Power supply monitoring method and apparatus

Abstract

The present application discloses a power supply monitoring method and apparatus. The method comprises: acquiring voltage performance indexes of a plurality of periods (S1); performing feature extraction on the voltage performance indexes to obtain a first statistical feature value (S2); detecting the degree of anomaly of the first statistical feature value by using an unsupervised anomaly detection model (S3); and obtaining an input feature vector of the first statistical feature value on the basis of the degree of anomaly, and inputting the input feature vector and the degree of anomaly into a supervised machine learning model for classification recognition to obtain a mode classification recognition result of voltage anomaly (S4).

Description

Power supply monitoring method and device

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to Chinese patent application No. 202311144341.1 filed in China on September 5, 2023, the entire contents of which are incorporated herein by reference. Technical Field

[0003] The present application relates to the technical field of wireless communication services, and in particular to a method and apparatus for power supply monitoring. Background Art

[0004] Traditional communication network wireless master control equipment and radio frequency equipment are typically deployed in the same computer room, and the monitoring of the supporting power supply equipment generally relies on the support of the entire computer room. With the centralized deployment of wireless base station equipment in communication networks, such as the centralized deployment of Cloud-Radio Access Networks (CRAN), the number of minimalist base stations (sites with remotely deployed radio frequency equipment) is increasing, and their proportion is showing an upward trend. However, the monitoring conditions for the supporting power supply equipment of existing minimalist base stations are relatively simple, and there are weaknesses in power supply monitoring. For example, it is difficult to quickly and effectively detect mains power outages and effectively monitor power supply stability.

[0005] Existing methods for connecting external alarms to RF devices can monitor power outages, but older RF device models may not support external alarms. Furthermore, most external alarm access cables only support a 15-meter distance, making them unsuitable for on-site deployment. Even if deployment conditions are met, they still require one cable per site and the associated human resources for installation and commissioning. Adding an IoT power monitoring unit on-site can also provide effective power monitoring, but this method also requires additional physical resources per site and the human resources required for installation and commissioning.

[0006] Monitoring the power supply and stability of minimalist base stations using existing external alarm access for RF devices or adding an IoT power monitoring unit presents the following issues. Implementing power monitoring requires high investment, low efficiency, and a poor cost-effectiveness ratio. Both existing monitoring methods require additional on-site physical resources and require human resources for installation and commissioning. Each site requires a specific investment, meaning that implementing network-wide minimalist base station power monitoring requires significant investment, resulting in a poor cost-effectiveness ratio.

[0007] Summary of the Invention

[0008] To this end, the first purpose of this application is to propose a power supply monitoring method. By finding a method that does not increase physical investment and centrally configures without the need for one site at a time, it can efficiently and accurately realize the function of monitoring the power supply of the radio frequency equipment of the minimalist station, and can be deployed in the existing network at a lower cost to realize the monitoring function of the supporting power supply of the minimalist station in the entire network.

[0009] The second objective of this application is to provide a power supply monitoring device.

[0010] To achieve the above objectives, a first embodiment of the present application provides a power supply monitoring method, comprising:

[0011] Obtain voltage performance indicators for multiple cycles;

[0012] Performing feature extraction on the voltage performance indicator to obtain a first statistical characteristic value;

[0013] Detecting the degree of abnormality of the first statistical eigenvalue using an unsupervised anomaly detection model;

[0014] An input feature vector of a first statistical feature value is obtained based on the abnormality degree, and the input feature vector and the abnormality degree are input into a supervised machine learning model for classification and recognition to obtain a pattern classification and recognition result of the voltage abnormality.

[0015] The power supply monitoring method according to an embodiment of the present application may also have the following additional technical features:

[0016] In one embodiment of the present application, obtaining voltage performance indicators of multiple cycles includes:

[0017] The base station radio frequency device's own counter is used to periodically and continuously detect the input voltage of the base station's remote radio frequency device to obtain voltage performance indicators for multiple cycles.

[0018] In one embodiment of the present application, performing feature extraction on the voltage performance indicator to obtain a first statistical feature value includes:

[0019] Performing voltage value aggregation statistics on the voltage performance indicator based on feature engineering processing to obtain a second statistical feature value;

[0020] Calculating the voltage performance index of the plurality of cycles using a MeanShift mean shift algorithm to obtain a third statistical characteristic value;

[0021] The first statistical feature value is obtained based on the second statistical feature value and the third statistical feature value.

[0022] In one embodiment of the present application, calculating the voltage performance index of the multiple cycles using the MeanShift mean shift algorithm to obtain the third statistical characteristic value includes:

[0023] Representing the voltage performance indicators of the multiple cycles as voltage performance indicator data points, and randomly selecting a point as a first center point;

[0024] Obtaining a data point set based on all voltage performance indicator data points within a preset bandwidth from the first center point;

[0025] Calculating a vector from the first center point to each element in the data point set to obtain an offset vector;

[0026] Based on the movement distance of the first center point in the preset direction, a second center point is determined when the magnitude of the obtained offset vector meets a preset threshold requirement;

[0027] The voltage mean shift degree is set as the distance between the voltage performance indicator data point and the second center point to obtain a third statistical characteristic value according to the distance setting result.

[0028] In one embodiment of the present application, the second statistical characteristic value includes: the mean, median, minimum, maximum and standard deviation of the voltage; the third statistical characteristic value includes the mean, median, minimum, maximum and standard deviation of the voltage mean drift degree; the first statistical characteristic value also includes at least: the mean, median, minimum, maximum, standard deviation of the current granularity of the voltage, and the voltage change values ​​of multiple consecutive statistical periods before and after.

[0029] In one embodiment of the present application, detecting the abnormality of the first statistical feature value using an unsupervised anomaly detection model includes:

[0030] Inputting the first statistical characteristic value into an n-sigma anomaly detection model to calculate a confidence interval band;

[0031] The first statistical characteristic value is numerically compared with the confidence interval band, and the abnormality degree of the first statistical characteristic value is determined according to the numerical comparison result.

[0032] In one embodiment of the present application, inputting the first statistical characteristic value into an n-sigma anomaly detection model to calculate a confidence interval band includes:

[0033] Calculate the median and MAD variance of the first statistical eigenvalue using the n-sigma anomaly detection model;

[0034] The confidence interval band is calculated based on the median and the MAD variance; wherein the confidence interval band includes an upper bound and a lower bound, the upper bound Upper bound = Median + n * MAD, and the lower bound Lower bound = Median - n * MAD.

