Terminal timing feature detection method, device and medium

Through the combination of ensemble prediction algorithm integrated learning and average absolute percentage error rate, the problem of insufficient detection accuracy of a single ensemble prediction algorithm in terminal security products is solved, and the timing feature detection with higher accuracy and robustness is achieved. It is suitable for Multi-scenario terminal equipment.

CN114254307BActive Publication Date: 2024-11-19HARBIN ANTIY TECH
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202111489030.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-08
Publication Date
2024-11-19
Estimated Expiration
2041-12-08

AI Technical Summary

Technical Problem

Existing terminal security products are difficult to adapt to different scenarios based on a single timing prediction algorithm, resulting in the inability to guarantee the accuracy of the detection results, and it is difficult to achieve high-precision timing abnormality detection.

Method used

The timing prediction algorithm integrated learning method is adopted, combined with the average absolute percentage error rate, and a timing prediction model is built to predict the terminal timing characteristics. Through the integration and weight allocation of multiple prediction algorithms, the final timing feature prediction value is generated.

Benefits of technology

It improves the accuracy and robustness of timing feature prediction values, enhances detection efficiency and accuracy, reduces false alarm rate, and is suitable for terminal equipment in multiple scenarios and long-term operation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114254307B_ABST
    Figure CN114254307B_ABST
Patent Text Reader

Abstract

The present invention relates to a terminal timing feature detection method, device and medium. First, the timing feature prediction value is calculated through a timing prediction model based on the integrated learning of the timing prediction algorithm and the mean absolute percentage error rate algorithm. Then, the terminal detects the actually generated timing features based on the timing feature prediction value to timely discover timing anomalies. The present invention avoids the bottleneck of predicting the terminal timing features by a single timing prediction algorithm, embodies a strong robust prediction capability, effectively improves the accuracy of the terminal timing feature prediction value, and detects the terminal timing anomaly based on such a timing feature prediction value, which can effectively improve the detection efficiency and detection accuracy, and is suitable for multiple scenarios.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of information security technology, and in particular to a terminal timing feature detection method, device and medium. Background Art

[0002] Security products installed on user terminals must have the ability to detect timing anomalies, that is, based on the timing information of user terminals in the past period of time, including terminal asset information, terminal behavior information, and threat warning information, etc., predict the timing characteristics of user terminals in the future period of time. When the actual timing characteristics do not meet the predicted values, they are judged as abnormal. It can be seen that terminal security products have extremely high requirements for the accuracy of the predicted values ​​of terminal timing characteristics. If the accuracy is not achieved, it is difficult to accurately detect the timing anomalies of the terminal, and the terminal security status cannot be judged in time, which makes it difficult to protect the data security of the user terminal.

[0003] Most of the existing terminal security products implement the terminal timing anomaly detection function based on a single timing prediction algorithm. However, due to its limitations, a single timing prediction algorithm is difficult to adapt to different scenarios and cannot guarantee the accuracy of the detection results in each scenario. Therefore, there is an urgent need for a terminal timing anomaly detection method that is robust, adaptive, and applicable to various scenarios. Summary of the invention

[0004] In view of this, the present invention provides a terminal timing feature detection method, device and medium, which are based on the integrated learning of the timing prediction algorithm and combined with the average absolute percentage error rate to obtain the terminal timing feature prediction value, which is used to detect the timing anomaly of the terminal. The present invention avoids the bottleneck of a single timing prediction algorithm for detecting timing anomalies, and adopts the integrated learning method of multiple timing prediction algorithms to predict timing features, which effectively improves the accuracy of the timing feature prediction value and at least partially solves the problems existing in the prior art.

[0005] The specific content of the invention is:

[0006] A terminal timing feature detection method, comprising:

[0007] Get terminal log data;

[0008] Performing statistics on the log data to obtain timing characteristic information of the terminal;

[0009] Constructing a time series training set and a time series verification set according to the time series feature information;

[0010] The time series training set and the time series verification set are calculated by a time series prediction model based on time series prediction algorithm ensemble learning to obtain a model prediction value matrix and a model verification value matrix respectively;

[0011] The model prediction value matrix and the model verification value matrix are calculated by a mean absolute percentage error rate algorithm to obtain a prediction value range coefficient;

[0012] The product of each prediction value in the model prediction value matrix and its corresponding prediction value range coefficient is calculated to obtain the timing feature prediction value of the terminal, and the timing feature prediction value is used to detect whether the timing feature of the terminal is abnormal.

