Intelligent transfer and self-help guidance assisted decision-making method and system based on federated learning

Abstract

The application provides an intelligent transfer and self-help guidance assisted decision-making method and system based on federal learning, relates to the technical field of intelligent medical treatment, and comprises the following steps: acquiring emergency help information, constructing an emergency rescue knowledge graph, extracting physiological characteristic parameters to identify diseases, and generating a self-help scheme; based on the diseases and the position of a user, searching for medical resources, adopting an ant colony algorithm to solve a multi-objective optimization function to obtain an optimal transfer path; and inputting the transfer path and the self-help scheme into a federal learning model to obtain a comprehensive decision. The application realizes accurate and efficient emergency rescue and self-help guidance, and improves the medical resource deployment efficiency and the first-aid success rate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent medical treatment, and in particular to an intelligent transfer and self-help guidance assisted decision-making method and system based on federated learning. BACKGROUND

[0002] With the development of society and the improvement of medical health awareness, the public's demand for emergency medical rescue services is increasing. When a sudden illness or accidental injury occurs, the patient or the surrounding personnel often needs to quickly obtain professional self-help guidance and quickly transfer to the most suitable medical institution. The traditional medical emergency system mainly relies on telephone call centers to provide rescue guidance and dispatch ambulances, and with the development of artificial intelligence, big data and mobile internet technology, intelligent medical emergency decision-making systems have become a research hotspot.

[0003] In recent years, medical emergency technology based on artificial intelligence has made some progress. Some systems provide preliminary diagnosis and self-help suggestions by collecting patient physiological parameters and symptom descriptions combined with a medical knowledge base; intelligent navigation technology is applied to ambulance dispatching and optimal path planning; knowledge graph technology is used to build a medical knowledge system to support medical decision-making. However, the organic integration of these technologies and the protection of medical data security have become important challenges for current medical emergency decision-making systems. SUMMARY

[0004] The embodiments of the present application provide an intelligent transfer and self-help guidance assisted decision-making method and system based on federated learning, which can solve the problems in the prior art.

[0005] In a first aspect, the embodiments of the present application provide an intelligent transfer and self-help guidance assisted decision-making method based on federated learning, comprising:

[0006] Obtaining emergency help information sent by a user terminal;

[0007] Extracting knowledge entities and relationships from a medical emergency knowledge base and a historical rescue case base to construct an emergency rescue knowledge graph; based on the emergency help information, extracting physiological feature parameters and identifying the user's current illness, searching for a path in the emergency rescue knowledge graph, and generating a self-help guidance scheme in combination with the on-site environment;

[0008] Based on the user's current illness and user location, retrieving matching medical resource requirements, determining multiple candidate medical institutions, and constructing a multi-objective optimization function in combination with the traffic conditions of each road segment at the current time, and using an ant colony algorithm to solve it to obtain an optimal transfer path;

[0009] Inputting the optimal transfer path and the self-help guidance scheme into a federated learning model to obtain a comprehensive decision, and sending the comprehensive decision to the user terminal.

[0010] extracting knowledge entities and relationships from a medical first-aid knowledge base and a historical rescue case base, and constructing an emergency rescue knowledge graph, including:

[0011] obtaining diagnosis and treatment specifications and first-aid guidelines from the medical first-aid knowledge base, generating structured medical knowledge texts, constructing a medical field word vector space, mapping medical terms in the medical knowledge texts to semantic vectors, extracting text features of the semantic vectors, and identifying medical knowledge entities according to the text features;

[0012] dividing the medical knowledge entities into multiple entity categories according to a medical professional tree classification system, calculating correlation degrees between the entity categories, marking entity pairs with correlation degrees greater than a correlation threshold as candidate relationship pairs, screening the candidate relationship pairs through confidence evaluation, and obtaining an initial knowledge graph;

[0013] identifying new type knowledge entities and relationships from the historical rescue case base, verifying semantic compatibility of the new type entities and relationships with existing knowledge in the initial knowledge graph, integrating the verified new type entities and relationships into the initial knowledge graph, and obtaining an emergency rescue knowledge graph.

[0014] based on the emergency help information, extracting physiological feature parameters and identifying the current illness of the user, performing path search in the emergency rescue knowledge graph, and generating a self-help guidance scheme in combination with the on-site environment, including:

[0015] extracting symptom feature vectors and physiological feature data of the user from the emergency help information, performing time alignment and synchronous calibration on the physiological feature data, and generating physiological feature vectors;

[0016] based on the symptom feature vectors and the physiological feature vectors, in combination with a symptom-disease association network in the emergency rescue knowledge graph, calculating an illness identification probability of the user at the current time, and determining the current illness of the user;

[0017] taking the current illness as a starting state node, determining a target rehabilitation state as a target state node, calculating an initial candidate rescue path from the starting state node to the target state node, pruning the initial candidate rescue path based on resource consumption upper limit, time consumption upper limit and environmental adaptability constraints, and obtaining an optimized candidate rescue path;

[0018] calculating a feasibility score of each of the optimized candidate rescue paths, selecting the candidate rescue path with the highest feasibility score, decomposing the candidate rescue path according to execution steps, identifying target rescue actions, and obtaining a self-help guidance scheme.

[0019] The initial candidate rescue path from the starting state node to the target state node is calculated, and the initial candidate rescue path is pruned based on resource consumption upper limit, time consumption upper limit and environmental adaptability constraint to obtain an optimized candidate rescue path, comprising:

[0020] Starting from the starting state node, adjacent state nodes are sequentially traversed through depth search, node feature vectors of the adjacent state nodes are extracted, and reachability scores of the adjacent state nodes to the target state node are calculated, and the adjacent state nodes with reachability scores greater than a reachability threshold are selected as next search nodes to generate multiple initial candidate rescue paths;

[0021] For each initial candidate rescue path, the total resource consumption of all state nodes thereof is calculated, the execution time and the conversion delay time of each state node are counted to obtain a total execution time, and the environmental adaptability score is calculated based on the current environment; when any one of the total resource consumption, the total execution time and the environmental adaptability score exceeds a default limit threshold, the candidate rescue path is marked as an infeasible path, and multiple feasible candidate rescue paths are obtained;

[0022] The rescue action similarity between the feasible candidate rescue paths is calculated, and the rescue paths with rescue action similarity higher than a path merging threshold are merged to obtain the optimized candidate rescue paths.

[0023] Based on the current illness of the user and the user location, a matched medical resource demand is retrieved, multiple candidate medical institutions are determined, and a multi-objective optimization function is constructed in combination with the traffic state of each road segment at the current time, and an ant colony algorithm is used for solving to obtain an optimal transfer path, comprising:

[0024] Based on the current illness of the user, the importance weight of each type of medical resource is calculated, and in combination with the standard configuration demand amount of each type of medical resource, candidate medical institutions meeting the basic demand are screened out;

[0025] The bed occupancy, expert on-duty and equipment usage of the candidate medical institutions are obtained, a resource availability score is calculated, and based on the resource availability score and the transfer distance, a comprehensive score of each candidate medical institution is obtained;

[0026] The road congestion intensity is calculated according to the road traffic volume data and the average speed data, and the real-time traffic time of each road segment is predicted in combination with the signal timing data; a multi-objective optimization function is constructed based on the comprehensive score and the real-time traffic time, and a maximum allowable transfer time is set;

[0027] A plurality of search units are set in a road grid, a transfer path is determined based on path pheromone concentration and dynamic heuristic information, the pheromone update amount is adjusted according to the multi-objective optimization function, and the optimal transfer path is obtained through iterative optimization.

[0028] The plurality of search units are arranged in a road grid, a transfer path is determined based on path pheromone concentration and heuristic strength, the pheromone update amount is adjusted according to the multi-objective optimization function, and the optimal transfer path is obtained through iterative optimization, comprising:

[0029] A plurality of search units are randomly deployed in a road grid, each search unit contains a current position, a moving direction and a visited path, the heuristic strength of each search unit is calculated by combining historical heuristic strength and real-time travel time of each road section;

[0030] An initial pheromone concentration is randomly placed in the road grid, the product of the heuristic strength and the initial pheromone concentration is taken as the path selection probability of the search unit, the transfer direction of the next step is determined based on the path selection probability, and the transfer path formed by each search unit in the current iteration cycle is obtained;

[0031] The multi-objective optimization function value of each transfer path is calculated, which is mapped as a pheromone increment, and the pheromone concentration in the road grid is updated according to the pheromone evaporation rate;

[0032] The distribution entropy of the multi-objective optimization function value is calculated, and when the distribution entropy change amount of continuous multiple iterations is less than an entropy change threshold, the transfer path with the maximum multi-objective function value is selected from the transfer paths in the current iteration cycle as the optimal transfer path.

