A hospital operation process optimization and anomaly detection system and method

Abstract

The application is a hospital operation process optimization and abnormality detection system and method, belonging to the technical field of hospital informationization and operation management. The technical problem to be solved is that the existing hospital operation management has the problems of serious data island, static process modeling, lagging abnormality discovery, lack of data support for optimization and insufficient real-time monitoring capability, resulting in invisible process execution, untimely problem handling and difficult quantification of optimization effect. The technical solution points are: non-intrusive collection of full operation logs of each business system of the hospital to build a unified process event model; multi-level primary key association is adopted to form a complete process event sequence, and an improved Alpha++ algorithm is used to automatically discover the actual business process model; multi-dimensional identification of process bottlenecks is adopted, and an algorithm combining cosine similarity and edit distance is used to detect abnormalities in real time and give graded early warnings; case-based reasoning and discrete event simulation are used to generate implementable optimization suggestions, forming a closed-loop operation management system.

Description

Technical Field

[0001] This invention belongs to the field of hospital information and operation management technology, specifically relating to a hospital operation process optimization and anomaly detection system and method. Background Technology

[0002] With the continuous advancement of hospital informatization, business systems such as Hospital Information System (HIS), Laboratory Information System (LIS), Picture Archiving and Communication System (PAC), Electronic Medical Record (EMR), and Hospital Resource Planning (DRP) have been widely applied in daily hospital operations. However, existing hospital operation and management models still have significant technical shortcomings: (1) The hospital’s various business systems operate independently, forming serious data silos. They cannot build a complete end-to-end process event chain, resulting in the process execution process being in a “black box” state, and managers cannot intuitively understand the actual flow of business.

[0003] (2) Traditional process management relies on static standard processes that are manually sorted out. The deviation rate between these processes and actual business execution is as high as 30%-50%. This makes it impossible to reflect the complex and ever-changing business patterns of hospitals and also difficult to adapt to the personalized process needs of different departments and different diseases.

[0004] (3) The discovery of anomalies is seriously delayed. Traditional manual inspections can only discover obvious and post-event process problems, but cannot identify hidden process bottlenecks, frequent rework paths and key control gaps, which leads to the accumulation and spread of problems, seriously affecting patients' medical experience and hospital operation efficiency.

[0005] (4) The process optimization lacks data support and mainly relies on the experience judgment of managers. The optimization plan is not targeted and effective enough, and the optimization effect cannot be quantitatively evaluated, making it difficult to form a continuous closed-loop improvement mechanism.

[0006] (5) The existing system lacks real-time process monitoring capabilities and cannot provide timely warnings and interventions for abnormal behaviors during process execution, which leads to increased medical error risks and high operating costs.

[0007] In view of this, the present invention is hereby proposed. Summary of the Invention

[0008] To address the aforementioned technical problems in existing technologies, this invention provides a hospital operation process optimization and anomaly detection system and method. By non-intrusively collecting full operation logs from various hospital business systems, a unified process event model is constructed. An improved Alpha++ process mining algorithm is used to automatically discover actual business processes, achieving digital mapping of the entire business process, multi-dimensional bottleneck identification, real-time anomaly detection, and data-driven intelligent optimization. Ultimately, this forms a closed-loop operation management system of "data collection - model discovery - problem identification - optimization suggestions - effect evaluation".

[0009] To achieve the above objectives, the technical solution of the present invention is as follows: Firstly, a hospital operation process optimization and anomaly detection system includes: Multi-source data acquisition and standardization module: used for non-intrusive acquisition of full operation logs from various business systems in the hospital, data cleaning and standardization processing, and construction of a unified process event model; Process event sequence construction module: used to aggregate standardized events according to business cases to form a complete process event sequence, and to preprocess and store it; Process mining and model discovery module: used to automatically discover actual business process models from process event sequences, compare them with standard process models, and dynamically update the process models; Bottleneck identification and anomaly detection module: used to identify process bottlenecks from multiple dimensions, detect abnormal behaviors and control gaps that deviate from the standard path in real time, and provide graded early warnings; Process optimization and visualization application module: It is used to automatically generate actionable optimization suggestions based on historical optimization cases and process simulations, quantitatively evaluate the optimization effect, and provide visualization and report output.

[0010] Furthermore, the multi-source data acquisition and standardization module includes: Non-intrusive data acquisition unit: Used to collect operation logs, database change logs, and API call logs of HIS, LIS, PACS, EMR, HRP, and OA systems through log file parsing, database CDC, message queue subscription, and API interface calls, and defines standardized collection fields for different systems; Data cleaning and preprocessing unit: used to remove duplicates, complete data, convert formats, and filter noise from raw data, and to handle missing values, outliers, and duplicate events; Unified event model building unit: used to define standard process event formats and map heterogeneous data from different systems to unified process events through a field mapping table.

