Method and system for predicting emergency resources by using machine learning

Abstract

The invention provides a method and a system for predicting emergency resources by using machine learning. The method comprises the following steps: acquiring and analyzing real-time emergency request information, and generating context awareness data; measuring the complex relationship between the current request and the historical cases to obtain a historical case set with high similarity, and analyzing key influence factors; weight parameters in the resource allocation priority model are dynamically adjusted, multiple scenes are simulated, and an optimized resource allocation scheme for the current emergency request is generated; by combining geographic information system data and traffic flow prediction, various scheduling scenes are simulated, an optimal path and time arrangement are selected, and a scheduling instruction for ensuring that first-aid resources arrive at the scene at the highest speed is generated; in the process of executing the scheduling instruction, the event progress and the environment change are continuously monitored, and the decision of the intelligent scheduling platform is corrected in real time according to the latest observation result. According to the technical scheme provided by the invention, the resource allocation efficiency and the response speed are improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of intelligent scheduling of emergency resources, and in particular, to a method and system for predicting emergency resources using machine learning. Background Art

[0002] Emergency resource dispatch is a key link in the emergency response system, especially in emergencies and emergencies. The rapid and accurate deployment of medical resources can significantly improve the success rate of treatment. With the acceleration of urbanization and the increase in population density, traditional emergency resource dispatch methods have been unable to meet the needs of efficiency and intelligence. Modern emergency dispatch not only needs to process a large amount of real-time emergency request information, but also needs to integrate data from different sensors to generate situational awareness data in order to have a more comprehensive understanding of the situation at the scene of the incident. In addition, the dispatch system must have highly intelligent decision-making capabilities, and be able to quickly make optimal path planning and time arrangements in a complex and changing environment to ensure that emergency resources arrive at the scene as quickly as possible.

[0003] Currently, most emergency resource dispatch systems rely on preset rules and fixed priority models to allocate resources. These systems are usually based on historical data and manual experience, and use static algorithms and limited parameter adjustments to formulate dispatch plans. For example, some systems will dispatch emergency vehicles based on the principle of closest distance, or allocate resources based on a pre-set priority order. In addition, some advanced systems have begun to introduce geographic information systems and traffic flow forecasts, but these functions mostly remain at the level of auxiliary decision-making and lack the ability of deep learning and automated optimization.

[0004] However, most existing scheduling systems are based on static rules and fixed models and cannot dynamically adapt to complex actual situations. Especially in the face of emergencies, fixed priorities and path planning may not be able to respond to changes in a timely manner, resulting in resource waste or delays. Traditional systems have limited processing capabilities for real-time data and cannot effectively integrate information from multiple sensors to generate comprehensive situational awareness data. This makes scheduling decisions often based on incomplete information, affecting the accuracy and reliability of decisions. Existing scheduling solutions lack the ability to flexibly adjust and cannot modify decisions in real time based on the latest observations. Once a scheduling instruction is issued, subsequent changes and feedback are difficult to take into account in a timely manner, reducing the flexibility and response speed of responding to emergencies.

[0005] In order to solve the above problems, a method of using machine learning to predict emergency resources has emerged. This method combines advanced technologies such as graph neural networks, reinforcement learning, and spatiotemporal graph convolutional networks to achieve intelligent and dynamic resource scheduling, significantly improving the scheduling efficiency and response speed of emergency resources. Summary of the invention

[0006] An embodiment of the present application provides a method and system for predicting first-aid resources using machine learning to solve the problems of low efficiency and slow response speed of resource allocation in the prior art.

[0007] In a first aspect, an embodiment of the present application provides a method for predicting first-aid resources using machine learning, including:

[0008] Obtain and parse real-time emergency request information, extract key metadata of the event, and at the same time integrate information from different sensors to generate situation awareness data;

[0009] Using the situation awareness data and the historical first-aid case database, a similarity evaluation system based on a graph neural network is used to measure the complex relationship between the current request and historical cases, obtain a set of historical cases with high similarity, and analyze key influencing factors from the set of historical cases;

[0010] According to the key influencing factors, dynamically adjust the weight parameters in the resource allocation priority model, and use a reinforcement learning algorithm to simulate multiple scenarios to generate an optimized resource allocation plan for the current emergency request;

[0011] Based on the optimized resource allocation plan, combined with geographic information system data and traffic flow prediction, a spatio-temporal graph convolutional network is used to simulate multiple scheduling scenarios, select the optimal path and time arrangement, and generate a scheduling instruction to ensure that the first-aid resources arrive at the scene at the fastest speed;

[0012] During the execution of the scheduling instruction, continuously monitor the progress of the event and environmental changes, and real-time correct the decision of the intelligent scheduling platform according to the latest observation results.

[0013] Optionally, using the situation awareness data and the historical first-aid case database, a similarity evaluation system based on a graph neural network is used to measure the complex relationship between the current request and historical cases, obtain a set of historical cases with high similarity, and analyze key influencing factors from the set of historical cases, including:

[0014] Based on the situation awareness data and the historical first-aid case database, a similarity evaluation system based on a graph neural network is constructed. The situation awareness data is generated by obtaining and parsing real-time emergency request information, extracting key metadata of the event, and at the same time integrating information from different sensors;

[0015] Using the similarity evaluation system, structurally process the situation awareness data and historical cases to obtain a current request feature vector and a historical case feature vector, ensuring that the two are compared in the same feature space;

[0016] According to the current request feature vector and the historical case feature vector, the graph neural network model is trained so that the graph neural network model learns to automatically capture and quantify the complex relationship between nodes, thereby obtaining a trained graph neural network model;

[0017] Based on the trained graph neural network model, the similarity scores between the current request feature vector and all historical case nodes are calculated and processed, and several historical cases with the highest similarity scores are screened out to obtain a set of historical cases with high similarity;

[0018] Based on the highly similar historical case collection, analyze the commonalities and differences therein and identify the key factors that have a significant impact on the rescue effect;

[0019] Based on the key influencing factors, information that helps optimize resource allocation is generated, providing a basis for the subsequent dynamic adjustment of weight parameters in the resource allocation priority model.

[0020] Optionally, the graph neural network model is trained according to the current request feature vector and the historical case feature vector, so that the graph neural network model learns to automatically capture and quantify the complex relationship between nodes, and a trained graph neural network model is obtained, including:

[0021] Using the current request feature vector and the historical case feature vector, a graph-structured data set is constructed, wherein each feature vector is used as a node, and the edge between the nodes represents the potential association or similarity between two cases, to obtain a graph-structured data set for training;

[0022] Based on the trained graph structure dataset, the parameters of the graph neural network model are initialized, and a loss function is defined to evaluate the difference between the predicted similarity score and the actual similarity score, thereby generating an initialized graph neural network model;

[0023] Using the initialized graph neural network model, forward propagation calculation processing is performed on the graph structure data set, and information is propagated through a multi-layer network so that the model can learn the local and global relationships between nodes and obtain a preliminary similarity evaluation result;

[0024] According to the preliminary similarity evaluation results, the weights and other parameters in the graph neural network are adjusted using the back propagation algorithm to minimize the loss function value, ensure that the similarity score output by the model is close to the actual situation, and obtain the optimized graph neural network parameters;

[0025] Based on the optimized graph neural network parameters, the graph structure data set is forward propagated and processed again, and the iterative optimization process of forward propagation and back propagation is repeated, and the model parameters are continuously updated until the model converges or reaches the preset performance indicators, thereby generating a further optimized graph neural network model;

[0026] Based on the further optimized graph neural network model, after sufficient training, a trained graph neural network model is finally obtained. The trained graph neural network model can automatically capture and quantify the complex relationships between nodes, providing support for subsequent similarity evaluation.

[0027] Optionally, dynamically adjusting the weight parameters in the resource allocation priority model according to the key influencing factors, using a reinforcement learning algorithm to simulate multiple scenarios, and generating an optimized resource allocation plan for the current emergency request, including:

[0028] Using the key influencing factors analyzed from a collection of highly similar historical cases, the key influencing factors are evaluated for importance, and an importance evaluation result reflecting the degree of influence of each factor on the resource allocation decision is obtained;

[0029] Based on the importance evaluation result, the weight parameters in the resource allocation priority model are updated to obtain an updated weight configuration reflecting the importance of each factor;

[0030] Using the updated weight configuration, the reinforcement learning environment is initialized, and a state space, an action space, and a reward function are defined, wherein the state space represents all emergency scenarios, the action space contains all feasible resource allocation decisions, and the reward function is used to quantify the effect of each decision, thereby generating an initialized reinforcement learning environment;

[0031] According to the initialized reinforcement learning environment, a deep reinforcement learning algorithm is used to simulate and train multiple scenarios. In each iteration, the agent selects an action according to the current state, observes the generated new state and immediate reward, and learns the optimal or near-optimal resource allocation plan by continuously trying different action strategies, thereby obtaining the result of simulation training;

[0032] Using the results of the simulation training, record and evaluate the resource allocation effects under different scenarios, focus on solutions that significantly improve rescue efficiency or improve patient prognosis, and consider the constraints in actual operations to obtain optimized scenario simulation results;

[0033] Based on the optimized scenario simulation results, a variety of uncertainties and dynamic change factors are comprehensively considered, and finally a set of optimized resource allocation solutions are output. The optimized resource allocation solutions reflect the weight adjustment of key influencing factors and fully reflect the best practices of resource allocation under different scenarios.

[0034] Optionally, according to the initialized reinforcement learning environment, a deep reinforcement learning algorithm is used to perform simulation training processing on multiple scenarios. In each iteration, the intelligent agent selects an action according to the current state, observes the generated new state and immediate reward, and learns the optimal or near-optimal resource allocation plan by continuously trying different action strategies, and obtains the results of the simulation training, including:

[0035] Based on the initialized reinforcement learning environment, construct a learning model of the intelligent agent, wherein the learning model enables the intelligent agent to learn the best action strategy by interacting with the environment, thereby obtaining an initial learning model;

[0036] At the beginning of each iteration, using the initial learning model, an action is selected according to the current state, wherein the current state selection is based on the agent's current strategy to ensure that new possibilities can be explored while fully utilizing known best practices to obtain the selected action;

[0037] According to the selected action, emergency resources are configured and processed, and after the selected action is executed, a new state and an immediate reward are observed, wherein the new state reflects the allocation of emergency resources after the action is executed, and the immediate reward measures the contribution of the selected action to improving rescue efficiency or improving patient prognosis, thereby generating an action execution result;

[0038] Using the results of the action execution, the learning model parameters of the agent are updated to optimize the agent's prediction ability and action strategy for future rewards, and generate an updated learning model;

[0039] In the next iteration, the updated learning model is used to select an action again according to the current state, continue to execute the selected action, observe the new state and immediate reward, and further update the learning model. This process is repeated. The agent gradually accumulates data on the effects of resource allocation in different scenarios by constantly trying different action strategies, and learns which resource allocation decisions can bring optimal or near-optimal results in different emergency scenarios.

[0040] After multiple rounds of iterative training, a set of optimized resource allocation plans are finally output based on the updated learning model. The optimized resource allocation plans reflect the learning outcomes of the intelligent agent in different scenarios, embody the optimal or near-optimal resource allocation decisions, and obtain the results of simulation training.

[0041] Optionally, based on the optimized resource allocation scheme, combined with geographic information system data and traffic flow forecast, a variety of dispatch scenarios are simulated through a spatiotemporal graph convolutional network, the optimal path and time schedule are selected, and dispatch instructions are generated to ensure that emergency resources arrive at the scene as quickly as possible, including:

[0042] Using the optimized resource allocation scheme, combined with geographic information system data and real-time traffic flow forecast information, a comprehensive scenario simulation environment is constructed, wherein the comprehensive scenario simulation environment can reflect the current location of the emergency request and its surrounding geographical and traffic conditions, thereby obtaining a comprehensive scenario simulation environment;

[0043] Based on the comprehensive scenario simulation environment, a spatiotemporal graph convolutional network is initialized, nodes and edges are defined, the nodes represent emergency resource deployment points and target locations, the edges represent connection relationships and traffic conditions between these locations, and an initialized spatiotemporal graph convolutional network is generated;

[0044] According to the resource allocation decision in the optimized resource configuration scheme, setting the input parameters of the initialized spatiotemporal graph convolutional network, including resource type, quantity, availability and estimated response time, to obtain a configured spatiotemporal graph convolutional network;

[0045] Using the configured spatiotemporal graph convolutional network, multiple scheduling scenarios are simulated. In each simulation, the agent calculates the selection and time arrangement of different paths based on the current geographic information system data and traffic flow forecast information, evaluates the resource scheduling effect under each scenario, and generates simulation results of multiple scheduling scenarios;

[0046] Based on the simulation results of the various dispatch scenarios, the effects of different dispatch scenarios are analyzed and compared, focusing on the solution that can deliver emergency resources to the destination in the shortest time, and considering the constraints in the actual situation to obtain the optimal path and time arrangement;

[0047] Using the preferred path and time schedule, a dispatch instruction is ultimately generated to ensure that emergency resources arrive at the scene as quickly as possible. The dispatch instruction reflects the impact of geographic information system data and traffic flow forecasts, and comprehensively considers the requirements of the optimized resource allocation plan.