[0013] Furthermore, constructing a time series training set and a time series verification set according to the time series feature information specifically includes:

[0014] Performing row-to-column calculation on the time series feature information to obtain a time series value matrix of each time series feature;

[0015] Filtering the time series value matrix whose eigenvalues ​​do not meet the specified threshold;

[0016] According to the preset time range rule, each time series value matrix obtained by filtering is divided according to the time range to obtain a time series training set and a time series verification set.

[0017] Furthermore, the time series prediction model based on the time series prediction algorithm ensemble learning is used to calculate the time series training set and the time series verification set to obtain a model prediction value matrix and a model verification value matrix, respectively, which specifically includes:

[0018] Determine the time series prediction algorithm applicable to the scenario, and debug and configure each time series prediction algorithm according to the scenario requirements;

[0019] The time series prediction algorithms and the Pearson correlation coefficient algorithm are used to perform cross-validation calculations on the time series training set and the time series verification set to meet a preset number of times, so as to obtain a model prediction value matrix and a model verification value matrix.

[0020] Furthermore, the cross-validation calculation specifically includes:

[0021] Inputting the time series training set into each time series prediction algorithm in the form of parameters to obtain the time series prediction value under each time series prediction algorithm;

[0022] Inputting the timing verification set into each timing prediction algorithm in the form of parameters to obtain the timing verification value under each timing prediction algorithm;

[0023] The Pearson correlation coefficient algorithm is used to calculate the Pearson correlation coefficient between the time series prediction value and the time series verification value under each time series prediction algorithm.

[0024] Further, after performing cross-validation calculations on the time series training set and the time series verification set to meet a preset number of times through the time series prediction algorithms and the Pearson correlation coefficient algorithm, the method further includes:

[0025] Calculate the mean of the Pearson correlation coefficients obtained under each time series prediction algorithm;

[0026] According to the mean calculation result, a weight is assigned to each time series prediction algorithm; the larger the mean result, the greater the weight assigned to the corresponding time series prediction algorithm.

[0027] Furthermore, obtaining the model prediction value matrix and the model verification value matrix specifically includes:

[0028] According to the weights of each time series prediction algorithm, the time series prediction values ​​obtained under each time series prediction algorithm are weighted averaged to obtain a model prediction value matrix; the time series verification values ​​obtained under each time series prediction algorithm are weighted averaged to obtain a model verification value matrix.

[0029] Furthermore, the model prediction value matrix and the model verification value matrix are calculated by the mean absolute percentage error rate algorithm to obtain the prediction value range coefficient, which specifically includes:

[0030] By using the mean absolute percentage error rate algorithm, each predicted value in the model prediction value matrix and its corresponding verification value in the model verification value matrix are calculated respectively, and the calculation results are filtered to obtain the fluctuation range coefficient corresponding to each predicted value in the model prediction value matrix.

[0031] Furthermore, it also includes: obtaining the current timing feature information of the terminal, matching the timing feature information with the timing feature prediction value, and if the matching fails, determining an abnormality and issuing an early warning.

[0032] An electronic device, comprising: a housing, a processor, a memory, a circuit board and a power supply circuit, wherein the circuit board is placed inside a space enclosed by the housing, and the processor and the memory are arranged on the circuit board; the power supply circuit is used to supply power to various circuits or devices of the above-mentioned electronic device; the memory is used to store executable program codes; the processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to execute the above-mentioned method.

[0033] A computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the above method.

[0034] The beneficial effects of the present invention are embodied in:

[0035] The present invention calculates the predicted value of the terminal timing characteristics through a timing prediction model based on the integrated learning of the timing prediction algorithm, thereby avoiding the bottleneck of a single timing prediction algorithm predicting the terminal timing characteristics, reflecting a strong robust prediction capability, and effectively improving the accuracy of the predicted value of the terminal timing characteristics. Considering the changes in the trend amplitude of the timing characteristics, the mean absolute percentage error rate is introduced to calculate the prediction value range coefficient. The predicted value obtained by the timing prediction model based on the integrated learning of the timing prediction algorithm and the predicted value range coefficient are comprehensively calculated to obtain the trend fluctuation range of the terminal timing characteristic prediction value, and the trend fluctuation range is integrated into the terminal timing characteristic prediction value. Based on such a timing characteristic prediction value, the terminal timing anomaly is detected, which can effectively improve the detection efficiency and detection accuracy, while reducing the detection false alarm rate. The present invention can adapt to multiple scenarios and is suitable for long-term operation on the client side. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0037] Figure 1 This is a flow chart of a terminal timing feature detection method according to an embodiment of the present invention;