[0011] Furthermore, the mathematical representation of the unified process event is as follows:

[0012] in, As a unique identifier for the event, Used as a unique identifier for the case. The name of the event. As an executor, The start time of the event. This is the end time of the event. For the resource ID used, This is a set of business attributes.

[0013] Furthermore, the process event sequence construction module includes: Case association unit: used to aggregate scattered events according to business rules and data relationships, such as patient visits, drug procurement, equipment maintenance, etc., to form a complete process event sequence; Event sequence preprocessing unit: used for time sorting of event sequences, filtering of abnormal events and merging of events, and handling parallel activities, cyclic activities and nested processes; Hybrid storage unit: Uses InfluxDB, a time-series database, to store raw event logs, and Neo4j, a graph database, to store process models and event relationships, supporting efficient process querying and analysis.

[0014] Furthermore, the process mining and model discovery module includes: Actual process model discovery unit: Used to automatically discover actual business process models from event logs using improved Alpha++ algorithm and heuristic mining algorithm, and generate flowcharts in BPMN2.0 standard format; Multi-dimensional process analysis unit: used to integrate time, resource and role dimensions for in-depth analysis of activity execution time distribution, resource utilization, role collaboration relationships and process flow patterns; Process Model Comparison Unit: Used to compare the actual process model with the hospital's pre-set standard process model to identify process deviations and differences; Dynamic update unit for process model: used to periodically re-execute the process mining algorithm to dynamically update the process model.

[0015] Furthermore, the improved Alpha++ algorithm introduces event frequency weights to filter low-frequency abnormal events, and its event filtering formula is as follows:

[0016] in, For a valid set of events, Let e ​​be the frequency of occurrence of event e. This is the frequency threshold, with a value range of 0.01-0.05.

[0017] Furthermore, the bottleneck identification and anomaly detection module includes: Multi-dimensional bottleneck identification unit: used to identify process bottlenecks from the dimensions of time, resources and process structure, and to locate global bottlenecks using the critical path method and bottleneck theory; Offline anomaly detection unit: Used to discover abnormal process patterns from historical event logs using DBSCAN clustering algorithm and Isolation Forest algorithm; Real-time anomaly detection unit: Used with Flink stream processing technology and sliding window mechanism to monitor the process execution in real time and detect abnormal behavior that deviates from the standard path through sequence alignment algorithm; Control Deficiency Identification and Early Warning Unit: Based on the compliance rule engine, it detects missing approval steps, unauthorized operations, and behaviors that do not comply with hospital rules and regulations in the process, and provides graded early warnings according to the severity of the anomalies.

[0018] Furthermore, the real-time anomaly detection employs an attention-based Transformer encoder to encode the process event sequence, and the anomaly score is calculated using the following formula:

[0019] in, This represents the cosine similarity of the feature vectors after global average pooling. The edit distance similarity of the process sequence; The 512-dimensional feature vector is obtained by encoding the actual process sequence using a Transformer encoder and then performing global average pooling. This is a 512-dimensional feature vector obtained after the standard process sequence undergoes the same encoding and pooling operations. A sequence of activity names for the actual process sequence; This is the sequence of activity names corresponding to the standard process; when When a general warning is triggered, An emergency warning is triggered at that time.

[0020] Furthermore, the process optimization and visualization application module includes: Optimization suggestion generation unit: Based on multi-dimensional case reasoning technology, it matches the currently identified problem with historical optimization cases based on similarity and generates executable optimization suggestions by combining process simulation results; Process simulation and verification unit: used to build a digital twin model of a process based on discrete event simulation technology, and to simulate the execution effect of different optimization schemes; Visualization unit: Used to provide a web-based visualization interface to display the actual process model, process execution status, bottleneck locations, abnormal events, and optimization suggestions; Effectiveness Evaluation and Closed-Loop Management Unit: Used to track the implementation of optimization plans, compare process indicators before and after optimization, evaluate the optimization effect, and store successful optimization cases in the case library.

[0021] Secondly, a method for optimizing and detecting anomalies in hospital operations processes, applied to the hospital operations process optimization and anomaly detection system described in any of the preceding statements, includes the following steps: S1. Non-invasively collect all operation logs from various business systems in the hospital, perform data cleaning and standardization processing, and build a unified process event model; S2. Aggregate the standardized events according to business cases to form a complete process event sequence, and preprocess and store them; S3. Automatically discover actual business process models from process event sequences, compare them with standard process models, and dynamically update the process models; S4. Multi-dimensional identification of process bottlenecks, real-time detection of abnormal behaviors and control gaps that deviate from the standard path, and hierarchical early warning; S5 automatically generates actionable optimization suggestions based on historical optimization cases and process simulations, quantitatively evaluates optimization effects, and provides visual displays and report outputs.

