Medical institution dynamic risk early warning method and device and storage medium

Abstract

The invention discloses a medical institution dynamic risk early warning method and device and a storage medium. Relates to the technical field of smart medical. According to the method, equipment parameters are collected in real time, and load self-adaptive dynamic threshold early warning is realized based on a quantile regression forest model; quantifying an influence score of the associated equipment by using a pre-constructed knowledge graph; and fault injection simulation is carried out by means of a digital twin model, the root cause is verified by comparing simulation data with real data, and finally accurate early warning information is pushed to a person in charge. According to the method, the problems of high false alarm and missing alarm rate and difficulty in multi-alarm root cause positioning of traditional fixed threshold early warning are solved, and accurate and automatic fault diagnosis and early warning are realized.

Description

Technical Field

[0001] This invention relates to the field of medical safety technology, specifically to a method, device, and storage medium for dynamic risk early warning in medical institutions. Background Technology

[0002] Medical institutions, especially large hospitals, are complex and dynamic systems comprised of buildings, medical equipment, information systems, and personnel activities. Their safe and stable operation is directly related to patients' health and the continuity of medical services. In recent years, with the expansion of hospital scale and the widespread use of advanced medical equipment, the complexity of infrastructure has increased significantly, making traditional manual inspection and static threshold monitoring models ineffective in addressing the growing security risks.

[0003] Currently, risk warnings in medical institutions primarily rely on sensors and monitoring systems deployed on critical equipment. These systems typically trigger alarms based on preset, fixed thresholds. For example, an alarm might be triggered when the room temperature in an ICU exceeds 28°C or the central oxygen supply pressure drops below 0.4 MPa. However, this static threshold method has significant drawbacks: First, the normal operating parameters of medical equipment fluctuate dynamically with its load (such as the frequency of operating room use and the number of patients admitted) and environmental conditions (such as seasonal changes). Fixed thresholds cannot adapt to this dynamism, potentially leading to false alarms during low-load periods and missed alarms during high-load periods due to overly broad threshold settings, resulting in low accuracy of warnings.

[0004] Secondly, when multiple devices or systems malfunction simultaneously, existing technologies lack effective root cause localization capabilities. For example, when abnormal operating room pressure, air conditioning system alarms, and medical gas pressure fluctuations occur simultaneously, maintenance personnel find it difficult to quickly determine which core device (such as a refrigeration unit, vacuum pump, or zone valve) caused this chain reaction. Furthermore, each monitoring system typically operates independently, forming "information silos." Summary of the Invention

[0005] The technical problem to be solved by this application is to overcome the shortcomings of the prior art and provide a method, device and storage medium for dynamic risk early warning in medical institutions.

[0006] Firstly, a dynamic risk early warning method for medical institutions is provided, including the following:

[0007] Real-time collection of parameter information from relevant equipment at monitoring points in medical institutions;

[0008] Based on the collected parameter information, the load of relevant equipment at the monitoring point is obtained. According to the current load, the threshold for triggering early warning is dynamically set based on the pre-trained quantile regression forest model. When the load exceeds the historical data range, the operating status of the equipment under the current load is simulated by a digital twin model pre-built based on all equipment at the monitoring point, and the simulated equipment operating parameters are fed back to the quantile regression forest model for incremental learning. The load is the amount of work undertaken or the pressure borne by the corresponding equipment to meet external demands per unit time.

[0009] When multiple device alerts occur within a set time range, the influence score of the devices associated with the alert locations is quantified based on a pre-built knowledge graph. The knowledge graph is constructed based on the physical connection relationships, logical dependencies, and spatial relationships between the devices related to the locations to be monitored. The physical connection relationship refers to the tangible entity connection between devices or systems that can transmit energy, matter, or signals. The logical dependency relationship refers to the functional dependence and support between devices or systems. The spatial relationship relationship refers to the relationship between devices or systems caused by their physical location and spatial layout.

[0010] Based on the influence score, the candidate devices associated with the warning location are sorted or combined, and the operating parameters of each candidate device are obtained at this time.

