Hospital internet of things equipment intelligent monitoring and fault early warning system based on edge computing

Abstract

The application discloses a hospital Internet of Things equipment intelligent monitoring and fault early warning system based on edge computing, relates to the technical field of equipment monitoring and early warning, and aims to solve the technical problem of lagged fault discovery of the existing intelligent monitoring and fault early warning system in a high real-time scene, and comprises a data access and processing module, an Internet of Things base station and a hospital Internet of Things edge computing platform.The data access and processing module is used for collecting various types of data of the hospital Internet of Things equipment and sending the data to a preset storage position, including a local database, a data processing module and an abnormal data judging unit.The Internet of Things base station is used for connecting the Internet of Things equipment, collecting equipment data and realizing intelligent monitoring and fault early warning through an algorithm model.The algorithm model is constructed based on a core algorithm and rules.The hospital Internet of Things edge computing platform is used for performing edge computing management on the data sent by the Internet of Things base station, feeding back analysis results and storing data, and a center cloud.The application has the advantage of improving the fault discovery speed of the Internet of Things equipment.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of equipment monitoring and early warning, in particular to a hospital Internet of Things equipment intelligent monitoring and fault early warning system based on edge computing. BACKGROUND

[0002] In the field of hospital Internet of Things, the stable operation of various medical equipment (such as intensive care monitors, intelligent infusion pumps, disinfection equipment, etc.) is directly related to the quality of medical services and patient safety. At present, the existing hospital Internet of Things equipment monitoring system mainly relies on a cloud centralized data processing mode, and the equipment operation data need to be uploaded to the cloud for analysis and fault judgment, which leads to high data transmission delay and lag of the edge end in responding to equipment abnormalities. In high real-time scenarios such as surgery and intensive care, this lag may lead to equipment failure that cannot be discovered and handled in time, which may cause medical risks and become a key technical problem restricting the efficient and safe operation of hospital Internet of Things equipment. Therefore, we propose a hospital Internet of Things equipment intelligent monitoring and fault early warning system based on edge computing. SUMMARY

[0003] The application aims to provide a hospital Internet of Things equipment intelligent monitoring and fault early warning system based on edge computing to solve the problem of lag in fault discovery of intelligent monitoring and fault early warning systems in high real-time scenarios.

[0004] To solve the above technical problems, the application provides the following technical scheme: a hospital Internet of Things equipment intelligent monitoring and fault early warning system based on edge computing, comprising:

[0005] A data access and processing module is used to collect various data of hospital Internet of Things equipment and send them to a preset storage location, including a local database, a data processing module and an abnormal data judgment unit.

[0006] An Internet of Things base station is used to connect Internet of Things equipment, collect equipment data and realize intelligent monitoring and fault early warning through an algorithm model, and the algorithm model is based on core algorithms and rules.

[0007] A hospital Internet of Things edge computing platform is used to perform edge computing management on the data sent by the Internet of Things base station, analyze the feedback results and store the data.

[0008] A central cloud is used to receive edge computing data and perform cloud computing analysis, model updating and generate analysis reports.

[0009] A fault judgment and processing module is used to detect and handle equipment faults in real time, generate fault alarms and processing records, and includes a fault processing database and a background, and the background includes a system data center.

[0010] A predictive maintenance module is configured to utilize historical data to predict failures, develop preventive maintenance plans, and determine equipment health status.

[0011] The system further comprises core algorithms and rules and a module coordination mechanism, the core algorithms and rules provide algorithm support for data processing and analysis of each module, and the module coordination mechanism is configured to realize coordinated work among the modules.

[0012] Preferably, the data access and processing module further comprises a data acquisition module and a data sending module.

[0013] The data acquisition module is connected through an interface matched with the hospital Internet of Things equipment, and real-time data, statistical data, alarm data and abnormal data are acquired at a set acquisition interval.

[0014] The data sending module encapsulates the acquired data, adds a unique device identifier and an accurate acquisition timestamp, and then synchronously sends the data to a local database and an Internet of Things base station through an internal data transmission bus.

[0015] The data acquisition module preliminarily marks the continuously acquired abnormal data, and the marking logic calculates the abnormal deviation degree of the data through an abnormal deviation degree formula. When the abnormal deviation degree is greater than a set fault judgment threshold, the abnormal data is marked as abnormal candidate data.

[0016] Preferably, the Internet of Things base station is connected with the Internet of Things equipment, and comprises a device driver, a data processing module, a remote control module, a data communication module, a rule engine, a local storage module, an edge computing algorithm, an edge computing rule, an abnormality analysis module and a prediction analysis module.

[0017] The Internet of Things base station sends the generated fault warning information records to an Internet of Things edge computing platform after sorting the records according to emergency levels.

[0018] The data processing module performs secondary verification on the abnormal candidate data, and calculates the fluctuation degree of the data through a sliding window variance formula.

[0019] Preferably, the core algorithms and rules comprise an edge computing algorithm model, rule content and a data processing model, wherein the edge computing algorithm model and the data processing model are deployed in the Internet of Things base station, and training and optimization of the algorithm model are completed in cooperation with a hospital Internet of Things edge computing platform and a central cloud.

[0020] In the edge computing algorithm model, an LSTM model is deployed in the Internet of Things base station, and comprises an input layer, multiple hidden layers and an output layer, and is trained by using an optimizer; a decision tree algorithm is deployed in the Internet of Things base station, and is configured to classify faults, with common fault types of the equipment as classification targets, and multiple equipment operation features related to the faults as classification bases.​

[0021] In the rule content, the fault judgment threshold is set as a fixed proportion of the normal operation parameter range according to the equipment type, and the equipment state evaluation is divided into three levels of health, slight abnormality and serious abnormality;

[0022] In the data processing model, the processing logic is deployed in the Internet of Things base station, the mean value filling method is used to replace the mean value of the surrounding effective data for a small amount of continuous missing values, and the missing values reaching a set number of continuous collection periods or a set proportion of single batch data missing are marked as data interruption;

[0023] Abnormal values are detected and removed by statistical principles, data is stored in a time series database, and is indexed according to equipment identification and time stamp;

[0024] The potential correlation between faults and equipment parameters is mined by an association rule algorithm, and the complex rule iteration of the algorithm is completed by the center cloud;

[0025] The parameter change trend is displayed by a line chart, and the fault rates of different equipment are compared by a column chart, and the visualized data is updated according to a set period;

[0026] The abnormal data judgment unit of the data processing module of the data access and processing module uses the above data processing model to mark the faults in the fault history data that meet the serious abnormality characteristics, and the abnormal data judgment unit includes an equipment data judgment unit, which stores the historical normal operation parameter range of each equipment as a judgment reference.

[0027] Preferably, the hospital Internet of Things edge computing platform is connected with the Internet of Things base station through a stable network connection mode, including a big data cloud computing cluster, an Internet of Things data storage module, an Internet of Things data uploading module and an edge computing management end;

[0028] Through the Internet of Things data download interface, the Internet of Things data from the center cloud is received and stored in the local storage device according to the equipment type, and remote access to the data by different departments is realized through permission management, ensuring the safety and pertinence of data access;

[0029] The edge computing management end realizes the access management and configuration management of the Internet of Things equipment through the Web console, allocates edge computing tasks through a task scheduling algorithm, optimizes the network path through a dynamic routing algorithm, and automatically switches to a backup path when the delay of a certain path is too high;

[0030] The center cloud is deployed in the hospital data center, including a central Internet of Things display and monitoring module, a central Internet of Things management module and a center cloud database;

[0031] The edge computing algorithm model, rules and Internet of Things data from the Internet of Things edge computing platform are received through the cloud data interaction interface, cloud computing and analysis are performed on the Internet of Things data, the algorithm model and rules are updated and optimized, and an analysis report is generated and stored in the central cloud database.