[0022] Compared with existing technologies, the hospital operation process optimization and anomaly detection system and method provided by this invention breaks down hospital data silos through non-invasive multi-source data acquisition and constructs a complete end-to-end process event chain; it uses an improved Alpha++ process mining algorithm to automatically discover actual business processes, solving the problem of large deviations between static process models and actual business; it achieves multi-dimensional identification of process bottlenecks and real-time detection of abnormal behaviors, enabling early detection of hidden problems that are difficult to identify using traditional methods; and it combines multi-dimensional case reasoning and discrete event simulation to generate data-driven optimization suggestions, forming a closed-loop improvement mechanism, which significantly improves hospital operation efficiency and patient medical experience. Attached Figure Description

[0023] Figure 1 This is an architecture diagram of a hospital operation process optimization and anomaly detection system provided in an embodiment of the present invention. Detailed Implementation

[0024] The technical solution of the present invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are not all embodiments of the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0025] It should be noted that, unless otherwise specifically stated, the relative arrangement and numerical expressions of the components and steps described in these embodiments should not be construed as limiting the scope of the invention.

[0026] The following description of exemplary embodiments is merely illustrative and is not intended to limit the invention or its application or use in any way. Techniques, methods, and apparatus known to those skilled in the art may not be discussed in detail herein, but where applicable, such techniques, methods, and apparatus should be considered part of this specification.

[0027] Example 1 See Figure 1 , Figure 1 This is an architecture diagram of a hospital operation process optimization and anomaly detection system proposed in this invention. The system collects full operation logs from all hospital business systems, constructs a unified process event model, and uses an improved Alpha++ process mining algorithm to automatically discover actual business processes, achieving digital mapping, bottleneck identification, anomaly detection, and intelligent optimization of the entire business process. Specifically, it may include: M1, Multi-Source Data Acquisition and Standardization Module: Used for non-intrusive acquisition of full operation logs from various hospital business systems, performing data cleaning and standardization processing, and constructing a unified process event model; specifically including: M11, Non-Intrusive Data Acquisition Unit: Employing multiple methods including log file parsing, database CDC (Database Control Center), message queue subscription, and API interface calls, this unit collects operation logs, database change logs, and API call logs from systems such as Hospital Information System (HIS), Laboratory Information System (LIS), Picture Archiving and Communication System (PACS), Electronic Medical Record System (EMR), and Hospital Resource Planning System (HRP). The acquisition process does not disrupt the normal operation of existing business systems and supports both real-time and batch acquisition modes, with real-time acquisition latency less than 1 second.

[0028] Standardized data collection fields are defined for different systems: HIS system: Collects patient ID, consultation number, operation type, operator ID, operation time, clinic ID, and fee amount; LIS system: Collects test request number, patient ID, specimen ID, test item, collector ID, collection time, submission time, report time, and reviewer ID; PACS system: Collects examination request number, patient ID, examination equipment ID, examination time, report generation time, and reporting physician ID; EMR system: Collects medical record ID, patient ID, doctor ID, writing time, modification time, and medical record status; HRP system: Collects purchase order number, requesting department, applicant, approver, approval time, warehousing time, and outbound time; M12, Data Cleaning and Preprocessing Unit: Performs deduplication, completion, format conversion, and noise filtering on the raw data. It removes redundant information and garbled characters from logs using regular expressions, completes missing timestamps and executor information using linear interpolation, and filters duplicate event records and obvious outliers (such as events with negative execution times). M13, Unified Event Model Building Unit: Defines a standard process event format and maps heterogeneous data from different systems to unified process events through a field mapping table. For example, it maps the "registration time" of the HIS system to the "start time" of a unified event, maps "registration staff ID" to "executor", and maps "clinic number" to "resource ID". The mathematical representation of a unified process event is:

[0029] in, A unique identifier for the event, generated using a UUID; To ensure each case is uniquely identified, the outpatient process uses the patient ID plus the date of visit, while the inpatient process uses the inpatient number. The name of the activity, such as "registration", "triage", "medical treatment", etc. For the implementer, such as the doctor's name, nurse's name, etc.; The start time of the event. This is the end time of the event. For the resource ID used, such as clinic ID, examination equipment ID, etc.; This is a set of business attributes, such as patient age, gender, and disease type.

[0030] M2, Process Event Sequence Construction Module: This module aggregates standardized events according to business cases to form a complete process event sequence, and performs preprocessing and storage; specifically, it includes: M21, Case Association Unit: A multi-level primary key association mechanism is used to aggregate events scattered across different systems according to business cases. Outpatient visit case: The primary association key is "Patient ID + Visit Date", and the secondary association key is "Outpatient Number", which associates all events of the same patient on the same day in the HIS, LIS, PACS and pharmacy systems; Inpatient visit case: The primary associated key is "Hospitalization Number", which is associated with all examination, treatment, nursing, surgery, and medication events during hospitalization; Drug procurement case: The primary associated key is "purchase order number", which is associated with the entire process of application, approval, warehousing, delivery, and settlement. Equipment maintenance case: The main associated key is "maintenance order number", which is associated with the entire process of reporting, dispatching, maintenance, acceptance and settlement. For events that are not directly related across systems, a combination of time window matching (30 minutes before and after) and fuzzy matching of patient IDs is used for association, with an association accuracy of ≥99.5%.

