Fault prediction method, device, electronic device, storage medium and program product

Abstract

The present disclosure provides a fault prediction method, apparatus, electronic device, storage medium and program product. The present disclosure classifies the device status data, and predicts and determines a first prediction value for the gradual data in the device status data, and predicts and determines a second prediction value for the burst data in the device status data. Finally, based on the first prediction value and the second prediction value, the final fault prediction result is determined. In summary, the technical solution provided by the present disclosure can classify a variety of device status data, and accurately predict different types of data in a targeted manner to obtain the final prediction result. In this way, the influence of special types of fault data (such as burst fault data, etc.) on part or all of the data can be eliminated, the accuracy of the prediction is improved, and the actual objective needs can be met.

Description

Technical Field

[0001] The present disclosure relates to the field of communication technology, and in particular to a fault prediction method, device, electronic device, storage medium and program product. Background Art

[0002] At present, most existing fault prediction methods use network models to directly predict and obtain prediction results. This method does not classify the fault types and ignores the impact of some fault type data (such as sudden fault data, etc.) on partial or overall data, which in turn causes the accuracy of the prediction results to decrease and is difficult to meet actual objective needs. Summary of the invention

[0003] The present disclosure provides a fault prediction method, device, electronic device, storage medium and program product, which are used to solve the problem that the fault types are not classified to a certain extent, and the influence of some fault type data (such as sudden fault data, etc.) on part or all data is ignored, which leads to a decrease in the accuracy of the prediction results and difficulty in meeting actual objective needs.

[0004] According to one aspect of the present disclosure, a fault prediction method is provided, the method comprising: classifying device status data; determining a first prediction value for gradual data in the device status data; determining a second prediction value for burst data in the device status data; and determining a fault prediction result based on the first prediction value and the second prediction value.

[0005] In addition, according to a method in one aspect of the present disclosure, device status data is classified, including: for any type of device status data, determining a first probability that the device status data belongs to a gradual failure Gaussian model, and determining a second probability that the device status data belongs to a sudden failure Gaussian model; based on the model category corresponding to the higher of the first probability and the second probability, determining the category of the device status data.

[0006] In addition, according to a method in one aspect of the present disclosure, for any type of device status data, determining the first probability that the device status data belongs to a gradual fault Gaussian model includes: determining the first parameter of the current gradual fault Gaussian model, the first parameter includes: a first variance, a first mean, a first covariance and a first probability density value; determining the second parameter of the device status data in the current gradual fault Gaussian model, the second parameter includes a second variance, a second mean, a second covariance and a second probability density value; obtaining the ratio of the first parameter to the second parameter to obtain the first probability.

[0007] In addition, according to a method in one aspect of the present disclosure, a first prediction value is determined for the progressive data in the device status data, including: using a trained neural network model to process the progressive data to obtain the first prediction value.

[0008] In addition, according to a method in one aspect of the present disclosure, the neural network model is obtained by iteratively training the input data based on the target error function; wherein the target error function is a fusion function of the smoothing loss function and the cosine similarity distance function.

[0009] In addition, according to a method in one aspect of the present disclosure, a second prediction value is determined for burst data in device status data, including: determining a third probability based on a third probability density value; the third probability density value is obtained by processing the probability density function of a Gaussian model of a burst failure; obtaining a third prediction value for the burst data; and determining a second prediction value based on the third probability and the third prediction value.

[0010] In addition, according to a method in one aspect of the present disclosure, a fault prediction result is determined based on a first prediction value and a second prediction value, including: adding the corresponding first prediction value and second prediction value to obtain a fault prediction result.

[0011] In addition, according to a method in one aspect of the present disclosure, before determining a first prediction value for first data in the device status data, the method also includes: preprocessing the device status data; wherein the preprocessing includes at least one of the following: data cleaning, data normalization, and feature extraction.

[0012] According to another aspect of the present disclosure, a network operation and maintenance device is provided, including: a Gaussian mixture GMM model, used to determine the category of equipment status data; a neural network model, used to predict the classified equipment status data to obtain a fault prediction result.

[0013] In addition, according to an apparatus of one aspect of the present disclosure, the GMM model at least includes: a gradual fault Gaussian model and a sudden fault Gaussian model.

[0014] According to another aspect of the present disclosure, a fault prediction device is provided, which includes: a classification unit for classifying device status data; a first determination unit for determining a first prediction value for gradual data in the device status data; a second determination unit for determining a second prediction value for burst data in the device status data; and a third determination unit for determining a fault prediction result based on the first prediction value and the second prediction value.

[0015] According to another aspect of the present disclosure, an electronic device is provided, comprising: a memory for storing computer-readable instructions; and a processor for executing the computer-readable instructions so that the electronic device executes the method described in any embodiment of the first aspect.

[0016] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium is provided for storing computer-readable instructions. When the computer-readable instructions are executed by a processor, the processor executes the method described in any embodiment of the first aspect.

[0017] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, wherein when the computer program is executed by a processor, the method described in any embodiment of the first aspect is implemented.

[0018] The present disclosure provides a fault prediction method, apparatus, electronic device, storage medium and program product. The present disclosure classifies the device status data, and predicts and determines a first prediction value for the gradual data in the device status data, and predicts and determines a second prediction value for the burst data in the device status data. Finally, based on the first prediction value and the second prediction value, a final fault prediction result is determined. In summary, the technical solution provided by the present disclosure can classify a variety of device status data, and accurately predict different types of data in a targeted manner to obtain a final prediction result. In this way, the influence of special types of fault data (such as burst fault data, etc.) on part or all of the data can be eliminated, the accuracy of the prediction is improved, and the actual objective needs can be met.

[0019] It is to be understood that both the foregoing general description and the following detailed description are exemplary, and are intended to provide further explanation of the technology as claimed. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The above and other purposes, features and advantages of the present disclosure will become more apparent by describing the embodiments of the present disclosure in more detail in conjunction with the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present disclosure and constitute a part of the specification. Together with the embodiments of the present disclosure, they are used to explain the present disclosure and do not constitute a limitation of the present disclosure. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0021] Figure 1 A flowchart of a fault prediction method provided by an embodiment of the present disclosure;

[0022] Figure 2 A flowchart of a complete fault prediction method provided by an embodiment of the present disclosure;

[0023] Figure 3A complete flow chart of a network equipment fault prediction model based on a BP neural network provided in an embodiment of the present disclosure;

[0024] Figure 4 A schematic diagram of a complete network operation and maintenance device provided in an embodiment of the present disclosure;

[0025] Figure 5 A complete network equipment fault prediction device and a self-intelligent network-network intelligent operation and maintenance system for the network equipment fault prediction device provided in the embodiments of the present disclosure;

[0026] Figure 6 A structural block diagram of a fault prediction device provided in an embodiment of the present disclosure;

[0027] Figure 7 A hardware block diagram of an electronic device provided in an embodiment of the present disclosure;

[0028] Figure 8 A schematic diagram of a computer-readable storage medium provided for an embodiment of the present disclosure. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the present disclosure more obvious, the exemplary embodiments according to the present disclosure will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all the embodiments of the present disclosure, and it should be understood that the present disclosure is not limited to the exemplary embodiments described here.