[0038] Figure 2 This is a flow chart of a method for generating a terminal timing feature prediction value according to an embodiment of the present invention;

[0039] Figure 3 A flow chart of a method for detecting terminal timing anomalies based on timing feature prediction values ​​according to an embodiment of the present invention;

[0040] Figure 4 The figure is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0042] It should be noted that the following embodiments and features in the embodiments may be combined with each other in the absence of conflict; and, based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making any creative work are within the scope of protection of the present disclosure.

[0043] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.

[0044] The present invention provides an embodiment of a terminal timing feature detection method, such as Figure 1 As shown, including:

[0045] S11: Obtain terminal log data;

[0046] S12: Performing statistics on the log data to obtain timing characteristic information of the terminal;

[0047] S13: constructing a time series training set and a time series verification set according to the time series feature information;

[0048] S14: calculating the time series training set and the time series verification set through a time series prediction model based on time series prediction algorithm ensemble learning, and obtaining a model prediction value matrix and a model verification value matrix respectively;

[0049] S15: Calculate the model prediction value matrix and the model verification value matrix by using a mean absolute percentage error rate algorithm to obtain a prediction value range coefficient;

[0050] S16: Calculate the product of each prediction value in the model prediction value matrix and its corresponding prediction value range coefficient to obtain the timing feature prediction value of the terminal, and the timing feature prediction value is used to detect whether the timing feature of the terminal is abnormal.

[0051] The present invention calculates the predicted value of the terminal time series characteristics based on the time series prediction model of the time series prediction algorithm ensemble learning, which avoids the bottleneck of a single time series prediction algorithm predicting the terminal time series characteristics, reflects a strong robust prediction ability, and effectively improves the accuracy of the predicted value of the terminal time series characteristics. Taking into account the changes in the trend amplitude of the time series characteristics, the mean absolute percentage error rate is introduced to calculate the prediction value range coefficient to obtain the fluctuation range of the predicted value based on the true value. The trend fluctuation range is integrated into the predicted value of the terminal time series characteristics, which effectively improves the prediction ability and the accuracy of the predicted value of the terminal time series characteristics, and can adapt to the fluctuation changes of different time series scenarios.

[0052] Preferably, constructing a time series training set and a time series verification set according to the time series feature information specifically includes:

[0053] Performing row-to-column calculation on the time series feature information to obtain a time series value matrix of each time series feature;

[0054] Filtering the time series value matrix whose eigenvalues ​​do not meet the specified threshold;

[0055] According to the preset time range rule, each time series value matrix obtained by filtering is divided according to the time range to obtain a time series training set and a time series verification set.

[0056] The purpose of filtering the time series value matrices whose eigenvalues ​​do not meet the specified threshold in the above preferred scheme is to filter out sparse matrices. The filtering rules can be customized according to actual needs. For example, filtering is performed in units of weeks. The specified threshold is set to the number of days with a single-day eigenvalue of 0 accounting for more than 80% of the days being less than 4 days. Then the time series value matrices corresponding to the situation in which the daily eigenvalue is 0 accounting for more than 80% of the days in a week are filtered out. The time series value matrices obtained by filtering are divided according to the time range according to the preset time range rule to obtain a time series training set and a time series verification set. For example, the training set is preset to be divided according to 15 weeks, and the verification set is divided according to 1 week. According to the number of training samples required, each filtered time series value matrix can be divided multiple times, such as dividing it into 5 training sets of 15 weeks and 1 week verification sets for subsequent cross-validation.

[0057] Preferably, the time series prediction model based on the time series prediction algorithm ensemble learning is used to calculate the time series training set and the time series verification set to obtain a model prediction value matrix and a model verification value matrix, respectively, specifically including:

[0058] Determine the time series prediction algorithm applicable to the scenario, and debug and configure each time series prediction algorithm according to the scenario requirements; perform cross-validation calculations on the time series training set and the time series verification set to meet the preset number of times through each time series prediction algorithm and the Pearson correlation coefficient algorithm to obtain a model prediction value matrix and a model verification value matrix.

