KPI anomaly detection method, system and medium based on function-type data analysis

Abstract

The present invention provides a KPI anomaly detection method, system, and medium based on functional data analysis, comprising the following steps: fitting discrete original KPI time series data into a dynamic function curve; extracting features of the dynamic function curve; inputting the extracted features of the dynamic function curve into a classification model, and performing anomaly identification on the features based on the classification model; and obtaining an identification result of whether the original KPI time series data output by the classification model is abnormal. In order to solve the problems of low detection accuracy, low efficiency, and inability to be applied to multi-scenario or multi-dimensional anomaly detection in the prior art KPI anomaly detection methods, the present invention converts the original KPI data into a dynamic function curve and converts the original data into functional data for analysis, which has stronger universality, reduces the workload of operation and maintenance sample data collection and cleaning, and effectively improves detection accuracy and efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of KPI anomaly detection, and in particular relates to a KPI anomaly detection method, system and medium thereof based on function-type data analysis. Background Art

[0002] The development of network information technology is driving advancements in enterprise IT systems. Faced with increasingly complex information systems, traditional manual operations and maintenance (O&M) are no longer able to meet the intelligent management requirements of enterprises. This has necessitated the emergence of intelligent O&M technology. Intelligent O&M encompasses many key scenarios and technologies, including monitoring, analysis, and decision-making for large-scale distributed systems. Key performance indicator (KPI) anomaly detection is a core underlying technology for intelligent O&M of internet services, and most intelligent O&M scenarios rely on the results of KPI anomaly detection.

[0003] KPI data is a type of meaningful time series data acquired through periodic sampling, formatted as (timestamp, value). Depending on the scenario and application requirements, it can be categorized into single-dimensional KPI anomaly detection and multi-dimensional KPI anomaly detection. Single-dimensional KPI anomaly detection focuses on anomalies at the indicator level (such as sudden spikes or jitter in KPIs) and is currently a major research direction. Unlike single-dimensional anomalies, multi-dimensional KPI anomaly detection focuses on anomalies at the entity level (e.g., servers, spacecraft, and other industrial equipment). Abnormal events in an entity often cause simultaneous anomalies in multiple indicators, often with certain correlations between them. Therefore, multi-dimensional KPI anomaly detection is more challenging. Existing anomaly detection models can be roughly divided into two categories: traditional methods and machine learning-based methods. However, due to issues such as periodic diversity, concept drift, and a scarcity of anomaly label samples, some current models lack high accuracy in practical applications.

[0004] Through the above analysis, the problems and defects of the existing technology are: the existing KPI anomaly detection method has low detection accuracy and low detection efficiency, and cannot be applied to multi-scenario or multi-dimensional KPI anomaly detection. Summary of the Invention

[0005] The purpose of the present invention is to address the deficiencies of the above-mentioned background technology and provide a KPI anomaly detection method based on functional data analysis to improve detection accuracy and efficiency.

[0006] The technical solution adopted by the present invention is: a KPI anomaly detection method based on functional data analysis, comprising the following steps:

[0007] Fit the discrete original KPI time series data into a dynamic function curve;

[0008] Extract the features of dynamic function curves;

[0009] Inputting the extracted features of the dynamic function curve into a classification model, and performing abnormality identification on the features based on the classification model;

[0010] Obtain an identification result of whether the original KPI time series data output by the classification model is abnormal.

[0011] In the above technical solution, the process of fitting the discrete original KPI time series data into a dynamic function curve includes: calculating the variance of the original KPI time series data and the fitted dynamic function curve and penalizing the smoothing term as the sum of the square errors between the original KPI time series data and the experimental prediction value; by minimizing the sum of the square errors, solving the basis function coefficients of the dynamic function curve corresponding to the original KPI time series data; calculating the dynamic function curve after fitting the original KPI time series data based on the basis function coefficients and the basis function, and the type of basis function can be selected as Fourier basis function or B-spline basis function according to the actual type.

