Equipment operation state self-checking method and system based on machine learning

Abstract

The invention relates to the technical field of data processing, in particular to an equipment operation state self-checking method and system based on machine learning, and the method comprises the steps: obtaining a current sequence of monitoring equipment, employing the noise level of the current sequence, adaptively adjusting the size of a subsample of an SCIFrest algorithm, and enabling the size of the subsample to be in negative correlation with the noise level; and randomly selecting a plurality of data sets with the same size as the subsamples from the current sequence, constructing a decision tree by using an SCIForest algorithm, and deploying the trained decision tree to the monitoring equipment, so as to perform self-inspection on the running state of the monitoring equipment based on a comparison result of an abnormal score determined by the trained decision tree and a preset value. The self-checking efficiency of the running state of the monitoring equipment can be improved, and meanwhile the accuracy of the self-checking result is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing. More specifically, the present invention relates to a method and system for self-checking the operating state of a device based on machine learning. Background Art

[0002] The self-checking function for monitoring the operating state of a device refers to the device automatically detecting and diagnosing faults based on its own operating parameters. With the continuous development of monitoring devices, performing real-time self-checking of the operating state can effectively improve the device efficiency and reduce the probability of faults. For monitoring devices, abnormal current data is often a precursor to an impending device fault. By real-time monitoring of abnormal current data, abnormalities in device operation can be detected in a timely manner, and preventive measures can be taken to avoid faults.

[0003] In the related art, as disclosed in the patent application document with the publication number CN118708936A, a method and system for monitoring device power-off are disclosed. The method includes: collecting in real time the device performance data corresponding to the regional devices, and extracting the power-related data in the device performance data; capturing the temporal correlation corresponding to the power-related data, and extracting the time-domain features corresponding to the power-related data; constructing a data isolation tree corresponding to the power-related data, and calculating the abnormal score value corresponding to the power-related data; identifying the abnormal power data in the power-related data, and screening out the abnormal devices in the regional devices according to the abnormal power data.

[0004] However, the above solution does not consider the influence of noise when identifying abnormal power data. In practical applications, the performance data of devices is often interfered by factors such as environmental changes, power fluctuations, and sensor errors, generating noise. Noise will disrupt the normal fluctuations of the data, thereby affecting the accuracy of the abnormal detection model. Due to the strong randomness and volatility of noise, it may be misjudged as normal changes, which in turn affects the construction of the isolation tree, resulting in normal data being misjudged as abnormal, or the true abnormal data not being correctly identified, thus affecting the judgment of the device state. Summary of the Invention

[0005] To solve the problem of being unable to accurately identify the abnormal operating state of a device, the present invention provides a method and system for self-checking the operating state of a device based on machine learning.

[0006] According to the first aspect of the present invention, there is provided a method for self-checking the operating state of a device based on machine learning, including: Obtaining the current sequence of the monitoring device; Using the noise level of the current sequence to adaptively adjust the subsample size of the SCIForest algorithm, where the subsample size is negatively correlated with the noise level; Randomly select several data sets from the current sequence with the same size as the sub-sample, use the SCIForest algorithm to construct decision trees, deploy the trained decision trees to the monitoring devices, and self-check the status of the monitoring devices based on the comparison result between the anomaly score determined by the trained decision trees and the preset value; The method for obtaining the noise level is as follows: calculate the noise performance of any data in the current sequence, and the noise performance is positively correlated with the fitting loss of the data within the preset local range centered on this data, as well as the ratio of the value range and the number of data categories of the data within this local range; With the goal of maximizing the between-class variance, divide the data in the current sequence into two categories, and use the ratio of the amount of data in the category with a larger noise performance to the amount of data in the other category to weight the average noise performance of the current sequence to obtain the noise level of the current sequence.