[0011] Initialize the digital twin model pre-built based on all devices at the monitoring points, and reset the state of all devices in the digital twin to the normal state before the fault occurred;

[0012] Based on the operating parameters of the candidate point devices, the operating conditions of the candidate devices in the digital twin model are set in an arranged order or combination to simulate fault injection.

[0013] Start the simulation engine of the digital twin model and let it run dynamically based on physical laws. The simulation time covers the complete period from the injection of fault to the stable manifestation of all derivative phenomena. Record the simulation process and statistically analyze the parameter curves of the relevant equipment at the points that triggered the alarm in the real world in the digital twin model.

[0014] The similarity between the simulated parameter curves and the parameters of the mapped devices in the real world is compared to determine whether a set threshold has been reached. The relevant parameters of the candidate devices that have reached the set threshold are pushed to the corresponding maintenance personnel, and an early warning is issued.

[0015] Furthermore, the method for setting the threshold for triggering the early warning includes:

[0016] Acquire data on the equipment under normal historical conditions, construct feature variables to characterize the equipment's operating conditions and the external environment, such as load indicators, ambient temperature, and equipment operating time, and construct target variables for key operating parameters of the equipment that need to be monitored and evaluated for abnormality.

[0017] A conditional quantile prediction model is trained using the quantile regression forest algorithm to learn the positive band statistical distribution of key operating parameters of equipment under specific operating conditions. The conditional quantile prediction model includes:

[0018] The first quantile regression forest model is used to learn and predict the upper quantile of the normal fluctuation of the target variable when the device is in a specific operating condition defined by the feature variables.

[0019] The second quantile regression forest model is used to learn and predict the lower quantile of the normal fluctuation of the target variable when the device is in a specific operating condition defined by the feature variables.

[0020] The system collects current characteristic variable data of the equipment in real time to characterize the current real-time operating condition of the equipment. The conditional quantile prediction model outputs a warning threshold that matches the current operating condition based on the upper and lower quantiles.

[0021] Furthermore, the method for constructing the knowledge graph includes:

[0022] The entities to be included in the knowledge graph are identified and defined. The entities include at least device entities and monitoring point entities. However, no unique identifier is assigned to each entity and attribute information is recorded.

[0023] Based on predefined physical connections, logical dependencies, and spatial relationships, directed connections are established between entities.

[0024] Each established relationship edge is assigned a weight value, which is used to quantify the strength or influence of the relationship.

[0025] Furthermore, the method for quantifying the influence score of the location associated with the early warning location includes:

[0026] Starting with the devices at all warning locations, the system traverses the pre-built knowledge graph along physical connections, logical dependencies, and spatial relationships to extract associated devices and connections, and constructs a warning association subgraph.

[0027] The topological influence score of each node in the subgraph is calculated using the weighted PageRank algorithm. The formula is: PR(v) = (1-d) / N + d × Σ[(PR(u) × w(u→v)) / Σw(u→k)]. Where: PR(v) represents the PageRank value of node v, d is the damping coefficient, ranging from 0.7 to 0.9, N is the total number of nodes in the subgraph, u is the set of all source nodes pointing to node v, PR(u) represents the PageRank value of source node u, w(u→v) is the weight of the relation edge from node u to node v, and Σw(u→k) is the sum of the weights of all outgoing edges from node u.

[0028] The calculation of the topology influence score includes the following sub-steps: initializing the PageRank value of each node to 1 / N; iteratively calculating the PageRank value of each node until the change in the PageRank value of all nodes is less than a preset threshold; normalizing the final PageRank value of all nodes to obtain the topology influence score of each node.

[0029] Furthermore, the method for calculating the similarity between the simulated parameter curves and the parameters of the corresponding devices in the real world includes:

[0030] Align the simulation parameter curves with the parameters of the corresponding devices in the real world on the time axis, and normalize the category data so that its value range is [0,1].