[0030] At present, most existing fault prediction methods use network models to directly predict and obtain prediction results. This method does not classify the fault types and ignores the impact of some fault type data (such as sudden fault data, etc.) on partial or overall data, which in turn causes the accuracy of the prediction results to decrease and is difficult to meet actual objective needs.

[0031] Therefore, this application proposes a fault prediction method, which can eliminate the influence of special types of fault data (such as sudden fault data, etc.) on part or all of the data, improve the accuracy of the prediction, and meet the actual objective needs. Figure 1 , Figure 1 A schematic diagram of a fault prediction method provided by an embodiment of the present disclosure. Figure 1 As shown, the method includes:

[0032] In step S101 , device status data is classified.

[0033] In the present disclosure, the device status data can be understood as all relevant data that characterize the status of the device. These relevant data may include but are not limited to the operating parameters of the device (such as the temperature, pressure, speed and other values ​​of the device), the performance indicators of the device (such as the production efficiency, accuracy and other data of the device), the fault records of the device (the time when the historical fault occurred, the type of fault and other related information), and the status information of each component of the device (such as the wear degree of a key component, the connection status and other data). The specific device status data will be described in conjunction with the embodiments below.

[0034] In the embodiments of the present disclosure, there may be multiple device status data. By classifying multiple device status data, a solid foundation can be laid for subsequent processing and analysis of different types of data and accurate prediction of device failures. Specifically, the classification method for device status data may include but is not limited to at least one of the following: classification according to historical records, classification by inputting a classification model, analysis using a classification algorithm, manual classification, etc. In specific applications, it can be flexibly selected or combined according to actual needs, and is not limited here.

[0035] In step S102, a first prediction value is determined for progressive data in the device status data.

[0036] In the present disclosure, progressive data can be understood as data that can change gradually over time, usually showing a slow but continuous trend. For example, the temperature of the device gradually increases, the vibration gradually increases, and other gradually changing data. The first prediction value can be understood as an estimated value of the future state of the progressive data.

[0037] In the disclosed embodiment, the device status data may be one or more, and correspondingly, the gradual data in the device status data may also be one or more. For one or more gradual data, a corresponding prediction model (such as at least one of a linear regression model, an autoregressive integrated moving average ARIMA model, a multivariate linear regression model, a neural network model, etc., which are not exhaustive here) may be used to determine a first prediction value. The purpose of determining the first prediction value is to eliminate the influence of other types of data (such as burst data, etc.) on the gradual data, so as to accurately predict the gradual failure of the equipment.

[0038] In step S103, a second prediction value is determined for the burst data in the device status data.

[0039] In the present disclosure, burst data can be understood as data that suddenly deviates from the normal state and is sudden and unpredictable during the operation of the device. For example, when the device is operating normally, the current is stable within a certain range, and a momentary current peak suddenly appears. For another example, when the device is operating normally, the vibration amplitude is at a low level, and suddenly a strong vibration occurs. These momentary abnormal data are all burst data. The second prediction value can be understood as an estimated value of the future state of the burst data.

[0040] In the disclosed embodiment, the device status data may be one or more, and correspondingly, the burst data in the device status data may also be one or more. For one or more gradual data, a corresponding probability model (such as at least one of Gaussian distribution, Poisson distribution, binomial distribution, etc., not exhaustive) may be used to determine the second prediction value. The purpose of determining the second prediction value is to accurately estimate the burst data and eliminate its influence on other data (such as gradual data, device status data, etc.) other than itself, which is conducive to the accurate estimation of fault prediction.

[0041] In step S104, a fault prediction result is determined based on the first prediction value and the second prediction value.

[0042] In the present disclosure, the fault prediction result can be understood as the predicted total number of equipment failures in the future state (including but not limited to the number of gradual failures, the number of sudden failures, etc.).

[0043] In the disclosed embodiment, the final fault prediction result can be obtained directly according to the prediction values ​​determined by different types, such as the first prediction value, the second prediction value, etc. Specifically, the first prediction value and the second prediction value can be added to obtain the fault prediction result; the first prediction value and the second prediction value can be weighted summed to obtain the fault prediction result; or the fault prediction result can be obtained according to the probability ratio of the first prediction value and the second prediction value, and the specific method is not limited here.

[0044] In summary, the technical solution provided by the present disclosure can classify a variety of equipment status data, and accurately predict different types of data in a targeted manner to obtain the final prediction results. In this way, the influence of special types of fault data (such as sudden fault data, etc.) on part or all of the data can be eliminated, the accuracy of the prediction is improved, and the actual objective needs can be met.

[0045] As mentioned above, the device status data disclosed in the present invention can be used to characterize all relevant data of the device status. Specific device status data will be described below.

[0046] In the present disclosure, device status data includes at least one of the following: device identification, environmental status data and device life portrait data; wherein, the environmental status data includes at least one of the following: region, temperature, humidity and pressure; the device life portrait data includes at least one of the following: network element affiliation, device status, usage intensity, degree of wear and tear, fault time and fault content.

[0047] In the disclosed embodiment, the device status data mainly includes three factors: device identification (ID), environmental status data, and device life portrait data.

[0048] Specifically, the first factor, device identification, is information used to uniquely identify a device, which can be the device's serial number, physical (MAC) address (or a custom coding system, etc.). Device identification helps track, query, and manage specific devices. When a problem occurs, the specific device can be quickly located based on the device identification, and then other status data related to the device can be viewed, providing a solid foundation for determining fault prediction results.