[0035] In one embodiment of the present application, the input feature vector and the abnormality degree are input into a supervised machine learning model for classification and recognition to obtain a pattern classification and recognition result of voltage abnormality, including:

[0036] Inputting the input feature vector and the abnormality degree into a random forest algorithm recognition model for classification recognition, and calculating a first pattern classification recognition result of whether the voltage fluctuation is frequent based on the number of times the first-order difference in the first statistical feature value is greater than a relative percentage of the reference value or the number of times the first-order difference is less than a relative percentage of the reference value and the corresponding first abnormality degree;

[0037] Calculating a second pattern classification recognition result of whether the voltage fluctuation is frequent by using the amplitude and variance of the first statistical characteristic value and the corresponding second abnormality degree;

[0038] A third pattern classification and identification result of whether the voltage is too low, too high, or the device is powered off is calculated based on the maximum and minimum voltage values ​​in the first statistical characteristic values ​​and the corresponding third abnormality degree;

[0039] A fourth pattern classification and identification result of whether a voltage jump occurs is calculated by calculating the maximum and minimum values ​​of the mean shift degree in the first statistical characteristic value and the corresponding fourth abnormality degree; and

[0040] The fifth mode classification recognition result of whether the power supply is converted is calculated based on the relative difference between the beginning and the end or the absolute difference between the beginning and the end in the first statistical characteristic value and the corresponding fifth abnormality degree.

[0041] To achieve the above objectives, a second embodiment of the present application provides a power supply monitoring device, comprising:

[0042] A voltage index acquisition module is used to obtain voltage performance indicators of multiple cycles;

[0043] an eigenvalue calculation module, configured to extract characteristics of the voltage performance indicator to obtain a first statistical eigenvalue;

[0044] an abnormality degree calculation module, configured to detect the abnormality degree of the first statistical eigenvalue using an unsupervised anomaly detection model;

[0045] A voltage anomaly identification module is used to obtain an input feature vector of a first statistical feature value based on the abnormality degree, and input the input feature vector and the abnormality degree into a supervised machine learning model for classification and identification to obtain a pattern classification and identification result of the voltage anomaly.

[0046] The power monitoring method and apparatus of the present invention, by finding a centralized configuration method without requiring site-by-site configuration without additional physical investment, can efficiently and accurately monitor the power supply of minimalist station radio frequency equipment. These methods and apparatus can be deployed on existing networks at a low cost to enable network-wide monitoring of the power supply of minimalist stations. This solves the high investment, low efficiency, and poor cost-performance inherent in existing approaches of integrating external alarms into radio frequency equipment or adding IoT power monitoring units.

[0047] To achieve the above-mentioned purpose, the third aspect embodiment of the present application proposes a computer device, comprising: a processor and a memory; wherein the processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to implement the power supply monitoring method as described in the first aspect embodiment.

[0048] To achieve the above-mentioned purpose, the fourth embodiment of the present application proposes a non-temporary computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the power supply monitoring method as described in the first embodiment is implemented.

[0049] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0051] FIG1 is a flow chart of a method for power monitoring according to an embodiment of the present application;

[0052] FIG2 is a schematic diagram showing the principle of an n-sigma algorithm according to an embodiment of the present application;

[0053] FIG3 is a framework diagram of a method for power monitoring according to an embodiment of the present application;

[0054] FIG4 is a flow chart of another power supply monitoring method according to an embodiment of the present application;

[0055] FIG5 is a diagram of input data without feature engineering according to an embodiment of the present application;

[0056] FIG6 is a diagram of input data after unsupervised anomaly detection and feature engineering conversion according to an embodiment of the present application;

[0057] FIG7 is a flowchart of another power monitoring method according to an embodiment of the present application;

[0058] FIG8 is a flowchart of another power monitoring method according to an embodiment of the present application;

[0059] FIG9 is a schematic diagram of model verification according to an embodiment of the present application;

[0060] FIG10 is a schematic structural diagram of a power supply monitoring device according to an embodiment of the present application. DETAILED DESCRIPTION

[0061] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0062] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0063] The following describes a method and apparatus for power supply monitoring according to embodiments of the present application with reference to the accompanying drawings.

[0064] FIG1 is a flow chart of a method for power monitoring according to an embodiment of the present application.

[0065] As shown in FIG1 , the method includes but is not limited to the following steps:

[0066] S1, obtain voltage performance indicators for multiple cycles.

[0067] It is understandable that the present application can utilize the counter of the base station radio frequency device itself to periodically and continuously record the input voltage value of the remote radio frequency device of the base station.

[0068] It is understandable that the base station radio frequency equipment can be powered by an integrated power supply system. In addition to converting the AC power of the mains into DC power, the integrated power supply system is also equipped with an external battery. When the AC power of the mains is interrupted, the battery is used for power supply.

[0069] It is understandable that the rated output voltage of the AC-DC converters of these power supply systems is higher than the rated output voltage of the external battery.

[0070] It is understandable that there is a voltage difference between the two rated output voltages. This solution uses the changes in the input voltage values ​​recorded by the base station RF equipment itself to identify the changes in the output voltage of the external power supply of the base station RF equipment, and then determine whether the power supply equipment has changed.

[0071] In one embodiment of the present application, the present application mainly continuously collects voltage performance indicators of multiple 15-minute cycles. The voltage indicators of the 15-minute cycle are represented as a sequence of 150 points, with one voltage collection point every 6 seconds. The voltage is the average input voltage value of the board, which is arranged in an array form in chronological order.

[0072] S2: Extract characteristics of the voltage performance indicator to obtain a first statistical characteristic value.

[0073] In the embodiment of the present application, a “mean shift” algorithm is used on the input voltage value recorded by the base station radio frequency device to create a data feature value.

[0074] It's understandable that the "mean shift" characteristic value of the base station radio equipment input voltage has a strong correlation with sudden drops or increases in the input voltage. Based on the characteristics of the input voltage changes, it can be used to determine whether the external AC-to-DC power supply of the remote radio equipment in the base station has been interrupted and whether to switch to an external battery. Alternatively, it can be used to determine whether the external power supply has been restored and whether to switch from battery power to AC-to-DC power from the mains.

