A nursing operation flow multi-terminal cooperation method and system

Abstract

The present application relates to the technical field of health care management, and particularly relates to a nursing operation flow multi-terminal collaboration method and system. The method comprises the following steps: obtaining nursing basic data, and performing nursing task semantic construction according to the nursing basic data to obtain nursing task ontology data; performing nursing operation flow chart construction according to the nursing task ontology data to obtain nursing operation flow chart data; obtaining a nursing terminal knowledge graph according to the nursing operation flow chart data; performing multi-agent collaboration model construction according to the nursing terminal knowledge graph to obtain a nursing collaboration model; performing task collaboration scheduling according to the nursing collaboration model to obtain nursing task collaboration data; obtaining historical nursing data, and performing task collaboration efficiency optimization on the nursing task collaboration data according to the historical nursing data to obtain nursing task optimization data. The present application is based on multi-agent collaboration scheduling and feedback optimization, and realizes more efficient, more intelligent and more adaptive nursing task management and execution control.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of health care management, and in particular to a nursing operation flow multi-terminal collaboration method and system. BACKGROUND

[0002] With the continuous rise of hospital nursing workload and the continuous increase of types of ward equipment, nursing tasks show the characteristics of high frequency, multiple types, strong dependence and cross-terminal execution. The traditional nursing scheduling system mainly dispatches tasks based on static process templates and manual scheduling strategies. And most of the current terminal scheduling systems only stay in the one-to-one mapping relationship of "task and equipment", and cannot support concurrent execution, collaborative cooperation or terminal replacement strategy of tasks. In the face of different ward layouts and clinical emergencies, it lacks dynamic scheduling and system recovery ability, which seriously restricts the continuity and flexible response level of nursing service. SUMMARY

[0003] The present application relates to the technical field of health care management, and in particular to a nursing operation flow multi-terminal collaboration method and system.

[0004] The present application provides a nursing operation flow multi-terminal collaboration method, which comprises:

[0005] S1, acquiring nursing basic data, and constructing nursing task semantics according to the nursing basic data to obtain nursing task ontology data; constructing a nursing operation flow chart according to the nursing task ontology data to obtain nursing operation flow chart data;

[0006] S2, constructing a nursing terminal knowledge graph according to the nursing operation flow chart data to obtain a nursing terminal knowledge graph; constructing a multi-agent collaboration model according to the nursing terminal knowledge graph to obtain a nursing collaboration model;

[0007] S3, performing task collaboration scheduling according to the nursing collaboration model to obtain nursing task collaboration data;

[0008] S4, acquiring historical nursing data, and optimizing the task collaboration efficiency of the nursing task collaboration data according to the historical nursing data to obtain nursing task optimization data.

[0009] The application can realize task dynamic decomposition, terminal capability matching and distributed intelligent scheduling for actual nursing scenes, and improve the real-time and accuracy of nursing tasks. The system uses nursing task ontology and job flow graph to ensure clear and controllable task logic dependence; through semantic abstraction and structured expression of execution resources by terminal knowledge graph, the collaborative adaptation capability between terminals is enhanced; combined with multi-agent model construction, self-organizing task negotiation of heterogeneous terminals in complex nursing situations is realized; at the same time, the historical data feedback mechanism is introduced, which can continuously optimize the scheduling efficiency and improve the robustness, scalability and intelligent level of the system.

[0010] Optionally, the nursing task semantic construction comprises:

[0011] Obtaining nursing basic data;

[0012] Performing task intention unit processing on the nursing basic data to obtain task intention unit data;

[0013] Performing nursing task semantic graph construction according to the task intention unit data to obtain nursing task semantic graph data;

[0014] Performing semantic clustering on the nursing task semantic graph data to obtain nursing task cluster graph data;

[0015] Performing nursing task ontology model construction on the nursing task cluster graph data to obtain nursing task ontology data.

[0016] In the application, by performing task intention unit processing and semantic graph modeling on the nursing basic data, the task expression structure with semantic logic can be extracted from the original record, and the explainability and hierarchical clarity of task modeling are improved. Task intention unit processing helps to decompose nursing behavior into basic action units with target orientation, so that task representation is more fine-grained; by constructing a nursing task semantic graph, the timing dependence, role relationship and target coupling structure between tasks can be preserved, and structured semantic organization is realized; by generating a task cluster graph through semantic clustering, high-frequency task patterns can be induced into standardized templates, which is conducive to unifying the task scheduling model; the constructed nursing task ontology model has semantic consistency and reasoning ability, and can provide a high-quality knowledge base for task scheduling, terminal matching and optimization.

[0017] Optionally, the nursing job flow graph construction comprises:

[0018] Performing task ontology instantiation according to the nursing task ontology data to obtain task instance data;

[0019] Performing constraint mapping on the task instance data to obtain task constraint data, wherein the constraint mapping comprises role constraint mapping, physical constraint mapping, time constraint mapping and space constraint mapping;

[0020] The task constraint data is subjected to job task node construction, and job task node data is obtained;

[0021] The job task node data is subjected to division, and node division data is obtained;

[0022] The node division data is subjected to multi-dimensional dependent edge construction, and multi-dimensional dependent edge data is obtained;

[0023] The node division data and the multi-dimensional dependent edge data are subjected to job logic diagram generation, and nursing job flowchart data is obtained.

[0024] In the application, through instantiation, constraint mapping and structured graph modeling of the nursing task ontology, abstract task knowledge can be effectively converted into executable task flowchart, and the logical clarity and operation controllability of nursing task scheduling are significantly improved. The task ontology instantiation can generate specific task instances with strong pertinence in combination with the individual state of the patient, and the individualized adaptation ability of the model is enhanced. The multi-dimensional constraint mapping mechanism of role, physics, time and space ensures the rationality and executability of each task under specific resources and scenes. Through job task node construction and division, a node set with segmentation scheduling and load decomposition ability is formed. The multi-dimensional dependent edge constructed on this basis covers sequential logic, resource conflict and collaborative relationship, so that the generated job flowchart not only retains the causality of task execution, but also has the ability of collaborative scheduling and dynamic reconstruction.

[0025] Optionally, the nursing terminal knowledge graph construction comprises:

[0026] The nursing job flowchart data is subjected to terminal entity extraction, and terminal entity data is obtained;

[0027] The terminal entity data is subjected to terminal and task mapping relationship extraction, and task terminal association data is obtained;

[0028] The task terminal association data is subjected to function hierarchical classification tree construction, and function tree data is obtained;

[0029] The function tree data is subjected to ontology subgraph subnetwork construction, and the nursing terminal knowledge graph is obtained.

[0030] In the application, terminal entities involved in nursing operation flowchart data are extracted, semantic modeling and ontology atlas generation are performed, structured expression and semantic association of terminal resources are realized, and high-quality knowledge support is provided for multi-terminal collaborative scheduling. Terminal entity extraction ensures that the system comprehensively perceives available execution nodes in the ward, including fixed devices and mobile terminals. Task and terminal mapping relationship extraction helps to identify the functional role and service frequency of each terminal in different tasks, and enhances the semantic matching ability of resources and tasks. The function hierarchical classification tree construction classifies terminals according to service capabilities, which not only improves the accuracy of terminal scheduling, but also enhances the substitutability and collaboration between terminals. The constructed ontology subgraph subnet has semantic properties such as superior-inferior relationship, capability constraint and collaboration rule, so that the entire knowledge atlas has the characteristics of query, reasoning and expansion.

[0031] Optionally, the function hierarchical classification tree construction comprises:

[0032] According to the task-terminal association data, semantic labels are generated to obtain association semantic label data;

[0033] According to the association semantic label data, graph function clustering is performed to obtain association clustering data;

[0034] The association clustering data is subjected to function hierarchical classification structure tree construction to obtain preliminary function tree data;

[0035] According to the preliminary function tree data, environment perception adaptation is performed to obtain function tree data.

[0036] In the application, through semantic label generation, graph clustering analysis and environment adaptation mechanism, multi-type nursing terminals can be structured and classified and intelligent function abstraction can be performed, which significantly improves the semantic organization ability and adaptive scheduling efficiency of terminal resources. By generating semantic labels from task-terminal association data, the feature information of terminals in the dimensions of task response, execution frequency and function type can be accurately extracted, and the semantic recognition accuracy of terminals can be enhanced. Based on the semantic label data, graph function clustering is performed, which helps to discover the functional similarity and collaborative relationship between terminals, thereby realizing function role induction. The function hierarchical classification structure tree constructed by the system enables terminal resources to have multi-level abstract expression ability, supports cross-level task mapping and function inheritance. Finally, the environment perception adaptation mechanism is introduced, which can dynamically adjust the function tree structure according to the actual application scene (such as night, emergency, ICU ward), and realize the application scene self-adaptation of terminal capability.

[0037] Optionally, the ontology subgraph subnet construction comprises:

[0038] The function tree data is subjected to node ontology mapping to obtain node mapping data, wherein the node ontology mapping comprises capability ontology mapping, service area mapping and response time mapping;

[0039] The semantic relationship edge data is obtained by constructing semantic relationship edges on the node mapping data.

[0040] The ontology subgraph structure data is obtained by generating ontology subgraph structure according to the semantic relationship edge data.

[0041] The nursing terminal knowledge graph is obtained by adding perception activation to the ontology subgraph structure data.

[0042] In the application, through the function node semantic mapping, relationship edge modeling and perception activation mechanism, the terminal function hierarchical structure can be converted into a nursing terminal knowledge graph with reasoning, adaptability and reconfigurability, which significantly enhances the semantic expression ability and intelligent decision support of the system. Through node ontology mapping, the terminal can be semantically expressed in key attributes such as capability characteristics, service area coverage and response timeliness, forming high-quality ontology node corpus; the construction of semantic relationship edges clearly defines the function inheritance, collaborative service and replaceable path between terminals, so that the ontology structure has clear logical graph relationship expression ability; by generating ontology subgraph structure, the system can sub-network the graph according to function clusters or task domains, improving reasoning efficiency and modular management ability; the system performs perception activation, supports the system to dynamically activate or freeze part of the graph nodes according to real-time context (such as time period, ward type, equipment state), so that the knowledge graph has environmental adaptability and runtime controllability.

[0043] Optionally, the multi-agent collaborative model construction includes:

[0044] The agent entity data is obtained by extracting agent entities according to the nursing terminal knowledge graph.

