Reservation and resource scheduling system based on cognitive intelligence and self-evolution rule engine

Abstract

The invention provides a reservation and resource scheduling system based on cognitive intelligence and a self-evolution rule engine, and the system comprises a rule cognition and self-evolution engine, an enhanced visual configuration, deduction and auditing engine, and a dynamic loading, real-time decision and predictive optimization module. A rule cognition and self-evolution engine of the multi-target intelligent conflict resolution and collaborative scheduling module is used for constructing a knowledge graph and generating a reservation rule in a current state by adopting a deep learning algorithm; the enhanced visual configuration, deduction and auditing engine is used for providing a rule editing function for a user; the rule demand edited by the user is transmitted to the rule cognition and self-evolution engine for detection; the dynamic loading, real-time decision-making and predictive optimization module is used for enabling the changed rule to take effect immediately; and the multi-target intelligent conflict resolution and cooperative scheduling module is used for processing the conflict problem of the reservation requests or the conflict problem between the reservation requests and available resources.

Description

Technical Field

[0001] This invention belongs to the field of medical technology, and in particular relates to an appointment and resource scheduling system based on cognitive intelligence and a self-evolving rule engine. Background Technology

[0002] Currently, medical institutions widely use various appointment systems to improve service efficiency. However, existing technologies have the following limitations:

[0003] 1. Difficulty in rule solidification and maintenance: Traditional medical appointment systems mostly rely on hard-coded business rules. When medical processes or policies (such as adjustments to the time slots for specific examinations or changes in medical insurance reimbursement rules) change, code-level modifications and redeployment are often required, resulting in long development cycles, slow response times, and a lack of deep semantic understanding and dynamic adaptability to complex medical scenarios.

[0004] 2. Insufficient Visual Configuration Capabilities: Although some systems provide visual configuration tools, these tools typically only support basic conditional judgments and simple logical combinations. For complex, multi-dimensional, and highly dependent composite rules (e.g., appointment scenarios that simultaneously meet multiple conditions such as "patient needs to fast for examination," "patient is over 70 years old," "patient has a specific chronic disease history," "qualification needs to be verified through the medical insurance interface," and "patient specifies a doctor with specific qualifications"), existing visual tools are difficult to effectively express and configure.

[0005] 3. Rigid conflict resolution mechanisms: When faced with conflicts in appointment rules (such as emergency patients cutting in line and regular appointment patients competing for resources), existing systems mostly use preset fixed priority strategies. This approach lacks awareness and adaptation to the real-time dynamic changes in the medical environment, making it difficult to balance multiple goals such as ensuring fairness, efficiency, and patient satisfaction, and to achieve global optimization and personalized scheduling.

[0006] 4. Data Silos and Knowledge Barriers: Medical information systems (such as Electronic Medical Records (EMR), Picture Archiving and Communication Systems (PACS), and Laboratory Information Systems (LIS)) accumulate massive amounts of heterogeneous data, but this data often exists in silos, making it difficult to effectively integrate, understand, and utilize. Existing appointment systems lack the ability to extract medical knowledge from this data and apply it to appointment scheduling decisions, resulting in insufficient accuracy and rationality in appointment scheduling.

[0007] 5. Large hospitals need to process tens of thousands of appointment requests every day. Due to rule conflicts, resource misallocation, and misunderstandings of patient needs, it is estimated that about 8% of appointment requests fail to be processed. This not only affects the patient's medical experience, but may also bring potential medical risks.

[0008] 6. Functions such as online medical insurance verification, intelligent triage rules, and clinical pathway guideline recommendations often rely on scattered third-party system interfaces, lacking unified rule arrangement, intelligent decision-making, and collaborative calling capabilities, resulting in low overall system efficiency.

[0009] 7. In the event of a public health emergency, the response time for manually adjusting appointment rules and resource allocation strategies often exceeds 15 minutes, and it is difficult to make predictive resource allocation and emergency plan activation based on the development trend of the event.

[0010] 8. Insufficient fulfillment of personalized medical needs: Traditional systems struggle to achieve highly personalized appointment scheduling and precise matching of medical resources based on a patient's specific condition severity, urgency, past medical history, drug allergy history, and even genetic information.

[0011] Therefore, there is an urgent need for a new medical appointment and resource scheduling solution that can intelligently understand medical rules, dynamically adapt to changes in the medical environment, efficiently handle complex appointment scenarios, and continuously optimize itself. Summary of the Invention

[0012] Purpose of the invention: To address the problems of rigid rules, inflexible conflict resolution mechanisms, insufficient data utilization, and inability to dynamically adapt to complex medical scenarios in existing medical appointment systems, this invention proposes an appointment and resource scheduling system based on cognitive intelligence and a self-evolving rule engine to achieve intelligent generation, dynamic optimization, and efficient execution of appointment rules.

[0013] Technical solution: This invention discloses a reservation and resource scheduling system based on cognitive intelligence and a self-evolving rule engine, including a rule cognition and self-evolving engine, an enhanced visual configuration, deduction and auditing engine, a dynamic loading, real-time decision-making and predictive optimization module, and a multi-objective intelligent conflict resolution and collaborative scheduling module.

[0014] The rule cognition and self-evolution engine is used to construct a knowledge graph and uses deep learning algorithms to generate reservation rules in the current state.

[0015] An enhanced visual configuration, deduction, and auditing engine provides users with rule editing capabilities and transmits the user-edited rule requirements to the rule cognition and self-evolution engine for testing.

[0016] The dynamic loading, real-time decision-making, and predictive optimization modules are used to ensure that changed rules take effect immediately.

[0017] The multi-objective intelligent conflict resolution and collaborative scheduling module is used to handle conflicts between reservation requests or between reservation requests and available resources.

[0018] Furthermore, the rule cognition and self-evolution engine includes a rule definition module, a knowledge graph module, and a rule effect evaluation and self-evolution module;

[0019] The rule definition module includes a natural language processing module, an API callback rule integration module, and a clinical guideline document parsing module;

[0020] The natural language processing module converts the appointment rule requirements described by medical staff or users in natural language into structured, machine-executable language;

[0021] The API callback rule integration module is used to call external rules via API when the executed reservation rule needs to call external rules.

[0022] The clinical guideline document parsing module uses NLP technology to automatically extract and structure rules related to appointment scheduling from clinical guidelines, and transforms them into knowledge in a knowledge graph or directly generates candidate rules.

[0023] The knowledge graph module includes a data extraction module, a knowledge graph construction module, and a knowledge graph learning module;

[0024] The data extraction module is used to extract entities, relationships between entities, and entity attributes from structured and unstructured data. The structured data is data with a fixed format and stored in a database, including appointment records, patient information, and doctor schedules. The unstructured data is data without a fixed format, including electronic medical record texts, medical literature abstracts, clinical treatment guidelines, and imaging reports.

[0025] The knowledge graph construction module is used to construct a medical appointment knowledge graph based on entities, relationships between entities, and entity attributes.

[0026] The knowledge graph learning module uses a graph neural network to learn the medical appointment knowledge graph to generate vector representations of nodes and edges, thereby capturing deep semantic relationships between entities.

[0027] The rule effectiveness evaluation and self-evolution module is used to generate reservation rules based on the rules defined by the knowledge graph and the rule definition module.