[0075] It can be understood that feature extraction is to provide effective input to machine learning models and algorithms by transforming relevant data.

[0076] Understandably, because voltage anomalies include multiple patterns, with varying sequences and potential overlap, simply analyzing 15-minute data for time series analysis is insufficient to identify and distinguish various anomaly patterns. Therefore, MeanShift and other statistical measures are employed to provide data statistics at different dimensions, fully characterizing various anomalies within multiple 15-minute periods.

[0077] In one embodiment of the present application, the characteristic value of the present application is the voltage statistical variable characteristic of the board, that is, the first statistical characteristic value, including but not limited to: the reference value of the board voltage mean, median, minimum, maximum, and standard deviation, the mean, median, minimum, maximum, and standard deviation of the current granularity of the board voltage, the mean, median, minimum, maximum, and standard deviation of the board voltage mean drift degree τ, and the voltage change value of the board for three consecutive statistical cycles before and after.

[0078] It can be understood that the input voltage value changes of the base station radio frequency equipment for multiple consecutive statistical cycles are used as characteristic values ​​to determine whether the external power supply is interrupted and whether to switch to using an external battery for power supply; or to determine whether the external power supply is restored and switch from battery power supply to AC-DC power supply.

[0079] Specifically, if the voltage in the first cycle is the rated output voltage of the AC-DC power supply, and the voltage in the third cycle is the rated output voltage of the battery, it is determined that the mains power outage may have occurred in the second cycle, and the battery was used for power supply. Alternatively, if the voltage in the first cycle is the rated output voltage of the battery, and the voltage in the third cycle is the rated output voltage of the AC-DC power supply, it is determined that the mains power supply may have been restored in the second cycle, and the AC-DC power supply was used for power supply.

[0080] For example, consider three consecutive statistical periods, T1 / T2 / T3. If the average voltage during T1 is the rated output voltage of the AC-DC converter, and the average voltage during T3 is the rated output voltage of the battery, it can be assumed that a power supply switch occurred at T2 with a high probability, with the base station's RF equipment switching from AC-DC power to external battery power.

[0081] S3, using an unsupervised anomaly detection model to detect the degree of anomaly of the first statistical eigenvalue.

[0082] In one embodiment of the present application, an n-sigma algorithm is used to detect anomalies in the above input voltage statistical variable characteristics to screen out as many anomalies as possible. According to the results of statistical analysis of the initial labels of anomalies, currently selecting n=3 can cover as many anomalies as possible.

[0083] As you can understand, the principles of the n-sigma algorithm are commonly used in engineering science. As shown in Figure 2, the n-sigma principle can be simply described as follows: if the data follows a normal distribution, an outlier is defined as a value in a set of results that deviates from the mean by more than n standard deviations. In other words, under the assumption of a normal distribution, the probability of a value that is n sigma (standard deviation) away from the mean is very small, and therefore it can be considered an outlier. If the data does not follow a normal distribution, it can also be described by a number of standard deviations away from the mean. (This makes the principle applicable to different business scenarios, but the value of n in n-sigma needs to be determined empirically, and this value can be considered a threshold.)

[0084] In one embodiment of the present application, n-sigma anomaly detection is performed for different input features, and feature calculation can be performed based on the voltage statistical variable features of the board in the past 7 days.

[0085] It is understandable that additional rules and ranges need to be added based on the business requirements of different scenarios. For example, voltage values ​​cannot be negative, so the lower boundary is 0, etc. for boundary correction. For special scenarios, such as only including the degree of abnormality in one direction, only excessive or insufficient values ​​can be considered.

[0086] It is understandable that the degree of anomaly, by describing the degree of anomaly detection, can make it possible to more accurately judge anomalies and characterize abnormal patterns in subsequent machine learning algorithms.

[0087] S4, obtaining an input feature vector of the first statistical feature value based on the abnormality degree, and inputting the input feature vector and the abnormality degree into a supervised machine learning model for classification and recognition to obtain a pattern classification and recognition result of the voltage abnormality.

[0088] In one embodiment of the present application, a supervised learning "random forest" algorithm model is used, and the "mean drift" characteristic value of the input voltage of the base station radio frequency equipment, the characteristic value of the input voltage change for multiple consecutive cycles, and the input voltage's value, median, minimum value, maximum value, standard deviation and other statistical characteristics of change are used as the main inputs to perform machine learning training on the model.

[0089] In one embodiment of the present application, all input features are converted into vectors, and the input feature vectors and the degree of abnormality are used as inputs to a supervised learning machine learning algorithm model to perform pattern classification and recognition of voltage anomalies.

[0090] It can be understood that this step mainly determines whether the voltage is abnormal and classifies and identifies the pattern of voltage abnormality.

[0091] Understandably, based on statistical analysis and expert feedback, the current abnormal modes mainly include voltage drop, voltage recovery, AC power switching to battery, battery switching to AC power, and frequent power fluctuations.

[0092] It is understandable that supervised machine learning algorithms are used for identification to ensure accurate classification and recognition. Therefore, the algorithm of choice is Random Forest. This algorithm is robust to abnormal data, nonlinear, highly accurate, and highly efficient.

[0093] In one embodiment of the present application, the input features of the random forest algorithm are the above-mentioned statistical variable features and their abnormality degree values, so as to calculate and output a plurality of output classification results.

[0094] The embodiments of this application combine unsupervised learning and supervised learning machine learning algorithm models to perform anomaly detection and pattern recognition on the input voltage of radio frequency equipment. The use of machine learning models is to provide robust voltage anomaly detection and pattern recognition based on differences in different regions; it can eliminate as much interference from voltage disturbances as possible on detection and recognition; through the unsupervised model, voltage anomalies can be quickly located with minimal resources, providing cold start deployment; through the supervised model, voltage anomalies can be accurately confirmed and abnormal patterns can be identified; through the supervised model with active learning feedback, adaptive evolution can be achieved in response to network changes, continuously improving accuracy.