[0012] In the above technical solution, the expression for the sum of square errors between the original KPI time series data and the experimental prediction value is:

[0013]

[0014]

[0015] Where sse represents the sum of squared errors, n represents the number of time series values; y iq represents the time series value of the qth sample point of the i-th original KPI time series data; x i (t q ) represents the value of the qth time point of the dynamic function curve corresponding to the i-th original KPI time series data; t represents the time point; λ represents the experimental prediction value; x i (s) represents the smooth function curve corresponding to the i-th original KPI time series data; D m Represents x i (t) is the mth derivative; ∫ds represents the integral of the square of the mth derivative of the function curve;

[0016] x i (t) represents the dynamic function curve corresponding to the i-th original KPI time series data; K represents the number of basis functions; c i，k represents the kth basis function coefficient of the dynamic function curve corresponding to the i-th original KPI time series data; φ k (t) represents the kth basis function.

[0017] In the above technical solution, the FLPP function-type feature extraction method is used to extract the features of the dynamic function curve.

[0018] The process of extracting the features of the dynamic function curve using the FLPP function-based feature extraction method in the above technical solution includes:

[0019] Calculate similarity matrix metrics based on the dynamic function curves corresponding to discrete KPI time series data;

[0020] The objective function of FLPP is obtained by calculating the total weight between the dynamic function curves in the low-dimensional space according to the similarity matrix measurement;

[0021] Convert the objective function of FLPP into eigenvalue decomposition;

[0022] Arrange the decomposed eigenvalues from small to large, use the eigenvectors corresponding to the top eigenvalues as the coefficients of the projection function, and calculate the projection eigenfunction based on the coefficients of the projection function;

[0023] The dynamic function curves in the original dimension are mapped to the low-dimensional space through the projection feature function to form the corresponding features of each dynamic function curve after dimensionality reduction.

[0024] The objective function of FLPP in the above technical solution is expressed as:

[0025]

[0026]

[0027] F represents the total weight between the dynamic function curves in the low-dimensional space; y i represents the dynamic function curve corresponding to the i-th original KPI time series data in the low-dimensional space; j Represents the dynamic function curve corresponding to the j-th original KPI time series data in the low-dimensional space; s ij Represents the dynamic function curve x corresponding to the i-th and j-th original KPI time series data under the original dimension i (t) and x j (t); N represents the number of original KPI time series data; c i Represents x i (t) basis function coefficients; c j Represents x j (t) basis function coefficients; c i ∈N o (c j ) represents x i (t) is the property of x j The basis function coefficient of one of the oth nearest neighbors of (t); p is the heat kernel parameter;

[0028] a(t) represents the projection function; x i(t) represents the dynamic function curve corresponding to the i-th original KPI time series data; x j (t) represents the dynamic function curve corresponding to the j-th original KPI time series data; <a,x i (t)> represents the projection of (t) under the projection function a(t); <a,x j (t)> represents x j (t) is projected under the projection function a(t).

[0029] In the above technical solution, the process of inputting the extracted features of the dynamic function curve into the classification model includes:

[0030] Determine whether the extracted dynamic function curve features are linearly separable or nonlinearly separable;

[0031] If the extracted features of the dynamic function curve are linearly separable, they are input into the classification model:

[0032] If the extracted features of the dynamic function curve are nonlinearly separable, the features of the dynamic function curve are converted into linearly separable features in a high-dimensional space through kernel function inner product calculation and then input into the classification model.

[0033] In the above technical solution, the process of building the classification model includes:

[0034] Get several KPI time series data that do not contain abnormal data.

[0035] Fit each KPI time series data into a dynamic function curve and extract its features to form a training data set;

[0036] The classification model is trained using the training data set to create the decision boundary of the classification model. After the training is completed, the parameters of the classification model are obtained.

[0037] The present invention provides a KPI anomaly detection system based on function-type data analysis, comprising: a dynamic function curve fitting module, a dynamic function curve feature extraction module and a KPI data anomaly identification module;

[0038] Dynamic function curve fitting module, used to obtain original KPI time series data and fit the original KPI time series data into a dynamic function curve;

[0039] A dynamic function curve feature extraction module is used to extract the features of the dynamic function curve using an FLPP function-based feature extraction method;

[0040] The KPI data anomaly identification module is used to identify anomalies in KPI data using a classification model based on the extracted features.

[0041] The present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to execute the steps of the KPI anomaly detection method based on function-type data analysis as described in the above technical solution.