[0028] Furthermore, the natural language processing module also includes an NLG module, used to convert the generated rules into natural language;

[0029] The rule effectiveness evaluation and self-evolution module includes a data collection module and a self-evolution module; the data collection module collects multi-dimensional data related to the execution of the appointment rules, including appointment success rate, patient waiting time, resource utilization rate, patient satisfaction score, clinical pathway compliance, and rule conflict rate.

[0030] The self-evolution module uses a deep reinforcement learning algorithm to treat the parameters and rule structure of the rule engine as actions, and the contribution of each rule or combination of rules to the key performance indicators as a reward function. The algorithm then performs adaptive optimization and evolution of the rules.

[0031] The key performance indicators include patient experience, operational efficiency, and medical quality and safety.

[0032] Furthermore, the enhanced visual configuration, inference, and auditing engine includes a visual rule editor, a priority dynamic adjustment and personalized strategy generation module, and a rule version control and traceable auditing module;

[0033] The visual rule editor provides a graphical interface with a component panel, a canvas, and a logic conversion module. The component panel contains conditional components, logic components, and action components. Users drag components onto the canvas area and combine and connect the components in the logic components with the components in the conditional and action components to form a logic flowchart. The logic conversion module is used to parse the rules saved on the canvas and translate them into a format that the rule engine can execute.

[0034] The enhanced visual configuration, inference, and auditing engine binds the rules set by the user in the canvas to the knowledge graph through the unique identifier in the rule element;

[0035] The priority dynamic adjustment and personalized strategy generation module is used by administrators to configure basic priority model parameters and view the dynamic priority strategy adjusted based on historical data and patient profiles.

[0036] The rule version control and traceable audit module is used to save all rules for deactivation, creation, modification, activation, and version rollback, forming a complete historical log.

[0037] Furthermore, the enhanced visualization configuration, simulation, and auditing engine also includes a simulation module, allowing users to create or modify rules in a sandbox environment and select to simulate real medical operation scenarios to verify the effectiveness and security of the rules.

[0038] Furthermore, the dynamic loading, real-time decision-making, and predictive optimization module includes a rule hot loading and incremental update module, a context awareness and real-time decision-making module, and a predictive analysis and proactive intervention module.

[0039] The rule hot loading and incremental update module adopts hot loading technology, which enables new rules or rule updates to be dynamically loaded into the running rule engine. The incremental update algorithm only transmits and replaces the affected decision nodes in the rule network or the relevant subgraphs in the knowledge graph, minimizing the performance jitter and service interruption time during system updates.

[0040] The context awareness and real-time decision-making module perceives the current medical service context information and dynamically adjusts the execution path, parameters, or resource reservation strategy of the rules; the current medical service context information includes the real-time number of patients in each department / doctor, the doctor's current working status, the real-time occupancy of examination equipment, the potential impact of future short-term weather forecasts on the number of patients in specific departments, and the patient's historical medical behavior patterns.

[0041] The predictive analytics and proactive intervention module uses time series forecasting models to predict peak appointment demand, patient volume in specific departments, and peak usage of specific examination equipment over a future period. When resource shortages, conflict risks, or service bottlenecks are predicted, it triggers early warnings and automatically initiates or recommends corresponding resource scheduling plans to the administrator.

[0042] Furthermore, when a conflict occurs, the multi-objective intelligent conflict resolution and collaborative scheduling module employs a multi-objective optimization algorithm to comprehensively consider multiple objective functions, including patient waiting time, resource utilization, medical quality and safety, in order to calculate and generate a Pareto optimal scheduling strategy.

[0043] Furthermore, the system also includes an external medical information system connected to the rule-based cognition and self-evolution engine, a third-party service module, and a user interaction module; the external medical information system includes an electronic medical record system, a hospital information system, a laboratory information system, and a medical image archiving and communication system; the third-party service module includes a medical insurance interface and a drug database.

[0044] Beneficial Effects: This invention achieves a deep cognitive understanding of medical appointment-related rules (including business rules, clinical guidelines, expert experience, etc.) and empowers the rule engine with the ability to learn and evolve based on actual operational results, dynamically optimizing rule parameters and the rule structure itself through technologies such as reinforcement learning and graph neural networks. This invention supports multimodal, context-aware intelligent rule generation and input methods, including Natural Language Understanding (NLP) and Generation (NLG), graphical drag-and-drop configuration, application programming interface (API) linkage, and automatic parsing and extraction of rules from documents such as clinical guidelines. This invention provides real-time appointment conflict prediction, intelligent conflict resolution mechanisms, and flexible resource scheduling strategies to ensure service quality and patient safety in scenarios such as emergency rooms, high-concurrency appointments, and complex multi-departmental collaborations. This invention constructs and utilizes a medical appointment knowledge graph to achieve knowledge-based reasoning and decision support, thereby improving the accuracy, rationality, and personalization of medical appointments. This invention can provide interpretable artificial intelligence (AI) decision-making paths, enhancing medical staff's and managers' understanding, trust, and controllability of the system's automated decision-making process. Attached Figure Description

[0045] Figure 1 This is an overall block diagram of the present invention.

[0046] Figure 2 For AI-driven rule cognition and self-evolution engine workflow.

[0047] Figure 3 This is a flowchart of the multimodal intelligent rule input and understanding module.

[0048] Figure 4 This is a flowchart of the predictive analytics and proactive intervention module.

[0049] Figure 5 A flowchart of the workflow for a multi-objective intelligent conflict resolution and collaborative scheduling algorithm in a medical setting. Detailed Implementation

[0050] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0051] like Figure 1 As shown, this invention discloses an appointment and resource scheduling system based on cognitive intelligence and a self-evolving rule engine. The core of the system is an AI-driven rule cognition and self-evolving engine 100. This engine works closely with an enhanced visualization configuration, deduction and auditing engine 200, a dynamic loading, real-time decision-making and predictive optimization module 300, and a multi-objective intelligent conflict resolution and collaborative scheduling module 400 for medical scenarios. It also interacts with external medical information systems (such as EMR, HIS, LIS, PACS), third-party services (medical insurance interface, drug database), and user interfaces (APP, Web) through a data interface 500.

[0052] like Figure 2 As shown, the AI-driven rule cognition and self-evolution engine is responsible for the lifecycle management of medical rules, including rule understanding, generation, evaluation, and evolution. The construction of the AI-driven rule cognition and self-evolution engine includes: a medical knowledge graph construction and fusion module 110; a multi-source rule intelligent understanding and generation module 120; and a rule effect evaluation and self-evolution module 130.

[0053] The medical knowledge graph construction and fusion module 110 includes the following modules:

[0054] Data Extraction and Fusion: From structured data (such as appointment records, patient basic information, and doctor schedules in databases) and unstructured / semi-structured data (such as electronic medical record texts, medical literature abstracts, clinical practice guidelines PDF / Word documents, and image reports), natural language processing (NLP) techniques (such as named entity recognition (NER) and relation extraction (RE)) and ontology engineering are used to extract medical entities (such as diseases, symptoms, examination items, surgeries, drugs, doctors, departments, and equipment), relationships between entities (such as disease-recommended examinations, doctor-specialty diseases, drug-examination contraindications, and examination-required equipment), and entity attributes.