[0031] For each monitored parameter, calculate the dynamic time warping distance between its simulated time series S and the actual time series R;

[0032] The dynamic time-normalized distance is converted into a similarity score, and the distance is mapped to the interval (0,1], where 1 represents complete similarity;

[0033] The similarity scores of multiple types of device parameters are weighted and fused to obtain a comprehensive similarity score.

[0034] In a second aspect, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory, characterized in that the processor executes the computer program to implement the method described in the first aspect.

[0035] Thirdly, a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the method described in the first aspect.

[0036] (ii) Beneficial effects

[0037] Compared with existing technologies, this application employs a dynamic threshold setting method optimized based on a quantile regression forest ensemble digital twin model. This allows the system to adaptively adjust the warning threshold according to the real-time load (i.e., actual working pressure) of the equipment. This overcomes the shortcomings of traditional fixed threshold methods, which suffer from false alarms due to overly strict thresholds at low loads and false alarms due to overly broad thresholds at high loads. In complex situations with multiple concurrent warnings, this method uses a pre-constructed knowledge graph to quantify the influence relationships between warning points. This method can automatically identify the most critical and potentially chain-reaction-prone equipment from interconnected alarm information as candidate root causes, transforming the traditional "isolated alarm" analysis mode into a "systematic problem localization" mode, thus shortening troubleshooting time. The fault state of the candidate root cause equipment is simulated in a highly realistic virtual model, and the consistency between the simulation results and actual alarm data is compared, providing objective evidence based on physical laws and data comparison for root cause judgment. This process reduces the subjectivity and uncertainty of judgments based solely on experience or single indicators, making the warning conclusions more reliable. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0039] Figure 1 This is a flowchart of the dynamic risk early warning method for medical institutions in this application. Detailed Implementation

[0040] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Unless otherwise defined, the technical or scientific terms used in this application should have the ordinary meaning understood by those skilled in the art. The words "first," "second," and similar terms used in this application do not indicate any order, quantity, or importance, but are only used to distinguish different components. The words "comprising" or "including," etc., mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but do not exclude other elements or objects. The words "connected," "coupled," or "connected," etc., are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," etc., are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0041] The dynamic risk early warning method for medical institutions in this embodiment is applicable to complex fault diagnosis scenarios of critical hospital infrastructure. Especially when multiple critical points such as operating rooms, ICUs, medical gas systems, and power supply and distribution systems simultaneously experience abnormal alarms, it can accurately locate the root cause equipment (such as central refrigeration units, main oxygen supply pipelines, or core power distribution cabinets) that triggers a chain reaction of faults through dynamic threshold early warning, knowledge graph correlation analysis, and digital twin simulation verification. The verified fault information is then pushed to the corresponding operation and maintenance team, thereby ensuring the continuity of medical services and patient safety. Its core framework includes a real-time data acquisition and load perception module, a dynamic threshold calculation module, a knowledge graph analysis module, a digital twin simulation verification module, and an early warning push module. The real-time data acquisition module is responsible for acquiring equipment operating parameters; the dynamic threshold module adaptively generates early warning thresholds based on the real-time load of the equipment using a quantile regression forest model; the knowledge graph module calculates influence scores through the physical, logical, and spatial relationships between devices; the digital twin module verifies the root cause through fault injection and physical simulation; and finally, the early warning push module accurately sends the confirmed root cause information to the responsible party.

[0042] The preferred embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application. Furthermore, the embodiments and features in the embodiments of this application can be combined with each other without conflict.

[0043] like Figure 1As shown in the embodiments of this application, the dynamic risk early warning method for medical institutions includes the following:

[0044] Real-time collection of parameter information from relevant equipment at monitoring points in medical institutions. These points can include: power distribution system equipment: high-voltage switchgear, transformers, low-voltage switchgear, UPS (Uninterruptible Power Supply), emergency generators, distribution boxes, voltage stabilizers; HVAC system equipment: chillers, cooling towers, circulating water pumps, air handling units, fan coil units, precision air conditioners, fresh air units, exhaust fans; medical gas system equipment; water supply and drainage system equipment: domestic water pumps, fire pumps, sewage treatment equipment, water tanks, sump pumps; fire protection system equipment: fire alarm controllers, smoke / heat detectors, fire pumps, sprinkler pumps, gas extinguishing devices; and information infrastructure equipment: core network switches, servers, storage arrays, and precision air conditioning in the computer room.