[0032] Preferably, the fault judgment and processing module further comprises a fault processing alarm unit and a fault record statistics unit.

[0033] The fault processing alarm unit sends the device fault alarm information and device fault notification information to the background through a sound-light alarm and an SMS gateway.

[0034] The fault processing alarm unit comprises a device fault alarm unit, a device fault notification unit and an alarm judgment unit.

[0035] The alarm judgment unit compares the current data of the device with the preset threshold in real time, and sends an alarm information when a plurality of continuous collection periods exceed the threshold, the alarm information comprising accurate alarm time, clear alarm reason and specific alarm data.

[0036] The fault record statistics unit is responsible for recording the complete information of a single fault and generating a fault detail file of a single device.

[0037] The alarm judgment unit determines whether to trigger an alarm through a fault confidence formula.

[0038] Preferably, the background further comprises a wireless gateway, an edge computing module and a remote terminal.

[0039] The edge computing module performs filtering and normalization processing on the device data.

[0040] The remote terminal further comprises a data classification unit, a data filtering unit and a data extraction unit.

[0041] When the Internet of Things device fails, the edge computing module obtains device state abnormal data and health data after processing the Internet of Things data, and the edge computing module transmits the device state abnormal data and health data to the data processing module.

[0042] Preferably, the predictive maintenance module comprises a prediction judgment unit, a prediction data storage unit, a prediction database, a cloud prediction module and a wireless cloud prediction module.

[0043] The prediction judgment unit predicts the probability of failure and possible failure causes in the subsequent use cycle of the device based on the accumulated operation data in the operation cycle and the historical fault records stored by the system.

[0044] Preferably, the module coordination mechanism comprises a three-level architecture coordination mechanism and a fault judgment and processing module and a predictive maintenance module coordination mechanism.

[0045] The three-level architecture coordination mechanism is that the Internet of Things base station uploads real-time data to the edge computing platform at a preset period, and the data format adopts a unified format, including the key fields of device identification, collection time, parameter name and value;

[0046] The edge computing platform uploads the summary data to the central cloud at a preset period;

[0047] The central cloud issues the updated model and rule to the edge computing platform at a preset period, and the file integrity is ensured through data verification before issuance;

[0048] The fault judgment and processing module and the prediction maintenance module coordinate mechanism synchronizes the time of all devices in the system by using the network time protocol;

[0049] The prediction maintenance module obtains the recent fault data from the fault judgment and processing module at a preset period, and performs fault prediction in combination with the locally stored device operation data; when obtaining the fault data, the prediction maintenance module filters the effective data according to the time stamp and verifies the data integrity;

[0050] When the central cloud needs to issue a control instruction to the Internet of Things base station, the instruction is forwarded through the edge computing platform, the edge computing platform records the instruction transmission state, and the execution result is fed back to the central cloud.

[0051] Preferably, it further comprises an Internet of Things service terminal;

[0052] The data of the Internet of Things service terminal is derived from the real-time alarm information of the fault judgment and processing module, the preventive maintenance plan of the prediction maintenance module, and the analysis report generated by the central cloud, and the data update frequency is consistent with the data interaction period in the module coordination mechanism;

[0053] The Internet of Things service terminal comprises a device state presentation end, a fault warning end, an alarm management end and a cloud data download end;

[0054] The device state presentation end directly displays the device state through the device health index.

[0055] Compared with the prior art, the beneficial effects of the present application are:

[0056] 1、The present application realizes local real-time processing and abnormality judgment of device data by deploying the edge computing algorithm model in the Internet of Things base station, without relying on cloud centralized calculation, greatly reduces the data transmission delay, can respond to device abnormalities in milliseconds, effectively solves the problem of fault discovery lag in high real-time scenarios, and guarantees the continuous and stable operation of medical devices.

[0057] 2. This invention also relies on the LSTM model in the core algorithm and rules to perform time-series prediction of equipment status parameters, and the decision tree algorithm to accurately classify faults. Combined with the data processing model to handle missing values ​​and outliers, it can significantly improve the accuracy of fault judgment at the edge, avoid misjudgment problems that may be caused by local rapid processing, and reduce the interference of invalid alarms on medical work.

[0058] 3. This invention also realizes dynamic data interaction and model collaborative optimization between IoT base stations, edge computing platforms and central cloud through a module collaboration mechanism. Combined with the predictive maintenance module to generate preventive maintenance plans based on historical data, the traditional passive fault handling mode can be transformed into an active prevention mode, reducing the probability of sudden equipment failure, further reducing the risk of medical service interruption due to equipment downtime, and improving the overall reliability of the hospital's IoT system. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of the system framework of the present invention. Detailed Implementation

[0060] To facilitate understanding of the technical solution of the present invention by those skilled in the art, the technical solution of the present invention will now be further described in conjunction with the accompanying drawings.

[0061] Example 1, such as Figure 1 As shown, this invention provides a hospital IoT device intelligent monitoring and fault early warning system based on edge computing, comprising:

[0062] The data access and processing module is used to collect various types of data from the hospital's IoT devices and send them to a preset storage location, including a local database, a data processing module, and an abnormal data judgment unit;

[0063] IoT base stations are used to connect IoT devices, collect device data, and realize intelligent monitoring and fault early warning through algorithm models. The algorithm models are built based on core algorithms and rules.

[0064] The hospital's IoT edge computing platform is used to manage, analyze, and store data sent by IoT base stations using edge computing.

[0065] The central cloud is used to receive data from edge computing and perform cloud computing analysis, model updates, and generate analysis reports.

[0066] The fault diagnosis and handling module is used to detect and handle equipment faults in real time, generate fault alarms and handling records, including a fault handling database and a backend, the backend of which includes a system data center;

[0067] The predictive maintenance module is used to predict faults using historical data, develop preventive maintenance plans, and determine the health status of equipment.

[0068] The system also includes core algorithms and rules, as well as a module collaboration mechanism. The core algorithms and rules provide algorithmic support for data processing and analysis in each module, while the module collaboration mechanism is used to enable collaborative work between modules.

[0069] In embodiments of the present invention, the data access and processing module further includes a data acquisition module and a data transmission module;

[0070] The data acquisition module establishes a connection with the hospital's IoT devices through an interface that matches the data acquisition data, and acquires real-time data, statistical data, alarm data and abnormal data at set acquisition intervals.

[0071] The data transmission module encapsulates the collected data, adds a unique device identifier and a precise collection timestamp, and then synchronously sends it to the local database and IoT base station via the internal data transmission bus. The data sent to the local database is compressed and stored at fixed intervals to save space, while the data sent to the IoT base station maintains its original precision to ensure the accuracy of the analysis.

[0072] The data acquisition module initially marks the continuously acquired abnormal data. The marking logic calculates the abnormal deviation of the data using the abnormal deviation formula. ,when When the value exceeds the set fault judgment threshold, it is marked as abnormal candidate data:

[0073] ;

[0074] in, Indicates the first The abnormal deviation of each collected data is a dimensionless parameter used to determine whether the data is an abnormal candidate data.

[0075] Indicates the first The collected device data reflects the specific parameter values ​​of the device at a certain moment;

[0076] This represents the historical average value of similar parameters for the equipment, serving as a benchmark value reflecting the long-term average level of the equipment's parameters.

[0077] This represents the historical standard deviation of similar parameters for the device, used to measure the dispersion of historical data.