[0033] The second aspect of the embodiment of the application provides an intelligent transfer and self-help guidance auxiliary decision-making system based on federated learning, comprising:

[0034] The first unit is used for acquiring emergency help information sent by a user terminal;

[0035] The second unit is used for extracting knowledge entities and relationships from a medical emergency knowledge base and a historical rescue case base to construct an emergency rescue knowledge graph, extracting physiological feature parameters and identifying a current illness of a user based on the emergency help information, performing path search in the emergency rescue knowledge graph, and generating a self-help guidance scheme in combination with a field environment;

[0036] The third unit is used for retrieving matched medical resource requirements based on the current illness of the user and the location of the user, determining a plurality of candidate medical institutions, constructing a multi-objective optimization function in combination with the travel state of each road section at the current moment, and obtaining an optimal transfer path by solving the multi-objective optimization function using an ant colony algorithm;

[0037] The fourth unit is used for inputting the optimal transfer path and the self-help guidance scheme into a federated learning model to obtain a comprehensive decision, and sending the comprehensive decision to the user terminal.

[0038] The third aspect of the embodiment of the application,

[0039] An electronic device is provided, comprising:

[0040] a processor;

[0041] a memory for storing processor-executable instructions;

[0042] wherein the processor is configured to invoke the instructions stored by the memory to perform the method as described above.

[0043] A fourth aspect of the embodiments of the present application,

[0044] A computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method as described above.

[0045] The beneficial effects of the present application are as follows:

[0046] The present application constructs an emergency rescue knowledge graph, extracts knowledge entities and relationships from a medical first aid knowledge base and a historical rescue case base, can quickly identify the current illness of the user and generate a targeted self-help guidance scheme, improves the self-help efficiency and scientificity in emergency situations, and effectively reduces the risk of illness deterioration.

[0047] The present application adopts an ant colony algorithm to construct a multi-objective optimization function, comprehensively considers the user's illness, location information and real-time traffic conditions of each section, can scientifically determine the optimal transfer path, greatly shortens the emergency response time, and improves the medical resource allocation efficiency.

[0048] The present application uses a federated learning model to comprehensively decide the self-help guidance scheme and the transfer path, guarantees the patient information security while realizing the optimal matching of medical resources and patient needs, forms a complete closed-loop decision support system from self-help guidance to medical transfer, and significantly improves the overall quality of emergency services. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 FIG. 1 is a flowchart of an intelligent transfer and self-help guidance auxiliary decision-making method based on federated learning according to an embodiment of the present application;

[0050] Figure 2 FIG. 2 is a flowchart of a self-help guidance scheme generation method according to an embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0052] The technical solutions of the present application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described in some embodiments.

[0053] Figure 1 The flowchart of the method for intelligent transfer and self-help guidance assisted decision-making based on federated learning in the embodiments of the present application is shown in FIG. 1. Figure 1 The method comprises the following steps.

[0054] Acquiring emergency help information sent by a user terminal;

[0055] Extracting knowledge entities and relationships from a medical first-aid knowledge base and a historical rescue case base to construct an emergency rescue knowledge graph; extracting physiological feature parameters and identifying a current illness of a user based on the emergency help information, performing path search in the emergency rescue knowledge graph, and generating a self-help guidance scheme in combination with a field environment;

[0056] Based on the current illness of the user and the location of the user, retrieving matched medical resource requirements, determining a plurality of candidate medical institutions, and constructing a multi-objective optimization function in combination with the traffic conditions of each road segment at the current time, and solving the multi-objective optimization function by using an ant colony algorithm to obtain an optimal transfer path;

[0057] Inputting the optimal transfer path and the self-help guidance scheme into a federated learning model to obtain a comprehensive decision, and sending the comprehensive decision to the user terminal.

[0058] In an optional embodiment, extracting knowledge entities and relationships from a medical first-aid knowledge base and a historical rescue case base to construct an emergency rescue knowledge graph comprises the following steps.

[0059] Acquiring diagnosis and treatment specifications and first-aid guidelines from the medical first-aid knowledge base to generate structured medical knowledge texts, constructing a medical field word vector space, mapping medical terms in the medical knowledge texts into semantic vectors, extracting text features of the semantic vectors, and identifying medical knowledge entities according to the text features;

[0060] The medical knowledge entity is divided into multiple entity categories according to a medical professional tree classification system, the correlation degree between the entity categories is calculated, entity pairs with a correlation degree greater than a correlation threshold are marked as candidate relationship pairs, the candidate relationship pairs are screened through confidence evaluation, and an initial knowledge graph is obtained;

[0061] New type knowledge entities and relationships are identified from a historical rescue case library, and semantic compatibility of the new type entities and relationships with existing knowledge in the initial knowledge graph is verified, the new type entities and relationships that pass the verification are integrated into the initial knowledge graph, and an emergency rescue knowledge graph is obtained.

[0062] In the process of constructing the emergency rescue knowledge graph, diagnosis and treatment specifications and first aid guidelines are obtained from a medical first aid knowledge base, multiple authoritative medical data sources are accessed, such as a standardized diagnosis and treatment file library of a first aid medical center and first aid guidelines published by various professional societies, these resources are collected through network crawler technology, long texts are divided into small paragraphs with complete semantics using document segmentation technology in natural language processing, for unstructured PDF documents, optical character recognition technology is applied for digital conversion, and regular expressions and pattern matching methods are used to extract document titles, chapters and key paragraphs, forming an initial knowledge text set.

[0063] To construct a medical field word vector space, a distributed representation learning method is used, in specific implementation, medical texts are subjected to word segmentation processing, a medical special word segmentation dictionary is used to ensure the integrity of professional terms, for example, "acute myocardial infarction with ventricular fibrillation" is treated as a whole word entry rather than being split, a 300-dimensional vector space is used for word vector training, the window size is set to 5, the negative sampling number is 5, and the iteration number is 15, and the generated word vector can capture the semantic relationship between medical terms. In actual application, the cosine similarity between "myocardial infarction" and "chest pain" is 0.82, and the similarity with "fracture" is only 0.14, proving that the vector space accurately reflects the semantic distance between medical concepts.

[0064] In the medical knowledge entity identification stage, a dual strategy based on rules and deep learning is combined, the rule part includes medical term dictionary matching and regular expression pattern recognition, which can quickly capture standardized medical entities, for example, matching "body temperature > 39°C" to identify "high fever" symptom entity, the deep learning part uses a bidirectional long short-term memory network structure, the input layer is a word vector, the number of hidden layer nodes is 128, and the output layer uses a conditional random field for sequence labeling, the F1 value of the model on the validation set reaches 0.92, and the entity identification results include "cardiopulmonary impairment", "traumatic shock", "dyspnea", "loss of consciousness", "chest compression", "tracheal intubation" and other entities.

[0065] The identified entities are classified into multiple categories according to the international medical professional classification system, specifically using a four-level classification structure, the first level includes six categories such as disease, symptom, treatment, drug, equipment, and first aid skill; the second level is further divided into 43 subcategories, such as cardiovascular disease, respiratory disease, etc. under the disease category; the third level is further refined into 217 specific categories; and the fourth level is specific entities. Through this hierarchical classification, "cardiopulmonary resuscitation" can be correctly classified as "first aid skill - basic life support technology - cardiopulmonary resuscitation" category.

[0066] The correlation degree between entities is calculated using a multi-dimensional feature fusion method. The initial correlation is constructed by the co-occurrence frequency of entities, such as "cardiac arrest" and "cardiopulmonary resuscitation" co-occur 723 times in the corpus. The vector cosine similarity between entities is calculated to capture semantic associations. Combined with entity type constraint rules, such as increasing the correlation weight between "symptom" and "disease" entities by 0.2, the above features are fused to calculate the comprehensive correlation score. Set the correlation threshold to 0.65, and mark the entity pairs greater than the threshold as candidate relationship pairs.

[0067] In the candidate relationship pair screening stage, the relationship classifier evaluates the confidence of each candidate entity pair. The relationship classifier receives the context syntax information of the entity pair and outputs the relationship type and confidence. For example, for the entity pair "chest compression" and "cardiac arrest", the system determines the relationship type as "treatment method" with a confidence of 0.94, so the relationship is retained. Candidate relationships with a confidence lower than 0.7 are filtered to ensure the accuracy of the initial knowledge graph. Through the above steps, the initial knowledge graph contains 5,823 entities and 12,746 relationships.

