Artificial Intelligence-Based Internet of Things Device Fault Diagnosis Method and System

Abstract

The present invention relates to the field of equipment fault diagnosis, specifically to an Internet of Things device fault diagnosis method and system based on artificial intelligence. This method collects the operating state data and alarm signals of Internet of Things devices and performs feature extraction to obtain the feature parameter set of the alarm signals; obtains historical false alarm signals, extracts the features of the false alarm signals, and generates a false alarm signal feature library; performs feature matching between the feature parameter set of the alarm signals and the false alarm signal feature library, identifies and marks the false alarm signal tags, and extracts the corresponding fault feature parameters from the false alarm signal feature library; determines the alarm signals with false alarm signal tags through the time-delay verification method. If the fault feature parameters of the alarm signals disappear, it is judged as a false alarm signal and removed; if the fault feature parameters of the alarm signals do not disappear, it is determined as a true alarm signal. It can achieve more accurate diagnosis of the true fault signals of the equipment, thereby improving the accuracy and efficiency of fault diagnosis.

Description

Technical Field

[0001] The present invention relates to the field of equipment fault diagnosis, and specifically to an Internet of Things (IoT) device fault diagnosis method and system based on artificial intelligence. Background Art

[0002] As the place in the hospital with the most stringent requirements for patient life support and monitoring, the intensive care unit (ICU) relies on various Internet of Things (IoT) devices to comprehensively monitor the patient's condition. These devices include vital sign monitors (such as electrocardiogram and blood oxygen saturation monitors), ventilators, infusion pumps, infusion monitoring systems, etc. The stability of their operating status is directly related to the safety of patients. However, in actual applications, due to the complex environment in the ICU, such as multi-device electromagnetic interference, temperature and humidity changes, and increased network transmission load, IoT devices are prone to abnormal operation.

[0003] Existing fault diagnosis technologies for ICU IoT devices mainly rely on traditional monitoring and maintenance methods, such as alarm systems based on single-point anomaly detection or regular manual inspections. Traditional technologies are difficult to effectively integrate and dynamically analyze multi-source information, unable to quickly judge the authenticity of alarm signals and the operating status of devices, which may lead to delayed diagnosis or false alarm interference. Moreover, due to the large number of ICU IoT devices and the resulting alarm signals, many false alarm signals are often generated, which will greatly interfere with the accurate monitoring of the patient's condition.

[0004] Therefore, there is an urgent need for a fault diagnosis method for ICU IoT devices that can accurately diagnose the true fault signals of devices, thereby improving the accuracy and efficiency of fault diagnosis. Summary of the Invention

[0005] (1) Technical Problems to be Solved

[0006] The purpose of the present invention is to provide an Internet of Things device fault diagnosis method and system based on artificial intelligence to solve the problem that many false alarm signals are often generated in the alarm signals generated by IoT devices, and it is impossible to effectively identify false alarm signals, which will greatly interfere with the safety monitoring of the state of the monitored object.

[0007] (2) Technical Solutions

[0008] To achieve the above purpose, on the one hand, the present invention provides an Internet of Things device fault diagnosis method based on artificial intelligence, and the method includes:

[0009] S1. Real-time collect the operation status data and alarm signals of the Internet of Things devices and record them as the first data. Perform data preprocessing on the first data to obtain the second data. Extract the characteristic parameter set of the alarm signals from the second data. The characteristic parameter set includes frequency change, signal duration, and amplitude outliers. The data preprocessing includes noise filtering, missing value filling, and normalization processing.

[0010] S2. Obtain the historical false alarm signals of the Internet of Things devices and perform feature analysis through a deep learning model. Extract the features of the false alarm signals and generate a false alarm signal feature library. Perform feature matching between the characteristic parameter set of the alarm signals and the false alarm signal feature library and calculate the cosine similarity. When the cosine similarity is greater than the preset similarity threshold, mark the alarm signal with a false alarm signal label and extract the corresponding fault characteristic parameters from the false alarm signal feature library.

[0011] S3. Determine the alarm signals with false alarm signal labels through the time-delay verification method. The time-delay verification method is to continuously collect the operation status data of the Internet of Things devices within the time-delay time threshold and record them as the third data. Perform dynamic trend analysis by combining the fault characteristic parameters with the third data to determine whether the fault characteristic parameters of the alarm signals disappear. If the fault characteristic parameters of the alarm signals disappear, it is determined as a false alarm signal and the alarm signal is removed. If the fault characteristic parameters of the alarm signals do not disappear, it is determined as a real alarm signal, and a fault alarm message is generated and the maintenance process of the Internet of Things devices is triggered.

[0012] Further, the time-delay verification method is to continuously collect the operation status data of the Internet of Things devices within the time-delay time threshold and record them as the third data. The method of performing dynamic trend analysis by combining the fault characteristic parameters with the third data to determine whether the fault characteristic parameters of the alarm signals disappear includes:

[0013] Obtain the historical fault data of the Internet of Things devices. Obtain the fault response time after each fault occurrence according to the historical fault data, and calculate the average fault response time and the standard deviation σ T , and calculate the time-delay time threshold T d ; The time-delay time threshold T d The calculation formula is:

[0014]

[0015] where β is a confidence factor used to determine the sensitivity of the time-delay time threshold.

[0016] Collect the operation status data of the Internet of Things devices at a preset sampling frequency within the time-delay time threshold T d and record it as the time-delay data set Dt , D t = {x1, …, x i , …, x n}, t ∈ [0, T d ; where x i represents the operating state data of the Internet of Things device at the i-th sampling moment; n is the number of sampling times within the time period [0, T d .

[0017] Combine the time-delay data set D t with the fault characteristic parameters of the Internet of Things device to construct a fault characteristic matrix X:

[0018]

[0019] where x i,j is the eigenvalue of the j-th fault at the i-th sampling moment; i ∈ [1, n], j ∈ [1, m]; m represents the number of fault characteristic parameters.

[0020] Calculate the alarm signal continuity index C of each fault characteristic parameter within the time-delay time threshold through a sliding window j ; The calculation formula of the continuity index C j is:

[0021]

[0022] where μ j is the historical average value of each fault characteristic parameter; ∈ j tolerance interval, indicating the allowable error range of the eigenvalue x i,j of the current sampling; I(·) is an indicator function used to judge whether the eigenvalue x i,j of the current sampling is within the tolerance interval.

[0023] Obtain the significance index S of each fault characteristic parameter through the significance index calculation formula j , and the calculation formula of the significance index is:

[0024]

[0025] where max(x i,j ) represents the maximum sampled eigenvalue of the j-th fault characteristic parameter within the time-delay time threshold T d , and min(x i,j ) represents the minimum sampled eigenvalue of the j-th fault characteristic parameter within the time-delay time threshold T d ; σ j is the historical standard deviation of the j-th fault characteristic parameter.

[0026] The continuity index C of the alarm signalj and the significance index S j Obtain a comprehensive evaluation value through a linear weighting algorithm, compare the comprehensive evaluation value with a preset disappearance threshold, and determine whether the fault characteristic parameters of the alarm signal disappear; if the comprehensive evaluation value is less than the preset disappearance threshold, it is determined that the fault characteristic parameters of the alarm signal disappear; if the comprehensive evaluation value is greater than the preset disappearance threshold, it is determined that the fault characteristic parameters of the alarm signal do not disappear.

[0027] Furthermore, the method further includes:

[0028] Obtain the historical alarm signals of the Internet of Things devices, and statistically analyze the false alarm signal probability and the true alarm signal probability of the historical alarm signals through data; combine the fault characteristic parameters of the alarm signals, and use the probability density estimation method to construct a conditional probability distribution.

[0029] Determine the false alarm suppression strategy of the Internet of Things device according to the conditional probability distribution; when the false alarm distribution probability in the conditional probability distribution is greater than a preset suppression threshold, trigger the false alarm suppression strategy and mark the alarm signal with a false alarm label, otherwise, adjust the time delay time threshold.

[0030] Adjust the time delay time threshold through a dynamic correction factor, where the dynamic correction factor is the ratio of the false alarm distribution probability to the true alarm distribution probability.

