vector and each current treatment priority score, as well as the current environmental disturbance parameters and the current emergency resource tension index, are all input into the three-party dynamic game model to solve the three-party Nash equilibrium of the pre-hospital emergency team, the in-hospital department, and the transport unit, to obtain the treatment strategy of the pre-hospital emergency team, the treatment strategy of the in-hospital department, and the treatment strategy of the transport unit; wherein, the three-party Nash equilibrium is used to represent the strategy chosen by each party in the dynamic game, so that when the strategies of other parties remain unchanged, this party cannot increase its benefits by unilaterally changing its strategy.

[0197] Optionally, the three-party strategy determination unit 901 is used to determine the current patient status vector of each patient based on each patient's physiological signals, medical history text and imaging data; and to determine the current treatment priority score of each patient based on each patient's SOFA score, vital sign variability, number of underlying diseases and age group.

[0198] Optionally, the tripartite strategy determination unit 901 is used to determine the current environmental disturbance parameters based on the current traffic congestion index, the current weather disaster level, the current disease transmission rate, and the current distance between hospitals; and to determine the current emergency resource shortage index based on the current number of available ICU beds, the current ventilator occupancy rate, the current ambulance response time, and the current operating room vacancy status parameters.

[0199] Optionally, the three-party strategy determination unit 901 is used to determine the resource needs of each patient based on the severity of their condition and clinical indicators: quantify the resource needs of each patient to obtain parameters such as the number of available ICU beds, the current ventilator occupancy rate, the current ambulance response time, and the current operating room vacancy status.

[0200] Optionally, the three-party strategy determination unit 901 is used to predict the severity of a patient's condition based on each patient's SOFA score, vital sign variability, number of underlying diseases, age group, and clinical indicators.

[0201] Optionally, the robust scheduling determination unit 902 is used to determine the system load rate used to determine the meta-learning update frequency based on the game strategy update frequency, data fusion delay time, and meta-learning inner loop steps; the system load rate and existing emergency resources, as well as the pre-hospital emergency team treatment strategy, in-hospital department treatment strategy, and transport unit treatment strategy are input into the meta-learning optimization model for robust scheduling, so as to obtain the robust scheduling strategy output by the meta-learning optimization model.

[0202] Optionally, the emergency resource scheduling unit 903 is used to determine the patient arrival status and emergency resource occupancy status based on the execution results; and to dynamically adjust the robust scheduling strategy based on the patient arrival status and emergency resource occupancy status to obtain the adjusted robust scheduling strategy, so as to execute ICU bed scheduling, operating room scheduling and ambulance scheduling according to the adjusted robust scheduling strategy.

[0203] It should be understood that the units and references recorded in the aforementioned emergency resource scheduling device based on multi-stage dynamic game theory are related to... Figure 1 The steps in the described method correspond accordingly. Therefore, the operations and features described above for the method also apply to the data access device and the units contained therein, and will not be repeated here. The emergency resource scheduling device based on multi-stage dynamic game theory can be pre-implemented in the browser or other secure applications of a computer device, or it can be loaded into the browser or its secure applications of a computer device through download or other means. The corresponding units in the emergency resource scheduling device based on multi-stage dynamic game theory can cooperate with the units in the computer device to implement the solution of the embodiments of this application.

[0204] The following is for reference. Figure 10 It shows a schematic diagram of the structure of a computer system 1000 suitable for implementing computer devices or servers in the embodiments of this application.

[0205] like Figure 10As shown, the computer system 1000 includes a central processing unit (CPU) 1001, which can perform various appropriate actions and processes according to programs stored in read-only memory (ROM) 1002 or programs loaded from storage section 1008 into random access memory (RAM) 1003. The RAM 1003 also stores various programs and data required for the operation of the computer system 1000. The CPU 1001, ROM 1002, and RAM 1003 are interconnected via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0206] The following components are connected to the input / output (I / O) interface 1005: an input section 1006 including a keyboard, mouse, etc.; an output section 1007 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN card, modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the input / output (I / O) interface 1005 as needed. A removable medium 1011, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1010 as needed so that computer programs read from it can be installed into the storage section 1008 as needed.

[0207] Specifically, according to embodiments of this application, the above references Figure 1 The described process can be implemented as a computer software program. For example, embodiments of this application include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program containing instructions for performing... Figure 1 The program code for the method. In such an embodiment, the computer program can be downloaded and installed from a network via communication section 1009, and / or installed from removable media 1011.

[0208] It should be noted that the computer-readable medium shown in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0209] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0210] The units or modules described in the embodiments of this application can be implemented in software or hardware. The described units or modules can also be located in a processor. The names of these units or modules do not, in certain circumstances, constitute a limitation on the unit or module itself.