[0078] This formula is used to measure the first The degree of deviation of each collected data point from the historical average level of similar parameters of the device. This is determined by calculating the... Data collected Compared with historical average The absolute difference, divided by the historical standard deviation. The dimensionless deviation is obtained. ,when When the data exceeds the set fault judgment threshold, it indicates that the data deviates significantly from the normal range. It is marked as abnormal candidate data, providing a preliminary screening basis for subsequent abnormal data verification.

[0079] Historical average and standard deviation Update via sliding window:

[0080] ;

[0081] The historical average is dynamically updated using a sliding window approach. (The last part, "current moment," appears to be an error and doesn't translate directly.) Data collection and the historical average at the previous moment Based on this, by updating the coefficients Perform a weighted calculation to obtain the historical average at the current time. This update method allows the historical average to be continuously adjusted as new data is added, making it more consistent with the real-time changes in equipment parameters;

[0082] ;

[0083] in, express The historical average of the device parameters at any given time is an average that has been dynamically updated.

[0084] This represents a dimensionless update coefficient, with a value between 0 and 1, used to adjust the weight of newly collected data and historical data in the update calculation;

[0085] express The device data collected in real time, and With the same dimensions, it is the latest data that participates in the updating of historical mean and variance;

[0086] express Historical average at time, and Same dimension, used for calculation Historical benchmark of historical average at any given time;

[0087] express Historical variance at time, and The square of a unitary quantity reflects the fluctuation of current equipment parameters;

[0088] express Data collection in real time and The square of the difference between the historical mean and the current time is used to calculate the contribution of the current data to the historical variance.

[0089] express Historical variance at time, and Squares of the same dimension are used for calculation. Historical benchmark of historical variance at any given moment;

[0090] Similar to the historical mean update logic, the historical variance is dynamically updated based on the data collected at the current moment. Compared with the current historical average The squared difference, combined with the historical variance at the previous time step. By updating the coefficients The historical variance at the current time is obtained by weighted calculation. Ensure that historical variance reflects the latest fluctuations in equipment parameters;

[0091] Precise screening of abnormal data: The abnormal deviation formula can objectively identify abnormal candidate data that deviates from the normal range by quantitatively comparing the collected data with the historical mean and standard deviation, avoiding the subjectivity and error of manual judgment and improving the accuracy of the initial screening of abnormal data.

[0092] Dynamically adapting to equipment changes: The sliding window update mechanism of historical mean and variance enables the benchmark value to be dynamically adjusted as the equipment operates over a long period of time, adapting to the slow drift of equipment parameters caused by factors such as aging and environmental changes, and ensuring the timeliness and applicability of the anomaly judgment benchmark.

[0093] Improve system response speed: Through differentiated processing of the data sending module, local database storage compresses data to save space, and IoT base stations receive raw data to ensure analysis accuracy. This balances data storage efficiency and subsequent analysis accuracy, providing a data foundation for the system to respond quickly and handle anomalies.

[0094] Providing reliable input for subsequent modules: The initially labeled abnormal candidate data provides high-quality input for subsequent processing steps such as secondary verification of IoT base stations, reduces the interference of invalid data on subsequent analysis, and improves the operating efficiency of the entire system and the reliability of analysis results.

[0095] In embodiments of the present invention, the IoT base station is connected to various types of IoT devices, including device drivers, data processing, remote control, data communication, rule engines, local storage, edge computing algorithms, edge computing rules, anomaly analysis modules, and predictive analysis modules;

[0096] The device drivers include dedicated drivers for different brands and models of devices, which can be dynamically loaded to quickly adapt to newly connected devices and ensure data acquisition compatibility.

[0097] The data processing module performs format conversion and unit standardization on the received raw data;

[0098] The rules engine stores device-specific fault diagnosis logic, which can be flexibly configured and modified through a visual operation interface to adapt to the characteristics of different devices;

[0099] Local storage uses stable storage media and stores recent raw data and analysis results by device category. When storage space is insufficient, it automatically overwrites the oldest historical data in chronological order.

[0100] The IoT base station records the generated fault warning information, sorts it by urgency, and sends it to the IoT edge computing platform;

[0101] Data communication is achieved through Wi-Fi to connect with the hospital's IoT edge computing platform. It adopts a wireless standard adapted to medical scenarios and is suitable for high-speed, short-range data transmission scenarios of high-precision equipment in operating rooms.

[0102] The data processing module performs secondary verification on the outlier candidate data, calculating the degree of data fluctuation using the sliding window variance formula:

[0103] ;

[0104] in, Indicates the first The variance of the data within a sliding window is used to measure the dispersion and volatility of the data within that window.

[0105] The size of the dimensionless sliding window, i.e. the number of data points contained within the window, determines the range of data involved in the calculation.

[0106] The time index represents the data and is used to identify data collected at different times within the window;

[0107] This represents the index of the sliding window, indicating that the current calculation is for the [index]. One window;

[0108] Indicates the first The mean of the data within each sliding window, and With the same dimensions, it serves as a benchmark value for measuring data fluctuations within a window;

[0109] This formula is used to calculate the degree of fluctuation in data within a specific window, serving as a secondary verification of outlier candidate data. (Based on window size) For the range, take the first... To the For each data point, first calculate the mean of each data point within the window. The differences are squared, and then the average of these squared values ​​is calculated to obtain the window variance. , The larger the value, the more drastic the data fluctuation within that window, combined with the anomaly deviation. This allows for more accurate identification of abnormal data;

[0110] when and (Converted to comparison of the same magnitude through threshold) Data that meets all conditions is identified as anomalous.

[0111] Window mean Calculated using the following formula:

[0112] ;

[0113] This mean serves as the benchmark for calculating data fluctuations, and is related to the dynamic mean. To form complementary verification between short-term and long-term benchmarks;

[0114] This formula is used to calculate the average level of data within a sliding window, serving as a benchmark for measuring data fluctuations within the window. By analyzing the data within the window... To the Sum the data points and then divide by the window size. , obtain the window mean This mean reflects the central tendency of the data in the short term, unlike the dynamic mean which reflects the long-term average level. They work together to evaluate the data from different time dimensions;

[0115] The sliding window variance formula calculates the fluctuation of data within a certain range, thus performing a secondary verification of initially labeled outlier candidates and reducing misjudgments caused by random factors. and Data is only identified as abnormal when all conditions are met, which greatly improves the accuracy of anomaly identification and avoids unnecessary warnings.

[0116] Window mean This reflects the short-term average level of the data, while the dynamic mean reflects the long-term average level. This combination creates a complementary relationship. It allows the system to consider both instantaneous fluctuations and long-term trends in data, enabling it to assess data status from multiple perspectives and improving the comprehensiveness of its assessment of equipment operating conditions.

[0117] The sliding window can be flexibly resized according to device characteristics. For devices with rapidly fluctuating data, a smaller window can be selected to quickly capture changes; for relatively stable devices, a larger window can be selected to smooth out short-term noise. This flexibility allows the system to better adapt to the operating characteristics of different devices, improving the system's adaptability.

[0118] The abnormal data, after being verified twice, provides high-quality input data for subsequent modules such as edge computing and fault diagnosis of IoT base stations. This reduces the interference of invalid data on subsequent analysis, improves the operating efficiency and decision reliability of the entire system, and helps to detect equipment faults in a timely manner and take measures.

[0119] In embodiments of the present invention, the core algorithms and rules include an edge computing algorithm model, rule content, and a data processing model, wherein the edge computing algorithm model and the data processing model are deployed on IoT base stations, and the training and optimization of the algorithm model are completed collaboratively by the hospital's IoT edge computing platform and the central cloud.

[0120] In the edge computing algorithm model, the LSTM model is deployed on the IoT base station and includes an input layer, multiple hidden layers, and an output layer. It is trained using an optimizer. The decision tree algorithm is deployed on the IoT base station for fault classification. It takes common equipment fault types as the classification target and selects multiple fault-related equipment operation features as the classification basis.