[0068] To continuously expand and improve the knowledge graph, access the historical rescue case library for new knowledge discovery. The rescue case library contains more than 10,000 real first aid records, each record contains injury description, on-site disposal measures, transportation process and treatment results, etc. Apply entity recognition algorithm to the case text to discover potential new entities. For the identified new entities, such as the special injury "explosive decompression injury" caused by a new type of disaster, calculate its semantic similarity with the existing entities in the knowledge graph. When the similarity with the most similar entity "blast injury" is 0.76, it is considered as a potential new entity.

[0069] In the new knowledge verification stage, semantic compatibility test is used to ensure that the new knowledge is consistent with the existing knowledge system. Semantic compatibility evaluation includes entity type consistency check, relationship logic conflict detection and expert rule verification. The verified new entities and relationships are integrated into the existing knowledge graph, and the correlation weight between related entities is updated. For example, the newly discovered relationship "mask ventilation" is the "emergency treatment method" of "airway obstruction", which does not conflict with the existing knowledge and is successfully integrated into the graph after verification.

[0070] The finally constructed emergency rescue knowledge graph contains 6,205 medical knowledge entities and 14,362 relationships, covering the main fields of emergency medicine and providing comprehensive knowledge support for emergency rescue decision-making.

[0071] Figure 2 The flowchart of the self-help guidance scheme generation method for emergency rescue. In an optional implementation, based on the emergency help information, the physiological characteristic parameters are extracted and the current illness of the user is identified, the path search is performed in the emergency rescue knowledge graph, and the self-help guidance scheme is generated in combination with the on-site environment, including:

[0072] From the emergency help information, the symptom feature vector and the physiological feature data of the user are extracted, the physiological feature data is time-aligned and synchronized calibrated, and the physiological feature vector is generated;

[0073] Based on the symptom feature vector and the physiological feature vector, in combination with the symptom-disease association network in the emergency rescue knowledge graph, the illness recognition probability of the user at the current time is calculated, and the current illness of the user is determined;

[0074] The current illness is taken as a starting state node, and a target rehabilitation state is determined as a target state node, an initial candidate rescue path from the starting state node to the target state node is calculated, the initial candidate rescue path is pruned based on the upper limit of resource consumption, the upper limit of time consumption and environmental adaptability constraints, and an optimized candidate rescue path is obtained;

[0075] The feasibility score of each of the optimized candidate rescue paths is calculated, the candidate rescue path with the highest feasibility score is selected, and the execution steps are decomposed to identify the target rescue action, and the self-help guidance scheme is obtained.

[0076] In this embodiment, emergency help information containing user physiological condition information is received, which comes from various channels such as intelligent wearable devices, mobile terminal devices or intelligent medical devices, etc. The emergency help information usually contains multi-modal information such as text description, voice recording and device monitoring data, etc. For the received emergency help information, information analysis and preprocessing are performed, and different formats of information are uniformly converted into a standardized data format.

[0077] For text-form symptom descriptions, use word segmentation techniques to divide the text into word units, such as "I feel chest tightness and shortness of breath, and heart rate acceleration" into "I", "feel", "chest tightness", "shortness of breath", "and", "heart rate", "acceleration", etc. Use part-of-speech tagging techniques to identify symptom-related words such as "chest tightness", "shortness of breath", "heart rate acceleration", etc. Establish a medical symptom knowledge base containing common symptom terms, synonyms, and related descriptions, which contains standard names, common expressions, and corresponding feature values of symptoms. For example, for the symptom "chest tightness", the knowledge base stores multiple expressions such as "chest tightness", "chest tightness", "chest tightness", etc. and maps them to a unified symptom identifier. Through text matching algorithms, the identified symptom description is matched with the standard symptoms in the knowledge base to generate a standardized symptom feature set.

[0078] For each identified symptom, extract its severity, duration, frequency, and other attribute information, such as "severe headache for two hours", extract the symptom "headache", severity "severe", and duration "two hours". Associate these attributes with the corresponding symptoms to construct a structured symptom description. Based on the extracted symptoms and their attributes, construct a symptom feature vector, which is a multi-dimensional array, each dimension corresponding to a specific symptom or attribute. Map the identified symptoms to the corresponding dimensions of the vector, and assign values to each dimension through the presence of symptoms, severity, and other attributes. For example, in a feature space containing 100 common symptoms, if "chest tightness" is identified and the degree is "moderate", assign a value of 0.6 to the corresponding "chest tightness" dimension (assuming the degree range is 0-1).

[0079] For physiological feature data, extract user's physiological indicators such as heart rate, blood pressure, blood oxygen saturation, and respiratory rate from emergency help information. These data come from different devices and have different collection times and frequencies. In order to accurately assess the user's health status, it is necessary to time-align and synchronize the heterogeneous data.

[0080] Check the timestamp format of each data source, convert timestamps of different formats to a standard time format, for example, convert "2025-09-15 08:30:45" and "15 / 09 / 2025 08:30:45" to ISO 8601 format "2025-09-15T08:30:45Z", identify the sampling frequency of each data source, such as heart rate data sampled once per second, while blood pressure data sampled once per minute, select a unified time reference, for example, take the smallest sampling interval as the reference, create a uniformly distributed time point sequence. For data with low sampling frequency, use interpolation algorithm to fill in the missing time points. Specifically, for linearly changing physiological indicators such as body temperature, linear interpolation is used; for indicators with greater volatility, such as heart rate, cubic spline interpolation is used to maintain the smoothness of the curve.

[0081] Considering the time bias between different devices, by analyzing the time sequence pattern of the data, such as the correlation between heart rate and respiratory rate, identify the time offset between data sequences, calculate the time delay between different physiological indicator sequences through the maximum cross-correlation method, and adjust the timestamp accordingly to ensure that data from different sources is aligned in time, detect and process outliers and noise in the data, for example, for the heart rate data sequence [75, 76, 150, 74, 75], identify 150 as an outlier, replace it with the median of the surrounding values through median filtering.

[0082] Based on the time-aligned and synchronized calibrated physiological data, construct a physiological feature vector, which contains multiple dimensions, each corresponding to a physiological indicator or its statistical feature, for time-series physiological data, extract statistical features such as mean, standard deviation, peak value, etc. to capture the distribution characteristics and trend of the data, for example, for heart rate data, not only record the current heart rate value, but also calculate the average heart rate, heart rate variability and trend within the past 5 minutes, etc. Organize these features into a structured physiological feature vector to facilitate subsequent health status assessment and emergency level determination.

[0083] Based on the symptom feature vector and the physiological feature vector, the symptom-disease association network in the emergency rescue knowledge graph is combined to calculate the disease identification probability of the user at the current time, determine the current disease of the user, and in this embodiment, the emergency rescue knowledge graph contains 5000 common first-aid diseases, each disease is associated with a specific symptom combination and physiological feature mode, and a deep neural network model is used to take the symptom feature vector and the physiological feature vector as input and output the disease type and its probability value suffered by the user. In the specific implementation process, the two feature vectors pass through their respective feature extraction layers, the features are fused, and the probability distribution of each disease is output through the softmax layer. For example, when the user's symptom features include "severe chest pain, cold sweat, nausea", and the physiological features show "rapid heart rate (110 times / min), high blood pressure (150 / 95 mmHg)", the output is "probability distribution of acute myocardial infarction: 85.6%, pulmonary embolism: 10.2%, gastric ulcer: 4.2%", and the current disease of the user is determined to be acute myocardial infarction.

[0084] The current disease is taken as the starting state node, the target rehabilitation state is determined as the target state node, and the initial candidate rescue path from the starting state node to the target state node is calculated. In the emergency rescue knowledge graph, each disease state is connected to the corresponding rescue measure, and the resources, time cost and environmental requirements required for executing the rescue measure are marked on each edge. An improved search algorithm is used to generate multiple candidate paths from the starting state to the target state in the knowledge graph. Taking acute myocardial infarction as an example, the following candidate paths are generated: Path 1: "take aspirin-take sublingual nitroglycerin-call emergency-keep semi-recumbent position-rest-continuously monitor vital signs"; Path 2: "cardiopulmonary resuscitation-use automatic external defibrillator-call emergency-continuously monitor vital signs"; Path 3: "take sublingual nitroglycerin-flat position-deep breathing relaxation-call emergency-continuously monitor vital signs".