[0045] The agent simulation data is obtained by simulating agents according to the agent entity data.

[0046] The nursing collaborative model is obtained by constructing structure perception collaborative modeling according to the agent simulation data.

[0047] In the application, by extracting agent entities from the nursing terminal knowledge graph, and combining the perception simulation and structural collaborative modeling mechanism, a nursing agent system with environment perception ability, behavior decision-making ability and collaborative execution ability can be effectively constructed, and the autonomous collaboration efficiency of multi-terminal resources and the flexibility of system scheduling are significantly improved. The agent entity extraction can convert the terminal device structure into an agent object with behavior ability, state attribute and communication interface, providing a unified expression basis for modeling. The perception simulation introduces multi-dimensional information such as position state, resource occupation and task allocation, realizes dynamic modeling and state prediction of the agent running scene, and enhances the response ability of the system to running risks and load changes. The structural perception collaborative modeling is based on the functional boundary, communication topology and task complementarity between agents, constructs the cooperation relationship network between agents, and enables the model to have the ability of flexible scheduling, resource reallocation and self-organization collaboration.

[0048] Optionally, S3 comprises:

[0049] According to the nursing collaborative model, the task-driven window is divided, and nursing task division data is obtained;

[0050] According to the nursing collaborative model and the nursing task division data, distributed task candidate matching is performed, and task candidate matching data is obtained;

[0051] According to the task candidate matching data, a collaborative scheduling graph is generated, and nursing task collaboration data is obtained.

[0052] In the application, through task-driven window division, candidate matching and collaborative graph generation, a high-precision, high-response and high-adaptability nursing task distribution strategy can be realized on the basis of the multi-agent nursing collaborative model, and the dynamic adaptability and resource utilization efficiency of the overall scheduling system are improved. The task-driven window division can dynamically generate an execution interval with time flexibility according to the task urgency, execution dependency and terminal state, and enhance the time flexibility and conflict avoidance ability of task scheduling. The distributed task candidate matching supports adaptive matching between tasks and terminals under multi-objective constraints by introducing an agent self-evaluation mechanism and local state perception, avoiding resource redundancy and inefficient assignment. The generation of the collaborative scheduling graph establishes a multi-dimensional dependence and collaborative execution path between task nodes and agent nodes, forming a computable and traceable scheduling graph structure.

[0053] Optionally, S4 comprises:

[0054] Obtain historical nursing data;

[0055] According to the historical nursing data, the historical nursing task trajectory of the nursing task collaboration data is reconstructed, and nursing task trajectory data is obtained;

[0056] Scheduling behavior extraction is performed on the nursing task trajectory data to obtain scheduling behavior data;

[0057] Scheduling efficiency calculation is performed according to the scheduling behavior data to obtain scheduling efficiency data;

[0058] Abnormal behavior attribution is performed according to the scheduling efficiency data to obtain scheduling bottleneck data;

[0059] Strategy graph updating is performed on the nursing task collaboration data according to the scheduling bottleneck data to obtain nursing task optimization data.

[0060] In the present application, through trajectory reconstruction driven by historical data, scheduling behavior analysis and strategy graph updating, feedback optimization and adaptive evolution of the nursing task collaboration mechanism are realized, and the continuous improvement capability and scheduling robustness of the system in the actual operation process are significantly improved. The historical nursing task trajectory reconstruction can accurately restore the actual execution path and state change of the task, construct the time and resource sequence of the real execution process, and provide data support for optimization; the scheduling behavior extraction can identify the key decision nodes, resource conflict points and collaboration failure conditions in the task execution process, and enhance the observability of the scheduling strategy execution quality; the scheduling efficiency calculation quantifies the scheduling system performance from the task response time, terminal utilization rate and successful execution rate, etc., forming a quantifiable evaluation index; the abnormal behavior attribution mechanism can locate specific bottleneck links (such as specific terminal congestion, specific task high failure rate, etc.), and provide targeted adjustment basis; through strategy graph updating, automatic correction of scheduling rules, matching logic and terminal preference weight is realized, so that the nursing scheduling system has continuous learning and structural adaptive ability.

[0061] Optionally, the present application also provides a nursing operation flow multi-terminal collaboration system for executing the nursing operation flow multi-terminal collaboration method as described above, the nursing operation flow multi-terminal collaboration system comprising:

[0062] A task ontology construction module is configured to obtain nursing basic data, and perform nursing task semantic construction according to the nursing basic data to obtain nursing task ontology data; and perform nursing operation flow graph construction according to the nursing task ontology data to obtain nursing operation flow graph data;

[0063] A terminal knowledge collaborative modeling module is configured to perform nursing terminal knowledge graph construction according to the nursing operation flow graph data to obtain a nursing terminal knowledge graph; and perform multi-agent collaborative model construction according to the nursing terminal knowledge graph to obtain a nursing collaboration model;

[0064] An intelligent scheduling execution module is configured to perform task collaboration scheduling according to the nursing collaboration model to obtain nursing task collaboration data;

[0065] The scheduling feedback optimization module is used for acquiring historical nursing data, and performing task coordination efficiency optimization on the nursing task coordination data according to the historical nursing data, to obtain nursing task optimization data.

[0066] The present application aims to effectively abstract the execution intention, behavior unit and dependency relationship of the task by constructing the semantics of the nursing basic data and generating the task ontology, to ensure that the task has clear structure and semantically interpretable expression ability in different scenarios; by constructing the job flow graph and combining the multi-dimensional dependency modeling of the task constraints, the calculability of the time sequence, space and resource relationship between tasks is realized, to provide logical support for the scheduling algorithm; by mapping the job graph to the terminal resource, constructing the terminal knowledge graph and extracting the multi-agent structure, the terminal perception and coordination ability are given, to realize the distributed response and cooperative scheduling of the heterogeneous terminal; at the scheduling level, based on the task window division and the agent capability matching mechanism, the scheduling graph structure is formed, to realize efficient and conflict-controllable task allocation; by reconstructing the nursing task trajectory from the historical data, evaluating the scheduling behavior and efficiency, and updating the scheduling strategy graph combined with the bottleneck attribution mechanism, the system has the ability of closed-loop optimization and long-term evolution. BRIEF DESCRIPTION OF DRAWINGS

[0067] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-restrictive embodiments made with reference to the attached drawings:

[0068] Figure 1 A step flow chart of a nursing job flow multi-terminal coordination method of an embodiment is shown;

[0069] Figure 2 A step flow chart of a nursing task semantic construction method of an embodiment is shown;

[0070] Figure 3 A step flow chart of a nursing terminal knowledge graph construction method of an embodiment is shown;

[0071] Figure 4 A step flow chart of a nursing task coordination method of an embodiment is shown;

[0072] Figure 5 A step flow chart of a nursing task coordination efficiency optimization method of an embodiment is shown;

[0073] The implementation of the present application, functional features and advantages will be further described with reference to the embodiments and the attached drawings. DETAILED DESCRIPTION

[0074] The technical method of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are 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 those skilled in the art without creative work fall within the scope of the present application.

[0075] In addition, the accompanying drawings are only schematic illustrations of the present application, and are not necessarily drawn to scale. The functional entities can be implemented in software, or in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0076] It should be understood that although the terms "first", "second" and the like can be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, without departing from the scope of the example embodiments, a first element can be called a second element, and similarly a second element can be called a first element. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0077] Referring to Figures 1 to 5 The present application provides a nursing operation flow multi-terminal cooperation method, which comprises the following steps:

[0078] S1, acquiring nursing basic data, and performing nursing task semantic construction according to the nursing basic data to obtain nursing task ontology data; performing nursing operation flow chart construction according to the nursing task ontology data to obtain nursing operation flow chart data;

[0079] In an embodiment, the nursing-related basic data is collected from multiple sources in a hospital information system, including an electronic medical record system (EHR) to obtain structured data such as patient status, nursing needs, task completion records, etc.; a nurse behavior record system to collect time series data such as nurse work logs and task execution times; a ward monitoring platform to obtain environmental information such as spatial position and equipment status; nursing documents and free text records, including nursing shift records, note texts and other unstructured data. The collected data covers but is not limited to dimensions such as task type, operation time, responsible person, trigger condition and ward location. For nursing task data, the task intent unit is extracted by behavior sequence segmentation. This process uses a combination of template matching and behavior frequency analysis to extract standardized intent units from behavior description texts. The extracted intent units are then used to construct a nursing operation flow chart, which is a directed graph composed of nodes and edges. Each node represents a nursing task, and each edge represents the execution sequence of the task. The nursing operation flow chart is used to guide the nursing process and improve the efficiency of nursing work. Figure ThreeThe tuple, structured as <action type, trigger condition, expected goal>, for example, extracts the standard triple from the original record "Observe and record surgical bleeding within 12 hours post-surgery": <Patrol, record surgical bleeding within 12 hours post-surgery>. This triple serves as a semantic unit in subsequent task modeling and reasoning. Based on the above task intent units, a task semantic graph structure is constructed. , where: node set Each node represents a nursing task, and each node carries the following attributes: task urgency, estimated time, and associated role; edge set. Representing relationships between tasks, including but not limited to temporal causal order (e.g., "a certain task must be performed first"); data dependencies (e.g., "this task depends on the output data of the previous task"); and mutual exclusion (e.g., "two tasks cannot be performed at the same time"). For each task node, four types of constraints are defined and bound, including role constraints (tasks can only be performed by nurses with specific qualifications or permissions, such as nurses with injection qualifications); time constraints (tasks must be completed within a specified time window, such as within 30 minutes post-operation); spatial constraints (tasks are limited to a specific physical area, such as only bed A7); and physical resource constraints (task execution requires specific auxiliary tools or equipment support, such as the need to configure a call button or mobile terminal). Based on the semantic graph and constraint mapping results, a nursing workflow graph structure is constructed.