[0059] In terminal security products, common time series prediction algorithms include HoltWinters (triple exponential smoothing of non-stationary series containing linear trends and periodic fluctuations), ARIMA (autoregressive integrated moving average), EWMA (exponentially weighted moving average) and ARCH (autoregressive conditional heteroskedasticity model). In the prior art, most of them are based on one of them for time series prediction, which is difficult to adapt to different scenarios, and the accuracy of the detection results in each scenario cannot be guaranteed. The present invention can integrate the above-mentioned prediction algorithms, other existing prediction algorithms or user-developed prediction algorithms to avoid the bottleneck of a single time series prediction algorithm for predicting terminal time series features, and can be applied to multiple scenarios, while enhancing the generalization ability of the application.

[0060] The definition of cross validation is that, in a given sample, most of the samples are randomly or sequentially taken out for model training prediction, and a small part of the samples are left for verification. Performing one cross validation calculation is called 1-fold cross validation, performing five cross validation calculations is called 5-fold cross validation, and so on. According to the example of the aforementioned preferred scheme, the five 15-week training sets divided into five parts are the majority of samples used for model training prediction, and the five 1-week validation sets divided into five parts are the small part of samples used for verification. In order to better ensure the accuracy of the model prediction value matrix and the model validation value matrix, and then ensure the accuracy of the time series feature prediction value, the present invention uses 5-fold cross validation to calculate the training set and the validation set. The validation set exemplified in the embodiment of the present invention collects data for one week, so what is obtained within the scope of the example is the time series feature prediction value of the terminal within the next week. In practical applications, the collection period of multiple validation sets can be set according to the specific prediction needs, such as 3 days, 15 days, one month, etc.

[0061] Preferably, the cross-validation calculation specifically includes:

[0062] The time series training set is input into each time series prediction algorithm in the form of parameters to obtain the time series prediction value under each time series prediction algorithm; the time series verification set is input into each time series prediction algorithm in the form of parameters to obtain the time series verification value under each time series prediction algorithm; the Pearson correlation coefficient between the time series prediction value and the time series verification value under each time series prediction algorithm is calculated by the Pearson correlation coefficient algorithm. This process is a 1-fold cross-validation process, and executing this process 5 times is a 5-fold cross-validation. The Pearson correlation algorithm is an algorithm for measuring the linear relationship between fixed-distance variables. Introducing this algorithm into the cross-validation calculation can further ensure the accuracy of the cross-validation calculation.

[0063] Preferably, after performing cross-validation calculations on the time series training set and the time series verification set to meet a preset number of times by using the time series prediction algorithms and the Pearson correlation coefficient algorithm, the method further includes:

[0064] Calculate the mean of the Pearson correlation coefficients obtained under each time series prediction algorithm;

[0065] According to the mean calculation result, a weight is assigned to each time series prediction algorithm; the larger the mean result, the greater the weight assigned to the corresponding time series prediction algorithm.

[0066] In order to further illustrate the above-mentioned cross-validation calculation and the preferred scheme for assigning weights to each time series prediction algorithm, based on the above-mentioned example, further examples are given as follows: the five 15-week training sets that have been divided are input into each time series prediction algorithm in the form of parameters to obtain the time series prediction values ​​of the five training sets under each time series prediction algorithm; the five 1-week verification sets that have been divided are input into each time series prediction algorithm in the form of parameters to obtain the time series verification values ​​of the five verification sets under each time series prediction algorithm, and the Pearson correlation coefficient algorithm is used to calculate the Pearson correlation coefficient between the time series prediction value and the time series verification value under each time series prediction algorithm, and the Pearson correlation coefficients obtained under each time series prediction algorithm are averaged, and the obtained averages represent the actual performance values ​​of the corresponding time series prediction algorithms. The higher the actual performance value, the greater the weight allocated to the corresponding time series prediction algorithm.