[0095] The power supply monitoring method of the embodiment of the present application introduces an input voltage counter of the base station radio frequency equipment, creatively applies the "mean shift" algorithm to establish a characteristic value that describes the amplitude of the input voltage change, applies the input voltage value changes of multiple consecutive statistical cycles as characteristic values, and applies the multi-dimensional statistical characteristic values ​​of the input voltage, combined with the use of unsupervised and supervised machine learning algorithm models, to achieve effective monitoring of the input voltage changes of the base station remote radio frequency equipment.

[0096] Figure 3 is a framework diagram of a power monitoring method according to an embodiment of the present application. As shown in Figure 3, the method includes: data acquisition, statistical feature extraction, unsupervised anomaly detection, supervised classification detection (anomaly detection and pattern recognition), and active anomaly detection feedback. The method monitors the power supply of the remote radio frequency equipment of the base station and recognizes the working status pattern.

[0097] FIG4 is another power monitoring method according to an embodiment of the present application. As shown in FIG4 , the power monitoring method includes but is not limited to the following steps:

[0098] S401: Obtain voltage performance indicators of multiple cycles.

[0099] It should be noted that the specific implementation of step 401 can refer to the above embodiment and will not be described in detail here.

[0100] S402 , performing voltage value aggregation statistics on the voltage performance indicator based on feature engineering processing to obtain a second statistical feature value.

[0101] It's understandable that using a 15-minute voltage data sequence as input for pattern recognition can determine whether the voltage sequence within that granularity is normal, whether battery switching, switching to mains power, frequent voltage fluctuations, or power outages have occurred. However, if 150 points are directly used as input, overfitting is likely to occur if there are insufficient labels. Therefore, feature engineering is required to extract features that are appropriate for the current scenario.

[0102] Among them, the input without feature engineering is shown in Figure 5, and the input after unsupervised anomaly detection and feature engineering conversion is shown in Figure 6.

[0103] It is understandable that feature engineering introduces the MeanShift algorithm, which can capture the pulses generated by voltage fluctuations; the voltage values ​​are aggregated and statistically analyzed, and 150 voltage values ​​within a 15-minute period are aggregated into the voltage mean, median, minimum, maximum, and standard deviation, which can reduce the impact of data disturbances; unsupervised anomaly detection uses the n-sigma algorithm to determine whether each statistical indicator is abnormal and the degree of abnormality, and performs normalization processing, which can reduce the impact of differences in specific voltage values ​​of different devices on model predictions.

[0104] In one embodiment of the present application, the input data is processed by feature engineering, and 150 voltage values ​​in each statistical period are aggregated into voltage mean, median, minimum, maximum, and standard deviation to be recorded as second statistical feature values.

[0105] S403 , using a MeanShift mean shift algorithm to calculate voltage performance indicators of multiple cycles to obtain a third statistical characteristic value.

[0106] In one embodiment of the present application, the MeanShift algorithm is used to calculate the mean, median, minimum, maximum, and standard deviation of the voltage mean shift τ to represent the voltage performance index of multiple cycles as voltage performance index data points, and randomly select a point as the first center point; a data point set is obtained based on all voltage performance index data points within a preset bandwidth from the first center point; a vector from the first center point to each element in the data point set is calculated to obtain an offset vector; based on the movement distance of the first center point in a preset direction, a second center point is determined when the size of the offset vector meets a preset threshold requirement; the voltage mean shift degree is set to the distance between the voltage performance index data point and the second center point to obtain a third statistical eigenvalue based on the distance setting result. The steps of using MeanShift mean shift to create a data eigenvalue recorded as the third statistical eigenvalue are as follows:

[0107] 1) Randomly select a point from the 150 voltage performance index data points of each 15-minute period of the board voltage as the center point;

[0108] 2) Find all voltage performance index data points within the bandwidth from the center point, and record them as set M;

[0109] 3) Calculate the vector from the center point to each element in the set M, add these vectors together to get the offset vector;

[0110] 4) The center point moves along the shift direction, and the moving distance is the modulus of the offset vector;

[0111] 5) Repeat steps 2, 3, and 4 until the size of the offset vector meets the set threshold requirement. The center point at this time is C.

[0112] 6) Take the voltage mean drift degree τ as the distance between 150 voltage performance index data points and the center point C;

[0113] 7) Calculate the mean, median, minimum, maximum, and standard deviation of the board voltage mean drift τ respectively. These values ​​are also used as statistical features and as effective inputs for machine learning models and algorithms.

[0114] S404: Obtain a first statistical eigenvalue based on the second statistical eigenvalue and the third statistical eigenvalue.

[0115] It can be understood that the second statistical eigenvalue extracted includes the mean, median, minimum, maximum and standard deviation of the voltage; the third statistical eigenvalue includes the mean, median, minimum, maximum and standard deviation of the voltage mean drift degree, and other statistical eigenvalues ​​are counted as the final eigenvalue and recorded as the first statistical eigenvalue.

[0116] In one embodiment of the present invention, the statistical features after extraction are shown in Table 1:

[0117] Table 1

[0118] S405: Detect the abnormality level of the first statistical eigenvalue using an unsupervised anomaly detection model.

[0119] S406: Obtain an input feature vector of the first statistical feature value based on the abnormality degree, and input the input feature vector and the abnormality degree into a supervised machine learning model for classification and recognition to obtain a pattern classification and recognition result of the voltage abnormality.

[0120] It should be noted that the specific implementation of steps S405-S406 can refer to the above embodiment and will not be described in detail here.

[0121] The power supply monitoring method of the embodiment of the present application obtains voltage performance indicators of multiple cycles, aggregates and statistics the voltage values ​​of the voltage performance indicators based on feature engineering processing to obtain a second statistical eigenvalue, calculates the voltage performance indicators of multiple cycles using the MeanShift mean shift algorithm to obtain a third statistical eigenvalue, obtains a first statistical eigenvalue based on the second statistical eigenvalue and the third statistical eigenvalue, and uses an unsupervised anomaly detection model to detect the degree of abnormality of the first statistical eigenvalue. An input feature vector of the first statistical eigenvalue is obtained based on the degree of abnormality, and the input feature vector and the degree of abnormality are input into a supervised machine learning model for classification and recognition to obtain a pattern classification recognition result of the voltage anomaly. The present application introduces an input voltage counter of a base station radio frequency device, creatively applies the "mean shift" algorithm to establish a feature value that describes the amplitude of the input voltage change, applies the input voltage value changes of multiple consecutive statistical cycles as feature values, and applies the multi-dimensional statistical feature values ​​of the input voltage, and combines the use of unsupervised and supervised machine learning algorithm models to achieve effective monitoring of the input voltage changes of the base station remote radio frequency device.