[0080] S2. Construct a nursing terminal knowledge graph based on the nursing workflow data to obtain the nursing terminal knowledge graph; construct a multi-agent collaborative model based on the nursing terminal knowledge graph to obtain the nursing collaborative model;

[0081] In one embodiment, terminal devices related to task execution are identified in the target nursing area, and entities are extracted according to device category, including but not limited to mobile devices such as nurse-held handheld terminals (PDAs) and tablets; fixed devices such as smart medicine cabinets, automatic infusion pumps, and environmental controllers; and intelligent auxiliary devices such as voice interaction devices, bedside displays, and augmented reality (AR) headsets. Each type of terminal is considered a potential collaborative node and included in the knowledge graph construction. For each task node in the nursing workflow graph, a semantic mapping relationship between the task and the terminal is established. This mapping is based on the matching relationship between terminal capability characteristics and task requirements, generating the following structured association data: :express Required abilities for task i )Depend on The terminal j provides. In the capability matching process, the terminal is assigned a set of capability labels (such as a terminal with "vital sign collection" capability is labeled as "hasCapability: vitalscan"), and the task requirement is mapped to a target function item, and vector matching is performed between the two. Based on the terminal capability label data, the system performs clustering and hierarchical analysis, and constructs a functional classification tree structure of the terminal. The tree structure reflects the belonging relationship and hierarchical relationship of different terminals in the nursing task function space. The structure is as follows: first-level node: vital sign monitoring class; second-level node: temperature monitoring; third-level node: infrared sensing device, wearable temperature patch and other terminal types. Based on the above functional classification tree and terminal entity characteristics, a nursing terminal knowledge graph is constructed. The graph is a structure diagram containing nodes and edges, and its definition is as follows: node attributes include function type: indicating the core function category possessed by the terminal; response time: the average delay of the terminal in responding to the request; coverage area: the physical space range of the service capability of the terminal; state label: the current state of the terminal, such as "active", "faulty" and the like. Semantic edge types include isCompatibleWith: indicating that two terminals can cooperatively execute the same task; canReplace: indicating that one terminal can replace another terminal in the task; conflictsWith: indicating that two terminals exist conflict in space, time or resource, and cannot participate in the task at the same time. On the basis of the graph structure, each terminal is abstracted as an independent agent entity (Agent). The attributes of each agent include capability vector: indicating its capability characteristics in the task function space; spatial position: current position information, affecting its task accessibility; available time: its task execution available time period on the future time axis; energy consumption state: reflecting its power, resource consumption and other running parameters. Based on the above attribute information, each agent is equipped with an internal decision strategy, and the strategy includes a greedy score model (such as minimum task delay); or an adaptive optimization strategy based on reinforcement learning (such as dynamically adjusting the priority under task congestion). At the same time, the communication interface between the agents is established, and task broadcasting, feedback response and collaborative negotiation are realized.

[0082] S3, task coordination scheduling is performed according to the nursing coordination model, and nursing task coordination data is obtained;

[0083] In an embodiment, for each nursing task to be scheduled, the system first performs scheduling period division according to the time constraint parameters (such as the earliest executable time and the latest completion time) set by the task and the current available time window of the agent. If the task has a clear time boundary limit (for example, completed within 30 minutes after the operation), the scheduling time window is generated based on this division; if the agent has an idle available window within the period, the system marks it as a schedulable interval; for tasks with high priority, the system can perform dynamic contraction or queuing operations on the scheduling window to prioritize its execution requirements. This processing ensures that the scheduling plan is time-sequentially feasible and takes into account task priority. The system constructs a multi-factor scoring mechanism to evaluate the adaptation scores of all candidate agents. The scoring indicators include the ability matching degree (CapMatch), which measures whether the functional attributes of the agent cover the operation capabilities required by the task; the spatial distance penalty term (DistPenalty), which sets a penalty coefficient according to the spatial distance between the task execution location and the current location of the candidate terminal, the farther the distance, the lower the score; and the load balancing (LoadBalancing), which reflects the task pressure currently carried by the candidate agent, the higher the load of the terminal, the lower its matching priority. The adaptation score is calculated by weighting the above three factors, and the system can select the top-K based on the set threshold or score as the effective candidate agent list. Based on the candidate matching results, the system constructs a collaborative scheduling graph structure between the tasks and the terminals. The graph includes task nodes representing the current nursing tasks to be allocated, agent nodes representing the candidate terminals participating in scheduling, and matching edges (weighted edges) representing the existence of a feasible match between the task and the terminal, and the edge weight represents the matching confidence or priority. After the graph structure is generated, the system detects scheduling conflicts in the presence of resource conflicts (such as two tasks competing for the same terminal). Conflict processing can be completed in the following ways: prioritizing by edge weight, retaining the highest matching scheduling edge; attempting alternative path matching or delayed queuing for eliminated tasks; if the conflict cannot be resolved, the system marks the scheduling failure state and includes it in the next round of scheduling cycle. The system supports specifying alternative terminals for each task to deal with temporary failure or offline of the main scheduling terminal. The system outputs the scheduling results of the nursing tasks, including task identification, main scheduling agent, scheduled execution period, candidate backup agent list, and task scheduling priority, etc.

[0084] S4, obtain historical nursing data, and perform task coordination efficiency optimization on the nursing task coordination data according to the historical nursing data to obtain nursing task optimization data.

[0085] In an embodiment, the system fuses historical scheduling logs and actual terminal execution records to reconstruct the execution trajectory of a nursing task. Each trajectory record contains fields such as task identification, assigned terminal identification, actual start and end time, execution result identification, etc. If the system detects abnormal states such as delayed start, terminal response failure, task cancellation, etc. during task execution, it automatically tags the state. The system extracts event path patterns from task trajectory data and identifies the following key behavior types: normal path, task successfully completed, terminal response without abnormality; conflict and re-scheduling path, task successfully completed after re-scheduling after initial allocation failure; path interruption, task not completed, including terminal failure, task timeout cancellation, etc. Behavior events can be constructed as a graph structure or a time series structure. After clustering the task trajectories, the system calculates the following performance indicators: average response time, average time consumption from task creation to the first scheduling response; task success rate, the proportion of successfully completed tasks among all tasks; terminal utilization rate, the frequency of effective use of each type of terminal per unit time; collaboration density, the average number of collaborative terminals mobilized for each task. For scheduling failure or low efficiency scenarios, the system applies statistical frequency analysis, association rule mining and decision tree attribution algorithms to extract the following typical bottleneck patterns: the failure rate of a certain type of task on a specific terminal is significantly higher; task response delay significantly increases during night hours; there is a scheduling path congestion in a certain area or device network; high-priority tasks are inefficiently scheduled. These bottleneck labels will serve as the basis for strategy adjustment. Combined with the bottleneck analysis results, the system dynamically modifies the original scheduling graph, including adjusting the agent matching weight to preferentially allocate high-performing terminals; adjusting the edge weight in the scheduling graph to weaken the conflict path; modifying the default time window or execution priority of specific task types; caching successful paths in history as reusable scheduling templates and adding them to the strategy pool. The system generates a set of optimization suggestion data that can be used for the next round of scheduling, including recommending to extend the scheduling window of a certain type of task; assigning a certain type of task to preferentially allocate to a certain type of terminal; avoiding the use of some frequently failed terminal nodes, etc.

[0086] Optionally, the nursing task semantic construction comprises:

[0087] S11, acquiring nursing basic data;

[0088] In an embodiment, the system extracts information from an electronic medical record (EHR) system, including but not limited to the following: basic demographic information of the patient (such as age, gender, hospital number, etc.), clinical diagnosis results, treatment and medication plans, preoperative and postoperative records, medical history, nursing schedule, and other structured field data. For unstructured data such as nursing documents, shift records, free-text nursing logs, and medical communication records, the system uses natural language processing techniques for semantic analysis, including but not limited to Chinese word segmentation, named entity recognition, keyword extraction, and syntactic dependency analysis, to extract key information elements reflecting the dynamic changes of the patient's condition and nursing behavior.

[0089] S12, performing task intention unit processing on the nursing basic data to obtain task intention unit data;

[0090] In an embodiment, each nursing behavior record is considered as a potential nursing task segment, which is structured and expressed by task intention units. The unit is used to explicitly describe the operation matters that the nursing staff should complete under specific time or space conditions. Examples include monitoring body temperature, changing dressings, assisting with turning over, observing bleeding in the surgical area, etc. The system identifies key verbs (representing actions) and target objects (representing nursing targets) in tasks based on nursing text descriptions, combined with dependency syntax analysis and topic modeling algorithms (such as LDA), to achieve structured extraction of task semantics. For example, for the sentence "record blood pressure and mark abnormalities" in the nursing record, the system can identify two basic intention units, "record blood pressure" and "mark abnormalities", and further combine them into a composite nursing task with an internal logical sequence. The system combines the context information in the electronic medical record, including time constraints (such as "within 12 hours after surgery", "every morning and evening") and spatial constraints (such as "ward A area", "ICU bed 3"), to clearly define the execution time and location of the task. A mapping rule library is constructed between nursing behaviors and task intention units. Based on the standard actions in the nursing standard process and clinical path, a bidirectional matching relationship of "behavior sentence → intention unit" is formed. For example, "monitor body temperature every 4 hours within 48 hours after surgery" is mapped to the basic unit form: "monitor body temperature" as the task action, and "within 48 hours after surgery, every 4 hours" as the trigger condition. Each task intention unit is composed of the following elements: task action (Task), describing specific nursing behavior; trigger condition (Trigger), limiting the time or state prerequisite for task execution; responsibility role (Role), indicating the personnel role (such as nurse, responsible nurse, etc.) performing the task.

[0091] S13, constructing a nursing task semantic graph according to the task intention unit data to obtain nursing task semantic graph data;

[0092] In an embodiment, the nursing task semantic graph is a directed graph structure, where each node represents a nursing task intent unit, and the edges in the graph represent the relationships between tasks, including sequential execution, conditional constraints, data dependencies, or mutual exclusion, etc. The graph structure can effectively express the operation steps and the sequence relationship in the complex nursing process. The system constructs the extracted task intent units as nodes of the semantic graph. Each node records the following core elements, including the nursing task action (such as "monitoring body temperature", "recording body temperature"); trigger condition (such as "within 12 hours after surgery"); execution role (such as "responsible nurse", "general nurse"); and contains task urgency, expected time consumption, etc. The system establishes edges between nodes through two types of relationships, such as sequential dependence relationship, if a task operation must be executed after another task is completed, a directed edge is established between the two. For example, "monitoring body temperature" should be immediately followed by "recording body temperature", indicating that the former is a prerequisite for the latter; conditional logical relationship, if the execution of a task is limited by the state or result of another task, a logical dependence edge is established between the two. For example, "dressing change" needs to be performed on the premise that "wound observation" and no infection are confirmed, and a dependence relationship based on judgment condition is constructed.

