A terminal health detection method, device, apparatus and storage medium

Abstract

The embodiment of the present application provides a terminal health detection method, equipment, device and storage medium, which relates to the field of data processing technology. The above method includes: obtaining terminal information representing the terminal operation status; inputting the terminal information into a pre-trained terminal health detection model for encoding and classification to obtain an output result, the output result represents: the probability that the terminal belongs to a dormant state without operation and belongs to different health levels during operation, and the health level represents the network access quality of the terminal during operation; according to the output result, determining whether the terminal is in a dormant state without operation, or the health level of the terminal during operation. Applying the solution provided by the embodiment of the present application can improve the accuracy of the terminal health detection result.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a terminal health detection method, equipment, device and storage medium. Background Art

[0002] In order to determine the network experience of the terminal in the network, it is necessary to detect the health of the terminal. In the related technology, after the operation and maintenance management platform receives the operation data reported by the terminal, it deducts the health score of the terminal based on the initial value of the health score of the terminal according to the fixed threshold manually set by the expert to obtain the health score of the terminal. Experts need to manually set fixed thresholds for each data in the operation data, as well as the deduction values ​​corresponding to different fixed thresholds. For example, for data A, the fixed threshold is set to be no deduction if the value of data A is greater than 10, 5 points are deducted if the value is 5-10, and 10 points are deducted if the value is less than 5.

[0003] After calculating the health score, the operation and maintenance management platform determines the health of the terminal based on the health score of the terminal. For example, a health score of 80-100 determines that the health of the terminal is excellent, a health score of 60-80 determines that the health of the terminal is good, a health score of 40-60 determines that the health of the terminal is average, and a health score below 40 determines that the health of the terminal is poor.

[0004] That is, in this process, the health score of the terminal is calculated based on the fixed threshold set by the expert, and then the health level of the terminal is determined. However, this process relies heavily on the experience of the expert, and the fixed threshold value is fixed and cannot be flexibly adjusted according to the network environment. Therefore, the health score calculated based on the above process is inaccurate, which in turn leads to inaccurate health level of the terminal detected. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide a terminal health detection method, device, apparatus and storage medium to improve the accuracy of terminal health detection. The specific technical solution is as follows:

[0006] In a first aspect, an embodiment of the present application provides a terminal health detection method, the method comprising:

[0007] Obtain terminal information indicating terminal operation status;

[0008] Inputting the terminal information into a pre-trained terminal health degree detection model for encoding and classification to obtain an output result, wherein the output result indicates: the probability of the terminal being in a dormant state and being in different health degrees during operation, wherein the health degree indicates the network access quality of the terminal during operation;

[0009] It is determined according to the output result that the terminal is in a dormant state or the health status of the terminal during operation.

[0010] In one embodiment of the present application, the terminal health detection model includes an encoding sub-model for encoding terminal information and a classification sub-model for generating output results.

[0011] In one embodiment of the present application, the encoding sub-model is trained in the following manner:

[0012] Inputting the first sample terminal information into the basic coding sub-model for encoding to obtain a first sample terminal information vector;

[0013] Inputting the first sample terminal information vector into a basic decoding sub-model for decoding to obtain a sample decoding result, wherein the size of the sample decoding result is the same as the size of the first sample terminal information;

[0014] Calculating a first loss based on the first sample terminal information and the sample decoding result;

[0015] Based on the first loss, the parameters of the basic encoding sub-model and the basic decoding sub-model are adjusted, and the process returns to the step of inputting the first sample terminal information into the basic encoding sub-model for encoding to obtain the first sample terminal information vector and continues to execute until the first preset training termination condition is reached to obtain the trained encoding sub-model.

[0016] In one embodiment of the present application, the step of inputting the first sample terminal information into a basic coding sub-model for encoding to obtain a first sample terminal information vector includes:

[0017] Based on the basic coding sub-model, converting the first sample terminal information into a one-dimensional vector of a first length;

[0018] Based on the basic coding sub-model, dividing the one-dimensional vector into a target mean and a target variance, wherein the lengths of the target mean and the target variance are both half of the first length;

[0019] Based on the basic coding sub-model, a first sample terminal information vector of a preset length is obtained by sampling random numbers that satisfy a normal distribution, wherein the mean of the random numbers is the target mean, and the variance of the random numbers is the target variance.

[0020] In one embodiment of the present application, the classification sub-model is trained in the following manner:

[0021] Obtaining a second sample terminal information vector obtained by encoding the second sample terminal information through the encoding sub-model;

[0022] Inputting the second sample terminal information vector into the basic classification sub-model for classification to obtain a sample output result;

[0023] Calculate a second loss based on the sample output result and the sample label corresponding to the second sample terminal information, wherein the sample label corresponds to the second sample terminal information one-to-one, indicating that the terminal corresponding to the corresponding second sample terminal information is in a dormant state or is in a healthy state during operation;

[0024] The parameters of the basic classification sub-model are adjusted based on the second loss, and the step of returning to the step of obtaining the second sample terminal information vector obtained after encoding the second sample terminal information through the encoding sub-model is continued until the second preset training termination condition is reached to obtain the trained classification sub-model.

[0025] In one embodiment of the present application, the terminal information includes at least one of the following parameters of the terminal within a preset time period: delay, packet loss rate, uplink traffic, downlink traffic, uplink speed selection, downlink speed selection, uplink retransmission rate, and downlink retransmission rate.