[0055] Knowledge Graph Construction: The extracted entities, relationships, and attributes are integrated to construct a Medical Appointment Knowledge Graph (MAKG). This knowledge graph provides rich background knowledge for subsequent rule understanding, reasoning, and decision-making.

[0056] Knowledge Graph Learning and Representation: Graph Neural Networks (GNNs) (111), such as GraphSAGE or GAT, are used to learn vector representations of nodes (entities) and edges (relationships) in a knowledge graph, capturing deep semantic relationships between entities to improve the accuracy of rule generation and reasoning ability. For example, GNNs can be used to discover strong associations between latent diseases and examination items, or to predict the preferences of specific patient groups for certain appointment arrangements.

[0057] like Figure 3 As shown, the multi-source rule intelligent understanding and generation module 120 includes a natural language processing and generation interface, an API callback rule integration 122, a clinical guideline document parsing module 123, and a graphical rule configuration interface 124.

[0058] The Natural Language Processing (NLP) and Generation (NLG) interface allows healthcare professionals or administrators to describe appointment rule requirements using natural language (text or voice input; voice input must first be converted to text via ASR speech recognition) (e.g., "Patients over 60 years of age with a history of diabetes and renal insufficiency must complete a renal function assessment within one week before undergoing a contrast-enhanced CT scan, and patients allergic to iodine-containing contrast agents will be automatically excluded"). The NLP module utilizes semantic analysis, intent recognition, and other technologies to transform these into structured, machine-executable rule representations (such as XML, JSON, or internal DSL). The NLG module then explains the machine rules or decision-making processes to the user using natural language.

[0059] API Callback Rule Integration 122: Supports calling external APIs within defined rule logic. For example, when executing an appointment rule, it can be configured to call the medical insurance verification interface to verify the patient's medical insurance eligibility and reimbursement scope in real time, or call the drug interaction database to query whether there are any contraindications between the patient's current medication and the scheduled examination.

[0060] In this embodiment, the decision to call an external API is embedded in the rule logic during the rule design phase. When administrators or medical staff create or edit rules through the graphical rule configuration interface 124, a node in the rule logic flow can be an "API call" action. Users can drag and drop this "API call" node into the rule's execution path (e.g., after a certain condition check) and configure specific parameters for it, including: Target API address (URL): The external service interface to be called. Input parameters: The data to be passed to the API, which can come from the currently processed appointment request (such as patient ID, application number). Output processing: Parsing the results returned by the API and using the results for subsequent rule checks or operations. Triggering conditions: Explicitly defining that this API call will only be executed when specific conditions are met.

[0061] Knowledge Graph Association: The medical knowledge graph 110 not only stores medical entities and relationships, but also metadata for "actions" or "services." For example, the concept of "test information verification" can be associated with an external "test API service" in the knowledge graph. When a rule uses the business logic of "verifying test data," the rule engine automatically determines, by querying the knowledge graph, that the corresponding external API needs to be called to complete this task. This approach decouples business logic from specific technical implementations, making the rules more readable and maintainable.

[0062] Rule templates: For common scenarios that require calling external APIs (such as "new patient registration" and "online medical insurance settlement"), the system can provide preset rule templates. When users create rules using templates, the steps for calling the API are already pre-defined, and users only need to fill in a small amount of necessary configuration information.

[0063] Clinical Guideline Document Parsing Module 123: Utilizes NLP technologies (such as document structure analysis and information extraction) to automatically extract and structure rules and clauses related to appointment scheduling from clinical guidelines and expert consensus documents in PDF, Word, and other formats, and transform them into knowledge in a knowledge graph or directly generate candidate rules.

[0064] The graphical rule configuration interface 124 provides a visual rule editor (integrated into module 200, but its rule generation logic is driven by this module). Users can build rules by dragging and dropping components and connecting logic gates (AND / OR / NOT). The system can intelligently recommend rule components or templates based on knowledge graphs and historical data.

[0065] Rule effectiveness evaluation and self-evolution module 130 includes the following modules:

[0066] Data collection and monitoring: Through system tracking, log analysis and other methods, we continuously collect multi-dimensional data related to the effectiveness of rule execution, including appointment success rate, patient waiting time, resource (doctor, equipment, room) utilization rate, patient satisfaction score, clinical pathway compliance, rule conflict rate and so on.

[0067] Performance evaluation model: Establish an evaluation model to quantify the contribution of each rule or combination of rules to key performance indicators (KPIs);

[0068] In this embodiment, key performance indicators (KPIs) refer to quantifiable performance metrics used to measure the effectiveness of the appointment and resource scheduling method described in this invention in achieving its optimization objectives. These KPIs are the direct source of the "reward signal" in the deep reinforcement learning model 131. Key performance indicators primarily include patient experience, operational efficiency, and medical quality and safety.

[0069] Patient experience includes: average wait time (the average time from when a patient successfully makes an appointment to when they actually receive the service (such as a consultation or examination). The goal is to minimize this time); appointment success rate (the percentage of patient-initiated appointment requests that are ultimately successfully confirmed. The goal is to maximize this rate); patient satisfaction rating (patient ratings of the appointment process and outcome collected through questionnaires and other methods); and appointment change / cancellation rate (the percentage of appointments changed or cancelled due to system issues or patient dissatisfaction).

[0070] Operational efficiency dimensions include: resource utilization (the proportion of effective usage time of key medical resources such as doctors' working hours, specific equipment hours, operating rooms, and consultation rooms to the total available time. The goal is to maximize this utilization and avoid resource idleness or over-occupancy); and average daily appointment processing volume (the number of appointment requests that the system can successfully process each day).

[0071] The dimensions of medical quality and safety include: clinical pathway compliance (the degree to which appointment scheduling follows relevant clinical guidelines or pathway recommendations), rule conflict rate (the frequency of rule conflicts during scheduling. The goal is to minimize this frequency), and high-risk event incidence rate (the incidence of medical risk events that may result from inappropriate scheduling (such as scheduling a patient with a drug allergy for a contraindicated examination).

[0072] Self-evolution mechanism: A deep reinforcement learning (DRL) model (131) is adopted, such as the A3C (Asynchronous Advantage Actor-Critic) or PPO (Proximal Policy Optimization) algorithm. The parameters pointed to by the rule engine (such as the appointment time threshold for specific examination items, the priority weight of emergency queue jumping, the tendency of resource allocation, etc.) and the rule structure (such as the addition and deletion of condition nodes, the activation / disabling of rules, and the adjustment of the triggering order between rules) are used as the action space of the DRL Agent, and the changes of KPIs are used as the reward signal. Through continuous "trial and error" and learning, the DRL Agent can dynamically adjust the rule parameters and structure to maximize the long-term cumulative reward, thereby realizing the adaptive optimization and evolution of rules.

[0073] In the deep reinforcement learning model 131, "state" is the complete set of information that the DRL Agent can observe at any given time point, describing the current system environment. This state is the basis for the Agent's decision to make the next "action" (i.e., adjust rule parameters or structure). In this invention, "state" is a high-dimensional vector, mainly including resource state, demand state, system performance state, current rule set state, and external environment state.

[0074] Resource status includes: 1. Real-time availability, future scheduling, and current load rate of each medical resource (doctors, equipment, clinics, beds, etc.). 2. Specific attributes of resources, such as doctor's expertise and equipment accuracy level.

