Internet-of-things equipment operation monitoring method and system based on edge computing

By calculating the fluctuation complexity and noise level of IoT devices, correcting the feature distance and applying the Relief algorithm, the problem of poor accuracy in monitoring the operating status of IoT devices is solved, and more accurate feature weight screening and equipment status monitoring are achieved.

CN118981737AInactive Publication Date: 2024-11-19SHANDONG ZHENGTU INFORMATION POLYTRON TECH INC
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411464010.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-21
Publication Date
2024-11-19
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The accuracy of IoT devices' operating status monitoring is poor, especially in the presence of noise samples, which affects the accuracy of feature weight calculation and device failure prediction.

Method used

By obtaining the historical operating data of IoT devices, dividing the data points of normal operating state and abnormal operating state, calculating the fluctuation complexity and noise level of each dimension, using the distance correction coefficient to correct the characteristic distance, generating a distance matrix and applying the Relief algorithm Perform feature weight calculations and filter out key features as monitoring indicators.

Benefits of technology

Effectively reduce the impact of noise data on feature weight calculation, improve the accuracy and reliability of IoT equipment operating status monitoring, and enhance the early prediction ability of equipment abnormalities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118981737A_ABST
    Figure CN118981737A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of data processing, in particular to an internet-of-things equipment operation monitoring method and system based on edge computing, and the method comprises the steps: obtaining historical operation data of internet-of-things equipment, and dividing the historical operation data into a positive example set and a negative example set; calculating the fluctuation complexity of each dimension of data in the historical operation data; calculating the noisy degree of each dimension data in the positive example set and the negative example set according to the fluctuation complexity; calculating a distance correction coefficient according to the noisy degree of each dimension of data in the positive example set and the negative example set, and correcting a feature distance between two data points in the historical operation data through the distance correction coefficient to obtain a feature distance correction value; and generating a distance matrix according to the feature distance correction value, carrying out feature weight calculation by using a Relief algorithm, and screening out a corresponding feature as a monitoring index according to the size of the feature weight so as to monitor the operation state of the Internet of Things equipment. According to the scheme of the invention, the monitoring effect of the operation state of the Internet of Things equipment is effectively improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data processing technology. More specifically, the present invention relates to an edge computing-based IoT device operation monitoring method and system. Background Art

[0002] In the IoT factory online management platform, IoT devices can be used to monitor the status of the production line in the manufacturing workshop, and abnormal conditions of production equipment can be eliminated in a timely manner to improve the stability of production line operation. However, during long-term use, IoT devices may experience abnormal conditions such as equipment failure or excessive load. IoT devices in abnormal conditions may find it difficult to effectively monitor the operating status of the production line. Therefore, it is necessary to monitor IoT devices using data collected by IoT devices through edge nodes.

[0003] At present, when monitoring IoT devices, the Relief (Relevant Features) algorithm can usually be used to determine whether the operation of the IoT device is in a normal operating state or an abnormal operating state through the operating data of the IoT device. For example, the Chinese patent with the authorization announcement number CN116975741B and the name of the energy equipment fault prediction method, device and electronic device based on the Internet of Things discloses an energy equipment fault prediction method based on the Internet of Things, which monitors and collects the current operating state and current operating parameters of the energy equipment in real time through the Internet of Things, and then extracts features from the collected current operating data based on the ReliefF algorithm, and then trains the classifier to obtain the current prediction result corresponding to the current operating state and operating data of the energy equipment.

[0004] However, the Relief algorithm calculates weights based on the feature distance between samples. If there are noise samples in the data set, these noise samples may affect the selection of neighboring features and thus affect the calculation of feature weights, resulting in misleading results.

[0005] Based on this, what needs to be solved urgently is the problem of poor accuracy in monitoring the operating status of IoT devices. Summary of the invention

[0006] In order to solve the technical problem of poor accuracy in monitoring the operating status of the above-mentioned IoT devices, the present invention provides solutions in the following aspects.