[0049] The second largest factor, environmental status data, can be used to describe the external environmental factors of the device communication. Comprehensive analysis and processing of external environmental factors can help accurately predict the fault results. In the present disclosure, environmental status data can include but is not limited to four dimensions: region, temperature, humidity and pressure. Specifically, region is a fault consideration factor in the device operating environment. Different regions have different climate types. For example, in tropical areas, the temperature is high all year round and the humidity is also high. Equipment under such climatic conditions is more susceptible to high temperature and high humidity. High temperature and high humidity may cause moisture inside the device, causing problems such as circuit short circuit or corrosion (i.e., failure). In this way, regional factors can effectively affect the accuracy of fault prediction results. When determining fault prediction results, regional factors need to be considered. Temperature is a fault consideration factor in the device operating environment. Different devices have different requirements and tolerance ranges for temperature. For example, electronic equipment usually works normally within a certain temperature range. If the temperature is too high, it may cause the performance of electronic components to decline, shorten their lifespan, or even damage (i.e., failure). In this way, temperature factors can effectively affect the accuracy of fault prediction results and are essential when predicting fault prediction results. Humidity is a failure consideration in the equipment operating environment. Different equipment has different requirements and tolerance ranges for humidity. For example, electronic equipment usually works properly within a certain humidity range. In a humid environment, the solder joints on the electronic circuit board may be oxidized, resulting in poor circuit connection (i.e., failure). Therefore, the humidity factor can effectively affect the accuracy of the fault prediction results, and it is essential when predicting the fault prediction results. Pressure is also an important environmental factor for certain specific equipment. For example, in aerospace equipment, the low air pressure in the high-altitude environment can affect the sealing performance, heat dissipation performance, etc. of the equipment (i.e., failure). Therefore, the pressure factor can effectively affect the accuracy of the fault prediction results, and it is essential when predicting the fault prediction results.

[0050] The third factor, the device life portrait data, can be used to describe the internal work or performance of the device communication. Comprehensive analysis and processing of the internal state of the device can help accurately predict the fault result. In the present disclosure, the device life portrait data may include but is not limited to six dimensions: network element ownership, device status, usage intensity, wear degree, fault time, and fault content. Specifically, the network element ownership indicates the ownership of the device in the network. The network element ownership can determine whether the device belongs to a specific network area, subnet or network cluster. For example, in a telecommunications network, the network element ownership of a base station device may point to a specific regional network, which helps to troubleshoot or determine the fault when predicting the fault result. The device status is a general description of the current operating status of the device, which is the main consideration for determining the fault and is indispensable when predicting the fault result. The usage intensity reflects the frequency of the device being used, which is the main consideration for determining the fault and is indispensable when predicting the fault result. The wear degree directly reflects the loss of the device components, which is the main consideration for determining the fault and is indispensable when predicting the fault result. The fault time records the specific time point or time period when the device fails. Accurate failure time data is critical for fault analysis and is the main consideration for determining faults. The fault content describes the specific circumstances when the equipment fails in detail, which can effectively affect the accuracy of fault prediction results.

[0051] Exemplarily, the following table 1 is a multi-dimensional network device failure influencing factor (i.e., the device status data of the present disclosure), as shown below:

[0052]

[0053] Table 1

[0054] As can be seen from Table 1 above, all network equipment failure influencing factors (i.e., the equipment status data disclosed in this disclosure) mainly revolve around three parts: "equipment ID", "X1 environmental factors", and "X2 equipment life portrait". The equipment ID can include network elements and geographic information. The X1 environmental factors mainly represent the external conditions of the communication network, including four dimensions: region, temperature, humidity, and pressure; the X2 equipment life portrait mainly represents the current working status and performance of the network equipment, including six dimensions: network element ownership, equipment status, usage intensity, degree of wear, failure time, and failure content. The network equipment failure influencing factors refer to a total of ten influencing factors. Specifically, the content of each dimension can be customized and identified, and there is no restriction here. For example: the equipment ID can be identified by A0123456, where A represents the network element. 0123456 represents geographic information such as provinces, cities, counties, and districts. When it is used specifically, it can be set with reference to the above table, and it will not be repeated here.

[0055] For example, Figure 2 Flow chart of a complete fault prediction method provided by an embodiment of the present disclosure. Figure 2 As shown:

[0056] The entire fault prediction can be divided into three major steps: fault classification, fault prediction, and prediction output. Specifically, fault prediction can be to establish a multi-dimensional network equipment fault influencing factor system, and then obtain a set of network equipment fault influencing factors (i.e., the device status data disclosed in the present invention), and classify the network equipment fault influencing factor set through a series of steps such as data cleaning, standardization, and input of a Gaussian mixture model (i.e., equivalent to the classification of device status data disclosed in the present invention). Next, it can be determined whether it is a gradual fault to perform fault prediction and obtain a gradual fault prediction value (i.e., the first prediction value disclosed in the present invention). It is also possible to predict a sudden fault and obtain a sudden fault prediction value (i.e., the second prediction value disclosed in the present invention). Finally, based on the gradual fault prediction value and the sudden fault prediction value, the final multi-granularity network fault prediction value (i.e., the fault prediction result disclosed in the present invention) is obtained. The above will be explained separately in conjunction with specific embodiments below.

[0057] As mentioned above, the device status data may be classified. The following will specifically describe possible implementation methods of the classification. Specifically, the method of classifying the device status data includes:

[0058] For any type of device status data, determine a first probability that the device status data belongs to a gradual failure Gaussian model, and determine a second probability that the device status data belongs to a sudden failure Gaussian model;

[0059] The category of the device status data is determined based on the model category corresponding to the higher one of the first probability and the second probability.

[0060] In the present disclosure, the first probability can be understood as the probability of the device status data appearing according to the gradual fault Gaussian model distribution. The second probability can be understood as the probability of the device status data appearing according to the sudden fault Gaussian model distribution. The category of the device status data can be understood to include but not limited to gradual data or sudden data.

[0061] In the disclosed embodiment, when determining the classification type of the equipment status data, a gradual fault Gaussian model and a sudden fault Gaussian model can be used. Among them, the gradual fault Gaussian model and the sudden fault Gaussian model can be collectively referred to as a Gaussian mixture (GMM) model. The GMM model can handle complex data distributions, and the accuracy of the data classification results obtained is high. For example: for gradual fault data, its change trend may not be described by a simple single Gaussian distribution. There may be multiple stages or a mixture of multiple distributions under the influence of different factors, and GMM can accurately fit it by adjusting the parameters of different Gaussian components (such as mean, covariance, etc.). For another example: for sudden fault data, due to the particularity of the data at the moment of the sudden, its distribution may be very different from the normal data distribution. The GMM model can comprehensively consider the special distribution of the sudden fault data and the distribution of the normal operation data as different Gaussian components, so as to better represent the distribution of the entire equipment status data.

[0062] Specifically, any device status data can be input into a trained gradual failure Gaussian model or a sudden failure Gaussian model to obtain the corresponding first probability or second probability. The device status data is divided into gradual failure or sudden failure by the first probability and the second probability. If the first probability is higher than the second probability, the category of the device status data is determined to be a gradual failure, which means that from a probability perspective, the device status data is more consistent with the pattern characteristics of a gradual failure, and may be a manifestation of the device being in a state of slow performance decline and gradually developing towards failure. Conversely, if the second probability is higher than the first probability, the category of the device status data is determined to be a sudden failure, indicating that the device status data is more likely to be related to a sudden failure. The specific model type and classification method are not limited here.