[0067] Preferably, obtaining the model prediction value matrix and the model verification value matrix specifically includes:

[0068] According to the weight of each time series prediction algorithm, the time series prediction values ​​obtained under each time series prediction algorithm are weighted averaged to obtain the model prediction value matrix; the time series verification values ​​obtained under each time series prediction algorithm are weighted averaged to obtain the model verification value matrix. Multiple time series prediction algorithms are weighted according to their actual performance, and then weights are assigned, and then they are integrated together using the weighted average method, which can show a strong and robust prediction ability.

[0069] Preferably, the calculation of the model prediction value matrix and the model verification value matrix by the mean absolute percentage error rate algorithm to obtain the prediction value range coefficient specifically includes:

[0070] Through the mean absolute percentage error rate algorithm, each predicted value in the model prediction value matrix and its corresponding verification value in the model verification value matrix are calculated respectively, and the calculation results are filtered to obtain the fluctuation range coefficient corresponding to each predicted value in the model prediction value matrix. The filtering of the calculation results is a process of correcting the edge of the fluctuation range. The filtering rules can be defined according to actual needs. For example, the calculation results can be sorted first, and then the calculation results of the front and back 10% are filtered out, and the remaining calculation results are rounded to four decimal places to obtain the fluctuation range coefficient. Combined with the above example, the fluctuation range coefficient represents the fluctuation range of the time series feature prediction value in the next week.

[0071] Preferably, the method further comprises: acquiring current timing feature information of the terminal, matching the timing feature information with the timing feature prediction value, and if the matching fails, determining an abnormality and issuing an early warning.

[0072] The present invention Figure 1 The embodiment first calculates the terminal timing feature prediction value based on the timing prediction model of the timing prediction algorithm ensemble learning, combined with the mean absolute percentage error rate algorithm, and then detects the terminal timing anomaly based on the timing feature prediction value, which can effectively improve the detection efficiency and detection accuracy, while reducing the detection false alarm rate. Among them, the process of calculating and generating the timing feature prediction value can be run on the server side or on the terminal. If it is run on the server side, the generated timing feature prediction value is sent to the terminal through remote data transmission for detecting the timing anomaly of the terminal, which is more suitable for use in terminal devices in Internet scenarios; if it is run on the terminal, the process of calculating and generating the timing feature prediction value and detecting the terminal timing anomaly runs on the same terminal, and the process of calculating and generating the timing feature prediction value runs in the terminal background, which can better provide timing anomaly detection support for stand-alone scenarios.

[0073] Accordingly, in order to further illustrate the calculation and generation process of the timing feature prediction value of the present invention, the present invention provides an embodiment of a method for generating a timing feature prediction value of a terminal, such as Figure 2 As shown, including:

[0074] S21: Acquire terminal log data, perform statistics on the log data, and obtain time series feature information of the terminal;

[0075] S22: performing row-to-column calculation on the time series feature information to obtain a time series value matrix of each time series feature, and filtering the time series value matrix containing feature values ​​that do not meet a specified threshold;

[0076] S23: dividing each time series value matrix obtained by filtering according to the preset time range rule by time range to obtain a time series training set and a time series verification set;

[0077] S24: performing cross-validation calculations on the time series training set and the time series verification set to meet a preset number of times through each time series prediction algorithm included in the time series prediction model based on the time series prediction algorithm ensemble learning, and the Pearson correlation coefficient algorithm; specifically comprising: inputting the time series training set into each time series prediction algorithm in the form of parameters to obtain the time series prediction value under each time series prediction algorithm; inputting the time series verification set into each time series prediction algorithm in the form of parameters to obtain the time series verification value under each time series prediction algorithm; and calculating the Pearson correlation coefficients of the time series prediction value and the time series verification value under each time series prediction algorithm respectively through the Pearson correlation coefficient algorithm;

[0078] S25: Calculate the mean of the Pearson correlation coefficients obtained under each time series prediction algorithm, and assign a weight to each time series prediction algorithm according to the mean calculation result;

[0079] S26: performing weighted average calculation on the time series prediction values ​​obtained under each time series prediction algorithm according to the weights of each time series prediction algorithm to obtain a model prediction value matrix; performing weighted average calculation on the time series verification values ​​obtained under each time series prediction algorithm to obtain a model verification value matrix;

[0080] S27: Calculate each predicted value in the model predicted value matrix and its corresponding verification value in the model verification value matrix respectively by using the mean absolute percentage error rate algorithm, and filter the calculation results to obtain the fluctuation range coefficient corresponding to each predicted value in the model predicted value matrix;

[0081] S28: Calculate the product of each prediction value in the model prediction value matrix and its corresponding prediction value range coefficient to obtain the timing feature prediction value of the terminal, and the timing feature prediction value is used to detect whether the timing feature of the terminal is abnormal.