[0007] In a first aspect, the present invention provides an edge computing-based method for monitoring the operation of an IoT device, comprising: obtaining historical operation data of the IoT device, and dividing the data points of the normal operation state into a positive example set, and dividing the data points of the abnormal operation state into a negative example set; calculating the fluctuation complexity of each dimension data in the historical operation data, wherein the fluctuation complexity is positively correlated with the standard deviation of the characteristic value of the corresponding dimension data in the historical operation data; calculating the noise content of each dimension data in the positive example set and the negative example set according to the fluctuation complexity, and the noise content is positively correlated with the fluctuation complexity; calculating a distance correction coefficient according to the noise content of each dimension data in the positive example set and the negative example set, and correcting the characteristic distance between two data points in the historical operation data by the distance correction coefficient to obtain a characteristic distance correction value, wherein the distance correction coefficient is positively correlated with the noise content; generating a distance matrix according to the characteristic distance correction value, and calculating feature weights by using the Relief algorithm, and selecting corresponding features as monitoring indicators according to the size of the feature weights, so as to monitor the operation status of the IoT device.

[0008] The present invention calculates the fluctuation complexity of each dimension of data based on the operation data, and calculates the noise level of the positive example set and the noise level of the negative example set based on the data characteristic values ​​of each dimension of all data points in the positive example set and the negative example set. The characteristic distance correction value is calculated according to the noise level of the data of the positive example set and the negative example set in each dimension, and the guessed adjacent data points and the wrong guessed adjacent data points are selected by the characteristic distance correction value, thereby avoiding the influence of noise data on the accurate calculation of feature weights, and screening out features that are more conducive to the operation monitoring of the Internet of Things device through the feature weights to monitor the operation status of the Internet of Things device, which is more convenient and effective to improve the effect of monitoring the operation status of the Internet of Things device.

[0009] Preferably, the data points in the normal operating state are divided into a positive example set, and the data points in the abnormal operating state are divided into a negative example set, including: normalizing the historical operating data and labeling the historical operating data according to the actual operating state of the Internet of Things device; dividing the data points labeled as the normal operating state into a positive example set, and dividing the data points labeled as the abnormal operating state into a negative example set.

[0010] Preferably, the step of correcting the fluctuation complexity is also included, including: determining a first deviation value between the standard deviation of the data eigenvalues ​​of the positive example set and all data points in the historical operation data on the corresponding dimension, and a second deviation value between the standard deviation of the data eigenvalues ​​of the negative example set and all data points in the historical operation data on the corresponding dimension; calculating the average value of the first deviation value and the second deviation value, and using the average value to correct the standard deviation of the eigenvalues ​​of the corresponding dimensional data in the historical operation data to obtain a corrected fluctuation complexity.

[0011] The present invention can effectively reduce the impact of the data fluctuation range on data analysis by calculating the fluctuation complexity of each dimensional data according to the operating data, realize accurate judgment of whether the data point is a noise point, and effectively improve the accuracy of the operating data analysis process.

[0012] Preferably, the calculation formula of fluctuation complexity is: ; in, For the The fluctuation complexity of the dimensional data, is all the data points in the positive example set. The standard deviation of the eigenvalues ​​of the dimensional data, is all the data points in the negative set The standard deviation of the eigenvalues ​​of the dimensional data, is the union of the positive set and the negative set or the number of all data points in the historical running data. The standard deviation of the eigenvalues ​​of the dimensional data, exp() is an exponential function with the natural constant e as the base.

[0013] The present invention corrects the overall fluctuation degree of the data by the difference between the overall data and the corresponding dimensional data in the positive example set and the negative example set, which can effectively improve the accuracy of the calculation of the complexity of data fluctuation and achieve accurate evaluation of the data status.