[0085] When pruning the initial candidate rescue path, three constraints are used: resource consumption upper limit, time consumption upper limit and environmental adaptability constraint. The resource consumption upper limit refers to the limitation of the rescue materials currently available to the user, such as the fact that ordinary families do not have automatic external defibrillators. The time consumption upper limit refers to the limitation of the golden rescue time window for some serious diseases. The environmental adaptability constraint refers to whether the current environment is suitable for executing a specific rescue measure, such as the fact that standard cardiopulmonary resuscitation cannot be performed in a small space. Through user location information, environmental description and available resource list, each candidate path is evaluated and paths that do not meet the constraint conditions are removed. In the above example, assuming that the user is at home and does not have a defibrillator, path 2 is pruned due to resource constraints; if the user describes the environment as crowded and unable to lie flat, path 3 is pruned due to environmental constraints; only path 1 remains as the optimized candidate rescue path.

[0086] The feasibility score of each optimized candidate rescue path is calculated by comprehensively considering success probability, execution difficulty and risk factors. The success probability is calculated based on the success rate of the path in historical rescue cases. The execution difficulty evaluates the operation complexity of each step in the path for ordinary people. The risk factors consider the negative effects generated in the execution process. Each factor is assigned a weight, and the weighted sum is obtained to get the comprehensive feasibility score. The candidate rescue path with the highest feasibility score is selected. The selected rescue path is divided into execution steps, target rescue actions are identified, and a self-rescue guidance scheme is generated. Taking path 1 as an example, the system divides it into the following specific action steps: 1) take out aspirin tablets (if any) and chew them; 2) take out nitroglycerin tablets and place them under the tongue for ingestion; 3) dial the emergency number and clearly describe the symptoms; 4) adjust to a semi-recumbent position with the upper body raised by 30 degrees; 5) keep breathing steady and record the heart rate every 5 minutes. The detailed level of the guidance scheme is adjusted according to the actual situation of the user. For example, if the user indicates that he does not understand what "semi-recumbent position" means, more detailed posture adjustment instructions and graphical guidance will be provided.

[0087] In actual application, assuming that the user sends a help message "I suddenly feel severe chest pain, like being pressed by a heavy object, accompanied by cold sweating and nausea, and now I am at home", and the wearable device transmits physiological data of heart rate 112 times / min and blood pressure 152 / 96 mmHg, it is identified that the user has acute myocardial infarction, and a self-rescue guidance scheme is generated: "Please immediately chew a piece of aspirin (if available); place a piece of nitroglycerin under the tongue (if available); dial the emergency number immediately; adopt a semi-recumbent position with the upper body raised by about 30 degrees using a pillow; keep calm and avoid any activity; continuously record your heart rate and symptom changes, and wait for the arrival of rescue personnel." The scheme takes into account the user's home environment conditions, optimizes the initial self-rescue measures for acute myocardial infarction, and improves the survival probability of the user during the waiting period for professional rescue.

[0088] In an optional implementation, an initial candidate rescue path from the starting state node to the target state node is calculated, and the initial candidate rescue path is pruned based on resource consumption upper limit, time consumption upper limit and environmental adaptability constraint to obtain an optimized candidate rescue path, including:

[0089] Starting from the starting state node, adjacent state nodes are sequentially traversed through depth search, the node feature vectors of the adjacent state nodes are extracted, and the reachability scores of the adjacent state nodes to the target state node are calculated. Adjacent state nodes with reachability scores greater than a reachability threshold are selected as next search nodes to generate multiple initial candidate rescue paths.

[0090] For each initial candidate rescue path, calculate the total resource consumption of all state nodes thereof, count the execution time and transition delay time of each state node, obtain the total execution time, and calculate the environmental adaptability score based on the current environment; when any of the total resource consumption, the total execution time and the environmental adaptability score exceeds the default limit threshold, mark the candidate rescue path as an infeasible path, and obtain a plurality of feasible candidate rescue paths;

[0091] Calculate the rescue action similarity between the feasible candidate rescue paths, merge the rescue paths with a rescue action similarity higher than a path merging threshold, and obtain the optimized candidate rescue paths.

[0092] In the specific embodiment, starting from the initial state node, adjacent state nodes are traversed by a depth-first search algorithm, for each adjacent state node, a node feature vector is extracted, including symptom relief degree, vital sign improvement degree, patient self-help difficulty, treatment response time and other key feature calculations, the accessibility score of the node to the target state node is calculated, which is determined by the symptom improvement speed, vital sign stability degree and treatment method effectiveness. For example, in a certain acute asthma attack self-help scene, node A represents the "use bronchodilator inhalation" state, the symptom relief degree is 0.7 (full score is 1), the vital sign improvement degree is 0.6 (full score is 1), and the treatment response time is 3 minutes, and the accessibility score is 0.75. Compare the accessibility score with the preset accessibility threshold (such as 0.6), select the node with a score higher than the threshold as the next search node, and continuously search to generate a plurality of initial candidate self-help guidance paths.

[0093] For each generated initial candidate self-rescue guidance path, three aspects of constraint evaluation are performed, and the total amount of resource consumption is calculated, including drug consumption, medical device usage, physiological energy consumption, etc. Taking a certain asthma self-rescue path as an example, the path contains 5 nodes in turn: "use bronchodilator inhalation", "take a sitting position and lean forward", "perform pursed-lip breathing", "use bronchodilator inhalation again", "keep calm and monitor symptoms". The drug consumption of each node is [1 dose, 0, 0, 1 dose, 0] units respectively, so the total amount of bronchodilator consumption of the path is 2 doses. If the preset upper limit of bronchodilator usage is 3 doses, the path is feasible in terms of drug consumption constraint. The total execution time is calculated, including the execution time of each state node and the transition delay time between adjacent nodes. Taking the above path as an example, the execution time of the 5 nodes is [2, 5, 8, 2, 10] minutes respectively, and the node transition delay is [1, 1, 2, 1] minutes respectively. The total execution time is 27+5=32 minutes, and the upper limit of time consumption is 45 minutes, so the path is also feasible in terms of time constraint. The environmental adaptability score is calculated based on the current environment, including the comprehensive evaluation of factors such as whether there is assistance from others, whether there is access to drugs, and whether the space conditions are suitable, etc. For example, if the current environment is a solitary state, the bronchodilator is carried on the body, and it is in a well-ventilated indoor environment, the environmental adaptability score of the path is 0.85 (full score is 1), and the environmental adaptability threshold is set to 0.6, so the path is also feasible in terms of environmental constraint.

[0094] Through the above three constraint evaluations, any candidate path whose indicators exceed the violation limit threshold is marked as an infeasible path, thereby obtaining multiple feasible candidate self-rescue guidance paths. To further optimize path selection, the rescue action similarity between these feasible candidate paths is calculated. The action similarity is determined by comparing the proportion of similar self-rescue action sequences in different paths. For example, path A contains the action sequence [use bronchodilator inhalation, take a sitting position and lean forward, perform pursed-lip breathing, use bronchodilator inhalation again, keep calm and monitor symptoms], and path B contains the action sequence [use bronchodilator inhalation, take a sitting position and lean forward, perform abdominal breathing, use bronchodilator inhalation again, keep calm and monitor symptoms]. They have 4 similar actions, and the similarity is 4 / 5=0.8. If the path merging threshold is set to 0.7, the two paths can be merged, and the merging process will retain the action selection with better symptom relief effect, lower execution difficulty or less resource consumption. For example, in the above example, the system will compare the indicators of the two difference actions "pursed-lip breathing" and "abdominal breathing". Assuming that "pursed-lip breathing" is slightly more complex to perform but has a more significant effect on relieving asthma symptoms, the "pursed-lip breathing" action is selected, and the merged optimal path is formed.

[0095] By path merging, several optimized candidate self-rescue guidance paths are obtained, which not only meet the upper limit of resource consumption, the upper limit of time consumption and the environmental adaptability constraint, but also are optimized by merging similar paths to reduce redundant choices. For example, in the process of generating a self-rescue guidance scheme for acute hypoglycemia, the system prunes 8 feasible paths from the initial 15 candidate paths after three constraint evaluations, and finally obtains 3 optimized self-rescue guidance paths with obvious differences and respective advantages through similarity calculation and path merging: the first path focuses on "rapid glucose supplementation → sitting rest → rechecking blood glucose → eating complex carbohydrates", which is suitable for patients with a blood glucose meter and clear consciousness; the second path focuses on "oral glucose tablets or sugary drinks → supine position → seeking help from others → eating complex carbohydrates", which is suitable for patients without a blood glucose meter but with others present; and the third path focuses on "using a pre-filled glucose syringe → lateral recumbent position → calling an emergency number → waiting for professional rescue", which is suitable for severe patients with blurred consciousness and alone. Such differentiated path selection greatly improves the pertinence and effectiveness of the self-rescue guidance scheme, and can provide the most suitable self-rescue guidance according to the specific situation and environmental conditions of the patient.