[0031] M22, Event Sequence Preprocessing Unit: Sorts the aggregated event sequences by time to ensure that events are arranged in the order of execution; filters out obviously abnormal events; merges the same activities executed by the same executor within 5 minutes; and processes parallel activities, cyclic activities, and nested processes to prepare for subsequent process mining.

[0032] M23, Hybrid Storage Unit: Uses the time-series database InfluxDB to store raw event logs, supporting high-throughput writes and fast time-range queries; uses the graph database Neo4j to store process models and event relationships, supporting complex process queries and analysis. The two databases maintain data consistency through a scheduled data synchronization mechanism.

[0033] M3, Process Mining and Model Discovery Module: Used to automatically discover actual business process models from process event sequences, compare them with standard process models, and dynamically update the process models; specifically including: M31, Actual Business Process Model Discovery Unit: This unit automatically discovers actual business process models from event logs using an improved Alpha++ algorithm. The improved Alpha++ algorithm adds three improvements to the traditional algorithm: M311, Event Frequency Filtering: Introduces event frequency weights to filter out low-frequency abnormal events that occur below a threshold, thereby improving the accuracy of the process model; M312, Cyclic Activity Identification: Introducing a cyclic marker L, when an activity sequence A→B→A is detected and occurs ≥3 times, it is marked as a cyclic activity, generating a cyclic structure instead of a parallel structure; M313, Multi-instance Activity Support: Adds an instance counter N. When the same activity is executed N times by different resources in the same case, it is marked as a multi-instance activity, supporting two modes: "multi-instance serial" and "multi-instance parallel". The event filtering formula is as follows:

[0034] in, For a valid set of events, Let e ​​be the frequency of occurrence of event e. This is the frequency threshold, with a value range of 0.01-0.05, which can be adjusted according to actual business needs.

[0035] The algorithm can accurately identify parallel activities, cyclic activities, and nested processes, and generate flowcharts that conform to the BPMN 2.0 standard.

[0036] M32, Multi-dimensional Process Analysis Unit: In-depth analysis integrating time, resource, and role dimensions. In the time dimension, it analyzes the average execution time, longest execution time, waiting time, and queuing time for each activity; in the resource dimension, it analyzes the utilization rate and load of resources such as medical staff, equipment, and beds; in the role dimension, it analyzes the collaborative relationships and process flow patterns between different roles.

[0037] At the role level, the specific implementation method is as follows: First, construct a role-activity association matrix and count all activity types and execution frequencies performed by each role; then generate a directed graph of role interactions, where nodes represent different roles and the weight of the edges represents the number of activity transitions between two roles; next, calculate the degree centrality, betweenness centrality, and proximity centrality indices for each role to identify core collaborative roles and key collaborative paths in the process; finally, through sliding window time series analysis, track the changes in role collaboration modes at different time periods to discover collaboration bottlenecks and unreasonable role divisions.

[0038] M33, Process Model Comparison Unit: This unit compares the actual discovered process models with the hospital's pre-set standard process models, calculates the similarity between the two models, counts the number and proportion of cases deviating from the standard path, and generates a process deviation analysis report. Process model similarity is calculated using a graph edit distance-based method. First, convert the two BPMN2.0 process models into directed graphs. and ,in A set of nodes (representing activities). Let the graph be a set of edges (representing the flow relationships between activities); then calculate the graph... Convert to image The minimum total cost of the required graph editing operations (including node insertion, node deletion, node replacement, edge insertion, edge deletion, and edge replacement) is represented by the graph editing distance d(G1,G2). The formula for the final process model similarity is:

[0039] Where |V| is the number of nodes and |E| is the number of edges.

[0040] The statistical method for deviations from the standard path is as follows: the actual event sequence of each case is compared element by element with the standard event sequence corresponding to the standard process. When the actual sequence contains activities not included in the standard sequence, missing necessary activities in the standard sequence, or the execution order of activities is inconsistent with the standard sequence, the case is determined to have deviated from the standard path. The number of cases that deviate from the standard path in all cases is counted, the proportion of the total number of cases is calculated, and the cases are classified and statistically analyzed according to three types of deviation: new activities, missing activities, and incorrect order.

[0041] M34, Dynamic Update Unit for Process Model: The process mining algorithm is automatically re-executed every morning at midnight, updating the process model based on the previous day's event logs. When the similarity of the process model to the previous version is less than 0.8, manual review is triggered to ensure the accuracy of the process model.

[0042] M4, Bottleneck Identification and Anomaly Detection Module: Used for multi-dimensional identification of process bottlenecks, real-time detection of abnormal behaviors and control gaps deviating from the standard path, and tiered early warning; specifically including: M41, Multi-Dimensional Bottleneck Identification Unit: Identifies process bottlenecks from the perspectives of time, resources, and process structure. In the time dimension, it identifies activities with average execution time and average waiting time exceeding thresholds. In the resource dimension, it identifies resources with utilization exceeding 80%. In the process structure dimension, it identifies long paths, unbalanced parallel activities, frequent loops, and rework paths. Using the Critical Path Method (CPM) and the Theory of Bottlenecks (TOC), it pinpoints global bottlenecks affecting the overall process efficiency.