[0082] Figure 2 The embodiment is based on Figure 1 The preferred solution of the time series feature prediction value calculation process corresponding to the embodiment is obtained, and the corresponding process is Figure 1 The embodiments are similar, so Figure 2 The description of the above embodiment is relatively simple. Please refer to Figure 1 The embodiment described.

[0083] Accordingly, in order to further illustrate the process of detecting the terminal timing anomaly according to the present invention, the present invention provides a method embodiment for detecting the terminal timing anomaly based on the timing feature prediction value, such as Figure 3 Said, including:

[0084] S31: Obtaining the current timing characteristic information of the terminal;

[0085] S32: matching the time series feature information with the time series prediction model based on the time series prediction algorithm ensemble learning and the time series feature prediction value calculated and generated by the mean absolute percentage error rate algorithm;

[0086] S33: Determine whether the match is successful. If not, determine that it is abnormal and issue a warning; if so, return to S32.

[0087] Figure 3 The embodiments can effectively improve detection accuracy and reduce the false alarm rate of detection.

[0088] The embodiment of the present invention further provides an electronic device, such as Figure 4As shown, the present invention can be implemented Figure 1-3 In the process of the embodiment shown, the electronic device includes: a housing 41, a processor 42, a memory 43, a circuit board 44 and a power circuit 45, wherein the circuit board 44 is arranged inside the space enclosed by the housing 41, and the processor 42 and the memory 43 are arranged on the circuit board 44; the power circuit 45 is used to supply power to various circuits or devices of the above-mentioned electronic device; the memory 43 is used to store executable program codes; the processor 42 runs the program corresponding to the executable program code by reading the executable program code stored in the memory, so as to execute the method described in the above-mentioned embodiment.

[0089] For details on the specific execution process of the above steps by the processor 42 and the steps further executed by the processor 42 by running the executable program code, please refer to the present invention. Figure 1-3 The description of the illustrated embodiment will not be repeated here.

[0090] This electronic device exists in many forms, including but not limited to:

[0091] (1) Mobile communication devices: These devices are characterized by their mobile communication functions and their main purpose is to provide voice and data communications. These terminals include: smart phones (such as iPhone), multimedia phones, functional phones, and low-end phones.

[0092] (2) Ultra-mobile personal computer devices: These devices fall into the category of personal computers, have computing and processing capabilities, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices, such as iPad.

[0093] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players (such as iPods), handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0094] (4) Server: A device that provides computing services. The server consists of a processor, hard disk, memory, system bus, etc. The server is similar to a general computer architecture, but because it needs to provide highly reliable services, it has higher requirements in terms of processing power, stability, reliability, security, scalability, and manageability.

[0095] (5) Other electronic devices with data interaction functions.

[0096] An embodiment of the present invention further provides a computer-readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the method described in the above embodiment.

[0097] The present invention avoids the bottleneck of predicting the terminal timing characteristics by a single timing prediction algorithm through the timing prediction model based on the integrated learning of the timing prediction algorithm and the calculation of the terminal timing characteristics prediction value, reflects the prediction ability of strong robustness, and effectively improves the accuracy of the terminal timing characteristics prediction value. Considering the changes in the trend amplitude of the timing characteristics, the mean absolute percentage error rate is introduced to calculate the prediction value range coefficient. The prediction value obtained by the timing prediction model based on the integrated learning of the timing prediction algorithm and the prediction value range coefficient are comprehensively calculated to obtain the trend fluctuation range of the terminal timing characteristics prediction value, and the trend fluctuation range is integrated into the terminal timing characteristics prediction value. Based on such a timing characteristics prediction value, the terminal timing anomaly is detected, which can effectively improve the detection efficiency and detection accuracy, while reducing the detection false alarm rate. The present invention can adapt to multiple scenarios and is suitable for long-term operation on the client side.