[0048] Optionally, during the execution of the scheduling instruction, the progress of events and environmental changes are continuously monitored, and the decision of the intelligent scheduling platform is corrected in real time according to the latest observation results, including:

[0049] Using the generated dispatch instructions, start the dispatch process of emergency resources, set up and start the monitoring system, the monitoring system integrates data from multiple information sources, including real-time traffic conditions, weather conditions, live video streams, and feedback from emergency personnel to obtain initial monitoring data;

[0050] Based on the initial monitoring data, dynamically monitor the incident site and its surrounding environment, evaluate the effectiveness and feasibility of the current dispatch instructions, and obtain the latest observation results;

[0051] Using the latest observations, identify any factors that affect the arrival time or efficiency of emergency resources, and assess the impact of the factors on the existing dispatch plan to generate an impact assessment report;

[0052] According to the impact assessment report, the algorithm model built into the intelligent dispatching platform is used to make real-time adjustments to the current dispatching instructions, wherein the real-time adjustments involve re-planning the route, changing resource allocation, or adjusting the estimated arrival time, so as to ensure that emergency resources can arrive at the scene quickly and safely, and generate updated dispatching instructions;

[0053] Using the updated dispatch instructions, maintain communication with on-site emergency personnel, and promptly convey the updated dispatch instructions and related information to ensure that front-line personnel can take appropriate actions based on the latest situation and obtain confirmation feedback;

[0054] Based on the confirmation feedback, the process of monitoring, evaluating and adjusting is continuously cycled until the emergency task is completed, and each cycle is optimized based on the latest observation results to ensure that the dispatch decision is always adapted to the latest actual situation and generate the final dispatch record;

[0055] After the task is completed, the final scheduling record is used to collect and record all data in the entire scheduling process for subsequent analysis and improvement.

[0056] In a second aspect, an embodiment of the present application provides a system for predicting emergency resources using machine learning, including:

[0057] The extraction and integration module is used to obtain and parse real-time emergency request information, extract key metadata of the event, and integrate information from different sensors to generate situational awareness data;

[0058] A measurement and analysis module, which is used to measure the complex relationship between the current request and the historical cases by using the context-aware data and the historical emergency case database, based on a similarity evaluation system built on a graph neural network, to obtain a set of historical cases with high similarity, and to analyze key influencing factors from the set of historical cases;

[0059] An adjustment generation module is used to dynamically adjust the weight parameters in the resource allocation priority model according to the key influencing factors, use a reinforcement learning algorithm to simulate multiple scenarios, and generate an optimized resource allocation plan for the current emergency request;

[0060] A simulation selection module is used to simulate various dispatch scenarios based on the optimized resource allocation scheme, combined with geographic information system data and traffic flow forecasts, through a spatiotemporal graph convolutional network, select the optimal path and time schedule, and generate dispatch instructions to ensure that emergency resources arrive at the scene as quickly as possible;

[0061] The monitoring and correction module is used to continuously monitor the progress of events and environmental changes during the execution of the scheduling instructions, and to correct the decisions of the intelligent scheduling platform in real time based on the latest observation results.

[0062] In a third aspect, an embodiment of the present application provides a computing device, comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a method for predicting emergency resources using machine learning as described in any one of the first aspects.

[0063] In a fourth aspect, an embodiment of the present application provides a computer storage medium storing a computer program. When the computer program is executed by a computer, it implements a method for predicting emergency resources using machine learning as described in any one of the first aspects.

[0064] In an embodiment of the present application, real-time emergency request information is obtained and parsed, key metadata of the event is extracted, and information from different sensors is integrated to generate context-aware data; the context-aware data and the historical emergency case database are used to measure the complex relationship between the current request and the historical cases based on a similarity evaluation system constructed by a graph neural network, and a set of historical cases with high similarity is obtained, and key influencing factors are analyzed from the set of historical cases; according to the key influencing factors, the weight parameters in the resource allocation priority model are dynamically adjusted, and a reinforcement learning algorithm is used to simulate multiple scenarios to generate an optimized resource allocation plan for the current emergency request; based on the optimized resource allocation plan, in combination with geographic information system data and traffic flow forecasts, a variety of scheduling scenarios are simulated through a spatiotemporal graph convolutional network, the optimal path and time schedule are selected, and a scheduling instruction is generated to ensure that emergency resources arrive at the scene as quickly as possible; in the process of executing the scheduling instruction, the progress of the event and environmental changes are continuously monitored, and the decision of the intelligent scheduling platform is corrected in real time according to the latest observations.

[0065] The technical solution of this application has the following beneficial effects:

[0066] This application automatically parses and processes large amounts of data through machine learning models, reducing manual intervention and improving scheduling efficiency. Using graph neural networks and reinforcement learning algorithms, the similarity between current requests and historical cases can be more accurately evaluated, key influencing factors can be identified, and more reasonable resource allocation decisions can be made. Combining geographic information systems and traffic flow forecasts, a variety of scheduling scenarios are simulated through spatiotemporal graph convolutional networks to select the optimal path and schedule to ensure that emergency resources can arrive at the scene as quickly as possible. In the process of executing dispatch instructions, the progress of events and environmental changes are continuously monitored, and decisions are corrected in real time based on the latest observations, enhancing the flexibility and responsiveness to respond to emergencies. Dynamically adjust the weight parameters in the resource allocation priority model so that limited emergency resources can be used more reasonably and effectively to avoid waste of resources. The entire process is automated and intelligent, reducing human delays, significantly improving the response speed of emergency resources, and winning precious time to save lives.

[0067] Furthermore, by constructing a similarity evaluation system based on a graph neural network, the complex relationship between the current request and historical emergency cases is measured, a collection of historical cases with high similarity is screened out, and key influencing factors are identified from them. Specifically, firstly, based on the context-aware data and the historical emergency case database, a graph neural network model is constructed and trained so that it can automatically capture and quantify the complex relationship between nodes. Then, the trained model is used to calculate the similarity score between the current request and the historical case, and the historical case with the highest similarity is screened out. Next, the commonalities and differences in these historical cases are analyzed to identify the key factors that have a significant impact on the rescue effect. Finally, based on these key factors, information for optimizing resource allocation is generated to provide a basis for the subsequent dynamic adjustment of the weight parameters in the resource allocation priority model. Furthermore, a reinforcement learning algorithm is used to simulate multiple scenarios, the weight configuration is updated according to the importance evaluation results, the reinforcement learning environment is initialized, and the resource allocation plan is continuously optimized through the deep reinforcement learning algorithm, and finally a set of optimal resource allocation plans that comprehensively consider multiple uncertainties and dynamic change factors are output.

[0068] Through the above methods, the efficiency and response speed of emergency resource dispatching have been significantly improved through intelligent means. On the one hand, through the graph neural network, a large amount of historical data and real-time data are deeply analyzed to accurately identify key influencing factors, ensuring the scientificity and accuracy of decision-making; on the other hand, the reinforcement learning algorithm is used to simulate multiple scenarios and dynamically adjust the resource allocation priority to achieve the optimal allocation of resources, significantly improving the rescue efficiency and patient prognosis. The entire process is automated and intelligent, reducing human delays, enhancing flexibility and responsiveness in dealing with emergencies, and providing strong support for public safety and medical services.

[0069] These and other aspects of the present application will become more clearly understood in the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0071] Figure 1 A flowchart of a method for predicting emergency resources using machine learning provided in an embodiment of the present application;

[0072] Figure 2 A schematic diagram of the structure of a system for predicting emergency resources using machine learning provided in an embodiment of the present application;

[0073] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0074] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0075] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0076] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0077] Figure 1 A flowchart of a method for predicting emergency resources using machine learning is provided for an embodiment of the present application, such as Figure 1 As shown, the method includes:

[0078] 101. Obtain and parse real-time emergency request information, extract key metadata of events, and integrate information from different sensors to generate situational awareness data;

[0079] It involves the acquisition and analysis of real-time emergency request information, including collecting emergency call data from multiple sources and extracting key metadata from it, such as time, location, event type and severity. In addition, it also integrates information from different sensors, such as GPS positioning, camera monitoring, environmental sensors, and medical equipment monitoring data. These multi-source information together constitute situational awareness data, providing a comprehensive and accurate data foundation for subsequent intelligent decision-making.

[0080] First, the system receives emergency request information from various channels through interfaces or APIs, performs preliminary filtering and formatting to ensure data consistency and integrity. Then, natural language processing technology is used to parse the text content and extract key metadata. At the same time, the system will capture relevant data from various sensor networks and aggregate all information into a unified situational awareness data set through data fusion technology. This process not only improves the quality of the data, but also enhances the depth of understanding of the situation at the scene of the incident.

[0081] When implemented in a city emergency center, the system received a car accident alarm call, automatically parsed the time and location of the accident, obtained real-time video streams from nearby cameras, and confirmed the location of the accident vehicle in combination with GPS positioning. At the same time, weather and road condition information provided by environmental sensors was also included in the analysis. By integrating these multi-source information, the system generated detailed situational awareness data, providing a solid foundation for subsequent resource scheduling.

[0082] 102. Using the context-aware data and the historical emergency case database, a similarity evaluation system based on a graph neural network is used to measure the complex relationship between the current request and the historical cases, obtain a historical case set with high similarity, and analyze key influencing factors from the historical case set;

[0083] The core is to use graph neural networks to build a similarity assessment system, which performs structured processing on situational awareness data and information in historical emergency case databases, converting the two into feature vectors for comparison in the same feature space. The graph neural network model can capture the complex relationships between nodes, quantify these relationships, and learn to identify historical cases with high similarity through training. Ultimately, the system selects several of the most similar historical cases, further analyzes the similarities and differences among them, and identifies the key factors that have a significant impact on the rescue effect.

[0084] The system first preprocesses the context-aware data and historical case data, extracts features and converts them into feature vectors. Then, the graph neural network model is used to train these feature vectors so that they can automatically learn the complex relationships between nodes. After training, the system uses the model to calculate the similarity score between the current request feature vector and the historical case feature vector, and selects the historical cases with the highest similarity. Finally, through in-depth analysis of these historical cases, key influencing factors such as response time, traffic conditions, and resource types are identified.

[0085] Continuing with the previous embodiment, after generating the situational awareness data, the system calls the pre-trained graph neural network model to compare the current car accident case with similar cases in the historical database. The model identifies several highly similar cases and finds that in these cases, arriving at the scene quickly and allocating ambulances and fire trucks reasonably are key factors in improving the success rate of rescue. Based on this, the system generates a report containing these key factors, providing a basis for the next step of optimizing resource allocation.

[0086] 103. According to the key influencing factors, dynamically adjust the weight parameters in the resource allocation priority model, use the reinforcement learning algorithm to simulate multiple scenarios, and generate an optimized resource allocation plan for the current emergency request;

[0087] The aim is to dynamically adjust the weight parameters in the resource allocation priority model based on the key influencing factors identified in the previous step. This involves evaluating the importance of each influencing factor and updating the weight configuration of the model based on the evaluation results. Subsequently, the system initializes the reinforcement learning environment, defines the state space, action space, and reward function. Through the deep reinforcement learning algorithm, the system simulates multiple scenarios, continuously tries different action strategies, and finally learns the optimal or near-optimal resource allocation plan.

[0088] The system first evaluates the importance of key influencing factors to determine the degree of influence of each factor on resource allocation decisions. Then, based on the evaluation results, the weight parameters in the resource allocation priority model are updated to make the model better reflect the actual situation. Next, the reinforcement learning environment is initialized and the initial state and goal are set. The system uses a deep reinforcement learning algorithm. In each iteration, the agent selects an action based on the current state and observes the resulting new state and immediate reward. Through continuous trial and feedback, the system gradually optimizes the resource allocation plan until the optimal solution is found.

[0089] Continuing with the previous implementation example, the system updates the weight parameters of the resource allocation priority model based on the identified key factors. Next, the reinforcement learning environment is initialized, and the state space is set to the current accident site and its surrounding area. The action space includes dispatching emergency vehicles of different types and numbers, and the reward function is based on arrival time and rescue efficiency. After multiple simulation trainings, the system found an optimal resource allocation plan: first dispatch two ambulances and a fire truck to the scene, and arrange backup resources on standby.

[0090] 104. Based on the optimized resource allocation scheme, combined with geographic information system data and traffic flow forecast, a variety of dispatch scenarios are simulated through a spatiotemporal graph convolutional network, the optimal path and time schedule are selected, and dispatch instructions are generated to ensure that emergency resources arrive at the scene as quickly as possible;

[0091] The focus is to combine geographic information system data and traffic flow forecasts, simulate various dispatch scenarios through spatiotemporal graph convolutional networks, and select the optimal path and schedule. GIS data provides detailed geographic information, such as road networks and building distribution; traffic flow forecasts help to foresee future traffic conditions. The spatiotemporal graph convolutional network model can simulate the selection and schedule of different paths based on time and space factors, and ultimately generate dispatch instructions to ensure that emergency resources arrive at the scene as quickly as possible.

[0092] The system first combines GIS data and traffic flow forecasts to build a detailed geographic and traffic model. Then, it uses a spatiotemporal graph convolutional network to simulate multiple dispatch scenarios, considering the selection and timing of different paths. The system evaluates the feasibility of each scenario by calculating the estimated travel time and cost. Finally, it selects one or more optimal paths and generates specific dispatch instructions to ensure that emergency resources can arrive at the scene as quickly as possible.