[0093] S14, performing semantic clustering on the nursing task semantic graph data to obtain nursing task cluster graph data;

[0094] In an embodiment, the semantic clustering refers to classifying multiple task intent units with similar behavior characteristics or execution conditions in the semantic graph into the same task cluster. The tasks within the cluster have high similarity in execution mode, target object or context condition, and can be regarded as specific instances of the same type of nursing operation. The system performs semantic similarity calculation based on multiple feature dimensions in the task intent unit, mainly including nursing actions (such as “monitoring”, “recording”, “replacement”); trigger conditions (such as “within 12 hours after operation”, “morning and evening every day”); execution role (such as “nurse”, “nurse in charge”); To measure the semantic similarity between two task intent units, the system can use the following measurement methods, such as Jaccard similarity; cosine similarity (such as Word2Vec or BERT encoding). According to the task sample size and clustering accuracy requirements, the system can use any of the following algorithms to perform clustering operations, K-means clustering algorithm, by pre-setting the cluster number, the task vector is divided into a number of sets near the center point, suitable for large-scale task scenarios; hierarchical clustering algorithm, without pre-setting the cluster number, by constructing a task similarity tree structure, merging tasks from top to bottom or from bottom to top. After clustering is completed, the system organizes the task intent units belonging to the same task cluster into a unified cluster structure, and each task cluster has the following elements, cluster name, automatically named according to the common characteristics of the tasks within the cluster, for example “monitoring type task”, “nursing operation type task” and the like; cluster task list, including all task names or identifiers belonging to the cluster; extension field, including typical behavior description of the cluster, number of covered tasks, average execution time and other indicators.

[0095] S15, performing nursing task ontology model construction on the nursing task cluster graph data to obtain nursing task ontology data.

[0096] In an embodiment, the ontology model is a knowledge representation structure used to systematically describe the concepts (e.g., task categories), attributes (e.g., roles, tools, trigger conditions), and constraints (e.g., execution time, space conditions, etc.) in the nursing task cluster. By constructing the ontology, the system can achieve a unified expression, machine processing, and automatic reasoning of the semantic of nursing tasks. Based on the aforementioned task clustering results, the system defines the corresponding ontology concept categories for each task cluster. For example, tasks such as "monitoring body temperature" and "monitoring blood pressure" are classified into the "monitoring task" concept, and tasks such as "changing dressings" and "turning over patients" are classified into the "nursing operation task" concept. For each task ontology instance, the necessary attribute items are defined, covering the core information related to task execution, including but not limited to the execution role, indicating the personnel category (e.g., nurse, responsible nurse, doctor, etc.) performing the task; trigger conditions, specifying the situation under which the task needs to be executed (e.g., "within 12 hours after surgery," "twice a day"); execution method, describing the technical means or tools relied upon for task execution (e.g., "using an infrared thermometer," "using a closed-loop irrigation device," etc.); and task objective, used to express the task intent or expected effect (e.g., "monitoring vital sign changes," "preventing pressure ulcers"). The system defines execution constraint conditions for the ontology task instance, including spatial constraints, the physical location where the task should be executed (e.g., "ward A area," "operation room 4"); time constraints, the time limit or periodic frequency for completing the task (e.g., "every 4 hours," "once within 24 hours"); and equipment constraints, the instruments or auxiliary devices relied upon for task execution (e.g., "infrared thermometer," "intravenous infusion pump").

[0097] Optionally, the nursing work flow chart construction comprises:

[0098] According to the nursing task ontology data, the task ontology instantiation is performed to obtain task instance data.

[0099] In an embodiment, the system iterates through each task ontology record and instantiates the task in combination with the following contextual environment data: patient information (e.g., patient identification, medical record summary, postoperative recovery stage); time information (e.g., key event time points, postoperative stage, medication timing); spatial information (e.g., ward number, bed number, available equipment deployment location); equipment status and scheduling resources (e.g., terminal availability, priority allocation strategy). The instantiation rules include the following processing flow: if the task is "monitoring within X hours after surgery," the system automatically calculates the specific time window; if the task needs to be bound to a specific patient, a unique task identification is generated and associated with the patient; if the task is parallel, multiple instances are generated and assigned to different terminals or time slices.

[0100] The constraint mapping is performed on the task instance data to obtain task constraint data, wherein the constraint mapping includes role constraint mapping, physical constraint mapping, time constraint mapping, and spatial constraint mapping.

[0101] In an embodiment, the role constraint mapping is that the system automatically labels the role execution constraint for each task according to the task type and the in-hospital nursing operation permission specification. This constraint defines that the task must be performed by a role with corresponding qualifications, preventing the occurrence of irregular operations and capacity mismatch. Through the matching of task type and execution requirements, the corresponding role category (such as registered nurse, intern nurse, attending physician, etc.) is identified. For example, for a certain type of intravenous injection task, the system can set "only registered nurses can perform". The physical constraint mapping is that the system determines whether the task depends on a specific hardware terminal or auxiliary equipment during execution, and establishes a corresponding relationship. This constraint ensures that the task dispatch is only directed to nursing terminals with related equipment capabilities. Analyze the calling requirements for device functions in the task content, and extract the required device types in combination with the device resource library. For example, if the task needs to implement continuous infusion monitoring, the system will bind it to a terminal with "infusion pump control capability". The time constraint mapping is that the system generates the execution time window of the task according to the task trigger condition and the current state of the patient. This time window is used to guide the scheduling system to reasonably arrange the start and end time of the task, avoiding missing critical nursing nodes. Perform timestamp calculation on the time-dependent requirements in the task conditions (such as postoperative duration, medication interval), and set the executable interval. For example, the postoperative body temperature monitoring task requires to be completed within "4 hours after surgery", so the system automatically calculates the specific execution time period according to the surgical record. The space constraint mapping is that the system specifies that the task must be completed in a specific physical area in combination with the patient location information bound to the task. This constraint can prevent resource misplacement or cross-area mismatch, ensuring the on-site accessibility and environmental adaptability of task execution. Read the space identifier of the patient's bed or nursing target to impose spatial positioning restrictions on the task. For example, if the nursing object is located in "ward A-bed C05", the task is restricted to be dispatched and executed within the ward area.

[0102] The task constraint data is used to construct the job task node to obtain job task node data;

[0103] In an embodiment, the system abstracts each task instance as a graph node object. The node contains the following core fields, including task identifier (task_id) for uniquely identifying the task instance; task type (task_type) such as monitoring body temperature, injecting medication, etc.; execution priority (priority) in the form of a numerical value, ranging from [0, 1], the higher the value, the stronger the scheduling priority; required role (required_role) such as registered nurse, intern nurse, etc.; required device capability (required_device) such as the need for a handheld PDA or infusion pump support; time execution window (time_window) for the start and end time period during which the task can be scheduled; spatial execution scope (spatial_scope) for specifying the ward, room or bed in which the task should be completed. The calculation method of the execution priority is as follows: wherein is the task priority, is the urgency score weighting coefficient, taking a value of 0.5, is the urgency score (such as the score of a time-limited task marked by a doctor), is the patient risk weighting coefficient, taking a value of 0.3, is the patient condition risk level (such as a higher score for elderly, postoperative, ICU patients, obtained based on a pre-set condition risk table), is the task overlap weighting coefficient, taking a value of 0.2, is the degree of overlap with other tasks in time and space (such as a higher score for multiple tasks at the same time / location).

[0104] According to the job task node data, node division data is obtained;

[0105] In an embodiment, the task nodes are divided according to a time axis, including division based on an hour granularity, shift segmentation (such as an early shift and a night shift), or a clinical event window (such as pre-operation and post-operation); for example, a task occurs within 24 hours after operation, and is classified into a "post-operation initial window" set. Spatial execution locations of the task, such as a ward, a floor, and a nursing unit, are used to divide the task nodes spatially; for example, if a task action area is ward A, the task node is classified into a "ward A node set". The task nodes are classified according to a nursing responsibility relationship, such as a nurse number or a nursing group, according to patients; for example, all patient task nodes of a nurse N01 are classified into a "responsibility group N01 node set". When performing the division, the system classifies the nodes into corresponding division areas based on preset rules. For example, if an execution time of a task node is within 24 hours after a certain clinical event (such as post-operation), the system automatically classifies the task node into a time window corresponding to the event; if a spatial attribute of the task node points to a specific ward or a bed number, the system classifies the task node into a region group matching the space; and if the task node has an explicit mapping relationship with a patient responsibility, the task node is classified into a task cluster under a corresponding responsibility unit.

[0106] Multi-dimensional dependency edge data is obtained by constructing multi-dimensional dependency edges according to the node division data.

[0107] In an embodiment, the dependency edge construction is a sequential dependency edge (E_seq), which is used to represent that there is a sequence constraint between tasks, that is, a task must be executed after another task is completed; for example, if a task T001 is post-operation vital sign collection, and a task T004 is post-operation record submission, the two tasks need to maintain a time-dependent sequence, and a sequential dependency edge T001 → T004 is constructed. A resource conflict edge (E_conflict) is constructed when two tasks share the same device or are executed by the same nursing staff, a conflict dependency edge is established to avoid concurrent conflicts; for example, if T002 and T003 both need to use the same mobile PDA, the system will construct a conflict edge between the two tasks. A spatial flow edge (E_spatial) is used to represent the movement path dependency of tasks in space, which is commonly used in the optimization of physical execution paths of tasks such as patrol and sampling; for example, if T005 needs to be completed in ward A, and T006 needs to be completed in adjacent ward B, and there is an explicit patrol sequence requirement, a spatial path edge T005 → T006 is established. A task complement edge (E_complement) is used to describe a logical complementary relationship between tasks, for example, the completion of one task is a prerequisite for triggering another task; for example, an identity check task needs to be completed before an injection task, and a complement edge is established between the two tasks.

[0108] Nursing work flow diagram data is obtained by generating a work logic diagram according to the node division data and the multi-dimensional dependency edge data.