[0045] The collected parameters are mainly divided into three categories:

[0046] 1. Operating status parameters (directly reflect the equipment's "load" and health status).

[0047] Electrical parameters: voltage, current, active power, reactive power, power factor, frequency, and electrical energy (power consumption).

[0048] Mechanical / thermal parameters: temperature (bearing temperature, coil temperature, water temperature, oil temperature), pressure (water pressure, air pressure, oil pressure), flow rate (water flow rate, gas flow rate), speed (pump, fan), vibration amplitude.

[0049] Status indications: Start / Stop status, Run / Standby / Fault status, Valve opening degree, Switch status.

[0050] 2. Environmental parameters (reflecting the spatial conditions of the equipment).

[0051] Computer room / equipment room environment: temperature, humidity, water leakage monitoring, smoke concentration.

[0052] Medical service area environment: room temperature, humidity, pressure difference (used to control cleanliness), carbon dioxide concentration.

[0053] 3. Output performance parameters (reflecting the quality of service provided by the equipment).

[0054] Power supply and distribution system: output voltage stability, harmonic distortion rate, UPS battery backup time.

[0055] Air conditioning system: supply / return air temperature, cooling / heating output, filter differential pressure (reflecting the degree of blockage).

[0056] Medical gas systems: outlet pressure, gas purity, dew point (a measure of dryness).

[0057] Information systems: network port traffic, packet loss rate, latency, server CPU / memory / disk utilization.

[0058] Based on the collected parameter information, the load of relevant equipment at the monitored locations is obtained. Based on the current load, a pre-trained quantile regression forest model is used to dynamically set the threshold for triggering early warnings. Load refers to the workload or pressure borne by the corresponding equipment per unit time to meet external demands. The quantification method and specific indicators of load vary depending on the equipment type. For example, for power distribution equipment (UPS, transformers), the load is the transmitted electrical power, i.e., the total power demand of all downstream electrical equipment. Real-time load rate = (current output active power / rated capacity) × 100%. For medical gas equipment (liquid oxygen pumps, vacuum pumps), the load is the volumetric flow rate of the gas to be delivered, i.e., the instantaneous gas consumption of the entire hospital or area. For example, in the afternoon of summer, the entire hospital operates at full capacity, and operating rooms use a large number of electrosurgical units and other equipment. The load rate of the power distribution circuit supplying power to the operating rooms reaches 85%. Due to the large current, the contact temperature of the circuit breakers in the distribution cabinet will normally rise to 65°C. Under 85% load, the 95th quantile (normal upper limit) of the contact temperature may be 70°C. Therefore, the alarm only sounds when the temperature exceeds 70°C. Late at night, only the emergency room and ICU require basic power. The load on the same power distribution circuit is only 15%. With low current, the contact temperature should remain stable at around 40°C.

[0059] At 15% load, the upper limit of the 95th percentile of temperature is likely 45°C. If the temperature reaches 50°C at this time, although the absolute value is not high, it is significantly abnormal relative to the current load, which will trigger an alarm, indicating that there may be hidden dangers such as poor contact.

[0060] The training methods for the quantile regression forest model include:

[0061] Step 1: Construct the feature variable (X) and the target variable (y).

[0062] First, extract data from the historical database during the period when the device was operating normally and without any faults.

[0063] Characteristic variables (X) are used to describe the operating conditions of the equipment and the external environment. These include load parameters, ambient temperature parameters, and equipment operating time parameters. Ambient temperature refers to the temperature of the environment in which the equipment is located. It represents the main external factor affecting the equipment's heat dissipation efficiency. Equipment operating time refers to the continuous operating time from this startup to the current moment, taking into account time-related effects such as potential heat accumulation and performance degradation.