[0093] Based on the above-mentioned optimized resource allocation scheme, the system combined GIS data and real-time traffic flow forecasts to simulate multiple possible path options. Through the spatiotemporal graph convolutional network, the system calculated the best path from the emergency center to the accident site, and selected a route that avoided peak congestion considering current and future traffic conditions. The system then generated a dispatch instruction, instructing the nearest ambulance to go to the scene along this path, and notified the traffic management departments along the way to cooperate in diverting traffic.

[0094] 105. In the process of executing the dispatch instructions, the progress of events and environmental changes are continuously monitored, and the decisions of the intelligent dispatch platform are corrected in real time according to the latest observation results.

[0095] The system emphasizes continuous monitoring of event progress and environmental changes throughout the dispatch process to ensure real-time and flexible decision-making. The system obtains the latest information through various means and corrects dispatch decisions in real time based on this information. This dynamic adjustment mechanism enables the system to respond to changes in emergencies and maintain the best response state.

[0096] While executing dispatch instructions, the system starts the real-time monitoring module to track the progress of events and environmental changes. Once new situations or changes are discovered, the system immediately analyzes their impact and adjusts the dispatch instructions based on the latest observations. For example, if there is a sudden traffic jam on the way, the system will recalculate the route and update the dispatch instructions in time to ensure that emergency resources can still arrive at the scene as soon as possible.

[0097] As the ambulance was heading to the accident site, the system discovered through GPS tracking and real-time video monitoring that the road ahead was severely congested due to another traffic accident. The system quickly recalculated the route, selected an alternative route, and immediately updated the dispatch instructions through the intelligent dispatch platform. At the same time, the system notified the traffic management departments along the way to assist in diverting traffic and ensure the smooth passage of the ambulance. In the end, the ambulance arrived at the scene on time and successfully completed the rescue mission.

[0098] Through the implementation of steps 101 to 105, this method realizes the automation and intelligent management of the whole process from data acquisition to intelligent scheduling, and significantly improves the efficiency and response speed of emergency resource scheduling. First, by comprehensively acquiring and parsing real-time emergency request information, the accuracy and completeness of the decision-making basis are ensured. Secondly, the similarity evaluation system constructed by the graph neural network is used to accurately identify key influencing factors, providing a scientific basis for resource optimization allocation. Thirdly, the weight parameters in the resource allocation priority model are dynamically adjusted, and multiple scenarios are simulated by the reinforcement learning algorithm to generate the optimal resource allocation plan, which greatly improves the flexibility and adaptability of decision-making. Then, combined with GIS data and traffic flow prediction, the optimal path and time schedule are selected through the spatiotemporal graph convolution network to ensure that resources arrive at the scene as quickly as possible. Finally, the real-time monitoring and dynamic adjustment mechanism ensures the efficiency and accuracy of the entire scheduling process and enhances the ability to respond to emergencies. In summary, this method not only improves the scientificity and accuracy of emergency resource scheduling, but also greatly enhances the overall effectiveness of the emergency response system, providing strong support for public safety and medical services.

[0099] In order to solve the problem of insufficient capture of complex relationships between nodes during the training of the graph neural network model, in some embodiments, the step 102 of training the graph neural network model according to the current request feature vector and the historical case feature vector so that the graph neural network model learns to automatically capture and quantify the complex relationships between nodes, and obtains a trained graph neural network model, including:

[0100] Utilizing the situation-aware data and the historical emergency case database, a similarity evaluation system based on a graph neural network is constructed to measure the complex relationship between the current request and the historical cases, to obtain a historical case collection with high similarity, and to analyze key influencing factors from the historical case collection, including: constructing a similarity evaluation system based on a graph neural network based on the situation-aware data and the historical emergency case database, wherein the situation-aware data is generated by acquiring and parsing real-time emergency request information, extracting key metadata of the event, and integrating information from different sensors; utilizing the similarity evaluation system to perform structured processing on the situation-aware data and the historical cases, to obtain a current request feature vector and a historical case feature vector, to ensure that the two are compared in the same feature space; according to the current The graph neural network model is trained based on the request feature vector and the historical case feature vector, so that the graph neural network model can learn to automatically capture and quantify the complex relationships between nodes, thereby obtaining a trained graph neural network model; based on the trained graph neural network model, the similarity scores between the current request feature vector and all historical case nodes are calculated and processed, and several historical cases with the highest similarity scores are screened out to obtain a set of historical cases with high similarity; based on the set of historical cases with high similarity, the commonalities and differences therein are analyzed to identify the key influencing factors that have a significant impact on the rescue effect; based on the key influencing factors, information that helps to optimize resource allocation is generated to provide a basis for the subsequent dynamic adjustment of the weight parameters in the resource allocation priority model.

[0101] The graph neural network model is trained according to the current request feature vector and the historical case feature vector, so that the graph neural network model learns to automatically capture and quantify the complex relationship between nodes to obtain a trained graph neural network model, including: using the current request feature vector and the historical case feature vector to construct a graph structured data set, wherein each feature vector is used as a node, and the edge between the nodes represents the potential association or similarity between two cases, to obtain a graph structured data set for training; based on the trained graph structured data set, the parameters of the graph neural network model are initialized, and a loss function is defined to evaluate the difference between the predicted similarity score and the actual similarity score to generate an initialized graph neural network model; using the initialized graph neural network model, forward propagation calculation processing is performed on the graph structured data set, and information is propagated through a multi-layer network, so that the model can learn the relationship between nodes. local and global relationships, and obtain preliminary similarity evaluation results; based on the preliminary similarity evaluation results, use the back propagation algorithm to adjust the weights and other parameters in the graph neural network to minimize the loss function value, ensure that the similarity score output by the model is close to the actual situation, and obtain optimized graph neural network parameters; based on the optimized graph neural network parameters, perform forward propagation calculation processing on the graph structure data set again, and repeat the iterative optimization process of forward propagation and back propagation, continuously update the model parameters, until the model converges or reaches the preset performance indicators, and generate a further optimized graph neural network model; based on the further optimized graph neural network model, after sufficient training, finally obtain a trained graph neural network model, which can automatically capture and quantify the complex relationships between nodes, and provide support for subsequent similarity evaluation.

[0102] In this embodiment, the graph structure data set includes the current request feature vector and the historical case feature vector, each feature vector is a node, and the edges between the nodes represent the potential association or similarity between the two cases. The weights of these edges can be adjusted according to the actual similarity score to more accurately reflect the relationship between the nodes. Initializing the graph neural network model refers to setting the initial parameters of the model and defining the loss function to evaluate the difference between the predicted similarity score and the actual similarity score. The choice of loss function is crucial, which determines the direction and effect of model optimization. Forward propagation refers to the process of information transmission layer by layer in the network to calculate the output results; back propagation is to adjust the model parameters according to the value of the loss function to minimize the error. This process is iterated repeatedly until the model converges or reaches the preset performance index.

[0103] In an embodiment of the present application, first, a graph-structured data set is constructed using the current request feature vector and the historical case feature vector, wherein each feature vector is used as a node, and the edge between the nodes represents the potential association or similarity between the two cases, thereby obtaining a graph-structured data set for training. Next, based on the graph-structured data set, the parameters of the graph neural network model are initialized, and a loss function is defined to evaluate the difference between the predicted similarity score and the actual similarity score, thereby generating an initialized graph neural network model. Then, using the initialized graph neural network model, a forward propagation calculation is performed on the graph-structured data set, and information is propagated through a multi-layer network, so that the model can learn the local and global relationships between nodes, and obtain preliminary similarity evaluation results. Based on these preliminary results, the weights and other parameters in the graph neural network are adjusted using a back-propagation algorithm to minimize the loss function value, ensure that the similarity score output by the model is close to the actual situation, and obtain optimized graph neural network parameters. Finally, based on the optimized graph neural network parameters, the graph structure data set is forward propagated again, and the iterative optimization process of forward propagation and back propagation is repeated, and the model parameters are continuously updated until the model converges or reaches the preset performance indicators. Finally, a trained graph neural network model is generated, which can automatically capture and quantify the complex relationships between nodes, providing support for subsequent similarity evaluation.

[0104] Here is a specific example:

[0105] Suppose in a city emergency dispatch system, the system receives a car accident alarm call. After the context-aware data generation step, the feature vector of the current request is obtained. At the same time, feature vectors of multiple historical cases are extracted from the historical emergency case database. In order to train an efficient graph neural network model, the system first constructs these feature vectors into a graph-structured dataset, where each feature vector is a node and the edges between nodes represent the potential association or similarity between two cases. For example, if two cases occur in the same place and have similar accident types, the edge weight between them is higher.

[0106] Next, the system initializes the parameters of the graph neural network model and defines a loss function to evaluate the difference between the similarity score predicted by the model and the actual situation. The system begins to perform forward propagation calculations on the graph structure dataset, propagating information through multiple layers of the network so that the model gradually learns the local and global relationships between nodes. For example, the model may find that certain types of traffic accidents are usually accompanied by higher rescue difficulties and longer response times.

[0107] Based on the preliminary similarity evaluation results, the system uses the back-propagation algorithm to adjust the weights and other parameters in the graph neural network to minimize the loss function value. This process is iterated repeatedly until the model converges or reaches the preset performance indicators. In the end, the system generates a trained graph neural network model that can accurately capture and quantify the complex relationships between nodes. When faced with new emergency requests, the system can quickly identify the most similar historical cases and analyze key influencing factors, such as traffic conditions, weather conditions, etc., to provide a scientific basis for optimal resource allocation.

[0108] By introducing a graph structure data set and an iterative optimization process, this embodiment significantly enhances the graph neural network model's ability to learn complex relationships between nodes. This not only improves the accuracy of similarity assessment, but also enables the model to more accurately identify the most similar historical cases when faced with new requests, thereby providing reliable decision support for resource optimization and allocation. This method is particularly suitable for the field of emergency resource management that requires efficient and intelligent scheduling, and can greatly improve the speed and effectiveness of emergency response.

[0109] In order to solve the problem of insufficient importance assessment of key influencing factors in resource allocation decisions and insufficient optimization of resource allocation schemes, in some embodiments, the step 103 uses a deep reinforcement learning algorithm based on the initialized reinforcement learning environment to simulate training for multiple scenarios. In each iteration, the agent selects an action based on the current state and observes the generated new state and immediate reward. By continuously trying different action strategies, the agent learns the optimal or near-optimal resource allocation scheme, and obtains the results of the simulation training, including:

[0110] According to the key influencing factors, the weight parameters in the resource allocation priority model are dynamically adjusted, and a reinforcement learning algorithm is used to simulate multiple scenarios to generate an optimized resource allocation plan for the current emergency request, including: using the key influencing factors analyzed from a highly similar historical case set, the key influencing factors are evaluated for importance, and an importance evaluation result reflecting the degree of influence of each factor on the resource allocation decision is obtained; based on the importance evaluation result, the weight parameters in the resource allocation priority model are updated to obtain an updated weight configuration reflecting the importance of each factor; using the updated weight configuration, the reinforcement learning environment is initialized, and the state space, action space and reward function are defined, wherein the state space represents all emergency scenarios, the action space contains all feasible resource allocation decisions, and the reward function is used to quantify the effect of each decision, and generate an initialized reinforcement learning environment. learning environment; according to the initialized reinforcement learning environment, using a deep reinforcement learning algorithm, simulate training processing for multiple scenarios, in each iteration, the intelligent agent selects an action according to the current state, and observes the generated new state and immediate reward, and learns the optimal or near-optimal resource allocation plan by continuously trying different action strategies, and obtains the result of simulation training; using the results of the simulation training, recording and evaluating the resource allocation effects under different scenarios, focusing on the plans that significantly improve the rescue efficiency or improve the patient's prognosis, and considering the constraints in actual operations, to obtain optimized scenario simulation results; based on the optimized scenario simulation results, comprehensively considering multiple uncertainties and dynamic change factors, finally outputting a set of optimized resource allocation plans, the optimized resource allocation plans reflect the weight adjustment of key influencing factors, and fully embody the best practices of resource allocation under different scenarios.

[0111] According to the initialized reinforcement learning environment, a deep reinforcement learning algorithm is used to simulate and train multiple scenarios. In each iteration, the agent selects an action according to the current state, observes the generated new state and immediate reward, and learns the optimal or near-optimal resource allocation plan by continuously trying different action strategies, thereby obtaining the result of the simulation training, including: constructing a learning model of the agent based on the initialized reinforcement learning environment, wherein the learning model enables the agent to learn the best action strategy by interacting with the environment, thereby obtaining an initial learning model; at the beginning of each iteration, using the initial learning model, selecting an action according to the current state, wherein the current state selection is based on the current strategy of the agent, thereby ensuring that new possibilities can be explored and known best practices can be fully utilized, thereby obtaining the selected action; configuring emergency resources according to the selected action, observing the generated new state and immediate reward after executing the selected action, wherein the new state reflects the allocation of emergency resources after the action is executed ... The immediate reward measures the contribution of the selected action to improving rescue efficiency or improving patient prognosis, and generates an action execution result; using the action execution result, the learning model parameters of the agent are updated to optimize the agent's prediction ability and action strategy for future rewards, and generate an updated learning model; in the next iteration, using the updated learning model, an action is selected again according to the current state, the selected action is continued to be executed, the new state and immediate reward are observed, and the learning model is further updated. This process is repeated, and the agent gradually accumulates data on the effects of resource allocation in different scenarios by constantly trying different action strategies, and learns which resource allocation decisions can bring optimal or near-optimal results in different emergency scenarios; after multiple rounds of iterative training, a set of optimized resource allocation plans are finally output based on the updated learning model. The optimized resource allocation plans reflect the learning outcomes of the agent in different scenarios, embody the optimal or near-optimal resource allocation decisions, and obtain the results of simulation training.

