Hospital medical material circulation data anomaly early warning method and system

Abstract

The application discloses a hospital medical material flow data abnormality early warning method and system, and belongs to the technical field of electric digital data processing. The method constructs a dynamic flow topology graph by acquiring multi-source business data, uses a conservation consistency check and a graph matching algorithm to establish a mapping between end event nodes and billing nodes, and generates a closed loop alignment graph data structure; then, a reverse graph traversal is performed to calculate a quantity deviation, an attribute deviation degree and an isolated billing node quantity to generate an abnormal feature vector, and classification processing is performed to output an early warning result containing a local topology picture section. The application realizes fine-grained closed loop association of medical instrument life cycle data and department billing consumption data, solves the problem that multi-source traceability data is difficult to align in a medical material dynamic disassembly and assembly scene, improves the identification accuracy of abnormal material consumption, data chain breakage and inconsistent resource consumption data features, and realizes accurate early warning and traceable supervision of hospital medical material flow data abnormalities.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing technology, and in particular to a method and system for early warning of abnormal data in the circulation of hospital medical supplies. Background Technology

[0002] With the continuous advancement of healthcare informatization, the scale of data generated by hospital business systems continues to grow. A large amount of business and management data has been generated in areas such as drug and consumable procurement, medical equipment management, medical insurance settlement, and medical service behavior. Because the data sources involved in hospital operations are wide-ranging and the business relationships are complex, traditional risk control methods relying on manual inspection and experience-based judgment are insufficient to fully uncover the potential risk characteristics within the data, and cannot achieve real-time perception and accurate identification of risk factors.

[0003] Currently, early warning technologies for abnormal circulation of hospital medical supplies mainly revolve around the collection, integration, and analysis of medical business data and management data. Existing methods typically establish a data collection platform to aggregate diagnostic and treatment data, procurement data, financial data, asset data, and personnel information data from hospital information systems, and then analyze and evaluate relevant business behaviors in conjunction with a risk indicator system. In terms of data processing, some technologies employ data cleaning, data fusion, and statistical analysis to standardize data from different business systems, enabling the application of multi-source data. Regarding risk identification, existing research is gradually introducing technologies such as data mining, machine learning, and anomaly detection. By analyzing behavioral characteristics and correlation patterns in historical business data, risk identification models are formed, and risk alerts and warnings are issued based on the analysis results.

[0004] However, in the process of implementing the inventive technical solution in the embodiments of this application, it was found that the above-mentioned technology has at least the following technical problems:

[0005] In the process of early warning of abnormal data in the circulation of hospital medical supplies, for the scenarios of Central Sterile Supply Department (CSSD) and infection control material management, although existing technologies have introduced barcode traceability of sterilization packages and accumulated a large amount of IoT sensing data, there are still technical gaps at the underlying data processing level. Because disinfection items frequently undergo "unpacking" and "repackaging" operations during the multiple processes of collection, cleaning, packaging, sterilization, and distribution to departments, an extremely complex dynamic mapping relationship forms between the underlying IoT sensing data and the final departmental billing data. Existing data processing technologies lack graph matching algorithms to handle this dynamic combination topology, and cannot complete fine-grained closed-loop alignment of "reusable device lifecycle data" and "departmental billing consumption data" at the underlying data level. This break in the data association chain prevents the early warning model from accurately distinguishing between normal material consumption and maliciously inflated data characteristics through cross-comparison of multi-source data. Ultimately, this makes it difficult for the abnormal data identification model to effectively penetrate and accurately intercept hidden data anomalies such as quantity gaps and attribute mismatches. Summary of the Invention

[0006] To address the technical problems of existing technologies, such as the difficulty in accurately linking multi-source traceability data and operational consumption data due to the dynamic unpacking and repacking characteristics of disinfection supply materials in hospital hygiene management, which prevents closed-loop monitoring of the entire lifecycle of materials and makes it difficult for conventional anomaly detection models to accurately identify and effectively intercept hidden abnormal material consumption and abnormal data links, this invention provides a method and system for early warning of anomalies in hospital hygiene material circulation data. The technical solution is as follows:

[0007] On the one hand, a method for early warning of abnormalities in hospital medical supplies circulation data is provided. This method includes: acquiring multi-source business data within a target time period, including terminal department billing data and time-series physical circulation records of reusable medical devices; extracting device identifiers, device package identifiers, and operation type identifiers from the time-series physical circulation records; mapping device identifiers and device package identifiers to entity nodes, and operation type identifiers to event nodes corresponding to the operation type; establishing aggregated directed edges and branch directed edges between entity nodes and corresponding event nodes based on the merging and unmerging operations in the time-series physical circulation records to construct a dynamic circulation topology graph in memory; parsing terminal department billing data into billing nodes; verifying the conservation consistency between the number of input entities of the merging event node associated with the same device package identifier and the number of output entities of the corresponding unmerging event node in the dynamic circulation topology graph; and ensuring that the conservation consistency is satisfied. A graph matching algorithm is used to establish mapping edges between end event nodes and billing nodes in the dynamic flow topology graph to generate a closed-loop aligned graph data structure. Within this closed-loop aligned graph data structure, a reverse graph traversal algorithm is executed, starting from the billing node. This algorithm calculates the quantity deviation between the expected number of processed devices corresponding to the target processing event node and the number of traceable individual device entity nodes, the attribute deviation between entity nodes and billing nodes associated through mapping edges, and the number of isolated billing nodes lacking preceding physical flow paths. These calculations are then used to concatenate and generate anomaly feature vectors in the graph structure. A pre-defined risk identification model is invoked to classify these anomaly feature vectors to obtain a risk assessment status. When the risk assessment status is abnormal, corresponding local topology image segments are extracted from the closed-loop aligned graph data structure to generate and output a flow data anomaly warning result containing these local topology image segments.