[0096] In actual application, the thresholds are dynamically adjusted according to different disease scenarios. For example, in an emergency, the upper limit of time consumption can be appropriately relaxed; in a resource shortage situation, the resource consumption threshold can be reduced; and in an extreme environment, the environmental adaptability requirement can be increased. Such dynamic adjustment mechanism makes the rescue path calculation more flexible and practical, and can cope with various complex rescue scenarios.

[0097] In an optional implementation, based on the current illness of the user and the user location, a matching medical resource demand is retrieved, a plurality of candidate medical institutions are determined, a multi-objective optimization function is constructed in combination with the passing state of each road segment at the current time, an ant colony algorithm is used for solving, and an optimal transfer path is obtained, including:

[0098] Based on the current illness of the user, the importance weight of each type of medical resource is calculated, and in combination with the standard configuration demand amount of each type of medical resource, candidate medical institutions that meet the basic demand are screened out;

[0099] The bed occupancy, specialist on-duty situation and equipment usage of the candidate medical institutions are obtained, the resource availability score is calculated, and based on the resource availability score and the transfer distance, a comprehensive score of each candidate medical institution is obtained;

[0100] The road congestion intensity is calculated according to the road traffic volume data and the average speed data, the real-time passing time of each road segment is predicted in combination with the signal timing data, a multi-objective optimization function is constructed based on the comprehensive score and the real-time passing time, and the maximum allowable transfer time is set;

[0101] A plurality of search units are arranged in a road grid, a transfer path is determined based on path pheromone concentration and dynamic heuristic information, a pheromone update amount is adjusted according to the multi-objective optimization function, and an optimal transfer path is obtained through iterative optimization.

[0102] In the specific embodiment, the current disease information of the user is obtained, such as "acute ST segment elevation myocardial infarction", and the current geographic location coordinates of the user are obtained through GPS or mobile network positioning. For different diseases, a medical resource weight database is constructed, which records the dependence of various diseases on different medical resources. Medical resources are divided into four categories: equipment resources, drug resources, medical personnel resources, and specialist resources. Each category is further divided into multiple specific projects. For example, equipment resources include electrocardiogram monitors, defibrillators, ventilators, etc.; drug resources include thrombolytic drugs, anti-platelet drugs, nitrate ester drugs, etc.; medical personnel resources include cardiologists, intensive care physicians, cardiac intervention specialists, etc.; specialist resources include cardiac catheterization rooms, coronary care units, etc.

[0103] For the above "acute ST segment elevation myocardial infarction", relevant medical resources and their initial importance values are retrieved from the database. For example, the initial importance of the cardiac catheterization room is 0.95, the coronary care unit is 0.90, the electrocardiogram monitor is 0.85, the defibrillator is 0.80, the thrombolytic drug is 0.75, the anti-platelet drug is 0.70, the cardiologist is 0.90, and the cardiac intervention specialist is 0.95. The initial importance values are based on a large amount of clinical data and expert consensus, reflecting the basic importance of different medical resources for treating the disease. These initial importance values are dynamically adjusted according to the user's specific situation, including the user's age, complication status, past medical history, and time of onset. The adjustment mechanism uses a weighted correction method and is realized through a series of pre-set condition judgment rules. For example, when the user's age is greater than 75 years old, the importance of thrombolytic drugs will be reduced by 0.15, and when the time of onset is more than 12 hours, the importance of the cardiac catheterization room will be reduced by 0.25. In this example, assuming that the user is 62 years old, has a history of hypertension, and the time of onset is about 2 hours, the adjusted importance weights of various medical resources are calculated as follows: cardiac catheterization room 0.98, coronary care unit 0.92, electrocardiogram monitor 0.85, defibrillator 0.82, thrombolytic drug 0.72, anti-platelet drug 0.75, cardiologist 0.90, and cardiac intervention specialist 0.98.

[0104] After the medical resource importance weight calculation is completed, enter the medical resource standard configuration demand determination stage, retrieve the standard configuration demand amount of each type of medical resource required for treating "acute ST segment elevation myocardial infarction" from the medical specification database. The standard configuration demand amount represents the minimum resource configuration standard required for treating the disease and is formulated according to international medical guidelines and clinical practice experience. In this example, the retrieved standard configuration demand amount includes: at least 1 cardiac catheterization room, at least 1 bed in the coronary heart disease intensive care unit, at least 1 electrocardiogram monitor, at least 1 defibrillator, at least 3 common categories of thrombolytic drugs, at least 2 common categories of antiplatelet drugs, at least 2 cardiologists, at least 1 cardiac intervention specialist, etc.

[0105] Combine the medical resource importance weight with the standard configuration demand amount, center on the user's location, set a search radius of 30 kilometers, extract the resource configuration information of the surrounding medical institutions from the medical institution resource database, retrieve medical institutions that meet the basic resource configuration requirements, and form a list of candidate medical institutions.

[0106] Obtain real-time resource state data of the candidate medical institutions through the hospital information system interface, including specialist bed occupancy rate, related specialist on-duty status, and key medical equipment usage status. Take A Hospital as an example, the data shows that the cardiology department bed occupancy rate is 78%, there is 1 idle cardiac catheterization room, there are 3 on-duty cardiology specialists, there are 3 idle ICU beds, and the resource availability score is generated using a weighted calculation method, such as the resource availability score of A Hospital is 0.82. Calculate the straight-line distance from the user's location to each candidate medical institution and perform standardization processing, for example, the user's distance to A Hospital is 8.5 kilometers, and the standardized distance score is 0.72. Perform comprehensive scoring calculation on the resource availability score and the distance score, with a resource availability weight of 0.7 and a distance weight of 0.3, and obtain a comprehensive score of A Hospital of 0.79.

[0107] Obtain real-time road traffic data through the traffic monitoring network and road condition information service, take road segment "R1" as an example, the current traffic volume is 320 vehicles / hour, the average speed on this road segment is 35 km / h, and the congestion intensity is calculated to be 0.65 according to the preset congestion evaluation model, and the traffic signal timing data is obtained, such as the red light duration of intersection "J1" is 60 seconds and the green light duration is 40 seconds, combined with the traffic volume and average speed, the real-time travel time through this road segment is predicted to be 12 minutes, and similar calculations are performed on all major road segments in the region to construct a real-time road condition network model.

[0108] Construct a multi-objective optimization function, including maximizing the comprehensive score of medical institutions and minimizing the transfer time, and set the maximum allowable transfer time to 30 minutes, with a weight of 0.6 for the comprehensive score of medical institutions and a weight of 0.4 for the transfer time in the optimization function.

[0109] The road is divided into a plurality of grids, a search unit is arranged in the road grid, each unit represents an intersection or a road segment node, the ant colony algorithm parameters are initialized, including the number of ants is set to 50, the pheromone evaporation coefficient is set to 0.5, the iteration number is set to 100, each ant starts from the user position, and the next search unit is selected according to the path pheromone concentration and dynamic heuristic information, and the heuristic information is determined by the reciprocal of the real-time traffic time of the road segment.

[0110] After each ant completes the path search, the path score is calculated according to the multi-objective optimization function, and the pheromone update amount is adjusted accordingly, and the path with a high score obtains more pheromone enhancement, for example, the path score of a path leading to A hospital is 0.85, and the pheromone increment is 0.17, through multiple iterations, the optimal transfer path scheme is gradually converged.

[0111] In an actual case, a user reports a "severe chest pain" symptom, located at coordinates (119.9876°E, 30.1234°N), the current illness is acute myocardial infarction, and the medical resources with higher priority are the cardiac catheterization room and the cardiology expert. Five candidate medical institutions are searched, and their resource availability scores are 0.82, 0.75, 0.68, 0.79 and 0.71 respectively. Combined with the distance factor, the comprehensive scores are 0.79, 0.80, 0.65, 0.82, 0.66, and multiple transfer paths are calculated. After 50 iterations of the ant colony algorithm, the optimal path is determined as "start point → road R5 → road R8 → road R12 → A hospital", and the estimated transfer time is 18 minutes, passing through 3 traffic lights, and the total distance is 9.7 kilometers.