[0121] The rules stipulate that the fault judgment threshold is set as a fixed proportion of the normal operating parameter range according to the equipment type, and the equipment status assessment is divided into three levels: healthy, slightly abnormal, and seriously abnormal.

[0122] In the data processing model, the processing logic is deployed on the IoT base station. The mean imputation method is used to replace a small number of consecutive missing values ​​with the mean of the surrounding valid data. Missing values ​​that reach a set number of consecutive collection cycles or a set proportion of missing data in a single batch are marked as data interruptions.

[0123] Outliers are detected and removed using statistical principles, data is stored in a time-series database, and indexes are created by device identifier and timestamp to improve query efficiency.

[0124] The algorithm uses association rule analysis to uncover potential correlations between faults and equipment parameters. The complex rule iterations of this algorithm are completed by the central cloud.

[0125] Line charts are used to display parameter change trends, and bar charts are used to compare the failure rates of different devices. The visualized data is updated according to a set period.

[0126] The abnormal data judgment unit of the data processing module of the data access and processing module adopts the above data processing model to mark the faults that meet the serious abnormal characteristics in the fault history data. The abnormal data judgment unit includes a device data judgment unit, which stores the historical normal operating parameter range of each device as the judgment benchmark.

[0127] The LSTM model's prediction formula for device state parameters is as follows:

[0128] ;

[0129] in, express The predicted values ​​of the equipment status parameters at any given time are of the same dimension as the equipment parameters and are used to predict the future operating status of the equipment.

[0130] The Sigmoid activation function is a dimensionless function, and its formula is: Its function is to map input values ​​to... An interval is used to determine the proportion of control information transmitted in a gating mechanism.

[0131] This represents the weight matrix of the output layer. It is dimensionless and is used to weight the processed cell state values, thus affecting the output of the prediction results.

[0132] The hyperbolic tangent function is dimensionless and its formula is: Input values ​​can be mapped to Intervals are used to process cell states and candidate cell states.

[0133] express The cell state at any given time is dimensionless and is the core of the LSTM model, used to store long-term historical information.

[0134] This represents the bias term of the output layer, and Same dimension, used to adjust the baseline value of the output layer;

[0135] This formula is used to predict the device's performance. The state parameters at any given time. First, Cellular state at any given moment After the hyperbolic tangent function Processing, mapping values ​​to The interval, and then the output layer weight matrix. Multiply, plus the output layer bias Finally, it is activated by the Sigmoid function. Map the results to The interval is obtained Predicted value at time By comparing the predicted value with the actual value, the deviation can be used as the basis for triggering an early warning.

[0136] Cell state The update relies on the combined action of the input gate, forget gate, and output gate:

[0137] ;

[0138] The input gate determines which new information is stored in the cell state. Hidden state of time and Input data at any time After concatenation, it is combined with the input gate weight matrix. Multiply, plus input gate bias Then it goes through the Sigmoid activation function Processing yields the input gate output. Its value is Between these values, the closer the value is to 1, the more important that part of the information is.

[0139] ;

[0140] The forget gate is used to determine which information is discarded from the cell state. Similarly, and After concatenation, it is combined with the forget gate weight matrix. Multiply, plus the forget gate bias After activation by the Sigmoid function Processing yields the output of the forget gate. Value in Between these values, the closer to 1, the more historical information is retained;

[0141] ;

[0142] This formula generates the candidate cell states at the current moment, and After concatenation, the weight matrix of the candidate states is used. Multiply, plus candidate state bias Then through Function processing yields a range within Candidate cell states between ;

[0143] ;

[0144] Cell state updates combine historical and new information. Forget gate output. Cell state at the previous moment Multiplication preserves necessary historical information; input gate output With candidate cell state Multiplying them yields the new information that needs to be added; adding them together gives the cell state at the current moment. ;

[0145] in, express The output of the input gate is dimensionless and its value is in the range of time. Between these points, the proportion of new information entering the cell state is controlled;

[0146] The weight matrix represents the input gate. It is dimensionless and is used to weight the input information, affecting the output of the input gate.

[0147] express The hidden state of a time is dimensionless and contains important information from past time points, which is used for calculations at the current time point.

[0148] express The input data, after normalization, is dimensionless and... They have the same meaning and represent the input information of the model at the current moment;

[0149] This represents the bias term of the input gate, which is dimensionless and used to adjust the calculation basis of the input gate.

[0150] express The output of the forget gate is dimensionless and its value is in Between these, the proportion of historical information forgotten from the cellular state is controlled;

[0151] The weight matrix represents the forget gate. It is dimensionless and is used to weight the input information, affecting the output of the forget gate.

[0152] The bias term of the forget gate is dimensionless and is used to adjust the calculation basis of the forget gate.

[0153] express The candidate cell state at a given time is dimensionless and contains new information for the current time, which is used for cell state updates.

[0154] The weight matrix, which represents the candidate cell state, is dimensionless and is used to weight the input information, thus influencing the generation of candidate cell states.

[0155] A dimensionless bias term representing the candidate cell state, used to adjust the calculation benchmark for the candidate cell state;

[0156] express The cell state at a given moment is dimensionless and stores long-term information from the previous moment, providing the basis for updating the cell state at the current moment.

[0157] LSTM models, through their unique gating mechanism and cell states, can effectively capture the time-series features of device state parameters. The system accurately predicts the real-time status of equipment. This allows the system to anticipate potential equipment anomalies, providing a reliable basis for fault warnings and facilitating timely preventative measures.

[0158] Traditional recurrent neural networks are prone to gradient vanishing or exploding problems when processing long sequence data. In contrast, LSTM models, through cell states and gating mechanisms, can selectively remember and forget information, which effectively solves the problem of long-term dependence. They are suitable for processing large amounts of time-series data generated by the long-term operation of hospital IoT devices, ensuring the accuracy of long-term trend analysis of the devices.

[0159] LSTM models are deployed at IoT base stations, enabling data processing and prediction directly at the edge. This reduces the amount of data transmitted to the cloud, lowers network bandwidth pressure, and speeds up prediction response, allowing edge computing to play a more efficient role in device monitoring.

[0160] The training and optimization of the model are completed in collaboration between the edge computing platform and the central cloud. The model parameters can be continuously adjusted according to new device operation data, so that the model can adapt to the characteristics of different devices and the changes of devices in different operating stages, thereby improving the model's adaptability and generalization ability in complex medical IoT environments.

[0161] By combining decision tree algorithms to classify faults, the prediction results of LSTM models can provide more dimensional feature information for fault classification, which helps to more accurately determine the fault type, provide more targeted guidance for subsequent fault handling and maintenance, and improve the fault handling efficiency of the system.

[0162] In an embodiment of the present invention, the hospital IoT edge computing platform and the IoT base station establish a connection through a stable network connection method, including a big data cloud computing cluster, an IoT data storage module, an IoT data upload module, and an edge computing management terminal;

[0163] The system receives IoT data from the central cloud via an IoT data download interface and stores it on local storage devices according to device type. At the same time, it enables remote access to the data by different departments through access control, ensuring the security and relevance of data access.

[0164] The edge computing management terminal realizes IoT device access management and configuration management through a web console, allocates edge computing tasks through task scheduling algorithms, optimizes network paths with dynamic routing algorithms, and automatically switches to backup paths when a certain path has too high latency.

[0165] Task scheduling algorithms allocate resources by calculating device priority coefficients:

[0166] ;

[0167] in, Indicates the first The priority coefficient of each device is dimensionless and is used to measure the priority of the device in task scheduling. The higher the value, the higher the priority.