[0211] On the other hand, this application also provides a computer-readable storage medium, which may be included in the computer device described in the above embodiments, or may exist independently and not assembled into the computer device. The aforementioned computer-readable storage medium stores one or more programs that, when used by one or more processors, execute the methods described in this application. For example, it may execute... Figure 1 The steps of the method shown are as follows.

[0212] This application provides a computer program product including instructions that, when executed, cause the method described in this application to be performed. For example, it can execute... Figure 1 The steps of the method shown are as follows.

[0213] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0214] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. An emergency resource scheduling method based on multi-stage dynamic game, characterized in that, The method comprises: According to the three-party dynamic game model corresponding to the pre-hospital first aid team, the hospital department and the transfer unit, and the medical multi-modal data of each patient in the current treatment scene, the pre-hospital first aid team treatment strategy, the hospital department treatment strategy and the transfer unit treatment strategy representing the balance of interests of each party are determined. According to the pre-hospital first aid team treatment strategy, the hospital department treatment strategy and the transfer unit treatment strategy, the robust scheduling based on meta-learning is carried out on the existing emergency resources to determine a robust scheduling strategy; the robust scheduling strategy includes the ICU beds, operating rooms and ambulances required for scheduling to complete the treatment of all patients within a preset short time. According to the robust scheduling strategy, ICU bed scheduling, operating room scheduling and ambulance scheduling are performed.

2. The method of claim 1, wherein, According to the three-party dynamic game model corresponding to the pre-hospital first aid team, the hospital department and the transfer unit, and the medical multi-modal data of each patient in the current treatment scene, the pre-hospital first aid team treatment strategy, the hospital department treatment strategy and the transfer unit treatment strategy representing the balance of interests of each party are determined. According to the medical multi-modal data of each patient, the current patient state vector and the current treatment priority score of each patient are determined. Each current patient state vector and each current treatment priority score, as well as the current environmental disturbance parameter and the current emergency resource shortage index, are input into the three-party dynamic game model to solve the three-party Nash equilibrium of the pre-hospital first aid team, the hospital department and the transfer unit, and obtain the pre-hospital first aid team treatment strategy, the hospital department treatment strategy and the transfer unit treatment strategy. The three-party Nash equilibrium is used to represent the strategy selected by each party in the dynamic game, so that the party cannot increase its interest by changing the strategy unilaterally when the strategies of other parties remain unchanged.

3. The method of claim 2, wherein, According to the medical multi-modal data of each patient, the current patient state vector and the current treatment priority score of each patient are determined. According to the physiological signals, medical history texts and image data of each patient, the current patient state vector of each patient is determined. According to the SOFA score, vital sign fluctuation rate, number of underlying diseases and age grouping of each patient, the current treatment priority score of each patient is determined.

4. The method of claim 2, wherein, The current environmental disturbance parameter and the current emergency resource shortage index are determined. According to the current traffic congestion index, the current weather disaster level, the current disease transmission rate and the current hospital distance, the current environmental disturbance parameter is determined. According to the current available ICU bed number, the current ventilator occupancy rate, the current ambulance response time and the current operating room idle state parameter, the current emergency resource shortage index is determined.

5. The method of claim 4, wherein, The method further comprises: According to the severity of the disease and the clinical indicators of each patient, the resource demand corresponding to the patient is determined: The resource demand of each patient is quantified to obtain the current available ICU bed number, the current ventilator occupancy rate, the current ambulance response time and the current operating room idle state parameter.

6. The method of claim 5, wherein, The severity of the disease of each patient is determined. According to the SOFA score, vital sign fluctuation rate, number of underlying diseases, age grouping and clinical indicators of each patient, the severity of the patient's condition is determined.

7. The method according to any one of claims 1 to 6, characterized in that, According to the pre-hospital first aid team treatment strategy, the in-hospital department treatment strategy and the transfer unit treatment strategy, the existing emergency resources are robustly scheduled based on meta-learning to determine a robust scheduling strategy, including: Based on the game strategy update frequency, data fusion delay time and meta-learning inner loop step number, a system load rate for determining the meta-learning update frequency is determined; The system load rate and the existing emergency resources, as well as the pre-hospital first aid team treatment strategy, the in-hospital department treatment strategy and the transfer unit treatment strategy are input into a meta-learning optimization model for robust scheduling, and the robust scheduling strategy output by the meta-learning optimization model is obtained.

8. The method according to any one of claims 1 to 6, characterized in that, The method further comprises: According to the execution result, the patient arrival and the emergency resource occupation are determined; According to the patient arrival and the emergency resource occupation, the robust scheduling strategy is dynamically adjusted to obtain an adjusted robust scheduling strategy, so as to perform ICU bed scheduling, operating room scheduling and ambulance scheduling according to the adjusted robust scheduling strategy.

9. A computer device comprising a processor, a memory, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to realize the emergency resource scheduling method based on multi-stage dynamic game according to any one of claims 1 to 8.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the emergency resource scheduling method based on multi-stage dynamic game according to any one of claims 1 to 8.

Images