[0112] The final transfer scheme is displayed to the user or ambulance driver through a mobile application, including the transfer route, the estimated arrival time, the basic information of the target hospital and the contact information of the expert, and at the same time, the patient's estimated arrival notification is sent to the target hospital, so that the medical team can prepare for the reception work in advance.

[0113] In an alternative embodiment, a plurality of search units are arranged in the road grid, the transfer path is determined based on the path pheromone concentration and the heuristic strength, the pheromone update amount is adjusted according to the multi-objective optimization function, and the optimal transfer path is obtained by iterative optimization, comprising:

[0114] A plurality of search units are randomly deployed in the road grid, each search unit contains the current position, the moving direction and the visited path, the heuristic strength of each search unit is calculated by combining the historical heuristic strength and the real-time traffic time of each road segment;

[0115] Randomly placing initial pheromone concentration in the road grid, taking the product of the heuristic strength and the initial pheromone concentration as the path selection probability of the search unit, determining the transfer direction of the next step based on the path selection probability, and obtaining the transfer path formed by each search unit in the current iteration period;

[0116] Calculating the multi-objective optimization function value of each transfer path, mapping it as a pheromone increment, and updating the pheromone concentration in the road grid according to the pheromone evaporation rate;

[0117] Calculating the distribution entropy of the multi-objective optimization function value, and when the distribution entropy change of continuous multiple iterations is less than the entropy change threshold, selecting the transfer path with the maximum multi-objective function value from the transfer paths in the current iteration period as the optimal transfer path.

[0118] In the road grid, a plurality of search units are randomly deployed, for example, 50 search units are deployed in a 100x100 city road grid, each search unit contains three basic attributes: current position coordinates (x, y), selectable moving direction (up, down, left, right) and visited path record set P, the standard deviation of historical travel time is calculated for each road segment as the basis value of heuristic strength, for example, the average travel time of a road segment is 15 minutes during peak hours, and the standard deviation is 3 minutes, the average travel time is 5 minutes during off-peak hours, and the standard deviation is 1 minute, then the basic heuristic strength of the road segment can be set to 3 / 15=0.2. At the same time, the current traffic conditions of each road segment are collected in real time, for example, the current travel time of a road segment is 10 minutes, then the real-time heuristic strength of the road segment is 15-10=5 minutes. The historical heuristic strength and the real-time heuristic strength are combined according to the weight of 0.4:0.6 to obtain the comprehensive heuristic strength value of the road segment.

[0119] Randomly placing initial pheromone concentration in the road grid, the initial concentration range is set to a random value between 0.1 and 0.5, taking the current position of a search unit as (50, 50) for example, the road segments in its four selectable directions have the following pheromone concentrations: up (0.2), down (0.3), left (0.4), right (0.1), and the corresponding heuristic strengths are: up (0.6), down (0.4), left (0.5), right (0.8). Calculate the path selection probability of each direction, that is, the product of the pheromone concentration and the heuristic strength, to get: up (0.12), down (0.12), left (0.2), right (0.08). Normalize these product values to probability distribution: [0.23, 0.23, 0.38, 0.16], use roulette selection method to generate a random number between 0 and 1, for example, 0.3, then the search unit selects the left direction as the next moving direction. Repeat this process until each search unit reaches the destination from the starting point to form a complete transfer path.

[0120] The transport paths formed by all search units in the current iteration period are evaluated, and the multi-objective optimization function value of each path is calculated. Taking a path as an example, the comprehensive score of the end hospital is 0.82, and the expected travel time is 40 minutes. After weighted summation, the multi-objective optimization function value of the path is 0.75.

[0121] According to the multi-objective optimization function value, the pheromone increment of each path is calculated. For example, the path with an optimization function value of 0.75 has a pheromone increment of 0.75 multiplied by the pheromone enhancement factor 0.1, which is 0.075. The system updates the pheromone concentration in the road grid according to the preset pheromone evaporation rate (e.g. 0.2). For example, the current pheromone concentration of a road segment is 0.4, and the pheromone increment brought by the transport path passing through the road segment is 0.075. Therefore, the updated pheromone concentration of the road segment is 0.4*(1-0.2)+0.075=0.395.

[0122] After each iteration is completed, the distribution entropy of the multi-objective optimization function value of all transport paths in the current iteration period is calculated. Assuming that 50 transport paths are generated in the current iteration period, the multi-objective optimization function values are divided into 10 intervals, and the number of paths in each interval is counted, such as [3, 5, 8, 12, 10, 7, 3, 1, 1, 0]. According to this distribution, the information entropy is calculated, and the distribution entropy value of the current iteration period is 2.05. The distribution entropy values of the previous three iterations are 2.15, 2.08, and 2.05, respectively. The entropy change of the continuous iterations is 0.07 and 0.03. When the entropy change of the continuous 10 iterations is less than the preset entropy change threshold 0.05, it is considered that the search process has converged, and the iteration is stopped.

[0123] After the algorithm converges, the transport path with the maximum multi-objective function value is selected from the transport path set of the current iteration period as the optimal transport path. For example, the path with the maximum multi-objective function value in the current iteration period is 0.82, and the corresponding specific path is from the starting point (10, 10) to the end point A hospital (25, 25) through the nodes (12, 10), (12, 15), (20, 15), etc., with a total length of 23 kilometers and an expected travel time of 35 minutes. This path is taken as the final output optimal transport path and provided to the dispatching system for execution.

[0124] In practical application scenarios, this method can quickly calculate the optimal transport path that balances time, distance, and hospital matching degree according to the dynamic changes of the road network and the real-time adjustment of the transport demand. Tests show that compared with traditional single-objective path planning algorithms, this method can save 15% of the transport time on average and reduce 20% of the transport distance, making it particularly suitable for scenarios such as emergency rescue vehicle dispatching that require consideration of multiple optimization objectives.

[0125] The application also provides a specific embodiment of extracting features and converting formats of the optimal transport path and the self-rescue guidance scheme to generate a data format suitable for the input of the federated learning model. For the optimal transport path, key features such as path length, transport time, road condition, and transport risk are extracted. For the self-rescue guidance scheme, key features such as execution difficulty, resource demand, and expected effect of each step are extracted.

[0126] The federated learning model is composed of multiple local models distributed among different participants, and the collaborative training and reasoning are achieved through a secure aggregation mechanism. In the medical rescue scenario, typical institutions participating in federated learning include hospitals, emergency centers, traffic management departments, and meteorological service centers. Each participant trains a rescue decision-making model in a specific field based on their own professional field and data resources. For example, the local model of a hospital focuses on medical resource scheduling and disease assessment; the local model of an emergency center focuses on ambulance scheduling and on-site first aid; the local model of a traffic management department focuses on road condition analysis and transport time prediction; and the local model of a meteorological service center focuses on the assessment of the impact of weather on rescue operations.

[0127] The federated learning system adopts a horizontal federated learning architecture, and the local models of each participant have the same structure but different training data. In this embodiment, each local model adopts a deep neural network structure, including an input layer, multiple hidden layers, and an output layer. The input layer receives the feature vectors of the optimal transport path and the self-rescue guidance scheme; the hidden layers extract high-level features through nonlinear transformation; and the output layer generates a comprehensive decision score and decision recommendation. During training, each participant trains the model using local data, only exchanging model parameter gradients without sharing raw data, protecting data security while achieving collaborative learning. The model training adopts a federated averaging algorithm, and the central server periodically collects the model parameter gradients uploaded by each participant, performs weighted averaging, updates the global model, and distributes the updated global model parameters to each participant, achieving iterative optimization of the model.

[0128] To ensure the security of the federated learning process, multiple technical measures are adopted. Random noise is added to the model parameter gradients to prevent sensitive information from being leaked; secure multi-party computation allows participants to jointly compute functions without revealing their inputs; homomorphic encryption allows computation on encrypted data to ensure data security during transmission; and blockchain technology records model updates and participation, ensuring transparency and traceability. In this example, random noise conforming to the Laplace distribution is added to the model parameter gradients, with a noise intensity parameter ε set to 0.5, ensuring model accuracy while protecting data security.

[0129] The converted optimal transport path feature vector [23, 0.85, 0.15, 4] and the self-rescue guidance scheme feature vector [0.3, 0.2, 0.7, 0.8] are input into the trained federated learning model, and the forward propagation calculation is performed locally at each participant to generate local inference results. The hospital model generates the result [0.75, "recommend priority implementation of self-rescue measures and wait for professional rescue"], the emergency center model generates the result [0.82, "recommend immediate start of transport program"], the traffic department model generates the result [0.78, "transport path is smooth, recommend transport as soon as possible"], and the weather center model generates the result [0.7, "weather conditions are good, no need to adjust transport plan"]. These local inference results are securely transmitted to the central server for comprehensive analysis.