[0168] The weighting coefficients representing the importance score of equipment are dimensionless and related to... Together, they determine the weight of the two evaluation indicators in the priority calculation, and ;

[0169] Indicates the first The importance score of each device is dimensionless and ranges from 1 to 10. The higher the score, the more important the device is in the hospital's Internet of Things system.

[0170] The weighting coefficient represents the current load rate of the equipment; it is dimensionless and related to... To facilitate a comprehensive evaluation of equipment priorities;

[0171] Indicates the first The current load rate of the device is dimensionless and ranges from 0 to 1. It reflects the current resource usage of the device. The closer the value is to 1, the heavier the load.

[0172] This formula is used to calculate the first... The priority coefficients of each device are used to determine the order in which edge computing tasks are allocated. This is achieved by scoring the importance of each device. and current load rate of the equipment Multiply by the corresponding weighting coefficients respectively and Then add the two together to get the priority coefficient. ,because This ensures that the weighting of the two indicators in the comprehensive evaluation is reasonable. The higher the value, the higher the priority the device receives in task scheduling, and the more likely it is to obtain computing resources.

[0173] The equipment load rate U_m is calculated using the following formula:

[0174] ;

[0175] in, It represents the number of tasks currently being processed by the device, is dimensionless, and is one of the important indicators for evaluating device load;

[0176] This indicates the maximum number of tasks that the device can handle; it is dimensionless and reflects the upper limit of the device's task processing capacity.

[0177] This represents the weight of the number of tasks in the load rate calculation; it is dimensionless and its value ranges from [value missing]. This is used to adjust the weight of task count and CPU utilization in load assessment;

[0178] It represents the current CPU utilization rate of the device, is dimensionless, and is usually expressed as a percentage, reflecting the usage of the device's computing resources;

[0179] This formula is used to comprehensively evaluate the load situation of the equipment. First, calculate the ratio of the current number of tasks to the maximum number of tasks that can be handled, and the ratio of the current CPU utilization to 100%, then multiply each by the task weight. and Finally, the results are added together to obtain the equipment load rate. This formula takes into account both the number of tasks and CPU utilization, comprehensively reflecting the resource usage status of the device and providing an accurate basis for calculating the priority coefficient.

[0180] By using the device priority coefficient formula, edge computing resources can be rationally allocated based on the importance of the devices and their current load. Devices with high importance and low load are given priority in receiving resources, ensuring that the computing needs of critical equipment (such as high-precision equipment in operating rooms) are met first, thus improving the rationality and effectiveness of resource utilization.

[0181] The device load rate formula comprehensively considers both the number of tasks and CPU utilization, avoiding the one-sidedness of evaluation based on a single indicator. It reflects both the amount of tasks the device is currently processing and the degree of computing resource utilization, making load assessment more accurate, providing a reliable basis for priority calculation, and helping to prevent performance degradation due to excessive load.

[0182] Reasonable task scheduling enables edge computing tasks to run on the most suitable devices, reducing task queuing time and device idle time, improving the operating efficiency of the entire hospital IoT edge computing platform, ensuring that data processing and analysis can be completed quickly, and providing timely support for equipment monitoring and fault early warning;

[0183] Weighting coefficient , and The system can be flexibly adjusted according to actual conditions, allowing task scheduling strategies to adapt to different application scenarios and equipment characteristics. For example, in emergency medical scenarios, the weight of equipment importance can be increased to ensure that critical equipment operates first; when equipment load is generally high, the weight of load rate can be adjusted to balance equipment burden, enhancing the system's adaptability and flexibility.

[0184] The calculation of equipment load rate is linked to the computational requirements of the LSTM model. When the LSTM model has high computational requirements, the system can allocate more suitable equipment resources for it through task scheduling algorithm to ensure the efficient operation of the model. This realizes the collaborative work between modules and improves the stability and reliability of the entire system.

[0185] In an embodiment of the present invention, the central cloud is deployed in the hospital data center and includes a central IoT display and monitoring module, a central IoT management module, and a central cloud database;

[0186] The system receives edge computing algorithm models, rules, and IoT data from the IoT edge computing platform through a cloud data interaction interface. It performs cloud computing and analysis on the IoT data, updates and optimizes the algorithm models and rules, generates analysis reports, and stores them in the central cloud database.

[0187] The optimization of the edge computing model in the central cloud is achieved through iterative loss function:

[0188] ;

[0189] in, The loss function is dimensionless and comprehensively reflects the model's prediction error and parameter complexity. It is the objective function for model optimization, and the smaller the value, the better the model performance.

[0190] This represents the number of samples, which is dimensionless and represents the total number of data samples used in model training.

[0191] This represents the sample index, used to identify different training samples;

[0192] Indicates the first The actual value of each sample, which has the same dimension as the equipment parameters, serves as the target reference value for model prediction.

[0193] Indicates the first The model prediction value for each sample, and The meaning is consistent, that is, the prediction result output by the model;

[0194] This represents the regularization coefficient, which is dimensionless and used to adjust the weight of the regularization term in the loss function. The larger the value, the stronger the penalty on the model parameters.

[0195] Represents a regularization term, which is dimensionless and measures the complexity of the model parameters;

[0196] Represents the set of model parameters, which is dimensionless and includes all parameters in the model that need to be learned and optimized;

[0197] This formula measures the deviation between the predicted and actual values ​​of an edge computing model and is a core indicator for model optimization. First, the actual value for each sample is calculated. Compared with the predicted value The mean squared error term is obtained by averaging the squared differences, which reflects the prediction accuracy of the model; this is further increased by the regularization coefficient. With regularization term The regularization term in the multiplication is used to control the complexity of the model parameters. The loss function is minimized using gradient descent. By continuously adjusting the model parameters, model optimization can be achieved.

[0198] Regularization term Using L2 regularization:

[0199] ;

[0200] in, This represents the parameter index, used to identify different parameters in the model;

[0201] The model represents the first Each parameter is dimensionless and represents the specific parameters that need to be adjusted during the model optimization process.

[0202] This formula is used to calculate the regularization term by applying it to all parameters of the model. The sum of the squares is obtained. Its function is to penalize excessively large model parameters, avoid overfitting the model to the training data, and make the model more concise in parameters while maintaining prediction accuracy, thus improving its adaptability to new data and different devices;

[0203] The loss function takes into account both prediction error and parameter complexity. Minimizing the loss function through gradient descent can improve the prediction accuracy of the model while avoiding excessive parameter complexity, thereby achieving precise optimization of the edge computing model and enhancing the model's ability to analyze and predict data from hospital IoT devices.

[0204] L2 regularization effectively suppresses model overfitting by penalizing excessively large model parameters. In hospital IoT scenarios, where device types are diverse and data varies, regularization prevents the model from overly relying on training data from specific devices, ensuring stable operation across different devices and data scenarios and improving the model's generalization ability.

[0205] The optimized model has higher prediction accuracy and stronger adaptability, and can more accurately analyze equipment operation data and judge equipment status, providing a reliable basis for the analysis report generated in the central cloud, and improving the entire system's intelligent monitoring and fault early warning capabilities for hospital IoT devices;

[0206] The central cloud continuously optimizes the model based on the data received from the edge computing platform, enabling the model to evolve as equipment operation data accumulates, adapt to performance changes and new failure modes during long-term equipment operation, ensure that the model performance is always maintained at a high level, and extend the effective life cycle of the model.

[0207] The model optimization mechanism enables the edge computing model and the central cloud model to keep updated in tandem, ensuring that the processing logic and analysis standards for device data are consistent between the edge and the cloud, avoiding data interpretation deviations caused by model differences, and improving the consistency and reliability of data processing in the entire system.