[0064] The target variable (y) refers to the key parameter that needs to be monitored and its abnormality determined. Examples include bearing temperature, winding temperature, outlet pressure, and vibration amplitude. The health status of these parameters directly reflects the health status of the equipment.

[0065] For each normal time point t in history, there is a record: (X_t, y_t) = (load_t, ambient temperature_t, running time_t, ..., target parameter_t).

[0066] Quantile regression forests predict the conditional quantiles of y, that is, what percentage probability y is lower than the conditional mean under given conditions.

[0067] A first quantile regression forest model is trained to provide early warning of upper limits. Specifically, a function Q_high(X) is learned such that P(y≤Q_high(X)|X)=τ_high. τ_high is a preset high quantile, for example, 0.95. Its significance lies in the fact that when the equipment is in operating condition X, there is a 95% probability that the value of its target parameter y will fall below the value of Q_high(X). Therefore, Q_high(X) is the normal upper limit under this operating condition.

[0068] A second quantile regression forest model is trained to provide an early warning lower limit. Specifically, a function Q_low(X) is learned such that P(y≤Q_low(X)|X)=τ_low. τ_low is a preset low quantile, for example, 0.05. Its significance is that when the equipment is in operating condition X, there is only a 5% probability that the value of its target parameter y will be lower than Q_low(X). Therefore, Q_low(X) is the normal lower limit under this operating condition.

[0069] Once the quantile regression forest model is trained, the operating parameters of the equipment are collected in real time to obtain the current operating load of the equipment, and the threshold for triggering early warnings is dynamically set through the quantile regression forest model.

[0070] When the load exceeds the historical data range, a newly built department is put into operation, and the load reaches its historical peak. At this time, the reliability of the quantile regression forest model, which is pre-trained based on historical data statistics, will be greatly reduced. Therefore, in this embodiment, a digital twin model is used for hypothesis analysis to supplement and optimize the historical data. Specifically, a digital twin model pre-built based on all equipment at the monitoring points is used to simulate the operating status of the equipment under the current load, and the simulated equipment operating parameters are fed back to the quantile regression forest model for incremental learning to correct and optimize future dynamic thresholds.

[0071] The knowledge graph is constructed based on the physical connection relationships, logical dependencies, and spatial association relationships between related devices at the monitoring points. The physical connection relationship refers to the tangible physical connection between devices or systems that can transmit energy, matter, or signals. The logical dependency relationship refers to the functional dependence and support between devices or systems. The spatial association relationship refers to the relationship between devices or systems caused by their physical location and spatial layout.

[0072] The methods for constructing knowledge graphs include:

[0073] Identify and define entities to be included in the knowledge graph. These entities include at least equipment entities (e.g., UPS-001, precision air conditioner-A01, server-SVR-05), monitoring point entities (e.g., temperature sensor-TS1001, pressure sensor-PS2002), and spatial entities (e.g., Building 1, East Zone of the Third Floor, Operating Room-301, Information Room-DC01). Assign a unique identifier to each entity and record its attribute information.

[0074] Directed connections are established between entities based on predefined physical connections, logical dependencies, and spatial relationships. Each edge is assigned a weight value, which quantifies the strength or influence of the relationship. Once constructed, the connections are stored in a graph database, such as Neo4j, and complex relational queries can be efficiently handled using query languages ​​like Cypher.

[0075] When the system detects multiple devices triggering warnings within a set timeframe, it traverses a pre-built hospital infrastructure knowledge graph, starting with these warning devices. The traversal employs a breadth-first search (BFS) algorithm, proceeding along the aforementioned relational edges to extract all device nodes directly or indirectly associated with the warning devices and their connections, thus constructing a warning association subgraph for subsequent analysis. This subgraph retains all relational attributes from the original graph.