[0130] The central server integrates the inference results of each local model using a weighted voting mechanism, with weight distribution based on the professional relevance and historical accuracy of each participant. In this case, the hospital model weight is 0.4, the emergency center model weight is 0.3, the traffic department model weight is 0.2, and the weather center model weight is 0.1. Through weighted voting, the comprehensive decision score is 0.772 and the comprehensive decision suggestion is "consider patient condition and transport conditions, recommend immediate administration of nitroglycerin tablets and adopt prone position, and start transport program. Monitor symptoms during transport and maintain smooth communication".

[0131] According to the comprehensive decision result, a specific execution schedule is generated, which clearly indicates the execution time points of each self-rescue measure and transport step. In this case, the execution schedule is "0 minutes: administer nitroglycerin tablets; 1 minute: adopt prone position and loosen tight clothing; 2 minutes: call emergency services; 5 minutes: ambulance arrives and transport begins; 28 minutes: estimated arrival at medical institution". This time arrangement takes into account the urgency of self-rescue measures and the time prediction of transport paths, ensuring that rescue operations proceed in an orderly manner.

[0132] The generated comprehensive decision content is formatted to be suitable for display on user terminals. For smartphones and other mobile terminals, multimedia content containing text descriptions, icon indications, and voice prompts is generated. For smartwatches and other wearable devices, simplified content is generated, highlighting key steps and time points. For car-mounted devices, content focused on transport paths and navigation is generated. In this case, for the user's smartphone, a comprehensive content package containing self-rescue step illustrations, transport paths, time reminders, and voice guidance is generated.

[0133] The integrated decision content is sent to the user terminal through an encrypted communication channel, and according to network conditions, the most suitable transmission mode is automatically selected, such as a 4G / 5G network, Wi-Fi or a short message channel, in order to ensure transmission reliability, breakpoint resume and data compression techniques are used, and a degradation scheme for weak network environments is designed to ensure that core rescue information can still be transmitted under poor network conditions. In this case, the integrated decision content is successfully sent to the user's smartphone through the 4G network, and the transmission takes 0.8 seconds. The user terminal immediately displays the self-rescue guidance interface and plays a voice prompt.

[0134] Through the above method, the optimal transport path and self-rescue guidance scheme are successfully input into the federated learning model, obtaining an integrated decision that takes into account the medical professionalism, emergency timeliness, traffic practicality and environmental adaptability, and safely and efficiently sending it to the user terminal, providing an effective solution for intelligent rescue decision-making in emergency situations.

[0135] The application also provides a specific embodiment. User A is driving a vehicle on a highway and encounters a traffic accident. The vehicle collides with a guardrail, and user A's chest is hit, suspected of rib fracture and accompanied by difficulty breathing. The incident occurred at K127+500 on the highway, and there were no other vehicles or personnel nearby. User A sent an emergency help message through the vehicle-mounted intelligent terminal: "My car hit the guardrail, my chest hurts, I have difficulty breathing, the location is K127+500, and I request rescue." After receiving the emergency help message, the rescue process is immediately started.

[0136] Based on the emergency help information of user A, the keywords "chest pain" and "difficulty breathing" are extracted, and combined with the physiological data uploaded by the vehicle-mounted intelligent terminal, including heart rate 95 times / min, blood pressure 135 / 85 mmHg, respiratory rate 24 times / min, and blood oxygen saturation 94%, the data is analyzed by a symptom recognition algorithm to identify the current condition of user A as "closed rib fracture with mild pneumothorax" with a confidence level of 0.87. In the emergency rescue knowledge graph, "closed rib fracture with mild pneumothorax" is taken as the starting point, and "stable vital signs" is taken as the end point, a path search is performed, a potential rescue action sequence is generated, and combined with the on-site environmental information, including user alone, highway environment, temperature 28℃, and straight-line distance to the nearest medical institution 15 kilometers, a self-rescue guidance scheme is generated: maintain a semi-recumbent position, loosen tight clothing; take slow and shallow breaths, avoid deep breaths and violent coughing; gently press the injured side of the chest with a soft cushion or clothing; keep the car ventilated to avoid high temperatures; wait for professional rescue to arrive and avoid moving on your own. The execution difficulty score of the self-rescue scheme is 0.35 (full score is 1), the expected stability score is 0.82 (full score is 1), and the resource demand score is 0.2 (full score is 1).

[0137] Based on the current condition of user A "closed rib fracture with mild pneumothorax" and location information "highway K127+500", the matching medical resource demand is retrieved, and the importance weight of each type of medical resource is calculated for this condition: thoracic surgeon 0.85, chest CT device 0.90, thoracentesis device 0.65, respiratory support device 0.75, and pain management drug 0.70. From the medical institution database, 4 medical institutions within a 30-kilometer range from user A are retrieved: medical institution M1 is 18 kilometers away, has a complete thoracic surgery medical team and equipment; medical institution M2 is 12 kilometers away, has basic equipment but no professional thoracic surgery team; medical institution M3 is 25 kilometers away, has a thoracic surgery team but the CT device is under maintenance; medical institution M4 is 15 kilometers away, is a primary medical center with limited equipment and specialist resources. The system calculates the resource matching degrees of each medical institution as 0.92, 0.65, 0.78, and 0.45 respectively, and determines M1, M2, and M3 as candidate medical institutions.

[0138] Combined with real-time traffic information, a transfer path network from user A's location to each candidate medical institution is constructed. The traffic information shows that there is a small traffic jam near the highway exit leading to M1, with an average speed of 35 kilometers per hour; the road leading to M2 is smooth but involves multiple turns; the highway section leading to M3 is under road maintenance and is one-way. A multi-objective optimization function is constructed, considering factors such as transfer time, traffic risk, medical resource matching degree, and treatment timeliness. Ant colony algorithm is used for path optimization, with iteration number set to 100, ant number set to 20, pheromone importance factor set to 1.0, heuristic factor set to 2.0, and pheromone evaporation coefficient set to 0.5. After solving, the system obtains the optimal transfer path as "current location → highway K132 exit → county road X103 → medical institution M1", with an estimated transfer time of 24 minutes, a traffic risk score of 0.28 (full score 1), and a comprehensive score of 0.85 (full score 1).

[0139] The optimal transport path and self-rescue guidance scheme are input into a federated learning model, which is trained by 5 participants: a general hospital model, an emergency center model, a traffic management department model, a meteorological service center model, and a vehicle rescue service model. Based on their respective professional fields and data resources, each participant trains a professional model for traffic accident rescue, and collaborates in reasoning through the federated learning framework. The transport path feature vector [24, 0.28, 0.92, 18] and the self-rescue scheme feature vector [0.35, 0.2, 0.82, 0.75] are input into the federated learning model, and each participant's local model generates a local reasoning result and securely transmits it to the central server. The general hospital model suggests "giving priority to professional transport, while fixing the chest"; the emergency center model suggests "immediately starting ambulance transport, maintaining a semi-recumbent position"; the traffic management department model suggests "choosing the K132 exit transport route to avoid congested road sections"; the meteorological service center model points out "good weather, no precipitation, suitable for air rescue"; and the vehicle rescue service model suggests "waiting for professional rescue to arrive, avoiding moving the injured person on your own". The central server generates a comprehensive decision through a weighted voting mechanism: "considering the risk of rib fracture with mild pneumothorax and the professional thoracic surgery resources of M1 hospital, it is recommended to immediately transport to M1 hospital through the K132 exit, maintaining a semi-recumbent position and performing simple chest fixation before transport, while considering starting air rescue assessment, and expecting a rescue helicopter to arrive at the scene within 15 minutes, or an ambulance to arrive within 20 minutes and complete the transport within 24 minutes."

[0140] The comprehensive decision is sent to the user terminal through multiple channels, and the user A's vehicle-mounted intelligent terminal immediately displays the self-rescue guidance interface, including a semi-recumbent position diagram, a chest fixation method diagram, and voice prompts; at the same time, the transport path is sent to the vehicle navigation system, and the user's situation and location information are pushed to the nearest rescue unit through the emergency communication network, and rescue progress information is sent to user A: "ambulance has been dispatched, expected to arrive in 20 minutes; rescue helicopter is being evaluated, may arrive in 15 minutes; please keep the communication smooth and operate according to the self-rescue guidance."