[0026] In a second aspect, an embodiment of the present application provides a terminal health detection device, the device comprising:

[0027] processor;

[0028] Transceiver;

[0029] A machine-readable storage medium storing machine-executable instructions that can be executed by the processor; the machine-executable instructions prompt the processor to execute any one of the method steps in the first aspect.

[0030] In a third aspect, an embodiment of the present application provides a terminal health detection device, the device comprising:

[0031] An information acquisition module, used to acquire terminal information indicating terminal operation status;

[0032] A result output module is used to input the terminal information into a pre-trained terminal health detection model for encoding and classification to obtain an output result, wherein the output result indicates: the probability of the terminal being in a dormant state and being in different health levels during operation, and the health level indicates the network access quality of the terminal during operation;

[0033] The health determination module is used to determine whether the terminal is in a dormant state or the health status of the terminal during operation according to the output result.

[0034] In one embodiment of the present application, the terminal health detection model includes an encoding sub-model for encoding terminal information and a classification sub-model for generating output results.

[0035] In one embodiment of the present application, the encoding sub-model is obtained by training the following modules:

[0036] A first encoding module, used for inputting the first sample terminal information into the basic encoding sub-model for encoding to obtain a first sample terminal information vector;

[0037] a decoding module, configured to input the first sample terminal information vector into a basic decoding sub-model for decoding, to obtain a sample decoding result, wherein the size of the sample decoding result is the same as the size of the first sample terminal information;

[0038] A first loss calculation module, used for calculating a first loss based on the first sample terminal information and the sample decoding result;

[0039] The first parameter adjustment module is used to adjust the parameters of the basic encoding sub-model and the basic decoding sub-model based on the first loss, and return to continue execution from the first encoding module until a first preset training termination condition is reached to obtain a trained encoding sub-model.

[0040] In one embodiment of the present application, the first encoding module is specifically used to:

[0041] Based on the basic coding sub-model, converting the first sample terminal information into a one-dimensional vector of a first length;

[0042] Based on the basic coding sub-model, dividing the one-dimensional vector into a target mean and a target variance, wherein the lengths of the target mean and the target variance are both half of the first length;

[0043] Based on the basic coding sub-model, a first sample terminal information vector of a preset length is obtained by sampling from random numbers that satisfy a normal distribution, wherein the mean of the random numbers is the target mean, and the variance of the random numbers is the target variance.

[0044] In one embodiment of the present application, the classification sub-model is obtained by training the following modules:

[0045] A second encoding module, used to obtain a second sample terminal information vector obtained by encoding the second sample terminal information through the encoding sub-model;

[0046] A classification module, used for inputting the second sample terminal information vector into a basic classification sub-model for classification to obtain a sample output result;

[0047] A second loss calculation module, configured to calculate a second loss based on the sample output result and a sample label corresponding to the second sample terminal information, wherein the sample label corresponds to the second sample terminal information one-to-one, indicating that the terminal corresponding to the corresponding second sample terminal information is in a dormant state without operation, or the health of the terminal during operation;

[0048] The second parameter adjustment module is used to adjust the parameters of the basic classification sub-model based on the second loss, and return to continue execution from the second encoding module until a second preset training termination condition is reached to obtain a trained classification sub-model.

[0049] In one embodiment of the present application, the terminal information includes at least one of the following parameters of the terminal within a preset time period: delay, packet loss rate, uplink traffic, downlink traffic, uplink speed selection, downlink speed selection, uplink retransmission rate, and downlink retransmission rate.

[0050] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, any method step described in the first aspect is implemented.

[0051] In a fifth aspect, an embodiment of the present application further provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute any of the methods described in the first aspect above.

[0052] Beneficial effects of the embodiments of the present application:

[0053] In the embodiment of the present application, the terminal information of the terminal can be analyzed and processed by the terminal health detection model to obtain the output result, without the need to calculate the health score based on a fixed threshold manually set by an expert. The terminal health detection model obtained by training based on a large amount of data can learn the features in the data, perform accurate terminal health analysis, and save labor costs. And the output result can not only analyze the health of the terminal, but also distinguish whether the terminal is in an unused dormant state, which can avoid the terminal being defaulted to a used state during the terminal health program detection process, and the health of the terminal in a dormant state is mistakenly detected as poor health, thereby avoiding the problem of incorrect terminal health evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application, and a person skilled in the art can also obtain other drawings based on these drawings.

[0055] Figure 1 A schematic diagram of the first network topology provided in the embodiment of the present application;

[0056] Figure 2 A schematic diagram of a second network topology provided in an embodiment of the present application;

[0057] Figure 3 A flowchart of a terminal health detection method provided in an embodiment of the present application;

[0058] Figure 4 A schematic diagram of the structure of a terminal health detection model provided in an embodiment of the present application;

[0059] Figure 5 A schematic diagram of the first encoding sub-model training process provided in an embodiment of the present application;

[0060] Figure 6 A schematic diagram of the second encoding sub-model training process provided in an embodiment of the present application;

[0061] Figure 7 A schematic diagram of a classification sub-model training process provided in an embodiment of the present application;

[0062] Figure 8 A schematic diagram of a model effect provided in an embodiment of the present application;

[0063] Fig. 9 A schematic diagram of the structure of a terminal health detection device provided in an embodiment of the present application;

[0064] Fig.10 A schematic diagram of the structure of a terminal health detection device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0065] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field based on the present application belong to the scope of protection of the present application.