[0076] An initial PageRank value is assigned to each node in the warning association subgraph. Initialization typically employs a uniform distribution strategy. The topological influence score of each node in the subgraph is calculated based on the weighted PageRank algorithm, using the formula: PR(v) = (1-d) / N + d × Σ[(PR(u) × w(u→v)) / Σw(u→k)]. Where: PR(v) represents the PageRank value of node v, d is the damping coefficient, ranging from 0.7 to 0.9, N is the total number of nodes in the subgraph, u is the set of all source nodes pointing to node v, PR(u) represents the PageRank value of source node u, w(u→v) is the weight of the relational edge from node u to node v, and Σw(u→k) is the sum of the weights of all outgoing edges from node u.

[0077] The iterative process is as follows:

[0078] 1. Based on the initial PR value, calculate the new PR value for all nodes according to the formula above.

[0079] 2. Assign the newly calculated PR values ​​to each node, replacing the old values.

[0080] 3. Repeat steps 1 and 2 until the PR value of all nodes changes less than a preset convergence threshold (e.g., 0.00001), indicating that the calculation results have stabilized.

[0081] 4. The PR values ​​of each node obtained at this point are the final, unnormalized raw influence scores.

[0082] The original PageRank values ​​of all nodes obtained after iterative convergence are normalized so that they fall within the interval [0,1], thus obtaining the final topological influence score.

[0083] Candidate devices associated with warning locations are sorted or combined based on their influence scores, and the operating parameters of each selected device are obtained. They can be sorted from highest to lowest score and fed into the digital twin model one by one for fault simulation until the fitting effect meets expectations. If individual simulations fail to meet expectations, two devices can be simulated at a time until the fitting effect reaches expectations. Similarly, if the fault simulation of two devices is insufficient, the number of simulated faulty devices can be increased. The specific digital twin system simulation verification method is as follows:

[0084] Initialize a pre-built digital twin model based on the relevant equipment at the monitoring points, and reset the state of all equipment in the digital twin to the normal state before the fault occurred; according to the operating parameters of the candidate point equipment, set the operating conditions of the candidate point equipment in the digital twin model in an arranged order or combination to simulate fault injection; start the simulation engine of the digital twin model, allowing it to run dynamically based on physical laws, with the simulation time covering the complete period from fault injection to the stable manifestation of all derivative phenomena; record the simulation process, and statistically analyze the parameter curves of the relevant equipment at the points that triggered alarms in the real world in the digital twin model;

[0085] The similarity between the simulated parameter curves and the actual parameters of the corresponding devices in the real world is compared to determine whether the set threshold has been reached. The relevant parameters of the devices at the candidate points that have reached the set threshold are pushed to the corresponding responsible personnel, and an early warning is issued.

[0086] The method for calculating the similarity between the simulated parameter curve and the corresponding parameters of the real-world device includes: aligning the simulated parameter curve and the parameters of the real-world device on the time axis, and normalizing the category data to make its value range [0,1]; for each monitored parameter, calculating the dynamic time warping distance between its simulated time series S and the real time series R; converting the dynamic time warping distance into a similarity score, mapping the distance to the (0,1] interval, where 1 represents complete similarity; finally, weighted fusion of the similarity scores of multiple types of device parameters to obtain a comprehensive similarity score.

[0087] In this embodiment of the application, the method for constructing a digital twin model includes the following steps:

[0088] Based on Building Information Modeling (BIM), laser scanning, or photogrammetry, a three-dimensional geometric model containing precise spatial information of equipment, pipes, and cables is constructed. At the same time, multi-source heterogeneous data is accessed and integrated, including real-time data from IoT sensors, business data from hospital information systems, equipment management data, and environmental data, forming a unified data foundation.

[0089] Based on the aforementioned three-dimensional geometric model, multiple dynamic behavior models are integrated, including:

[0090] Simulation models based on physical laws are used to simulate the system behavior of thermal energy, fluids, and circuits; data analysis models based on machine learning are used to predict equipment performance degradation, load forecasting, and anomaly detection; rules and knowledge models based on knowledge graphs are used to encode domain standards and expert experience.

[0091] Historical data is used to simulate and verify the integrated digital twin model, and the consistency between the simulation output and the actual records is compared. The model parameters are then iteratively optimized through parameter calibration to make the simulation behavior closer to the real system behavior.