[0008] On the other hand, a hospital medical supplies circulation data anomaly early warning system is provided, including: a medical data acquisition module, a dynamic topology construction module, a closed-loop alignment generation module, a risk feature extraction module, and a data anomaly early warning module. The medical data acquisition module is used to acquire multi-source business data within a target time period, including terminal department billing data and time-series physical circulation records of reusable medical devices. The dynamic topology construction module is used to extract device identifiers, device package identifiers, and operation type identifiers from the time-series physical circulation records, map device identifiers and device package identifiers to entity nodes, map operation type identifiers to event nodes of the corresponding operation type, and establish aggregated directed edges and branch directed edges between entity nodes and corresponding event nodes based on the merging and unmerging operations in the time-series physical circulation records to construct a dynamic circulation topology graph in memory. The closed-loop alignment generation module is used to parse terminal department billing data into billing nodes and verify the number of input entities of the merging event nodes associated with the same device package identifier in the dynamic circulation topology graph. The unpacking event node outputs a consistent quantity of entities. When this consistency is satisfied, a graph matching algorithm is executed to establish mapping edges between end event nodes and billing nodes in the dynamic flow topology graph, generating a closed-loop aligned graph data structure. The risk feature extraction module performs a reverse graph traversal algorithm starting from the billing node in the closed-loop aligned graph data structure. It calculates the quantity deviation between the expected number of processed devices and the number of traceable single device entity nodes corresponding to the target processing event node, the attribute deviation between the entity node and the billing node associated through the mapping edge, and counts the number of isolated billing nodes lacking a preceding physical flow path, to splice together and generate a graph structure anomaly feature vector. The data anomaly warning module calls a pre-set risk identification model to classify the graph structure anomaly feature vector to obtain a risk assessment status. When the risk assessment status is abnormal, it extracts the corresponding local topology image segment from the closed-loop aligned graph data structure to generate and output a flow data anomaly warning result containing the local topology image segment.

[0009] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following:

[0010] Compared to existing technologies that primarily rely on barcode traceability of sterilization packages, single-business data analysis, or manual spot checks for risk supervision, this invention addresses the problem of inaccurate correlation between physical traceability data and departmental billing data caused by frequent unpacking and repacking of reusable medical devices during the processes of recycling, cleaning, packaging, sterilization, distribution, and use in disinfection supply centers. By acquiring multi-source business data, including terminal departmental billing data and time-series physical flow records of medical devices, this invention constructs a dynamic flow topology graph containing entity nodes, event nodes, and associated edges, achieving a structured representation of the complex flow process of medical devices. Furthermore, it establishes a mapping relationship between physical flow events and billing data through a graph matching algorithm, forming a closed-loop aligned graph data structure of business consumption and entity flow, thus solving the problems of broken multi-source data links and the inability to achieve full lifecycle traceability in existing technologies. Building upon this foundation, by reverse traversing the closed-loop alignment graph, the system comprehensively analyzes graph structure anomalies such as device quantity deviations, entity attribute deviations, and isolated billing nodes. Combined with a risk identification model for anomaly classification, this enables the system to accurately identify hidden graph structure anomalies at the data association level, including broken traceability chains, abnormal material consumption, and inconsistent billing attributes. Compared to existing detection methods based on fixed rules or single indicators, this invention fully utilizes the topological relationships of medical device circulation and the association characteristics of business data, improving the accuracy and penetration of risk identification. Simultaneously, by outputting early warning results containing abnormal local topological paths, it achieves rapid location and traceable analysis of risk anomalies, enhancing the intelligence, precision, and effectiveness of hospital health risk supervision. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A flowchart of a method for early warning of abnormal hospital medical supplies circulation data provided in this application embodiment;

[0013] Figure 2 This is a schematic diagram of the structure of the hospital medical supplies circulation data anomaly early warning system provided in this application embodiment. Detailed Implementation

[0014] Embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of the present disclosure are shown in the drawings, it should be understood that embodiments of the present disclosure may be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure.

[0015] It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure. In the description of the embodiments of this disclosure, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "this embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc., may refer to different or the same objects.

[0016] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0017] With the continuous improvement of hospital information technology and intelligent management, the Central Sterile Supply Department (CSSD) and infection control material management departments are gradually introducing barcode traceability, IoT sensing, intelligent equipment management, and business information systems. This enables data collection and recording of reusable medical devices throughout the entire process from collection, cleaning, disinfection, inspection, packaging, sterilization, storage, distribution to clinical use. Through these technologies, hospitals can generate multi-dimensional business data covering medical device processing procedures, circulation paths, departments using the devices, and consumption status, providing a data foundation for improving infection control management, ensuring medical safety, and conducting health management risk prevention and control.

[0018] However, in actual hospital operations, the management of reusable medical devices and infection control supplies is highly dynamic and complex. On the one hand, medical devices are typically processed and circulated in device packages. Frequent unpacking and repacking operations occur during cleaning, assembly, sterilization, disassembly, and clinical use, meaning that the relationship between the device, the device package, and the corresponding processing events is not fixed but rather dynamically related as the business process changes. On the other hand, existing hospital information systems typically record data separately for different business stages. The traceability system primarily records the physical processing and circulation of medical devices, while the clinical business system mainly records departmental usage and consumption. Differences exist between these systems in data structure, recording granularity, and business logic, making it difficult to establish a continuous and accurate correlation between the actual circulation and processing of medical devices and their clinical use and consumption.

[0019] Existing methods for monitoring health management risks typically rely on manual verification, periodic spot checks, or rule comparisons based on single business data. These methods are ill-suited to the analytical needs of complex relationships within the dynamic circulation of medical devices. For example, in the management of hospital infection control supplies, health management risks may arise, including anomalies in medical device circulation records, discrepancies between processed and actual usage quantities, missing material consumption paths, abnormal usage behaviors, and non-standard process execution. These anomalies are often hidden across multiple business processes and data nodes, making timely detection difficult using only single system data or manual experience. Furthermore, the lack of correlation analysis methods for the dynamic disassembly and repackaging process of medical devices hinders the achievement of closed-loop verification between the entire lifecycle circulation data of reusable medical devices and clinical usage and consumption data. This results in insufficient accuracy in identifying health management risks, incomplete traceability paths, and difficulties in anomaly localization.

[0020] Therefore, how to establish an effective data association analysis mechanism for the dynamic unpacking and repacking process of reusable medical devices in the disinfection supply center, accurately link physical circulation data with departmental billing and consumption data in a closed loop, and accurately identify and warn of potential health management anomalies based on multi-dimensional association characteristics has become a technical problem that urgently needs to be solved in the current intelligent supervision of hospital health risks.