[0007] The anomaly detection algorithm adopted by the present invention when constructing decision trees can effectively solve the problem of poor performance of the isolation forest algorithm in detecting cluster anomalies, and can process massive data, improving the efficiency of anomaly detection. At the same time, according to the noise level of the current sequence, the sub-sample size is adaptively adjusted, which can improve the noise resistance of anomaly detection and improve the calculation accuracy. Thus, the accuracy of the anomaly detection result is ensured, and the accuracy of the self-check result of the operating status of the monitoring device is improved.

[0008] Preferably, the noise performance of any data in the current sequence satisfies the following relational expression: ; In the formula, is the noise performance of the th data in the current sequence; is the data ordinal number in the current sequence; is the th data value range within the local range of the th data; is the number of data categories within the local range of the th data; is the th data fitting value within the local range of the th data; is the th data value within the local range of the th data; is the th data total number within the local range of the th data; is the absolute value symbol; is the hyperbolic tangent function.

[0009] The present invention can avoid misjudging normal data changes as changes caused by noise, ensuring the accuracy of noise performance.

[0010] Preferably, the method for obtaining the fitting value includes: Performing curve fitting on the data within a local range using the least squares method to obtain the fitting values of the data within the local range.

[0011] Preferably, the subsample size of the SCIForest algorithm is adaptively adjusted using the noise level of the current sequence to satisfy the following relationship: ; In the formula, is the adjusted subsample size; is the preset initial subsample size; is the noise level of the current sequence; is the exponential function with the natural constant as the base; is the ceiling symbol.

[0012] In the present invention, the subsample size is adaptively adjusted. When the noise level is high, by reducing the subsample size, the influence of noise can be weakened; when the noise level is high, by increasing the subsample size, the noise can be prevented from being ignored, thereby improving the calculation accuracy.

[0013] Preferably, the method for obtaining the anomaly score includes: Obtaining the current at the current moment, and traversing all the trained decision trees using the current at the current moment to obtain the path lengths of the current at the current moment in each of the trained decision trees; Calculating the average value of the path lengths of the current at the current moment in the trained decision trees and normalizing it to obtain the anomaly score of the current at the current moment.

[0014] Preferably, based on the comparison result between the anomaly score determined by the trained decision tree and a preset value, the state of the self-check monitoring device includes: Obtaining the preset value. When the anomaly score is greater than or equal to the preset value, it is determined that there is an anomaly in the operating state of the monitoring device at the current moment; When the anomaly score is less than the preset value, it is determined that the operating state of the monitoring device is normal at the current moment.

[0015] The present invention can accurately identify the anomaly in the operating state of the monitoring device.

[0016] Preferably, the method further includes: Presetting an update period; Within one update period, using the decision tree constructed with the current sequence collected in the first half period of the update period to determine the anomaly scores at each sampling moment in the second half period of the update period.

[0017] In order to avoid the influence of the difference in the current change trend on the detection result, the present invention introduces a mechanism of dynamically updating the period, which can more accurately track the trend change of the current fluctuation, avoid misjudgment caused by different trends, and thus improve the accuracy of the self-check result of the operation state of the monitoring device.

[0018] According to a second aspect of the present invention, there is provided a self-check system for the operation state of a device based on machine learning. The system includes a memory and a processor. A computer program is stored on the memory, and the processor executes the computer program to implement the steps of the first aspect of the present invention.

[0019] The present invention has the following effects: 1. The anomaly detection algorithm utilized by the present invention can process and analyze data in parallel by constructing multiple decision trees, thereby improving the self-check efficiency of the operation state of the monitoring device.

[0020] 2. The present invention adaptively adjusts the subsample size in combination with the noise level of the current sequence, which can improve the anti-noise ability of the anomaly detection algorithm. When the noise level is high, by setting a smaller subsample size, the interference of noise can be reduced; when the noise level is low, by setting a larger subsample size, the noise can be prevented from being ignored and the calculation accuracy can be improved, so that the abnormal operation state of the monitoring device can be accurately identified and the accuracy of the self-check result of the operation state of the monitoring device can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] 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 drawings, several embodiments of the present invention are shown by way of illustration and not limitation, and the same or corresponding reference numerals indicate the same or corresponding parts, wherein: Figure 1 is a schematic flowchart of the steps of a method for self-checking the operation state of a device based on machine learning according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, embodiments of the present invention. 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.