[0031] Furthermore, the method further includes:

[0032] Real-time collect network signal transmission parameters through a network monitoring tool and record them according to a time series, where the network signal transmission parameters include transmission delay, packet loss rate, and jitter value.

[0033] Obtain the transmission characteristics of the alarm signal under different network conditions by using a weighting algorithm for the network signal transmission parameters, and the transmission characteristics are:

[0034] d t = d0 + α·q + β·p;

[0035] where: d t is the comprehensive transmission delay of the signal in the network; α represents the weight factor of the jitter value q on the transmission delay, β represents the weight factor of the packet loss rate p on the transmission delay, and α and β are obtained through curve fitting analysis of historical data; d0 represents the transmission time required for the signal to be transmitted from the Internet of Things device to the monitoring system.

[0036] Obtain the standard transmission delay d l under the ideal network state, calculate the delay error E t between the comprehensive transmission delay d l and the standard transmission delay d d .

[0037] E d = d t −d l ; When E d is greater than a preset error threshold, it is determined that the current network state affects the transmission of the alarm signal, and the delay time threshold is corrected according to the delay error E d ; When E d is less than the preset error threshold, it is determined that the current network state has no effect on the transmission of the alarm signal.

[0038] Obtain the safety monitoring time threshold of the Internet of Things device, and the corrected delay time threshold is less than the safety monitoring time threshold.

[0039] Furthermore, the method further includes:

[0040] Obtain the historical monitoring data D of the monitoring object of the Internet of Things device l ; Obtain the physiological parameters P of the monitoring object collected in real time s and form a time series:

[0041] P s = {p t |t = 1, 2,..., T};

[0042] p t is the real-time physiological parameter at the acquisition time t; Combine the historical monitoring data D l and the physiological parameter P s to form an analysis data set:

[0043] D f = {D l , P s};

[0044] Construct an abnormal state distribution model of the monitoring object through the historical monitoring data D l to obtain the abnormal state P a (x):

[0045]

[0046] where μ a is the mean of the physiological parameters in the abnormal state in the historical monitoring data; is the variance of the physiological parameters in the abnormal state in the historical monitoring data.

[0047] Calculate the degree of deviation D s between the physiological parameter P a and the abnormal state P p through the degree of deviation calculation method:

[0048]

[0049] Degree of deviation D p Indicates the degree of abnormal state of the monitored object; σ a Is the standard deviation of the physiological parameters in the abnormal state among the historical monitoring data; According to the degree of deviation D p Dynamically adjust the security monitoring time threshold T of the Internet of Things device s :

[0050]

[0051] Among them, T s0 Is the initially set security monitoring time threshold; λ is a correction factor used to control the sensitivity of the adjustment of the security monitoring time threshold to the degree of deviation; Is an exponential correction term that decreases as the degree of deviation D p Increases.

[0052] Based on the same inventive concept, the present invention also provides an Internet of Things device fault diagnosis system based on artificial intelligence. The system includes: a feature parameter set acquisition module, an alarm signal analysis module, and an alarm signal verification and determination module, and the modules are communicatively connected in sequence;

[0053] The feature parameter set acquisition module is used to collect the operation state data and alarm signals of the Internet of Things device in real time and record them as the first data, preprocess the first data to obtain the second data, and extract the feature parameter set of the alarm signal from the second data. The feature parameter set includes frequency change, signal duration, and amplitude anomaly value; The data preprocessing includes noise filtering, missing value filling, and standardization processing.

[0054] The alarm signal analysis module is used to obtain the historical false alarm signals of the Internet of Things device and perform feature analysis through a deep learning model, extract the features of the false alarm signals and generate a false alarm signal feature library; Match the feature parameter set of the alarm signal with the false alarm signal feature library and calculate the cosine similarity. When the cosine similarity is greater than the preset similarity threshold, mark the alarm signal with a false alarm signal label and extract the corresponding fault feature parameters from the false alarm signal feature library.

[0055] An alarm signal verification and determination module is used to determine the alarm signal with a false alarm signal label through a time-delay verification method. The time-delay verification method is to continuously collect the operation status data of the Internet of Things device within the time-delay time threshold and record it as the third data, and perform dynamic trend analysis by combining the fault characteristic parameters with the third data to determine whether the fault characteristic parameters of the alarm signal disappear. If the fault characteristic parameters of the alarm signal disappear, it is determined as a false alarm signal, and the alarm signal is removed; if the fault characteristic parameters of the alarm signal do not disappear, it is determined as a true alarm signal, and a fault alarm message is generated and the maintenance process of the Internet of Things device is triggered.

[0056] Further, the alarm signal verification and determination module includes the following steps:

[0057] Obtain the historical fault data of the Internet of Things device, obtain the fault response time after each fault according to the historical fault data, and calculate the average fault response time and the standard deviation σ T , and calculate the time-delay time threshold T d ; The time-delay time threshold T d The calculation formula is:

[0058]

[0059] where β is a confidence factor used to determine the sensitivity of the time-delay time threshold.

[0060] Within the time-delay time threshold T d , collect the operation status data of the Internet of Things device at a preset sampling frequency and record it as the time-delay data set D t , D t ={x1,…,x i ,…,x n}, t∈[0,T d ; where, x i represents the operation status data of the Internet of Things device at the i-th sampling moment; n is the number of sampling times within the time period [0,T d .

[0061] Combine the time-delay data set D t with the fault characteristic parameters of the Internet of Things device to construct a fault characteristic matrix X:

[0062]

[0063] where, x i,j is the characteristic value of the j-th fault at the i-th sampling moment; i∈[1,n], j∈[1,m]; m represents the number of fault characteristic parameters.

[0064] Calculate the alarm signal continuity index C of each fault feature parameter within the time delay threshold through a sliding window j ; The continuity index C j The calculation formula is:

[0065]

[0066] where μ j is the historical average value of each fault feature parameter; ∈ j is the tolerance interval, representing the allowable error range of the current sampled eigenvalue x i,j ; I(·) is an indicator function used to determine whether the current sampled eigenvalue x i,j is within the tolerance interval;

[0067] Obtain the significance index S of each fault feature parameter through the significance index calculation formula j , and the significance index calculation formula is:

[0068]

[0069] where max(x i,j ) represents the maximum sampled eigenvalue of the jth fault feature parameter within the time delay threshold T d , and min(x i,j ) represents the minimum sampled eigenvalue of the jth fault feature parameter within the time delay threshold T d ; ο j is the historical standard deviation of the jth fault feature parameter.

[0070] Obtain the comprehensive evaluation value by linearly weighting the continuity index C j and the significance index S j of the alarm signal, and compare the comprehensive evaluation value with the preset disappearance threshold to determine whether the fault feature parameter of the alarm signal has disappeared; if the comprehensive evaluation value is less than the preset disappearance threshold, it is judged that the fault feature parameter of the alarm signal has disappeared; if the comprehensive evaluation value is greater than the preset disappearance threshold, it is judged that the fault feature parameter of the alarm signal has not disappeared.

[0071] Furthermore, the system further includes the following steps:

[0072] Obtain the historical alarm signals of the Internet of Things devices, and statistically analyze the false alarm signal probability and true alarm signal probability of the historical alarm signals; combine the fault feature parameters of the alarm signals and use the probability density estimation method to construct a conditional probability distribution.

[0073] Determine the false alarm suppression strategy for the Internet of Things device according to the conditional probability distribution; when the false alarm distribution probability in the conditional probability distribution is greater than the preset suppression threshold, trigger the false alarm suppression strategy and mark the alarm signal with a false alarm label, otherwise, adjust the delay time threshold.

[0074] Adjust the delay time threshold through a dynamic correction factor, where the dynamic correction factor is the ratio of the false alarm distribution probability to the true alarm distribution probability.

[0075] Furthermore, the system further includes the following steps:

[0076] Real-time collect network signal transmission parameters through a network monitoring tool and record them according to the time series. The network signal transmission parameters include transmission delay, packet loss rate, and jitter value.

[0077] Obtain the transmission characteristics of the alarm signal under different network conditions for the network signal transmission parameters through a weighted algorithm. The transmission characteristics are:

[0078] d t = d0 + α·q + β·p.