[0042] The beneficial effects of the present invention are: the present invention proposes a KPI anomaly detection method based on functional data analysis. Compared with traditional data analysis (such as multivariate statistical analysis), the present invention converts the original KPI data into a dynamic function curve, and converts the original data into functional data for analysis, which has stronger universality. The present invention fits the original KPI data through dynamic functions, which can greatly reduce the workload of operation and maintenance sample data collection and cleaning. At the same time, it can identify the deeper dynamic evolution laws in the KPI data and analyze the essential characteristics of complex data. The present invention further identifies the deeper dynamic evolution laws in the KPI data by extracting features from the dynamic function curve, and deeply analyzes the essential characteristics of complex data. The present invention classifies whether the KPI data is abnormal based on the characteristics of the dynamic function curve through a classification model, thereby ensuring the efficiency of the identification process.

[0043] The present invention uses basis functions and penalty coefficients to fit the original KPI data to obtain a dynamic function curve, which can quickly calculate non-periodic and periodic data and has better fitting advantages, thereby fitting better local features. Since the feature dimension D is much larger than the number of data N, it is difficult to use feature decomposition to obtain the correct eigenvalues and eigenvectors. The present invention uses a method unique to functional data and uses functional data calculations that are only related to the number of basis functions (m) and the number of data (n). The difference between m and n is greatly reduced, and it is relatively simple to calculate eigenvalues and characteristic functions, thereby obtaining a more reasonable and intuitive data interpretation.

[0044] This paper uses the FLPP functional feature extraction method to extract the features of dynamic function curves, reducing the computational effort while maintaining the proximity relationships of high-dimensional function spaces within low-dimensional function spaces, thereby ensuring data accuracy. This method can identify deeper dynamic evolution patterns within KPI data and analyze the essential characteristics underlying complex data. Compared to traditional methods, FLPP produces more detailed, intuitive, and reasonable features for KPI sample analysis after dimensionality reduction, enabling more efficient KPI anomaly detection.

[0045] The present invention proposes corresponding data processing methods for extracted nonlinearly separable and linearly separable features, making the classification model more adaptable and the feature extraction more detailed. The classification model of the present invention uses a linearly separable method, which is computationally simple, highly interpretable, and has fewer parameters, making it more computationally efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1This is a schematic diagram of a KPI anomaly detection method based on functional data analysis provided by an embodiment of the present invention;

[0047] Figure 2 This is a flow chart of a KPI anomaly detection method based on functional data analysis provided by an embodiment of the present invention;

[0048] Figure 3 Schematic diagram of a KPI anomaly detection system based on functional data analysis provided by an embodiment of the present invention;

[0049] Figure 4 This is a comparison diagram of the two-dimensional visualization of low-noise analog data by FLPP used in the present invention provided by an embodiment of the present invention and the prior art;

[0050] Figure 5 This is a comparison diagram of the FLPP used in the present invention for two-dimensional visualization of high-noise simulated data provided by an embodiment of the present invention and the prior art;

[0051] Figure 6 This is a performance graph of the FLPP model verified using the ECG public dataset provided by an embodiment of the present invention;

[0052] Figure 7 This is a performance graph of the FLPP model verified using the FaceAll public dataset provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments to facilitate a clear understanding of the present invention, but they do not constitute a limitation to the present invention.

[0054] like Figure 1 As shown, the KPI anomaly detection method based on functional data analysis (FDA) provided in an embodiment of the present invention includes:

[0055] S1, fitting the discrete original KPI time series data into a dynamic function curve;

[0056] S2, extracting the features of the dynamic function curve;

[0057] S3, inputting the extracted features of the dynamic function curve into a classification model, and performing abnormality identification on the features based on the classification model;

[0058] S4, obtaining a recognition result of whether the original KPI time series data output by the classification model is abnormal.

[0059] In step S1, the original KPI time series data is first obtained, and then the discrete original KPI time series data is fitted into a dynamic function curve.

[0060] In step S2, the features of each dynamic function curve are extracted using the FLPP function-based feature extraction method.

[0061] In step S3, the extracted dynamic function curve features are respectively input into the classification model. The classification model identifies whether the corresponding original KPI time series data is abnormal based on the input corresponding dynamic function curve features, thereby identifying abnormal KPI data in the original discrete KPI data.