[0023] The following will describe the specific embodiments of the present invention in detail with reference to the accompanying drawings.

[0024] Refer to Figure 1 , a method for self-checking the operation state of a device based on machine learning, includes steps S1 - S3, specifically as follows: S1: Obtain the current sequence of the monitoring device.

[0025] Specifically, a current sensor electrically connected to the monitoring device can be used to collect the current of the monitoring device at a certain frequency, such as 10 Hz, to obtain the current sequence of the monitoring device. The present invention does not particularly limit the data collection frequency.

[0026] S2: Adaptively adjust the subsample size of the SCIForest algorithm using the noise level of the current sequence. The subsample size is negatively correlated with the noise level.

[0027] It should be noted that the SCIForest algorithm is a commonly used isolation-based machine learning anomaly detection algorithm, which is improved based on the IForest algorithm. Aiming at the problem of poor performance of the original algorithm in detecting cluster anomalies, SCIForest proposes a new splitting criterion function and a method for constructing random hyperplanes. In the SCIForest algorithm, the size of the subsample set is usually fixed. However, since monitoring devices are mostly located outdoors or in noisy environments, the current may be affected by noise to varying degrees, and the impact of noise changes over time and is uncertain. A fixed-size subsample set may not be able to fully adapt to such noise fluctuations, resulting in inaccurate anomaly detection results, that is, noise may be misjudged as an anomaly, or normal data may be misjudged as an anomaly due to the influence of noise. Therefore, the present invention improves the traditional SCIForest algorithm by analyzing the noise level of the current sequence and dynamically adjusting the subsample size of the SCIForest algorithm, thereby enhancing the anti-noise ability of the algorithm and improving the accuracy and efficiency of current anomaly detection of the monitoring device.

[0028] Among them, the noise level refers to the data that can reflect the influence of the current sequence by noise. For example, when the current sequence is greatly affected by noise, the noise level of the current sequence is high; conversely, if the current sequence is less affected by noise, the noise level of the current sequence is low.

[0029] Optionally, when the noise level of the current sequence is large, reducing the subsample size of the SCIForest algorithm can reduce the influence of noise on the SCIForest algorithm and reduce the computational amount at the same time; when the noise level of the current sequence is small, increasing the subsample size of the SCIForest algorithm can improve the computational accuracy of the SCIForest algorithm.

[0030] Specifically, the determination of the noise level of the current sequence can be achieved through the following steps: Step 1: Calculate the noise performance of any data in the current sequence. The noise performance is positively correlated with the fitting loss of the data within a preset local range centered on this data, as well as the ratio of the value range and the number of data categories of the data within this local range. It should be noted that due to the complex working environment of the monitoring device, there are often various types of noise in the current sequence collected by the current sensor, such as random noise caused by internal electronic components of the device, and burst noise caused by external environmental interference. These noises usually cause violent fluctuations of the current data within a certain range. Therefore, the present invention utilizes this feature to calculate the ratio of the value range of the data within a local range to the number of data categories, and can evaluate the possibility that the data is noise.

[0031] It should be further noted that the load change of the monitoring device may cause large fluctuations in the current, thus misjudging the normal data change as the fluctuation caused by noise. However, since the normal data change usually shows a gradual change of the current within a short period of time, and its change trend is relatively regular, while the noise usually shows random fluctuations. Therefore, the present invention utilizes this feature to measure the fitting difficulty of the data within a local range through the fitting loss of the data within the local range, so as to be able to distinguish the reason for the data change and avoid misjudging the normal data change as the fluctuation caused by noise.