[0063] As mentioned above, any device status data can be input into a trained gradual fault Gaussian model or a sudden fault Gaussian model to obtain the corresponding first probability or second probability. The following will specifically describe possible implementation methods for obtaining the first probability or the second probability:

[0064] Here, the method of obtaining the first probability is used as an example for explanation. The principle of the method of obtaining the second probability is similar and will not be repeated. At this time, the method of obtaining the first probability includes:

[0065] Determine first parameters of the current progressive fault Gaussian model, the first parameters including: a first variance, a first mean, a first covariance, and a first probability density value;

[0066] Determine a second parameter of the device status data in the current progressive fault Gaussian model, the second parameter including a second variance, a second mean, a second covariance, and a second probability density value;

[0067] A ratio of the first parameter to the second parameter is obtained to obtain a first probability.

[0068] In the present disclosure, the first parameter can be understood as the standard parameter of the current progressive fault Gaussian type sample, which can be used as a standard for comparison when performing probability calculation. The second parameter can be understood as the parameter obtained when the device status data is input into the progressive fault Gaussian model, which can be compared with the standard first parameter to obtain the first probability.

[0069] Specifically, the first parameter includes but is not limited to at least one of the following: a first variance, a first mean, a first covariance, and a first probability density value. Among them, the first variance is an important indicator for measuring the degree of dispersion of the Gaussian type sample data of the gradual fault. If the first variance is small, it means that in the ideal gradual fault development process, the data is relatively concentrated near the mean, the fault development is relatively regular, and the probability value of the gradual fault increases; on the contrary, if the first variance is large, it means that the data fluctuates greatly, and there may be multiple factors affecting the gradual development of the fault, and the probability value of the gradual fault decreases. The first mean represents the central trend of the Gaussian type sample data of the gradual fault. The first mean represents the average level of the data under the typical gradual fault development situation. If the mean of the data of the actual device deviates greatly from the first mean, the probability value of the gradual fault decreases. If the mean of the data of the actual device is similar to the first mean, the probability value of the gradual fault increases. When multiple data are used to describe the gradual fault, the first covariance can reflect the linear relationship between these variables. If the first covariance of data A and data B is positive, it means that during the development of a gradual fault, the increase of data A is often accompanied by the increase of data B; if it is negative, it means that the two are in an inverse relationship. The first covariance helps to fully understand the correlation between different factors in the process of gradual faults, so as to more accurately construct the Gaussian model. The first probability density value is calculated based on the probability density function of the Gaussian model of gradual faults, which can represent the relative probability of the occurrence of gradual fault samples. The area with a higher probability density value indicates that during the development of a gradual fault, the data is more likely to be in the state of the gradual fault, which is consistent with the increased probability of gradual faults. The principle of the second parameter is similar and will not be repeated here.

[0070] In the disclosed embodiment, the first parameter of the current gradual fault Gaussian model can be determined, which can be obtained from a sample of the gradual fault Gaussian model. The sample can be obtained based on historical data; it can also be a sample obtained by an expert after precise judgment, which is not limited here. Preferably, it can be a sample obtained by an expert after precise judgment. Then, any device status data is input into the gradual fault Gaussian model to obtain the corresponding second parameter. Using the ratio of the first parameter to the second parameter, the probability (i.e., the first probability) that the device status data belongs to the gradual fault Gaussian model is determined.

[0071] Specifically, the first probability can satisfy the following formula:

[0072]

[0073] in, represents the variance of the current progressive fault Gaussian distribution; Represents the mean of the current progressive fault Gaussian distribution; represents the covariance of the current progressive fault Gaussian distribution; represents the probability density function of the current progressive fault Gaussian distribution; K represents the total amount of data; l represents the lth newly input data; Represents the jth sample data.

[0074] It can be seen from the above formula that by comparing the parameters of the newly input data with the parameters of the sample, the probability of conforming to the model can be obtained.

[0075] Similarly, the second probability can satisfy the following formula:

[0076]

[0077] in, Indicates the variance of the Gaussian distribution of the current sudden fault; Indicates the mean of the Gaussian distribution of the current sudden fault; Represents the covariance of the Gaussian distribution of the current sudden fault; represents the probability density function of the Gaussian distribution of the current sudden fault; K represents the total amount of data; l represents the lth newly input data; Represents the jth sample data.

[0078] It can be seen from the above formula that by comparing the parameters of the newly input data with the parameters of the sample, the probability of conforming to the model can be obtained.

[0079] Exemplarily, the complete classification of the device status data can be determined through the following steps:

[0080] Step 1: Data collection: Collect device ID data, device status, etc. through the resource management system, and obtain environmental operating parameters, device operating status, fault time content and other data through the network operation and maintenance platform. After obtaining relevant data, perform data integration, match the device ID with X1 and X2 to form a network equipment fault information database.

[0081] Step 2: Data preprocessing: Clean and preprocess the collected historical fault information data. Data cleaning: During data acquisition, some equipment lacks operating environment parameters (such as region, temperature, etc.) or operating status (no record of wear degree). Fault data with abnormal or missing indicators are deleted to ensure data quality and accuracy. Data preprocessing: Based on factor rules, the values ​​of various factors of fault information are standardized to support the operation and calculation of subsequent models.

[0082] Step 3: Feature extraction: Based on the network equipment failure influencing factor system, the dummy vector factors are eliminated and six factors including temperature, humidity, pressure, equipment status, usage intensity, and wear degree are selected as feature vectors to distinguish gradual failures from sudden failures.

[0083] Step 4: Build and apply a Gaussian mixture model:

[0084] (1) Set the number of components of the initial Gaussian mixture model to 2 (i.e., the number of classification categories is equal to 2), i.e., K = 2. In order to further guide the clustering results and improve the model operation efficiency, firstly, input 100 data samples after expert judgment, including 50 gradual fault data and 50 sudden fault data, and fit them to form two initial Gaussian models of gradual fault and sudden fault, and calculate the mean, variance, and covariance of the gradual fault data samples and the sudden fault data samples, i.e. , , and , , , as the initial Gaussian distribution parameters of the two clusters.

[0085] (2) Enter new data points , calculate the probability that each data point belongs to the gradual failure Gaussian model and the sudden failure Gaussian model , The closer the data point is to the center of the Gaussian distribution, the greater the probability, that is, the higher the probability of belonging to the cluster. The specific calculation method can be referred to above.

[0086] Step 5: Result output: Based on the classification of all fault data sets by the Gaussian model, output the gradual fault and sudden fault data sets according to the device ID.

[0087] At the same time, the probability density function of the Gaussian mixture model trained based on historical data can also be output as a parameter factor for subsequent fault prediction.