[0062] like Figure 2 As shown, the KPI anomaly detection method based on functional data analysis provided by an embodiment of the present invention specifically includes the following steps:

[0063] Read in KPI historical data, which is discrete original KPI time series data;

[0064] Fitting into a dynamic function curve: Fit each piece of original KPI time series data into a dynamic function curve;

[0065] Functional local projection: Use functional local projection to extract the features of dynamic function curves;

[0066] Support vector machine anomaly detection: inputting the extracted features of the dynamic function curve into a classification model, and performing anomaly identification on the features based on the classification model;

[0067] Obtain an identification result of whether the original KPI time series data output by the classification model is abnormal.

[0068] The discrete original KPI time series data includes several original KPI time series data. The original N KPI time series data include x1, x2, x3...x n , each KPI time series data x i The vector consists of KPI values at different time periods.

[0069] The present invention proposes a KPI anomaly detection method based on functional data analysis. Compared with traditional data analysis (such as multivariate statistical analysis), the present invention converts the original KPI data into a dynamic function curve, and converts the original data into functional data for analysis, which has stronger universality. The present invention fits the original KPI data with a dynamic function, which can greatly reduce the workload of operation and maintenance sample data collection and cleaning. At the same time, it can identify the deeper dynamic evolution laws in the KPI data and analyze the essential characteristics of complex data. The present invention further identifies the deeper dynamic evolution laws in the KPI data by extracting features from the dynamic function curve, and deeply analyzes the essential characteristics of complex data. The present invention classifies whether the KPI data is abnormal based on the characteristics of the dynamic function curve through a classification model, thereby ensuring the efficiency of the identification process.

[0070] The process of fitting any original KPI time series data into a dynamic function curve provided by the embodiment of the present invention includes:

[0071] Calculate the variance of the original KPI time series data and the fitted dynamic function curve and penalize the smoothing term as the sum of the square errors between the original KPI time series data and the experimental prediction value; by minimizing the sum of the square errors, obtain the basis function coefficients of the dynamic function curve corresponding to the original KPI time series data; calculate the dynamic function curve after fitting the original KPI time series data based on the basis function coefficients and the basis function.

[0072] The expression for the sum of square errors between the original KPI time series data and the experimental prediction value is:

[0073]

[0074]

[0075] Where sse represents the sum of squares for error, n represents the number of time series values; y iq represents the time series value of the qth sample point of the i-th original KPI time series data; x i (t q ) represents the value of the qth time point of the dynamic function curve corresponding to the i-th original KPI time series data; t represents the time point; λ represents the experimental prediction value, which usually defaults to 10 -8 To 1, increase exponentially to get the lambda with the smallest sse; x i (s) represents the smooth function curve corresponding to the i-th original KPI time series data; D m Represents x i (t) is the mth derivative; ∫ds represents the integral of the square of the mth derivative of the function curve;

[0076] x i (t) represents the dynamic function curve corresponding to the i-th original KPI time series data; K represents the number of basis functions; c ik represents the kth basis function coefficient of the dynamic function curve corresponding to the i-th original KPI time series data; φ k (t) represents the kth basis function, and B-spline or Fourier basis function can be selected according to the characteristics of the data.

[0077] x i Substitute the expression of (t) into sse λ After applying formula (i), we can get the following formula:

[0078]

[0079] This formula is used to express the error of the original KPI data in the dynamic function curve after fitting, and solve the basis function coefficient c by minimizing the error. ik , that is, the coefficient of the function curve after fitting is obtained by the least square method. Finally, the coefficient of the basis function c is obtained according to ik Calculate the dynamic function curve x i The above process transforms the original finite-dimensional discrete data into a continuous infinite-dimensional continuous curve, which can be applied to anomaly detection of more complex and diverse KPI samples. This is why this model has greater universality than traditional data analysis.

[0080] The present invention uses basis functions and penalty coefficients to fit the original KPI data to obtain a dynamic function curve, which can quickly calculate non-periodic and periodic data and has better fitting advantages, thereby fitting better local features. Since the feature dimension D is much larger than the number of data N, it is difficult to use feature decomposition to obtain the correct eigenvalues and eigenvectors. The present invention uses a method unique to functional data and uses functional data calculations that are only related to the number of basis functions (m) and the number of data (n). The difference between m and n is greatly reduced, and it is relatively simple to calculate eigenvalues and characteristic functions, thereby obtaining a more reasonable and intuitive data interpretation.