[0032] Among them, the noise manifestation refers to the possibility that the data is noise. The local range refers to a data segment centered on any data and with a length that contains multiple, such as 15 data. In this embodiment, the number of data included in the local range is 15, and the present embodiment does not make a special limitation on the number of data included in the local range.

[0033] Optionally, when the length of the data segment on either side of any data in the current sequence is less than 7, the data segment on that side is discarded, and the continuous data segment composed of the 7 data on the other side and this data is used as the local range of this data.

[0034] Specifically, the noise manifestation of any data in the current sequence satisfies the following relational expression: ; In the formula, is the noise manifestation of the th data in the current sequence; is the data ordinal number in the current sequence; is the value range of the data within the local range of the th data; is the number of data categories within the local range of the th data. Among them, the number of categories refers to the number of different values existing within the local range. For example, when the data within the local range is {1, 3, 4, 2, 1, 3}, the number of data categories within this local range is 4, that is, {1, 3, 4, 2}; is the fitting value of the th data within the local range of the th data; For the The local range of the data The value of each data; For the The total number of data within the local range of each data; is the absolute value symbol; is the hyperbolic tangent function.

[0035] in, , For the The maximum value of the data in the local range of data, For the The minimum value of the data in the local range of data.

[0036] Reflects the first The change intensity of the data in the local range of the data. The larger the value, the more drastic the change of the data in the local range. On the contrary, the smaller the value is, the more likely the data in the local range is to be noise. The smaller the possibility that a data is noise, the smaller the noise performance of the corresponding data.

[0037] The first The average fitting loss of the data within the local range of each data. The larger the value, the more difficult it is to fit the data within the local range, that is, the change of the data within the local range is more random, indicating that the possibility that the change intensity within the local range is caused by noise is greater; on the contrary, the smaller the value, the smaller the fitting difficulty of the data within the local range, that is, the change of the data within the local range is regular, indicating that the possibility that the change intensity within the local range is normal current change is greater.

[0038] In another embodiment, the noise performance may also be calculated in other ways, such as by a summation formula instead of a product formula.

[0039] In an exemplary embodiment of the present invention, the fitting value determination of data can be achieved by the following steps: The least square method is used to perform curve fitting on the data within the local range to obtain the fitting values of each data within the local range.

[0040] Optionally, the ridge regression method can also be used to perform curve fitting on the data within a local range. Of course, a suitable method can also be selected according to specific circumstances for curve fitting. This embodiment does not make a special limitation on the selected fitting method. It should be noted that the method of least squares fitting for data is a prior art, and this embodiment will not elaborate on it here.

[0041] Step 2: With the goal of maximizing the between-class variance, divide the data in the current sequence into two categories, and use the ratio of the data volume of the category with a larger noise performance to that of the other category to weight the average noise performance of the current sequence to obtain the noise level of the current sequence.

[0042] It should be noted that the average noise performance of the current sequence can only reflect the overall noise performance of the current sequence, but cannot reflect the true distribution of the noise in the current sequence. For example, when there is less data with a large noise performance and more data with a small noise performance in the current sequence, the overall noise performance of the current sequence will be low after taking the average value, which may be misjudged as the current sequence being less affected by noise, resulting in a low sub-sample size set for the SCIForest algorithm. Therefore, the present invention divides the current sequence into two categories by maximizing the between-class variance, and uses the ratio of the data volume of the category with a larger noise performance to that of the other category to weight the average noise performance of the current sequence to accurately evaluate the influence of noise on the current sequence.

[0043] Specifically, the noise level of the current sequence satisfies the following relational expression: ; In the formula, is the noise level of the current sequence; is the average value of the noise performance of the data in the current sequence; is the data volume of the category with a larger noise performance in the classification result; is the data volume of the category with a smaller noise performance in the classification result.