[0088]

[0089]

[0090] Among them, D is the dimension of the data, which is determined based on the number of characteristic factors extracted by the model, namely, temperature, humidity, pressure, equipment status, usage intensity, and wear degree. All belong to D*D matrices, | |、| | are the determinants of the covariance matrices of the Gaussian distributions of gradual and sudden faults, respectively. is the inverse of the covariance matrix of the Gaussian distribution of gradual and sudden faults, T represents the transpose of the matrix, is an exponential function.

[0091] As mentioned above, the device status data obtained by the above method is gradual data or burst data. The following will specifically describe a possible implementation method of obtaining the first prediction value by using burst data. At this time, the method of determining the first prediction value includes:

[0092] The trained neural network model is used to process the gradual data to obtain a first prediction value.

[0093] In the present disclosure, the neural network model includes but is not limited to at least one of the following: Multilayer Perceptron (MLP), Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU), Back Propagation Neural Network (BP), etc. These neural network models have their own advantages in processing different types of data and problems, and can be flexibly adjusted according to actual needs and user needs, and are not limited here. Preferably, the neural network model of the present disclosure can be a BP neural network. Because the BP neural network can learn and approximate any complex nonlinear relationship. And for gradual data, its change trend is often not a simple linear relationship, and may be affected by a combination of multiple factors. In this way, using the BP neural network to process gradual data can automatically capture these complex nonlinear features, thereby more accurately predicting the future development trend of the data with high accuracy. At the same time, the BP neural network can process multiple input data at the same time, and the calculations between each neuron are performed independently. This enables it to quickly process a large amount of gradual data and improve the efficiency of prediction.

[0094] Specifically, the gradual data can be directly input into a trained neural network model (such as a BP neural network) to obtain a predicted value of the gradual data (i.e., the first predicted value of the present disclosure). The first predicted value can be understood as an estimated fault value of the future state of the gradual data.

[0095] It should be noted that the neural network model of the present invention can be obtained by iteratively training the input data based on the target error function; wherein the target error function can be a fusion function of the smoothing loss function and the cosine similarity distance function.

[0096] In the present disclosure, the target error function can be understood as a mathematical expression for measuring the difference between the prediction result of the neural network model and the actual result. Its function is to guide the neural network to continuously adjust the weights and biases during the training process to minimize the prediction error, thereby improving the accuracy and generalization ability of the model. The smooth loss function can make the model pay more attention to the overall trend and continuity of the data during the training process to avoid overfitting. It usually has a certain robustness to outliers and noise, and can make the model learn the characteristics of the data more stably. The cosine similarity distance function focuses on measuring the directional similarity between two vectors. In the neural network, it can help the model better learn the feature space distribution of the input data and improve the model's recognition and distinction ability for similar data. In this way, by integrating the smooth loss function and the cosine similarity distance function, the target error function can comprehensively consider multiple factors such as the overall trend, continuity and feature space distribution of the data, thereby more comprehensively measuring the prediction error of the model. This can make the neural network learn the characteristics of the data more effectively during the training process and improve the accuracy and generalization ability of the model.

[0097] Exemplarily, the objective function of the present disclosure may satisfy the following formula:

[0098]

[0099]

[0100] in, represents the cosine similarity distance function loss;

[0101] represents the smooth loss function loss;

[0102] It represents the estimated value of network equipment failure, that is, the estimated value obtained by inputting data into the model;

[0103] Indicates the true fault value of the network device, that is, the standard comparison value;

[0104] represents the weight factor;

[0105] Indicates the threshold value that controls the smoothing range.

[0106] As shown in the above formula, it can be seen that the smooth loss function is brought into the cosine similarity distance function for fusion to obtain the objective function. Input data (such as gradual data) is input into the neural network model until the input data satisfies the above objective function, and the predicted output result is determined.

[0107] The present disclosure also provides a complete neural network model training process and a process for obtaining a prediction value of gradual data (i.e., the first prediction value of the present disclosure). The following is a specific description in conjunction with an exemplary embodiment:

[0108] For example, the following is the complete process of training a neural network:

[0109] Basic principle: Build and train a BP neural network to obtain a usable network equipment failure prediction model. The topological structure of the BP neural network consists of three layers of neural networks: input layer, hidden layer, and output layer. Each layer contains several interrelated processing units. The input layer is the preprocessed data or feature values ​​input into the neural network model, that is, the factors affecting network equipment failure, including environmental factors and equipment life portraits; the output layer is the predicted value of the model output, that is, the number of network equipment failures in the next year; the neurons in the hidden layer perform nonlinear transformations on the input data through activation functions, and continuously transmit them to the next network layer.

[0110] The information flow of the BP neural network is divided into feedforward prediction and feedback learning. In the feedforward prediction process, the information flow is transmitted forward in the order of input layer-hidden layer-output layer. The feature value is input from the input layer, transmitted forward along the neural network, multiplied and summed with the weights between the input unit and the hidden layer neurons, converted and processed by the activation function, and continued to be transmitted into the next layer of the network, and finally the calculation result is output. In the feedback learning process, the information flow is fed back and propagated in the order of output layer-hidden layer-input layer, and the error between the predicted value and the true value is calculated using the improved loss function. If the error value is higher than expected, the error signal is returned along the original neuron to adjust and update the weights and thresholds of each layer of the network. This cycle is repeated until the error value meets expectations, the network training is completed, and a usable network equipment fault prediction model is obtained.

[0111] The specific training steps are as follows:

[0112] Step 1: Data standardization. The data can be standardized using the standard score (Z-score) method. Specifically, considering the different ranges of variation of each numerical variable, in order to improve the accuracy of network training and prediction and maximize the fault characteristics, Z-score is used for standardization. The formula can be:

[0113]

[0114] Where X represents the standardized network equipment failure factor value;

[0115] Indicates the original network device failure factor value;

[0116] Represents the mean value of network equipment failure factor;

[0117] Represents the standard deviation of the network failure factor.

[0118] The above formula is used to standardize the data, and at the same time, some nominal factors, such as region and network element ownership, can be processed as dummy variables, that is, the redundant dummy vectors in the vector factor are removed to obtain standard vector data. For example, for the region, the mountain area can be directly set to [1,0,0], the hills to [0,1,0], and the plains to [0,0,1].

[0119] Step 2: Construct BP neural network and initialize the weights and thresholds of the network layer. Construct BP neural network with 2 hidden layers. Initialize the weights between the input layer and the hidden layer, and between the hidden layers as , , the weights of the hidden layer and the output layer are , initialize the threshold between the input layer and the hidden layer, and the hidden layer as , , the weights of the hidden layer and the output layer are .