[0109] In an embodiment, the system integrates all the task node data of the nursing tasks to form a unified node set, denoted as node set , where each node represents a nursing task instance and contains information such as task type, constraint attribute, priority score, time window, and spatial range. Meanwhile, the multi-dimensional dependency edge data is combined to form a multi-type edge set, edge set , where each edge contains edge type (e.g., sequential dependency, resource conflict, spatial flow, task coordination), direction attribute, weight score, and other fields. All divided task nodes are included in the graph; all edges that satisfy the dependency relationship are added to form an attributed directed graph . The graph structure can be processed by patient dimension or time window to form multiple patient schedule-level or shift-level subgraphs.

[0110] Optionally, the construction of the nursing terminal knowledge graph comprises:

[0111] S21, terminal entity extraction is performed according to the nursing operation flowchart data to obtain terminal entity data;

[0112] In an embodiment, the system divides terminals into the following categories according to their physical form and functional use, such as mobile terminals including handheld PDA devices carried by nurses, wearable sensing terminals (such as wristbands or finger clip monitors), and mobile medicine carts; fixed terminals such as infusion pumps at the bedside, bedside call systems, and bedside information display screens, whose positions are basically fixed; intelligent environment terminals including air disinfection devices installed in the ward environment, intelligent lighting control systems, and temperature and humidity monitoring modules; and nursing staff terminals, which abstract the nursing role as a "human terminal" in the form of "Agent Nurse" to participate in task scheduling and resource management, and their behavior ability can be regarded as a schedulable service node.

[0113] S22, terminal and task mapping relationship extraction is performed on the terminal entity data to obtain task terminal association data;

[0114] In one embodiment, the system extracts data fields for each task node from the constructed nursing workflow graph. These fields primarily include: task type (e.g., "identity verification," "dressing change," "medication reminder," etc.); required equipment or functional requirements (e.g., "with barcode scanning function," "requires connection to an infusion pump," etc.); specified or default execution role (e.g., "responsible nurse," "intern nurse"); execution location or task coverage area (e.g., "ward A7," "ICU ward 2"). The system constructs semantic mapping rules between tasks and terminals, organized in triplets as follows: <task number, required features, terminal type>. The matching logic includes: if the task type is "medication verification," matching terminal devices with "barcode scanning" capability and "mobility"; if the task is "infusion change," matching terminals that can be bound to an "infusion pump" device, such as a bedside control screen; if the task limits the execution role to "nurse," only matching terminals bound to nursing permissions (e.g., nurse PDAs); if the task specifies a spatial location, only matching terminal entities whose deployment location covers that area. Task type (task_type) Terminal capabilities are determined by matching the terminal's supported operational capabilities to the functionalities required for the task; task role is defined as follows. Terminal role binding ensures that tasks can only be controlled by terminals with the appropriate permissions; task location. Terminal deployment location (terminal_location_coverage) prioritizes terminal devices that are in the same or nearby spatial locations.

[0115] S23. Construct a functional hierarchy classification tree based on the task terminal association data to obtain functional tree data;

[0116] In an embodiment, all terminal entities and their capability sets are input, and the task mapping result (terminal task association distribution) is obtained. The terminal capabilities are semantically labeled, such as "code scanning" -> ID_Verification; "identity check + infusion confirmation" -> PreInjection_SafetyCheck; and "call response" -> Passive_AlarmReceiver. A terminal function similarity matrix is established, and the Jaccard similarity coefficient and capability vector embedding are used. Spectral clustering or DBSCAN is applied to cluster the terminal capabilities in the function semantic space. In the function tree construction process, the clustering results are used as the basis for constructing the second layer "function subclass" nodes. The system calculates the capability similarity matrix based on the terminal capability labels (such as code scanning, infusion confirmation, etc.), and uses clustering algorithms (such as spectral clustering or DBSCAN) to divide the terminal capabilities into several capability similarity clusters. Each cluster represents a set of terminals with similar semantic functions. The system extracts the "semantic center label" of each cluster based on the commonality of the main capability labels in the cluster, and uses it to name the function subclass node corresponding to the cluster. For example, cluster A includes "blood pressure monitoring device", "body temperature detector", and "multi-parameter monitor", and the common label is "vital sign monitoring". Therefore, the function subclass node VitalsMonitoring is established. Cluster B includes "code scanning device" and "drug identification assistant", and the commonality is "medication verification". Therefore, the subclass node MedicationVerification is established. The function classification tree adopts a multi-layer organizational structure, and is refined layer by layer according to "task category -> function category -> function instance". The first layer is the task category (monitoring class, execution class, and auxiliary class). The second layer is the function subclass (such as temperature monitoring and blood pressure monitoring under the monitoring class). The third layer is the function instance node (terminal model / device capability combination).

[0117] S24, ontology subgraph subnet construction is performed on the function tree data to obtain a nursing terminal knowledge graph.

[0118] In an embodiment, the system converts each functional node, terminal entity and its attributes in the functional hierarchical classification tree into semantic elements in the ontology structure based on the functional hierarchical classification tree, including specifically class node (OWL Class) construction, each functional class in the functional tree (such as "temperature monitoring", "call response") is converted into an OWL class; individual node (OWL Individual) construction, each terminal entity instance (such as PDA, infusion pump) is converted into a specific individual (Individual); functional capability (such as "code scanning", "identity checking") is modeled as an object attribute (ObjectProperty) for describing the connection relationship between the terminal and its supported functions; quantitative indicators such as execution delay and power consumption are modeled as data attributes (DataProperty) for representing quantitative characteristics of the terminal, such as "execLatency (execution delay)" and "batteryLife (battery life)". The system defines the following semantic relationship edge types, including inheritance relationship (subClassOf), which is used to represent the hierarchical structure between functional classes, for example, "temperature monitoring" is a subclass of "vital sign monitoring"; functional equivalence relationship (canBeSubstitutedBy), which represents that two terminals are interchangeable at the functional level and is used for scheduling substitution; spatial deployment relationship (locatedIn), which is used to describe the spatial subordination relationship between the terminal entity and its deployment location; context activation relationship (hasContextualActivation), which is used to model the activation rules of the terminal in a specific context (such as time, area, patient state).

[0119] Optionally, the hierarchical classification tree construction includes:

[0120] According to the task terminal association data, semantic label generation is performed to obtain association semantic label data;

[0121] In one embodiment, the semantic tag generation process is based on the following three types of data: task terminal association data, including the set of tasks supported or executed by each terminal entity; task meta-information, including task type tags (such as monitoring, verification), execution requirements (such as required device type, latency limits, and permission roles); and terminal execution behavior records, including historical execution frequency, success rate, typical service areas, and other operational indicator data. The system defines a terminal semantic tag generation function `Tag(terminal_id)` to abstract input information into multiple semantic tags. For example, task category tag generation extracts the types of tasks executed by the terminal and assigns them corresponding semantic tags. For instance, if the terminal has executed a body temperature monitoring task, the tag is "vital signs monitoring"; if it has executed a drug scanning task, the tag is "medication verification". Functional attribute tag generation is based on the terminal's hardware capabilities or system configuration. For example, if it supports wearable use, the tag is "supports wearing"; if it is a mobile terminal, the tag is "mobile"; if the system has a high response rate, the tag is "real-time response". Performance capability tag generation analyzes the terminal's historical behavior data to reflect its operational performance. For example, a response latency of less than 3 seconds is tagged as "Low Latency"; high usage rate in ICU wards is tagged as "ICU Permanent"; high success rate is tagged as "High Success Rate". Tag types include, but are not limited to, task category tags, generated based on the type of task the terminal is involved in, such as VitalSignsMonitoring; MedicationScan; AlarmResponse. Functional attribute tags reflect the terminal's functional characteristics, such as Mobile; WearableSupport; RealTime; SafetyCritical. Performance capability tags are extracted from the terminal's execution records, including ResponseLatency<3s; BatteryPowered; UsedInICU; HighSuccessRate.

[0122] Graph function clustering is performed based on the associated semantic label data to obtain associated clustering data;

[0123] In one embodiment, a functional similarity graph G=(V,E) is constructed, where V is the terminal node and E is the functional similarity edge between any two terminals. The weights are calculated as follows: ,in For functional similarity, For the first One terminal node, For the first terminal nodes, is the Jaccard similarity, is the terminal associated semantic tag set, is the terminal associated semantic tag set, or wherein is the functional similarity, is the first terminal node, is the first terminal node, is the embedding vector cosine similarity, is the terminal associated functional embedding vector, is the terminal associated functional embedding vector; spectral clustering is used for small-scale data sets; the Louvain community discovery algorithm or DeepWalk+KMeans is used for large-scale graphs. Each terminal is allowed to appear in multiple clusters (multiple roles), i.e., a soft clustering structure is constructed.

[0124] The associated clustering data is subjected to a functional hierarchical classification structure tree construction, and preliminary functional tree data is obtained;

[0125] In an embodiment, the constructed functional tree is a directed hierarchical tree structure composed of multiple nodes, and the node types and levels are as follows: the root node represents the top-level functional category and is uniformly named "task support terminal" or "nursing capability terminal set"; the intermediate node represents a functional clustering unit and corresponds to the semantic functional cluster obtained in step S26, such as "vital sign class", "alarm response class", etc.; and the leaf node represents a specific terminal entity instance, such as a handheld PDA, a wristband terminal, and a bedside monitoring device. Each node has the following additional attributes, including source tag information indicating the semantic clustering tag set from which the node originates; inherited capability characteristics describing the functional tags inherited from the upper layer of clustering, such as "mobility", "real-time response", and "ICU availability"; and replaceable relationship markers for marking whether the node has a function equivalent or partially overlapping replacement candidate. The system generates a functional hierarchical path according to the following construction principles, including clustering as an intermediate layer, taking the clustering result as the intermediate node level of the tree to realize the structural unification of semantic clustering and functional classification; terminal multiple membership processing, if a terminal entity appears in multiple clusters, the system generates leaf nodes for it under multiple functional paths, marks them as "cross nodes", and records their path intersection information; and functional path refinement logic, the tree path follows the order from "general functional category" to "dedicated functional subcategory" to "entity terminal" for construction, such as "task support terminal → vital sign class → mobile vital sign terminal → PDA-01".

[0126] According to the preliminary function tree data, environment sensing adaptation is performed to obtain function tree data.