[0014] Preferably, the calculation formula for the noise level is: ; ; in, The data of the positive example set is The noise level in each dimension is The data of the negative example set is The noise level in each dimension is is the number of data points contained in the positive example set, is the number of data points contained in the negative example set, For the positive example The data in the dimensions The number of data points outside The negative example set The data in the dimensions The number of data points outside is all the data points in the positive example set. The mean of the feature values ​​of the dimension data, is all the data points in the negative set The mean of the feature values ​​of the dimensional data.

[0015] The present invention calculates the noise content of the positive example set and the noise content of the negative example set by using the data feature values ​​of each dimension of all data points in the positive example set and the negative example set, which can effectively reduce the influence of noise data on the feature weights of the Relief algorithm, thereby improving the accuracy and reliability of the IoT device operation data analysis process.

[0016] Preferably, the method for obtaining the characteristic distance correction value includes: calculating the distance correction coefficient based on the noise level of each dimension data in the positive example set and the negative example set, wherein the distance correction coefficient is positively correlated with the noise level and is positively correlated with the difference between the average values ​​of the data point and the set to which it belongs in the corresponding dimension; and correcting the characteristic distance between two data points in the historical operation data by the distance correction coefficient to obtain the characteristic distance correction value.

[0017] In order to further reduce the impact of noise data on the reliability of operating data, the present invention calculates the correction value of the feature distance according to the noise content of the data in the positive example set in each dimension, the noise content of the data in the negative example set in each dimension and the value of the data point in each dimension, thereby realizing accurate calculation of the feature distance and effectively improving the reliability of the operating data monitoring process.

[0018] Preferably, the characteristic distance correction value is calculated as follows: ; in, is the correction value of the characteristic distance between data points, and The two data points are The eigenvalues ​​in dimensions, and They are The corresponding data point belongs to the set and The corresponding data point belongs to the set The noise level in each dimension is for The data point in the set to which the corresponding data point belongs is The mean value in each dimension, for The data point in the set to which the corresponding data point belongs is The mean value in each dimension, is the number of features contained in the training set data.

[0019] Preferably, a distance matrix is ​​generated according to the feature distance correction value, and the feature weight is calculated using the Relief algorithm, and the corresponding features are selected as monitoring indicators according to the size of the feature weight, including: generating a distance matrix according to the feature distance correction value, and randomly selecting a sample from the union of the positive example set and the negative example set; according to the Relief algorithm and the distance matrix, determining the sample's guessed neighbors in the positive example set, and determining the sample's wrongly guessed neighbors in the negative example set; calculating the weight of each feature according to the guessed neighbors and the wrongly guessed neighbors; taking the average of the feature weights corresponding to each sample as the final feature weight, and selecting multiple features with the largest weights as monitoring indicators.

[0020] The present invention generates a distance matrix based on the characteristic distance correction value, and determines the appropriate characteristics as monitoring indicators through weight calculation and indicator screening, which can effectively reduce the impact of noise data on the operating status monitoring of the Internet of Things device and realize reliable monitoring of the operating status of the Internet of Things device.

[0021] Preferably, the historical operation data includes one or more of the following data: temperature, humidity, vibration amplitude, current and voltage.

[0022] In a second aspect, the present invention also provides an edge computing-based IoT device operation monitoring system, comprising: a processor; a memory storing computer program instructions, which, when executed by the processor, implements the edge computing-based IoT device operation monitoring method according to one or more of the aforementioned embodiments.

[0023] The beneficial effect of the present invention is that the present invention calculates the fluctuation complexity of each dimension data according to the operation data, and measures the change of each dimension data from the overall data, so as to better calculate the noise level of the positive example set and the noise level of the negative example set. The noise level of the positive example set and the noise level of the negative example set can be used to measure the noise level in the data collected when the IoT device is in different operating states. The distance correction coefficient is calculated according to the noise level of the data of the positive example set in each dimension and the noise level of the data of the negative example set in the corresponding dimension, as well as the value of the data point in each dimension, and the characteristic distance between the data points is corrected by the distance correction coefficient. The corrected characteristic distance is used to select the guessed adjacent data points and the guessed incorrect adjacent data points, which can reduce the influence of noise data on the calculation of feature weights, and the feature weights are used to screen out features that are more conducive to the operation monitoring of the IoT device to monitor the operation status of the IoT device, which is more convenient and effective to improve the effect of monitoring the operation status of the IoT device. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein: Figure 1 is a flow chart schematically illustrating a method for monitoring the operation of an IoT device based on edge computing according to an embodiment of the present invention; Figure 2 It is a schematic diagram schematically showing an IoT device operation monitoring system based on edge computing according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0026] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0027] Figure 1 Schematically shows a flow chart of an IoT device operation monitoring method 100 based on edge computing according to an embodiment of the present invention.