[0075] Demand status includes: 1. The length of the current queue of pending appointment requests and the distribution of request types (e.g., emergency, general, follow-up). 2. Forecast values ​​of short-term and long-term appointment demand. 3. Patient information, such as the patient's disease type, severity, VIP level, and historical medical behavior (e.g., no-show rate).

[0076] System performance status includes the actual values ​​of key performance indicators (KPIs) over a recent period, such as average wait time over the past hour, daily resource utilization, and the number of recent conflicts. This provides the agent with immediate feedback on the effectiveness of its previous actions.

[0077] The current state of the rule set includes key parameter values ​​from the currently effective rule set, such as the weighting coefficients for emergency priority, time window settings for each examination item, and bias scores for resource allocation. The agent needs to know these parameters to determine how to adjust the rules.

[0078] External environmental conditions include: 1. Time information (such as time of day, day of the week, and whether it is a holiday).

[0079] If the system detects that the rule of "scheduling certain types of fasting examinations between 8-10 AM" leads to excessive concentration of appointments during that time slot, while the resource availability rate is high between 10-12 PM, and the knowledge graph shows a potential interaction between this examination and certain medications taken by some patients after dinner (affecting their fasting status), the self-evolutionary module may not only adjust through DRL (Device Detailed Rule) to suggest extending the "fasting period" for this examination to 11 AM, but may also generate a new auxiliary rule or prompt based on knowledge graph learning: "If the patient took XXX medication last night, it is recommended that they schedule a fasting examination after 10 AM," and add guidance on medication taken after dinner to the appointment reminder.

[0080] The enhanced visual configuration, simulation, and auditing engine was built and put into operation. This engine provides users with a user-friendly, powerful, and controllable interface for rule management and system monitoring.

[0081] 2.1 Visual Rule Editor: Provides a user-friendly graphical interface that supports drag-and-drop rule creation, editing, and combination. Rule elements can be associated with entities in the medical knowledge graph to achieve intelligent suggestions and verification.

[0082] The implementation of drag-and-drop rule creation in this embodiment is mainly divided into two levels: front-end interaction and back-end transformation.

[0083] Front-end implementation (user interface layer):

[0084] Component Palette: Provides a "toolbox" on the left or top of the interface, containing the basic elements that make up the rules, such as:

[0085] Conditional Components (IF): Used for judgments such as "patient age", "examination items", "disease diagnosis", etc.

[0086] Logic components: such as AND, OR, and NOT, used to combine multiple conditions. Action components: such as "Approve Appointment," "Send Reminder," "Call API," and "Recommend Doctor."

[0087] Canvas: Users can drag and drop components from the panel onto the main workspace (canvas).

[0088] Connections and Layout: Users can use the mouse to "draw lines" between connection points of components to represent the logical flow (for example, connecting the "true" exit of an "IF" component to an "ACTION" component). Front-end frameworks (such as React Flow or GoJS) are responsible for rendering these nodes and connections, forming a visual logical flowchart.

[0089] Configuration Panel: Clicking on any component on the canvas will bring up its property configuration panel on the right. For example, clicking on a "Patient Age" condition component allows you to enter a comparison operator (greater than, less than, equal to) and a specific value (such as "60").

[0090] Backend implementation (logic transformation layer):

[0091] When a user saves the rules on the canvas, the front-end serializes this visual flowchart into a structured data format, typically JSON. This JSON object fully describes the type, configuration parameters, and connections between each node (component). Upon receiving this JSON, the back-end's AI-driven rule recognition and self-evolution engine 100 parses it and translates it into a format that the rule engine can understand and execute (e.g., a domain-specific language (DSL), an XML format like Drools's DRL file, or directly compiled into an executable decision tree / network structure).

[0092] The process of associating user-defined rules with entities in the medical knowledge graph is as follows:

[0093] Triggering association: In the configuration panel on the front end, when a user needs to specify a medical concept for a condition or action component (for example, selecting "disease" to be judged in the condition component), the system will trigger an interaction with the knowledge graph.

[0094] Intelligent Search and Recommendation: An input box will be provided in the configuration panel. When the user enters keywords (such as "diabetes" or "kidney disease"), the system will initiate a query to the medical knowledge graph (110) in real time.

[0095] Returning structured entities: The knowledge graph does not return simple text matching, but a structured list of entities. For example, inputting "kidney disease" may return multiple specific entities such as "chronic kidney disease (CKD)" and "nephrotic syndrome", each with a unique identifier (URI) and hierarchical relationship (such as stages 1-5 under CKD).

[0096] User selection and binding: The user selects the most accurate entity from the list, such as "chronic kidney disease (CKD)". In the backend, the JSON representation of this rule component no longer stores a vague string "kidney disease", but instead stores the unique ID or URI of the entity "chronic kidney disease (CKD)" in the knowledge graph.

[0097] Context awareness and attribute population: Once a knowledge graph entity is bound, the system can obtain richer context. For example, after binding "chronic kidney disease (CKD)," the configuration panel can automatically load all relevant attributes of the disease (such as "stage" and "common complications") from the knowledge graph and allow the user to directly select these attributes for the next step of judgment (for example, selecting the "stage" attribute and setting the condition to "greater than or equal to 3").

[0098] 2.2 Rule Sandbox and Simulation Environment 210:

[0099] Users can create or modify rules in the sandbox environment and choose to simulate real medical operation scenarios (such as importing historical high-concurrency registration data, simulating sudden failures of specific equipment, and setting parameters for seasonal high-incidence periods of diseases).

[0100] The system visually displays the multi-dimensional impact predictions that rule changes may bring (such as changes in the distribution of average patient waiting time, changes in resource load rates of specific departments / doctors, and the potential number and extent of new conflict points). After verifying the effectiveness and security of the rules, it supports automated canary releases or A / B testing to gradually promote the new rules.

[0101] 2.3 Priority Dynamic Adjustment and Personalized Strategy Generation Interface 220: Allows administrators to configure basic priority model parameters and view the dynamic priority strategy trained and adjusted by AI based on historical data and patient profiles (including disease stage, risk level, allergy history, VIP level, historical compliance, etc.).

[0102] When the system detects a rule conflict, it can not only ensure priority for emergency care, but also generate and apply personalized appointment adjustment strategies for different patient groups (such as pregnant women, patients undergoing radiotherapy and chemotherapy for tumors, patients with rare diseases, and elderly and frail patients) based on preset strategies and real-time assessments.

[0103] The rules in this embodiment come from various sources and are not all created by the learning model. Rules can be created by humans through a visual interface or natural language, or they can be generated by parsing from clinical guideline documents. The learning model (especially the DRL model 131) plays more of the role of "optimizer" and "evolver", that is, based on the existing rule set, it fine-tunes the parameters or proposes structural modification suggestions based on the performance data.

[0104] Conflict Resolution Mechanism: When the system detects a rule conflict (e.g., two different high-priority rules simultaneously vying for the same indivisible resource, such as the last operating room), the system does not directly modify the rules themselves. Instead, it initiates a real-time, multi-objective decision-making process to resolve this specific conflict instance. This process is handled by the medical scenario multi-objective intelligent conflict resolution and collaborative scheduling module 400, and the specific steps are as follows:

[0105] Conflict identification: When performing reservation allocation, the real-time decision module 320 detects resource contention or logical contradictions, thereby identifying conflicts.