[0092] Develop an application programming interface (API) or a visual interactive interface for the digital twin model to support status monitoring, scenario simulation, and early warning decision-making.

[0093] Corresponding to the above method, this application embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the above method.

[0094] This application employs a dynamic threshold setting method optimized based on a quantile regression forest ensemble digital twin model. The system can adaptively adjust the warning threshold according to the real-time load (i.e., actual working pressure) of the equipment. This overcomes the shortcomings of traditional fixed threshold methods, which suffer from false alarms due to overly strict thresholds at low loads and false alarms due to overly broad thresholds at high loads. In complex situations with multiple concurrent warnings, this method uses a pre-constructed knowledge graph to quantify the influence relationships between warning points. This method can automatically identify the most critical and potentially chain-reaction-prone equipment from interconnected alarm information as candidate root causes, transforming the traditional "isolated alarm" analysis mode into a "systematic problem localization" mode, thus shortening fault diagnosis time. The fault state of the candidate root cause equipment is simulated in a highly realistic virtual model, and the consistency between the simulation results and actual alarm data is compared, providing objective evidence based on physical laws and data comparison for root cause judgment. This process reduces the subjectivity and uncertainty of judgments based solely on experience or single indicators, making the warning conclusions more reliable.

[0095] The functions of each functional unit of the electronic device provided in the above embodiments of this application can be implemented through the above methods and steps. Therefore, the specific working process and beneficial effects of each unit in the electronic device provided in the embodiments of this application will not be repeated here.

[0096] Corresponding to the above method, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above method.

[0097] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform any of the methods described in the above embodiments.

[0098] Those skilled in the art will understand that the embodiments in this application can be provided as methods, systems, or computer program products. Therefore, the embodiments in this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments in this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0099] This application describes embodiments of methods, apparatus (systems), and computer program products according to embodiments of this application with reference to flowchart illustrations and / or block diagrams. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0100] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0101] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0102] Although preferred embodiments have been described in this application, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of this application.

[0103] Obviously, those skilled in the art can make various modifications and variations to the embodiments of this application without departing from the spirit and scope of the embodiments of this application. Therefore, if these modifications and variations to the embodiments of this application fall within the scope of the claims in this application and their equivalents, then this application also intends to include these modifications and variations.

Claims

1. A medical institution dynamic risk early warning method, characterized in that, Comprise the following: Real-time acquisition of parameter information of all devices at the monitoring point of the medical institution; According to the acquired parameter information, the load of the device at the monitoring point is obtained, and the threshold value of the early warning trigger is dynamically set based on the pre-trained quantile regression forest model according to the current load. When the load exceeds the historical data range, the running state of the device under the current load is simulated through the pre-constructed digital twin model of all devices at the monitoring point, and the simulated device running parameters are fed back to the quantile regression forest model for incremental learning. The load is the workload or pressure borne by the corresponding device in unit time to meet external demand; When multiple devices are warned within a set time range, the influence score of the device associated with the warning point is quantified based on the pre-constructed knowledge graph; The knowledge graph is constructed based on the physical connection relationship, logical dependency relationship and spatial correlation relationship between the related devices at the monitoring point. The physical connection relationship refers to the tangible entity connection between devices or systems that can transmit energy, matter or signals. The logical dependency relationship refers to the functional dependency and support between devices or systems. The spatial correlation relationship refers to the relationship between devices or systems resulting from their physical location and spatial layout. According to the influence score, the candidate devices associated with the warning point are sorted or combined, and the running parameters of each candidate device at this time are obtained; Initialize the digital twin model pre-constructed based on all devices at the monitoring point, and reset the state of all devices in the digital twin to the normal state before the fault occurs; According to the running parameters of the candidate point device, set the running condition of the candidate device in the digital twin model according to the order or combination, and simulate fault injection; Start the simulation engine of the digital twin model to dynamically run based on physical laws, and the simulation time covers the complete period from fault injection to the stable appearance of all derived phenomena; Record the simulation process and count the parameter curves of the related devices at the point in the real world that triggered the alarm in the digital twin model; According to the similarity comparison between the simulated parameter curves and the parameters of the devices mapped in the real world, it is judged whether the set threshold is reached, and the candidate device related parameters that reach the set threshold are pushed to the corresponding maintenance personnel and a warning is given.