[0112] In this embodiment, the agent refers to an entity that performs decision-making tasks in a reinforcement learning environment. It gradually learns how to take the best actions by interacting with the environment. The learning model is a mathematical model used by the agent to predict future rewards and select action strategies. Initially, the learning model is based on existing data or randomly initialized. As the number of iterations increases, the model parameters are continuously updated and the performance gradually improves. The state space represents a set of all possible first aid scenarios, and each state represents a specific first aid situation, such as the location, time, weather conditions, etc. of the accident. The action space contains all feasible resource allocation decisions, such as dispatching different types of emergency vehicles, dispatching different numbers of medical staff, etc. The immediate reward is a measure of the contribution of the selected action to improving rescue efficiency or improving patient prognosis. The design of the immediate reward is crucial and directly affects the learning direction and effect of the agent.

[0113] In the embodiment of the present application, first, based on the initialized reinforcement learning environment, a learning model of the intelligent agent is constructed, so that the intelligent agent can learn the best action strategy by interacting with the environment, and obtain an initial learning model. Then, at the beginning of each iteration, the initial learning model is used to select an action according to the current state, ensuring that new possibilities can be explored and known best practices can be fully utilized to obtain the selected action. Next, according to the selected action, the emergency resources are configured and processed, and after the selected action is executed, the new state and immediate reward generated are observed, and the action execution result is generated. These results reflect the allocation of emergency resources after the action is executed and its impact on rescue efficiency. Subsequently, the learning model parameters of the intelligent agent are updated using the action execution results to optimize the intelligent agent's prediction ability and action strategy for future rewards, and an updated learning model is generated. Finally, in the next iteration, the updated learning model is used to select an action again according to the current state, continue to execute the selected action, observe the new state and immediate reward, and further update the learning model. This process is repeated, and the intelligent agent gradually accumulates data on the effect of resource allocation in different scenarios by constantly trying different action strategies, and learns which resource allocation decisions can bring optimal or near-optimal results in different emergency scenarios. After multiple rounds of iterative training, a set of optimized resource allocation plans are finally output based on the updated learning model, which reflects the optimal or near-optimal resource allocation decision and obtains the result of simulation training.

[0114] Here is a specific example:

[0115] Suppose in a city emergency dispatch system, the system has identified the key influencing factors of the accident scene, such as traffic conditions, weather conditions, accident types, etc. Based on these key influencing factors, the system initializes a reinforcement learning environment, defines the state space (all possible emergency scenarios), action space (all feasible resource allocation decisions), and sets the reward function (quantifying the effect of each decision). Then, the system builds a learning model for the agent, enabling the agent to learn the best action strategy by interacting with the environment.

[0116] In the first iteration, the agent selected a resource allocation plan based on the initial learning model, such as sending two ambulances to the scene. After executing this action, the system observed the new state (the time when the ambulance arrived at the scene, the on-site treatment situation, etc.) and the immediate reward (the degree of improvement in rescue efficiency). Based on this feedback, the system updated the agent's learning model parameters and optimized its ability to predict future rewards and action strategies.

[0117] In subsequent iterations, the agent continues to select actions based on the updated learning model, and after executing the selected action, it observes the new state and immediate reward again, and further updates the learning model. As the number of iterations increases, the agent gradually accumulates a large amount of data on the effects of resource allocation in different scenarios, and learns to make optimal or near-optimal resource allocation decisions in different emergency scenarios.

[0118] After multiple rounds of iterative training, the system finally outputted a set of optimized resource allocation plans. For example, the agent found that in similar situations, sending an ambulance and a fire truck first can control the scene faster and improve the overall rescue efficiency. This set of optimization plans not only reflects the learning results of the agent in different scenarios, but also reflects the optimal or near-optimal resource allocation decisions, providing a scientific basis for actual operations.

[0119] By introducing a deep reinforcement learning algorithm, this embodiment significantly enhances the accuracy and effectiveness of resource allocation decisions. By continuously interacting with the environment, the agent gradually learns which resource allocation decisions can bring optimal or near-optimal results in different emergency scenarios. This method is particularly suitable for the field of emergency resource management that requires efficient and intelligent scheduling. It can greatly improve the speed and effectiveness of emergency response and ensure that the most reasonable resource allocation decisions are made in a complex and changing environment.

[0120] In order to solve the problem of insufficient optimization of path selection and time arrangement in the resource scheduling process, in some embodiments, the optimized resource allocation scheme in step 104 is combined with geographic information system data and traffic flow prediction, and multiple scheduling scenarios are simulated through a spatiotemporal graph convolutional network to select the optimal path and time arrangement, and generate a scheduling instruction to ensure that emergency resources arrive at the scene as quickly as possible, including:

[0121] Based on the optimized resource allocation scheme, combined with geographic information system data and traffic flow forecast, a variety of scheduling scenarios are simulated through a spatiotemporal graph convolutional network, the optimal path and time schedule are selected, and a scheduling instruction is generated to ensure that emergency resources arrive at the scene as quickly as possible, including: using the optimized resource allocation scheme, combined with geographic information system data and real-time traffic flow forecast information, to build a comprehensive scenario simulation environment, the comprehensive scenario simulation environment can reflect the current location of the emergency request and its surrounding geographical and traffic conditions, and obtain a comprehensive scenario simulation environment; based on the comprehensive scenario simulation environment, initialize the spatiotemporal graph convolutional network, define nodes and edges, the nodes represent emergency resource deployment points and target locations, the edges represent the connection relationship and traffic conditions between these locations, and generate an initialized spatiotemporal graph convolutional network; according to the resource allocation decision in the optimized resource allocation scheme, set the input parameters of the initialized spatiotemporal graph convolutional network data, including resource type, quantity, availability and expected response time, to obtain a configured spatiotemporal graph convolutional network; using the configured spatiotemporal graph convolutional network, simulate and process multiple scheduling scenarios, in each simulation, the agent calculates the selection and time schedule of different paths according to the current geographic information system data and traffic flow forecast information, evaluates the resource scheduling effect under each scenario, and generates simulation results of multiple scheduling scenarios; based on the simulation results of the multiple scheduling scenarios, analyze and compare the effects of different scheduling scenarios, pay attention to the scheme that can deliver emergency resources to the destination in the shortest time, and consider the constraints in the actual situation to obtain the preferred path and time schedule; using the preferred path and time schedule, finally generate a scheduling instruction that ensures that the emergency resources arrive at the scene as quickly as possible, the scheduling instruction reflects the influence of the geographic information system data and traffic flow forecast, and comprehensively considers the requirements of the optimized resource allocation plan.

[0122] In this embodiment, the comprehensive scenario simulation environment includes geographic information system data and real-time traffic flow forecast information to reflect the current emergency request location and its surrounding geographical and traffic conditions. These data provide detailed geographic information and dynamic traffic conditions to help the model simulate the real world more accurately. The spatiotemporal graph convolutional network is a deep learning model that can simultaneously process data in spatial and temporal dimensions and capture complex relationships between nodes. In the network, nodes represent emergency resource deployment points and target locations, and edges represent the connection relationship and traffic conditions between these locations. Input parameters refer to resource allocation decisions in the optimal resource allocation scheme, including resource type, quantity, availability, and expected response time. These parameters are configured in the spatiotemporal graph convolutional network as input to the model to simulate different scheduling scenarios. The simulation results of the scheduling scenario are the results generated after each simulation, including the selection and time schedule of different paths, as well as the resource scheduling effect under each scenario. These results are used for analysis and comparison to select the optimal path and time schedule.

[0123] In the embodiment of the present application, first, the optimized resource allocation scheme is used, combined with geographic information system data and real-time traffic flow forecast information, to construct a comprehensive scenario simulation environment, which can reflect the current location of the emergency request and its surrounding geographical and traffic conditions, and obtain a comprehensive scenario simulation environment. Then, based on the comprehensive scenario simulation environment, the spatiotemporal graph convolutional network is initialized, nodes and edges are defined, where nodes represent emergency resource deployment points and target locations, and edges represent the connection relationship and traffic conditions between these locations, and an initialized spatiotemporal graph convolutional network is generated. Next, according to the resource allocation decision in the optimized resource allocation scheme, the input parameters of the initialized spatiotemporal graph convolutional network are set, including resource type, quantity, availability and expected response time, to obtain a configured spatiotemporal graph convolutional network. Subsequently, the configured spatiotemporal graph convolutional network is used to simulate a variety of scheduling scenarios. In each simulation, the agent calculates the selection and time arrangement of different paths based on the current geographic information system data and traffic flow forecast information, evaluates the resource scheduling effect under each scenario, and generates simulation results of a variety of scheduling scenarios. Finally, based on the simulation results of the various dispatch scenarios, the effects of different dispatch scenarios are analyzed and compared, focusing on the plan that can deliver emergency resources to the destination in the shortest time, and considering the constraints in the actual situation, obtaining the optimal path and time schedule, and finally generating a dispatch instruction to ensure that emergency resources arrive at the scene as quickly as possible. The instruction reflects the influence of geographic information system data and traffic flow forecasts, and comprehensively considers the requirements of the optimal resource allocation plan.

[0124] Here is a specific example:

[0125] Suppose in a city emergency dispatch system, the system has generated an optimized resource allocation plan and decided to send two ambulances to the accident scene. In order to determine the optimal route and schedule, the system first combines geographic information system data and real-time traffic flow forecast information to build a comprehensive scenario simulation environment, which reflects the road network, building distribution and current traffic conditions in and around the accident site in detail.

[0126] Next, the system initializes the spatiotemporal graph convolutional network and defines nodes (such as emergency centers and accident sites) and edges (such as traffic conditions on different road sections). Then, based on the resource allocation decisions in the optimal resource allocation plan (such as the number, availability, and expected response time of two ambulances), the input parameters of the spatiotemporal graph convolutional network are set to generate a configured spatiotemporal graph convolutional network.

[0127] In each simulation, the agent calculates the choice and timing of different routes based on current GIS data and traffic forecast information. For example, one route may pass through a main road but encounter congestion during peak hours, while another route may be farther but have better road conditions. The system evaluates the resource scheduling effect under each scenario and records the time consumption and rescue efficiency of different routes.

[0128] After multiple simulations, the system analyzed and compared the effects of different dispatch scenarios, and ultimately selected the optimal route and schedule that could deliver emergency resources to the destination in the shortest possible time. For example, the system found that choosing an alternative route that avoids peak congestion could reduce response time by about 15 minutes. Ultimately, the system generated a set of dispatch instructions to ensure that emergency resources arrive at the scene as quickly as possible. The instructions not only took into account the impact of geographic information system data and traffic flow forecasts, but also fully reflected the requirements of optimizing resource allocation plans to ensure that ambulances can arrive at the accident site on time and successfully complete the rescue mission.

[0129] By introducing a spatiotemporal graph convolutional network and a comprehensive scenario simulation environment, this embodiment significantly enhances the path selection and time scheduling optimization capabilities in the resource scheduling process. This method is particularly suitable for the field of emergency resource management that requires efficient and intelligent scheduling. It can greatly improve the speed and effectiveness of emergency response and ensure that the most reasonable resource allocation decisions are made in a complex and changing environment. By interacting with the environment, the agent gradually learns which paths and time arrangements can bring optimal or near-optimal results in different emergency scenarios, providing a scientific basis for actual operations.

[0130] In order to solve the problem of untimely response to event progress and environmental changes during the execution of the scheduling instruction, in some embodiments, in the process of executing the scheduling instruction in step 105, the event progress and environmental changes are continuously monitored, and the decision of the intelligent scheduling platform is corrected in real time according to the latest observation results, including:

[0131] In the process of executing the dispatch instruction, the progress of the event and environmental changes are continuously monitored, and the decision of the intelligent dispatch platform is corrected in real time according to the latest observation results, including: using the generated dispatch instruction to start the dispatch process of emergency resources, setting up and starting the monitoring system, the monitoring system integrates data from multiple information sources, including real-time traffic conditions, weather conditions, on-site video streams, and feedback from emergency personnel to obtain initial monitoring data; based on the initial monitoring data, the event site and its surrounding environment are dynamically monitored to evaluate the effectiveness and feasibility of the current dispatch instruction to obtain the latest observation results; using the latest observation results, any factors that affect the arrival time or efficiency of emergency resources are identified, and the impact of the factors on the existing dispatch plan is evaluated to generate an impact assessment report; according to the impact assessment report, the built-in algorithm of the intelligent dispatch platform is used to generate an impact assessment report; The method model is used to adjust the current dispatch instruction in real time, and the real-time adjustment involves re-planning the route, changing the resource allocation or adjusting the estimated arrival time, so as to ensure that the emergency resources can arrive at the scene quickly and safely, and generate an updated dispatch instruction; using the updated dispatch instruction, maintain communication with the on-site emergency personnel, and promptly convey the updated dispatch instruction and related information to ensure that the front-line staff can take appropriate actions according to the latest situation and obtain confirmation feedback; based on the confirmation feedback, continuously cycle the monitoring, evaluation and adjustment process until the emergency task is completed, and each cycle is optimized based on the latest observation results to ensure that the dispatch decision is always adapted to the latest actual situation and generate a final dispatch record; after the task is completed, use the final dispatch record to collect and record all data in the entire dispatch process for subsequent analysis and improvement.