[0021] It should be noted that although this embodiment uses the circulation of reusable medical devices in a sterilization supply center as a typical scenario for detailed explanation, the dynamic circulation topology and closed-loop alignment graph data structure constructed by this invention are also applicable to risk monitoring of hospital drugs and disposable medical consumables (SPD supplies). When processing drug and consumable data, the "device identifier" mentioned in the claims and embodiments of this invention can be equivalently mapped to the unique traceability code of the drug or the unique device identifier (UDI) of the consumable; the "device package identifier" can be equivalently mapped to the barcode or batch number of the large packaging of the drug or consumable; the "unpacking" and "packaging" operations can be equivalently mapped to physical circulation operations such as "breaking down large packaging", "allocation and distribution according to base quantity" or "assembly of materials" in the material circulation process; at the same time, the "target processing event nodes (such as cleaning, sterilization, and packaging)" in the reverse graph traversal can be equivalently mapped to key quality control event nodes such as "warehousing inspection", "front warehouse shelving" or "prescription review" in the drug and consumable supply chain. Its underlying node mapping, consistency check, and graph matching algorithm logic remain completely consistent.

[0022] Firstly, such as Figure 1 The diagram shown is a flowchart of a method for early warning of abnormal hospital medical supplies circulation data provided in an embodiment of this application; Figure 1Furthermore, it should be noted that the various steps in this invention are interconnected and work in synergy to form a complete closed-loop risk analysis process. By acquiring multi-source business data, a data foundation is provided for constructing a dynamic circulation topology graph. By modeling device entities, device packages, and operational events into a graph structure, a continuous representation of the dynamic disassembly and assembly process of medical devices is achieved, providing a unified carrier for subsequent data association. Based on quantity conservation verification and graph matching algorithms, physical circulation events are accurately mapped to departmental billing data, forming a closed-loop aligned graph data structure, realizing the association and integration of the material circulation process and the business consumption process. On this basis, through graph traversal and abnormal feature extraction, potential abnormal relationships can be mined from the complete business chain, and risk judgment and location can be achieved by combining a risk identification model. Each step supports the others through data transmission and logical association, solving the problems of the separation between physical traceability data and business billing data, the difficulty in tracking dynamic disassembly and assembly scenarios, and the difficulty in discovering hidden risks in existing technologies. This achieves closed-loop supervision of the entire life cycle circulation and business consumption of medical devices, improving the accuracy, real-time performance, and interpretability of abnormal early warnings for hospital medical material circulation data.

[0023] refer to Figure 1 The first step of the hospital healthcare supplies circulation data anomaly early warning method is to acquire multi-source business data within the target time period. This multi-source business data includes terminal department billing data and time-series physical circulation records of reusable medical devices. This step provides a complete data foundation for hospital healthcare management and risk supervision by aggregating data from multiple aspects such as medical device management, department usage, and billing management. This helps to improve the transparency of medical resource utilization and reduce regulatory blind spots caused by information silos.

[0024] Specifically, this step achieves the acquisition and preprocessing of multi-source heterogeneous data through the following sub-steps:

[0025] S11: Determine the target time period and trigger the data extraction task. The system is equipped with a scheduled task manager that automatically triggers the data extraction process according to a preset risk warning cycle (e.g., 00:00 every day or 02:00 every Monday). The system determines the target time period based on the current time. This period defines the time window for a risk warning analysis.

[0026] S12: Obtain terminal department billing data. Extract data from the inpatient and outpatient billing details tables of the Hospital Information System (HIS) using standard interfaces (such as Web Service interfaces) or ETL tools provided by the Hospital Data Integration Platform (HIP). The terminal department billing data is parsed into a structured collection of billing records. Each billing record includes at least: a unique billing identifier, the consuming department identifier, the patient identifier, the billing item code, the billing quantity, the billing unit price, and the billing service time.

[0027] S13: Obtain the time-series physical flow records of reusable medical devices. Extract all material flow logs for the target time period from the Central Sterile Supply Department (CSSD) log database. These records are generated by scanning with IoT sensing devices (including PDA handheld terminals, fixed barcode scanners, and RFID readers) deployed at each stage of collection, cleaning, packing, sterilization, distribution, and use. The time-series physical flow records contain topological information characterizing the dynamic combination relationships of materials. Specific fields include: device package / single item identifier, operation type identifier (clearly marked as "collection," "packing and repacking," "sterilization," or "departmental use unpacking"), operation timestamp, and operator identifier. For the "packing and repacking" operation, the record contains the aggregation relationship between the parent package ID and the child package ID; for the "departmental use unpacking" operation, the record contains the branching relationship between the parent package ID and the split device ID.

[0028] In applications targeting drug and consumable monitoring, this step simultaneously extracts the time-series physical flow records of drugs and consumables within the target time period from the hospital supplies management system (SPD) or hospital resource planning system (HRP). The topology information of these records also includes: consumable UDI or drug traceability code (mapped to a single item identifier), packaging barcode (mapped to an instrument pack identifier), and operation type identifier (such as warehousing, distribution, and dismantling events), serving as a unified input source for constructing a dynamic flow topology map.

[0029] S14: Perform data cleaning and time-series alignment. The acquired heterogeneous data is loaded into the cache queue of the in-memory database for preprocessing to ensure data quality. Specific processing rules include: for duplicate scan records caused by signal jitter from IoT devices, a rule is set to deduplicate records of the same device identifier within a 2-second time window as redundant; the time difference between billing service time and physical flow time is verified, and clock deviations between different system servers are corrected to ensure logical consistency between physical flow time and billing time; records with missing key fields are marked as "abnormal data to be verified" and stored separately.

[0030] This step integrates medical device physical circulation data with departmental operational usage data to achieve comprehensive understanding of the handling, circulation, and use of medical devices in hospital health management. This avoids information gaps caused by a single source of operational data and improves the data integrity and regulatory capabilities of the medical device management process.

[0031] The second step of the hospital medical supplies circulation data anomaly early warning method is as follows: extract the device identifier, device package identifier, and operation type identifier from the time-series physical circulation record; map the device identifier and device package identifier to entity nodes; map the operation type identifier to the corresponding operation type event node; and establish aggregated directed edges and branch directed edges between entity nodes and corresponding event nodes based on the packing and unpacking operations in the time-series physical circulation record to construct a dynamic circulation topology graph in memory. This step achieves visualization and traceability management of the medical supplies circulation process by dynamically tracking the cleaning, packing, sterilization, distribution, and use of medical devices. This is conducive to standardizing the medical device usage process and preventing health management risks caused by incomplete circulation records and inadequate management.