[0106] Activate the multi-objective optimization model 410: The system takes the conflicting parties (such as two appointment requests) and related contextual information (patient information, request urgency, resource status, etc.) as input and activates the preset multi-objective optimization algorithm (such as NSGA-II 411).

[0107] Evaluating Alternatives: The optimization algorithm quickly generates and evaluates multiple possible solutions. Decision-Making Based on Multiple Objectives: The algorithm scores each alternative based on several pre-defined objective functions (i.e., the aforementioned KPIs, such as fairness, efficiency, patient experience, and medical safety). This is where the dynamic prioritization strategy comes into play: it's not a simple static list of "Emergency > VIP > General." It's a dynamic weighting model that temporarily increases the weight of certain KPIs based on the patient's specific information (e.g., pregnant women, patients undergoing chemotherapy, elderly or frail patients) and the current medical environment. For example, for high-risk pregnant women, the weights of "medical safety" and "urgency" will be much higher than "resource utilization."

[0108] Selecting the optimal solution: The algorithm will find one or a set of "Pareto optimal solutions" (i.e. solutions that cannot improve any other objective without harming any other objective). The system can automatically select one of them (such as the one with the highest overall score) or recommend this set of optimal solutions to a human operator for final decision-making.

[0109] Execution and Recording: The system executes the final selected solution and records in detail the occurrence, decision-making process, and outcome of the conflict. These records serve as valuable data and are fed back to the rule self-evolution module (130). The DRL model learns from these successful conflict resolution cases and may optimize the rules in the future to fundamentally reduce the occurrence of similar conflicts.

[0110] 2.4 Rule Version Control and Traceable Audit Module 230: This module meticulously records all rule creation, modification, activation, deactivation, and version rollback operations (including rule changes generated by AI-driven automatic evolution), forming a complete change history log. This ensures that the rule change process is traceable and auditable, meeting the compliance requirements of the healthcare industry.

[0111] 2.5 Explainable AI (XAI) Interface 240):

[0112] For rules automatically adjusted or generated by AI, and complex decision-making processes (such as conflict resolution), provide a visual explanation of the decision-making path. For example, using techniques such as decision trees, SHAP values ​​(SHapley Additive exPlanations), or LIME (Local Interpretable Model-agnostic Explanations), explain to users which data features, knowledge from the knowledge graph, and logical reasoning process the decision is based on, thereby enhancing users' trust in and control over AI decision-making.

[0113] The module for dynamic loading, real-time decision-making, and predictive optimization is 300. This module is responsible for the immediate application of rules and proactive adjustments based on predictions.

[0114] 3.1 Rule Hot Loading and Incremental Update Mechanism 310:

[0115] When rules change (either manually or through AI self-evolution), there is no need to restart the entire system or service. Utilizing hot-loading rule technology, new rules or rule updates can be dynamically loaded into the running rule engine.

[0116] By using incremental update algorithms (such as those based on Delta Encoding), only affected decision nodes in the rule network or relevant subgraphs in the knowledge graph are transmitted and replaced, minimizing performance jitter and service interruption time during system updates.

[0117] 3.2 Context Awareness and Real-Time Decision Making 320:

[0118] When executing rules, the rule engine not only relies on static rules but also perceives the current medical service context in real time. This information includes: the real-time number of patients in each department / doctor, the doctor's current work status (e.g., in surgery, in an outpatient clinic, temporarily away), the real-time occupancy of examination equipment, the potential impact of short-term weather forecasts on the number of patients in specific departments (e.g., orthopedics, respiratory medicine), patients' historical medical behavior patterns (e.g., habitual lateness, cancellation rate), and the behavior patterns of similar patient groups. Based on this real-time context, the engine dynamically adjusts the rule execution path, parameters, or resource reservation strategies.

[0119] 3.3 Predictive Analysis and Proactive Intervention Module 330, this module as follows Figure 4 As shown:

[0120] Based on historical appointment data, patient flow data, seasonal disease spectrum data, and even external public health event data (such as epidemic reports), time series prediction models (such as LSTM-Long Short-Term Memory (331), ARIMA, Prophet, etc.) are used to predict peak appointment demand, visit volume in specific departments, and peak usage of specific examination equipment in the future (such as the next few hours, days, or weeks). When potential resource shortages, high conflict risks, or service bottlenecks are predicted, the system can trigger early warnings and automatically activate or recommend corresponding resource scheduling plans to the administrator (such as suggesting increased doctor scheduling, temporarily opening more appointment slots, guiding non-emergency patients to seek medical treatment at off-peak times, and pre-allocating mobile medical resources, etc.).

[0121] 4. Implementation of Multi-Objective Intelligent Conflict Resolution and Cooperative Scheduling Algorithms in Medical Scenarios (400 characters) Figure 5 As shown, this module is the core of the system for handling complex reservation scenarios and resolving resource conflicts.

[0122] 4.1 Multi-objective optimization model 410: When conflicts arise between appointment requests or between appointment requests and available resources, a multi-objective optimization algorithm is activated. This algorithm comprehensively considers multiple, sometimes contradictory, objectives, such as fairness (e.g., following a first-come, first-served principle, prioritizing ethical considerations (e.g., prioritizing critically ill patients)), efficiency (e.g., maximizing equipment utilization, balancing the workload of medical staff, and shortening the overall patient waiting time), urgency (ensuring timely treatment for emergency and high-priority patients), patient experience (minimizing patient waiting time and satisfying patients' doctor or time preferences as much as possible), and medical quality and safety (following clinical pathway guidelines and reducing the risk of complications or medical errors due to inappropriate scheduling).

[0123] The algorithms used in this embodiment may include improved evolutionary algorithms such as NSGA-II (Non-dominated Sorting Genetic Algorithm II) (411) and SPEA2 (Strength Pareto Evolutionary Algorithm 2), or methods based on multi-agent reinforcement learning (MARL). These algorithms can find a set of Pareto optimal solutions or reach a Nash equilibrium in a game environment for the system to select or recommend to the operator.

[0124] 4.2 Intelligent Conflict Resolution Strategy:

[0125] Based on the results of multi-objective optimization, the system automatically or semi-automatically performs conflict resolution. For example, in the event of a sudden CT equipment malfunction, the system automatically triggers the following example rule (this rule itself can also be generated or optimized by AI):

[0126] This rule, named "Equipment Failure Emergency and Intelligent Referral Scheduling_v2," defines how the system should automatically and intelligently handle affected patient appointments when a critical piece of equipment named CT01 fails. Its detailed logical steps are as follows:

[0127] Triggering condition: This rule will be automatically activated when the system detects that the status of the CT01 device has changed to "fault".

[0128] Immediate Response: Once the rule is triggered, the current appointment processing mode will be immediately adjusted to "emergency priority" to ensure that critically ill patients can receive the highest priority treatment when resources are scarce.

[0129] Intelligent Decision-Making Process (Multi-Level Degradation Strategy): Next, the system will initiate an intelligent, multi-level solution search process:

[0130] Step 1: Find alternative resources within the hospital.

[0131] The system will query its built-in medical knowledge graph to find other imaging devices within the hospital that can replace CT01. This query not only finds similar devices, but also considers the patient's specific needs (such as the urgency of the examination and the requirements for image quality) to ensure that the recommended alternative device is clinically appropriate.

[0132] Step 2: Perform branching based on the search results within the hospital.