[0208] In embodiments of the present invention, the fault judgment and processing module further includes a fault processing alarm unit and a fault recording and statistics unit;

[0209] The fault handling alarm unit sends equipment fault alarm information and equipment fault notification information to the backend through the audible and visual alarm and SMS gateway;

[0210] The fault handling alarm unit includes an equipment fault alarm unit, an equipment fault notification unit, and an alarm judgment unit;

[0211] The alarm judgment unit compares the current data of the device with the preset threshold in real time. When the threshold is exceeded for multiple consecutive collection cycles, an alarm message is issued. The alarm message includes the precise alarm time, the clear alarm reason and the specific alarm data.

[0212] The fault recording and statistics unit is responsible for recording complete information about a single fault and generating a fault detail file for each device.

[0213] The alarm judgment unit determines whether an alarm is triggered using a fault confidence formula:

[0214] ;

[0215] in, It represents the confidence level of a fault, is dimensionless, and comprehensively reflects the likelihood of a fault occurring in the equipment. It is the core indicator for determining whether an alarm should be triggered.

[0216] This represents the number of consecutive abnormal cycles, dimensionless, and refers to the number of cycles in which the device data continuously exceeds the preset threshold within the judgment window;

[0217] This represents the total number of cycles in the judgment window, which is dimensionless. It is the total number of cycles contained in the window used to determine whether the device is faulty.

[0218] Represents the maximum abnormal deviation within the window, dimensionless, and... The meaning is consistent; it refers to the maximum value among all abnormal deviations within the judgment window, reflecting the severity of the abnormality.

[0219] This formula is used to quantify the reliability of equipment failure and is a key basis for triggering alarms. First, the number of consecutive abnormal cycles is calculated. Total number of cycles in the judgment window The ratio reflects the frequency of outlier occurrences; then multiplied by the maximum outlier deviation within the window. The degree of anomaly is considered comprehensively. The product of the two factors yields the fault confidence level. , The higher the value, the greater the possibility of equipment failure. When the alarm threshold is exceeded, an alarm will be triggered.

[0220] Alarm thresholds are dynamically adjusted based on device type:

[0221] ;

[0222] in, Indicates the first The alarm threshold of the device is dimensionless and is the critical value that triggers an alarm, based on the fault confidence level. An alarm is triggered when this value is exceeded;

[0223] This represents the adjustment coefficient, which is dimensionless and has a value greater than 1. It is used to adjust the alarm threshold based on the historical average fault confidence level, so that the threshold is more in line with the actual situation of the equipment.

[0224] Indicates the first The historical average failure confidence level of the equipment is dimensionless and represents the average failure confidence level of the equipment in the past, reflecting the historical failure characteristics of the equipment.

[0225] This represents a device index used to identify different devices;

[0226] This formula is used to dynamically determine alarm thresholds for different devices. It is based on the device's historical average fault confidence level. Based on, multiplied by an adjustment factor (Greater than 1) This gives the alarm threshold for the device. This setting allows the threshold to be adapted to the historical fault characteristics of the device, avoiding mismatch in alarm sensitivity for different devices due to a uniform threshold, and improving the accuracy and relevance of alarms.

[0227] The fault confidence formula comprehensively considers the frequency and severity of abnormal data, avoiding false alarms triggered by single, accidental anomalies and improving alarm accuracy. Alarms are only triggered when anomalies are persistent and severe, reducing the interference of invalid alarms on staff.

[0228] The alarm threshold formula dynamically adjusts the threshold based on the device's historical average failure confidence level, ensuring that different types of devices and usage scenarios have suitable alarm standards. For example, for devices with a high failure rate, the threshold will be increased accordingly to avoid frequent alarms; for critical equipment, the threshold can be adjusted... This improves alarm sensitivity and ensures timely fault detection;

[0229] Accurate alarms enable designated personnel in the back-end to quickly obtain effective fault information, including alarm time, cause and data, which facilitates staff to quickly locate the problem and take measures to deal with it, shorten the fault handling time and ensure the normal operation of the hospital's IoT devices;

[0230] A reasonable alarm judgment mechanism and dynamic threshold adjustment enable the fault diagnosis and handling module to detect equipment faults more reliably, reducing missed and false alarms. Meanwhile, the fault record statistics unit generates detailed fault files, providing data support for equipment maintenance and analysis, and contributing to continuous optimization of system performance.

[0231] The maximum anomaly deviation used in fault confidence calculation Deviation from abnormality This interconnected approach enables seamless data processing from acquisition and anomaly labeling to fault diagnosis, demonstrating the synergy between the various modules of the system and ensuring the integrity and effectiveness of the technical solution.

[0232] In embodiments of the present invention, the backend also includes a wireless gateway, an edge computing module, and a remote terminal;

[0233] The edge computing module filters and normalizes device data;

[0234] The remote terminal also includes a data classification unit, a data filtering unit, and a data extraction unit;

[0235] When an IoT device malfunctions, the edge computing module processes the IoT data to obtain abnormal device status data and health data, and then transmits the abnormal device status data and health data to the data processing module.

[0236] The formula for normalizing data in the edge computing module is:

[0237] ;

[0238] in, This represents normalized data, dimensionless, with values ​​in... This eliminates the differences in units and magnitudes between the original data, which is then used for subsequent fault diagnosis calculations.

[0239] The historical minimum value refers to the lowest value that the equipment parameter has ever appeared in history, and is used as one of the benchmarks for normalization.

[0240] Represents the historical maximum value, which refers to the largest value that the equipment parameter has ever appeared in history. It is another benchmark for normalization.

[0241] This formula is used to convert raw equipment data into dimensionless data. It involves calculating the raw data... Compared with historical minimum The difference, divided by the historical maximum value. Compared with historical minimum The difference is used to obtain the normalized data. Its value range is This processing eliminates the influence of differences in units and magnitudes of parameters from different equipment, making the data comparable and facilitating unified calculation and analysis by the subsequent fault diagnosis unit;

[0242] Historical extremes , Optimized through a regular update mechanism:

[0243] ;

[0244] This formula is used to dynamically update the historical minimum value. It takes the historical minimum value from the previous time step. and the current update cycle The smaller of the two minimum values ​​from the internal sampled data is taken as the historical minimum value at the current moment. This ensures that the historical minimum value reflects the latest minimum level of the equipment parameters and adapts to parameter drift during long-term operation of the equipment.

[0245] ;

[0246] in, Indicates the update period, dimensionless, refers to the time interval between updates of historical extreme values, and determines the range of sampled data included in the calculation for each update;

[0247] This represents the sampled data index, used to identify different sampled data within the update period;

[0248] This represents the sampled data within the update period, and... Having the same dimension is the basis for updating historical extreme values;

[0249] express The historical maximum value at a given time, and With the same dimensions, it reflects the maximum level of the current equipment parameters after dynamic updating;

[0250] express The historical maximum value at a given time, and It has the same dimensions, providing a historical benchmark for updating the historical maximum value at the current moment;

[0251] This formula is similar to the historical minimum update logic and is used to dynamically update the historical maximum. It retrieves the historical maximum value from the previous time step. and the current update cycle The larger of the two values ​​of the internally sampled data is taken as the historical maximum value at the current time. This ensures that the historical maximum value reflects the latest maximum level of the equipment parameters, making the normalized benchmark more consistent with the actual operating conditions of the equipment.