[0079] Where: d t is the comprehensive transmission delay of the signal in the network; α represents the weight factor of the jitter value q on the transmission delay, β represents the weight factor of the packet loss rate p on the transmission delay, and α, β are obtained through historical data fitting analysis; d0 represents the transmission time required for the signal to be transmitted from the Internet of Things device to the monitoring system.

[0080] Obtain the standard transmission delay d l under the ideal network state, calculate the delay error E t between the comprehensive transmission delay d l and the standard transmission delay d d .

[0081] E d = d t ―d l ; when E d is greater than the preset error threshold, it is determined that the current network state affects the transmission of the alarm signal, and the delay time threshold is corrected according to the delay error E d ; when E d is less than the preset error threshold, it is determined that the current network state has no impact on the transmission of the alarm signal.

[0082] Obtain the safety monitoring time threshold of the Internet of Things device, and the corrected delay time threshold is less than the safety monitoring time threshold.

[0083] Furthermore, the system further includes the following steps:

[0084] Obtain the historical monitoring data D of the monitoring object of the Internet of Things device l ; Obtain the physiological parameters P of the monitoring object collected in real time s And form a time series:

[0085] P s ={p t |t = 1, 2, …, T}.

[0086] p t is the real-time physiological parameter at the collection time t; Combine the historical monitoring data D l and the physiological parameter P s to jointly constitute an analysis data set:

[0087] D f ={D l , P s};

[0088] Construct an abnormal state distribution model of the monitoring object through the historical monitoring data D to obtain the abnormal state P l (x): a where μ

[0089]

[0090] is the mean value of the physiological parameters in the abnormal state in the historical monitoring data; a is the variance of the physiological parameters in the abnormal state in the historical monitoring data.

[0091] Obtain the deviation degree D of the physiological parameter P s and the abnormal state P a (x) through the deviation degree calculation method p :

[0092]

[0093] The deviation degree D p represents the abnormal state degree of the monitoring object; σ a is the standard deviation of the physiological parameters in the abnormal state in the historical monitoring data; According to the deviation degree D p dynamically adjust the security monitoring time threshold T of the Internet of Things device s :

[0094]

[0095] where T s0 is the initially set security monitoring time threshold; λ is a correction factor used to control the sensitivity of the deviation degree to the adjustment of the security monitoring time threshold; is a function of the deviation degree D pAn exponentially corrected term that decreases as it increases.

[0096] (3) Beneficial effects

[0097] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0098] 1. By analyzing the characteristics of false alarm signals through a deep learning model and constructing a feature library, false alarm signals are effectively identified and marked through feature matching and cosine similarity calculation. Further, combined with the time delay verification method, dynamic analysis of fault feature parameters is carried out, thereby improving the recognition accuracy of false alarm signals and reducing the resource waste and equipment maintenance costs caused by false alarms.

[0099] 2. By comprehensively considering the degree of deviation, network transmission delay, and device operating status, the safety monitoring time threshold is dynamically adjusted to ensure the reliability of alarm signals under complex network and operating conditions. At the same time, combined with a dynamic correction factor, the time delay time threshold is flexibly adjusted to avoid missed alarms or false alarms caused by a fixed time delay time threshold, thereby improving the real-time performance and accuracy of device fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0100] Figure 1 It is a flowchart of a method for fault diagnosis of Internet of Things devices based on artificial intelligence according to Embodiment 1 of the present invention.

[0101] Figure 2 It is a schematic diagram of the module composition of a system for fault diagnosis of Internet of Things devices based on artificial intelligence according to Embodiment 2 of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0102] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0103] Before giving examples, it is necessary to elaborate on the application scenarios of the inventive concept of the present invention. The present invention is a method and system for fault diagnosis of Internet of Things devices based on artificial intelligence. In the intensive care unit (ICU), Internet of Things devices such as vital sign monitors, ventilators, and infusion pumps are crucial for the health monitoring of patients. However, the operation of the devices in a complex environment may be affected by factors such as sensor failures, network delays, and false alarms, which are likely to lead to inaccurate alarm information or delayed responses, thus threatening the safety of patients.

[0104] Embodiment 1: As Figure 1 shown, this embodiment provides a method for fault diagnosis of Internet of Things devices based on artificial intelligence, and the method includes:

[0105] S1. Real-time collect the operation status data and alarm signals of Internet of Things devices and record them as the first data. Perform data preprocessing on the first data to obtain the second data, and perform feature extraction on the second data to obtain the feature parameter set of the alarm signal. The feature parameter set includes frequency change, signal duration, and amplitude anomaly value. The data preprocessing includes noise filtering, missing value filling, and normalization processing. The Internet of Things devices in the ICU record the real-time operation status data (such as the electrocardiogram signal of the electrocardiograph, the pressure waveform of the ventilator, the infusion rate of the infusion pump, etc.) and generate alarm signals (such as heart rate exceeding the standard, blood oxygen saturation lower than the safe range) in case of abnormalities. These operation status data and alarm signals are the basis for fault diagnosis. The goal of data preprocessing is to improve the data quality and ensure the accuracy and reliability of subsequent analysis. Noise filtering is used to remove random interference in the operation status data, such as signal spikes caused by poor sensor contact or high-frequency noise generated by electromagnetic interference. Missing value filling is due to the possible occurrence of data missing during device operation, such as intermittent data loss caused by network transmission problems. Common methods include mean filling, interpolation method, or prediction filling based on machine learning. Normalization processing is to make the data of different dimensions comparable, and scale the data values to a unified range according to a certain ratio, such as normalizing the values to the interval [0,1]. Perform feature extraction on the second data, and analyze the feature parameters in the alarm signal, including: Frequency change refers to the abnormal change of the vibration frequency in the signal. For example, the irregular change of the heart rate waveform in the electrocardiogram signal may indicate sensor problems or arrhythmia in the patient. Signal duration is that the duration of some alarm signals is abnormally short or long, which can help distinguish false alarms from real alarms. For example, short-term fluctuations in blood oxygen saturation may be false alarms caused by patient movement. Amplitude anomaly value is the out-of-range change of the monitoring signal amplitude. For example, a sudden fluctuation in the infusion rate of the infusion pump exceeding the set value may indicate abnormal operation of the device. The parameter set after feature extraction is called the feature parameter set of the alarm signal and is used for subsequent feature matching and fault diagnosis.

[0106] S2. Obtain the historical false alarm signals of the Internet of Things devices and conduct feature analysis through a deep learning model, extract the features of the false alarm signals and generate a false alarm signal feature library; match the feature parameter set of the alarm signal with the features in the false alarm signal feature library and calculate the cosine similarity. When the cosine similarity is greater than the preset similarity threshold, mark the alarm signal with a false alarm signal label and extract the corresponding fault feature parameters from the false alarm signal feature library. Internet of Things devices (such as electrocardiogram monitors, ventilators) may generate false alarm signals during operation. For example, false alarms may be caused by patient movement, sensor position movement, or environmental interference. By collecting and storing the data of these false alarm signals and conducting feature analysis on the signals through a deep learning model (such as convolutional neural network, recurrent neural network), the features of the false alarm signals can be extracted and a false alarm signal feature library can be generated. The false alarm signal feature library is a structured database that contains the feature parameters of each false alarm signal, such as frequency features, amplitude features, time features, etc. When the device generates a new alarm signal, extract its feature parameter set and match it with the features in the false alarm signal feature library. During the matching process, calculate the cosine similarity between the alarm signal feature parameter set and the features in the library through the cosine similarity formula to quantify the similarity between the two. If the similarity exceeds the preset similarity threshold (such as 0.9), it is determined that the alarm signal is a false alarm signal. When the alarm signal is determined to be a false alarm signal, the system will mark it with a "false alarm signal" label and extract the corresponding fault feature parameters from the feature library (such as abnormal amplitude caused by sensor movement, frequency mutation caused by environmental interference, etc.) to provide a basis for subsequent fault location and device maintenance.