[0032] Specifically, this step uses an attribute graph model to construct the topology of material flow in an in-memory database. The specific implementation process includes the following sub-steps:

[0033] S21: Node Mapping and Instantiation Based on Consistent Hashing Algorithm. Traverse the time-series physical flow records obtained in Step 1, performing entity and event extraction and mapping operations for each record: Entity Node Construction: Using the consistent hashing algorithm, calculate the hash value with "device identifier" or "device package identifier" as input to generate a unique node ID in the graph database. If the ID does not exist in the in-memory graph, instantiate a new entity node. The node attributes include entity_type (enumeration value: single entity / package) and base_info (JSON format storage name, specifications); if it exists, read it directly to ensure the uniqueness of the physical entity mapping in the graph. Event Node Construction: Create event nodes. The node attributes include timestamp (Unix timestamp format), operator_id, and event_type.

[0034] S22: Establishment and Attribute Definition of Aggregated Directed Edges Based on Bundling Operations. When an event node's event_type is identified as a bundling event, a graph construction algorithm is executed based on the "input list" and "output package ID" in the flow record: Incoming Edge Construction: Traverse all individual device entity nodes in the input list and establish a directed edge pointing to the bundling event node. The edge type is defined as AGGREGATE (aggregation). Edge Attribute Definition: Store the quantity and timestamp in the edge's attribute table to record the quantity characteristics during aggregation. Outgoing Edge Construction: Establish a directed edge from the bundling event node to the newly generated device package entity node. The type is defined as GENERATE (generation), thus completing the "multiple-in, one-out" topology connection.

[0035] S23: Branch-oriented edge construction and attribute definition based on unpacking operation. When the event_type of an event node is identified as an unpacking event, the graph construction algorithm is executed according to the "input package ID" and "output list" in the flow record: Incoming edge construction: Establish a directed edge from the split instrument package entity node to the unpacking event node. Outgoing edge construction: Traverse all individual instrument entity nodes in the output list and establish a directed edge from the unpacking event node to each individual instrument. The edge type is defined as BRANCH. Edge attribute definition: Store the quantity (default is 1) and the splitting timestamp in the attribute table of the branch-oriented edge to complete the "one-in, many-out" topology connection.

[0036] S24: Temporal Topology Graph Generation and Loop Detection Based on Kahn's Algorithm. Based on the timestamp attribute of each event node, temporally consecutive event nodes are connected by temporal edges. To ensure the rigor of the flow logic, Kahn's algorithm is introduced for loop detection when generating the final topology graph: the in-degree of all nodes is calculated, and nodes with an in-degree of 0 (the starting node) are added to the queue; nodes are removed sequentially, and the in-degree of their neighboring nodes is reduced. If there are any unvisited nodes, a loop anomaly is determined in the graph (i.e., a logically impossible circular flow exists), and an error reporting mechanism is triggered; finally, a strict directed acyclic graph (DAG) structure is generated as the dynamic flow topology graph. This graph is stored in memory in the form of an adjacency list, supporting subsequent efficient graph traversal.

[0037] By establishing aggregated directed edges and branched directed edges based on packaging and unpacking operations, and constructing a dynamic flow topology graph with temporal dependencies using timestamps, the dynamic transformation process of reusable medical devices from individual devices to device packages and then back to individual devices is accurately represented. This method can improve the precision of medical device flow management in hospital sterilization supply centers, enabling complete traceability of the device flow process and reducing hygiene management loopholes caused by incomplete flow records and unclear entity relationships.

[0038] The third step of the hospital medical supplies circulation data anomaly early warning method is as follows: Parse the terminal department billing data into billing nodes, verify the conservation consistency between the number of input entities of the merging event node associated with the same device package identifier and the number of output entities of the corresponding unpacking event node in the dynamic circulation topology graph, and, when the conservation consistency is satisfied, execute a graph matching algorithm to establish mapping edges between the end event nodes and billing nodes in the dynamic circulation topology graph to generate a closed-loop aligned graph data structure. This step, by establishing the correspondence between the actual circulation of medical devices, clinical use, and billing records, enables full-process supervision of hospital medical supplies consumption, and can promptly detect inconsistencies between medical device use and billing, abnormal consumption, and illegal billing, thereby improving the accuracy and effectiveness of hospital management.

[0039] As a further step, the specific implementation process of this step includes the following sub-steps:

[0040] S31: Billing Node Parsing and Attribute Index Construction. Traverse the terminal department billing data obtained in Step 1 and parse it into billing nodes in the graph database. Configure attribute fields for each billing node, including: department identifier, billing item code, billing quantity, billing timestamp, and patient identifier. Simultaneously, construct a hash index in memory with (department identifier, billing timestamp) as the key to accelerate subsequent graph matching queries.

[0041] S32: Conservation consistency check based on set-symmetric difference operation. Traverse the dynamic flow topology graph and perform strict entity conservation checks on the flow cycles associated with the same instrument package identifier:

[0042] Retrieve merge event nodes belonging to the same medical kit circulation cycle in the topology graph. With unpacking event node And must meet timestamp Strictly less than timestamp .

[0043] Extracting the target Input single device entity node ID set and by The output set of individual device entity node IDs Call the set symmetric difference algorithm to calculate... and The degree of difference.

[0044] Set the judgment threshold .like If the elements of the two sets completely overlap, it is determined that the data conservation consistency is satisfied; otherwise, it is determined that the data is abnormal. The system generates a "data chain break" mark for this flow branch in the topology graph (the attribute field is_valid is set to false) and triggers the interception logic to skip the execution of the subsequent graph matching algorithm for this node branch.

[0045] S33: Establishing mapping edges based on a temporal sliding window and a greedy matching algorithm. For flow branches that pass the consistency check, execute the cross-domain graph matching algorithm:

[0046] Setting up a time-series sliding window model ,in This is the timestamp for the issuance of the end event node. Preferably, a lower threshold window is set. Hour, upper limit threshold Hour.