[0028] like Figure 1 As shown, in step S101, the historical operation data is divided into a positive example set and a negative example set. The historical operation data of the IoT device is obtained, and the data points of the normal operation state are divided into the positive example set, and the data points of the abnormal operation state are divided into the negative example set.

[0029] Since different data have different data characteristics, the normal fluctuation range of data in some dimensions is larger, while the fluctuation range of some data is smaller. This phenomenon may affect the judgment of whether a data point is a noise point. Therefore, the present invention calculates the fluctuation complexity of each dimension of data based on the running data.

[0030] In some embodiments, the historical operation data is normalized and labeled according to the actual operation status of the IoT device. The data points labeled as normal operation status are divided into a positive example set, and the data points labeled as abnormal operation status are divided into a negative example set.

[0031] Taking S seconds as the time interval (in this embodiment, S=1 is taken as an example, and the implementer can adjust it according to the actual situation), the historical operation data of the IoT device in the historical operation process is collected through the sensor equipped by the IoT device, including but not limited to data features of dimensions such as temperature, humidity, vibration amplitude, current and voltage, and the data features of each dimension are normalized by a linear normalization method. The collected operation data is labeled according to the actual operation status of the IoT device in the historical operation process, that is, the data of each dimension at a certain moment is taken as a data point, and the data points are labeled. The data points corresponding to the normal operation status are labeled as the normal operation status, and the data points corresponding to the abnormal operation status are labeled as the abnormal operation status. The set consisting of the data points belonging to the normal operation status is used as the positive example set, and the set consisting of the data points belonging to the abnormal operation status is used as the negative example set.

[0032] At step S102, the fluctuation complexity is calculated. The fluctuation complexity of each dimension data in the historical operation data is calculated, wherein the fluctuation complexity is positively correlated with the standard deviation of the characteristic value of the corresponding dimension data in the historical operation data. Furthermore, the fluctuation complexity can be corrected using the deviations of the positive example set and the negative example set from the overall data. Specifically, a first deviation value between the standard deviation of the data characteristic values ​​of the positive example set and all data points in the historical operation data on the corresponding dimension, and a second deviation value between the standard deviation of the data characteristic values ​​of the negative example set and all data points in the historical operation data on the corresponding dimension are determined. Calculate the average value of the first deviation value and the second deviation value, and use the average value to correct the standard deviation of the characteristic value of the corresponding dimension data in the historical operation data to obtain the corrected fluctuation complexity.

[0033] Calculate the standard deviation of each dimension of data in the positive set, calculate the standard deviation of each dimension of data in the negative set, and calculate the standard deviation of each dimension of data in the union of the positive set and the negative set. Calculate the fluctuation complexity of each dimension of data based on the above standard deviations. The calculation method of the fluctuation complexity of each dimension of data is as follows:

[0034] in, For the The fluctuation complexity of the dimensional data, is all the data points in the positive example set. The standard deviation of the eigenvalues ​​of the dimensional data, is all the data points in the negative set The standard deviation of the eigenvalues ​​of the dimensional data, is the union of the positive set and the negative set or the number of all data points in the historical running data. The standard deviation of the eigenvalues ​​of the dimensional data, exp() is an exponential function with the natural constant e as the base.