[0132] In this embodiment, the monitoring system integrates data from multiple information sources, including real-time traffic conditions, weather conditions, on-site video streams, and feedback from emergency personnel, for dynamic monitoring of the incident site and its surrounding environment. The initial monitoring data refers to the data collected by the monitoring system for the first time, which is used to evaluate the effectiveness and feasibility of the current dispatch instructions. The latest observations are based on the initial monitoring data and subsequent dynamic monitoring. The system continuously updates the latest information, reflecting the changes in the incident site and its surrounding environment. The impact assessment report is to identify the factors that affect the arrival time or efficiency of emergency resources through analysis of the latest observations, and evaluate their impact on the existing dispatch plan. Real-time adjustment is to adjust the current dispatch instructions based on the impact assessment report using the algorithm model built into the intelligent dispatch platform, involving operations such as re-planning the path, changing resource allocation, or adjusting the estimated arrival time. Confirmation feedback is to maintain communication with on-site emergency personnel, promptly convey updated dispatch instructions and related information, ensure that front-line staff can take appropriate actions based on the latest situation, and obtain their confirmation feedback. The final dispatch record is to collect and record all data in the entire dispatch process after the task is completed for subsequent analysis and improvement.

[0133] In the embodiment of the present application, first, the dispatching process of emergency resources is started using the generated dispatching instructions, and the monitoring system is set and started. The monitoring system integrates data from multiple information sources, including real-time traffic conditions, weather conditions, on-site video streams, and feedback from emergency personnel to obtain initial monitoring data. Then, based on the initial monitoring data, the event site and its surrounding environment are dynamically monitored to evaluate the effectiveness and feasibility of the current dispatching instructions and obtain the latest observation results. Next, using the latest observation results, any factors that affect the arrival time or efficiency of emergency resources are identified, and the impact of the factors on the existing dispatching plan is evaluated to generate an impact assessment report. Subsequently, according to the impact assessment report, the algorithm model built into the intelligent dispatching platform is used to adjust the current dispatching instructions in real time, involving re-planning the path, changing resource allocation or adjusting the estimated arrival time, ensuring that emergency resources can arrive at the scene quickly and safely, and generating updated dispatching instructions. Then, using the updated dispatching instructions, maintain communication with the on-site emergency personnel, and promptly convey the updated dispatching instructions and related information to ensure that front-line staff can take appropriate actions based on the latest situation and obtain confirmation feedback. Finally, based on the confirmation feedback, the monitoring, evaluation and adjustment process is continuously cycled until the emergency mission is completed. Each cycle is optimized based on the latest observations to ensure that the dispatch decision is always adapted to the latest actual situation and generate the final dispatch record. After the mission is completed, the final dispatch record is used to collect and record all data in the entire dispatch process for subsequent analysis and improvement.

[0134] Here is a specific example:

[0135] Suppose in a city emergency dispatch system, the system has generated a set of dispatch instructions and decided to send two ambulances to the scene of a car accident. In order to ensure the effectiveness and feasibility of the dispatch instructions, the system starts a monitoring system that integrates data from multiple information sources, including:

[0136] Real-time traffic conditions: Understand road conditions through real-time traffic data provided by traffic management departments.

[0137] Weather conditions: temperature, humidity, wind speed and other data obtained from the weather station to assess the impact of weather on rescue.

[0138] On-site video streaming: Real-time video obtained by cameras installed near the scene can provide an intuitive understanding of the situation at the accident scene.

[0139] First Responder Feedback: Receive instant feedback from first responders on the scene via radio communications or mobile apps.

[0140] The data collected by the monitoring system for the first time (initial monitoring data) showed that a main road leading to the accident site was severely congested due to another traffic accident. Based on this information, the system immediately identified that this factor could significantly delay the arrival of the ambulance and generated an impact assessment report. According to the report, the intelligent dispatch platform replanned an alternative route to avoid congested sections and notified traffic management departments along the way to assist in diverting traffic. At the same time, the system adjusted the estimated arrival time and communicated the updated dispatch instructions to the on-site emergency personnel to ensure that they can respond according to the latest situation.

[0141] Throughout the emergency mission, the system continuously monitors the progress of events and environmental changes, such as discovering that the weather suddenly deteriorates, which may cause the road surface to be slippery and increase driving risks. The system once again makes real-time adjustments based on the latest observations, reminding drivers to pay attention to safety and adjusting the allocation of some resources. After each adjustment, the system communicates with the on-site emergency personnel to ensure that they receive and understand the new instructions and obtain confirmation feedback. After the mission is completed, the system generates a final dispatch record, which records all the data in the entire dispatch process in detail for subsequent analysis and improvement, so as to make better decisions in similar situations in the future.

[0142] By introducing dynamic monitoring and real-time adjustment mechanisms, this embodiment significantly enhances the real-time performance and accuracy of the dispatch instruction execution process. This method is particularly suitable for the field of emergency resource management that requires efficient and intelligent dispatch. It can greatly improve the speed and effectiveness of emergency response and ensure that the most reasonable resource allocation decisions are made in a complex and changing environment. The intelligent dispatch platform constantly monitors, evaluates and adjusts to keep dispatch decisions adapted to the latest actual situation, providing a scientific basis for actual operations, and continuously improves the performance and reliability of the system by collecting and analyzing dispatch records.

[0143] This application considers that in the emergency resource scheduling system, accurately identifying and quantifying the similarities between current requests and historical cases is a key step in achieving intelligent scheduling. Traditional rule-based methods have difficulty handling complex nonlinear relationships, while graph neural networks, as a powerful deep learning tool, can capture complex relationships between nodes and provide a new solution for similarity assessment. However, in order to ensure the effectiveness and robustness of the model, it is necessary to design appropriate loss functions and similarity calculation methods in order to optimize model parameters during training and accurately measure similarity in practical applications.

[0144] To this end, the R&D team developed a similarity assessment system based on graph neural networks. The system can not only automatically capture and quantify the complex relationships between nodes, but also introduce a weighted exponential average method to comprehensively consider the importance of different factors, so as to more accurately identify key influencing factors and guide resource allocation decisions. Therefore, a new optional solution is proposed, which includes:

[0145] Using the situational awareness data and the historical emergency case database, a similarity evaluation system built based on a graph neural network is used to measure the complex relationship between the current request and historical cases, obtain a historical case set with high similarity, and analyze key influencing factors from the historical case set, including:

[0146] Based on the context-aware data and the historical emergency case database, a similarity evaluation system based on graph neural network is built. The context-aware data is generated by acquiring and parsing real-time emergency request information, extracting key metadata of the event, and integrating information from different sensors.

[0147] Using the similarity evaluation system, the context-aware data and historical cases are structured to obtain the current request feature vector f current and the historical case feature vector f historical To ensure that the two are compared in the same feature space, the feature vector f is expressed as:

[0148] f=[f 1 , f 2 , …, fn ]

[0149] Among them, f i Represents the value on the i-th feature dimension, n is the number of dimensions of the feature vector, i is the index variable, and represents the i-th historical case or data sample;

[0150] According to the current request feature vector f current and the historical case feature vector f historical , training the graph neural network model so that the graph neural network model learns to automatically capture and quantify the complex relationships between nodes, thereby obtaining a trained graph neural network model;

[0151] The loss function L used in the training process is defined by the following formula:

[0152]

[0153] Where N is the number of samples, y i is the true similarity score, is the predicted similarity score, w i is the weight of each sample, which is dynamically adjusted according to the importance or confidence of the sample;

[0154] Based on the trained graph neural network model, the current request feature vector f current Calculate the similarity scores between all historical case nodes, select several historical cases with the highest similarity scores, and obtain a set of historical cases with high similarity;

[0155] The similarity score similarity(f current , f historical ):

[0156]

[0157] Among them, W is a diagonal matrix, and its diagonal elements represent the weights of each feature dimension, ∥·∥ W Represents the weighted modulus, defined as f current Represents the current request feature vector; f historical Represents the current request feature vector;

[0158] Based on the highly similar historical case collection, analyze the commonalities and differences therein and identify the key factors that have a significant impact on the rescue effect;

[0159] Based on the key influencing factors, information that helps optimize resource allocation is generated to provide a basis for subsequent dynamic adjustment of weight parameters in the resource allocation priority model;

[0160]

[0161] Among them, w i is the impact weight of each case, g i is the value of the key influencing factor, α is a positive scaling factor used to adjust the impact of the index; Weighted_Exponential_Average is a weighted exponential average method.

[0162] The following is a detailed explanation of each parameter:

[0163] f: is an n-dimensional vector, each element f i Represents the value of the i-th feature dimension.

[0164] n ： is the dimension of the feature vector.

[0165] i: represents the i-th historical case or data sample.

[0166] f current : The feature vector representing the emergency request currently received. It reflects the specific circumstances of the current event and is used for comparison with historical cases.

[0167] f historical : Represents the feature vector of each historical case extracted from the historical emergency case database. It is used to compare with the current request feature vector to find similar historical cases.

[0168] N is the number of samples and defines the size of the training set.

[0169] y i : is the true similarity score, usually obtained by expert annotation or other means.

[0170] is the predicted similarity score, predicted by the graph neural network model.

[0171] w i : The weight of each sample is dynamically adjusted according to the importance or confidence of the sample.

[0172] W: Diagonal matrix, whose diagonal elements represent the weights of each feature dimension, which is used to reflect the importance of different features.

[0173] ||·|| W Represents the weighted modulus, defined as Consider the weights of feature dimensions.

[0174] w i : The impact weight of each case.

[0175] g i : is the value of the key influencing factor.

[0176] α: is a positive scaling factor used to adjust the influence of the exponent.

[0177] Weighted-Exponential-Average: is a weighted exponential averaging method.

[0178] The use of the above parameters and formulas is intended to build an efficient similarity assessment system, which automatically captures and quantifies the complex relationships between nodes through the graph neural network model, and then selects historical cases with the highest similarity scores, analyzes the key influencing factors, and ultimately provides a scientific basis for resource allocation decisions. This method not only improves the accuracy of similarity assessment, but also enhances the intelligence level of the system, and can make the most reasonable resource allocation decisions in a complex and changing environment.

[0179] Here is a specific example:

[0180] Assume that in a city emergency dispatch system, the system receives a car accident alarm call. After the context-aware data generation step, the feature vector f of the current request is obtained. ourrent At the same time, the feature vectors f of multiple historical cases were extracted from the historical emergency case database. historical In order to train an efficient graph neural network model, the system constructs these feature vectors into a graph-structured dataset, where each feature vector serves as a node and the edges between nodes represent the potential association or similarity between two cases.

[0181] The loss function L used in the training process is defined by the following formula:

[0182]

[0183] Where N is the number of samples, y i is the true similarity score, is the predicted similarity score, w i is the weight of each sample, which is dynamically adjusted according to the importance or confidence of the sample;

[0184] The similarity score similarity(f current , f historical ):

[0185]

[0186] Among them, W is a diagonal matrix whose diagonal elements represent the weights of each feature dimension, ||·|| WRepresents the weighted modulus, defined as f current Represents the current request feature vector; f historical Represents the current request feature vector;

[0187] The weighted exponential average Weighted_Exponential_Average is calculated using the following formula:

[0188]

[0189] Among them, w i is the impact weight of each case, g i is the value of the key influencing factor, α is a positive scaling factor used to adjust the impact of the index; Weighted_Exponential_Average is a weighted exponential average method.

[0190] Assume the following specific values:

[0191] Eigenvector: f current =[1, 0.8, 0.6, 0.4, 0.2]; f historical =[0.9, 0.7, 0.5, 0.3, 0.1]

[0192] Weight matrix W: diagonal matrix, diagonal elements are [1, 1, 1, 1]

[0193] Scaling factor α: set to 1

[0194] Sample weight w i :Assume uniform distribution, that is, all w i =1;

[0195] First, calculate

[0196]

[0197] Next, calculate the weighted modulus || f current || W and ||f historical || W :

[0198]

[0199] Finally calculate the similarity score:

[0200]

[0201] Calculate weighted exponential average

[0202] Assume that all w i=1, and the value of the key influencing factor g i are [0.9, 0.7, 0.5, 0.3, 0.1] respectively, then:

[0203]

[0204] The calculation results show that the similarity score is about 0.989, indicating that the current request has a very high similarity with the selected historical case. This shows that the two have a high degree of match in multiple feature dimensions and can be used as reference cases for further analysis.

[0205] The calculation results show that the average weighted index is about 0.631, which reflects the comprehensive importance of each key influencing factor. A higher value means that some factors have a more significant impact on the rescue effect, which helps to optimize resource allocation decisions.