[0098] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A terminal timing feature detection method, characterized in that: include: Get terminal log data; Performing statistics on the log data to obtain timing characteristic information of the terminal; Constructing a time series training set and a time series verification set according to the time series feature information; The time series training set and the time series verification set are calculated by a time series prediction model based on time series prediction algorithm ensemble learning to obtain a model prediction value matrix and a model verification value matrix respectively; The model prediction value matrix and the model verification value matrix are calculated by a mean absolute percentage error rate algorithm to obtain a prediction value range coefficient; Calculate the product of each prediction value in the model prediction value matrix and its corresponding prediction value range coefficient to obtain the time series feature prediction value of the terminal, and the time series feature prediction value is used to detect whether the time series feature of the terminal is abnormal; The time series prediction model based on the time series prediction algorithm ensemble learning is used to calculate the time series training set and the time series verification set to obtain a model prediction value matrix and a model verification value matrix, respectively, including: Determine the time series prediction algorithm applicable to the scenario, and debug and configure each time series prediction algorithm according to the scenario requirements; Perform cross-validation calculations on the time series training set and the time series verification set to meet a preset number of times through the time series prediction algorithms and the Pearson correlation coefficient algorithm to obtain a model prediction value matrix and a model verification value matrix; The model prediction value matrix and the model verification value matrix are calculated by the mean absolute percentage error rate algorithm to obtain the prediction value range coefficient, including: By using the mean absolute percentage error rate algorithm, each predicted value in the model prediction value matrix and its corresponding verification value in the model verification value matrix are calculated respectively, and the calculation results are filtered to obtain the fluctuation range coefficient corresponding to each predicted value in the model prediction value matrix.

2. The method according to claim 1, characterized in that The step of constructing a time series training set and a time series verification set according to the time series feature information specifically includes: Performing row-to-column calculation on the time series feature information to obtain a time series value matrix of each time series feature; Filtering the time series value matrix whose eigenvalues ​​do not meet the specified threshold; According to the preset time range rule, each time series value matrix obtained by filtering is divided according to the time range to obtain a time series training set and a time series verification set.

3. The method according to claim 2, characterized in that The cross validation calculation specifically includes: Inputting the time series training set into each time series prediction algorithm in the form of parameters to obtain the time series prediction value under each time series prediction algorithm; Inputting the timing verification set into each timing prediction algorithm in the form of parameters to obtain the timing verification value under each timing prediction algorithm; The Pearson correlation coefficient algorithm is used to calculate the Pearson correlation coefficient between the time series prediction value and the time series verification value under each time series prediction algorithm.

4. The method according to claim 3, characterized in that After performing cross-validation calculations on the time series training set and the time series verification set to meet a preset number of times through the time series prediction algorithms and the Pearson correlation coefficient algorithm, the method further includes: Calculate the mean of the Pearson correlation coefficients obtained under each time series prediction algorithm; According to the mean calculation result, a weight is assigned to each time series prediction algorithm; the larger the mean result, the greater the weight assigned to the corresponding time series prediction algorithm.

5. The method according to claim 4, characterized in that The obtaining of the model prediction value matrix and the model verification value matrix specifically includes: According to the weights of each time series prediction algorithm, the time series prediction values ​​obtained under each time series prediction algorithm are weighted averaged to obtain a model prediction value matrix; the time series verification values ​​obtained under each time series prediction algorithm are weighted averaged to obtain a model verification value matrix.

6. The method according to claim 1, characterized in that Also includes: The current timing feature information of the terminal is obtained, and the timing feature information is matched with the timing feature prediction value. If the match fails, it is determined to be abnormal and an early warning is issued.

7. An electronic device, characterized in that: The electronic device comprises: a housing, a processor, a memory, a circuit board and a power supply circuit, wherein the circuit board is placed inside the space enclosed by the housing, and the processor and the memory are arranged on the circuit board; the power supply circuit is used to supply power to various circuits or devices of the above-mentioned electronic device; the memory is used to store executable program codes; the processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to execute any method described in claims 1-6.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the method described in any one of claims 1-6.

Citation Information

Patent Citations

  • Transaction index abnormality monitoring method based on deep learning model LSTM

    CN108197845A

  • Abnormal flow monitoring method, device and apparatus and storage medium

    CN110839016A

  • Model-based traffic anomaly monitoring method, device and equipment and storage medium

    CN110880984A

  • Hybrid cloud scene-oriented time series data anomaly prediction method based on ensemble learning technology

    CN112131212A

  • Abnormal behavior prediction method and device

    CN112995195A