2. The dynamic risk early warning method of the medical institution according to claim 1: the threshold value setting method of the early warning trigger comprises: Obtain the data of the device in the historical normal running state, construct the feature variables representing the load index of the device running condition and external environment, environmental temperature and device running time, and construct the target variables of the key running parameters of the device that need to be monitored and evaluated for abnormality; A conditional quantile prediction model is trained and generated using a quantile regression forest algorithm to learn the statistical distribution of the key running parameters of the device under a specific working condition. The conditional quantile prediction model comprises: A first quantile regression forest model is used to learn and predict the upper quantile of the normal fluctuation of the target variable when the device is in a specific working condition defined by the feature variables. A second quantile regression forest model is used to learn and predict: when the device is in a specific working condition defined by the feature variables, the lower quantile of the normal fluctuation of the target variable; Real-time acquisition of the current feature variable data of the device to represent the current real-time running working condition of the device, and the conditional quantile prediction model outputs a warning threshold matched with the current working condition according to the upper quantile and the lower quantile.

3. The medical institution dynamic risk early warning method according to claim 2, characterized in that: The method for constructing the knowledge graph comprises: Identifying and defining entities to be included in the knowledge graph, the entities at least including device entities and monitoring point entities, and not assigning a unique identifier to each entity and recording attribute information; Based on the pre-defined physical connection relationship, logical dependency relationship and spatial correlation relationship, a directed connection between entities is established, Assign a weight value to each established relationship edge, and the weight value is used to quantify the strength or influence degree of the relationship.

4. The medical institution dynamic risk early warning method according to claim 3, characterized in that: The method for quantifying the point influence score associated with the warning point comprises: Taking the device of all warning points as the starting point, traversing along the physical connection, logical dependency and spatial correlation relationship in the pre-constructed knowledge graph, extracting the associated devices and connection relationships, and constructing a warning correlation subgraph; Based on the weighted PageRank algorithm, the topological influence score of each node in the subgraph is calculated, and the calculation formula is: PR(v)=(1-d) / N+d×Σ[(PR(u)×w(u→v)) / Σw(u→k)]; Wherein: PR(v) represents the PageRank value of node v, d is a damping coefficient, the value range is 0.7-0.9, N is the total number of nodes in the subgraph, u is the source node set of all nodes pointing to node v, PR(u) represents the PageRank value of the source node u, w(u→v) is the relationship edge weight from node u to node v, and Σw(u→k) is the sum of the weights of all outgoing edges of node u; The calculation of the topological influence score comprises the following sub-steps: initializing the PageRank value of each node to 1 / N; iteratively calculating the PageRank value of each node until the PageRank value of all nodes changes less than a preset threshold; and normalizing the final PageRank value of all nodes to obtain the topological influence score of each node.

5. The medical institution dynamic risk early warning method according to claim 4, characterized in that: The method for calculating the similarity of the simulated parameter curve and the parameter of the corresponding device in the real world comprises: Aligning the simulation parameter curve and the parameter of the corresponding device in the real world on the time axis, and normalizing the type data to make the value range [0, 1]; For each monitored parameter, the dynamic time warping distance between the simulation time series S and the real time series R is calculated respectively; Convert the dynamic time warping distance into a similarity score, and map the distance to the interval (0, 1], and 1 represents complete similarity; Weighted fusion is performed on the similarity scores of multiple types of device parameters to obtain a comprehensive similarity score.

6. An electronic device comprising a memory, a processor, and a computer program stored on the memory, wherein the computer program comprises instructions that, when executed by the processor, cause the electronic device to perform the method of any one of claims 1-5. The processor executes the computer program to implement the method of any one of claims 1 to 5.

7. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method of any one of claims 1 to 7.

Images