[0206] By introducing a graph neural network model and a specific loss function, similarity calculation method, and weighted exponential average method, this embodiment significantly enhances the intelligence level of the emergency resource scheduling system. The system can not only accurately identify and quantify the similarities between the current request and historical cases, but also comprehensively consider the importance of different factors and provide a scientific basis for optimizing resource allocation. This method is particularly suitable for the field of emergency resource management that requires efficient and intelligent scheduling. It can greatly improve the speed and effectiveness of emergency response and ensure that the most reasonable resource allocation decisions are made in a complex and changing environment.

[0207] This application takes into account that in the emergency resource dispatch system, it is crucial to ensure that emergency resources can arrive at the scene as quickly as possible. However, traditional dispatch methods find it difficult to fully consider the complex interactions of geographic information system data, real-time traffic flow forecast information, and multiple dispatch scenarios. In order to meet this challenge, the R&D team introduced a spatiotemporal graph convolutional network for scenario simulation and selected the optimal path and schedule through a series of optimization formulas. This method not only improves the accuracy and timeliness of dispatch instructions, but also makes the best decision in a complex and changing environment. Therefore, a new optional solution is proposed, which includes:

[0208] Based on the above-mentioned optimized resource allocation scheme, combined with geographic information system data and traffic flow forecast, a variety of dispatch scenarios are simulated through the spatiotemporal graph convolutional network, the optimal path and time schedule are selected, and dispatch instructions are generated to ensure that emergency resources arrive at the scene as quickly as possible, including:

[0209] Using the optimized resource allocation scheme, combined with geographic information system data and real-time traffic flow forecast information, a comprehensive scenario simulation environment is constructed, wherein the scenario simulation environment can reflect the current geographic and traffic conditions of the location of the emergency request and its surroundings, and obtain a comprehensive scenario simulation environment;

[0210] Based on the comprehensive scenario simulation environment, the spatiotemporal graph convolutional network is initialized, nodes and edges are defined, the nodes represent emergency resource deployment points and target locations, the edges represent the connection relationship and traffic conditions between these locations, and the initialized spatiotemporal graph convolutional network is generated;

[0211] According to the resource allocation decision in the optimized resource configuration scheme, setting the input parameters of the initialized spatiotemporal graph convolutional network, including resource type, quantity, availability and estimated response time, to obtain a configured spatiotemporal graph convolutional network;

[0212] Using the configured spatiotemporal graph convolutional network, multiple scheduling scenarios are simulated. In each simulation, the agent calculates the selection and time arrangement of different paths based on the current geographic information system data and traffic flow forecast information, evaluates the resource scheduling effect under each scenario, and generates simulation results of multiple scheduling scenarios;

[0213] The path selection probability P is calculated by the following formula: ij :

[0214]

[0215] Among them, t ij is the estimated travel time from node i to node j; β is a positive scaling factor used to adjust the impact of time; γ is a positive adjustment factor used to adjust the impact of the target node weight; w j is the weight of the target node j, reflecting its importance or urgency; t ik is the estimated travel time from node i to node k; k is an index variable representing all other target nodes reachable from the current node i;

[0216] The resource scheduling effect score S is calculated using the following formula: i :

[0217]

[0218] Among them, w j is the weight of the target node j; P ij The probability of path selection from node i to node j; δ is a positive adjustment factor used to adjust time sensitivity; η is a positive time decay factor used to reduce the impact of long time paths; j is an index variable representing a specific target node or destination node;

[0219] The total scheduling effect score T is calculated using the following formula:

[0220]

[0221] Among them, S iis the effectiveness score of the i-th scheduling scenario; θ is a positive power adjustment factor used to adjust the non-linear impact of the score; i is an index variable representing each scheduling scenario or path selection plan;

[0222] Based on the simulation results of the multiple scheduling scenarios, analyze and compare the effects of different scheduling scenarios, focus on the plan that can deliver the emergency resources to the destination in the shortest time, and consider the limiting conditions in the actual situation to obtain the preferred path and time arrangement;

[0223] Calculate the average scheduling score WIAS through the following formula:

[0224]

[0225] where w i is the weight of each scheduling scenario, dynamically adjusted according to its urgency or other factors; S i is the effectiveness score of the i-th scheduling scenario; α is a positive scaling factor used to adjust the degree of influence of the score; λ is a trade-off factor to balance the influence of the effectiveness score and cost; C i is the cost of the i-th scheduling scenario, considering traffic flow and other limiting conditions; N represents the total number of scheduling scenarios;

[0226] Using the preferred path and time arrangement, finally generate a scheduling instruction to ensure that the emergency resources arrive at the scene at the fastest speed. The scheduling instruction reflects the influence of geographic information system data and traffic flow prediction, and comprehensively considers the requirements of the optimized resource allocation plan;

[0227] Calculate the optimal path selection P through the following formula opt :

[0228]

[0229] where is the set of all possible paths; WIAS(P) is the weighted exponential average scheduling score of the given path P; μ is a positive adjustment factor used to adjust the influence of distance; D P is the total distance of path P;

[0230] Calculate the optimal time arrangement T through the following formula opt :

[0231]

[0232] where is the set of all possible schedules; C(T) is the cost function of schedule T, taking into account traffic flow and other constraints; λ is a trade-off factor that balances the impact of time and cost; v is a positive adjustment factor used to adjust the exponential decay of time; ρ is a positive time decay factor used to reduce the impact of long schedules.

[0233] The following is a detailed explanation of each parameter:

[0234] t ij : Estimated travel time from node i to node j. Calculated using GIS data and real-time traffic flow forecast information. Usually obtained using map service APIs or internally developed traffic forecast models.

[0235] β: A positive scaling factor used to adjust the impact of time. It is set based on historical data and experience. The optimal value can be found through experiments and parameter adjustment to ensure that the time factor has a reasonable weight in path selection. The introduction of β allows the impact of time to be flexibly adjusted to meet the needs of different scenarios.

[0236] w: The weight of the target node j, reflecting its importance or urgency. Based on a comprehensive evaluation of event type, location attributes, historical rescue case analysis, etc. For example, a hospital may be more important than an ordinary building. Consider the importance of the target node to ensure that high-priority locations get a higher probability of path selection.

[0237] γ: A positive adjustment factor used to adjust the impact of the target node weight. It is also set based on historical data and experience. The optimal value can be found through experiments and parameter adjustment to ensure that the node weight is a reasonable weight in path selection. The introduction of γ allows the impact of node weight to be flexibly adjusted to meet the needs of different scenarios.

[0238] k: index variable, representing all other target nodes reachable from the current node. Defined by the graph structure. In the network graph, all nodes k directly connected to node i can be used as candidate target nodes.

[0239] t ik : Estimated travel time from point i to point i. Calculated using GIS data and real-time traffic flow forecast information.

[0240] w j : The weight of the target node j, reflecting its importance or urgency. Based on a comprehensive evaluation of event type, location attributes (such as hospital, school, etc.), historical rescue case analysis, etc. For example, the importance of a hospital may be higher than that of an ordinary building. Consider the importance of the target node to ensure that high-priority locations receive higher scores.

[0241] P ij:The path selection probability from node i to node j. It is calculated by the previous formula, which comprehensively considers time and node importance. The path selection probability reflects the likelihood of choosing a certain path under the current conditions and is an important basis for evaluating the scheduling effect.

[0242] δ: A positive adjustment factor used to adjust time sensitivity. It is also set according to historical data and experience. The optimal value can be found through experiments and parameter tuning to ensure that the impact of long paths is appropriately weakened.

[0243] η: A positive time decay factor used to reduce the impact of long paths; j is an index variable representing a specific target node or destination node; introducing η makes the impact of long paths gradually decrease, ensuring that scheduling decisions tend to choose shorter paths.

[0244] j: An index variable representing a specific target node or destination node. It is defined through the graph structure. In the network graph, all nodes j directly connected to node i can be used as candidate target nodes.

[0245] S i : The effect score of the i-th scheduling scenario;

[0246] θ: A positive power adjustment factor used to adjust the non-linear impact of the score; it is set according to historical data and experience. The optimal value can be found through experiments and parameter tuning to ensure that the non-linear impact of the score is reasonable in different scenarios.

[0247] i: An index variable representing each scheduling scenario or path selection scheme. Ensure that all possible scheduling scenarios are evaluated and the optimal scheme is selected from them.

[0248] T: The total scheduling effect score. Comprehensively evaluate the effects of all scheduling scenarios to ensure that the final score can reflect the advantages and disadvantages of the overall scheduling scheme.

[0249] w i : The weight of the i-th scheduling scenario, which is dynamically adjusted according to its urgency or other factors. Ensure that the weight of each scheduling scenario can reflect its actual importance and urgency, making the final score more in line with the actual situation.

[0250] α: A positive scaling factor used to adjust the degree of influence of the score; introducing α makes the influence of the score adjustable flexibly to meet the requirements in different scenarios. A higher α value will amplify the score difference, while a lower α value makes the score difference smoother.

[0251] λ: A trade-off factor that balances the influence of the effect score and cost; introducing λ makes it possible to flexibly adjust between the effect score and cost to ensure a reasonable balance between the two.

[0252] C i: The cost function of the i-th dispatch scenario, taking into account traffic flow and other constraints; taking into account actual costs to ensure that dispatch decisions reflect real economic impacts. Higher costs mean higher resource consumption or more complex operations.

[0253] N: represents the total number of scheduling scenarios; ensures that all possible scheduling scenarios are evaluated and the best solution is selected.

[0254] WIAS: stands for Weighted Exponential Average Scheduling Score. It comprehensively evaluates the effects of all scheduling scenarios to ensure that the final score reflects the pros and cons of the overall scheduling solution.

[0255] : The set of all possible paths; ensure that all possible paths are evaluated and the best path is selected from them.

[0256] WIAS(P): Weighted exponential average dispatch score for a given path P; used to evaluate the overall dispatch effect of each path to ensure that the selected path is not only efficient but also cost-effective.

[0257] μ: A positive adjustment factor used to adjust the impact of distance. The introduction of μ allows the impact of distance to be flexibly adjusted to meet the needs of different scenarios.

[0258] D P : The total distance of path P; taking into account the actual travel distance to ensure that scheduling decisions reflect real geographical conditions. Shorter distances generally mean faster response times.

[0259] P opt : Optimal path selection; ensure that the selected path has the highest scheduling effect score and takes into account the actual travel distance to achieve comprehensive optimization.

[0260] : The set of all possible time arrangements; ensure that all possible times are evaluated and the best time is selected from them.

[0261] C(T): Cost function for scheduling T, taking into account traffic flow and other constraints; taking into account actual costs to ensure that scheduling decisions reflect real economic impacts. Higher costs mean higher resource consumption or more complex operations.

[0262] ν: A positive adjustment factor used to adjust the exponential decay of time. The introduction of ν allows the exponential decay of time to be flexibly adjusted to meet the needs of different scenarios.

[0263] ρ: A positive time decay factor used to reduce the impact of long-time schedules. The introduction of ρ gradually reduces the impact of long-time schedules, ensuring that scheduling decisions are more inclined to choose shorter-time schedules.

[0264] The following is a brief introduction to the design reasons of each sub-item:

[0265] Exponential decay term :This item is used to represent the time cost of going from node j to node j. ij As the number increases, the probability of path selection decreases exponentially. This ensures that shorter paths have a higher probability of selection, which meets the demand for fast response in reality.

[0266] Normalization term : This item is used to ensure that the sum of the selection probabilities of all possible paths is 1. By normalizing the time cost of each path, it is possible to avoid the selection probability of a single path being too large or too small, and ensure the rationality of the probability distribution.

[0267] Node weight adjustment term 1+γ·log(1+w j ): This item is used to adjust the importance of the target node. j By performing a logarithmic transformation and multiplying it by the adjustment factor γ, the selection probability of high-priority nodes can be appropriately increased without making the weight too large and affecting the overall result.

[0268] This formula multiplies each item to comprehensively consider the two factors of time and node importance. The exponential decay term reflects the time cost, while the node weight adjustment term reflects the importance of the node. Multiplying the two can take these two important factors into account at the same time, ensuring that the final selected path is both fast and reaches important target nodes first.

[0269] This formula aims to consider not only the time cost but also the importance of the target node when selecting the optimal path. By introducing the scaling factor β and the adjustment factor γ, the balance between the time cost and the importance of the node can be flexibly adjusted to ensure that the scheduling decision can respond quickly and effectively to emergencies. In addition, the normalization process ensures the rationality of the selection probability, enabling the entire system to make the best decision in a complex and changing environment.

[0270] Time sensitivity adjustment term 1+δ·exp(-ηt ij ): The basic term 1 ensures that the path selection has a certain basic weight even without additional time adjustment. The exponential decay term exp(-ηt ij )With t ij As the time increases, the term decreases exponentially, reducing the impact of long-term paths. This meets the needs of rapid response in reality. The adjustment factor δ allows flexible adjustment of time sensitivity to meet the needs of different scenarios.

[0271] This formula multiplies the sub-items to comprehensively consider these three important factors. The node weight reflects the importance of the target node, the path selection probability reflects the feasibility of the path, and the time sensitivity adjustment term ensures the requirement of fast response. The multiplication of the three can comprehensively evaluate the effect of each path, ensuring that the final score takes into account both importance and feasibility and response speed.