[0127] In an embodiment, the system pre-defines a set of environment adaptation rules for determining whether a specific environment meets the applicable scenario of a function node, and the activation function is used to express , wherein is the activation function of the function node, is the i-th function node, is the current running environment or context of the system, such as "night shift", "emergency area", "high pressure state", etc., is the activation result is true, is the conditional judgment logic, is the context set in which the function node is allowed to be activated, which can be pre-set as multiple tags or condition combinations. If the context is "night shift + B ward", the bed-side monitoring terminal node supporting quiet mode is activated; if it is in the ICU area and the patient is an elderly group, the priority of the high-precision vital sign monitoring node is increased. The system marks "whether available" state for each terminal node in the function tree according to the context activation result, and records the activation reason; for example, the terminal "T-Watch-03" is in the "activated" state in the current "night shift ward" scenario, and the reason recorded is "suitable for night mobile monitoring"; if a terminal is determined to be unavailable due to failure or inapplicability, the system will find a "replaceable sibling node" with equivalent ability in the same layer or adjacent layer of the function tree, and temporarily reassign the path for replacement scheduling; the replacement relationship is sorted and selected according to "functional similarity label" and "context adaptation overlap degree".

[0128] Optionally, the ontology subgraph subnetwork structure comprises:

[0129] The function tree data is subjected to node ontology mapping to obtain node mapping data, wherein the node ontology mapping comprises capability ontology mapping, service area mapping, and response time mapping;

[0130] ​In an embodiment, the input is a leaf node in a function tree structure, each leaf node representing a specific care terminal unit, such as a bedside monitoring device, a mobile care terminal or a call response device, etc. Each terminal node contains its unique identifier and function type information. The capability ontology is mapped to the "has capability" attribute field according to the function modules or sensing capabilities possessed by the terminal. For example, if a device has the function of real-time vital sign acquisition, its capability label can be "real-time vital sign monitoring"; if a mobile device supports drug code scanning and medication verification, the capability label is "code scanning and drug verification". This mapping can be uniformly expressed as "has the following capabilities". The service area is mapped to the deployment location or coverage area of each terminal, which clearly indicates its physical distribution or service reach. By setting the "deployed in a certain area" attribute, the area boundary constraint is formed, such as the device only serves a certain ward or care unit. The response time is mapped to the response performance setting attribute of the terminal, mainly including the maximum response time, startup time and other time-sensitive indicators. This attribute can quantitatively reflect the interaction performance of the terminal. The system integrates the above three types of attributes to form a node ontology knowledge item oriented to the terminal, and outputs a structured triple set, each record being expressed in the form of "entity-attribute-attribute value".

[0131] The node mapping data is subjected to semantic relationship edge construction to obtain semantic relationship edge data;

[0132] In an embodiment, based on the node ontology mapping data, the semantic relationship edges between terminal devices are constructed to form a directed connection structure with semantic logical meaning. For example, the substitutability relationship (isSubstitutableWith) is that if two terminals have highly similar capability tags and the difference in their response times is within a preset tolerance range (e.g., no more than 1.5 seconds), it is considered that they have functional substitutability in a specific task scenario. Such edges are used to automatically recommend or switch to an alternative terminal to perform the same task in the case of terminal failure or task failure. The functional dependency relationship (isFunctionallyDependentOn) is that if the normal operation or task execution of a terminal device must rely on the pre-function or state support of another terminal (for example, the position information collection of a ward call depends on the bedside monitor), a "functional dependency" relationship is established between the two terminals. This edge is used to describe the prerequisite or hardware binding relationship in the function hierarchy. The co-execution history relationship (hasCoExecutionHistoryWith) is based on historical nursing task execution trajectory data. If the system statistics find that two terminals have frequent co-working records in multiple nursing task processes (i.e., multiple times of participating in the same task or process segment at the same time), a co-history edge can be established between them to express the potential collaboration frequency and deployment linkage demand. The system traverses all terminal node pairs, and evaluates their attribute tags and historical trajectory data in turn. If any of the above relationship construction logics is met, a semantic relationship edge is generated.

[0133] According to the semantic relationship edge data, the ontology subgraph structure is generated to obtain ontology subgraph structure data;

[0134] In an embodiment, the input data includes a set of terminal nodes, all nursing terminal nodes that have passed through the ontology attribute mapping, including attribute information such as functional capabilities, deployment locations, and response capabilities; and a set of semantic relationship edges, a completed directed semantic edge data, indicating the relationships between terminals such as replaceability, functional dependence, and collaborative execution. The system uses a graph structure modeling tool (such as the graph database Neo4j, the semantic reasoning engine RDFLib, or the graph computing library NetworkX) to initialize the graph structure, and imports all nodes and semantic edges into the graph to form a global semantic graph. The system extracts subgraphs according to functional categories. For example, based on classification criteria such as “monitoring type”, “drug injection type”, and “recording type”, nodes with similar capability labels or collaborative records in historical tasks are grouped into the same subgraph. Each ontology subgraph should meet the following structural characteristics, including strong internal connectivity, a connection density between terminal nodes in the subgraph not less than a preset threshold (for example, a density ≥ 0.6), ensuring sufficient semantic interaction; functional redundancy support, at least one “replaceable relationship” path in the subgraph for task fault tolerance scheduling; collaborative execution capability, at least one “collaborative historical relationship” path in the subgraph representing the joint operation capability between devices; and functional anchor nodes, each subgraph should contain at least one main functional terminal node as the central anchor or functional representative node of the subgraph. In the constructed ontology subgraph, the system performs attribute labeling on the nodes and edges (node attributes include: functional type label (such as “vital sign monitoring”), deployment location (such as “ward A area”), and capability level (such as high, medium, and low levels); edge attributes include: semantic relationship type, historical collaboration weight (which can be calculated based on past task co-occurrence frequency), and response time difference value), obtaining ontology subgraph structure data.

[0135] The ontology subgraph structure data is subjected to perception activation addition to obtain a nursing terminal knowledge graph.

[0136] In an embodiment, external operating environment information is obtained, and context data such as nursing shifts (day shift / night shift), ward characteristics (ICU / general ward), and patient status (high risk / early postoperative) is embedded into the existing ontology subgraph structure, thereby dynamically activating or inhibiting the participation state of a specific terminal node, and realizing the intelligent adaptation capability of the graph in different nursing situations. The system defines context activation rules for determining whether a terminal node in the graph is in a “schedulable” or “preferentially enabled” state under the current environment. The activation rules include an activation input unit, a node representing a nursing terminal, and an environment context C representing the current shift type, ward scenario, and patient level; a similarity matching function ContextMatch( ), the matching degree of the attributes (such as deployment location, functional capability, response time, etc.) possessed by the computing node and the context condition; and activation determination, if the matching score is higher than a preset threshold (for example = 0.75), the node is activated, indicating that it should participate in task scheduling under the current situation. The activation determination formula can be expressed as if ContextMatch( ) ≥ , the node is marked as "activated", otherwise as "not activated". A set of typical situation-device activation rules are predefined in the system, for example, when the scene is "ICU ward + high-risk patient", the terminal device with high-precision continuous vital sign monitoring capability is preferentially activated; when the scene is "night shift + shortage of manpower", the terminal with automatic execution capability is activated, such as intelligent medicine dispensing device or voice navigation interactive device; if the environmental noise is high and manual interaction is difficult, the terminal device relying on voice input is disabled.

[0137] Optionally, the multi-agent collaborative model construction comprises:

[0138] According to the nursing terminal knowledge graph, agent entity extraction is performed to obtain agent entity data;

[0139] In an embodiment, the system performs candidate screening according to the following three functional attributes, including perceptibility, the terminal has input functions such as signal acquisition, monitoring and reception; communicability, the terminal can exchange information with the system platform or other terminals, supporting task receiving, state reporting or message broadcasting; and schedulability, the terminal supports control behaviors such as instruction issuing, task execution and state updating. The terminal node that meets any two of the above and is in the "activated" state under the current context can be used as an agent candidate node. For each terminal node that meets the condition, the system extracts its core structure information to form an agent entity, including an agent identifier (agent_id), which is the unique number of the terminal in the knowledge graph, used to identify the entity in the multi-agent environment; a capability set (capabilities), indicating the core operation types supported by the terminal, such as "vital sign monitoring", "alarm pushing", "nursing task execution", etc.; a deployment location (location), indicating the area unit currently served by the terminal, such as "ICU-A zone", "ward B group", etc.; and a resource status (resource_status), recording the current running load, including but not limited to battery percentage, availability label (such as "idle", "executing", "under maintenance"), device temperature, CPU occupancy, etc.

[0140] According to the agent entity data, agent perception simulation is performed to obtain agent simulation data;

[0141] In one embodiment, an agent entity is used as the modeling unit to construct a state evolution model reflecting its perception-response-transfer process in the current nursing environment context. The simulation objectives include predicting the agent's resource status (availability, idle time, etc.) in future time periods; simulating perceived environmental changes (such as sudden task density, ward level upgrades); and extrapolating possible response behaviors (such as immediate execution, delayed execution, proactive alerts, etc.). The model's state space and input variables mainly include the following three dimensions, including the state dimension (…). This reflects the operational state of an intelligent agent at a given moment, such as processing load (CPU usage), communication capability (bandwidth usage), and remaining battery power; environmental perception dimension ( ), reflecting the nursing context of the area where the agent is located, such as the density of nearby emergencies, the risk level of the ward, and the patient's vital sign load index; behavioral response dimension ( This represents the set of actions that the agent can currently perform (such as monitoring, reporting, and task execution), and the execution costs of these actions (such as estimated completion time and estimated energy consumption). The simulation uses a state transition function to model the agent's state evolution path at each time step, in the form: ,in Indicates the first An intelligent agent in The state at any given moment; Indicates in The set of environmental context parameters observed at any given time; This represents the action taken by the agent at the current moment (which can be an externally issued command or an autonomous decision response). If the simulation objective is rule-driven static scene evaluation, a rule-based state evolution method is used to explicitly update the state value through conditional statements. If the simulation objective is dynamic sequence prediction or long-term trend modeling, a recurrent neural network model (such as LSTM or GRU) is recommended to learn the nonlinear laws of state evolution over time and automatically generate future state sequences.

[0142] Based on the agent simulation data, structural perception collaborative modeling is performed to obtain the nursing collaboration model.