[0047] Extract the receiving department identifier from the end event node (referring to the "Department Disbursement" type node). In the billing node index, filter nodes that meet the following criteria: the receiving department identifier equals the invoice department identifier, and the billing timestamp... The nodes constitute the candidate node set. .

[0048] Traverse the candidate node set Construct the objective function A greedy matching strategy is adopted to select the option that best matches the objective function. The node with the smallest value (i.e., the smallest time span) As the only best match.

[0049] At the end event node and Establish a cross-domain mapping edge of type BILLING_MAPPING between them, with the edge's weight attribute recording the time span value. This generates a complete closed-loop aligned graph data structure.

[0050] It should be noted that this step, by performing a conservation consistency check on the number of input entities in a combined package and the number of output entities in an unpacked package within the same medical device package circulation cycle, can detect data breaks, missing entities, or abnormal changes during the medical device circulation process in advance, and prevent abnormal data from entering subsequent risk analysis processes. This method improves the authenticity and reliability of hospital medical supplies traceability data, avoids erroneous regulatory judgments due to abnormal basic data, and helps to strengthen the standardized management of medical device disinfection, use, and circulation, thereby reducing health management risks.

[0051] Furthermore, by combining the receiving department identifier, the billing department identifier, and the time window to establish a precise mapping relationship between the end event node and the billing node, an effective link is achieved between the actual distribution and use of medical devices and the department's business billing behavior. This method can improve the hospital's ability to monitor the consistency between medical resource consumption and billing behavior, promptly detect discrepancies between medical device usage records and billing records, and prevent risky behaviors such as abnormal billing, illegal use, or abnormal resource consumption.

[0052] The fourth step of the hospital's abnormal data flow early warning method for medical supplies is as follows: In the closed-loop aligned graph data structure, a reverse graph traversal algorithm is executed starting from the billing node. This algorithm calculates the quantity deviation between the expected number of medical devices to be processed corresponding to the target processing event node and the number of traceable individual medical device entity nodes, the attribute deviation between the entity node and the billing node associated through the mapping edge, and the number of isolated billing nodes lacking a preceding physical flow path. These steps are then used to generate an abnormal feature vector of the graph structure. This step, through comprehensive analysis of multi-dimensional correlation information such as the number, attributes, and flow paths of medical devices, transforms the traditional manual sampling inspection into intelligent risk analysis. It can discover hidden abnormal material consumption, inflated workload, illegal use, and irregular charges, thereby improving the hospital's proactive discovery capabilities in supervision and management.

[0053] Specifically, this step utilizes graph computing techniques to extract multi-dimensional risk characteristics, and the implementation process includes the following sub-steps:

[0054] S41: Calculation of quantity deviation based on inverse depth-first search (DFS).

[0055] Starting from any billing node in the closed-loop aligned graph data structure, the algorithm recursively traverses upstream along the reverse direction of the flow edge, pointing backwards to the end event node along the cross-domain mapping edge.

[0056] During the traversal, event nodes with node type labels of "cleaning," "sterilization," or "packaging" are identified as target processing event nodes. The pre-defined `expected_quantity` (expected number of instruments to be processed) is read from the node's attribute field and recorded as... .

[0057] Starting from the target event node, the process continues traversing backwards along the aggregated directed edges pointing to that node until the physical flow initiation layer is reached (i.e., the individual device entity node with an in-degree of zero). During the traversal, a weighted summation algorithm is used to accumulate and count the number of entity nodes visited, thus obtaining the number of traceable individual device entity nodes. .

[0058] Calculate quantity deviation .like If the quantity is abnormal, it is determined that there is an anomaly and it is taken as the first component of the feature vector; otherwise, the quantity is determined to be normal and the first component of the feature vector is assigned a value of 0.

[0059] S42: Attribute deviation calculation based on value classification tree model.

[0060] The system has a pre-installed medical supplies value dictionary, covering reusable medical devices, medicines, and disposable consumables. This dictionary is stored in a multi-branch tree structure, with each node corresponding to a value level code (e.g., Level 1 is a basic consumable package or basic essential medicine; Level 2 is a general surgical package or routine out-of-pocket medicine; Level 3 is a high-value precision medical device package, high-value implantable consumables, or special targeted drugs). By comparing the standard supply value level with the actual chargeable item level, it can not only reflect deviations in the attributes of medical device assets but also accurately detect irregularities such as falsely charging low-priced medicines / consumables as high-priced items.

[0061] Extract the original entity node IDs along the same traceability path, and look up their standard material value hierarchy codes in the dictionary using hash mapping. Extract the charge item code of the end billing node and query its actual charge item level code. .

[0062] Calculate the distance between two hierarchical codes along the path in the tree structure, i.e., the attribute deviation. Preferably, if If the condition is met, it is determined that there is a risk of material swapping, and this is used as the second component of the feature vector; otherwise, the attributes are determined to match, and the second component of the feature vector is assigned a value of 0.

[0063] S43: Statistics of isolated billing nodes based on in-degree detection.

[0064] Traverse all billing nodes in the closed-loop alignment graph and examine their set of incoming edges. For any target billing node, count the number of incoming edges whose source node type is "endpoint event node". .

[0065] Set the judgment threshold .like If a billing node lacks a preceding physical flow path, an ISOLATED tag is added to it. After traversal, the total number of nodes carrying this tag is counted. And use it as the third component of the feature vector.

[0066] S44: Generation of concatenated feature vectors of graph structure anomalies.

[0067] The values ​​of the three components are concatenated to generate a three-dimensional feature vector. Preferably, the Min-Max normalization algorithm is used to process each component in the vector, mapping its numerical range to the [0,1] interval, so as to serve as the standardized input data for subsequent models.

[0068] It should be understood that by comprehensively analyzing the medical device circulation paths, entity attribute relationships, and billing associations in the closed-loop alignment graph data structure, abnormal features are extracted from dimensions such as consistency of processing quantity, matching of material attributes, and integrity of billing paths. This enables intelligent verification of the consistency between actual use of medical devices, processing workload, material value attributes, and billing behavior. This method can effectively detect health management risks such as inflated processing workload, interchangeable material attributes, and abnormal billing without physical circulation evidence. It improves the transparency and standardization of hospital medical device use, disinfection, and billing management processes, providing more accurate and reliable data support for hospital risk prevention and control.