[0107] For example, an infusion pump device emits an alarm signal indicating abnormal liquid flow rate. However, a preliminary investigation finds that the liquid delivery is normal, which may be a false alarm. The system collects historical false alarm signals and trains a deep learning model to establish a feature library, which contains the following features of false alarm signals: False alarm type 1 (environmental vibration): small frequency change, short signal duration, amplitude slightly higher than the normal range. False alarm type 2 (sensor position change): frequency mutation, longer duration, significantly abnormal amplitude. False alarm type 3 (network interference): no obvious frequency change, but irregular signal duration. By comparing the feature parameter set of the alarm signal with the features in the historical false alarm feature library and calculating the cosine similarity to determine whether it is a false alarm signal, it reduces the interference of false alarm signals to medical staff and also provides a detailed fault analysis basis for the operating state of the device.

[0108] S3. Determine the alarm signals with false alarm signal tags through the time-delay verification method. The time-delay verification method is to continuously collect the operation status data of the Internet of Things device within the time-delay time threshold and record it as the third data, and perform dynamic trend analysis by combining the fault feature parameters with the third data to determine whether the fault feature parameters of the alarm signal disappear. If the fault feature parameters of the alarm signal disappear, it is determined as a false alarm signal, and the alarm signal is removed; if the fault feature parameters of the alarm signal do not disappear, it is determined as a real alarm signal, and a fault alarm message is generated and the Internet of Things device maintenance process is triggered. The time-delay verification method is a method that uses a time window, that is, the time-delay time threshold, to dynamically verify the authenticity of the alarm signal. By continuously monitoring the operation status data of the device within a period of time (such as 10 seconds, 30 seconds) after the alarm signal is generated, and combining the alarm signal feature parameters, it is determined whether the alarm signal disappears, so as to distinguish false alarm signals from real alarm signals. Within the time-delay time threshold, collect the operation status data of the device, including the core parameters of the device operation, such as sensor voltage, frequency change, working temperature, etc. Dynamic trend analysis is to combine the fault feature parameters (such as frequency change, amplitude abnormality extracted from the false alarm feature library) with the operation status data to analyze whether these features continue to exist or gradually disappear within the time-delay time. If parameters such as frequency and amplitude gradually return to the normal range (for example, the frequency fluctuation decreases and the amplitude tends to be stable), it indicates that the alarm signal may be caused by transient external factors (such as environmental interference or operation errors) and can be determined as a false alarm. If the fault feature parameters continue to exist or the trend deteriorates within the time-delay time (such as the frequency fluctuation intensifies and the amplitude abnormality expands), it indicates that the alarm signal may be a true reflection of the device failure. If the fault feature parameters disappear, the alarm signal is marked as a false alarm signal and removed to avoid affecting the system performance or misleading medical staff. If the fault feature parameters do not disappear, it is determined as a real alarm signal, the system generates a fault alarm message, prompts the medical staff and triggers the device maintenance process. The time-delay verification effectively avoids false alarms caused by short-term interference, reduces the workload of medical staff, and reduces unnecessary device maintenance or misoperations.

[0109] For example, in the intensive care unit, a ventilator emits an alarm signal indicating that the airway pressure is too high. However, the on-duty personnel's preliminary investigation reveals that there are no obvious abnormalities in the patient's breathing state and airway. At this time, the system conducts a time-delay verification on the alarm signal. The system sets the time-delay time threshold at 30 seconds and continuously collects the operating state data of the ventilator during this period, mainly including: the airway pressure change curve, the airflow rate, and the equipment operating state (such as the compressor operating frequency and the valve switch state). The fault characteristic parameters of the alarm signal are: sudden change in airway pressure frequency: the fluctuation range ≥ 5 Hz, abnormal rate: below the set value by 20%. The dynamic trend analysis shows that the airway pressure fluctuates significantly (8 Hz) within the first 15 seconds but gradually returns to the normal range (fluctuation < 2 Hz) in the next 15 seconds. The airflow rate shows no significant abnormality within 30 seconds and fluctuates within 5% of the normal value. Combining the dynamic trend analysis, it is found that the abnormality of the airway pressure is a temporary fluctuation, which may be caused by the patient's brief cough or environmental interference rather than equipment failure. The fault characteristic parameters disappear within the time-delay time threshold, and the alarm signal is removed by the system to avoid misleading medical staff. If the airway pressure still continues to fluctuate after 30 seconds and the airflow rate is low, the system will generate a fault alarm message to notify the medical staff and trigger the equipment maintenance process, such as checking the airway or replacing the sensor components of the ventilator.

[0110] The time-delay verification method is a method that continuously collects the operating state data of the Internet of Things device within the time-delay time threshold and records it as the third data, and combines the fault characteristic parameters with the third data for dynamic trend analysis to determine whether the fault characteristic parameters of the alarm signal disappear, including:

[0111] Obtain the historical fault data of the Internet of Things device, obtain the fault response time after each fault occurs according to the historical fault data, and calculate the average fault response time and the standard deviation σ T , and calculate the time-delay time threshold T d ; The time-delay time threshold T d The calculation formula is:

[0112]

[0113] where β is the confidence factor, used to determine the sensitivity of the time-delay time threshold, and the value range of β is 1.5 - 2.0;

[0114] Within the time-delay time threshold T d collect the operating state data of the Internet of Things device at a preset sampling frequency and record it as the time-delay data set D t , D t ={x1,…,x i ,…,x n}, t ∈ [0,T d ; where xi represents the operating status data of the Internet of Things device at the i-th sampling moment; n is the number of sampling times within the time period [0, T d ;

[0115] Combine the time delay data set D t to construct a fault feature matrix X in combination with the fault feature parameters of the Internet of Things device:

[0116]

[0117] where, x i,j is the eigenvalue of the j-th fault at the i-th sampling moment; i ∈ [1, n], j ∈ [1, m]; m represents the number of fault feature parameters;

[0118] Calculate the alarm signal continuity index C of each fault feature parameter within the time delay time threshold through a sliding window j ; The continuity index C j The calculation formula is:

[0119]

[0120] where, μ j is the historical average value of each fault feature parameter; ∈ j Tolerance interval, indicating the allowable error range of the eigenvalue x i,j of the current sample; I(·) is an indicator function used to determine whether the eigenvalue x i,j of the current sample is within the tolerance interval;

[0121] Obtain the significance index S of each fault feature parameter through the significance index calculation formula j , and the significance index calculation formula is:

[0122]

[0123] where, max(x i,j ) represents the maximum sampled eigenvalue of the j-th fault feature parameter within the time delay time threshold T d , min(x i,j ) represents the minimum sampled eigenvalue of the j-th fault feature parameter within the time delay time threshold T d ; σ j is the historical standard deviation of the j-th fault feature parameter;

[0124] Combine the continuity index C j of the alarm signal and the significance index S jThe comprehensive evaluation value is obtained through a linear weighting algorithm, and the comprehensive evaluation value is compared with a preset disappearance threshold to determine whether the fault characteristic parameters of the alarm signal disappear; if the comprehensive evaluation value is less than the preset disappearance threshold, it is determined that the fault characteristic parameters of the alarm signal disappear; if the comprehensive evaluation value is greater than the preset disappearance threshold, it is determined that the fault characteristic parameters of the alarm signal do not disappear.