[0133] If a suitable alternative within the hospital is found, the system will automatically select the best available device. Then, it will automatically reschedule the patient for this new device and its next available time slot. Finally, the system will send the patient a personalized notification, roughly stating: "Dear patient, your originally scheduled CT01 device has experienced a malfunction. The system has automatically rescheduled you for another device in our hospital at a new time… You can choose to accept this, request another rescheduling, or inquire about other options." This empowers patients with the right to know and the right to choose.

[0134] If no suitable alternative equipment is available within the hospital: the system will automatically expand its search scope and query the knowledge graph again, this time looking for available equipment resources at partner hospitals. Factors such as geographical location and cooperation agreements will be considered during the query. Based on the cross-hospital search results, further branching processing will be performed: if available resources are found at a partner hospital: the system will automatically initiate the cross-hospital referral coordination process and notify the patient: "Equipment malfunction, no alternative available within the hospital, we are coordinating your examination at a partner hospital, please wait for further notification." If even partner hospitals do not have available resources: the system determines that the problem is beyond its automated processing capacity. In this case, it will immediately report it to manual processing. The system will automatically generate an emergency work order, clearly describing the problem: "CT01 equipment malfunction, and there is no internal or partner backup equipment, affecting the examination of patient [specific urgency level]." This allows relevant personnel to quickly intervene and coordinate manually.

[0135] 4.3 Multi-Resource Collaborative Scheduling: For complex medical events requiring the coordination of multiple resources (such as major surgeries), the system not only considers the availability of core resources (such as operating rooms), but also needs to synchronously coordinate the surgeon team (considering their qualifications, experience, and schedules), anesthesiologist team, circulating / scrubbing nurses, special surgical instruments (such as the da Vinci robot), intraoperative consumables, postoperative ICU beds, and even blood sources from the transfusion department through a rule engine and collaborative scheduling algorithm. By utilizing a Constraint Satisfaction Problem (CSP) solver or combining it with multi-objective optimization algorithms, the system ensures that all relevant resources are effectively matched and reserved in terms of both time and logic.

[0136] Example 1: Intelligent green channel decision-making and resource reservation for emergency PCI patients

[0137] Scenario: A patient initiates a request through the emergency vehicle network system (or a family member initiates an emergency help request through the hospital's APP) stating "Suspected acute myocardial infarction, conscious but with unstable blood pressure, requiring emergency PCI surgery".

[0138] System processing flow:

[0139] Request Reception and Preliminary Parsing: The system receives the request. The NLP component 121 of the multi-source rule intelligent understanding and generation module 120 in the AI-driven rule cognition and self-evolution engine 100 parses the keywords "acute myocardial infarction" and "emergency PCI" in the request.

[0140] Knowledge Graph Query and Resource Requirements Confirmation: The system queries the Medical Knowledge Graph (MAKG) 110 to confirm the typical resource requirements for "PCI surgery for acute myocardial infarction", including: a qualified catheterization lab (including DSA equipment), a team of interventional cardiologists with PCI qualifications (chief surgeon and assistants), anesthesiologists, surgical nurses, specific types of catheters, stents and other consumables, and possible CCU beds after the operation.

[0141] Priority marking and conflict prediction: The multi-objective intelligent conflict resolution and collaborative scheduling module 400 in medical scenarios marks this request as the highest priority. The system immediately checks the current occupancy status of the catheterization lab and the scheduling for the next 2-3 hours.

[0142] Dynamic resource scheduling and reservation: If a non-emergency elective surgery is currently underway or about to begin in the catheterization lab, the multi-objective optimization model 410 triggers an "emergency priority" strategy, automatically assessing the impact of pausing or postponing the elective surgery (such as patient waiting time and physician scheduling conflicts), and generating instructions. The context-aware and real-time decision-making module 320 queries interventional physicians, anesthesiologists, and nurses who meet the qualifications and whose current status is "ready to be called upon" or "can finish the current task in a short time," and automatically sends an emergency team formation notification through the hospital's communication system.

[0143] The system queries the HIS / supply chain system to check the real-time inventory of key consumables required for PCI (such as specific model stents). If the inventory is insufficient or located in different warehouses, an emergency allocation process is triggered or an emergency requisition notice is sent to the pharmacy and medical device department.

[0144] Predict and reserve the time window required for surgery (e.g., an average of 2 hours) and postoperative CCU bed availability (e.g., an average of 24-48 hours).

[0145] Information Synchronization and Notification: The system automatically pushes PCI surgery information to the electronic whiteboards and surgery scheduling systems of relevant departments (catheterization lab, CCU, anesthesiology). It also sends basic patient information, estimated arrival time, and current vital signs to the personal terminals (e.g., apps, SMS) of members of the established medical team. If the patient is referred from a primary care hospital, the system attempts to automatically retrieve the patient's preliminary examination results (e.g., ECG, troponin) from the referring hospital via preset interfaces or standards (e.g., FHIR), providing decision-making support for the in-hospital emergency team and saving time.

[0146] Rule self-evolution: Data on various time points of this emergency treatment (from receiving the request to the patient entering the catheterization lab, from balloon dilation to restoration of blood flow, etc.), resource matching efficiency, and patient prognosis were collected. The DRL model 131 of the rule effectiveness evaluation and self-evolution module 130 will use this data to optimize resource matching rules, team notification priorities, and consumable warning thresholds in the emergency PCI process during subsequent training, with the aim of responding to similar events faster and better in the future.

[0147] Example 2: Generation of Complex Reservation Rules and Enhancement of Knowledge Graph Based on Natural Language Input

[0148] Scenario: Outpatient doctors set appointment rules for a specific patient group through voice or text input on the doctor's workstation: "For all patients with chronic kidney disease (CKD) stage 3 and above who need to undergo abdominal enhanced CT scans, the system should automatically prompt them to have a kidney function re-examination within one week when the patient makes an appointment, and prioritize these examinations to be completed in the morning. At the same time, patients should be reminded to fast for 4 hours before the examination and to pay attention to the risk of contrast agent kidney damage."

[0149] System processing flow:

[0150] Natural Language Input and Parsing: Doctors input the above requirements via microphone voice input (converted by ASR) or keyboard text input. The NLP component 121 in the AI-driven rule cognition and self-evolution engine (100) performs semantic analysis on the text: 1. Identify the core intent: Create an appointment rule for a specific examination for a specific patient group. 2. Extract key entities: Examination items (abdominal enhanced CT), patient conditions (chronic kidney disease stage 3 and above), actions (prompt for renal function re-examination, priority for morning, fasting for 4 hours, attention to contrast agent risks).

[0151] Knowledge Graph Query and Information Enhancement: 1. The NLP component links "Chronic Kidney Disease Stage 3 and Above" with the disease ontology (such as the CKD staging definition based on ICD-10 encoding) in the Medical Knowledge Graph (MAKG) (110) to clarify its precise definition. 2. Querying the knowledge graph to find the association between "Abdominal Enhanced CT" and "Chronic Kidney Disease" may reveal relevant clinical guidelines recommending that for patients with CKD Stage 3 and above, hypotonic or isotonic non-ionic iodine contrast agents should be used, and adequate hydration should be performed to reduce the risk of contrast-induced nephropathy (CIN).