[0066] Because in the related art, the terminal health detection relies on the fixed threshold set manually by experts, this process is heavily dependent on the experience of experts, and the value of the fixed threshold cannot be flexibly adjusted according to the network environment. Furthermore, the terminal is often not in a state of being used continuously. When the terminal is in an unused dormant state, terminal information may also be generated, and in this case, the value of some data in the terminal information may be within the fixed threshold set manually by experts that requires deduction, resulting in the terminal health being detected as average or poor. But in fact, the terminal is not in operation at all at this time. Therefore, when the terminal is not running, the health level is judged to be at a low level, resulting in serious errors in the detection results of the terminal health level.

[0067] In order to solve the above problems, the embodiments of the present application provide a terminal health detection method, device, apparatus and storage medium.

[0068] First, the network topology of the application scenario of this application is explained.

[0069] See also Figure 1 , which is a schematic diagram of the first network topology provided in an embodiment of the present application.

[0070] The figure includes the detected terminal, the access point (AP) that provides wireless signals to the terminal, the switching equipment, the artificial intelligence (AI) server, and the access controller (AC).

[0071] In the embodiment of the present application, after the AP obtains the terminal information of the terminal, it sends it to the AC through the switching device. The AC then forwards the collected terminal information to the artificial intelligence server through the switching device, and the artificial intelligence server executes the terminal health detection method provided in the embodiment of the present application.

[0072] Figure 1 The scenario shown is a wireless network scenario, and the terminal in the figure is a wireless terminal.

[0073] See also Figure 2 , which is a schematic diagram of the second network topology provided in an embodiment of the present application.

[0074] The figure includes the detected terminal, switching device, router, and artificial intelligence server. In the embodiment of the present application, after the router obtains the terminal information of the terminal, it forwards the terminal information to the artificial intelligence server through the switching device, and the artificial intelligence server executes the terminal health detection method provided in the embodiment of the present application.

[0075] Figure 2The scenario shown is a wired network scenario, and the terminals in the figure are wired terminals.

[0076] See also Figure 3 , which is a flow chart of a terminal health detection method provided in an embodiment of the present application, including the following steps S301-S303.

[0077] S301: Acquire terminal information indicating terminal operation status.

[0078] The terminal may be a wired terminal or a wireless terminal. The terminal information includes at least one of the following parameters of the terminal within a preset time period: latency, packet loss rate, uplink traffic, downlink traffic, uplink speed selection, downlink speed selection, uplink retransmission rate, and downlink retransmission rate.

[0079] In one embodiment of the present application, a preset time interval may be set, and parameters included in the terminal information may be collected once according to a preset period within the preset time interval, and all parameters collected within the preset time interval may be collectively used as the terminal information.

[0080] For example, if the above terminal information includes delay, packet loss rate, uplink traffic, downlink traffic, uplink speed selection, downlink speed selection, uplink retransmission rate, and downlink retransmission rate, a total of 8 parameters, the above preset time interval is 450s, and the preset period is 10s, that is, within 450s, the above 8 parameters are collected once every 10s, and finally 45 groups of terminal information are obtained, each group containing 8 parameters. The terminal information can be expressed as [45,1,8], where 45 represents the number of parameter groups contained in the terminal information, 1 represents that all parameters together constitute 1 terminal information, and 8 represents that each group contains 8 parameters.

[0081] The above is only an example. In fact, the above preset time interval, the duration of the preset cycle, the number of parameter groups included in each terminal information, and the number of parameter items included in each parameter group can all be set based on actual conditions.

[0082] S302: Input the above terminal information into a pre-trained terminal health detection model for encoding and classification to obtain an output result.

[0083] The output result indicates the probability of the terminal being in a dormant state and being in different health levels during operation, and the health levels indicate the network access quality of the terminal during operation.

[0084] In one embodiment of the present application, the terminal health degree detection model may be a logistic regression model, a support vector machine model, a random forest model, a neural network model, etc. In one example, a neural network model may be selected as the terminal health degree detection model.

[0085] In the embodiment of the present application, the original data of the terminal information is selected as the input of the terminal health detection model, so that the most complete information in the original data can be retained.

[0086] Specifically, the terminal health detection model first encodes the terminal information to generate a one-dimensional terminal information vector of a preset length, and then classifies the terminal information vector to obtain an output result. The output result indicates the probability that the terminal is in a dormant state and has various degrees of health during operation.

[0087] For example, the preset length may be 16, 32, etc.

[0088] In one embodiment of the present application, the terminal health detection model includes an encoding sub-model for encoding terminal information and a classification sub-model for generating output results.

[0089] See also Figure 4 , which is a structural diagram of a terminal health detection model provided in an embodiment of the present application.

[0090] The figure includes a coding sub-model and a classification sub-model. The original terminal information is input into the coding sub-model, and the terminal information vector output by the coding sub-model is input into the classification sub-model. The classification sub-model outputs the output result of the terminal health detection model.

[0091] S303: Determine, based on the output result, whether the terminal is in a dormant state or not in operation, or the health of the terminal during operation.

[0092] In one embodiment of the present application, the state with the highest probability in the output result may be determined as the actual state of the terminal.