[0081] After converting discrete data into functional data, feature extraction is performed using methods specific to functional data. Common models include Functional Principal Component Analysis (FPCA) and Functional Local Projection (FLPP). The main idea of FPCA is to find a projection feature function based on the functional data that maximizes the variance of the reduced data. The original dimensional data is then projected into a low-dimensional space using this projection feature function. FPCA-based principal component feature extraction removes redundant information from the data, achieving dimensionality reduction. The main idea of FLPP, on the other hand, is to use the similarity graph of the original dimensional space to maintain the local manifold structure of the data in a low-dimensional space. While preserving this local manifold structure, the projected feature function that maximizes the local variance of the reduced data is found. The original dimensional data is then projected into a low-dimensional space using this projection feature function for feature extraction. This removes redundant information while preserving the local structure, achieving dimensionality reduction. Functional feature extraction based on FPCA focuses more on the global features of KPI samples, has unclear relationships between local features, and is sensitive to outliers. However, the functional feature extraction method based on FLPP focuses more on local features between samples. Therefore, FLPP performs better in KPI anomaly detection.

[0082] The FLPP function-based feature extraction method uses the similarity graph of the original dimensional function space to maintain the manifold structure of the data in the low-dimensional function space, and the similarity matrix measurement is also calculated based on discrete sample points.

[0083] The process of extracting the features of the dynamic function curve using the FLPP function-based feature extraction method in this specific embodiment includes:

[0084] The similarity matrix metric is calculated based on the dynamic function curves corresponding to the discrete KPI time series data. The purpose is to maintain the similarity of the original dimensional space in the low-dimensional space and use the similarity graph of the original dimensional space to maintain the local manifold structure of the data in the low-dimensional space.

[0085] The objective function of FLPP is obtained by calculating the total weight between each dynamic function curve in the low-dimensional space based on the similarity matrix metric. The objective function of FLPP is converted into eigenvalue decomposition. The decomposed eigenvalues are arranged from small to large, and the eigenvectors corresponding to the top eigenvalues are used as the coefficients of the projection function. The projection characteristic function is calculated based on the coefficients of the projection function. That is, on the basis of preserving the local epidemic structure, the projection characteristic function that maintains the maximum local variance of the local epidemic data after dimensionality reduction is determined.

[0086] The dynamic function curves in the original dimension are mapped to the low-dimensional space through the projection feature function to form the corresponding features of each dynamic function curve after dimensionality reduction.

[0087] The specific process of extracting the features of the dynamic function curve using the FLPP function-based feature extraction method in this specific embodiment includes:

[0088] (1) The following formula is used to calculate the dynamic function curve corresponding to the discrete KPI time series data and calculate the similarity matrix measurement:

[0089]

[0090] s ij Represents the dynamic function curve x corresponding to the i-th and j-th original KPI time series data in the original dimensional space i (t) and x j (t) similarity measure, c i Represents x i (t) basis function coefficients; c j Represents x j (t) basis function coefficients; its calculation method uses x i (t) The corresponding basis function coefficient c i Calculation; c i ∈N o (c j ) represents x i (t) is the property of xj The basis function coefficient of one of the oth nearest neighbors of (t); p is the heat kernel parameter, ranging from 0 to 1, and defaults to 1.

[0091] (2) The following formula is used to obtain the total weight between the dynamic function curves in the low-dimensional space to obtain the objective function of FLPP;

[0092]

[0093] F represents the total weight between the dynamic function curves in the low-dimensional space; y i represents the dynamic function curve corresponding to the i-th original KPI time series data in the low-dimensional space; j represents the dynamic function curve corresponding to the jth original KPI time series data in the low-dimensional space; N represents the number of original KPI time series data; a(t) represents the projection function; x i (t) represents the dynamic function curve corresponding to the i-th original KPI time series data; x j (t) represents the dynamic function curve corresponding to the j-th original KPI time series data; <a,x i (t)> represents the projection of (t) under the projection function a(t); <a,x j (t)> represents x j (t) is projected under the projection function a(t).