[0120] Step 3: Enter the sample set and select The function performs weighted summation and activation of each hidden layer and outputs the calculation result. function, increasing the nonlinear expression ability of the model. Compared with other activation functions, such as function, The function has a value range of [-1,1], which can alleviate the gradient disappearance problem. The mathematical expression of the function is: , where e is the base of the natural logarithm, approximately equal to 2.71828. The specific formula is as follows:

[0121] 1. Input layer: is the n training set input of the neural network, expressed as a vector form, namely .

[0122] 2. Weighted summation and activation of the first hidden layer: The neurons in the first hidden layer complete the weighted summation of the input signal, add a threshold device, and then pass the activation function The output obtained by nonlinear transformation is .

[0123] in, Represents the weights of the input layer and the hidden layer;

[0124] Represents the threshold of the input layer and the hidden layer.

[0125] 3. Weighted summation and activation of the second hidden layer: The neurons in the second hidden layer complete the weighted summation of the input signal, add a threshold device, and then perform a nonlinear transformation through the activation function tanh2 to obtain the output: .

[0126] in, Represents the weights of the first and second hidden layers and the second hidden layer;

[0127] Represents the threshold between the first and second hidden layers.

[0128] 4. Output layer: The final output of the output layer neurons is calculated by weighted summation and activation function. y represents the number of failures classified by device ID. The activation function of the output layer is selected Function, that is .

[0129] Where v represents the weights of the second hidden layer and the output layer;

[0130] C represents the threshold of the second hidden layer and the output layer.

[0131] Step 4: Correct the weights and thresholds of each layer according to the cosine-smoothed improved error function. The improved error function can be understood as the objective function mentioned above. The specific formula can be found above. Based on the output and feedback errors of the network layer, the improved error function is used for evaluation. If the set requirements are met, the training phase is completed; if not, the improved error function and weights are updated and the training is iterated.

[0132] Step 5: Loop feedback until the error value meets expectations, complete network training, and form a BP neural network fault prediction model. After multiple iterations, the improved BP neural network has a smaller weight and is smoother than the traditional network response, greatly reducing overfitting.

[0133] Exemplarily, the following is a complete process of obtaining the predicted value of the gradual data (i.e., the first predicted value of the present disclosure) using the above-trained neural network model:

[0134] Step 1: Collect data and perform standardization. This is consistent with the process of obtaining network equipment fault samples. Collect real-time equipment status information according to the equipment ID. Equipment ID data, equipment status, etc. are obtained through the resource management system. Environmental operating parameters, equipment operating status, fault time and other data are obtained through the network operation and maintenance platform. The fault content factor adopts the manual assignment method, and r dummy vector values ​​are assigned in turn. Each equipment ID has r original equipment information sets. Consistent with the training phase method, the Z-score method is used for standardization to obtain the equipment information set related to the fault.

[0135] Step 2: Calculate the predicted value y1 of the number of failures of gradual failures. is the n actual value input of the neural network, expressed as a vector form, that is .

[0136] The output function is:

[0137]

[0138] Among them, sigmoid and tanh represent neural network activation functions;

[0139] v and W represent the weights between neurons in the neural network;

[0140] R and c represent the thresholds between neural network layers.

[0141] As described above, the process of obtaining the first prediction value based on the gradual data is described in detail. The following is a possible implementation method of obtaining the second prediction value for the burst data. At this time, the method of obtaining the second prediction value includes:

[0142] Based on the third probability density value, determining a third probability; the third probability density value is obtained based on the probability density function of the Gaussian model of the sudden failure;

[0143] Obtaining a third prediction value of the burst data;

[0144] Based on the third probability and the third predicted value, a second predicted value is determined.

[0145] In the present disclosure, the third probability density value can be understood as the probability density of the burst data using the Gaussian model of the burst fault. The probability density of the Gaussian model of the burst fault can satisfy the following formula:

[0146]

[0147] Among them, D is the dimension of the data, which is determined based on the number of characteristic factors extracted by the model, namely, temperature, humidity, pressure, equipment status, usage intensity, and wear degree. All are D*D matrices, | | is the determinant of the covariance matrix of Gaussian distribution of sudden failures, is the inverse of the Gaussian distribution covariance matrix of sudden faults, T represents the transpose of the matrix, is an exponential function.

[0148] It is worth noting that the third prediction value can be understood as the estimated value obtained by multiplexing burst data with gradual data into the neural network model. Then, the second estimated value is determined by directly multiplying the probability of the above-mentioned burst failure Gaussian with the third estimated value.

[0149] Exemplarily, the process of obtaining the second estimated value can be understood as follows:

[0150] The number of sudden failures is equal to the total number multiplied by the probability value, and the calculation function is derived as follows: +c.

[0151] Where p is the sum of the probability values ​​of sudden failures of the device ID and the fault content, and the p value can be the p value of the above formula. Sigmoid and tanh represent neural network activation functions; v and W represent the weights between neurons in the neural network; R and c represent the thresholds between neural network layers.

[0152] As mentioned above, the method for determining the first prediction value and the second prediction value is specifically explained. Below, it will be specifically explained how to use the first prediction value and the second prediction value to determine the final fault prediction result. At this time, the method includes: adding the corresponding first prediction value and the second prediction value to obtain the fault prediction result.

[0153] In the disclosed embodiment, the first prediction value (i.e., the gradual prediction result) and the second prediction value (the sudden prediction result) obtained above can be directly added to obtain the final fault prediction result. Exemplarily, the fault prediction result can satisfy the following formula:

[0154]

[0155] Among them, y1 represents the predicted value of the number of gradual failures of the equipment; y2 represents the predicted value of the number of sudden failures of the equipment; p represents the total probability of sudden failures of the equipment ID in the fault content; sigmoid and tanh represent the activation functions of the neural network; v and W represent the weights between neurons of the neural network; R and c represent the thresholds between neural network layers.

[0156] The total prediction times of all fault contents are counted according to the device ID. The expression is:

[0157] , where r is the number of fault content types.

[0158] It should be noted that in any of the above embodiments, the device status data can be preprocessed before the predicted value is determined. The processing disclosed herein includes but is not limited to at least one of the following: data cleaning, data normalization, and feature extraction. Data cleaning can improve data quality and the accuracy of money-making data analysis. Data normalization can eliminate the influence of other factors and accelerate the convergence of the model. Feature extraction can reduce data dimensions and extract key information. The present disclosure does not impose any special restrictions on the specific means of data cleaning, data normalization, and feature extraction.

[0159] For example, Figure 3 A complete flow chart of the network equipment fault prediction model based on BP neural network provided in the embodiment of the present disclosure:

[0160] like Figure 3 As shown: It can be determined whether the equipment status data is a gradual fault. When it is determined to be a gradual fault, the data can be standardized and then input into the BP neural network model. The function performs weighted summation and activation of each hidden layer, and uses the improved error function to calculate the error. When the error meets the requirements, the network training ends, the actual value is input, and the predicted value y1 is output. When the error does not meet the requirements, the weights and thresholds of each layer of the network can be adjusted and updated until the error meets the requirements. When it is determined that it is not a sudden fault, the fault prediction value can be directly determined according to the device ID. That is, y1 is used to calculate the number of faults y2. And the predicted value y is output by adding y1 and y2 to obtain the final fault prediction result.