[0069] The fifth step of the hospital healthcare supplies circulation data anomaly early warning method is as follows: A pre-set risk identification model is invoked to classify the graph structure anomaly feature vectors to obtain a risk assessment status. When the risk assessment status is abnormal, the corresponding local topological image segment is extracted from the closed-loop aligned graph data structure to generate and output a circulation data anomaly early warning result containing the local topological image segment. This step, by automatically identifying risks and providing anomaly correlation paths, enables rapid location of the risk occurrence link, involved objects, and causes. It provides traceable regulatory evidence for hospital discipline inspection, auditing, medical affairs, and health management departments, improving the timeliness, accuracy, and intelligence level of risk prevention and control.

[0070] Specifically, this step uses an unsupervised machine learning algorithm to determine anomalies in feature vectors and combines graph visualization technology to generate early warning evidence. The specific implementation process includes the following sub-steps:

[0071] S51: Risk assessment state classification based on the isolated forest algorithm.

[0072] The pre-built risk identification model based on the Isolation Forest algorithm is invoked. The specific hyperparameters configured during the model's construction phase are: number of decision trees. Subsampling size .

[0073] The graph structure anomaly feature vector generated in step four Input model. The model constructs a binary search tree by randomly selecting features and split values, and calculates the average path length of the feature vectors in the tree. And through the normalization formula Calculate the anomaly isolation score ,in The score is adjusted by the average path length correction factor for the binary search tree. The value range is [0,1].

[0074] Setting dynamic risk thresholds Preferably, a sliding window mechanism is used to calculate the dynamic risk threshold. During the system cold start phase, initial presets... The value is set to 0.7. Subsequently, the system periodically (e.g., weekly) calculates the average anomaly isolation score for all samples from the previous period. and standard deviation ,in accordance with The threshold is dynamically updated based on a criterion (e.g., k=3) to adapt to distribution changes caused by fluctuations in the workload of different hospitals. If the feature vector is found to be abnormal, the risk assessment status is set to "high risk"; otherwise, it is set to "normal".

[0075] S52: Dynamic extraction and cropping of local topological image segments.

[0076] Identify key feature components that trigger abnormal states (such as quantity deviation) Or attribute deviation The billing node associated with this component is locked as the root node. .

[0077] by Starting from the first element, a breadth-first search (BFS) algorithm is used to traverse the closed-loop alignment graph along the reverse direction of the mapping edges and the flow edges, extracting a complete subgraph containing all upstream entity nodes and event nodes. .

[0078] Traversing subgraphs For all nodes, set the clipping criterion as follows: if the number of nodes deviates from the attribute... And attribute deviation If the incoming edge type is normal, it is considered a normal node and deleted. Abnormal backbone links with quantity gaps, attribute deviations, or missing paths are retained, generating a simplified abnormal topology fragment. .

[0079] S53: Visualization rendering and early warning result output.

[0080] abnormal topology fragments Input to a graph rendering engine (such as D3.js or ECharts). Set specific rendering rules: render abnormal event nodes in red (RGB value: #FF0000), render entity nodes with attribute deviations in orange (RGB value: #FFA500), render normal nodes in gray, and label the specific deviation value next to the node with a text label.

[0081] The rendered graph data segments are compared with the risk assessment status and anomaly isolation score. The data is combined and packaged to generate a structured JSON data packet or PDF warning report, which is then pushed to the monitoring terminal of the discipline inspection and supervision department via the hospital's local area network.

[0082] It should be added that by using an intelligent risk identification model based on graph structure anomaly feature vectors for anomaly classification, and combining it with local topological image segment extraction and visualization, the system can automatically identify, accurately locate, and traceably analyze potential risks in the hospital's medical device circulation and billing processes. This approach overcomes the limitations of traditional manual sampling and fixed-rule supervision, improving the timeliness, accuracy, and intelligence of hospital healthcare supply circulation data anomaly warnings. It enables supervisors to quickly grasp the entities, business processes, and related paths involved in anomalies, providing intuitive and effective risk evidence for hospital disciplinary supervision, internal auditing, and healthcare management decisions.

[0083] Secondly, such as Figure 2 The diagram shown is a structural schematic of the hospital medical supplies circulation data anomaly early warning system provided in this application embodiment. (Refer to...) Figure 2 The system includes: a medical data acquisition module, a dynamic topology construction module, a closed-loop alignment generation module, a risk feature extraction module, and a data anomaly early warning module.

[0084] Specifically, the medical data acquisition module acquires multi-source business data within the target time period, including terminal department billing data and time-series physical flow records of reusable medical devices. The dynamic topology construction module extracts device identifiers, device package identifiers, and operation type identifiers from the time-series physical flow records. It maps device identifiers and device package identifiers to entity nodes, and operation type identifiers to event nodes corresponding to the operation types. Based on the merging and unmerging operations in the time-series physical flow records, it establishes aggregated directed edges and branch directed edges between entity nodes and corresponding event nodes to construct a dynamic flow topology graph in memory. The closed-loop alignment generation module parses terminal department billing data into billing nodes, verifies the conservation consistency between the number of input entities of merging event nodes associated with the same device package identifier and the number of output entities of the corresponding unmerging event nodes in the dynamic flow topology graph, and executes a graph matching algorithm to establish mapping edges between end event nodes and billing nodes in the dynamic flow topology graph when the conservation consistency is satisfied, thereby generating a closed-loop aligned graph data structure. The risk feature extraction module performs a reverse graph traversal algorithm starting from the billing node in the closed-loop aligned graph data structure. It calculates the quantity deviation between the expected number of processed devices and the number of traceable individual device entity nodes corresponding to the target processing event node, the attribute deviation between the entity node and the billing node associated through mapping edges, and counts the number of isolated billing nodes lacking preceding physical flow paths. These calculations are then used to generate an abnormal feature vector for the graph structure. The data anomaly warning module calls a pre-built risk identification model to classify the abnormal feature vector for the graph structure to obtain a risk assessment status. When the risk assessment status is abnormal, it extracts the corresponding local topology image segment from the closed-loop aligned graph data structure to generate and output a flow data anomaly warning result containing the local topology image segment.