[0125] The continuity index of the alarm signal is used to measure the continuity degree of the fault characteristic parameters within the time delay time threshold. The closer this value is to 1, the more stable the fault characteristic parameters tend to be, with smaller changes and higher continuity; on the contrary, if this value is close to 0, it indicates that the changes of the characteristic parameters are large and the continuity is poor. The sample size of the time delay data set represents the number of samples collected within the time delay time threshold. For example, if 100 data points are collected, then n = 100. For example, in a certain Internet of Things device, j may represent current, voltage, or temperature, and i is each specific sampling moment within the time delay time threshold. μ j is the historical mean of the fault characteristic parameters, which is the average value of the fault characteristics in historical data and is usually used to judge the deviation degree between the current sampled characteristic value and the historical mean. ∈ j The tolerance interval defines an allowable error range, indicating that the current sampled characteristic value x i,j can be regarded as "continuous" within the deviation range from the historical mean μ j . The selection of the tolerance interval is based on the actual situation. For example, the current may fluctuate within a range of 0.1A and still be regarded as continuous. Therefore, ∈ j can take ±0.1A. The I(·) indicator function is used to judge whether the current sampling point is within the set tolerance interval, and its definition is as follows: if |x i,j ―μ j |<∈ k , then I = 1, that is, the current sampled fault characteristic value is within the tolerance interval, indicating that the fault characteristic parameters change stably and have good continuity. If |x i,j ―μ j |≥∈ k, then I = 0, that is, the current sampling point is outside the tolerance interval, indicating that the characteristic parameter changes greatly and the continuity is poor. Judge all sampling moments and sum the samples marked as 1. The summation result represents the proportion of fault eigenvalue that remains within the tolerance interval of the historical mean within the given time delay threshold. Divide the accumulated result (i.e., the number of consecutive samples) by the total number of samples to obtain the continuity index of the alarm signal, which reflects the stability of the fault eigenvalue. A high significance index indicates that the change of the fault feature in the time delay dataset is very drastic, which may indicate that the device is experiencing a relatively serious fault or abnormal state. Therefore, the alarm signal has a high warning value. A low significance index indicates that the change of the fault feature is relatively stable, indicating that the device state is relatively normal and the urgency of the alarm signal is low. According to the continuity index C j and the significance index S j , calculate the comprehensive evaluation value by linear weighting: C 综合 =α·C j +(1 - α)·S j ; C 综合 is the comprehensive evaluation value, which comprehensively considers the continuity and significance of the alarm signal. α is the weighting coefficient, and its value range is between 0 and 1, which is used to adjust the influence of the continuity index C j and the significance index S j on the comprehensive evaluation value. If α is larger, it means that the continuity index has a higher weight in the comprehensive evaluation value. If α is larger, it means that the significance index has a higher weight in the comprehensive evaluation value. The preset disappearance threshold is used to judge whether the fault characteristic parameter of the alarm signal has disappeared. If the comprehensive evaluation value is less than the disappearance threshold, it is considered that the fault characteristic parameter of the alarm signal has disappeared and it may be a false alarm signal; if the comprehensive evaluation value is greater than the disappearance threshold, it is considered that the fault characteristic parameter of the alarm signal has not disappeared and the alarm signal is still valid, indicating a real fault signal.

[0126] The method further includes:

[0127] Obtain the historical alarm signals of the IoT device, and statistically analyze the false alarm signal probability and the real alarm signal probability of the historical alarm signals; combine the fault characteristic parameters of the alarm signal and construct a conditional probability distribution using the probability density estimation method;

[0128] Determine the false alarm suppression strategy of the IoT device according to the conditional probability distribution; when the false alarm distribution probability in the conditional probability distribution is greater than the preset suppression threshold, trigger the false alarm suppression strategy and mark the alarm signal with a false alarm label, otherwise, adjust the time delay threshold;

[0129] Adjust the time delay threshold through a dynamic correction factor, and the dynamic correction factor is the ratio of the false alarm distribution probability to the real alarm distribution probability.

[0130] When the false alarm distribution probability is greater than the true alarm distribution probability, the time delay threshold is extended according to the dynamic correction factor; when the false alarm distribution probability is less than the true alarm distribution probability, the time delay threshold is shortened according to the dynamic correction factor. Combining the fault characteristic parameters of the alarm signal (such as temperature, voltage, current, etc.) and the historical data of false alarm signals / true alarm signals, the conditional probability distribution of the alarm signal is calculated using probability density estimation methods (such as kernel density estimation, histogram estimation, etc.). For example, if the false alarm probability in the conditional probability distribution is greater than the preset suppression threshold, it is determined that the alarm signal has a high probability of being a false alarm, thereby triggering a suppression strategy and marking the alarm signal as a false alarm signal. After triggering the suppression strategy, the system can perform the following operations: Mark the alarm signal as a false alarm. Send a "false alarm signal" prompt to the monitoring personnel. In subsequent monitoring, the system can take additional measures, such as reducing the priority of the alarm signal or reducing the response time of the alarm signal. To adjust the response mechanism of the alarm signal, the system adjusts the time delay threshold according to the ratio of historical false alarm signals and true alarm signals, further optimizing the sensitivity and accuracy of the alarm. For example: If the probability of false alarms is high and the dynamic correction factor is large, the time delay threshold can be appropriately increased to extend the observation time of the alarm signal and avoid misjudging some normal alarm signals prematurely. If the probability of false alarms is low and the dynamic correction factor is small, the time delay threshold can be appropriately reduced to improve the system's fast response ability to fault signals.

[0131] The method includes:

[0132] Real-time collect network signal transmission parameters through a network monitoring tool and record them according to a time series, where the network signal transmission parameters include transmission delay, packet loss rate, and jitter value;

[0133] The network signal transmission parameters are used to obtain the transmission characteristics of the alarm signal under different network conditions through a weighted algorithm, and the transmission characteristics are:

[0134] d t = d0 + α·q + β·p;

[0135] Where: d t is the comprehensive transmission time delay of the signal in the network; α represents the weight factor of the jitter value q on the transmission delay, β represents the weight factor of the packet loss rate p on the transmission delay, and α, β are obtained through fitting analysis of historical data; d0 represents the transmission time required for the signal to be transmitted from the Internet of Things device to the monitoring system;

[0136] Obtain the standard transmission time delay d l under the ideal network state, calculate the time delay error E between the comprehensive transmission time delay d t and the standard transmission time delay d l ​d ;

[0137] E d = d t ―d l ; When E d is greater than the preset error threshold, it is determined that the current network state affects the transmission of the alarm signal, and the delay time threshold is corrected according to the delay error E d ; When E d is less than the preset error threshold, it is determined that the current network state has no influence on the transmission of the alarm signal;

[0138] Obtain the safety monitoring time threshold of the Internet of Things device, and the corrected delay time threshold is less than the safety monitoring time threshold. Use a network monitoring tool (such as Wireshark or a dedicated network monitoring module) to collect the transmission parameters of the Internet of Things alarm signal in the network in real time. The packet loss rate is the proportion of lost data packets during transmission, usually expressed as a percentage. The jitter value is the degree of fluctuation of the network delay, reflecting the transmission stability. The weight factor is obtained by fitting historical data. For example, analyze the influence of the packet loss rate and jitter value on the delay under different network states in the past to obtain the best fitting parameter values. In practice, the jitter value has a greater impact on signals with high real-time requirements. Obtaining the standard transmission delay in the ideal network state is the transmission delay without jitter and packet loss. Correcting the delay time threshold according to the delay error E d adopts a piecewise dynamic adjustment method. For example, the correction factor function f(E d ):

[0139]

[0140] That is, the correction is triggered only when the preset error threshold range is exceeded. k is a correction coefficient used to control the correction amplitude and is determined by analyzing experimental data. Then the corrected delay time threshold is T d + f(E d ), and at the same time, the corrected delay time threshold shall not exceed the safety monitoring time threshold of the Internet of Things device, so as to ensure that the alarm signal will not exceed the response time limit of the device safety monitoring even when the network conditions are poor.

[0141] The method includes:

[0142] Obtain the historical monitoring parameters D of the monitoring object of the Internet of Things device l ; Obtain the physiological parameters P of the monitoring object collected in real time s and form a time series:

[0143] P s = {p t ∣t = 1, 2, …, T};

[0144] p t is the real-time physiological parameter at the acquisition time t; the historical monitoring data D l and the physiological parameter P s jointly constitute an analysis data set:

[0145] D f ={D l ,P s};

[0146] Construct an abnormal state distribution model of the monitored object through the historical monitoring data D l to obtain the abnormal state P a (x):

[0147]

[0148] where μ a is the mean value of the physiological parameters in the abnormal state of the historical monitoring data; is the variance of the physiological parameters in the abnormal state of the historical monitoring data;

[0149] Compare the physiological parameter P s with the abnormal state P a (x) to obtain the deviation degree D p by a deviation degree calculation method:

[0150]

[0151] The deviation degree D p represents the abnormal state degree of the monitored object; σ a is the standard deviation of the physiological parameters in the abnormal state of the historical monitoring data; According to the deviation degree D p dynamically adjust the security monitoring time threshold T s of the Internet of Things device:

[0152]

[0153] where T s0 is the initially set security monitoring time threshold; λ is a correction factor used to control the sensitivity of the deviation degree to the adjustment of the security monitoring time threshold; is a function that changes with the deviation degree D pAn exponentially corrected term that decreases as it increases. The historical monitoring data of the monitored object, which in this embodiment is the historical physiological parameters such as heart rate, blood pressure, etc. By calculating the deviation degree between the real-time physiological parameters and the abnormal state distribution, the abnormal risk of the current state of the monitored object is determined. When the deviation degree is large, it indicates that there is an abnormal risk for the monitored object, and the safety monitoring time threshold is dynamically shortened to ensure the safety of the monitored object. When the deviation degree is small, it indicates that the state of the monitored object is relatively normal, and a relatively long safety monitoring time threshold is allowed.