[0122] FIG7 is a flow chart of another power monitoring method according to an embodiment of the present application, as shown in FIG7 , including but not limited to the following steps:

[0123] S701: Obtain voltage performance indicators of multiple cycles.

[0124] S702: Perform feature extraction on the voltage performance indicator to obtain a first statistical feature value.

[0125] It should be noted that the specific implementation of steps S701-S702 can refer to the above embodiment and will not be described in detail here.

[0126] S703: Input the first statistical characteristic value into an n-sigma anomaly detection model to calculate a confidence interval band.

[0127] It can be understood that this application uses the n-sigma algorithm to determine whether each statistical indicator is abnormal and the degree of abnormality, compresses information, reduces the impact of data disturbances, and reduces the dimension of the input data, which can make the subsequent classification model more stable.

[0128] In one embodiment of the present application, the input characteristic value is the voltage statistical variable characteristic of the board, including: the reference value of the board voltage mean, median, minimum, maximum, and standard deviation, the mean, median, minimum, maximum, and standard deviation of the current granularity of the board voltage, the mean, median, minimum, maximum, and standard deviation of the board voltage mean drift degree τ, and the voltage change value of the board for three consecutive statistical cycles before and after.

[0129] In one embodiment of the present application, the n-sigma anomaly detection model is used to calculate the median and MAD variance of the first statistical eigenvalue; the confidence interval band is calculated based on the median and MAD variance; wherein the confidence interval band includes a boundary and a lower boundary, the upper bound Upper bound = Median + n * MAD, and the lower bound Lower bound = Median - n * MAD.

[0130] Specifically, n-sigma anomaly detection is performed on the above variable characteristics respectively, the median and MAD variance are calculated, and then the confidence interval band is calculated: upper bound and lower bound, upper bound Upper bound = Median + n*MAD, lower bound Lower bound = Median - n*MAD.

[0131] S704: Numerically compare the first statistical characteristic value with the confidence interval band, and determine the abnormality degree of the first statistical characteristic value according to the result of the numerical comparison.

[0132] In one embodiment of the present application, the above-mentioned variable characteristics are compared with the confidence interval band. If the above-mentioned variable characteristics are greater than or equal to the upper boundary, the above-mentioned variable characteristics are abnormally too large; if the above-mentioned variable characteristics are less than or equal to the lower boundary, the above-mentioned variable characteristics are abnormally too small; in other cases, the above-mentioned variable characteristics are normal.

[0133] In one embodiment of the present application, the degree of abnormality = (input feature value - Median) / MAD; the larger the absolute value of the degree of abnormality, the more obvious the abnormality. The positive and negative signs indicate whether the abnormality is too large or too small.

[0134] In one embodiment of the present application, all input features are finally converted into vectors to describe (f1, f2, f3…) based on these anomalies, and the output is (ad1, ad2, ad3…) as the input of the supervised learning machine learning algorithm model.

[0135] S705 , obtaining an input feature vector of the first statistical feature value based on the abnormality degree, and inputting the input feature vector and the abnormality degree into a supervised machine learning model for classification and recognition to obtain a pattern classification and recognition result of the voltage abnormality.

[0136] It should be noted that the specific implementation of step S705 can refer to the above embodiment and will not be described in detail here.

[0137] The power supply monitoring method of the embodiment of the present application obtains voltage performance indicators of multiple cycles, performs feature extraction on the voltage performance indicators to obtain a first statistical eigenvalue, inputs the first statistical eigenvalue into an n-sigma anomaly detection model to calculate a confidence interval band, numerically compares the first statistical eigenvalue with the confidence interval band, determines the degree of abnormality of the first statistical eigenvalue based on the numerical comparison result, obtains an input eigenvector of the first statistical eigenvalue based on the degree of abnormality, and inputs the input eigenvector and the degree of abnormality into a supervised machine learning model for classification and recognition to obtain a pattern classification recognition result of the voltage anomaly. The present application introduces an input voltage counter of a base station radio frequency device, creatively applies a "mean shift" algorithm to establish a eigenvalue that describes the amplitude of input voltage changes, applies the input voltage value changes of multiple consecutive statistical cycles as eigenvalues, and applies multi-dimensional statistical eigenvalues ​​of the input voltage, and combines unsupervised and supervised machine learning algorithm models to achieve effective monitoring of input voltage changes of remote radio frequency devices of base stations.

[0138] FIG8 is a flow chart of another power monitoring method according to an embodiment of the present application, as shown in FIG8 , including but not limited to the following steps:

[0139] S801: Obtain voltage performance indicators for multiple cycles.

[0140] S802: Perform feature extraction on the voltage performance indicator to obtain a first statistical feature value.

[0141] S803: Detect the abnormality level of the first statistical eigenvalue using an unsupervised anomaly detection model.

[0142] It should be noted that the specific implementation of steps S801-S803 can refer to the above embodiment and will not be described in detail here.

[0143] S804: Input the input feature vector and the abnormality level into a random forest algorithm recognition model for row classification recognition. A first pattern classification recognition result indicating whether voltage fluctuations are frequent is calculated based on the number of times the first-order difference in the first statistical feature value is greater than a relative percentage of the reference value, or the number of times the first-order difference is less than a relative percentage of the reference value, and the corresponding first abnormality level.

[0144] S805, calculating a second pattern classification result of whether the voltage fluctuation is frequent based on the amplitude and variance of the first statistical characteristic value and the corresponding second abnormality degree;

[0145] S806, calculating a third pattern classification result of whether the voltage is too low, too high, or the device is powered off based on the maximum and minimum voltage values ​​in the first statistical characteristic value and the corresponding third abnormality level;

[0146] S807, calculating a fourth pattern classification result of whether a voltage jump occurs based on the maximum and minimum values ​​of the mean shift degree in the first statistical characteristic value and the corresponding fourth abnormality degree; and

[0147] S808: Calculate a fifth mode classification recognition result of whether the power supply is switched based on the relative difference or absolute difference between the beginning and the end of the first statistical characteristic value and the corresponding fifth abnormality degree.