[0161] The present disclosure also provides a network operation and maintenance device, which may include:

[0162] Gaussian mixture GMM model, used to determine the category of equipment status data;

[0163] The neural network model is used to predict the classified equipment status data and obtain fault prediction results.

[0164] In the present disclosure, the GMM model at least includes: a gradual fault Gaussian model and a sudden fault Gaussian model.

[0165] For example, Figure 4 A schematic diagram of a complete network operation and maintenance device provided in an embodiment of the present disclosure, such as Figure 4 As shown:

[0166] The self-smart network is the result of China's huge and growing market demand as well as its leading network and technology drive, providing digital experiences such as "zero waiting", "zero contact", "zero failure" and "zero risk" to consumers and vertical industry users.

[0167] The network intelligent operation and maintenance device is a technical solution that applies fault classification methods and prediction models to self-intelligent networks to improve the intelligence of network operation and maintenance.

[0168] The exclusive big model for intelligent operation and maintenance of network equipment is the basic resource layer of the network intelligent operation and maintenance device. It includes the network equipment fault classification and fault prediction model in the proposal. On this basis, it combines the internal big language to provide technical functions such as big model data training and machine learning.

[0169] As the data management layer of network intelligent operation and maintenance, the network equipment resource visualization system mainly includes three sub-devices: equipment archives, fault warning center, and operation and maintenance knowledge Q&A.

[0170] (1) Equipment files: Label the equipment environment, life status, operating status and other information to achieve “one device, one file” management for network equipment.

[0171] (2) Fault Warning Center: Based on the multi-granularity fault prediction data output by Device 2, fault warning is carried out in the early stage of equipment degradation according to network elements, regions, fault types, and fault occurrence times, and the faulty equipment and fault causes are identified to carry out preliminary treatment.

[0172] (3) Operation and maintenance knowledge question and answer: Establish a knowledge sharing platform to realize the knowledge of fault prediction data, work order handling results, and alarm handling plans, so as to achieve the best and fastest handling of network faults and improve operation and maintenance efficiency.

[0173] As the network intelligent operation and maintenance service support layer, the network equipment fault integration center realizes the intelligent operation and maintenance goals of active fault perception, fault location and demarcation through fault discovery, fault location, fault identification, fault handling and fault reporting. The fault integration center further feeds back the fault handling data to the network equipment resource visualization system for real-time dynamic update.

[0174] For example, Figure 5 A complete network device fault prediction device and a self-intelligent network-network intelligent operation and maintenance system for a network device fault prediction device provided by an embodiment of the present disclosure, such as Figure 5 As shown:

[0175] Specifically, it includes firstly building a multi-dimensional network equipment failure influencing factor system around the three major factors of "equipment ID, environmental factors, and equipment life status", clustering training of network equipment failures based on the Gaussian mixture model algorithm, and obtaining the probability distribution function information of gradual and sudden network equipment failures; secondly, building a network equipment failure data prediction model based on BP neural network, and taking the Gaussian mixture model fault classification results as an important input for prediction. Among them, an improved error function based on Huber function and cosine similarity distance function is proposed to predict the probability of gradual network equipment failure and output multi-granularity network equipment failure prediction numbers; finally, building a network intelligent operation and maintenance device under a self-intelligent network system, taking the fault classification method and prediction model as the important technical support of the network intelligent operation and maintenance system, providing intelligent means for network operation, and providing intelligent capabilities for network equipment resource management and fault closed-loop tracking.

[0176] The present disclosure also provides a fault prediction device. Figure 6 A structural block diagram of a fault prediction device provided by an embodiment of the present disclosure, such as Figure 6 As shown, the fault prediction device 600 includes:

[0177] A classification unit 601, used to classify device status data;

[0178] A first determining unit 602, configured to determine a first prediction value for progressive data in the device status data;

[0179] A second determining unit 603, configured to determine a second prediction value for the burst data in the device status data;

[0180] The third determining unit 604 is configured to determine a fault prediction result based on the first prediction value and the second prediction value.

[0181] In an exemplary embodiment, classifying device status data includes: for any type of device status data, determining a first probability that the device status data belongs to a gradual failure Gaussian model, and determining a second probability that the device status data belongs to a sudden failure Gaussian model; and determining the category of the device status data based on the model category corresponding to the higher of the first probability and the second probability.

[0182] In an exemplary embodiment, for any type of device status data, determining a first probability that the device status data belongs to a gradual fault Gaussian model includes: determining a first parameter of the current gradual fault Gaussian model, the first parameter including: a first variance, a first mean, a first covariance, and a first probability density value; determining a second parameter of the device status data in the current gradual fault Gaussian model, the second parameter including a second variance, a second mean, a second covariance, and a second probability density value; obtaining a ratio of the first parameter to the second parameter to obtain a first probability.

[0183] In an exemplary embodiment, determining a first prediction value for the progressive data in the device status data includes: processing the progressive data using a trained neural network model to obtain the first prediction value.

[0184] In an exemplary embodiment, the neural network model is obtained by iteratively training the input data based on the target error function; wherein the target error function is a fusion function of the smoothing loss function and the cosine similarity distance function.

[0185] In an exemplary embodiment, a second prediction value is determined for burst data in device status data, including: determining a third probability based on a third probability density value; the third probability density value is obtained by processing the probability density function of a Gaussian model of a burst failure; obtaining a third prediction value for the burst data; and determining a second prediction value based on the third probability and the third prediction value.

[0186] In an exemplary embodiment, determining a fault prediction result based on a first prediction value and a second prediction value includes: adding the corresponding first prediction value and the second prediction value to obtain a fault prediction result.

[0187] In an exemplary embodiment, before determining the first prediction value for the first data in the device status data, the method further includes: preprocessing the device status data; wherein the preprocessing includes at least one of the following: data cleaning, data normalization, and feature extraction.

[0188] Figure 7 A hardware block diagram of an electronic device provided in an embodiment of the present disclosure. The electronic device 700 according to an embodiment of the present disclosure includes at least a processor; and a memory for storing computer-readable instructions. When the computer-readable instructions are loaded and executed by the processor, the processor executes the fault prediction method described in any of the above embodiments of the present disclosure.