[0154] Embodiment 2: Based on the same inventive concept, this embodiment also provides an Internet of Things device fault diagnosis system based on artificial intelligence. The system includes: a feature parameter set acquisition module, an alarm signal analysis module, and an alarm signal verification and determination module, which are communicatively connected in sequence;

[0155] The feature parameter set acquisition module is used to collect the operation state data and alarm signals of the Internet of Things device in real time and record them as the first data, preprocess the first data to obtain the second data, and extract the feature parameter set of the alarm signal from the second data. The feature parameter set includes frequency change, signal duration, and amplitude anomaly value; the data preprocessing includes noise filtering, missing value filling, and normalization processing.

[0156] The alarm signal analysis module is used to obtain the historical false alarm signals of the Internet of Things device and perform feature analysis through a deep learning model, extract the features of the false alarm signals and generate a false alarm signal feature library; match the feature parameter set of the alarm signal with the false alarm signal feature library and calculate the cosine similarity. When the cosine similarity is greater than the preset similarity threshold, mark the alarm signal with a false alarm signal label and extract the corresponding fault feature parameters from the false alarm signal feature library.

[0157] The alarm signal verification and determination module is used to determine the alarm signal with a false alarm signal label through the time delay verification method. The time delay verification method is to continuously collect the operation state data of the Internet of Things device within the time delay time threshold and record it as the third data, perform dynamic trend analysis on the fault feature parameters combined with the third data, and judge whether the fault feature parameters of the alarm signal disappear. If the fault feature parameters of the alarm signal disappear, it is judged as a false alarm signal and the alarm signal is removed; if the fault feature parameters of the alarm signal do not disappear, it is determined as a real alarm signal, and a fault alarm message is generated and the Internet of Things device maintenance process is triggered.

[0158] The alarm signal verification and determination module includes the following steps:

[0159] Obtain the historical failure data of the Internet of Things device, obtain the failure response time after each failure according to the historical failure data, and calculate the average failure response time and the standard deviation σ T , and calculate the delay time threshold T d ; The delay time threshold T d The calculation formula is:

[0160]

[0161] where β is the confidence factor, used to determine the sensitivity of the delay time threshold, and the value range of β is 1.5 - 2.0.

[0162] Collect the operating state data of the Internet of Things device at a preset sampling frequency within the delay time threshold T d and record it as the delay data set D t , D t ={x1,…,x i ,…,x n}, t ∈ [0, T d ; where x i represents the operating state data of the Internet of Things device at the i-th sampling moment; n is the number of sampling times within the time period [0, T d .

[0163] Combine the delay data set D t with the failure characteristic parameters of the Internet of Things device to construct a failure characteristic matrix X:

[0164]

[0165] where x i,j is the eigenvalue of the j-th failure at the i-th sampling moment; i ∈ [1, n], j ∈ [1, m]; m represents the number of failure characteristic parameters.

[0166] Calculate the alarm signal continuity index C of each failure characteristic parameter within the delay time threshold through a sliding window j ; The continuity index C j The calculation formula is:

[0167]

[0168] where μ j is the historical average value of each failure characteristic parameter; ∈ j tolerance interval, indicating the allowable error range of the eigenvalue x i,j of the current sampling; I(·) is an indicator function, used to determine whether the eigenvalue x i,j of the current sampling is within the tolerance interval.

[0169] The significance index S of each fault feature parameter is obtained through the significance index calculation formula j , and the significance index calculation formula is as follows:

[0170]

[0171] where max(x i,j ) represents the maximum sampling eigenvalue of the j-th fault feature parameter within the time delay threshold T d , and min(x i,j ) represents the minimum sampling eigenvalue of the j-th fault feature parameter within the time delay threshold T d ; σ j is the historical standard deviation of the j-th fault feature parameter.

[0172] The continuity index C j of the alarm signal and the significance index S j are used to obtain a comprehensive evaluation value through a linear weighting algorithm. The comprehensive evaluation value is compared with a preset disappearance threshold to determine whether the fault feature parameter of the alarm signal has disappeared; if the comprehensive evaluation value is less than the preset disappearance threshold, it is determined that the fault feature parameter of the alarm signal has disappeared; if the comprehensive evaluation value is greater than the preset disappearance threshold, it is determined that the fault feature parameter of the alarm signal has not disappeared.

[0173] The system further includes the following steps:

[0174] Obtain the historical alarm signals of the Internet of Things devices, and statistically analyze the false alarm signal probability and true alarm signal probability of the historical alarm signals; combine the fault feature parameters of the alarm signals and use the probability density estimation method to construct a conditional probability distribution.

[0175] Determine the false alarm suppression strategy of the Internet of Things devices according to the conditional probability distribution; when the false alarm distribution probability in the conditional probability distribution is greater than the preset suppression threshold, trigger the false alarm suppression strategy and mark the alarm signal with a false alarm label, otherwise, adjust the time delay threshold.

[0176] Adjust the time delay threshold through a dynamic correction factor, and the dynamic correction factor is the ratio of the false alarm distribution probability to the true alarm distribution probability.

[0177] The system further includes the following steps:

[0178] Real-time collect network signal transmission parameters through a network monitoring tool and record them according to a time series. The network signal transmission parameters include transmission delay, packet loss rate, and jitter value.

[0179] The transmission characteristics of the alarm signal under different network conditions are obtained by using a weighting algorithm for the network signal transmission parameters, and the transmission characteristics are as follows:

[0180] d t = d0 + α·q + β·p.

[0181] Where: d t is the comprehensive transmission delay of the signal in the network; α represents the weight factor of the jitter value q on the transmission delay, β represents the weight factor of the packet loss rate p on the transmission delay, and α and β are obtained through fitting analysis of historical data; d0 represents the transmission time required for the signal to be transmitted from the Internet of Things device to the monitoring system.

[0182] Obtain the standard transmission delay d l under the ideal network state, calculate the delay error E t between the comprehensive transmission delay d l and the standard transmission delay d d ;

[0183] E d = d t ―d l ; When E d is greater than the preset error threshold, it is determined that the current network state affects the transmission of the alarm signal, and the delay time threshold is corrected according to the delay error E d ; When E d is less than the preset error threshold, it is determined that the current network state has no impact on the transmission of the alarm signal.

[0184] Obtain the safety monitoring time threshold of the Internet of Things device, and the corrected delay time threshold is less than the safety monitoring time threshold.

[0185] The system further includes the following steps:

[0186] Obtain the historical monitoring data D l of the monitoring object of the Internet of Things device; obtain the physiological parameters P s of the monitoring object collected in real time and form a time series:

[0187] P s = {p t ∣t = 1, 2, …, T}.

[0188] p t is the real-time physiological parameter at the collection time t; combine the historical monitoring data D l and the physiological parameters P s to form an analysis data set:

[0189] D f = {D l , P s}}。

[0190] Through the historical monitoring data D l Construct an abnormal state distribution model of the monitoring object to obtain the abnormal state P a (x):

[0191]

[0192] Wherein, μ a Is the mean value of the physiological parameters in the abnormal state in the historical monitoring data; Is the variance of the physiological parameters in the abnormal state in the historical monitoring data.