[0043] M42, Offline Anomaly Detection Unit: Employs the DBSCAN clustering algorithm and the Isolation Forest algorithm to discover abnormal process patterns from historical event logs. The DBSCAN algorithm is used to cluster normal process sequences, while the Isolation Forest algorithm is used to detect outlier abnormal sequences. It can detect various anomaly patterns such as abnormal jumps, abnormal execution sequences, and abnormal execution times.

[0044] M43, Real-time Anomaly Detection Unit: Employs Flink stream processing technology and a sliding window mechanism to monitor the process execution in real time. The event coding process is as follows: M431, Event Vectorization: The "Event Name", "Executor" and "Resource ID" of each event are encoded into 128-dimensional vectors using a pre-trained Word2Vec model, and then concatenated to obtain a 384-dimensional event vector. M432, Sequence padding: Pad process sequences of less than 32 with zero vectors to a fixed length of 32; M433, Transformer encoding: A 2-layer Transformer encoder is used, with a hidden layer dimension of 512, and the output is a 32×512-dimensional sequence feature vector; M434. Similarity Calculation: After performing global average pooling on the feature vectors of the actual sequence and the feature vectors of the standard sequence, calculate the anomaly score by fusing cosine similarity and edit distance.

[0045] The formula for calculating the anomaly score is as follows:

[0046] in, This represents the cosine similarity of the feature vectors after global average pooling. For the edit distance similarity of the process sequence, Edit distance similarity of the process sequence; The 512-dimensional feature vector is obtained by encoding the actual process sequence using a Transformer encoder and then performing global average pooling. This is a 512-dimensional feature vector obtained after the standard process sequence undergoes the same encoding and pooling operations. A sequence of activity names for the actual process sequence; This is the sequence of activity names corresponding to the standard process; when When a general warning is triggered, An emergency warning is triggered at that time.

[0047] Among them, the edit distance similarity of the process sequence The calculation formula is:

[0048] in, The Levenstein edit distance between the actual active sequence and the standard active sequence is the minimum number of operations required to convert the actual sequence into the standard sequence through insertion, deletion, and replacement operations. The length of the actual activity sequence. The length of the standard activity sequence.

[0049] M44, Control Deficiency Identification and Early Warning Unit: Based on a compliance rule engine in the form of "condition-action," this unit detects missing approval steps, unauthorized access operations, and behaviors that do not comply with hospital regulations in the workflow. The compliance rule engine supports three types of rules: M441, Sequence Rules: These rules specify the order in which activities are performed, such as "Examination requests must be submitted only after the doctor has reviewed them." M442, Permission Rules: Specifies the execution permissions for activities, such as "The use of high-value consumables must be approved by associate chief physicians or above"; M443. Time Rules: These rules specify time limits for the execution of activities, such as "Emergency laboratory reports must be issued within 30 minutes".

[0050] Rule Example: IF Inspection Request. Reviewer IDISNULLTHEN triggers "Inspection Request Not Reviewed" alert; Tiered alert handling process: General warning ( (0.6-0.7): Send to the department head nurse, requiring verification and processing within 2 hours, and enter the processing result into the system. Important warning ( 0.7-0.8): Send to the department head and operations management specialist, requiring verification and processing within 1 hour. Emergency warning ( >0.8): Send to the hospital's vice president in charge, department head, and operations management director, requesting verification and processing within 30 minutes. All warnings are set with timeout reminders. Warnings that are not processed within the processing time limit will be automatically escalated and a warning processing statistics report will be generated.

[0051] M5, Process Optimization and Visualization Application Module: This module automatically generates actionable optimization suggestions based on historical optimization cases and process simulations, quantitatively evaluates optimization effectiveness, and provides visual displays and report outputs; specifically including: M51, Optimization Suggestion Generation Unit: Based on multi-dimensional case reasoning technology, this unit matches the currently identified problem with historical optimization cases based on their similarity. Similarity calculation uses multi-dimensional weighted cosine similarity, matching the top three cases with the highest similarity. Combined with process simulation results, it automatically generates specific and executable process optimization suggestions, including resource allocation suggestions, process reengineering suggestions, and system improvement suggestions.

[0052] The formula for calculating the similarity of case reasoning is:

[0053] in, As a current problem case, For the first case in the historical case library One case, The semantic similarity of the problem descriptions. For the similarity of process structure, For the similarity of resource allocation, These are the weighting coefficients, and The default value is .