[0252] Data normalization formulas convert raw equipment data of different dimensions and magnitudes to... The interval eliminates the differences between data, making parameters that were not originally directly comparable (such as temperature and pressure) comparable, which facilitates the unified analysis and calculation of the fault diagnosis unit and improves the accuracy of the analysis results;

[0253] The periodic update mechanism of historical extreme values ​​enables and It can dynamically adjust as parameters drift slowly during long-term equipment operation, avoiding normalization distortion caused by fixed baseline values. For example, equipment aging may cause overall parameters to be too high; updated historical extreme values ​​can better adapt to this change, ensuring the validity of normalized data.

[0254] The edge computing module normalizes data locally, reducing the amount of raw data transmitted to the cloud and alleviating network transmission pressure. Simultaneously, the normalized data is more suitable for rapid computation and analysis at the edge, improving edge computing efficiency and enabling rapid completion of tasks such as fault diagnosis at the edge.

[0255] The normalized data provides standardized input for the fault judgment unit, ensuring the consistency of calculations such as fault confidence. The historical extreme value update and the dynamic mean update complement each other, dynamically tracking changes in equipment parameters from both extreme value and mean dimensions, comprehensively reflecting the equipment operating status, and enhancing the accuracy of the system's assessment of equipment status.

[0256] A stable normalized baseline and dynamic update mechanism ensure data processing stability, unaffected by short-term fluctuations and long-term drift in equipment parameters. This provides high-quality data support for subsequent fault alarm and predictive maintenance modules, contributing to improved overall system stability and reliability.

[0257] In an embodiment of the present invention, the prediction maintenance module includes a prediction judgment unit, a prediction data storage unit, a prediction database, a cloud prediction module, and a wireless cloud prediction module;

[0258] The prediction and judgment unit predicts the probability of failure and possible causes of failure in the subsequent use cycle of the equipment based on the operating data accumulated during the equipment's operating cycle and the historical fault records stored in the system. The equipment operating cycle and the subsequent use cycle are both preset according to the equipment type and the hospital's operation and maintenance needs.

[0259] The predictive maintenance module sets up a predictive model. The training data for the predictive model comes from historical fault data and equipment operation data accumulated over a long period of time. The model parameters are adjusted by cross-validation. Accuracy, recall, and F1 score are used as evaluation indicators. When the indicators reach the preset standards, the model training is completed and put into use. The model is periodically retrained to adapt to changes in equipment performance.

[0260] The prediction and judgment unit outputs the future failure probability using the failure probability prediction formula:

[0261] ;

[0262] in, It represents the probability of failure within a preset period in the future. It is dimensionless and ranges from 0 to 1. It is used to indicate the likelihood of equipment failure in the future.

[0263] This represents the Sigmoid function, which maps the input composite variables to the 0-1 interval to obtain the failure probability;

[0264] and These represent dimensionless coefficients, used to adjust the historical average fault confidence level. and historical average volatility Weights in failure probability prediction;

[0265] It represents the historical average failure confidence level, is dimensionless, and reflects the average level of the equipment's historical failure confidence level;

[0266] This represents the historical average volatility, which is dimensionless. The normalized average value reflects the average degree of fluctuation in the equipment's historical data;

[0267] This represents the bias term, used to adjust the baseline value of the comprehensive variables so that the failure probability prediction is more in line with the actual situation of the equipment.

[0268] This formula is used to predict the probability of equipment failure within a preset future period. First, the historical average failure confidence level is calculated. and historical average volatility Multiply by the coefficient respectively and In addition to the bias term This yields a composite variable; then, this composite variable is input into the Sigmoid function. Mapping the results to the range of 0-1 yields the failure probability. , The closer the value is to 1, the greater the likelihood of future equipment failure, providing a basis for developing preventative maintenance plans;

[0269] Historical average failure confidence The calculation method is as follows:

[0270] ;

[0271] in, This represents the number of historical fault samples, which is dimensionless and is used to calculate the historical average fault confidence level.

[0272] This represents the historical fault index, used to identify different historical fault records;

[0273] Indicates the first The confidence level of each historical failure is dimensionless and is the basic data for calculating the confidence level of the historical average failure.

[0274] In embodiments of the present invention, the module collaboration mechanism includes a three-level architecture collaboration mechanism and a fault judgment and processing module and a predictive maintenance module collaboration mechanism;

[0275] The three-tier architecture collaboration mechanism enables IoT base stations to upload real-time data to the edge computing platform at preset intervals. The data format adopts a unified format and includes key fields such as device identifier, collection time, parameter name, and value.

[0276] The edge computing platform uploads aggregated data to the central cloud at preset intervals;

[0277] The central cloud distributes updated models and rules to the edge computing platform at preset cycles. Before distribution, data verification is performed to ensure file integrity, and an exception handling mechanism is in place.

[0278] The collaborative mechanism between the fault diagnosis and handling module and the predictive maintenance module is to use the Network Time Protocol to synchronize the time of all devices in the system, ensure the consistency of data time, and perform periodic automatic calibration.

[0279] The predictive maintenance module obtains recent fault data from the fault diagnosis and handling module at a preset cycle, and combines it with locally stored equipment operation data to predict faults. When obtaining fault data, the predictive maintenance module filters valid data based on timestamps and verifies data integrity.

[0280] In the three-tier architecture collaboration, the exception handling mechanism is that when data transmission fails, the sender will retransmit the data after waiting for a set time. The number of retransmissions will not exceed the set number. If it still fails, the transmission log will be recorded and manual intervention will be triggered.

[0281] When an instruction is lost, if the receiver does not receive the instruction within a set time, it will send an instruction loss notification to the sender, and the sender will resend the instruction.

[0282] When the central cloud needs to issue control commands to the IoT base station, the commands are forwarded through the edge computing platform. The edge computing platform records the command transmission status and feeds back the execution results to the central cloud.

[0283] In embodiments of the present invention, an Internet of Things (IoT) service terminal is also included;

[0284] The data of the IoT service terminal comes from real-time alarm information from the fault diagnosis and handling module, preventive maintenance plans from the predictive maintenance module, and analysis reports generated by the central cloud. The data update frequency is consistent with the data interaction cycle in the module collaboration mechanism.

[0285] IoT service terminals include device status display terminals, fault warning terminals, alarm management terminals, and cloud data download terminals;

[0286] The device status display interface visually shows the device status through a device health index:

[0287] ;

[0288] in, For health index, This is the sum of the failure probabilities over multiple future cycles. This is the sum of recent fault confidence levels. The normalization coefficient is... The closer to 1, the healthier the device is.

[0289] The embodiments disclosed in this invention are preferred embodiments, but are not limited thereto. Those skilled in the art can easily understand the spirit of this invention based on the above embodiments and make different extensions and variations, but as long as they do not depart from the spirit of this invention, they are all within the protection scope of this invention.