[0093] As can be seen from the above, in the embodiment of the present application, the terminal information of the terminal can be analyzed and processed by the terminal health detection model to obtain the output result, without the need to calculate the health score based on a fixed threshold manually set by an expert. The terminal health detection model obtained by training based on a large amount of data can learn the features in the data, perform accurate terminal health analysis, and save labor costs. And the output result can not only analyze the health of the terminal, but also distinguish whether the terminal is in an unused dormant state, which can avoid the terminal being defaulted to a used state during the terminal health program detection process, and the health of the terminal in a dormant state is mistakenly detected as poor health, thereby avoiding the problem of incorrect terminal health evaluation.

[0094] The structures of the encoding sub-model and the classification sub-model are described below.

[0095] In one embodiment of the present application, the above-mentioned encoding sub-model may be a VAE (Variational AutoEncoder) model.

[0096] The encoding sub-model may include a series of sequentially connected convolutional layers, each of which contains a convolution kernel and an activation function. The first output data of the last convolutional layer is then flattened into one-dimensional data of a first length, and then the one-dimensional data is processed through a series of sequentially connected fully connected layers to obtain second output data. The second output data is divided into two one-dimensional vectors, and the length of the two one-dimensional vectors is half of the second output data. Specifically, the second output data is divided in half to obtain two one-dimensional vectors. The first half is used as the target mean, and the second half is used as the target variance. A series of random numbers are generated, the random numbers are normally distributed, the mean of the random numbers is the target mean, and the variance of the random numbers is the target variance. Then, a reconstructed latent vector of a preset length is sampled from the random numbers as the terminal information vector.

[0097] In the above example, the terminal information is [45,1,8], and the terminal information is input into the encoding sub-model. It is first processed by a 2d convolution layer. The convolution kernel size in the 2d convolution layer is [5,1], the step size is 1, and the dimension is 50. The activation function in the 2d convolution layer can be a ReLU (Linear rectification function). After the processing of the 2d convolution layer, the size of the obtained data becomes [41,1,50]. The data is further processed by the second 2d convolution layer. The convolution kernel size in the second 2d convolution layer is [3,1], the step size is 1, and the dimension is 3. The activation function can be a ReLU function, and the size of the obtained data becomes [39,1,3].

[0098] The data is then flattened into one-dimensional data of length

[108] and processed by the first fully connected layer. The dimension of the fully connected layer is 32. The activation function in the fully connected layer can be the ReLU function, and the output data size is

[32] .

[0099] The data is then processed by the second fully connected layer. The dimension of the fully connected layer is 32. The activation function in the fully connected layer can be the ReLU function, and the output data size is

[32] .

[0100] The data is divided into two one-dimensional vectors of length

[16] , the two one-dimensional vectors are the target mean and the target variance. Then a series of random numbers are generated, the random numbers are normally distributed, the mean of the random numbers is the target mean, and the variance of the random numbers is the target variance. Then a reconstructed latent vector of a preset length of

[16] is sampled from the random numbers as the terminal information vector.

[0101] In addition, the classification sub-model may include a series of fully connected layers, and the length of the final output result is the number of classifications of the terminal health detection result. For example, the terminal health detection result includes the terminal being in a dormant state, the terminal being in operation and having an excellent health level, and the terminal being in operation and having a poor health level. The number of classifications is 3, and the length of the output result is also 3. The output result contains 3 data, each of which corresponds to a classification of the terminal health detection result, indicating the probability that the terminal health detection result is the classification.

[0102] Based on the fact that the length of the terminal information vector in the above example is

[16] , the dimension of the first fully connected layer included in the above classification sub-model can be 64, the activation function can be the ReLU function, and the length of the output data is

[64] .

[0103] The dimension of the second fully connected layer can be 32, the activation function can be the ReLU function, and the length of the output data is

[32] .

[0104] The dimension of the third fully connected layer can be 3, the activation function can be the SoftMax function, and the length of the output data is [3].

[0105] It should be noted that, in addition to the above examples, the number of convolutional layers included in the above encoding sub-model, as well as the step size, dimension, convolution kernel size, and activation function of each convolutional layer can be set as required, as long as the length and dimension of the data output by the subsequent convolutional layer are less than or equal to the length and dimension of the data output by the previous convolutional layer. The number of fully connected layers included in the above classification sub-model, as well as the dimension of each fully connected layer, and the activation function included can be set as required, as long as the length of the data output by the subsequent fully connected layer is less than or equal to the length of the data output by the previous fully connected layer, and the length of the data output by the last fully connected layer is equal to the total number of different health levels and sleep states in the terminal's health detection result.

[0106] In one embodiment of the present application, the above encoding sub-model is implemented as follows Figure 5 The process shown is trained.

[0107] See also Figure 5 , is a schematic diagram of the first encoding sub-model training process provided in the embodiment of the present application, including the following steps S501-S504.

[0108] S501: Inputting first sample terminal information into a basic coding sub-model for encoding to obtain a first sample terminal information vector.

[0109] The structure of the basic coding sub-model is the same as that of the coding sub-model, and the only difference is the parameters of the two. The basic coding sub-model will not be described in detail here.

[0110] In addition, the method of obtaining the first sample terminal information is similar to the aforementioned method of obtaining the terminal information, and will not be repeated here.

[0111] In one embodiment of the present application, the above step S501 can be implemented by following steps A-C.