[0152] Structured rule generation: NLP component 121 uses enhanced information obtained from the knowledge graph to transform natural language descriptions into structured rules.

[0153] Rule validation and deployment: The generated rules can be previewed and simulated in the rule sandbox (210) of the enhanced visual configuration, simulation and auditing engine (200), and doctors can confirm whether its logic meets expectations. After confirmation, the rules are deployed to the production environment through a dynamic loading and incremental update mechanism (310) and take effect immediately.

[0154] Explainability: The XAI interface 240 can show doctors how this rule was generated step by step from their natural language input and in conjunction with which entries in the knowledge graph (such as CKD staging definitions and contrast agent guidelines).

[0155] Rule self-evolution: Subsequently, the system will monitor the execution effect of this rule, such as whether any patients have delayed their examinations due to failure to have their kidney function re-examinations in a timely manner, or whether such patients are too concentrated in the morning. The DRL model (131) may fine-tune the time window of the "kidney function re-examination" reminder based on data feedback, or adjust the weight of "prioritizing the morning", or even suggest a more refined triage strategy.

[0156] Example 3: Dynamic Allocation of Outpatient Appointment Resources for Seasonal Respiratory Diseases Based on LSTM Prediction Model: With the arrival of autumn and winter, the system needs to cope with the surge in outpatient and examination demand brought about by the high incidence of respiratory diseases such as influenza. System Processing Flow:

[0157] Data Collection and Model Training: Predictive Analytics and Proactive Intervention Module 330 Continuously collects historical data: daily / weekly outpatient appointments, actual visits, and requests for related examinations (such as chest X-ray / CT, influenza virus nucleic acid / antigen testing) for each department (especially respiratory, pediatric, and infectious disease departments).

[0158] Integrate external data: local meteorological data (temperature and humidity changes), influenza indices or weekly infectious disease surveillance reports released by public health departments.

[0159] Use these multi-source time series data to train an LSTM model 331 (or other applicable models such as Prophet) to predict trends in respiratory outpatient visits and related examination needs over the next 1-4 weeks.

[0160] Demand Forecasting and Early Warning: At the end of September, based on the latest input data (such as the start of a temperature drop and a small number of reported influenza cases), the LSTM model predicted that within the next two weeks, the demand for respiratory outpatient services might increase by 30% compared to the baseline, and the demand for related examinations might increase by 25%. The system automatically generates early warnings and alerts hospital administrators and relevant department heads via dashboards or notifications.

[0161] The rules engine triggers the contingency plan: the set of rules for "seasonal disease peak response" preset in the AI-driven rule cognition and self-evolution engine 100 (or generated by AI based on historical experience) is activated.

[0162] The multi-objective intelligent conflict resolution and collaborative scheduling module for medical scenarios has begun executing the scheduling strategy in the contingency plan: Outpatient scheduling adjustment suggestion: The system pushes a suggestion to the head of the medical affairs department or respiratory department: "According to the forecast, the pressure on the respiratory outpatient department will increase in the next two weeks. Suggestions: 1. Respiratory doctors should add temporary outpatient units (such as adding afternoon or weekend appointments). 2. Consider drawing some doctors with experience in the diagnosis and treatment of respiratory diseases from other subspecialties of internal medicine to support the respiratory outpatient department (the 'cross-departmental support' sub-rule in the contingency plan needs to be activated, and the doctor's qualifications and scheduling conflicts need to be checked)."

[0163] Dynamic allocation of examination resources: The allocation ratio of appointment pools for chest X-ray / CT equipment in the radiology department and influenza testing windows in the laboratory department is automatically adjusted to appropriately increase the number of appointments available for respiratory disease-related examinations.

[0164] Patient triage guidance: For patients who have scheduled routine follow-up visits to the respiratory department (non-urgent, with stable conditions), the system will intelligently recommend in the patient's APP or SMS reminder: "Recently, it is the peak season for respiratory diseases. If your condition is stable, you may consider postponing your follow-up visit or trying our hospital's online consultation service to avoid peak hours."

[0165] Drug and consumable inventory alert: Based on predicted increases in outpatient and examination volumes, the system automatically calculates potential consumption of antiviral drugs, antipyretics, cough suppressants, test kits, etc., and compares this with current inventory. If the inventory falls below the safety stock threshold, a replenishment alert is sent to the pharmacy department and consumable warehouse.

[0166] Patient education and guidance: The system can be configured to proactively push seasonal influenza prevention and health care knowledge and vaccination reminders to patient groups (such as patients with underlying respiratory diseases such as chronic obstructive pulmonary disease and asthma) through hospital apps, official accounts and other channels, and inform the hospital of the preparations made to cope with the peak (such as increasing outpatient clinics and opening rapid testing channels), and guide patients to seek medical treatment in an orderly manner.

[0167] Effectiveness monitoring and model / rule iteration: The system continuously monitors indicators such as actual outpatient volume, waiting time, and patient satisfaction, and compares them with predicted values.

[0168] The predictive analytics module uses new real-world data to incrementally train or periodically retrain the LSTM model to improve the accuracy of future predictions.

[0169] The rule effectiveness evaluation and self-evolution module assess the effectiveness of various rule measures implemented to address the peak. The DRL model 131 may adjust parameters such as the trigger thresholds and resource allocation ratios for various measures in the contingency plan. For example, if it is found that the effect of drawing on doctors for support is not good (due to insufficient training or patient non-acceptance), the weight of this measure may be reduced in the future, and more emphasis may be placed on guiding online consultations.

[0170] Compared with traditional solutions, this invention has achieved significant improvements in several key performance indicators, as detailed below:

[0171] 1. Significantly improved efficiency in rule configuration and management: Through multimodal intelligent rule input (natural language, document parsing, and visual configuration), AI intelligent recommendation, and assisted generation, the average configuration time for complex medical appointment rules is reduced from approximately 2 hours (including coding adjustments) in the traditional approach to approximately 5 minutes in this approach, representing an efficiency improvement of up to 95.8%.

[0172] 2. Significantly faster emergency response and treatment decision-making speed: Through real-time conflict prediction, intelligent conflict resolution and automatic triggering of contingency plans, the system reduces the average response and treatment decision-making delay for emergency queue jumping or sudden resource shortages from more than 15 minutes in traditional manual coordination to less than 2 seconds in this solution, a reduction of up to 98.9%.

[0173] 3. Significantly improved system stability and reservation success rate: Through multi-objective intelligent conflict resolution and predictive analysis, the daily reservation failure rate caused by rule conflicts, resource mismatch, etc., is controlled from about 8% in the traditional system to less than 0.2% in this solution, a reduction of 97.5%.

[0174] 4. Significantly reduced rule optimization and maintenance costs: The AI ​​rule cognition and self-evolution module automatically optimizes rules through technologies such as DRL, replacing the thousands of dollars in manual analysis and tuning costs per month in traditional solutions. This achieves a near 100% reduction in manual costs for rule optimization and maintenance, while continuously maintaining the best performance of the rule set.

[0175] 5. Significantly improved patient experience: Through personalized appointment strategies, reduced waiting time, increased success rate and precise guidance, patient appointment satisfaction has increased from about 75% in the traditional approach to over 90% in this approach, an improvement of more than 20%.

[0176] 6. Effective improvement in the efficiency of medical resource utilization: Through collaborative scheduling and predictive resource allocation, the overall average utilization rate of medical equipment (such as MRI and CT) has been increased from about 70% to over 85%, an increase of about 21.4%.