[0148] Specifically, by counting the number of times the first-order difference in the input feature is greater than a relative percentage of the reference value, or the number of times the first-order difference is less than a relative percentage of the reference value, it is possible to determine whether the voltage fluctuation is frequent. The more times, the more frequent the voltage fluctuation.

[0149] The amplitude and variance of the input features can be used to determine whether the voltage fluctuation is severe or frequent. The larger the amplitude and variance, the more frequent the power supply fluctuations.

[0150] By inputting the maximum and minimum voltage characteristics, you can determine whether the voltage is too low, too high, or the device has lost power. If the maximum voltage is higher than the rated output voltage of the power supply, the voltage is too high; if the maximum voltage is lower than the rated output voltage, the voltage is too low; if the minimum voltage is 0, the voltage has lost power; if the maximum, minimum, and rated output voltage of the power supply are equal, the voltage has recovered.

[0151] By inputting the maximum and minimum mean shift values ​​in the characteristics, we can determine whether the voltage has experienced a jump or is just a small fluctuation. The closer the maximum and minimum values ​​are, the smaller the voltage fluctuation; otherwise, a power supply switch has occurred.

[0152] By inputting the relative or absolute voltage drop from the beginning to the end as a feature, you can determine the final state after voltage fluctuations: whether the power transition is from mains to battery or vice versa. If the relative or absolute voltage drop from the beginning to the end is greater than 0, the power transition is from mains to battery; otherwise, it is from battery to mains.

[0153] The output indicates the input voltage mode of the radio equipment during the statistical period. Modes include "switching to battery power," "switching to mains power," "unstable and frequent voltage fluctuations," "voltage interruption," "overvoltage," and "undervoltage." If the input voltage mode is "switching to battery power," the base station radio equipment is determined to have switched from an external AC-to-DC power source to battery power. This information can be used to determine whether the voltage is abnormal and to categorize abnormal voltage modes, including: voltage drop, voltage recovery, switching from mains to battery, switching from battery to mains, and frequent power fluctuations.

[0154] Furthermore, the verification data used in this application is a benchmark label obtained by combining expert experience with existing warning rules. As the real result, it is split into a training set and a test set to verify the accuracy, precision, recall rate, and f1-score of the model respectively. The following is the result of ten-fold cross-validation. As shown in Figure 9, it can be seen that the f1-score scores of various modes are all above 0.9, so the model effect is still good.

[0155] Wherein, accuracy (Accuracy) A = (TP + TN) / (TP + FP + TN + FN);

[0156] Wherein, precision (Precision) P = TP / (TP+FP);

[0157] Among them, recall rate (Recall) R = TP / (TP+FN);

[0158] Among them, the comprehensive evaluation index (F-Measure) F1=2(P*R) / (P+R).

[0159] In summary, the present application does not require additional physical resource investment. Compared with the traditional external alarm access method of radio frequency equipment or the addition of an IoT power monitoring unit to achieve power monitoring, both of which require additional physical resource investment, the present application utilizes the existing counter data of the wireless base station master control equipment in the existing network. Through the support of innovative algorithms, relying solely on software investment can greatly reduce the investment in physical resources to achieve the same remote radio frequency equipment power monitoring function. It can reserve support space for subsequent new models of remote radio frequency equipment through a simple software upgrade. The present application is more efficient in achieving power monitoring. Compared with the traditional on-site one-time installation and debugging human resource investment method, the present application performs calculations and processing through a centralized big data platform; utilizing the existing network management system, there is no need to perform on-site configuration of a single site separately. A centralized background configuration can achieve the power monitoring function of the remote radio frequency equipment of the entire network, and work efficiency is greatly improved. The present application can also make full use of the existing wireless hardware equipment and facilities in the network to efficiently achieve the power monitoring function of the remote radio frequency equipment without increasing the investment of additional physical resources.

[0160] According to the power supply monitoring method of the embodiment of the present application, by introducing the input voltage counter of the base station radio frequency equipment, the "mean shift" algorithm is creatively applied to establish a characteristic value describing the amplitude of the input voltage change, the input voltage value changes of multiple consecutive statistical cycles are used as characteristic values, and the multi-dimensional statistical characteristic values ​​of the input voltage are applied. In combination with the use of unsupervised and supervised machine learning algorithm models, effective monitoring of the input voltage changes of the base station remote radio frequency equipment is achieved.

[0161] To implement the above embodiment, as shown in FIG10 , this embodiment further provides a power supply monitoring device 10 , which includes: a voltage index acquisition module 100 , a characteristic value calculation module 200 , an abnormality degree calculation module 300 , and a voltage abnormality identification module 400 ;

[0162] The voltage index acquisition module 100 is used to obtain voltage performance indicators of multiple cycles;

[0163] The characteristic value calculation module 200 is used to extract the characteristics of the voltage performance index to obtain a first statistical characteristic value;

[0164] An abnormality degree calculation module 300, configured to detect the abnormality degree of the first statistical eigenvalue using an unsupervised anomaly detection model;

[0165] The voltage anomaly identification module 400 is used to obtain an input feature vector of a first statistical feature value based on the degree of anomaly, and input the input feature vector and the degree of anomaly into a supervised machine learning model for classification and identification to obtain a pattern classification and identification result of the voltage anomaly.

[0166] Furthermore, the voltage indicator acquisition module 100 is further configured to:

[0167] The base station radio frequency device's own counter is used to periodically and continuously detect the input voltage of the base station's remote radio frequency device to obtain voltage performance indicators for multiple cycles.

[0168] Furthermore, the eigenvalue calculation module 200 further includes:

[0169] A first calculation subunit is configured to aggregate and count the voltage values ​​of the voltage performance indicator based on feature engineering processing to obtain a second statistical feature value;

[0170] a second calculation subunit, configured to calculate a voltage performance index of a plurality of cycles using a MeanShift mean shift algorithm to obtain a third statistical characteristic value;

[0171] The third calculation subunit is configured to obtain the first statistical characteristic value based on the second voltage characteristic value and the third voltage characteristic value.