[0141] Through the above method, the present application realizes intelligent processing of the whole process from emergency help to comprehensive rescue decision-making in the traffic accident scene, effectively integrates medical knowledge, path planning, and multi-party collaborative decision-making, provides timely and professional rescue services for users, and improves the timeliness and success rate of emergency rescue.

[0142] The intelligent transport and self-rescue guidance auxiliary decision-making system based on federated learning of the embodiment of the present application comprises:

[0143] The first unit is used for acquiring emergency help information sent by a user terminal;

[0144] The second unit is configured to extract knowledge entities and relationships from a medical first-aid knowledge base and a historical rescue case base, and construct an emergency rescue knowledge graph; based on the emergency help information, physiological characteristic parameters are extracted and the current illness of the user is identified, path search is performed in the emergency rescue knowledge graph, and a self-help guidance scheme is generated in combination with the on-site environment;

[0145] The third unit is configured to search for matched medical resource requirements based on the current illness of the user and the location of the user, determine a plurality of candidate medical institutions, construct a multi-objective optimization function in combination with the traffic conditions of each road segment at the current time, and obtain an optimal transfer path by using an ant colony algorithm to solve the multi-objective optimization function.

[0146] The fourth unit is configured to input the optimal transfer path and the self-help guidance scheme into a federated learning model to obtain a comprehensive decision, and send the comprehensive decision to a user terminal.

[0147] In a third aspect, an electronic device is provided, including:

[0148] a processor;

[0149] a memory for storing processor-executable instructions;

[0150] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0151] In a fourth aspect, a computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0152] The present application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions stored therein, which are used to perform various aspects of the present application.

[0153] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for intelligent transportation and self-rescue guidance and decision support based on federated learning, characterized in that, include: Obtain emergency help requests sent by user terminals; Knowledge entities and relationships are extracted from the medical emergency knowledge base and historical rescue case database to construct an emergency rescue knowledge graph, including: The medical emergency knowledge base retrieves treatment guidelines and emergency care instructions, generates structured medical knowledge text, constructs a medical domain word vector space, maps medical terms in the medical knowledge text to semantic vectors, extracts textual features from the semantic vectors, and identifies medical knowledge entities based on the textual features. The medical knowledge entities are divided into multiple entity categories according to a medical professional tree classification system. The correlation degree between the entity categories is calculated. Entity pairs with a correlation degree greater than the correlation threshold are marked as candidate relationship pairs. The candidate relationship pairs are screened by confidence evaluation to obtain the initial knowledge graph. New types of knowledge entities and relationships are identified from the historical rescue case database, and the semantic compatibility of the new entities and relationships with the existing knowledge in the initial knowledge graph is verified. The verified new entities and relationships are then integrated into the initial knowledge graph to obtain the emergency rescue knowledge graph. Based on the emergency help information, physiological characteristic parameters are extracted and the user's current symptoms are identified. Path search is performed in the emergency rescue knowledge graph, and a self-rescue guidance plan is generated in combination with the on-site environment. Based on the user's current symptoms and location, matching medical resource needs are retrieved, multiple candidate medical institutions are identified, and a multi-objective optimization function is constructed by combining the current traffic status of each road segment. This function is then solved using an ant colony optimization algorithm to obtain the optimal transfer path, including: Based on the user's current symptoms, the importance weights of various medical resources are calculated, and candidate medical institutions that meet the basic requirements are selected by combining the standard configuration requirements of various medical resources. The system obtains information on bed occupancy, expert availability, and equipment usage of the candidate medical institutions, calculates resource availability scores, and obtains a comprehensive score for each candidate medical institution based on the resource availability scores and transport distance. The congestion intensity of road segments is calculated based on road traffic flow data and average vehicle speed data, and the real-time travel time of each road segment is predicted by combining traffic light timing data; based on the comprehensive score and the real-time travel time, a multi-objective optimization function is constructed, and the maximum allowable transfer time is set. Multiple search units are set in the road grid. The transfer path is determined based on the path pheromone concentration and dynamic heuristic information. The pheromone update amount is adjusted according to the multi-objective optimization function, and the optimal transfer path is obtained through iterative optimization. The optimal transfer route and the self-rescue guidance plan are input into the federated learning model to obtain a comprehensive decision, which is then sent to the user terminal.

2. The method according to claim 1, characterized in that, Based on the emergency assistance information, physiological characteristic parameters are extracted and the user's current symptoms are identified. A path search is performed within the emergency rescue knowledge graph, and a self-rescue guidance plan is generated based on the on-site environment, including: From the emergency help information, extract the user's symptom feature vector and physiological feature data, perform time alignment and synchronization calibration on the physiological feature data, and generate physiological feature vector; Based on the symptom feature vector and the physiological feature vector, and combined with the symptom-disease association network in the emergency rescue knowledge graph, the probability of symptom identification of the user at the current moment is calculated, and the user's current symptom is determined. The current symptom is taken as the starting state node, the target recovery state is determined as the target state node, the initial candidate rescue path from the starting state node to the target state node is calculated, and the initial candidate rescue path is pruned based on the upper limit of resource consumption, the upper limit of time consumption and environmental adaptability constraints to obtain the optimized candidate rescue path. Calculate the feasibility score for each optimized candidate rescue path, select the candidate rescue path with the highest feasibility score, decompose it according to the execution steps, identify the target rescue action, and obtain a self-rescue guidance plan.

3. The method according to claim 2, characterized in that, Calculate the initial candidate rescue path from the starting state node to the target state node. Prune the initial candidate rescue path based on resource consumption limits, time consumption limits, and environmental adaptability constraints to obtain an optimized candidate rescue path, including: Starting from the initial state node, the adjacent state nodes are traversed sequentially through depth search. The node feature vectors of the adjacent state nodes are extracted, and their reachability scores to the target state node are calculated. The adjacent state nodes with reachability scores greater than the reachability threshold are selected as the next search nodes, generating multiple initial candidate rescue paths. For each initial candidate rescue path, calculate the total resource consumption of all its state nodes, and calculate the execution time and transition delay time of each state node to obtain the total execution time. Based on the current environment, calculate the environment adaptability score. When any one of the indicators of the total resource consumption, the total execution time, and the environment adaptability score exceeds the default limit threshold, mark the candidate rescue path as an infeasible path to obtain multiple feasible candidate rescue paths. Calculate the similarity of rescue actions among the feasible candidate rescue paths, and merge the rescue paths with a similarity of rescue actions higher than the path merging threshold to obtain the optimized candidate rescue paths.

4. The method according to claim 1, characterized in that, Multiple search units are set up in the road grid. The transport path is determined based on the path pheromone concentration and heuristic strength. The pheromone update amount is adjusted according to the multi-objective optimization function, and the optimal transport path is obtained through iterative optimization, including: Multiple search units are randomly deployed in the road grid. Each search unit contains the current location, direction of movement, and visited paths. The heuristic intensity of each search unit is calculated by combining the historical heuristic intensity and the real-time travel time of each road segment. An initial pheromone concentration is randomly distributed in the road grid. The product of the heuristic intensity and the initial pheromone concentration is used as the path selection probability of the search unit. The next transfer direction is determined based on the path selection probability, and the transfer path formed by each search unit in the current iteration cycle is obtained. Calculate the multi-objective optimization function value for each transport path, map it to the pheromone increment, and update the pheromone concentration in the road grid according to the pheromone evaporation rate; Calculate the distribution entropy of the multi-objective optimization function value. When the change in distribution entropy over multiple consecutive iterations is less than the entropy change threshold, select the transfer path with the largest multi-objective function value from the transfer paths in the current iteration cycle as the optimal transfer path.

5. A federated learning-based intelligent transfer and self-rescue guidance and decision support system, used to implement the method as described in any one of claims 1-4, characterized in that, include: The first unit is used to obtain emergency help information sent by user terminals; The second unit is used to extract knowledge entities and relationships from the medical emergency knowledge base and historical rescue case database to construct an emergency rescue knowledge graph. Based on the emergency help information, physiological characteristic parameters are extracted and the user's current symptoms are identified. Path search is performed in the emergency rescue knowledge graph, and a self-rescue guidance plan is generated in combination with the on-site environment. The third unit is used to retrieve matching medical resource needs based on the user's current symptoms and location, determine multiple candidate medical institutions, and construct a multi-objective optimization function by combining the current traffic status of each road segment, and solve it using the ant colony algorithm to obtain the optimal transfer path. The fourth unit is used to input the optimal transfer path and the self-rescue guidance plan into the federated learning model to obtain a comprehensive decision, and then send the comprehensive decision to the user terminal.

6. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 4.

Images