[0272] This formula is designed to evaluate the resource scheduling effect starting from a certain node. By comprehensively considering the importance of the target node (weight w j ), the probability of path selection (path selection probability P ij ) and time sensitivity (time sensitivity adjustment term). This formula can comprehensively evaluate the effect of each path and obtain a comprehensive scheduling effect score S by summarizing the scores of all possible paths. i This approach not only considers the speed of reaching each target node, but also the importance of the node, ensuring that the scheduling decision can respond quickly and effectively to emergencies. In addition, by introducing the adjustment factor δ and the time decay factor η, the influence of the time factor can be flexibly adjusted to meet the needs of different scenarios.

[0273] Weighted Exponential Average Scheduling Score (WIAS) (P): This sub-item is used to evaluate the overall scheduling effect of each path. By comprehensively considering the resource scheduling effect score and cost, it ensures that the selected path is not only efficient but also cost-effective. This is the core basis for selecting the optimal path.

[0274] Distance adjustment term μ·log(1+D P ): basic term log(1+D P )The logarithmic function is used to smooth the effect of distance to avoid excessively high scores due to large distances. P ) Ensure that when D P = 0, the item is 0 and does not affect the score. The adjustment factor μ allows the influence of distance to be flexibly adjusted to meet the needs of different scenarios.

[0275] This formula combines the weighted exponential average dispatch score WIAS(P) and the distance adjustment term μ·log(1+D P ) is added to comprehensively consider two important factors: scheduling effect and actual travel distance. The scheduling effect score reflects the overall performance of the path in resource scheduling, while the distance adjustment item ensures that the selected path is as short as possible. The addition of the two can comprehensively evaluate the effect of each path and ensure that the final selected path is both efficient and feasible.

[0276] When this formula aims to select the optimal path, it not only considers the scheduling effect (i.e., the resource scheduling effect score), but also takes into account the actual travel distance. By introducing the adjustment factor μ, the influence of the distance factor can be flexibly adjusted to ensure that the path selection is both efficient and adaptable to the actual situation. Maximize WIAS(P)+μ·log(1+D P ) The goal is to find a path that can provide the best scheduling effect and maintain a short distance, so as to achieve comprehensive optimization. This method is particularly applicable to the field of emergency resource management that requires efficient and intelligent scheduling, which can greatly improve the speed and effect of emergency response, and ensure the most reasonable resource allocation decision-making in a complex and changeable environment.

[0277] Time term T: It directly reflects the length of the time arrangement. A shorter time arrangement means a faster response speed, which is one of the core bases for selecting the optimal time arrangement.

[0278] Cost term λ·C(T): The basic cost function C(T) reflects the actual cost of implementing this time arrangement, including factors such as resource consumption and traffic flow. The trade-off factor λ allows flexible adjustment of the influence degree of the cost to ensure a reasonable balance between time and cost.

[0279] Time decay term ν·exp(-ρT): The basic logarithmic function exp(-ρT) is used to smooth the influence of time and avoid too high a score caused by too long a time. At the same time, exp(-ρT) ensures that when T is large, this term decreases significantly, punishing long time arrangements. The adjustment factor ν allows flexible adjustment of the influence degree of time decay to meet the requirements in different scenarios.

[0280] This formula adds up each sub-item to comprehensively consider the influence of response time, actual cost, and long time arrangements. The time term T reflects the response speed, the cost term λ·C(T) reflects the actual cost of implementing this time arrangement, and the time decay term ν·exp(-ρT) ensures that the selected arrangement is as short as possible. Adding the three together can comprehensively evaluate the effect of each time arrangement, ensuring that the finally selected time arrangement is both efficient and economical, and avoiding the negative impact brought by long time arrangements as much as possible.

[0281] This formula aims to select the optimal time schedule by considering not only the response time (i.e., arriving as quickly as possible) but also the actual cost and the impact of long-term scheduling. By introducing the trade-off factor λ, the adjustment factor ν, and the time decay factor ρ, the influence of each factor can be flexibly adjusted to ensure that the schedule can respond quickly and be economically reasonable, and to avoid the negative impact of long-term scheduling as much as possible. Minimize T+λ. The goal of C(T)+ν·exp(-ρT) is to find a schedule that can provide the best response speed while maintaining a low cost, thereby achieving comprehensive optimization. This method is particularly suitable for the field of emergency resource management that requires efficient and intelligent scheduling. It can greatly improve the speed and effect of emergency response and ensure that the most reasonable resource allocation decisions are made in a complex and changing environment.

[0282] Here is a specific example:

[0283] Suppose in a city emergency dispatch system, the system receives a car accident alarm call. In order to ensure that emergency resources can arrive at the scene as quickly as possible, the system combines geographic information system data and real-time traffic flow forecast information, uses spatiotemporal graph convolutional networks to simulate multiple dispatch scenarios, and selects the optimal path and schedule.

[0284] The following is the data preparation:

[0285] Nodes: emergency resource deployment points (such as hospitals, fire stations) and target locations (such as accident sites)

[0286] Edge: The connection relationship and traffic conditions between these locations

[0287] The following are the input parameters:

[0288] Resource type: ambulance, fire truck, etc.

[0289] Quantity: the quantity of each resource;

[0290] Availability: the amount of resources currently available;

[0291] Estimated response time: The estimated travel time from the deployment point to the target location.

[0292] Simulation processing: Combine geographic information system data and real-time traffic flow forecast information to build a comprehensive scenario simulation environment that reflects the current geographical and traffic conditions in and around the location of the emergency request.

[0293] Initialize the spatiotemporal graph convolutional network: define nodes and edges to generate an initialized spatiotemporal graph convolutional network. Set input parameters to get a configured spatiotemporal graph convolutional network.

[0294] Simulate multiple scheduling scenarios: In each simulation, the agent calculates the selection and time schedule of different paths based on the current GIS data and traffic flow forecast information, evaluates the resource scheduling effect under each scenario, and generates simulation results for multiple scheduling scenarios.

[0295] Use the following formula to calculate the path selection probability P ij :

[0296]

[0297] Among them, t ij is the estimated travel time from node i to node j; β is a positive scaling factor used to adjust the impact of time; γ is a positive adjustment factor used to adjust the impact of the target node weight; w j is the weight of the target node j, reflecting its importance or urgency; t ik is the estimated travel time from node i to node k; k is an index variable representing all other target nodes reachable from the current node i;

[0298] Use the following formula to calculate the resource scheduling effect score S i :

[0299]

[0300] Among them, w j is the weight of the target node j; P ij The probability of path selection from node i to node j; δ is a positive adjustment factor used to adjust time sensitivity; η is a positive time decay factor used to reduce the impact of long time paths; j is an index variable representing a specific target node or destination node;

[0301] Use the following formula to calculate the total scheduling effect score T:

[0302]

[0303] Among them, S i is the effect score of the i-th scheduling scenario; θ is a positive power adjustment factor used to adjust the nonlinear impact of the score; i is an index variable representing each scheduling scenario or path selection scheme;

[0304] The average dispatch score (WIAS) is calculated using the following formula:

[0305]

[0306] Among them, w i is the weight of each scheduling scenario, which is dynamically adjusted according to its urgency or other factors; S iis the effect score of the i-th scheduling scenario; α is a positive scaling factor used to adjust the impact of the score; λ is a trade-off factor that balances the impact of the effect score and cost; C i is the cost of the ith scheduling scenario, taking into account traffic flow and other constraints; N represents the total number of scheduling scenarios;

[0307] Use the following formula to calculate the optimal path selection P opt :

[0308]

[0309] in, is the set of all possible paths; WIAS(P) is the weighted exponential average scheduling score of a given path P; μ is a positive adjustment factor used to adjust the impact of distance; D P is the total distance of path P;

[0310] Use the following formula to calculate the optimal time schedule T opt :

[0311]

[0312] in, is the set of all possible schedules; C(T) is the cost function of schedule T, taking into account traffic flow and other constraints; λ is a trade-off factor that balances the impact of time and cost; v is a positive adjustment factor used to adjust the exponential decay of time; ρ is a positive time decay factor used to reduce the impact of long schedules.

[0313] Assume the following specific values:

[0314] Nodes: A (hospital), B (fire station), C (accident scene)

[0315] Side: AC, BC

[0316] Input parameters:

[0317] Resource Type: Ambulance

[0318] Quantity: 2

[0319] Availability: All available

[0320] Estimated response time: AC: 10 minutes, BC: 15 minutes

[0321] Calculate the path selection probability P ij :

[0322] Assume β = 0.1, γ = 0.5, w C =1:

[0323]

[0324] Calculate the resource scheduling effect score S i :

[0325] Assume δ = 0.1, η = 0.05, w C =1:

[0326] S A =1·0.6827·(1+0.1·exp(-0.05×10))≈0.6827·1.0951≈0.7475

[0327] S B =1·0.3173·(1+0.1·exp(-0.05×15))≈0.3173·1.0779≈0.3418

[0328] Calculate the total scheduling effect score T:

[0329] Let θ = 2:

[0330] T=(0.7475 2 +0.3418 2 ) 1 / 2 ≈(0.5587+0.1168) 1 / 2 ≈0.8219

[0331] Calculate the average dispatch score WIAS:

[0332] Assume w 1 =1,w 2 =1,α=1,λ=0.5,C 1 =0.5, C 2 =0.7:

[0333]

[0334] Calculate the optimal path selection P opt :

[0335] Assume μ = 0.1, D AC =10, D BC =15:

[0336] P opt =arg max(0.854+0.1·log(1+10),0.854+0.1·log(1+15))≈arg max(0.854+

[0337] 0.2303, 0.854+0.2708)≈arg max(1.0843, 1.1248)

[0338] Therefore, path BC is selected as the optimal path.

[0339] Calculate the optimal time schedule T opt :

[0340] Assume ν=0.1, ρ=0.05, C(T)=0.5:

[0341] T opt =min(10+0.5·0.5+0.1·exp(-0.05×10), 15+0.5·0.7+0.1·exp(-0.05×15))≈

[0342] min(10.25+0.0607,15.35+0.0375)≈min(10.3107,15.3875)

[0343] Therefore, the time of 10.3107 minutes is selected as the optimal schedule.

[0344] The calculation results show that the probability of going from B to C is low, but considering the actual traffic conditions and node weights, the path from B to C was finally selected. A higher score indicates that the path and schedule perform well in terms of efficiency and urgency. The score after considering multiple factors reflects the effectiveness of the overall scheduling plan. A higher score means that the plan performs stably in many situations. Path BC is selected as the optimal path to ensure that emergency resources can arrive at the scene as quickly as possible. 10.3107 minutes is selected as the optimal schedule, taking into account the impact of time and cost.

[0345] By introducing a spatiotemporal graph convolutional network for scenario simulation and using a series of optimization formulas to select the optimal path and schedule, this embodiment significantly enhances the intelligence level of the emergency resource scheduling system. The system can not only accurately identify and quantify the similarities between current requests and historical cases, but also comprehensively consider geographic information system data and real-time traffic flow forecast information to provide a scientific basis for optimizing resource allocation. This method is particularly suitable for the field of emergency resource management that requires efficient and intelligent scheduling. It can greatly improve the speed and effectiveness of emergency response and ensure that the most reasonable resource allocation decisions are made in a complex and changing environment.

[0346] Figure 2 A schematic diagram of a system for predicting emergency resources using machine learning is provided for an embodiment of the present application. Figure 2 As shown, the system includes:

[0347] The extraction and integration module 21 is used to obtain and parse real-time emergency request information, extract key metadata of the event, and integrate information from different sensors to generate situational awareness data;

[0348] A measurement and analysis module 22 is used to use the context-aware data and the historical emergency case database, and a similarity evaluation system based on a graph neural network to measure the complex relationship between the current request and the historical cases, obtain a set of historical cases with high similarity, and analyze key influencing factors from the set of historical cases;

[0349] An adjustment generation module 23 is used to dynamically adjust the weight parameters in the resource allocation priority model according to the key influencing factors, and use a reinforcement learning algorithm to simulate multiple scenarios to generate an optimized resource allocation plan for the current emergency request;

[0350] A simulation selection module 24 is used to simulate various dispatch scenarios based on the optimized resource allocation scheme, in combination with geographic information system data and traffic flow prediction, through a spatiotemporal graph convolutional network, select the optimal path and time schedule, and generate dispatch instructions to ensure that emergency resources arrive at the scene as quickly as possible;

[0351] The monitoring and correction module 25 is used to continuously monitor the progress of events and environmental changes during the execution of the scheduling instructions, and to correct the decisions of the intelligent scheduling platform in real time according to the latest observation results.

[0352] Figure 2 The system for predicting emergency resources using machine learning can be performed Figure 1 The implementation principle and technical effect of the method for predicting emergency resources using machine learning described in the illustrated embodiment will not be described in detail. The specific manner in which each module and unit performs operations in the system for predicting emergency resources using machine learning in the above embodiment has been described in detail in the embodiment of the method, and will not be described in detail here.