[0193] Compare the physiological parameter P s With the abnormal state P a (x) to obtain the deviation degree D through the deviation degree calculation method p :

[0194]

[0195] The deviation degree D p Represents the abnormal state degree of the monitoring object; σ a Is the standard deviation of the physiological parameters in the abnormal state in the historical monitoring data; According to the deviation degree D p Dynamically adjust the security monitoring time threshold T of the Internet of Things device s :

[0196]

[0197] Wherein, T s0 Is the initially set security monitoring time threshold; λ is a correction factor used to control the sensitivity of the deviation degree to the adjustment of the security monitoring time threshold; Is an exponential correction term that decreases as the deviation degree D p Increases.

[0198] It should be noted that regarding the system in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated herein.

[0199] Finally, it should be noted that: Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An Internet of Things device fault diagnosis method based on artificial intelligence, characterized in that The method includes: Collecting the operation status data and alarm signals of the Internet of Things devices in real time and recording them as the first data, performing data preprocessing on the first data to obtain the second data, and performing feature extraction on the second data to obtain the feature parameter set of the alarm signals. The feature parameter set includes frequency change, signal duration, and amplitude outliers; the data preprocessing includes noise filtering, missing value filling, and normalization processing. Obtaining the historical false alarm signals of the Internet of Things devices and performing feature analysis through a deep learning model, extracting the features of the false alarm signals and generating a false alarm signal feature library; performing feature matching on the feature parameter set of the alarm signals and the false alarm signal feature library and calculating the cosine similarity. When the cosine similarity is greater than the preset similarity threshold, marking the alarm signal with a false alarm signal label and extracting the corresponding fault feature parameters from the false alarm signal feature library. Determining the alarm signal with the false alarm signal label through the time delay verification method. The time delay verification method is to continuously collect the operation status data of the Internet of Things devices within the time delay time threshold and record them as the third data, performing dynamic trend analysis by combining the fault feature parameters with the third data, and judging whether the fault feature parameters of the alarm signal disappear. If the fault feature parameters of the alarm signal disappear, it is judged as a false alarm signal, and the alarm signal is removed; if the fault feature parameters of the alarm signal do not disappear, it is judged as a real alarm signal, and a fault alarm message is generated and the Internet of Things device maintenance process is triggered. Performing dynamic trend analysis by combining the fault feature parameters with the third data includes: calculating the alarm signal continuity index C of each fault feature parameter within the time delay time threshold through a sliding window j ; obtaining the significance index S of each fault feature parameter within the time delay time threshold through the significance index calculation formula j ; the continuity index C of the alarm signal j and the significance index S j are used to obtain a comprehensive evaluation value through a linear weighted algorithm, and the comprehensive evaluation value is compared with a preset disappearance threshold to determine whether the fault feature parameter of the alarm signal has disappeared; j is the j-th fault feature parameter, j ∈ [1, m], and m represents the number of fault feature parameters.

2. The method for fault diagnosis of Internet of Things devices based on artificial intelligence according to claim 1, wherein The time delay verification method is to continuously collect the operation status data of the Internet of Things devices within the time delay time threshold and record them as the third data. The method of performing dynamic trend analysis by combining the fault feature parameters with the third data and judging whether the fault feature parameters of the alarm signal disappear includes: Obtain the historical failure data of the Internet of Things device, obtain the failure response time after each failure according to the historical failure data, and calculate the average failure response time and the standard deviation σ T , and calculate the delay time threshold T d ; The delay time threshold T d The calculation formula is: where β is a confidence factor used to determine the sensitivity of the time delay time threshold. Within the time delay threshold T d collect the operation status data of the Internet of Things device at a preset sampling frequency and record it as the time delay data set D t , D t ={x1,...,x i ,…,x n}, t ∈ [0, T d ; where xi i represents the operation status data of the Internet of Things device at the i-th sampling moment; n is the number of sampling times within the time period [0, T d ; Combine the time delay data set D t and construct a fault feature matrix X by combining with the fault feature parameters of the Internet of Things devices: where x i,j is the eigenvalue of the j-th fault feature parameter at the i-th sampling moment; i ∈ [1, n], j ∈ [1, m]; m represents the number of fault feature parameters; The continuity index C j The calculation formula is as follows: where, μ j is the historical average value of each fault characteristic parameter; ∈ j is the tolerance interval, representing the allowable error range of the currently sampled eigenvalue x i,j ; I(·) is an indicator function used to determine whether the currently sampled eigenvalue x i,j is within the tolerance interval; The significance index S j The calculation formula is as follows: where, max(x i,j ) represents the maximum sampling eigenvalue of the j-th fault feature parameter within the time delay threshold T d , and min(x i,j ) represents the minimum sampling eigenvalue of the j-th fault feature parameter within the time delay threshold T d ; σ j is the historical standard deviation of the j-th fault feature parameter; If the comprehensive evaluation value is less than the preset disappearance threshold, it is judged that the fault feature parameters of the alarm signal disappear; if the comprehensive evaluation value is greater than the preset disappearance threshold, it is judged that the fault feature parameters of the alarm signal do not disappear.

3. The method for diagnosing faults of Internet of Things devices based on artificial intelligence according to claim 2, wherein The method further includes: Obtaining the historical alarm signals of the Internet of Things devices and statistically analyzing the false alarm signal probability and real alarm signal probability of the historical alarm signals; combining the fault feature parameters of the alarm signals and using the probability density estimation method to construct a conditional probability distribution. Determining the false alarm suppression strategy of the Internet of Things devices according to the conditional probability distribution; when the false alarm distribution probability in the conditional probability distribution is greater than the preset suppression threshold, triggering the false alarm suppression strategy and marking the alarm signal with a false alarm label, otherwise, adjusting the time delay time threshold. Adjusting the time delay time threshold through a dynamic correction factor, where the dynamic correction factor is the ratio of the false alarm distribution probability to the real alarm distribution probability.

4. The method for fault diagnosis of Internet of Things devices based on artificial intelligence according to claim 3, wherein The method further includes: Real-time collecting network signal transmission parameters through a network monitoring tool and recording them in a time series. The network signal transmission parameters include transmission delay, packet loss rate, and jitter value. Obtaining the transmission characteristics of the alarm signal under different network conditions by applying a weighted algorithm to the network signal transmission parameters. The transmission characteristics are: d t = d0 + α·q + β·p; Where: d t is the comprehensive transmission delay of the signal in the network; α represents the weight factor of the jitter value q on the transmission delay, β represents the weight factor of the packet loss rate p on the transmission delay, and α, β are obtained through fitting analysis of historical data; d0 represents the transmission time required for the signal to be transmitted from the Internet of Things device to the monitoring system; Obtain the standard transmission delay d under the ideal network state l , calculate the comprehensive transmission delay d t and the standard transmission delay d l for the delay error E d ; E d = d t - d l ; When E d is greater than a preset error threshold, it is determined that the current network state affects the transmission of the alarm signal, and the time delay time threshold is corrected according to the time delay error E d ; When E d is less than the preset error threshold, it is determined that the current network state has no influence on the transmission of the alarm signal; Obtain the security monitoring time threshold of the Internet of Things device, and the corrected delay time threshold is less than the security monitoring time threshold.

5. The method for diagnosing faults of Internet of Things devices based on artificial intelligence according to claim 4, wherein The method further includes: Obtain the historical monitoring data D of the monitoring object of the Internet of Things device l ; Obtain the physiological parameters P of the monitoring object collected in real time s And form a time series: P s = {p t | t = 1, 2, …, T}; p t is the real-time physiological parameter at the acquisition time t; combining the historical monitoring data D l and the physiological parameter P s to jointly form an analysis data set: D f = {D l , P s}; Through historical monitoring data D l Construct an abnormal state distribution model of the monitored object to obtain the abnormal state P a (x): where, μ a is the mean value of physiological parameters in the abnormal state among historical monitoring data; is the variance of physiological parameters in the abnormal state among historical monitoring data; The physiological parameter P s and the abnormal state P a (x) obtain the deviation degree D through the deviation degree calculation method p : Deviation degree D p represents the degree of abnormal state of the monitored object; σ a is the standard deviation of physiological parameters in the abnormal state of historical monitoring data; according to the deviation degree D p dynamically adjust the security monitoring time threshold T of the Internet of Things device s : Among them, T s0 is the initially set safety monitoring time threshold; λ is a correction factor used to control the sensitivity of the adjustment of the safety monitoring time threshold to the deviation degree; is an exponential correction term that decreases as the deviation degree D p increases.