[0172] Furthermore, the second computing subunit is further configured to:

[0173] Representing the voltage performance index of multiple cycles as voltage performance index data points, and randomly selecting a point as a first center point;

[0174] Obtaining a data point set based on all voltage performance indicator data points within a preset bandwidth from the first center point;

[0175] Calculate the vector from the first center point to each element in the data point set to obtain an offset vector;

[0176] Based on the movement distance of the first center point in the preset direction, a second center point is determined when the magnitude of the obtained offset vector meets a preset threshold requirement;

[0177] The voltage mean shift degree is set as the distance between the voltage performance indicator data point and the second center point to obtain a third statistical characteristic value according to the distance setting result.

[0178] Furthermore, the second voltage characteristic value includes: the mean, median, minimum, maximum and standard deviation of the voltage; the third voltage characteristic value includes the mean, median, minimum, maximum and standard deviation of the voltage mean drift degree; the first statistical characteristic value also includes at least: the mean, median, minimum, maximum, standard deviation of the current granularity of the voltage, and the voltage change values ​​of multiple consecutive statistical periods before and after.

[0179] Furthermore, the above abnormality degree calculation module 300 includes:

[0180] an interval band calculation subunit, configured to input the first statistical characteristic value into the n-sigma anomaly detection model to calculate a confidence interval band;

[0181] The numerical comparison subunit is used to compare the first statistical characteristic value with the confidence interval band, and determine the abnormality degree of the first statistical characteristic value according to the numerical comparison result.

[0182] Furthermore, the interval band calculation subunit is also used to:

[0183] Calculate the median and MAD variance of the first statistical eigenvalue using the n-sigma anomaly detection model;

[0184] The confidence interval band is calculated based on the median and the MAD variance. The confidence interval band includes an upper bound and a lower bound, where the upper bound is Median + n*MAD and the lower bound is Median - n*MAD.

[0185] Furthermore, the voltage anomaly identification module 400 is further configured to:

[0186] Input the input feature vector and the degree of abnormality into the random forest algorithm recognition model for row classification recognition, and obtain a first pattern classification recognition result of whether the voltage fluctuation is frequent by calculating the number of times the first-order difference in the first statistical feature value is greater than the relative percentage of the reference value or the number of times the first-order difference is less than the relative percentage of the reference value and the corresponding first degree of abnormality;

[0187] A second pattern classification recognition result of whether the voltage fluctuation is frequent is calculated by using the amplitude and variance of the first statistical characteristic value and the corresponding second abnormality degree;

[0188] The third mode classification recognition result of whether the voltage is too low, too high, or the device is powered off is calculated based on the maximum and minimum voltage values ​​in the first statistical characteristic value and the corresponding third abnormality degree;

[0189] A fourth pattern classification and identification result of whether a voltage jump occurs is calculated by calculating the maximum and minimum values ​​of the mean shift degree in the first statistical characteristic value and the corresponding fourth abnormality degree; and

[0190] The fifth mode classification recognition result of whether the power supply is converted is calculated by using the relative difference between the beginning and the end or the absolute difference between the beginning and the end in the first statistical characteristic value and the corresponding fifth abnormality degree.

[0191] The power supply monitoring device of the embodiment of the present application introduces an input voltage counter of the base station radio frequency equipment, creatively applies the "mean shift" algorithm to establish a characteristic value describing the amplitude of the input voltage change, uses the input voltage value changes of multiple consecutive statistical cycles as characteristic values, and applies multi-dimensional statistical characteristic values ​​of the input voltage, combined with the use of unsupervised and supervised machine learning algorithm models, to achieve effective monitoring of the input voltage changes of the base station remote radio frequency equipment.

[0192] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0193] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of such features. Throughout the description of this application, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined.

Claims

1. A method for power supply monitoring, the method comprising: Obtain voltage performance indicators for multiple cycles; Performing feature extraction on the voltage performance indicator to obtain a first statistical feature value; Detecting the degree of abnormality of the first statistical feature value using an unsupervised anomaly detection model; An input feature vector of a first statistical feature value is obtained based on the abnormality degree, and the input feature vector and the abnormality degree are input into a supervised machine learning model for classification and recognition to obtain a pattern classification and recognition result of voltage abnormality.

2. The method according to claim 1, wherein: The obtaining of voltage performance indicators of multiple cycles includes: The input voltage of the remote radio frequency device of the base station is periodically and continuously detected by using the counter of the radio frequency device of the base station to obtain voltage performance indicators of multiple cycles.

3. The method according to claim 2, wherein: Extracting a feature of the voltage performance indicator to obtain a first statistical feature value includes: Based on feature engineering processing, the voltage performance indicator is aggregated and counted to obtain a second statistical feature value; Calculating the voltage performance index of the plurality of cycles using a MeanShift mean shift algorithm to obtain a third statistical characteristic value; The first statistical feature value is obtained based on the second statistical feature value and the third statistical feature value.

4. The method according to claim 3, wherein: The calculating the voltage performance index of the plurality of cycles by using the MeanShift mean shift algorithm to obtain a third statistical characteristic value includes: Representing the voltage performance indicators of the multiple cycles as voltage performance indicator data points, and randomly selecting a point as a first center point; Obtaining a data point set based on all voltage performance indicator data points whose distance from the first center point is within a preset bandwidth; Calculating a vector from the first center point to each element in the data point set to obtain an offset vector; Based on the moving distance of the first center point in the preset direction, a second center point is determined when the size of the obtained offset vector meets a preset threshold requirement; The voltage mean shift degree is set as the distance between the voltage performance indicator data point and the second center point to obtain a third statistical characteristic value according to the distance setting result.

5. The method according to claim 4, wherein: The second statistical characteristic value includes: the mean, median, minimum, maximum and standard deviation of the voltage; the third statistical characteristic value includes the mean, median, minimum, maximum and standard deviation of the voltage mean drift degree; the first statistical characteristic value also includes at least: the mean, median, minimum, maximum and standard deviation of the current granularity of the voltage, and the voltage change values ​​of multiple consecutive statistical periods before and after.

6. The method according to claim 5, wherein: Detecting the abnormality of the first statistical characteristic value using an unsupervised anomaly detection model includes: Inputting the first statistical characteristic value into an n-sigma anomaly detection model to calculate a confidence interval band; The first statistical characteristic value is numerically compared with the confidence interval band, and the abnormality degree of the first statistical characteristic value is determined according to the numerical comparison result.