[0353] In one possible design, Figure 2 The system for predicting emergency resources using machine learning in the illustrated embodiment may be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0354] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0355] The processing component 32 is used to: obtain and parse real-time emergency request information, extract key metadata of the event, and integrate information from different sensors to generate situational awareness data; use the situational awareness data and the historical emergency case database, and a similarity evaluation system built based on a graph neural network to measure the complex relationship between the current request and the historical cases, and obtain a set of historical cases with high similarity, and analyze the key influencing factors from the set of historical cases; according to the key influencing factors, dynamically adjust the weight parameters in the resource allocation priority model, use a reinforcement learning algorithm to simulate multiple scenarios, and generate an optimized resource allocation plan for the current emergency request; based on the optimized resource allocation plan, combined with geographic information system data and traffic flow prediction, simulate multiple scheduling scenarios through a spatiotemporal graph convolutional network, select the optimal path and time schedule, and generate a scheduling instruction to ensure that emergency resources arrive at the scene as quickly as possible; in the process of executing the scheduling instruction, continuously monitor the progress of the event and environmental changes, and correct the decision of the intelligent scheduling platform in real time according to the latest observations.

[0356] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0357] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0358] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0359] The input / output interface provides an interface between the processing component and the peripheral interface module, which may be an output device, an input device, etc.

[0360] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0361] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0362] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The illustrated embodiment is a method for predicting emergency resources using machine learning.

[0363] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0364] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0365] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0366] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting emergency resources using machine learning, characterized in that: include: Obtain and parse real-time emergency request information, extract key metadata of events, and integrate information from different sensors to generate situational awareness data; Using the context-aware data and the historical emergency case database, a similarity evaluation system based on a graph neural network is used to measure the complex relationship between the current request and the historical cases, obtain a historical case set with high similarity, and analyze key influencing factors from the historical case set; According to the key influencing factors, the weight parameters in the resource allocation priority model are dynamically adjusted, and a reinforcement learning algorithm is used to simulate multiple scenarios to generate an optimized resource allocation plan for the current emergency request; Based on the optimized resource allocation scheme, combined with geographic information system data and traffic flow forecast, a variety of dispatch scenarios are simulated through a spatiotemporal graph convolutional network, the optimal path and time schedule are selected, and dispatch instructions are generated to ensure that emergency resources arrive at the scene as quickly as possible; In the process of executing the scheduling instructions, the progress of events and environmental changes are continuously monitored, and the decisions of the intelligent scheduling platform are corrected in real time based on the latest observations.

2. The method according to claim 1, characterized in that The similarity evaluation system constructed based on the graph neural network using the context-aware data and the historical emergency case database measures the complex relationship between the current request and the historical cases, obtains a historical case set with high similarity, and analyzes key influencing factors from the historical case set, including: Based on the context-aware data and the historical emergency case database, a similarity evaluation system based on graph neural network is built. The context-aware data is generated by acquiring and parsing real-time emergency request information, extracting key metadata of the event, and integrating information from different sensors. Using the similarity evaluation system, the context-aware data and historical cases are structured to obtain a current request feature vector and a historical case feature vector, ensuring that the two are compared in the same feature space; According to the current request feature vector and the historical case feature vector, the graph neural network model is trained so that the graph neural network model learns to automatically capture and quantify the complex relationship between nodes, thereby obtaining a trained graph neural network model; Based on the trained graph neural network model, the similarity scores between the current request feature vector and all historical case nodes are calculated and processed, and several historical cases with the highest similarity scores are screened out to obtain a set of historical cases with high similarity; Based on the highly similar historical case collection, analyze the commonalities and differences therein and identify the key factors that have a significant impact on the rescue effect; Based on the key influencing factors, information that helps optimize resource allocation is generated, providing a basis for the subsequent dynamic adjustment of weight parameters in the resource allocation priority model.

3. The method according to claim 2, characterized in that The step of training the graph neural network model according to the current request feature vector and the historical case feature vector so that the graph neural network model learns to automatically capture and quantify the complex relationship between nodes to obtain a trained graph neural network model includes: Using the current request feature vector and the historical case feature vector, a graph-structured data set is constructed, wherein each feature vector is used as a node, and the edge between the nodes represents the potential association or similarity between two cases, to obtain a graph-structured data set for training; Based on the trained graph structure dataset, the parameters of the graph neural network model are initialized, and a loss function is defined to evaluate the difference between the predicted similarity score and the actual similarity score, thereby generating an initialized graph neural network model; Using the initialized graph neural network model, forward propagation calculation processing is performed on the graph structure data set, and information is propagated through a multi-layer network so that the model can learn the local and global relationships between nodes and obtain a preliminary similarity evaluation result; According to the preliminary similarity evaluation results, the weights and other parameters in the graph neural network are adjusted using the back propagation algorithm to minimize the loss function value, ensure that the similarity score output by the model is close to the actual situation, and obtain the optimized graph neural network parameters; Based on the optimized graph neural network parameters, the graph structure data set is forward propagated and processed again, and the iterative optimization process of forward propagation and back propagation is repeated, and the model parameters are continuously updated until the model converges or reaches the preset performance indicators, thereby generating a further optimized graph neural network model; Based on the further optimized graph neural network model, after sufficient training, a trained graph neural network model is finally obtained. The trained graph neural network model can automatically capture and quantify the complex relationships between nodes, providing support for subsequent similarity evaluation.

4. The method according to claim 1, characterized in that: The method dynamically adjusts the weight parameters in the resource allocation priority model according to the key influencing factors, uses a reinforcement learning algorithm to simulate multiple scenarios, and generates an optimized resource allocation plan for the current emergency request, including: Using the key influencing factors analyzed from a collection of highly similar historical cases, the key influencing factors are evaluated for importance, and an importance evaluation result reflecting the degree of influence of each factor on the resource allocation decision is obtained; Based on the importance evaluation result, the weight parameters in the resource allocation priority model are updated to obtain an updated weight configuration reflecting the importance of each factor; Using the updated weight configuration, the reinforcement learning environment is initialized, and a state space, an action space, and a reward function are defined, wherein the state space represents all emergency scenarios, the action space contains all feasible resource allocation decisions, and the reward function is used to quantify the effect of each decision, thereby generating an initialized reinforcement learning environment; According to the initialized reinforcement learning environment, a deep reinforcement learning algorithm is used to simulate and train multiple scenarios. In each iteration, the agent selects an action according to the current state, observes the generated new state and immediate reward, and learns the optimal or near-optimal resource allocation plan by continuously trying different action strategies, thereby obtaining the result of simulation training; Using the results of the simulation training, record and evaluate the resource allocation effects under different scenarios, focus on solutions that significantly improve rescue efficiency or improve patient prognosis, and consider the constraints in actual operations to obtain optimized scenario simulation results; Based on the optimized scenario simulation results, a variety of uncertainties and dynamic change factors are comprehensively considered, and finally a set of optimized resource allocation solutions are output. The optimized resource allocation solutions reflect the weight adjustment of key influencing factors and fully reflect the best practices of resource allocation under different scenarios.

5. The method according to claim 4, characterized in that According to the initialized reinforcement learning environment, a deep reinforcement learning algorithm is used to simulate and train multiple scenarios. In each iteration, the agent selects an action according to the current state, observes the generated new state and immediate reward, and learns the optimal or near-optimal resource allocation plan by continuously trying different action strategies, and obtains the results of the simulation training, including: Based on the initialized reinforcement learning environment, construct a learning model of the intelligent agent, wherein the learning model enables the intelligent agent to learn the best action strategy by interacting with the environment, thereby obtaining an initial learning model; At the beginning of each iteration, using the initial learning model, an action is selected according to the current state, wherein the current state selection is based on the agent's current strategy to ensure that new possibilities can be explored while fully utilizing known best practices to obtain the selected action; According to the selected action, emergency resources are configured and processed, and after the selected action is executed, a new state and an immediate reward are observed, wherein the new state reflects the allocation of emergency resources after the action is executed, and the immediate reward measures the contribution of the selected action to improving rescue efficiency or improving patient prognosis, thereby generating an action execution result; Using the results of the action execution, the learning model parameters of the agent are updated to optimize the agent's prediction ability and action strategy for future rewards, and generate an updated learning model; In the next iteration, the updated learning model is used to select an action again according to the current state, continue to execute the selected action, observe the new state and immediate reward, and further update the learning model. This process is repeated. The agent gradually accumulates data on the effects of resource allocation in different scenarios by constantly trying different action strategies, and learns which resource allocation decisions can bring optimal or near-optimal results in different emergency scenarios. After multiple rounds of iterative training, a set of optimized resource allocation plans are finally output based on the updated learning model. The optimized resource allocation plans reflect the learning outcomes of the intelligent agent in different scenarios, embody the optimal or near-optimal resource allocation decisions, and obtain the results of simulation training.

6. The method according to claim 1, characterized in that Based on the optimized resource allocation scheme, combined with geographic information system data and traffic flow forecast, a variety of dispatch scenarios are simulated through a spatiotemporal graph convolutional network, the optimal path and time schedule are selected, and dispatch instructions are generated to ensure that emergency resources arrive at the scene as quickly as possible, including: Using the optimized resource allocation scheme, combined with geographic information system data and real-time traffic flow forecast information, a comprehensive scenario simulation environment is constructed, wherein the comprehensive scenario simulation environment can reflect the current location of the emergency request and its surrounding geographical and traffic conditions, thereby obtaining a comprehensive scenario simulation environment; Based on the comprehensive scenario simulation environment, a spatiotemporal graph convolutional network is initialized, nodes and edges are defined, the nodes represent emergency resource deployment points and target locations, the edges represent connection relationships and traffic conditions between these locations, and an initialized spatiotemporal graph convolutional network is generated; According to the resource allocation decision in the optimized resource configuration scheme, setting the input parameters of the initialized spatiotemporal graph convolutional network, including resource type, quantity, availability and estimated response time, to obtain a configured spatiotemporal graph convolutional network; Using the configured spatiotemporal graph convolutional network, multiple scheduling scenarios are simulated. In each simulation, the agent calculates the selection and time arrangement of different paths based on the current geographic information system data and traffic flow forecast information, evaluates the resource scheduling effect under each scenario, and generates simulation results of multiple scheduling scenarios; Based on the simulation results of the various dispatch scenarios, the effects of different dispatch scenarios are analyzed and compared, focusing on the solution that can deliver emergency resources to the destination in the shortest time, and considering the constraints in the actual situation to obtain the optimal path and time arrangement; Using the preferred path and time schedule, a dispatch instruction is ultimately generated to ensure that emergency resources arrive at the scene as quickly as possible. The dispatch instruction reflects the impact of geographic information system data and traffic flow forecasts, and comprehensively considers the requirements of the optimized resource allocation plan.

7. The method according to claim 1, characterized in that In the process of executing the dispatch instruction, the progress of events and environmental changes are continuously monitored, and the decision of the intelligent dispatch platform is corrected in real time according to the latest observation results, including: Using the generated dispatch instructions, start the dispatch process of emergency resources, set up and start the monitoring system, the monitoring system integrates data from multiple information sources, including real-time traffic conditions, weather conditions, live video streams, and feedback from emergency personnel to obtain initial monitoring data; Based on the initial monitoring data, dynamically monitor the incident site and its surrounding environment, evaluate the effectiveness and feasibility of the current dispatch instructions, and obtain the latest observation results; Using the latest observations, identify any factors that affect the arrival time or efficiency of emergency resources, and assess the impact of the factors on the existing dispatch plan to generate an impact assessment report; According to the impact assessment report, the algorithm model built into the intelligent dispatching platform is used to make real-time adjustments to the current dispatching instructions, wherein the real-time adjustments involve re-planning the route, changing resource allocation, or adjusting the estimated arrival time, so as to ensure that emergency resources can arrive at the scene quickly and safely, and generate updated dispatching instructions; Using the updated dispatch instructions, maintain communication with on-site emergency personnel, and promptly convey the updated dispatch instructions and related information to ensure that front-line personnel can take appropriate actions based on the latest situation and obtain confirmation feedback; Based on the confirmation feedback, the process of monitoring, evaluating and adjusting is continuously cycled until the emergency task is completed, and each cycle is optimized based on the latest observation results to ensure that the dispatch decision is always adapted to the latest actual situation and generate the final dispatch record; After the task is completed, the final scheduling record is used to collect and record all data in the entire scheduling process for subsequent analysis and improvement.

8. A system for predicting emergency resources using machine learning, characterized in that: include: The extraction and integration module is used to obtain and parse real-time emergency request information, extract key metadata of the event, and integrate information from different sensors to generate situational awareness data; A measurement and analysis module is used to use the situational awareness data and the historical emergency case database, and a similarity evaluation system built based on a graph neural network to measure the complex relationship between the current request and the historical cases, obtain a historical case set with high similarity, and analyze key influencing factors from the historical case set; An adjustment generation module is used to dynamically adjust the weight parameters in the resource allocation priority model according to the key influencing factors, and use a reinforcement learning algorithm to simulate multiple scenarios to generate an optimized resource allocation plan for the current emergency request; A simulation selection module is used to simulate various dispatch scenarios based on the optimized resource allocation scheme, combined with geographic information system data and traffic flow forecasts, through a spatiotemporal graph convolutional network, select the optimal path and time schedule, and generate dispatch instructions to ensure that emergency resources arrive at the scene as quickly as possible; The monitoring and correction module is used to continuously monitor the progress of events and environmental changes during the execution of the scheduling instructions, and to correct the decisions of the intelligent scheduling platform in real time based on the latest observation results.

9. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a method for predicting emergency resources using machine learning as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, a method for predicting emergency resources using machine learning as described in any one of claims 1 to 7 is implemented.

Images