[0085] Overall, this invention performs structured parsing of device identifiers, device package identifiers, and operation type identifiers in the time-series physical flow records, converting medical device entities, device packages, and processing operations into corresponding nodes. Based on unpacking and repacking operations, it establishes dynamic relationships between nodes, constructing a dynamic flow topology graph. By expressing the medical device processing and flow process through a graph structure, it effectively preserves the relationships between device entities and device packages that change with business processes. This solves the problem that traditional linear traceability methods struggle to describe changes in entity relationships under dynamic unpacking and repacking scenarios, improving the traceability and relationship expression capabilities of the medical device flow process.

[0086] Furthermore, this invention converts terminal department billing data into billing nodes and, combined with quantity conservation checks during the device packaging process and graph matching algorithms, establishes a mapping relationship between physical flow end-point event nodes and billing nodes, generating a closed-loop aligned graph data structure. This achieves data connectivity between the medical device processing, circulation, and clinical use processes. Through this closed-loop association method, inconsistencies between the actual physical flow path and business consumption records can be effectively identified, providing accurate data for discovering abnormal material usage, abnormal process execution, and hygiene management risks.

[0087] Building upon this foundation, this invention performs reverse graph traversal based on a closed-loop aligned graph data structure, tracing the corresponding medical device circulation path from the business consumption results. It then generates anomaly feature vectors for the graph structure by combining multi-dimensional correlation features such as quantity deviation, attribute deviation, and the number of isolated billing nodes. Compared to existing risk detection methods based on single indicators or fixed rules, this invention can comprehensively analyze the entire medical device process, fully uncovering abnormal patterns hidden between multiple stages and nodes, thus improving the accuracy and depth of identification and analysis of complex healthcare management risks.

[0088] Meanwhile, this invention classifies the abnormal feature vectors of the graph structure by calling the risk identification model, and extracts the corresponding local topological image segment when an abnormal risk is detected, generating an early warning result containing the risk association path. This enables the risk identification result to not only reflect the abnormal state, but also provide the associated nodes and flow paths of the abnormality, enhancing the interpretability and verifiability of the risk warning, and facilitating managers to quickly locate the source of the risk and take targeted regulatory measures.

[0089] In summary, this invention, through multi-source data fusion, dynamic topology modeling, closed-loop correlation analysis, abnormal feature extraction, and intelligent risk identification, forms a full-process risk early warning mechanism for hospital health management scenarios. It effectively solves the problems of data association difficulties, difficulty in corresponding business consumption and physical circulation, and difficulty in timely detection of abnormal behavior during the dynamic disassembly and repackaging of reusable medical devices. It realizes intelligent supervision of the medical device circulation process and business usage process, and improves the accuracy, real-time performance, traceability, and supervision efficiency of abnormal early warning of hospital health supplies circulation data.

[0090] Through the above description of the implementation methods, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the above functions can be divided into different functional modules to complete all or part of the functions described above.

[0091] In the embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or other forms.

[0092] The units described as separate components may or may not be physically separate. A component shown as a unit can be one or more physical units, located in one place or distributed in multiple different locations. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0093] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0094] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to the solution, or all or part of the technical solution, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0095] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for early warning of abnormal data in the circulation of hospital medical supplies, characterized in that, Includes the following steps: Acquire multi-source business data within the target time period, including terminal department billing data and time-series physical flow records of reusable medical devices; Extract the instrument identifier, instrument package identifier, and operation type identifier from the time-series physical flow record. Map the instrument identifier and instrument package identifier to entity nodes, and map the operation type identifier to the corresponding operation type event node. Based on the package merging and unpacking operations in the time-series physical flow record, establish aggregated directed edges and branch directed edges between entity nodes and corresponding event nodes to construct a dynamic flow topology graph in memory. The terminal department billing data is parsed into billing nodes. The conservation consistency of the number of input entities of the combined event node associated with the same instrument package identifier and the number of output entities of the corresponding unpacking event node in the dynamic circulation topology graph is verified. When the conservation consistency is satisfied, the graph matching algorithm is executed to establish the mapping edge between the end event node and the billing node in the dynamic circulation topology graph to generate a closed-loop aligned graph data structure. In the closed-loop aligned graph data structure, the reverse graph traversal algorithm is executed starting from the billing node. The number deviation between the expected number of devices to be processed and the number of traceable single device entity nodes corresponding to the target processing event node is calculated, the attribute deviation between the entity node and the billing node associated through the mapping edge is calculated, and the number of isolated billing nodes lacking the preceding physical flow path is counted, so as to splice and generate the graph structure abnormal feature vector. The pre-built risk identification model is invoked to classify the abnormal feature vectors of the graph structure to obtain the risk assessment status. When the risk assessment status is abnormal, the corresponding local topology image segment is extracted from the closed-loop aligned graph data structure to generate and output the flow data abnormal warning result containing the local topology image segment.

2. The method for early warning of abnormal hospital medical supplies circulation data as described in claim 1, characterized in that, The process of establishing aggregated directed edges and branched directed edges between entity nodes and corresponding event nodes based on the merging and unmerging operations in the temporal physical flow record, in order to construct a dynamic flow topology graph in memory, includes: When an operation type identifier is identified as a combined event node, the entity nodes corresponding to each individual device involved in the packaging are taken as the starting node, and the combined event node is taken as the target node, and an aggregated directed edge is established pointing to the combined event node. When the operation type is identified as a disassembly event node, the disassembly event node is used as the starting node, and the entity nodes corresponding to each individual device after disassembly are used as the target nodes. A directed edge branching out from the disassembly event node is established. The event nodes are associated and connected according to the order of their timestamps to generate a directed acyclic graph structure with temporal dependencies, which serves as a dynamic flow topology graph.

3. The method for early warning of abnormal hospital medical supplies circulation data as described in claim 1, characterized in that, The conservation consistency of the number of input entities for the merging event node associated with the same instrument package identifier and the number of output entities for the corresponding unpacking event node in the dynamic flow topology graph includes: In the location dynamic flow topology diagram, the event node pairs that belong to the same medical device package flow cycle and whose package merging event node occurs earlier than the corresponding unpacking event node; Extract the set of input individual device identifiers pointing to the packaged event node, and extract the set of output individual device identifiers issued by the unpackaged event node; The input set of individual device identifiers is compared with the output set of individual device identifiers. If the content of the elements in the two sets is the same and the number of elements in the sets is equal, it is determined that the conservation consistency is satisfied; otherwise, a data break mark is generated for the target node branch, and the subsequent graph matching algorithm is intercepted for the node branch carrying the data break mark.