[0143] In one embodiment, structure-aware collaborative modeling aims to construct a graph structure model reflecting the collaborative potential and collaborative paths of multiple agents based on the state evolution and functional interactions among them. The constructed collaborative graph structure is defined as a directed graph. , where the set of nodes Each node in the graph represents an extracted nursing agent entity; the edge set , represents that there is a certain collaborative relationship between the two agents; edge attribute type: used to represent the type of collaborative relationship, including functional complementarity, two agents have complementary operation capabilities (such as monitoring and reporting); task connection, two agents have a sequential connection relationship in the task flow; resource sharing, two agents share communication, energy consumption or task control resources at the physical location or platform level. The weight of each edge is used to measure the strength and priority of the collaboration between the two agents, and the weight calculation function is defined as follows: , wherein is the agent collaboration weight, representing the edge weight of the agent and the agent in the collaboration graph, is the collaboration frequency weight coefficient, taking a value of 0.5, represents the frequency of the collaborative relationship between the agent and in the historical task execution, is the functional similarity weight coefficient, taking a value of 0.3, represents the cosine similarity between the functional capability vectors of the two agents, is the spatial adjacency weight coefficient, taking a value of 0.2, represents whether the two are deployed in adjacent or overlapping areas in the physical space.

[0144] Optionally, S3 includes:

[0145] S31, performing task-driven window division according to the nursing collaboration model to obtain nursing task division data;

[0146] In an embodiment, the nursing tasks in a to-be-scheduled or to-be-executed state are divided into a plurality of task scheduling windows with independent identifiers according to their time sensitivity, spatial deployment characteristics, and task type priority. The window not only exists as a scheduling unit, but also can be used as a basic structure for load evaluation, agent scheduling, and path optimization. The time dimension division is performed according to the trigger period of the task, the service response time limit, and the high-frequency nursing response cycle summarized in the collaboration model to dynamically partition the task execution period. The generated form is a set of time period intervals , representing that the task has scheduling eligibility within the window. Spatial dimension division is performed by clustering the physical space where the task belongs to, such as by using the deployment location of the intelligent terminal, the geographical division of the ward (e.g. A zone, B zone), and the current flow density of the nursing staff, etc. Each spatial window corresponds to a service area label RkR_kRk, which is used for the spatial constraint convention of task-resource matching. Priority stratification strategy is performed by ranking the task types according to indicators such as the urgency of the task and the service impact range: first-aid or high-risk tasks enter the high-priority level, corresponding to a narrower time window and higher scheduling response requirements; routine nursing tasks enter the low-priority level, which can be assigned to a relatively loose time window to balance the system resource occupation. Each task scheduling window is represented in the form of a triple: , where is the window start time; is the window end time; is the spatial service area bound to the window. This structure serves as the basic unit of the divided scheduling window and can also carry the associated task type or task set.

[0147] S32, performing distributed task candidate matching according to the nursing coordination model and the nursing task division data, to obtain task candidate matching data;

[0148] In an embodiment, for each nursing task to be scheduled, a number of intelligent agents with good execution capability, response timeliness, resource state and execution history evaluation are selected from the available terminal resources according to the scheduling window and task characteristics where the task is located, to form a set of candidate intelligent agents. The matching mechanism is executed based on multi-dimensional evaluation criteria, mainly including the following four types of matching dimensions. The capability matching is to determine whether the operation capability required by the task is included in the capability list of the candidate intelligent agent. For example, if the task requires "information registration" and "barcode checking" functions, the intelligent agent's function description must cover the corresponding capability modules. The spatiotemporal accessibility matching includes spatial distance accessibility calculation, which calculates the spatial distance between the task location and the current location of the intelligent agent , requiring that it does not exceed the preset spatial threshold . The time response accessibility calculation requires that the available time of the intelligent agent is earlier than or equal to the expected start time of the task, i.e. . The resource state matching is to check whether the current running state of the intelligent agent meets the task execution requirements, including but not limited to indicators such as power level, the number of tasks currently being processed, communication bandwidth availability, etc. The historical execution reliability matching is to use the historical task execution data recorded in the system to calculate the success rate or stability indicators of a certain intelligent agent when executing this type of task, denoted as , where is the execution reliability score, which represents the historical success rate or stability indicators of a certain intelligent agent when executing a specified type of task. is an identifier of an intelligent agent, representing a specific entity participating in the execution of a nursing task in the system, which can be a digital human, a nursing robot, a mobile terminal, or an edge node with execution capability, is an identifier of a task type, used to represent a specific category of a nursing task, such as vital sign measurement, drug distribution, patrol reminder, etc. Only when the index is greater than or equal to a preset threshold , the task is considered as a valid candidate. The system needs to return at least N candidate intelligent agents for each task. If the task has no qualified candidate matching object under the current window, the task is marked as a “scheduling bottleneck” and automatically reported to the main control engine. After the system completes the matching calculation, the output is the candidate intelligent agent set corresponding to each task.

[0149] S33, generating a collaborative scheduling graph according to the task candidate matching data to obtain nursing task collaboration data.

[0150] In an embodiment, a scheduling graph model combining the bilateral relationship between tasks and resources is established, the matching priority relationship between tasks and schedulable intelligent agents is clarified, and the graph structure is endowed with multiple scheduling elements such as time window, load distribution, and priority control. Two vertex sets of the graph represent the set of tasks to be executed and the set of schedulable intelligent agents, respectively, wherein the task node represents each specific nursing task; the intelligent agent node represents a terminal resource with scheduling capability (such as a nursing PDA, a monitoring device, etc.). Each edge represents the matching relationship between a task and an intelligent agent, and is attached with the following key attribute fields, including matching weight (matching score value), reflecting the adaptation degree of the intelligent agent under the current task; time window (execution time period), indicating the scheduling time interval to which the task belongs; load balancing score (load state index), measuring the acceptable degree of the current load carried by the intelligent agent; priority score (task urgency), based on the urgency of the task itself and the patient state, etc. The edge selection can use the following strategies, using the Hungarian method or heuristic greedy algorithm, to preferentially match the task-intelligent agent pair with the largest matching weight in the graph, to maximize the overall matching score. If the tasks can be executed in parallel, a multiple matching algorithm can be used to generate multiple feasible subgraphs under the guarantee of resource constraints, to improve resource utilization. The scheduling “rollback” and “task exchange” strategies are supported to adapt to external environmental fluctuations, temporary resource failures, or sudden task insertion. For example, when an intelligent agent fails suddenly, the system can quickly switch the task target based on the reserved candidate edge.

[0151] Optionally, S4 includes:

[0152] S41, obtaining historical nursing data;

[0153] In an embodiment, the system interfaces with the nursing information system and the task execution log platform, extracts the historical execution record of each nursing task, mainly including task ID, which identifies the unique number of each nursing task; agent ID, which records which agent (such as a nurse or a terminal device) actually executes the task; start time and end time, which are used to calculate the task execution duration and the scheduling period; location, which represents the execution location or the associated ward of the task; status, which includes completed, canceled, failed, and other status labels; error code, which records the exception or failure reason in the task execution process, and is used for subsequent exception classification and risk modeling.

[0154] S42, reconstruct the historical nursing task trajectory of the nursing task coordination data according to the historical nursing data, and obtain the nursing task trajectory data;

[0155] In an embodiment, the complete execution sequence of the historical nursing task is restored, and is modeled as a trajectory graph structure in the time and space dimensions. The reconstruction content includes task sequence, spatial migration path, task connection logic, and node attribute system. Based on the obtained historical nursing task data, the trajectory is reconstructed in the following manner: according to the start and end time of the task executed by each agent (agent), the tasks are arranged in chronological order to generate an ordered task event sequence , which represents a set of task events from the first to the nth, and is used to form an event chain; the location field in the task execution record is extracted to form a spatial path set of the agent during the task process: , wherein represents the execution location of the th task, and the path is represented as a sequence of continuous spatial migration points, , and the value of n is 1…n; if the system is deployed with Internet of Things positioning devices (such as RFID, Bluetooth beacon, Wi-Fi trajectory record, etc.), the actual trajectory can be refined to jump; if not, the indirect inference is based on the task location record. The constructed task trajectory graph is a directed graph structure with attributes, including nodes representing each executed nursing task, node attributes including task number, start and end time, execution location, responsible subject, etc.; edges representing the association between two consecutive tasks, edge attributes including time span, representing the time interval between two tasks; path distance, representing the spatial jump distance, which can be approximately calculated according to the ward plan coordinate or unit number; and waiting time, representing the waiting time from the end of the previous task to the start of the next task.

[0156] S43, scheduling behavior extraction is performed on the nursing task trajectory data to obtain scheduling behavior data;

[0157] In an embodiment, the system extracts and models the key scheduling behavior characteristics in the historical task execution process based on the constructed nursing task trajectory data, to form structured scheduling behavior data. The system extracts the following key behavior characteristic indicators from the trajectory data, including response delay, indicating the time difference between the task being allocated and the actual start of execution, defined as the task start time minus the planned allocation time. Agent switching frequency, indicating the number of switches involving different execution agents (agents) in the same task chain (such as a continuous nursing process), in the same task chain (such as a continuous nursing operation process), the number of times the task is switched between different agents. Its calculation method is to count the number of times the execution agent is replaced in each task chain; task switching delay, which refers to the length of the connection between consecutive tasks, i.e. the time difference between the task being allocated and the actual start of execution, defined as the task start time minus the planned allocation time; the system combines the above multiple behavior characteristics to form a unified behavior description vector, defined as: wherein represents the response delay of the th task; represents the agent switching frequency; represents the task switching delay; the expandable items include the number of task retries, abnormal interruption markers, etc. The system can use clustering algorithms (such as K-means, DBSCAN, etc.) to cluster and analyze the behavior samples under the same task type based on the constructed behavior vector, to form scheduling behavior templates or typical behavior portraits.

[0158] S44, scheduling efficiency calculation is performed according to the scheduling behavior data to obtain scheduling efficiency data;

[0159] In an embodiment, the system quantitatively evaluates the execution efficiency in the nursing task scheduling process based on the historical scheduling behavior data, and constructs a multi-dimensional efficiency index set. Key indicators include average response time wherein is the average response time, indicating the average delay between the task being allocated and the start of execution, is the total number of tasks, is the task index number, is the start timestamp of the th task, is the allocation timestamp of the th task; average execution time wherein is the average execution time, is the total number of tasks, indexing the tasks, the start time stamp of the th task, the end time stamp of the th task; resource utilization (such as agent working time / total online time), calculating the effective working time proportion of each agent (such as a nursing terminal or personnel) in its online period. The working time can be calculated by accumulating the task execution time, and the total online time is obtained based on login, positioning or device heartbeat record. Multi-dimensional filtering calculation such as time period / region / task type is supported; output scheduling efficiency score matrix.