[0054] M52, Process Simulation Verification Unit: Constructs a digital twin model of the process based on Discrete Event Simulation (DES) technology. Specific implementation steps: M521. Import the actual process model and define the execution time distribution of activities (normal distribution, exponential distribution, etc.). M522. Define resource attributes, including resource quantity, working time, and service efficiency. M523. Set simulation parameters, including simulation duration, patient arrival rate, and resource scheduling strategy. M524. Run the simulation model and output the following metrics: total process time, resource utilization, patient waiting time, and queue length. M525. Compare the simulation results of different optimization schemes and select the optimal scheme.

[0055] The simulation model was implemented using AnyLogic software, and the error between the simulation results and the actual running results is less than 10%.

[0056] M53, Visualization Unit: Provides an intuitive web-based visualization interface, displaying the actual process model, process execution status, bottleneck locations, abnormal events, and optimization suggestions. It supports zooming, panning, and clicking to view details of the process model, and supports real-time monitoring on a large screen, providing decision support for managers.

[0057] M54. Effectiveness Evaluation and Closed-Loop Management Unit: Tracks the implementation of optimization plans, compares process indicators before and after optimization, such as average consultation time, examination and testing turnaround time, and patient satisfaction, to evaluate the optimization effect. Successful optimization cases are stored in a case library, continuously enriching the case library to improve the accuracy and relevance of optimization suggestions, forming a closed-loop management system of "problem identification - process optimization - effectiveness evaluation".

[0058] Example 2 This invention proposes a method for optimizing hospital operation processes and detecting anomalies, the specific steps of which include: S1. Non-invasively collect all operation logs from various business systems in the hospital, perform data cleaning and standardization processing, and build a unified process event model; S2. Aggregate the standardized events according to business cases to form a complete process event sequence, and preprocess and store them; S3. Automatically discover actual business process models from process event sequences, compare them with standard process models, and dynamically update the process models; S4. Multi-dimensional identification of process bottlenecks, real-time detection of abnormal behaviors and control gaps that deviate from the standard path, and hierarchical early warning; S5: Automatically generates actionable optimization suggestions based on historical optimization cases and process simulations, quantitatively evaluates optimization effects, and provides visual displays and report outputs. Example 3 This embodiment details the application of the hospital operation process optimization and anomaly detection system in outpatient and inpatient examination and testing processes. It provides algorithm performance test data and comparative experimental data with existing technologies, fully presenting the system's technical implementation path and application effects. Specifically, it includes: B1. Algorithm Performance Testing The algorithm's performance was tested using real operational data from a top-tier hospital over three months (a total of 1.2 million event records). The results are as follows: The improved Alpha++ algorithm achieved a process model accuracy of 92.3%, an improvement of 11.7% compared to the traditional Alpha++ algorithm, and a cyclic activity recognition accuracy of 89.6%. The real-time anomaly detection algorithm has an accuracy of 88.5%, a false positive rate of 4.2%, a false negative rate of 2.1%, and a detection time of less than 10ms for a single process sequence. The accuracy rate of matching optimization suggestions based on case reasoning was 85.7%, and the average time to generate an optimization suggestion was less than 2 seconds. B2. Comparison experiment with existing technologies Outpatient workflow data from the same hospital during the same time period was selected and managed using traditional manual management, a commercial workflow mining system, and the system of this invention, respectively. The results are compared below:

[0059] B3. Optimization of Outpatient Visit Process and Application of Anomaly Detection B31. Data Acquisition and Preprocessing: Collect operation logs from the HIS system for patient registration, triage, consultation, payment, examination, and medication dispensing. Collect one month's worth of outpatient data from a tertiary hospital, including 120,000 event records involving 35,000 patients.

[0060] B32. Process Model Discovery: An improved Alpha++ algorithm was used to discover the actual outpatient process model from the event logs. Significant differences were found between the actual outpatient process and the standard process. Approximately 40% of patients would undergo examinations after their initial consultation and then return to the doctor for a follow-up appointment, while the standard process did not include this follow-up step.

[0061] B33. Bottleneck Identification: Multi-dimensional analysis identified three main bottlenecks in the outpatient process: B331. The triage waiting time is too long, with an average waiting time of 25 minutes and a maximum waiting time of 60 minutes. B332. Insufficient doctor consultation time, with an average consultation time of only 6 minutes, leading to decreased patient satisfaction; B333: Long payment queue time, with an average payment time of 12 minutes.

[0062] B34. Anomaly Detection: Real-time detection of abnormal behaviors in the outpatient process, identifying the following common anomalies: B341. Approximately 5% of patients skip triage and go directly to their doctor. B342. Approximately 3% of doctors end a consultation without completing the medical record writing. B343. Approximately 2% of patients did not pick up their medication after paying.

[0063] The system provides real-time alerts for these abnormal behaviors, reminding administrators to handle them promptly.

[0064] B35. Optimization Suggestion Generation and Effect Evaluation: The system automatically generates the following optimization suggestions: B351. Increase the number of triage staff during peak hours (8:00-10:00 AM) from 2 to 4. B352. Implement time-slot appointments for medical treatment, with appointment times accurate to 15 minutes; B353. Promote self-service payment machines and mobile payment to reduce the number of manual payment windows.