[0035] Above The data set (in this embodiment, the data set refers to the union of the positive example set and the negative example set) is the whole in the first The standard deviation of the data in each dimension represents the overall fluctuation of the data. The larger the value, the smaller the standard deviation of the data. The data of the first dimension The greater the difference between the means of the data in the first dimension, the The greater the fluctuation complexity of the dimensional data, the smaller the value, indicating that The data of the first dimension The smaller the difference between the means of the data in each dimension, the The smaller the fluctuation complexity of the data in each dimension.

[0036] Since the data set as a whole contains both positive and negative examples, and the characteristics of the data in the positive set are different from those in the negative set, it will be difficult to directly The complexity of the fluctuations represented is relatively large, so right Make a downward correction. For the data set as a whole The standard deviation of the data in the dimension is the sum of all the data points in the positive example set. The difference between the standard deviations of the characteristic values ​​of the data in the dimensions. The larger the value, the smaller the data set as a whole. The larger the standard deviation of the data in the dimension, the more likely it is that the data set as a whole contains data of different states (positive and negative examples). In fact, the smaller the fluctuation complexity of the data should be, the smaller the value is, the more likely it is that the data set as a whole contains data of different states (positive and negative examples). The standard deviation of the data in the dimension and The smaller the gap between them, the closer the actual data fluctuation complexity is to . Similarly, I will not elaborate on this. Then we get and through right Make corrections, The bigger the The smaller the The greater the regulatory effect, the The smaller the fluctuation complexity of the dimensional data, The more hours, The larger the The smaller the regulatory effect, the The fluctuation complexity of the dimensional data is affected by The smaller the impact.

[0037] Since the built-in sensors of the IoT devices may be in different data collection states when they are in different working states, there are different degrees of noise in the positive example set and the negative example set. In order to reduce the impact of noise data on the feature weights of the Relief algorithm, the present invention calculates the noise degree of the positive example set and the noise degree of the negative example set according to the data feature values ​​of each dimension of all data points in the positive example set and the negative example set.

[0038] In step S103, the noise level is calculated. The noise level of each dimension data in the positive example set and the negative example set is calculated according to the fluctuation complexity, and the noise level is positively correlated with the fluctuation complexity.

[0039] In some embodiments, the mean and standard deviation of each dimension of data in the positive example set can be calculated, the mean and standard deviation of each dimension of data in the negative example set can be calculated, and then the mean of each dimension of data in the union of the positive example set and the negative example set can be calculated. The noise level of the positive example set and the noise level of the negative example set can be calculated based on the mean of each dimension of data and the fluctuation complexity of the data. The noise level is calculated as follows:

[0040]

[0041] in, and The data of the positive example set are The noise level in the dimension and the data of the negative example set are in The noise level in each dimension is For the The fluctuation complexity of the dimensional data, is the number of data points contained in the positive example set, is the number of data points contained in the negative example set, For the positive example The data in the dimensions The number of data points outside The negative example set The data in the dimensions The number of data points outside is all the data points in the positive example set. The mean of the feature values ​​of the dimension data, is all the data points in the negative set The mean of the feature values ​​of the dimension data, is all the data points in the positive example set. The standard deviation of the eigenvalues ​​of the dimensional data, is all the data points in the negative set The standard deviation of the eigenvalues ​​of the dimensional data, exp() is an exponential function with the natural constant e as the base.

[0042] In the above formula, is the first data point in the positive example set The values ​​in the dimensions are The ratio of the number of data points other than the positive example to the number of all data points in the positive example set. Usually, the noise in the data only occupies a small part of the overall data. According to statistical theory, the value of normal data should be within a certain multiple standard deviation range around the mean of the data set. Therefore, this value represents the number of data points in the positive example set. The value on the dimension is the same as the value on the positive example set. The proportion of values ​​with a large deviation compared to the overall data on each dimension, The larger the value, the more The more noise or abnormal data there may be in the data of the dimension, the more likely the positive example set is in the The greater the noise level in each dimension, The smaller the value, the more The less noise or abnormal data there may be in the data of the dimension, the more likely the positive example set is in the The smaller the noise in each dimension.