[0160] S45, according to the scheduling efficiency data, abnormal behavior attribution is performed to obtain scheduling bottleneck data;

[0161] In an embodiment, the system identifies and attributes the key bottleneck points affecting the efficiency of nursing task execution based on the scheduling efficiency data obtained by the foregoing calculation, and constructs a scheduling bottleneck data structure. The system introduces an abnormal score function to judge whether there is an efficiency anomaly in the task scheduling execution. The following multi-dimensional index thresholds are set: if any index exceeds the preset threshold, it is considered abnormal, including response delay (response_time) > threshold ; task execution time (execution_time) > threshold ; task failure rate (failure_rate) > threshold ; when any index in the historical task trajectory exceeds the corresponding threshold, the corresponding task instance and its associated agent will be marked as an "abnormal task". Traceability analysis is performed on all abnormal tasks, and the system uses the following attribution factor system, such as capability mismatch, the agent executing the task does not have complete support capability or part of the function cannot meet the standard; low path planning efficiency, unreasonable spatial path planning, leading to cross-zone or non-shortest path migration; concurrent resource congestion, task-intensive and concurrent conflict is serious during scheduling period, system scheduling capability is limited; environmental interference factors, such as sudden event insertion, near shift handover time, etc. The system can use the following two types of methods to model the causal path of abnormal behavior, including naive Bayes model, based on historical data distribution, learning the conditional probability relationship between each attribution factor and abnormal label, outputting the most possible abnormal trigger factor; decision tree model, constructing task attributes, scheduling state, environmental context and other input dimensions, identifying the most critical split condition and attribution path through tree structure, suitable for scenarios with higher interpretability requirements. The system outputs scheduling bottleneck data.

[0162] S46, according to the scheduling bottleneck data, the strategy graph of the nursing task coordination data is updated to obtain nursing task optimization data.

[0163] In an embodiment, based on the bottleneck attribution result in the historical task, the original collaborative graph is corrected to solve the problems of inefficient path, resource conflict or capacity imbalance, and the edge weight or path structure is updated to make the task preferentially match to the agent node with high success rate and reasonable load in future scheduling. The original scheduling strategy graph structure is denoted as , wherein is all the nursing agent nodes participating in scheduling; is the collaborative edge of the task reachable path, including the scheduling weight and the optional execution path; the attributes of each edge include the task type adaptation degree, the historical success rate, the collaborative frequency, the response delay, etc. The updated strategy graph structure is denoted as . If a certain agent frequently appears failure record in a specific task type, the matching weight of the edge corresponding to the task category is adjusted downward. The update rule is as follows , wherein is the weight value of the adjusted task scheduling edge; is the weight value of the original task scheduling edge; is a scheduling tolerance coefficient, which controls the sensitivity of the failure rate to the scheduling weight, and the value is 0.1-0.5; represents the historical task failure rate of the agent under the task type. If there is a high congestion, high failure rate edge in the original scheduling path, the system will re-match the agent path with similar ability and better performance in the candidate node set, and include it in the alternative path graph; the original path can be marked as “de-weighted path” or “invalid path”, and its priority is reduced in scheduling. The edges that frequently appear bottleneck problems (such as congestion, delay, failure) are deleted or de-weighted; the alternative agent path is introduced to re-match from the bottleneck task node, and is marked as “reconstructed edge”; it is ensured that the updated graph is still a connected graph to avoid task node isolation or no assignable target. After the update is completed, the system outputs a new optimized version of the scheduling strategy graph.

[0164] Optionally, the application also provides a nursing operation flow multi-terminal collaborative system for executing the nursing operation flow multi-terminal collaborative method as described above, and the nursing operation flow multi-terminal collaborative system comprises:

[0165] a task ontology construction module, configured to acquire nursing basic data, and perform nursing task semantic construction according to the nursing basic data to obtain nursing task ontology data; and perform nursing operation flow graph construction according to the nursing task ontology data to obtain nursing operation flow graph data;

[0166] a terminal knowledge collaborative modeling module, configured to perform nursing terminal knowledge graph construction according to the nursing operation flow graph data to obtain a nursing terminal knowledge graph; and perform multi-agent collaborative model construction according to the nursing terminal knowledge graph to obtain a nursing collaborative model;

[0167] The intelligent scheduling execution module is configured to perform task coordination scheduling according to the nursing coordination model, and obtain nursing task coordination data.

[0168] The scheduling feedback optimization module is configured to obtain historical nursing data, and perform task coordination efficiency optimization on the nursing task coordination data according to the historical nursing data, and obtain nursing task optimization data.

[0169] Therefore, from any viewpoint, the embodiments should be considered as being exemplary and not limiting, the scope of the application being defined by the appended claims and not by the above description, and all the changes which fall within the meaning and the scope of equivalent elements of the application file are therefore intended to be embraced therein.

[0170] The foregoing is considered as a specific implementation of the application, enabling those skilled in the art to understand or implement the application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the application. Therefore, the application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-terminal collaborative method for nursing workflow, characterized in that, The method includes: S1. Obtain basic nursing data, including patients' basic demographic information, clinical diagnosis results, treatment and medication plans, pre- and post-operative records, past medical history, nursing plans, and extract key information elements reflecting dynamic changes in patient condition and nursing behaviors from nursing documents, shift handover records, free-text nursing logs, and medical-nursing communication records; process the basic nursing data into task intent units to obtain task intent unit data; construct a nursing task semantic graph based on the task intent unit data to obtain nursing task semantic graph data; perform semantic clustering on the nursing task semantic graph data to obtain nursing task cluster graph data; and perform nursing task... The process involves: constructing an ontology model to obtain nursing task ontology data; instantiating the task ontology based on this data to obtain task instance data; performing constraint mapping on the task instance data to obtain task constraint data, including role constraint mapping, physical constraint mapping, time constraint mapping, and spatial constraint mapping; constructing task nodes from the task constraint data to obtain task node data; partitioning the task node data to obtain node partitioning data; constructing multidimensional dependency edges from the node partitioning data to obtain multidimensional dependency edge data; and generating a task logic graph based on the node partitioning data and the multidimensional dependency edge data to obtain nursing task flow graph data. S2. Based on the nursing workflow diagram data, terminal entities are extracted to obtain terminal entity data; the terminal entity data is then used to extract the terminal-task mapping relationship to obtain task-terminal association data; a functional hierarchy classification tree is constructed based on the task-terminal association data to obtain functional tree data; ontology subgraphs and subnets are constructed based on the functional tree data to obtain a nursing terminal knowledge graph; intelligent agent entities are extracted based on the nursing terminal knowledge graph to obtain intelligent agent entity data; intelligent agent perception simulation is performed based on the intelligent agent entity data to obtain intelligent agent simulation data; and structural perception collaborative modeling is performed based on the intelligent agent simulation data to obtain a nursing collaboration model. S3. Perform task coordination scheduling based on the nursing coordination model to obtain nursing task coordination data; S4. Obtain historical nursing data and optimize the task collaboration efficiency of nursing task collaboration data based on the historical nursing data to obtain optimized nursing task data.

2. The method according to claim 1, characterized in that, The construction of the functional hierarchical classification tree includes: Semantic tags are generated based on the associated data from the task terminal to obtain associated semantic tag data; Graph function clustering is performed based on the associated semantic label data to obtain associated clustering data; A functional hierarchical classification structure tree is constructed from the clustered data to obtain preliminary functional tree data. Environmental perception adaptation is performed based on the preliminary functional tree data to obtain functional tree data.

3. The method according to claim 1, characterized in that, The ontology subgraph subnet construction includes: Perform node ontology mapping on the function tree data to obtain node mapping data, where node ontology mapping includes capability ontology mapping, service area mapping, and response time mapping. Semantic relation edges are constructed from the node mapping data to obtain semantic relation edge data; The ontology subgraph structure is generated based on the semantic relation edge data to obtain the ontology subgraph structure data. By performing perceptual activation and adding to the ontology subgraph structure data, a nursing terminal knowledge graph is obtained.

4. The method according to claim 1, characterized in that, S3 include: Based on the nursing collaboration model, task-driven window partitioning is performed to obtain nursing task partitioning data; Distributed task candidate matching is performed based on the nursing collaboration model and nursing task segmentation data to obtain task candidate matching data; Based on the task candidate matching data, a collaborative scheduling graph is generated to obtain nursing task collaborative data.

5. The method according to claim 1, characterized in that, S4 include: Obtain historical nursing data; Based on historical nursing data, the historical nursing task trajectory is reconstructed from the nursing task collaboration data to obtain nursing task trajectory data; Scheduling behavior data is obtained by extracting scheduling behavior data from nursing task trajectory data. The scheduling efficiency data is obtained by calculating the scheduling behavior data. Attributing abnormal behavior to scheduling efficiency data yields scheduling bottleneck data. Based on the scheduling bottleneck data, the nursing task collaboration data is updated using a strategy graph to obtain optimized nursing task data.

6. A multi-terminal collaborative system for nursing workflow, characterized in that, For executing the multi-terminal collaborative method for nursing workflow as described in claim 1, the multi-terminal collaborative system for nursing workflow includes: The task ontology construction module is used to acquire basic nursing data, construct nursing task semantics based on the basic nursing data, and obtain nursing task ontology data; and construct nursing task flow graphs based on the nursing task ontology data to obtain nursing task flow graph data. The terminal knowledge collaborative modeling module is used to construct a nursing terminal knowledge graph based on nursing workflow data, and to construct a multi-agent collaborative model based on the nursing terminal knowledge graph, thereby obtaining a nursing collaborative model. The intelligent scheduling and execution module is used to perform task coordination scheduling based on the nursing collaboration model to obtain nursing task coordination data. The scheduling feedback optimization module is used to acquire historical nursing data and optimize the task collaboration efficiency of nursing task collaboration data based on the historical nursing data to obtain optimized nursing task data.

Images