[0112] Step A: Based on the basic coding sub-model, convert the first sample terminal information into a one-dimensional vector of a first length.

[0113] Specifically, the basic encoding sub-model can convert the terminal information into a one-dimensional vector of the first length through a series of convolution, activation and full connection processing.

[0114] Step B: Based on the basic coding sub-model, the one-dimensional vector is divided into a target mean and a target variance.

[0115] The lengths of the target mean and the target variance are both half of the first length.

[0116] In one embodiment of the present application, the one-dimensional vector may be split into two halves, with the first half serving as the target mean and the second half serving as the target variance.

[0117] Step C: Based on the above basic coding sub-model, a first sample terminal information vector of a preset length is obtained by sampling from random numbers that satisfy a normal distribution.

[0118] Among them, the mean of the above random numbers is the above target mean, and the variance of the above random numbers is the above target variance.

[0119] S502: Input the first sample terminal information vector into the basic decoding sub-model for decoding to obtain a sample decoding result.

[0120] The size of the sample decoding result is the same as the size of the first sample terminal information.

[0121] Specifically, the processing process of the above basic decoding sub-model on the input data is opposite to that of the basic encoding sub-model.

[0122] The above basic encoding sub-model includes a series of convolutional layers and a series of fully connected layers from front to back in the processing order, and the basic decoding sub-model includes a series of fully connected layers and a series of deconvolutional layers from front to back in the processing order.

[0123] The processing process of the input data by the above-mentioned basic decoding sub-model is opposite to that of the basic encoding sub-model, which means that the size of the output data of the last fully connected layer in the basic encoding sub-model is the same as the size of the input data of the first fully connected layer in the basic decoding sub-model; the size of the input data of the last fully connected layer in the basic encoding sub-model is the same as the size of the output data of the first fully connected layer in the basic decoding sub-model; the size of the output data of the second-to-last fully connected layer in the basic encoding sub-model is the same as the size of the input data of the second fully connected layer in the basic decoding sub-model; the size of the input data of the second-to-last fully connected layer in the basic encoding sub-model is the same as the size of the output data of the second fully connected layer in the basic decoding sub-model; and so on.

[0124] The size of the output data of the last convolutional layer in the basic encoding sub-model is the same as the size of the input data of the first deconvolutional layer in the basic decoding sub-model; the size of the input data of the last convolutional layer in the basic encoding sub-model is the same as the size of the output data of the first deconvolutional layer in the basic decoding sub-model; the size of the output data of the second-to-last convolutional layer in the basic encoding sub-model is the same as the size of the input data of the second deconvolutional layer in the basic decoding sub-model; the size of the input data of the second-to-last convolutional layer in the basic encoding sub-model is the same as the size of the output data of the second deconvolutional layer in the basic decoding sub-model; and so on.

[0125] Based on the above example of the encoding sub-model structure, the structure of the basic decoding sub-model can be exemplified as follows:

[0126] The dimension of the first fully connected layer in the basic decoding sub-model is 32, and the activation function contained therein can be the ReLU function, and the size of the output data is

[32] . The dimension of the second fully connected layer is 108, and the activation function contained therein can be the ReLU function, and the size of the output data is

[108] . The output data of size

[108] is then reorganized into data of size [39,1,3]. The data is processed by the first 2d deconvolution layer, the convolution kernel size of the first 2d deconvolution layer is [3,1], the step size is 1, the dimension is 10, the activation function can be ReLU, and the size of the output data is [41,1,10]; the output data is then processed by the second 2d deconvolution layer, the convolution kernel size of the second 2d deconvolution layer is [5,1], the step size is 1, the dimension is 8, there is no activation function, and the output data is the sample decoding result, the size is the same as the first sample terminal information, both are [45,1,8].

[0127] In addition, before executing step S502, a random error may be added to the first sample terminal vector output by the basic coding sub-model.

[0128] S503: Calculate a first loss based on the first sample terminal information and the sample decoding result.

[0129] In one embodiment of the present application, the first loss may be calculated based on a HuberLoss (smoothed L1 loss) function, or a loss function in other related technologies, such as a cross entropy function.

[0130] S504: Based on the first loss, adjust parameters of the basic encoding sub-model and the basic decoding sub-model.

[0131] Specifically, the first solver may be used to adjust parameters, the learning rate may be the first learning rate, and the batch-size may be the first preset size.

[0132] For example, the first solver may be an Adam (adaptive moment estimation) solver or a solver in other related technologies, the first learning rate may be set as required, such as 0.001, and the first preset size of the batch-size may also be set as required, such as 32, 64, etc.

[0133] If the first preset training termination condition is not met, the process returns to step S501 and continues to execute until the first preset training termination condition is met, thereby obtaining a trained encoding sub-model.

[0134] The first preset training termination condition may be that the first loss is lower than the first preset loss, or the number of training rounds reaches the first preset number of rounds, for example, the first preset number of rounds is 300 rounds, 400 rounds, etc.

[0135] See also Figure 6 , which is a schematic diagram of the second encoding sub-model training process provided in an embodiment of the present application.

[0136] The figure includes a basic encoding sub-model and a basic decoding sub-model. The input to the basic encoding sub-model is the first sample terminal information, and the output of the basic decoding sub-model is the sample decoding result. The training goal is: the sample decoding result is as similar as possible to the first sample terminal information.