[0065] Through process simulation verification, the optimized average patient consultation time can be reduced from 90 minutes to 65 minutes, a reduction of approximately 28%. One month after implementing the optimization plan, actual data shows that the average patient consultation time has been reduced to 68 minutes, the triage waiting time has been reduced to 12 minutes, the payment time has been reduced to 5 minutes, and the patient satisfaction score has increased from 82 to 91.

[0066] B4. Optimization of Inpatient Examination and Laboratory Procedures and Application of Abnormal Detection B41. Data Acquisition and Preprocessing: Collect operation logs from the HIS, LIS, and PACS systems for inpatient examination requests, specimen collection, specimen transportation, testing, report generation, and report review. A total of one month's worth of inpatient examination and testing data was collected, including 80,000 event records involving 12,000 patients.

[0067] B42. Process Model Findings: Frequent rework paths were found in the actual inpatient examination and testing process. Approximately 15% of specimens needed to be recollected due to non-compliance, and approximately 8% of reports needed to be re-reviewed due to errors.

[0068] B43. Control Deficiencies Identification: Based on the compliance rule engine, the following control deficiencies were identified: B431. Approximately 10% of examination requests were not reviewed by a doctor. B432, approximately 7% of the specimens were not identified; B433, Approximately 5% of the reports were not reviewed by a senior physician.

[0069] B44. Optimization Suggestion Generation and Effect Evaluation: The system automatically generates the following optimization suggestions: B441. Establish a specimen quality control system and strengthen the training of specimen collection personnel; B442. Optimize specimen transport routes and increase specimen transport frequency; B443. Improve the review process for inspection applications and reports, and mandate the review process.

[0070] One month after implementing the optimization plan, the specimen non-compliance rate dropped from 15% to 6%, the reporting error rate dropped from 8% to 3%, and the average turnaround time for inspection and testing was shortened from 24 hours to 16 hours, a reduction of approximately 33%.

[0071] B5. Overall Effect of Multiple Business Scenarios The system's application effects in other core business scenarios: B51. Drug Procurement Process: The procurement cycle has been shortened from an average of 7 days to 4 days, inventory turnover has increased by 35%, and the stockout rate has decreased from 8% to 2%. B52. Surgical Procedure: The average waiting time for surgery has been reduced from 3 days to 1.5 days, and the utilization rate of the operating room has increased from 65% to 82%. B53. Equipment maintenance process: The average equipment failure repair time has been reduced from 24 hours to 8 hours, and the equipment availability rate has been increased from 90% to 97%.

[0072] B6. Overall System Effectiveness Evaluation After three months of operation at a top-tier hospital, an overall effectiveness evaluation was conducted, and the results are as follows: B61. Process transparency is improved by 100%, realizing visualized management of the entire business process; B62. The time for detecting abnormal events has been reduced from an average of 24 hours to real-time, and the efficiency of abnormal event handling has been improved by 80%. B63. The average time for patients to seek medical treatment was reduced by 25%, and the average length of hospital stay was reduced by 12%. B64. Hospital operating costs decreased by 15%, and patient satisfaction increased by 12 percentage points.

[0073] The above specific embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to examples, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A hospital operation process optimization and anomaly detection system, characterized in that, include: Multi-source data acquisition and standardization module: used for non-intrusive acquisition of full operation logs from various business systems in the hospital, data cleaning and standardization processing, and construction of a unified process event model; Process event sequence construction module: used to aggregate standardized events according to business cases to form a complete process event sequence, and to preprocess and store it; Process mining and model discovery module: used to automatically discover actual business process models from process event sequences, compare them with standard process models, and dynamically update the process models; Bottleneck identification and anomaly detection module: used to identify process bottlenecks from multiple dimensions, detect abnormal behaviors and control gaps that deviate from the standard path in real time, and provide graded early warnings; Process optimization and visualization application module: It is used to automatically generate actionable optimization suggestions based on historical optimization cases and process simulations, quantitatively evaluate the optimization effect, and provide visualization and report output.

2. The hospital operation process optimization and anomaly detection system according to claim 1, characterized in that, The multi-source data acquisition and standardization module includes: Non-intrusive data acquisition unit: Used to collect operation logs, database change logs, and API call logs of HIS, LIS, PACS, EMR, HRP, and OA systems through log file parsing, database CDC, message queue subscription, and API interface calls, and defines standardized collection fields for different systems; Data cleaning and preprocessing unit: used to remove duplicates, complete data, convert formats, and filter noise from raw data, and to handle missing values, outliers, and duplicate events; Unified event model building unit: used to define standard process event formats and map heterogeneous data from different systems to unified process events through a field mapping table.

3. The hospital operation process optimization and anomaly detection system according to claim 2, characterized in that, The mathematical representation of the unified process event is as follows: in, As a unique identifier for the event, Used as a unique identifier for the case. The name of the event. As an executor, The start time of the event. This is the end time of the event. For the resource ID used, This is a set of business attributes.