6. An Internet of Things device fault diagnosis system based on artificial intelligence, characterized in that, The system includes: a feature parameter set acquisition module, an alarm signal analysis module, and an alarm signal verification and determination module, which are communicatively connected in sequence; The feature parameter set acquisition module is used to collect the operation status data and alarm signals of the Internet of Things device in real time and record them as the first data, perform data preprocessing on the first data to obtain the second data, and perform feature extraction on the second data to obtain the feature parameter set of the alarm signal. The feature parameter set includes frequency change, signal duration, and amplitude anomaly value; the data preprocessing includes noise filtering, missing value filling, and normalization processing; The alarm signal analysis module is used to obtain the historical false alarm signals of the Internet of Things device and perform feature analysis through a deep learning model, extract the features of the false alarm signals and generate a false alarm signal feature library; perform feature matching on the feature parameter set of the alarm signal and the false alarm signal feature library and calculate the cosine similarity. When the cosine similarity is greater than the preset similarity threshold, mark the alarm signal with a false alarm signal label and extract the corresponding fault feature parameters from the false alarm signal feature library; The alarm signal verification and determination module is used to determine the alarm signal with a false alarm signal label through a delay verification method. The delay verification method is to continuously collect the operation status data of the Internet of Things device within the delay time threshold and record it as the third data, perform dynamic trend analysis on the fault feature parameters combined with the third data, and determine whether the fault feature parameters of the alarm signal disappear. If the fault feature parameters of the alarm signal disappear, it is determined as a false alarm signal and the alarm signal is removed; if the fault feature parameters of the alarm signal do not disappear, it is determined as a real alarm signal, and a fault alarm message is generated and the Internet of Things device maintenance process is triggered; Combining the fault characteristic parameters with the third data for dynamic trend analysis includes: calculating the alarm signal continuity index C of each fault characteristic parameter within the time delay time threshold through a sliding window j ; obtaining the significance index S of each fault characteristic parameter within the time delay time threshold through the significance index calculation formula j ; the continuity index C of the alarm signal j and the significance index S j are used to obtain a comprehensive evaluation value through a linear weighted algorithm, and the comprehensive evaluation value is compared with a preset disappearance threshold to determine whether the fault characteristic parameter of the alarm signal has disappeared; j is the jth fault characteristic parameter, j ∈ [1, m], and m represents the number of fault characteristic parameters 7. The artificial intelligence-based Internet of Things device fault diagnosis system according to claim 6, characterized in that The alarm signal verification and determination module includes the following steps: Obtain the historical failure data of the Internet of Things device, obtain the failure response time after each failure according to the historical failure data, and calculate the average failure response time and the standard deviation σ T , and calculate the delay time threshold T d ; The delay time threshold T d The calculation formula is: where β is a confidence factor used to determine the sensitivity of the delay time threshold; Within the time delay threshold T d collect the operation status data of the Internet of Things device at a preset sampling frequency and record it as the time delay data set D t , D t ={x1,...,x i ,…,x n}, t∈[0, T d ; where xi i represents the operation status data of the Internet of Things device at the i-th sampling moment; n is the number of sampling times within the time period [0, T d ; Combine the time delay data set D t and construct a fault feature matrix X by combining with the fault feature parameters of the Internet of Things device: where x i,j is the eigenvalue of the j-th fault feature parameter at the i-th sampling moment; i ∈ [1, n], j ∈ [1, m]; m represents the number of fault feature parameters; The continuity index C j The calculation formula is as follows: where, μ j is the historical average value of each fault characteristic parameter; ∈ j is the tolerance interval, representing the allowable error range of the currently sampled eigenvalue x i,j ; I(·) is the indicator function used to determine whether the currently sampled eigenvalue x i,j is within the tolerance interval; The significance index S j The calculation formula is as follows: Among them, max(x i,j ) represents the maximum sampling eigenvalue of the j-th fault feature parameter within the time delay threshold T d , and min(x i,j ) represents the minimum sampling eigenvalue of the j-th fault feature parameter within the time delay threshold T d ; σ j is the historical standard deviation of the j-th fault feature parameter; If the comprehensive evaluation value is less than the preset disappearance threshold, it is determined that the fault feature parameters of the alarm signal disappear; if the comprehensive evaluation value is greater than the preset disappearance threshold, it is determined that the fault feature parameters of the alarm signal do not disappear.

8. The Internet of Things device fault diagnosis system based on artificial intelligence according to claim 7, wherein, The system further includes the following steps: Obtain the historical alarm signals of the Internet of Things device, and statistically analyze the false alarm signal probability and real alarm signal probability of the historical alarm signals; combine the fault feature parameters of the alarm signal and use the probability density estimation method to construct a conditional probability distribution; Determine the false alarm suppression strategy of the Internet of Things device according to the conditional probability distribution; when the false alarm distribution probability in the conditional probability distribution is greater than the preset suppression threshold, trigger the false alarm suppression strategy and mark the alarm signal with a false alarm label, otherwise, adjust the delay time threshold; Adjust the delay time threshold through a dynamic correction factor, and the dynamic correction factor is the ratio of the false alarm distribution probability to the real alarm distribution probability.

9. The artificial intelligence-based Internet of Things device fault diagnosis system according to claim 8, characterized in that, The system further includes the following steps: Real-time collect network signal transmission parameters through a network monitoring tool and record them according to a time series, where the network signal transmission parameters include transmission delay, packet loss rate, and jitter value; Obtain the transmission characteristics of the alarm signal under different network conditions by using a weighted algorithm for the network signal transmission parameters, and the transmission characteristics are: d t = d0 + α·q + β·p; where: d t is the comprehensive transmission delay of the signal in the network; α represents the weight factor of the jitter value q on the transmission delay, β represents the weight factor of the packet loss rate p on the transmission delay, and α, β are obtained through fitting analysis of historical data; d0 represents the transmission time required for the signal to be transmitted from the Internet of Things device to the monitoring system; Obtain the standard transmission delay d under the ideal network state l , calculate the comprehensive transmission delay d t and the standard transmission delay d l for the delay error E d ; E d = d t - d l ; When E d is greater than a preset error threshold, it is determined that the current network state affects the transmission of the alarm signal, and the delay time threshold is corrected according to the delay error E d ; When E d is less than the preset error threshold, it is determined that the current network state has no influence on the transmission of the alarm signal; Obtain the safety monitoring time threshold of the Internet of Things device, and the corrected delay time threshold is less than the safety monitoring time threshold.

10. The Internet of Things device fault diagnosis system based on artificial intelligence according to claim 9, wherein, The system further includes the following steps: Obtain the historical monitoring data D of the monitoring object of the Internet of Things device l ; Obtain the physiological parameters P of the monitoring object collected in real time s And form a time series: P s = {p t | t = 1, 2, …, T}; p t is the real-time physiological parameter at the acquisition time t; combining the historical monitoring data D l and the physiological parameter P s to jointly form an analysis data set: D f ={D l ,P s}; Through historical monitoring data D l Construct an abnormal state distribution model of the monitored object to obtain the abnormal state P a (x): Among them, μ a is the mean value of physiological parameters in the abnormal state of historical monitoring data; is the variance of physiological parameters in the abnormal state of historical monitoring data; The physiological parameter P s and the abnormal state P a (x) obtain the deviation degree D through the deviation degree calculation method p : Deviation degree D p represents the degree of abnormal state of the monitored object; σ a is the standard deviation of physiological parameters in the abnormal state of historical monitoring data; according to the deviation degree D p dynamically adjust the security monitoring time threshold T of the Internet of Things device s : Among them, T s0 is the initially set safety monitoring time threshold; λ is a correction factor used to control the sensitivity of the adjustment of the safety monitoring time threshold to the deviation degree; is an exponential correction term that decreases as the deviation degree D p increases.

Images