[0137] As can be seen from the above, in the embodiment of the present application, the basic encoding sub-model and the basic decoding sub-model are trained together, and finally the trained encoding sub-model is used for data processing. The process of training the encoding sub-model is an unsupervised training process, which can reduce the workload required to add labels to the first sample terminal information.

[0138] In another embodiment of the present application, the above classification sub-model is implemented by Figure 7 The process shown is trained.

[0139] See also Figure 7, is a schematic diagram of a classification sub-model training process provided in an embodiment of the present application.

[0140] S701: Obtain a second sample terminal information vector obtained by encoding the second sample terminal information through the above encoding sub-model.

[0141] In one embodiment of the present application, the encoding sub-model may be trained first, and then the sample terminal information may be processed by the trained encoding sub-model to obtain a second sample terminal information vector.

[0142] The second sample terminal information used when training the basic classification sub-model is the sample terminal information with the sample label added. That is, in the embodiment of the present application, the obtained sample terminal information can be divided into two parts, one part does not need to add the sample label, and is used as the first sample terminal information for training the encoding sub-model, and the other part adds the sample label and is used as the second sample terminal information for training the classification sub-model.

[0143] S702: Input the second sample terminal information vector into the basic classification sub-model for classification to obtain a sample output result.

[0144] Specifically, the structure of the above basic classification sub-model is the same as that of the aforementioned classification sub-model, except for the parameters. The basic classification sub-model will not be described in detail in the embodiments of the present application.

[0145] S703: Calculate the second loss based on the sample output result and the sample label corresponding to the second sample terminal information.

[0146] The sample tags correspond one-to-one to the second sample terminal information, indicating that the terminal corresponding to the second sample terminal information is in a dormant state or is not in operation, or the health of the terminal during operation.

[0147] Specifically, the sample label may be manually added to the sample terminal information.

[0148] In one embodiment of the present application, the second loss between the sample output result and the sample label can be calculated based on the cross entropy function or other loss calculation functions in related technologies (such as the HuberLoss function).

[0149] S704: Adjust parameters of the basic classification sub-model based on the second loss.

[0150] Specifically, the second solver may be used to adjust parameters, the learning rate is the second learning rate, and the batch-size is the second preset size.

[0151] For example, the second solver can be an Adam solver or a solver in other related technologies. The second learning rate can be set according to demand, such as 0.001, and the second preset size of batch-size can also be set according to demand, such as 16, 32, etc.

[0152] If the second preset training termination condition is not met, the process returns to step S701 and continues to execute until the second preset training termination condition is met, thereby obtaining a trained classification sub-model.

[0153] The second preset training termination condition may be that the second loss is lower than the second preset loss, or the number of training rounds reaches a second preset number of rounds, for example, the second preset number of rounds is 300 rounds, 400 rounds, etc.

[0154] It can be seen from the above that in the embodiment of the present application, the second sample terminal information with the sample label added can complete the training of the classification sub-model.

[0155] In summary, in the embodiment of the present application, the coding sub-model and the classification sub-model in the terminal health detection model are trained separately instead of being trained uniformly, which can improve the accuracy of the coding sub-model and the classification sub-model in the trained terminal health detection model. And because the first sample terminal information used in training the coding sub-model does not need to add sample labels, only the second sample terminal information used in training the classification sub-model needs to add sample labels, so compared with directly using the sample terminal information with sample labels added to the model to directly train the terminal health detection model at one time, training different sub-models separately can reduce the number of second sample terminal information that needs to add sample labels, and reduce the workload of the training process.

[0156] See also Figure 8 , which is a schematic diagram of a model effect provided by an embodiment of the present application. In the figure, each row corresponds to a result obtained after the terminal health detection model after training five times processes the test data, and each column corresponds to a label corresponding to the test data.

[0157] The first column indicates that among the test data labeled as dormant state, the number of terminals detected as dormant state by the terminal health detection model is 131, the number of terminals detected as poor health is 1, and the number of terminals detected as excellent health is 9.

[0158] The second column indicates that among the test data labeled as poor health, the number of devices detected as dormant by the terminal health detection model is 0, the number of devices detected as poor health is 73, and the number of devices detected as excellent health is 8.

[0159] The third column indicates that among the test data labeled as excellent health, 7 were detected as dormant by the terminal health detection model, 15 were detected as poor health, and 73 were detected as excellent health.

[0160] That is, after five model trainings using the above method, the terminal health detection model obtained has a detection accuracy of approximately 95% for the sleep state, approximately 92% for poor health, and approximately 88% for excellent health. In other words, accurate detection can be achieved based on the above terminal health detection model.

[0161] In addition, the method for collecting the above test data is the same as the method for collecting sample terminal information. Part of the collected data can be used as sample terminal information and the other part as test data. The embodiment of the present application does not limit the method for collecting test data.

[0162] Furthermore, in executing the aforementioned Figure 3 The terminal health detection method shown will also obtain a large amount of terminal information during the terminal health detection process. The obtained terminal information can be collected as new sample terminal information to further train the terminal health detection model, so that the trained terminal health detection model is more suitable for application scenarios.

[0163] Corresponding to the aforementioned method for detecting the health status of a terminal, an embodiment of the present application further provides a device for detecting the health status of a terminal.