7. The method according to claim 6, wherein: Inputting the first statistical characteristic value into the n-sigma anomaly detection model to calculate the confidence interval band includes: Calculate the median and MAD variance of the first statistical eigenvalue using the n-sigma anomaly detection model; The confidence interval band is calculated according to the median and the MAD variance; wherein the confidence interval band includes an upper bound and a lower bound, the upper bound Upper bound = Median + n * MAD, and the lower bound Lower bound = Median - n * MAD.

8. The method according to claim 7, wherein: Inputting the input feature vector and the abnormality degree into a supervised machine learning model for classification and recognition to obtain a pattern classification and recognition result of voltage abnormality, including: Input the input feature vector and the abnormality degree into the random forest algorithm recognition model for classification recognition, and calculate the first pattern classification recognition result of whether the voltage fluctuation is frequent by using the number of times the first-order difference in the first statistical feature value is greater than the relative percentage of the reference value or the number of times the first-order difference is less than the relative percentage of the reference value and the corresponding first abnormality degree; A second pattern classification recognition result of whether voltage fluctuation is frequent is calculated by using the amplitude and variance in the first statistical characteristic value and the corresponding second abnormality degree; The third mode classification recognition result of whether the voltage is too low, too high, or the device is powered off is calculated by using the maximum and minimum voltage values ​​in the first statistical characteristic value and the corresponding third abnormality degree; A fourth mode classification recognition result of whether a voltage jump occurs is calculated by calculating the maximum value and the minimum value of the mean shift degree in the first statistical characteristic value and the corresponding fourth abnormality degree; and The fifth mode classification recognition result of whether the power supply is converted is calculated by using the relative difference between the beginning and the end or the absolute difference between the beginning and the end in the first statistical characteristic value and the corresponding fifth abnormality degree.

9. A power supply monitoring device, comprising: A voltage index acquisition module is used to obtain voltage performance indicators of multiple cycles; An eigenvalue calculation module, used for performing feature extraction on the voltage performance indicator to obtain a first statistical eigenvalue; an abnormality degree calculation module, used for detecting the abnormality degree of the first statistical characteristic value by using an unsupervised anomaly detection model; A voltage anomaly identification module is used to obtain an input feature vector of a first statistical feature value based on the abnormality degree, and input the input feature vector and the abnormality degree into a supervised machine learning model for classification and identification to obtain a pattern classification and identification result of the voltage anomaly.

10. The device according to claim 9, wherein: The voltage indicator acquisition module is also used for: The input voltage of the remote radio frequency device of the base station is periodically and continuously detected by using the counter of the radio frequency device of the base station to obtain voltage performance indicators of multiple cycles.

11. The device according to claim 10, wherein: The eigenvalue calculation module further includes: A first calculation subunit is used to aggregate and count the voltage values ​​of the voltage performance indicator based on feature engineering processing to obtain a second statistical feature value; A second calculation subunit, configured to calculate the voltage performance index of the plurality of cycles by using a MeanShift mean shift algorithm to obtain a third statistical characteristic value; a third calculating subunit, configured to calculate the second statistical characteristic value and the third statistical characteristic value based on the second statistical characteristic value; The first statistical characteristic value is obtained.

12. The device according to claim 11, wherein The second computing subunit is further configured to: Representing the voltage performance indicators of the multiple cycles as voltage performance indicator data points, and randomly selecting a point as a first center point; Obtaining a data point set based on all voltage performance indicator data points whose distance from the first center point is within a preset bandwidth; Calculating a vector from the first center point to each element in the data point set to obtain an offset vector; Based on the moving distance of the first center point in the preset direction, a second center point is determined when the size of the obtained offset vector meets a preset threshold requirement; The voltage mean shift degree is set as the distance between the voltage performance indicator data point and the second center point to obtain a third statistical characteristic value according to the distance setting result.

13. The device according to claim 12, wherein: The second statistical characteristic value includes: the mean, median, minimum, maximum and standard deviation of the voltage; the third statistical characteristic value includes the mean, median, minimum, maximum and standard deviation of the voltage mean drift degree; the first statistical characteristic value also includes at least: the mean, median, minimum, maximum and standard deviation of the current granularity of the voltage, and the voltage change values ​​of multiple consecutive statistical periods before and after.

14. The device according to claim 13, wherein: The abnormality degree calculation module comprises: An interval band calculation subunit, used for inputting the first statistical characteristic value into an n-sigma anomaly detection model to calculate a confidence interval band; The numerical comparison subunit is used to compare the first statistical characteristic value with the confidence interval band, and determine the abnormality degree of the first statistical characteristic value according to the numerical comparison result.

15. The device according to claim 14, wherein: The interval band calculation subunit is further used for: Calculate the median and MAD variance of the first statistical eigenvalue using the n-sigma anomaly detection model; The confidence interval band is calculated according to the median and the MAD variance; wherein the confidence interval band includes an upper bound and a lower bound, the upper bound Upper bound = Median + n * MAD, and the lower bound Lower bound = Median - n * MAD.

16. The device according to claim 15, wherein: The voltage anomaly identification module is further used for: Input the input feature vector and the abnormality degree into the random forest algorithm recognition model for classification recognition, and calculate the first pattern classification recognition result of whether the voltage fluctuation is frequent by using the number of times the first-order difference in the first statistical feature value is greater than the relative percentage of the reference value or the number of times the first-order difference is less than the relative percentage of the reference value and the corresponding first abnormality degree; A second pattern classification recognition result of whether voltage fluctuation is frequent is calculated by using the amplitude and variance in the first statistical characteristic value and the corresponding second abnormality degree; The third mode classification recognition result of whether the voltage is too low, too high or the device is powered off is calculated by using the maximum and minimum voltage values ​​in the first statistical characteristic values ​​and the corresponding third abnormality degree; A fourth mode classification recognition result of whether a voltage jump occurs is calculated by calculating the maximum value and the minimum value of the mean shift degree in the first statistical characteristic value and the corresponding fourth abnormality degree; and The fifth mode classification recognition result of whether the power supply is converted is calculated by using the relative difference between the beginning and the end or the absolute difference between the beginning and the end in the first statistical characteristic value and the corresponding fifth abnormality degree.

Images