[0094] The objective function is simplified as follows:

[0095]

[0096]

[0097] F represents the simplification of the objective function; L represents the simplification of the objective function; ij The Laplace matrix corresponding to the similarity matrix formed; X represents the dynamic function curve x in the n original dimensional space i (t); W represents the inner product between basis functions φ(t); C represents the basis function coefficient matrix.

[0098] (3) The simplified FLPP objective function is converted into eigenvalue decomposition using the Lagrange multiplier method:

[0099] WCLC T Wa=λWCDC T Wa

[0100] Arrange the decomposed eigenvalues from small to large, and the eigenvectors corresponding to the first d eigenvalues are the projection function [a1(t),…,a d (t)]. Substitute the d projection function coefficients into the following formula to calculate the projection characteristic function a(t):

[0101]

[0102] Among them, a l represents the lth projection function coefficient; φ l (t) represents the lth basis function.

[0103] (4) Calculate the projection characteristic function a(t) and the dynamic function curve x in n original dimensional spaces i (t) The inner product of the function matrix X formed can extract n dynamic function curves after dimensionality reduction x i Characteristic curve after (t).

[0104] Dynamic function curve feature extraction of KPI data based on FLPP can identify deeper dynamic evolution patterns in KPI data and analyze the essential characteristics of complex data. Compared with the KPI sample features after dimensionality reduction obtained by traditional methods, the features are more detailed, intuitive, and reasonable, which makes it possible to detect KPI anomalies more efficiently.

[0105] Based on the significant features extracted by the functional data analysis model, the subsequent classification models such as support vector machines can be used to efficiently identify anomalies in KPI data.

[0106] This paper uses the FLPP functional feature extraction method to extract the features of dynamic function curves, reducing the computational effort while maintaining the proximity relationships of high-dimensional function spaces within low-dimensional function spaces, thereby ensuring data accuracy. This method can identify deeper dynamic evolution patterns within KPI data and analyze the essential characteristics underlying complex data. Compared to traditional methods, FLPP produces more detailed, intuitive, and reasonable features for KPI sample analysis after dimensionality reduction, enabling more efficient KPI anomaly detection.

[0107] In this specific embodiment, the process of inputting the extracted features of the dynamic function curve into the classification model includes:

[0108] Determine whether the extracted dynamic function curve features are linearly separable or nonlinearly separable;

[0109] If the extracted features of the dynamic function curve are linearly separable, they are input into the classification model:

[0110] If the extracted characteristics of the dynamic function curve are nonlinearly separable, by introducing the kernel function, the essential idea is to map the low-dimensional dynamic function curve characteristics to the high-dimensional space through the kernel function, so that the low-dimensional dynamic function curve characteristics are linearly separable in the high dimension; according to the inner product calculation of the kernel function, the low-dimensional dynamic function curve characteristics are converted into linear separable in the high-dimensional space, and then input into the classification model. The classification model is obtained according to the training, and the linear separable method is used to identify KPI abnormal data.

[0111] The present invention proposes corresponding data processing methods for the extracted nonlinearly separable and linearly separable features, respectively, so as to make the use of the classification model more adaptable and the feature extraction more detailed.

[0112] The classification model of this specific embodiment uses the characteristics of the low-dimensional dynamic function curve of the linear separable method for classification. The construction process of the classification model includes:

[0113] Get several KPI time series data that do not contain abnormal data.

[0114] Fit each KPI time series data into a dynamic function curve and extract its features to form a training data set;

[0115] The classification model is trained using the training data set to create the decision boundary of the classification model. After the training is completed, the parameters of the classification model are obtained.

[0116] The classification model of the present invention adopts a linear separable method, which is simple to calculate, has strong interpretability, has fewer parameters, and is more conducive to calculation.

[0117] The classification model of this specific embodiment adopts the support vector machine (SVM) algorithm, which is widely used in the field of machine learning and has achieved relatively good results. Its basic idea is to correctly distinguish data and maximize the geometric separation of the separating hyperplane. The maximum separation can ensure good classification and prediction capabilities for unknown new instances. The support vector machine is essentially solving a convex quadratic programming problem under constraints, transforming it into an optimization problem of the dual variable through Lagrangian duality, and obtaining the optimal solution of the original problem by solving the dual problem equivalent to the original problem.