[0164] See also Fig. 9 , is a schematic diagram of the structure of a terminal health detection device provided in an embodiment of the present application, the device comprising:

[0165] Processor 901;

[0166] transceiver 904;

[0167] A machine-readable storage medium 902 stores machine-executable instructions that can be executed by the processor; the machine-executable instructions prompt the processor to execute any one of the method steps in the terminal health detection method.

[0168] like Fig. 9As shown, the network device may further include a communication bus 903. The processor 901, the machine-readable storage medium 902, and the transceiver 904 communicate with each other through the communication bus 903. The communication bus 903 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The communication bus 903 may be divided into an address bus, a data bus, a control bus, and the like.

[0169] The transceiver 904 may be a wireless communication module. Under the control of the processor 901 , the transceiver 904 exchanges data with other devices.

[0170] The machine-readable storage medium 902 may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk storage. In addition, the machine-readable storage medium 902 may also be at least one storage device located away from the aforementioned processor.

[0171] The processor 901 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0172] As can be seen from the above, in the embodiment of the present application, the terminal information of the terminal can be analyzed and processed by the terminal health detection model to obtain the output result, without the need to calculate the health score based on a fixed threshold manually set by an expert. The terminal health detection model obtained by training based on a large amount of data can learn the features in the data, perform accurate terminal health analysis, and save labor costs. And the output result can not only analyze the health of the terminal, but also distinguish whether the terminal is in an unused dormant state, which can avoid the terminal being defaulted to a used state during the terminal health program detection process, and the health of the terminal in a dormant state is mistakenly detected as poor health, thereby avoiding the problem of incorrect terminal health evaluation.

[0173] Corresponding to the aforementioned terminal health detection method, the embodiment of the present application also provides a terminal health detection device.

[0174] See also Fig.10 , is a schematic diagram of the structure of a terminal health detection device provided in an embodiment of the present application, the device comprising:

[0175] The information acquisition module 1001 is used to acquire terminal information indicating the operation status of the terminal;

[0176] The result output module 1002 is used to input the terminal information into a pre-trained terminal health detection model for encoding and classification to obtain an output result, wherein the output result indicates: the probability of the terminal being in a dormant state and being in different health levels during operation, and the health level indicates the network access quality of the terminal during operation;

[0177] The health determination module 1003 is used to determine whether the terminal is in a dormant state or the health status of the terminal during operation according to the output result.

[0178] As can be seen from the above, in the embodiment of the present application, the terminal information of the terminal can be analyzed and processed by the terminal health detection model to obtain the output result, without the need to calculate the health score based on a fixed threshold manually set by an expert. The terminal health detection model obtained by training based on a large amount of data can learn the features in the data, perform accurate terminal health analysis, and save labor costs. And the output result can not only analyze the health of the terminal, but also distinguish whether the terminal is in an unused dormant state, which can avoid the terminal being defaulted to a used state during the terminal health program detection process, and the health of the terminal in a dormant state is mistakenly detected as poor health, thereby avoiding the problem of incorrect terminal health evaluation.

[0179] In one embodiment of the present application, the terminal health detection model includes an encoding sub-model for encoding terminal information and a classification sub-model for generating output results.

[0180] In one embodiment of the present application, the encoding sub-model is obtained by training the following modules:

[0181] A first encoding module, used for inputting the first sample terminal information into the basic encoding sub-model for encoding to obtain a first sample terminal information vector;

[0182] a decoding module, configured to input the first sample terminal information vector into a basic decoding sub-model for decoding, to obtain a sample decoding result, wherein the size of the sample decoding result is the same as the size of the first sample terminal information;

[0183] A first loss calculation module, used for calculating a first loss based on the first sample terminal information and the sample decoding result;

[0184] The first parameter adjustment module is used to adjust the parameters of the basic encoding sub-model and the basic decoding sub-model based on the first loss, and return to continue execution from the first encoding module until a first preset training termination condition is reached to obtain a trained encoding sub-model.

[0185] As can be seen from the above, in the embodiment of the present application, the basic encoding sub-model and the basic decoding sub-model are trained together, and finally the trained encoding sub-model is used for data processing. The process of training the encoding sub-model is an unsupervised training process, which can reduce the workload required to add labels to the first sample terminal information.

[0186] In one embodiment of the present application, the first encoding module is specifically used to:

[0187] Based on the basic coding sub-model, converting the first sample terminal information into a one-dimensional vector of a first length;

[0188] Based on the basic coding sub-model, dividing the one-dimensional vector into a target mean and a target variance, wherein the lengths of the target mean and the target variance are both half of the first length;

[0189] Based on the basic coding sub-model, a first sample terminal information vector of a preset length is obtained by sampling random numbers that satisfy a normal distribution, wherein the mean of the random numbers is the target mean, and the variance of the random numbers is the target variance.

[0190] In one embodiment of the present application, the classification sub-model is obtained by training the following modules:

[0191] A second encoding module, used to obtain a second sample terminal information vector obtained by encoding the second sample terminal information through the encoding sub-model;

[0192] A classification module, used for inputting the second sample terminal information vector into a basic classification sub-model for classification to obtain a sample output result;

[0193] A second loss calculation module, configured to calculate a second loss based on the sample output result and a sample label corresponding to the second sample terminal information, wherein the sample label corresponds to the second sample terminal information one-to-one, indicating that the terminal corresponding to the corresponding second sample terminal information is in a dormant state without operation, or the health of the terminal during operation;

[0194] The second parameter adjustment module is used to adjust the parameters of the basic classification sub-model based on the second loss, and return to continue execution from the second encoding module until a second preset training termination condition is reached to obtain a trained classification sub-model.