[0043] No. The complexity of fluctuations in the data The larger the value, the The data of each dimension has a large degree of fluctuation as a whole, which makes the boundary between normal fluctuations and noise points of the data more blurred. The positive and negative example sets may contain more noise data. The smaller the time, the more The data of the dimension has a smaller fluctuation degree, and there may be a more obvious boundary between the normal fluctuation of the data and the noise point. It is easier to determine the first The noise level in the dimension is thus right Make corrections, The bigger it is, The smaller the right The greater the degree of correction, the The bigger, The more hours, The larger the The closer it is to 1, the right The smaller the degree of correction, the The closer it is to 1, The size depends on , and then get the positive example set in the first More references to the noise level in each dimension .

[0044] for The principle is similar and will not be elaborated here.

[0045] The implementation of the Relief algorithm requires calculating the characteristic distance between each data point and each data point in the training set (the union of the positive set and the negative set), and selecting the data point with the smallest characteristic distance from the positive set as the guessed adjacent data point (the concept in the Relief algorithm), and selecting the data point with the smallest characteristic distance from the negative set as the wrongly guessed adjacent data point (the concept in the Relief algorithm). However, due to the existence of noise data, the noise data may be selected as the guessed adjacent data point or the wrongly guessed adjacent data point, which may affect the calculation of the feature weight. Therefore, the present invention calculates the correction value of the characteristic distance according to the noise level of the data in the positive set in each dimension, the noise level of the data in the negative set in each dimension, and the value of the data point in each dimension.

[0046] At step S104, a characteristic distance correction value is calculated. A distance correction coefficient is calculated according to the noise level of each dimensional data in the positive example set and the negative example set, and the characteristic distance between two data points in the historical operation data is corrected by the distance correction coefficient to obtain a characteristic distance correction value, wherein the distance correction coefficient is positively correlated with the noise level. In some embodiments, a distance correction coefficient is calculated according to the noise level of each dimensional data in the positive example set and the negative example set, wherein the distance correction coefficient is positively correlated with the noise level, and is positively correlated with the difference between the average values ​​of the data point and the set to which it belongs in the corresponding dimension. The characteristic distance between two data points in the historical operation data is corrected by the distance correction coefficient to obtain a characteristic distance correction value.

[0047] The correction value of the characteristic distance is calculated as follows: ; in, is the correction value of the characteristic distance between data points, and The two data points are The eigenvalues ​​in dimensions, and They are The corresponding data point belongs to the set and The corresponding data point belongs to the set The noise level in each dimension is for The data point in the set to which the corresponding data point belongs is The mean value in each dimension, for The data point in the set to which the corresponding data point belongs is The mean value in each dimension, is the number of features contained in the training set data.

[0048] for The corresponding data point belongs to the set The average value of the dimensions The larger the value, the greater the difference between The corresponding data point is The more the values ​​on the dimension deviate The overall level of the set to which the corresponding data point belongs, The corresponding data point is The more likely the value in a dimension is a noise value, the more it should be Towards Correction, the smaller the value, the The corresponding data point is The closer the values ​​on the dimensions are, The overall level of the set to which the corresponding data point belongs, The corresponding data point is The more likely the value in a dimension is not a noise value, the less it should be corrected. .

[0049] Since the noise data only occupies a small part of the overall data, in order to avoid Overregulation, through right Make corrections, The bigger the description The greater the noise level of the set to which the corresponding data point belongs, The more likely it is noise data, the more you should adjust , The smaller the description The less noisy the set to which the corresponding data point belongs, The smaller the possibility of being noise data, the The smaller it is, the The smaller the adjustment range, the Adjust to .for The principle is similar and will not be elaborated here. Corrected to .

[0050] In step S105, the monitoring index is screened. A distance matrix is ​​generated according to the feature distance correction value, and the feature weight is calculated using the relief algorithm. The corresponding features are screened out as monitoring indicators according to the size of the feature weight to monitor the operating status of the IoT device.