[0118] The classification model is based on Lagrange duality. The dual problem of the original problem is the minimax problem:

[0119]

[0120] For each low-dimensional characteristic inequality constraint of a dynamic function curve, the Lagrange multiplier α is introduced and the Lagrange function is defined:

[0121]

[0122] Among them, L(w,b,a) represents the loss function about parameters w, b, a, α represents the Lagrange multiplier constraint term, and the Lagrange multiplier method is iteratively solved; w represents the normal vector of the hyperplane segmentation, b represents the displacement, z i Represents the hyperplane function of the i-th low-dimensional dynamic function curve feature, h i Represents i low-dimensional dynamic function curve features; after training, the parameters are obtained and the low-dimensional dynamic function curve features are classified to detect whether they are abnormal. The output of the classification model is 0 or 1, where 0 indicates abnormality and 1 indicates normality.

[0123] like Figure 3 As shown, the KPI anomaly detection system based on functional data analysis provided by the present invention includes:

[0124] Dynamic function curve fitting module 1, used to obtain original KPI time series data and fit the discrete original KPI time series data into a dynamic function curve;

[0125] Dynamic function curve feature extraction module 2, used to extract the features of the dynamic function curve using the FLPP function-based feature extraction method;

[0126] The KPI data anomaly identification module 3 is used to identify anomalies in the KPI data using a classification model based on the extracted features.

[0127] The technical solution of the present invention will be further described below with reference to specific embodiments.

[0128] The present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor performs the following steps of the KPI anomaly detection method based on functional data analysis as described in the above technical solution:

[0129] An application embodiment of the present invention provides a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the KPI anomaly detection method based on functional data analysis.

[0130] An application embodiment of the present invention provides an information data processing terminal, characterized in that the information data processing terminal is used to execute the KPI anomaly detection method based on functional data analysis.

[0131] The embodiments of the present invention have achieved some positive results during the development or use process, and indeed have great advantages over the existing technology. The following content describes them in conjunction with data, charts, etc. from the experimental process.

[0132] We randomly generated 1200 simulated data points based on the following function (parameters were sampled from a Gaussian distribution, with 300 sample points per class and 200 feature dimensions). We then reduced the simulated data to a two-dimensional space for data visualization using traditional principal component analysis (PCA), locality preserving projection (LPP), functional principal component analysis (FPCA), and functional locality preserving projection (FLPP). We then compared the performance of different algorithms.

[0133] like Figure 4 、 5 As shown in the figure, two-dimensional visualization comparisons of low-noise and high-noise simulated data are performed. It can be seen that the FLPP used in the present invention has better feature dimensionality reduction performance, especially in tasks with a large number of noise points. It performs better and is therefore very suitable for feature extraction from high-dimensional noisy KPI data.

[0134] In addition, experiments on the ECG and FaceAll public datasets also verified the performance of the FLPP model. Dimensionality reduction based on PCA, LPP, FPCA, and FLPP was performed on these two datasets, and the low-dimensional features were input into KNN for classification. Figure 6 、 Figure 7 The classification accuracy of KNN under different dimensions is given, and the results show that the FLPP used in the present invention has better performance. Figure 6 Performance graph of the FLPP model validated for the ECG public dataset; Figure 7 Performance graph of the FLPP model validated on the FaceAll public dataset.

[0135] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art will appreciate that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0136] The contents not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A KPI anomaly detection method based on functional data analysis, characterized by: The following steps are involved: Fit the discrete original KPI time series data into a dynamic function curve; Extracting features of the dynamic function curve; Inputting the extracted features of the dynamic function curve into a classification model, and performing abnormality identification on the features based on the classification model; Obtaining an identification result of whether the original KPI time series data output by the classification model is abnormal; The process of fitting discrete original KPI time series data into a dynamic function curve includes: calculating the variance between the original KPI time series data and the fitted dynamic function curve and penalizing the smoothing term as the sum of squared errors between the original KPI time series data and the experimental prediction value; minimizing the sum of squared errors to obtain the basis function coefficients of the dynamic function curve corresponding to the original KPI time series data; and obtaining the dynamic function curve fitted to the original KPI time series data based on the basis function coefficients and the basis function calculation. The expression for the sum of square errors between the original KPI time series data and the experimental prediction value is: ； ； Where sse represents the sum of squared errors, and n represents the number of time series values; y iq represents the time series value of the qth sample point of the i-th original KPI time series data; x i (t q ) represents the value of the qth time point of the dynamic function curve corresponding to the i-th original KPI time series data; t represents the time point; λ represents the experimental prediction value; x i (s) represents the smooth function curve corresponding to the i-th original KPI time series data; D m express x i (t) is the mth derivative; ∫ds represents the integral of the square of the mth derivative of the function curve; x i (t) represents the dynamic function curve corresponding to the i-th original KPI time series data; K represents the number of basis functions; c i，k represents the kth basis function coefficient of the dynamic function curve corresponding to the i-th original KPI time series data; φ k (t) represents the kth basis function.