[0044] Furthermore, after determining the noise level of the current sequence, the sub-sample size of the SCIForest algorithm can be adaptively adjusted using the noise level of the current sequence. Specifically, using the noise level of the current sequence to adaptively adjust the sub-sample size of the SCIForest algorithm satisfies the following relational expression: ; In the formula, is the adjusted sub-sample size; is the preset initial sub-sample size, in the present invention = 100; is the noise level of the current sequence; is the natural constant an exponential function with a base of is the ceiling symbol.

[0045] It should be noted that the initial sub - sample size set in the present invention is a relatively large value, and the present invention only adjusts the initial sub - sample size downward. Of course, an appropriate initial sub - sample size can also be set according to specific circumstances, and the present embodiment does not make special limitations on the setting of the initial sub - sample size.

[0046] S3: Randomly select several data sets with the same size as the sub - sample from the current - time series, use the SCIForest algorithm to construct decision trees, and deploy the trained decision trees to the monitoring device to self - check the status of the monitoring device based on the comparison result between the anomaly score determined by the trained decision trees and the preset value.

[0047] Optionally, when randomly selecting several data sets with the same size as the sub - sample from the current - time series, several data sets can be selected by simple random sampling or by stratified sampling. The present invention does not make special limitations on the selected sampling method. It should be noted that the process of constructing decision trees using the SCIForest algorithm based on the selected data sets and training the constructed decision trees is prior art, and this embodiment does not describe it in detail here.

[0048] Among them, the ensemble scale set in the present invention is 50, that is, the number of constructed decision trees is 50, which also means the number of selected data sets is 50.

[0049] In an exemplary embodiment of the present invention, the determination of the anomaly score can be achieved through the following steps: Obtain the current - time current, use the current - time current to traverse all the trained decision trees, and obtain the path lengths of the current - time current in each of the trained decision trees; calculate the average value of the path lengths of the current - time current in the trained decision trees and normalize it to obtain the anomaly score of the current - time current.

[0050] Among them, in the SCIForest algorithm, the process of converting the average value of the path lengths into an anomaly score by normalization is prior art, and this embodiment does not elaborate on it here.

[0051] In an exemplary embodiment of the present invention, the self - check of the monitoring device can be achieved through the following steps: Obtain the preset value. When the anomaly score is greater than or equal to the preset value, it is determined that the operating state of the monitoring device is abnormal at the current time; when the anomaly score is less than the preset value, it is determined that the operating state of the monitoring device is normal at the current time.

[0052] Optionally, the preset value can be set to 0.56. If it is determined that the anomaly score of the current current is greater than or equal to 0.56, it is determined that the current at the current moment is abnormal, indicating that the monitoring device is in an abnormal operating state at the current moment. An alarm can be issued or a corresponding processing mechanism can be triggered to perform fault processing in a timely manner.

[0053] Optionally, if the anomaly score of the current at the current moment is less than 0.56, it is determined that the current at the current moment is normal data, indicating that the monitoring device is in a normal operating state at the current moment, thereby realizing the self-check of the operating state of the monitoring device.

[0054] In an exemplary embodiment of the present invention, it further includes: A preset update period; within one update period, a decision tree constructed using the current sequence collected in the first half of the update period is used to determine the anomaly scores at each sampling moment in the second half of the update period.

[0055] For example, the update period can be set to 2 , for any update period, the current data in the first period can be collected, a decision tree is constructed using the improved SCIForest algorithm, and the trained decision tree is deployed to the monitoring device to perform anomaly detection on the operating state of the monitoring device in the second period of the update period; after the anomaly detection in this period is completed, the power data in the first period of the next update period is obtained, a new decision tree is constructed using the improved SCIForest algorithm, and the trained new decision tree is redeployed to the monitoring device to perform anomaly detection on the operating state of the monitoring device in the second period of the update period, and so on, to realize the self-check of the operating state of the monitoring device. In this embodiment = 1 min.

[0056] The present invention also provides a device operating state self-check system based on machine learning. The system includes a memory and a processor, and a computer program is stored on the memory. The computer program integrates the functions of the device operating state self-check method based on machine learning. When the computer program is executed, the efficiency of the self-check of the operating state of the monitoring device can be improved, and at the same time, the accuracy of the self-check result can be improved.