4. The method for early warning of abnormal hospital medical supplies circulation data as described in claim 1, characterized in that, The execution graph matching algorithm establishes mapping edges between end event nodes and billing nodes in the dynamic flow topology graph, including: Extract the receiving department identifier and disbursement timestamp carried in the end event node, and extract the billing department identifier and billing timestamp carried in the billing node; Within the preset time-series sliding window, select end event nodes whose receiving department identifier matches the billing department identifier and whose issuance timestamp is earlier than the billing timestamp as the candidate node set; Based on the principle of minimizing the time span, a unique end event node is locked from the candidate node set, and a cross-domain mapping edge is established between the locked end event node and the corresponding billing node.

5. The method for early warning of abnormal hospital medical supplies circulation data as described in claim 1, characterized in that, The discrepancy between the expected number of processing devices corresponding to the target processing event node and the number of traceable individual device entity nodes includes: Starting from the billing node, trace upstream along the closed-loop alignment graph data structure to locate the target processing event node belonging to the target processing category, and obtain the expected number of processing devices corresponding to the target processing event node. Reverse recursion along the valid flow edge pointing to the target processing event node to the physical flow starting layer, and count the number of traceable individual device entity nodes. The numerical deviation between the expected number of devices to be processed and the number of traceable individual device entity nodes is calculated and used as a data anomaly feature component reflecting the inconsistency between the expected and traceable quantities.

6. The method for early warning of abnormal hospital medical supplies circulation data as described in claim 1, characterized in that, The calculation of attribute deviation between entity nodes and billing nodes associated through mapping edges includes: Extract the original entity nodes located on the same traceability path and the end billing nodes associated through the mapping edges from the closed-loop alignment graph data structure; Call the pre-built medical supplies value dictionary to query the standard supply value level corresponding to the original entity node and the actual chargeable item level corresponding to the end billing node. Calculate the hierarchical span distance between the standard material value level and the actual chargeable item level in the value classification tree data structure, and use the hierarchical span distance as the attribute deviation degree that reflects the abnormal mismatch between the entity and the chargeable item attributes.

7. The method for early warning of abnormal hospital medical supplies circulation data as described in claim 1, characterized in that, The statistics on the number of isolated billing nodes lacking a preceding physical transfer path include: In the closed-loop aligned graph data structure, traverse each billing node to determine the number of incoming edges for each billing node and the source node type of each incoming edge. Calculate the number of incoming edges from the end event node for each billing node; when the number of incoming edges from the end event node to the target billing node is zero, determine that the target billing node lacks a preceding physical flow path and add an isolated node label to the target billing node. The total number of billing nodes carrying the isolated node label is counted to obtain the number of isolated billing nodes, and the number of isolated billing nodes is used as an abnormal feature component reflecting the lack of correlation between billing records and physical flow records.

8. The method for early warning of abnormal hospital medical supplies circulation data as described in claim 1, characterized in that, The step of calling a pre-set risk identification model to classify the graph structure anomaly feature vectors to obtain a risk assessment status includes: Input the graph structure anomaly feature vectors into the risk identification model based on the isolated forest algorithm; The anomaly isolation score of graph structure anomaly feature vectors in a multidimensional feature space is calculated using a risk identification model. The system compares the anomaly isolation score with a preset dynamic risk threshold, and outputs a risk assessment status indicating an anomaly when the anomaly isolation score is greater than the dynamic risk threshold.

9. The method for early warning of abnormal hospital medical supplies circulation data as described in claim 1, characterized in that, The step of extracting corresponding local topology image segments from the closed-loop alignment graph data structure to generate and output flow data anomaly warning results containing local topology image segments includes: Using the billing node associated with the key feature component that triggers the abnormal state as the root node, extract the complete tracing path graph of the billing node in the closed-loop alignment graph data structure; Trim normal branches that do not contain abnormal feature data from the complete traceability path diagram, and retain abnormal main links that have quantity gaps, attribute deviations, or missing paths; The abnormal backbone link is rendered as a visual graph data image segment, and the graph data image segment is combined and packaged with the risk assessment status to generate and output a structured flow data anomaly warning result.

10. A hospital medical supplies circulation data anomaly early warning system, employing the hospital medical supplies circulation data anomaly early warning method as described in any one of claims 1-9, characterized in that, include: The module includes a medical data acquisition module, a dynamic topology construction module, a closed-loop alignment generation module, a risk feature extraction module, and a data anomaly early warning module. The medical data acquisition module is used to acquire multi-source business data within a target time period. The multi-source business data includes terminal department billing data and time-series physical flow records of reusable medical devices. The dynamic topology construction module is used to extract the instrument identifier, instrument package identifier, and operation type identifier from the time-series physical flow record, map the instrument identifier and instrument package identifier to entity nodes, map the operation type identifier to the corresponding operation type event node, and establish aggregated directed edges and branch directed edges between entity nodes and corresponding event nodes based on the package merging and unpacking operations in the time-series physical flow record, so as to construct a dynamic flow topology graph in memory. The closed-loop alignment generation module is used to parse the terminal department billing data into billing nodes, verify the conservation consistency between the number of input entities of the combined event node associated with the same instrument package identifier and the number of output entities of the corresponding unpacking event node in the dynamic flow topology graph, and execute the graph matching algorithm to establish the mapping edge between the end event node and the billing node in the dynamic flow topology graph when the conservation consistency is satisfied, so as to generate the closed-loop alignment graph data structure. The risk feature extraction module is used to execute a reverse graph traversal algorithm starting from the billing node in the closed-loop aligned graph data structure. It calculates the quantity deviation between the expected number of devices to be processed and the number of traceable single device entity nodes corresponding to the target processing event node, calculates the attribute deviation between the entity node and the billing node associated through the mapping edge, and counts the number of isolated billing nodes that lack a preceding physical flow path, so as to splice together the graph structure abnormal feature vector. The data anomaly early warning module is used to call a preset risk identification model to classify the graph structure anomaly feature vector to obtain the risk assessment status. When the risk assessment status is abnormal, it extracts the corresponding local topology image segment from the closed-loop aligned graph data structure to generate and output the flow data anomaly early warning result containing the local topology image segment.

Images