[0051] In some embodiments, a distance matrix can be generated based on the feature distance correction value, and a sample can be randomly selected from the union of the positive example set and the negative example set. According to the Relief algorithm and the distance matrix, the guessed neighbors of the sample are determined in the positive example set, and the wrong guessed neighbors of the sample are determined in the negative example set. The weight of each feature is calculated based on the guessed neighbors and the wrong guessed neighbors. The mean of the feature weights corresponding to each sample is used as the final feature weight, and multiple features with the largest weights are selected as monitoring indicators.

[0052] In practical applications, a distance matrix can be generated based on the correction value of the feature distance according to the existing technology of the Relief algorithm, and a sample can be randomly selected from the training set. The correct guessed neighbors of the sample can be found in the positive set according to the distance matrix, and the wrong guessed neighbors of the sample can be found in the negative set. The weight of each feature is calculated, and the above steps are repeated to select new samples (in this example, there is no limit on the number of selected samples, and the implementer can determine the number of selected samples according to the size of the training set, that is, the number of times the above steps are repeated), and the weight of each feature is calculated. The mean of the feature weight corresponding to each sample is used as the final feature weight, and the K features with the largest weights are selected (in this example, K=3, and the implementer can adjust it according to the actual situation) as monitoring indicators. Monitoring the operating status of IoT devices through this method can reduce the impact of noise data on the operating status monitoring of IoT devices, and can monitor IoT devices more conveniently and easily.

[0053] Figure 2 It is a schematic diagram schematically showing an IoT device operation monitoring system based on edge computing according to an embodiment of the present invention.

[0054] The present invention also provides an IoT device operation monitoring system based on edge computing. Figure 2 As shown, the system includes a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, the method for monitoring the operation of an IoT device based on edge computing is implemented.

[0055] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface, and their configuration and functions are known in the art, so they will not be described in detail here.

[0056] In the present invention, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, apparatus or device. For example, a computer-readable storage medium may be any appropriate magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc., or any other medium that can be used to store the required information and can be accessed by an application, a module, or both. Any such computer storage medium may be part of a device or accessible or connectable to a device. Any application or module described in the present invention may be implemented using computer-readable / executable instructions that may be stored or otherwise maintained by such a computer-readable medium.

[0057] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A method for monitoring the operation of IoT devices based on edge computing, characterized in that: include: Obtain the historical operation data of IoT devices, and divide the data points in normal operation status into positive example sets, and divide the data points in abnormal operation status into negative example sets; Calculate the fluctuation complexity of each dimension data in the historical operation data, where the fluctuation complexity is positively correlated with the standard deviation of the eigenvalue of the corresponding dimension data in the historical operation data; Calculate the noise level of each dimension data in the positive example set and the negative example set according to the fluctuation complexity, and the noise level is positively correlated with the fluctuation complexity; The distance correction coefficient is calculated according to the noise level of each dimension data in the positive example set and the negative example set, and the characteristic distance between two data points in the historical operation data is corrected by the distance correction coefficient to obtain a characteristic distance correction value, wherein the distance correction coefficient is positively correlated with the noise level; A distance matrix is ​​generated according to the feature distance correction value, and the feature weight is calculated using the Relief algorithm. The corresponding features are selected as monitoring indicators based on the size of the feature weight to monitor the operating status of the IoT device.

2. The method for monitoring the operation of IoT devices based on edge computing according to claim 1, characterized in that: The data points in the normal operation state are divided into the positive example set, and the data points in the abnormal operation state are divided into the negative example set, including: Normalize the historical operation data and mark it according to the actual operation status of the IoT devices; The data points marked as normal operation are divided into the positive example set, and the data points marked as abnormal operation are divided into the negative example set.