Claims

1. A hospital IoT device intelligent monitoring and fault early warning system based on edge computing, characterized in that, include: The data access and processing module is used to collect various types of data from the hospital's IoT devices and send them to a preset storage location, including a local database, a data processing module, and an abnormal data judgment unit; An IoT base station is used to connect IoT devices, collect device data, and realize intelligent monitoring and fault early warning through an algorithm model, which is built based on core algorithms and rules. The hospital's IoT edge computing platform is used to manage, analyze, and store data sent by IoT base stations using edge computing. The central cloud is used to receive data from edge computing and perform cloud computing analysis, model updates, and generate analysis reports. The fault diagnosis and handling module is used to detect and handle equipment faults in real time, generate fault alarms and handling records, including a fault handling database and a backend, the backend including a system data center; The predictive maintenance module is used to predict faults using historical data, develop preventive maintenance plans, and determine the health status of equipment. The system also includes core algorithms and rules, as well as a module collaboration mechanism. The core algorithms and rules provide algorithmic support for data processing and analysis of each module, while the module collaboration mechanism is used to enable collaborative work between modules. The core algorithms and rules include edge computing algorithm models, rule content, and data processing models. The edge computing algorithm models and data processing models are deployed on IoT base stations, and the training and optimization of the algorithm models are completed collaboratively by the hospital's IoT edge computing platform and the central cloud. In the edge computing algorithm model, the LSTM model is deployed on the Internet of Things base station and includes an input layer, multiple hidden layers, and an output layer. It is trained using an optimizer. The decision tree algorithm is deployed in IoT base stations for fault classification. It takes common fault types of equipment as the classification target and selects multiple fault-related equipment operating features as the classification basis. In the rules, the fault judgment threshold is set as a fixed proportion of the normal operating parameter range according to the equipment type, and the equipment status assessment is divided into three levels: healthy, slightly abnormal, and seriously abnormal. In the data processing model, the processing logic is deployed on the Internet of Things base station. The mean-filling method is used to replace a small number of consecutive missing values ​​with the mean of the surrounding valid data. Missing values ​​that have reached a set number of consecutive collection cycles or a set proportion of missing data in a single batch are marked as data interruptions. Outliers are detected and removed using statistical principles, and data is stored in a time-series database with indexes created based on device identifiers and timestamps. The algorithm uses association rule analysis to uncover potential correlations between faults and equipment parameters. The complex rule iterations of this algorithm are completed by the central cloud. Line charts are used to display parameter change trends, and bar charts are used to compare the failure rates of different devices. The visualized data is updated according to a set period. The abnormal data judgment unit of the data processing module of the data access and processing module adopts the above-mentioned data processing model to mark faults that meet the serious abnormality characteristics in the fault history data. The abnormal data judgment unit includes a device data judgment unit, which stores the historical normal operating parameter range of each device as the judgment benchmark.

2. The intelligent monitoring and fault early warning system for hospital IoT devices based on edge computing according to claim 1, characterized in that, The data access and processing module also includes a data acquisition module and a data transmission module; The data acquisition module establishes a connection with the hospital's IoT devices through an interface that matches the data acquisition data, and acquires real-time data, statistical data, alarm data and abnormal data at set acquisition intervals. The data transmission module encapsulates the collected data, adds a unique device identifier and a precise collection timestamp, and then synchronously sends it to the local database and IoT base station via the internal data transmission bus. The data acquisition module initially marks the continuously acquired abnormal data, and the marking logic calculates the abnormal deviation of the data using an anomaly deviation formula. ,when When the value exceeds the set fault judgment threshold, it is marked as abnormal candidate data.

3. The intelligent monitoring and fault early warning system for hospital IoT devices based on edge computing according to claim 2, characterized in that, The IoT base station is connected to IoT devices and includes device drivers, data processing, remote control, data communication, rule engine, local storage, edge computing algorithm, edge computing rules, anomaly analysis module, and predictive analysis module. The IoT base station will record the generated fault warning information, sort it by urgency, and send it to the IoT edge computing platform. The data processing module performs secondary verification on the abnormal candidate data and calculates the degree of data fluctuation using the sliding window variance formula.

4. The intelligent monitoring and fault early warning system for hospital IoT devices based on edge computing according to claim 3, characterized in that, The hospital's IoT edge computing platform establishes a connection with IoT base stations through a stable network connection method, including a big data cloud computing cluster, an IoT data storage module, an IoT data upload module, and an edge computing management terminal; The system receives IoT data from the central cloud via an IoT data download interface and stores it on local storage devices according to device type. At the same time, it enables remote access to the data by different departments through access control, ensuring the security and relevance of data access. The edge computing management terminal realizes IoT device access management and configuration management through a web console, allocates edge computing tasks through task scheduling algorithms, optimizes network paths with dynamic routing algorithms, and automatically switches to backup paths when a certain path has too high latency. The central cloud is deployed in the hospital's data center and includes a central IoT display and monitoring module, a central IoT management module, and a central cloud database. The system receives edge computing algorithm models, rules, and IoT data from the IoT edge computing platform through a cloud data interaction interface. It performs cloud computing and analysis on the IoT data, updates and optimizes the algorithm models and rules, generates analysis reports, and stores them in the central cloud database.

5. The intelligent monitoring and fault early warning system for hospital IoT devices based on edge computing according to claim 4, characterized in that, The fault diagnosis and processing module also includes a fault processing alarm unit and a fault record statistics unit; The fault handling alarm unit sends equipment fault alarm information and equipment fault notification information to the backend through the audible and visual alarm and SMS gateway; The fault handling alarm unit includes an equipment fault alarm unit, an equipment fault notification unit, and an alarm judgment unit; The alarm judgment unit compares the current data of the device with the preset threshold in real time. When the threshold is exceeded for multiple consecutive collection cycles, an alarm message is issued. The alarm message includes the precise alarm time, the clear alarm reason and the specific alarm data. The fault recording and statistics unit is responsible for recording complete information about a single fault and generating a fault detail file for each device. The alarm judgment unit determines whether an alarm is triggered by using a fault confidence formula.

6. The intelligent monitoring and fault early warning system for hospital IoT devices based on edge computing according to claim 5, characterized in that, The backend also includes a wireless gateway, an edge computing module, and a remote terminal; The edge computing module filters and normalizes device data; The remote terminal also includes a data classification unit, a data filtering unit, and a data extraction unit; When an IoT device malfunctions, the edge computing module processes the IoT data to obtain abnormal device status data and health data, and then transmits the abnormal device status data and health data to the data processing module.

7. The intelligent monitoring and fault early warning system for hospital IoT devices based on edge computing according to claim 6, characterized in that, The prediction maintenance module includes a prediction judgment unit, a prediction data storage unit, a prediction database, a cloud prediction module, and a wireless cloud prediction module; The prediction and judgment unit predicts the probability of failure and possible causes of failure in subsequent use cycles of the equipment based on the operating data accumulated during the equipment's operating cycle and the historical fault records stored in the system.

8. The intelligent monitoring and fault early warning system for hospital IoT devices based on edge computing according to claim 7, characterized in that, The module collaboration mechanism includes a three-level architecture collaboration mechanism and a collaboration mechanism between the fault judgment and handling module and the predictive maintenance module. The three-level architecture collaboration mechanism is that the IoT base station uploads real-time data to the edge computing platform at a preset period. The data format adopts a unified format and includes key fields such as device identifier, collection time, parameter name and value. The edge computing platform uploads aggregated data to the central cloud at preset intervals; The central cloud distributes updated models and rules to the edge computing platform at a preset cycle, and ensures file integrity through data verification before distribution. The collaborative mechanism between the fault diagnosis and handling module and the predictive maintenance module is to use the Network Time Protocol to synchronize the time of all devices in the system. The predictive maintenance module obtains recent fault data from the fault diagnosis and handling module at a preset cycle, and combines it with locally stored equipment operation data to predict faults. When obtaining fault data, the predictive maintenance module filters valid data based on timestamps and verifies data integrity. When the central cloud needs to issue control commands to the IoT base station, the commands are forwarded through the edge computing platform. The edge computing platform records the command transmission status and feeds back the execution results to the central cloud.

9. A hospital IoT device intelligent monitoring and fault early warning system based on edge computing according to claim 8, characterized in that, It also includes IoT service terminals; The data of the IoT service terminal comes from real-time alarm information from the fault diagnosis and handling module, preventive maintenance plans from the predictive maintenance module, and analysis reports generated by the central cloud. The data update frequency is consistent with the data interaction cycle in the module collaboration mechanism. The IoT service terminal includes a device status display terminal, a fault early warning terminal, an alarm management terminal, and a cloud data download terminal; The device status display interface visually shows the device status through the device health index.

Images