4. The hospital operation process optimization and anomaly detection system according to claim 1, characterized in that, The process event sequence construction module includes: Case association unit: used to aggregate scattered events according to business rules and data relationships, such as patient visits, drug procurement, equipment maintenance, etc., to form a complete process event sequence; Event sequence preprocessing unit: used for time sorting of event sequences, filtering of abnormal events and merging of events, and handling parallel activities, cyclic activities and nested processes; Hybrid storage unit: Uses InfluxDB, a time-series database, to store raw event logs, and Neo4j, a graph database, to store process models and event relationships, supporting efficient process querying and analysis.

5. The hospital operation process optimization and anomaly detection system according to claim 1, characterized in that, The process mining and model discovery module includes: Actual process model discovery unit: Used to automatically discover actual business process models from event logs using improved Alpha++ algorithm and heuristic mining algorithm, and generate flowcharts in BPMN2.0 standard format; Multi-dimensional process analysis unit: used to integrate time, resource and role dimensions for in-depth analysis of activity execution time distribution, resource utilization, role collaboration relationships and process flow patterns; Process Model Comparison Unit: Used to compare the actual process model with the hospital's pre-set standard process model to identify process deviations and differences; Dynamic update unit for process model: used to periodically re-execute the process mining algorithm to dynamically update the process model.

6. The hospital operation process optimization and anomaly detection system according to claim 5, characterized in that, The improved Alpha++ algorithm introduces event frequency weights to filter low-frequency abnormal events. Its event filtering formula is as follows: in, For a valid set of events, Let e ​​be the frequency of occurrence of event e. This is the frequency threshold, with a value range of 0.01-0.

05.

7. The hospital operation process optimization and anomaly detection system according to claim 1, characterized in that, The bottleneck identification and anomaly detection module includes: Multi-dimensional bottleneck identification unit: used to identify process bottlenecks from the dimensions of time, resources and process structure, and to locate global bottlenecks using the critical path method and bottleneck theory; Offline anomaly detection unit: Used to discover abnormal process patterns from historical event logs using DBSCAN clustering algorithm and Isolation Forest algorithm; Real-time anomaly detection unit: Used with Flink stream processing technology and sliding window mechanism to monitor the process execution in real time and detect abnormal behavior that deviates from the standard path through sequence alignment algorithm; Control Deficiency Identification and Early Warning Unit: Based on the compliance rule engine, it detects missing approval steps, unauthorized operations, and behaviors that do not comply with hospital rules and regulations in the process, and provides graded early warnings according to the severity of the anomalies.

8. The hospital operation process optimization and anomaly detection system according to claim 7, characterized in that, The real-time anomaly detection uses an attention-based Transformer encoder to encode the process event sequence, and the anomaly score is calculated using the following formula: in, This represents the cosine similarity of the feature vectors after global average pooling. The edit distance similarity of the process sequence; The 512-dimensional feature vector is obtained by encoding the actual process sequence using a Transformer encoder and then performing global average pooling. This is a 512-dimensional feature vector obtained after the standard process sequence undergoes the same encoding and pooling operations. A sequence of activity names for the actual process sequence; This is the sequence of activity names corresponding to the standard process; when When a general warning is triggered, An emergency warning is triggered at that time.

9. The hospital operation process optimization and anomaly detection system according to claim 1, characterized in that, The process optimization and visualization application module includes: Optimization suggestion generation unit: Based on multi-dimensional case reasoning technology, it matches the currently identified problem with historical optimization cases based on similarity and generates executable optimization suggestions by combining process simulation results; Process simulation and verification unit: used to build a digital twin model of a process based on discrete event simulation technology, and to simulate the execution effect of different optimization schemes; Visualization unit: Used to provide a web-based visualization interface to display the actual process model, process execution status, bottleneck locations, abnormal events, and optimization suggestions; Effectiveness Evaluation and Closed-Loop Management Unit: Used to track the implementation of optimization plans, compare process indicators before and after optimization, evaluate the optimization effect, and store successful optimization cases in the case library.

10. A method for optimizing hospital operation processes and detecting anomalies, characterized in that, The hospital operation process optimization and anomaly detection system according to any one of claims 1-9 includes the following steps: S1. Non-invasively collect all operation logs from various business systems in the hospital, perform data cleaning and standardization processing, and build a unified process event model; S2. Aggregate the standardized events according to business cases to form a complete process event sequence, and preprocess and store them; S3. Automatically discover actual business process models from process event sequences, compare them with standard process models, and dynamically update the process models; S4. Multi-dimensional identification of process bottlenecks, real-time detection of abnormal behaviors and control gaps that deviate from the standard path, and hierarchical early warning; S5 automatically generates actionable optimization suggestions based on historical optimization cases and process simulations, quantitatively evaluates optimization effects, and provides visual displays and report outputs.

Images