[0195] It can be seen from the above that in the embodiment of the present application, the second sample terminal information with the sample label added can complete the training of the classification sub-model.

[0196] In one embodiment of the present application, the terminal information includes at least one of the following parameters of the terminal within a preset time period: delay, packet loss rate, uplink traffic, downlink traffic, uplink speed selection, downlink speed selection, uplink retransmission rate, and downlink retransmission rate.

[0197] In another embodiment provided in the present application, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned terminal health detection methods are implemented.

[0198] In another embodiment provided in the present application, a computer program product including instructions is also provided, which, when executed on a computer, enables the computer to execute any terminal health detection method in the above embodiments.

[0199] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive Solid State Disk (SSD)), etc.

[0200] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0201] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, apparatus, computer-readable storage medium, and computer program product embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0202] The above description is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application are included in the protection scope of the present application.

Claims

1. A terminal health detection method, It is characterized in that The method comprises: Acquire terminal information indicating terminal operation status, where, if the terminal is a wireless terminal, the terminal is communicatively connected to a wireless access point, and, if the terminal is a wired terminal, the terminal is communicatively connected to a router; Inputting the terminal information into a pre-trained terminal health degree detection model for encoding and classification to obtain an output result, wherein the output result indicates: the probability of the terminal being in a dormant state and being in different health degrees during operation, wherein the health degree indicates the network access quality of the terminal during operation; It is determined according to the output result that the terminal is in a dormant state or the health status of the terminal during operation.

2. The method according to claim 1, It is characterized in that The terminal health detection model includes an encoding sub-model for encoding terminal information and a classification sub-model for generating output results.

3. The method according to claim 2, It is characterized in that The encoding sub-model is trained in the following way: Inputting the first sample terminal information into the basic coding sub-model for encoding to obtain a first sample terminal information vector; Inputting the first sample terminal information vector into a basic decoding sub-model for decoding to obtain a sample decoding result, wherein the size of the sample decoding result is the same as the size of the first sample terminal information; Calculating a first loss based on the first sample terminal information and the sample decoding result; Based on the first loss, the parameters of the basic encoding sub-model and the basic decoding sub-model are adjusted, and the process returns to the step of inputting the first sample terminal information into the basic encoding sub-model for encoding to obtain the first sample terminal information vector and continues to execute until the first preset training termination condition is reached to obtain the trained encoding sub-model.

4. The method according to claim 3, It is characterized in that The step of inputting the first sample terminal information into the basic coding sub-model for encoding to obtain the first sample terminal information vector includes: Based on the basic coding sub-model, converting the first sample terminal information into a one-dimensional vector of a first length; Based on the basic coding sub-model, dividing the one-dimensional vector into a target mean and a target variance, wherein the lengths of the target mean and the target variance are both half of the first length; Based on the basic coding sub-model, a first sample terminal information vector of a preset length is obtained by sampling random numbers that satisfy a normal distribution, wherein the mean of the random numbers is the target mean, and the variance of the random numbers is the target variance.

5. The method according to claim 2, It is characterized in that The classification sub-model is trained in the following way: Obtaining a second sample terminal information vector obtained by encoding the second sample terminal information through the encoding sub-model; Inputting the second sample terminal information vector into the basic classification sub-model for classification to obtain a sample output result; Calculate a second loss based on the sample output result and the sample label corresponding to the second sample terminal information, wherein the sample label corresponds to the second sample terminal information one-to-one, indicating that the terminal corresponding to the corresponding second sample terminal information is in a dormant state or is in a healthy state during operation; The parameters of the basic classification sub-model are adjusted based on the second loss, and the step of returning to the step of obtaining the second sample terminal information vector obtained after encoding the second sample terminal information through the encoding sub-model is continued until the second preset training termination condition is reached to obtain the trained classification sub-model.

6. The method according to any one of claims 1 to 5, It is characterized in that The terminal information includes at least one of the following parameters of the terminal within a preset time period: delay, packet loss rate, uplink traffic, downlink traffic, uplink speed selection, downlink speed selection, uplink retransmission rate, and downlink retransmission rate.

7. A terminal health detection device, It is characterized in that The device comprises: processor; Transceiver; A machine-readable storage medium storing machine-executable instructions that can be executed by the processor; the machine-executable instructions prompt the processor to execute the method steps of any one of claims 1-6.

8. A terminal health detection device, It is characterized in that The device comprises: An information acquisition module, used to acquire terminal information indicating the operation status of the terminal, where the terminal is a wireless terminal and is connected to a wireless access point in communication, and where the terminal is a wired terminal and is connected to a router in communication; A result output module is used to input the terminal information into a pre-trained terminal health detection model for encoding and classification to obtain an output result, wherein the output result indicates: the probability of the terminal being in a dormant state and being in different health levels during operation, and the health level indicates the network access quality of the terminal during operation; The health determination module is used to determine whether the terminal is in a dormant state or the health status of the terminal during operation according to the output result.

9. The device according to claim 8, It is characterized in that The terminal health detection model includes an encoding sub-model for encoding terminal information and a classification sub-model for generating output results.

10. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps described in any one of claims 1 to 6 are implemented.

Images