[0057] In the description of this specification, the meanings of "a plurality of" and "several" are at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

[0058] Although this specification has shown and described several 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. Many variations, changes and alternative means will occur to those skilled in the art without departing from the spirit and scope of the present invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in practicing the invention.

Claims

1. A method for self-checking the operating state of a device based on machine learning, characterized in that Including: Obtain the current sequence of the monitoring device; Adaptive adjust the sub-sample size of the SCIForest algorithm using the noise level of the current sequence, where the sub-sample size is negatively correlated with the noise level; Randomly select several data sets from the current sequence with the same size as the sub-sample size, construct a decision tree using the SCIForest algorithm, and deploy the trained decision tree to the monitoring device to self-check the status of the monitoring device based on the comparison result between the anomaly score determined by the trained decision tree and a preset value; The method for obtaining the noise level is: calculate the noise performance of any data in the current sequence, where the noise performance is positively correlated with the fitting loss of the data within a preset local range centered on this data, as well as the ratio of the value range of the data within this local range and the number of data categories; Taking the maximization of the between-class variance as the goal, divide the data in the current sequence into two categories, and use the ratio of the data volume of the category with a larger noise performance to the data volume of the other category to weight the average noise performance of the current sequence to obtain the noise level of the current sequence.

2. The method for self-checking the operating state of a device based on machine learning according to claim 1, wherein The noise performance of any data in the current sequence satisfies the following relational expression: ； Wherein, is the noise performance of the th data in the current sequence; is the data ordinal number in the current sequence; is the value range of the data within the local range of the th data; is the number of data categories within the local range of the th data; is the fitting value of the th data within the local range of the th data; is the value of the th data within the local range of the th data; is the total number of data within the local range of the th data; is the absolute value symbol; is the hyperbolic tangent function.

3. The method for self-checking the operating state of a device based on machine learning according to claim 2, wherein The method for obtaining the fitting value includes: Use the least squares method to perform curve fitting on the data within the local range to obtain the fitting values of each data within the local range.

4. The method for self-checking the operating state of a device based on machine learning according to claim 1, characterized in that, The adaptive adjustment of the sub-sample size of the SCIForest algorithm using the noise level of the current sequence satisfies the following relational expression: ； Wherein, is the adjusted sub-sample size; is the preset initial sub-sample size; is the noise level of the current sequence; is the exponential function with the natural constant as the base; is the ceiling symbol.

5. The method for self-checking the operating status of a device based on machine learning according to claim 1, characterized in that The method for obtaining the anomaly score includes: Obtain the current current, and use the current at the current moment to traverse all the trained decision trees to obtain the path length of the current at the current moment in each trained decision tree; Calculate the average value of the path lengths of the current at the current moment in the trained decision trees and normalize it to obtain the anomaly score of the current at the current moment.

6. The method for self-checking the operating state of a device based on machine learning according to claim 5, wherein The self-check of the status of the monitoring device based on the comparison result between the anomaly score determined by the trained decision tree and a preset value includes: Obtain the preset value. When the anomaly score is greater than or equal to the preset value, it is determined that there is an anomaly in the operating state of the monitoring device at the current moment; When the anomaly score is less than the preset value, it is determined that the operating state of the monitoring device is normal at the current moment.

7. The method for self-checking the operating state of a device based on machine learning according to claim 5, wherein The method further includes: Preset an update period; Within one update period, use the decision tree constructed from the current sequence collected in the first half of the update period to determine the anomaly scores at each sampling moment in the second half of the update period.

8. A self-checking system for the operating state of a device based on machine learning, characterized in that, The machine learning-based device operating state self-check system includes a memory and a processor. A computer program is stored on the memory, and the processor executes the computer program to implement the steps of the method according to any one of claims 1-7.

Images