[0189] Figure 7The electronic device 700 shown specifically includes: a central processing unit (CPU) 701, a graphics processing unit (GPU) 702 and a memory 703. These units are connected to each other via a bus 704. The central processing unit (CPU) 701 and / or the graphics processing unit (GPU) 702 can be used as the above-mentioned processor, and the memory 703 can be used as the above-mentioned memory for storing computer-readable instructions. In addition, the electronic device 700 may also include a communication unit 705, a storage unit 706, an output unit 707, an input unit 708 and an external device 709, which are also connected to the bus 704.

[0190] Figure 8 A schematic diagram of a computer-readable storage medium provided in an embodiment of the present disclosure. Figure 8 As shown, a computer-readable storage medium 800 according to an embodiment of the present disclosure has computer-readable instructions 801 stored thereon. When the computer-readable instructions 801 are executed by a processor, the fault prediction method described with reference to the above figures according to any of the foregoing embodiments of the present disclosure is executed. The computer-readable storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, optical disk, magnetic disk, etc.

[0191] The present disclosure further provides a computer program product, including a computer program, which, when executed by a processor, implements the fault prediction method described in any of the foregoing embodiments of the present disclosure.

[0192] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this disclosure.

[0193] The basic principles of the present disclosure are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present disclosure are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present disclosure. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, and are not limitations. The above details do not limit the present disclosure to the necessity of adopting the above specific details to be implemented.

[0194] The block diagrams of the devices, apparatuses, equipment, and systems involved in this disclosure are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagrams. As will be appreciated by those skilled in the art, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including," "comprising," "having," and the like are open words, referring to "including but not limited to," and can be used interchangeably therewith. The words "or" and "and" used herein refer to the words "and / or," and can be used interchangeably therewith, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to," and can be used interchangeably therewith.

[0195] Additionally, as used herein, "or" used in a list of items beginning with "at least one" indicates a separate list, so that, for example, a list of "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not mean that the example described is preferred or better than other examples.

[0196] It should also be noted that in the system and method of the present disclosure, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present disclosure.

[0197] Various changes, substitutions, and modifications of the techniques described herein may be made without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of the present disclosure is not limited to the specific aspects of the processes, machines, manufactures, compositions of events, means, methods, and actions described above. Currently existing or later to be developed processes, machines, manufactures, compositions of events, means, methods, or actions that perform substantially the same functions or achieve substantially the same results as the corresponding aspects described herein may be utilized. Thus, the appended claims include such processes, machines, manufactures, compositions of events, means, methods, or actions within their scope.

[0198] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present disclosure. Therefore, the present disclosure is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0199] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present disclosure to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A fault prediction method, characterized in that: The method comprises: Categorize device status data; Determine a first prediction value for the progressive data in the device status data; Determine a second prediction value for the burst data in the device status data; Determining a fault prediction result based on the first prediction value and the second prediction value; The classifying of the device status data includes: For any type of device status data, determine a first probability that the device status data belongs to a gradual failure Gaussian model, and determine a second probability that the device status data belongs to a sudden failure Gaussian model; Determining a category of the device status data based on a model category corresponding to a higher one of the first probability and the second probability; Wherein, for any type of device status data, determining a first probability that the device status data belongs to a progressive failure Gaussian model includes: Determine first parameters of the current progressive fault Gaussian model, the first parameters including: a first variance, a first mean, a first covariance, and a first probability density value; Determine second parameters of the device status data in the current progressive fault Gaussian model, the second parameters including a second variance, a second mean, a second covariance, and a second probability density value; The ratio of the first parameter to the second parameter is obtained to obtain the first probability.

2. The method according to claim 1, characterized in that The step of determining a first prediction value for the progressive data in the device status data includes: The gradual data is processed using the trained neural network model to obtain the first predicted value.

3. The method according to claim 2, characterized in that The neural network model is obtained by iteratively training the input data based on the target error function; The target error function is a fusion function of a smoothing loss function and a cosine similarity distance function.

4. The method according to claim 1, characterized in that The step of determining a second prediction value for the burst data in the device status data includes: Determine a third probability based on a third probability density value; the third probability density value is obtained based on a probability density function of a Gaussian model of a sudden failure; Obtaining a third predicted value of the burst data; Based on the third probability and the third predicted value, the second predicted value is determined.

5. The method according to claim 1, characterized in that The determining a fault prediction result based on the first prediction value and the second prediction value includes: The corresponding first prediction value and the second prediction value are added to obtain the fault prediction result.

6. The method according to claim 1, characterized in that Before determining the first prediction value for the first data in the device status data, the method further includes: Preprocessing the device status data; Wherein, the preprocessing includes at least one of the following: data cleaning, data normalization, and feature extraction.

7. A network operation and maintenance device, characterized in that: include: Gaussian mixture GMM model, used to determine the category of equipment status data; The neural network model is used to predict the classified equipment status data and obtain the fault prediction results; Wherein, the GMM model at least includes: a gradual fault Gaussian model and a sudden fault Gaussian model; Wherein, for any type of device status data, the gradual fault Gaussian model is used to determine the first probability that the device status data belongs to the gradual fault Gaussian model, and the sudden fault Gaussian model is used to determine the second probability that the device status data belongs to the sudden fault Gaussian model; the first probability and the second probability are used to determine the category of the device status data; The progressive fault Gaussian model is used to determine a first parameter and a second parameter, and the first parameter and the second parameter are used to determine the first probability.

8. A fault prediction device, characterized in that: The device comprises: A classification unit, used for classifying device status data; A first determining unit, configured to determine a first prediction value for the progressive data in the device status data; A second determining unit, configured to determine a second prediction value for the burst data in the device status data; A third determining unit, configured to determine a fault prediction result based on the first prediction value and the second prediction value; Wherein, the classification unit is specifically used for: For any type of device status data, determine a first probability that the device status data belongs to a gradual failure Gaussian model, and determine a second probability that the device status data belongs to a sudden failure Gaussian model; Determining a category of the device status data based on a model category corresponding to a higher one of the first probability and the second probability; Wherein, when the classification unit is used to determine the first probability that the device status data belongs to the gradual fault Gaussian model, it is specifically used to: Determine first parameters of the current progressive fault Gaussian model, the first parameters including: a first variance, a first mean, a first covariance, and a first probability density value; Determine a second parameter of the device status data in the current progressive fault Gaussian model, the second parameter comprising a second variance, a second mean, a second covariance, and a second probability density value; The ratio of the first parameter to the second parameter is obtained to obtain the first probability.

9. An electronic device, characterized in that: include: a memory for storing computer readable instructions; as well as A processor is used to run the computer-readable instructions so that the electronic device executes the method according to any one of claims 1 to 6.

10. A non-transitory computer-readable storage medium for storing computer-readable instructions, characterized in that: When the computer-readable instructions are executed by a processor, the processor is caused to perform the method according to any one of claims 1 to 6.

11. A computer program product, characterized in that The method comprises a computer program, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images