[0177] 7. Enhanced ability to respond to public health emergencies: By combining predictive analysis, the response speed for rule adjustments to emergencies has been shortened from more than 15 minutes in the traditional manual mode to less than 1 minute in this plan (including the lead time for predictive adjustments), an improvement of more than 93.3%.

[0178] 8. Reduced workload for medical staff: Automated and intelligent rule management and conflict coordination significantly reduce the administrative burden and cognitive load of medical staff and appointment center personnel.

[0179] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.

Claims

1. A reservation and resource scheduling system based on cognitive intelligence and a self-evolving rule engine, characterized in that, It includes a rule-aware and self-evolution engine, an enhanced visual configuration, deduction and auditing engine, a dynamic loading, real-time decision-making and predictive optimization module, and a multi-objective intelligent conflict resolution and collaborative scheduling module; The rule cognition and self-evolution engine is used to construct a knowledge graph and uses deep learning algorithms to generate reservation rules in the current state. An enhanced visual configuration, deduction, and auditing engine provides users with rule editing capabilities and transmits the user-edited rule requirements to the rule cognition and self-evolution engine for testing. The dynamic loading, real-time decision-making, and predictive optimization modules are used to ensure that changed rules take effect immediately. The multi-objective intelligent conflict resolution and collaborative scheduling module is used to handle conflicts between reservation requests or between reservation requests and available resources.

2. The reservation and resource scheduling system based on cognitive intelligence and a self-evolving rule engine according to claim 1, characterized in that, The rule cognition and self-evolution engine includes a rule definition module, a knowledge graph module, and a rule effect evaluation and self-evolution module; The rule definition module includes a natural language processing module, an API callback rule integration module, and a clinical guideline document parsing module; The natural language processing module converts the appointment rule requirements described by medical staff or users in natural language into structured, machine-executable language; The API callback rule integration module is used to call external rules via API when the executed reservation rule needs to call external rules. The clinical guideline document parsing module uses NLP technology to automatically extract and structure rules related to appointment scheduling from clinical guidelines, and transforms them into knowledge in a knowledge graph or directly generates candidate rules. The knowledge graph module includes a data extraction module, a knowledge graph construction module, and a knowledge graph learning module; The data extraction module is used to extract entities, relationships between entities, and entity attributes from structured and unstructured data. The structured data is data with a fixed format and stored in a database, including appointment records, patient information, and doctor schedules. The unstructured data is data without a fixed format, including electronic medical record texts, medical literature abstracts, clinical treatment guidelines, and imaging reports. The knowledge graph construction module is used to construct a medical appointment knowledge graph based on entities, relationships between entities, and entity attributes. The knowledge graph learning module uses a graph neural network to learn the medical appointment knowledge graph to generate vector representations of nodes and edges, thereby capturing deep semantic relationships between entities. The rule effectiveness evaluation and self-evolution module is used to generate reservation rules based on the rules defined by the knowledge graph and the rule definition module.

3. The reservation and resource scheduling system based on cognitive intelligence and a self-evolving rule engine according to claim 2, characterized in that, The natural language processing module also includes an NLG module, which is used to convert the generated rules into natural language; The rule effectiveness evaluation and self-evolution module includes a data collection module and a self-evolution module; the data collection module collects multi-dimensional data related to the execution of the appointment rules, including appointment success rate, patient waiting time, resource utilization rate, patient satisfaction score, clinical pathway compliance, and rule conflict rate. The self-evolution module uses a deep reinforcement learning algorithm to treat the parameters and rule structure of the rule engine as actions, and the contribution of each rule or combination of rules to the key performance indicators as a reward function. The algorithm then performs adaptive optimization and evolution of the rules. The key performance indicators include patient experience, operational efficiency, and medical quality and safety.

4. The reservation and resource scheduling system based on cognitive intelligence and a self-evolving rule engine according to claim 1, characterized in that, The enhanced visual configuration, inference, and auditing engine includes a visual rule editor, a priority dynamic adjustment and personalized strategy generation module, and a rule version control and traceable auditing module. The visual rule editor provides a graphical interface with a component panel, a canvas, and a logic transformation module. The component panel has conditional components, logic components, and action components. Users drag components to the canvas area and combine and connect the components in the conditional components and action components through the components in the logic components to form a logic flowchart. The logic conversion module is used to parse the rules saved on the canvas and translate them into a format that the rule engine can execute. The enhanced visual configuration, inference, and auditing engine binds the rules set by the user in the canvas to the knowledge graph through the unique identifier in the rule element; The priority dynamic adjustment and personalized strategy generation module is used by administrators to configure basic priority model parameters and view the dynamic priority strategy adjusted based on historical data and patient profiles. The rule version control and traceable audit module is used to save all rules for deactivation, creation, modification, activation, and version rollback, forming a complete historical log.

5. The reservation and resource scheduling system based on cognitive intelligence and a self-evolving rule engine according to claim 1, characterized in that, The enhanced visualization configuration, simulation, and auditing engine also includes a simulation module, where users can create or modify rules in a sandbox environment and choose to simulate real medical operation scenarios to verify the effectiveness and security of the rules.

6. The reservation and resource scheduling system based on cognitive intelligence and a self-evolving rule engine according to claim 1, characterized in that, The dynamic loading, real-time decision-making, and predictive optimization module includes a rule hot loading and incremental update module, a context awareness and real-time decision-making module, and a predictive analysis and proactive intervention module. The rule hot loading and incremental update module adopts hot loading technology, which enables new rules or rule updates to be dynamically loaded into the running rule engine. The incremental update algorithm only transmits and replaces the affected decision nodes in the rule network or the relevant subgraphs in the knowledge graph, minimizing the performance jitter and service interruption time during system updates. The context awareness and real-time decision-making module perceives the current medical service context information and dynamically adjusts the execution path, parameters, or resource reservation strategy of the rules; the current medical service context information includes the real-time number of patients in each department / doctor, the doctor's current working status, the real-time occupancy of examination equipment, the potential impact of future short-term weather forecasts on the number of patients in specific departments, and the patient's historical medical behavior patterns. The predictive analytics and proactive intervention module uses time series forecasting models to predict peak appointment demand, patient volume in specific departments, and peak usage of specific examination equipment over a future period. When resource shortages, conflict risks, or service bottlenecks are predicted, it triggers early warnings and automatically initiates or recommends corresponding resource scheduling plans to the administrator.

7. The reservation and resource scheduling system based on cognitive intelligence and a self-evolving rule engine according to claim 1, characterized in that, When a conflict occurs, the multi-objective intelligent conflict resolution and collaborative scheduling module employs a multi-objective optimization algorithm to comprehensively consider multiple objective functions, including patient waiting time, resource utilization, medical quality and safety, in order to calculate and generate a Pareto optimal scheduling strategy.

8. The reservation and resource scheduling system based on cognitive intelligence and a self-evolving rule engine according to claim 1, characterized in that, The system also includes an external medical information system connected to the rule-based cognition and self-evolution engine, a third-party service module, and a user interaction module; the external medical information system includes an electronic medical record system, a hospital information system, a laboratory information system, and a medical image archiving and communication system; The third-party service module includes a medical insurance interface and a drug database.

Images