3. The method for monitoring the operation of IoT devices based on edge computing according to claim 1, characterized in that: It also includes steps to correct for fluctuation complexity, including: Determine a first deviation value between the standard deviation of the data characteristic values ​​of the positive example set and all data points in the historical operation data on the corresponding dimension, and a second deviation value between the standard deviation of the data characteristic values ​​of the negative example set and all data points in the historical operation data on the corresponding dimension; An average value of the first deviation value and the second deviation value is calculated, and the standard deviation of the characteristic value of the corresponding dimensional data in the historical operation data is corrected by using the average value to obtain a corrected fluctuation complexity.

4. The method for monitoring the operation of IoT devices based on edge computing according to claim 3 is characterized in that: The calculation formula of fluctuation complexity is: ; in, For the The fluctuation complexity of the dimensional data, is all the data points in the positive example set. The standard deviation of the eigenvalues ​​of the dimensional data, is all the data points in the negative set The standard deviation of the eigenvalues ​​of the dimensional data, is the union of the positive set and the negative set or the number of all data points in the historical running data. The standard deviation of the eigenvalues ​​of the dimensional data, exp() is an exponential function with the natural constant e as the base.

5. The method for monitoring the operation of IoT devices based on edge computing according to claim 4 is characterized in that: The calculation formula for the noise level is: ; ; in, The data of the positive example set is The noise level in each dimension is The data of the negative example set is The noise level in each dimension is is the number of data points contained in the positive example set, is the number of data points contained in the negative example set, For the positive example The data in the dimensions The number of data points outside The negative example set The data in the dimensions The number of data points outside is all the data points in the positive example set. The mean of the feature values ​​of the dimension data, is all the data points in the negative set The mean of the feature values ​​of the dimensional data.

6. The method for monitoring the operation of IoT devices based on edge computing according to claim 1, characterized in that: Methods for obtaining feature distance correction values ​​include: The distance correction coefficient is calculated according to the noise level of each dimension data in the positive example set and the negative example set, where the distance correction coefficient is positively correlated with the noise level and the difference between the average value of the data point and its set in the corresponding dimension; The characteristic distance between two data points in the historical operation data is corrected by the distance correction coefficient to obtain a characteristic distance correction value.

7. The method for monitoring the operation of IoT devices based on edge computing according to claim 6, characterized in that: The characteristic distance correction value is calculated as follows: ; in, is the correction value of the characteristic distance between data points, and The two data points are The eigenvalues ​​in dimensions, and They are The corresponding data point belongs to the set and The corresponding data point belongs to the set The noise level in each dimension is for The data point in the set to which the corresponding data point belongs is The mean value in each dimension, for The data point in the set to which the corresponding data point belongs is The mean value in each dimension, is the number of features contained in the training set data.

8. The method for monitoring the operation of IoT devices based on edge computing according to claim 1, characterized in that: Generate a distance matrix based on the feature distance correction value, and use the Relief algorithm to calculate the feature weight. Select the corresponding features as monitoring indicators based on the size of the feature weight, including: Generate a distance matrix based on the feature distance correction value, and randomly select a sample from the union of the positive example set and the negative example set; According to the Relief algorithm and the distance matrix, the correct neighbors of the samples are determined in the positive example set, and the incorrect neighbors of the samples are determined in the negative example set; Calculate the weight of each feature based on the guessed nearest neighbors and the wrong guessed nearest neighbors; The mean of the feature weights corresponding to each sample is taken as the final feature weight, and multiple features with the largest weights are selected as monitoring indicators.

9. The method for monitoring the operation of IoT devices based on edge computing according to any one of claims 1 to 8, characterized in that: The historical operation data includes one or more of the following data: Temperature, humidity, vibration amplitude, current and voltage.

10. An IoT device operation monitoring system based on edge computing, characterized in that: include: processor; A memory storing computer program instructions, which, when executed by the processor, implements the method for monitoring the operation of an IoT device based on edge computing according to any one of claims 1 to 9.

Citation Information

Cited By

  • Digital energy air compression station information tracing method

    CN119312260A