2. The KPI anomaly detection method based on functional data analysis according to claim 1, characterized in that: The FLPP function-type feature extraction method is used to extract the features of the dynamic function curve.

3. The KPI anomaly detection method based on functional data analysis according to claim 2, characterized in that: The process of extracting the features of the dynamic function curve using the FLPP function-based feature extraction method includes: Calculate similarity matrix metrics based on the dynamic function curves corresponding to discrete KPI time series data; The objective function of FLPP is obtained by calculating the total weight between the dynamic function curves in the low-dimensional space according to the similarity matrix measurement; Convert the objective function of FLPP into eigenvalue decomposition; Arrange the decomposed eigenvalues from small to large, use the eigenvectors corresponding to the top eigenvalues as the coefficients of the projection function, and calculate the projection eigenfunction based on the coefficients of the projection function; The dynamic function curves in the original dimension are mapped to the low-dimensional space through the projection feature function to form the corresponding features of each dynamic function curve after dimensionality reduction.

4. The KPI anomaly detection method based on functional data analysis according to claim 3, characterized in that: The objective function of the FLPP is expressed as: ； ； F represents the total weight between the dynamic function curves in the low-dimensional space; y i represents the dynamic function curve corresponding to the i-th original KPI time series data in the low-dimensional space; j Represents the dynamic function curve corresponding to the j-th original KPI time series data in the low-dimensional space; s ij Represents the dynamic function curve x corresponding to the i-th and j-th original KPI time series data under the original dimension i (t) and x j (t); N represents the number of original KPI time series data; c i Represents x i (t) basis function coefficients; c j Represents x j (t) basis function coefficients; c i ∈N o (c j ) represents x i (t) is the property of x j The basis function coefficient of one of the oth nearest neighbors of (t); p is the heat kernel parameter; a(t) represents the projection function; x i (t) represents the dynamic function curve corresponding to the i-th original KPI time series data; x j (t) represents the dynamic function curve corresponding to the j-th original KPI time series data; <a,x i (t)> represents the projection of (t) under the projection function a(t); <a,x j (t)> represents x j (t) is projected under the projection function a(t).

5. The KPI anomaly detection method based on functional data analysis according to claim 1, characterized in that: The process of inputting the extracted features of the dynamic function curve into the classification model includes: Determine whether the extracted dynamic function curve features are linearly separable or nonlinearly separable; If the extracted features of the dynamic function curve are linearly separable, they are input into the classification model: If the extracted features of the dynamic function curve are nonlinearly separable, the features of the dynamic function curve are converted into linearly separable features in a high-dimensional space through kernel function inner product calculation and then input into the classification model.

6. The KPI anomaly detection method based on functional data analysis according to claim 5, characterized in that: The construction process of the classification model includes: Get several KPI time series data that do not contain abnormal data. Fit each KPI time series data into a dynamic function curve and extract its features to form a training data set; The classification model is trained using the training data set to create the decision boundary of the classification model. After the training is completed, the parameters of the classification model are obtained.

7. A KPI anomaly detection system based on functional data analysis, characterized by Used to implement the method according to any one of claims 1 to 6, comprising: a dynamic function curve fitting module, a dynamic function curve feature extraction module and a KPI data anomaly identification module; Dynamic function curve fitting module, used to obtain original KPI time series data and fit the original KPI time series data into a dynamic function curve; A dynamic function curve feature extraction module is used to extract the features of the dynamic function curve using an FLPP function-based feature extraction method; The KPI data anomaly identification module is used to identify anomalies in KPI data using a classification model based on the extracted features.

8. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the KPI anomaly detection method based on functional data analysis according to any one of claims 1 to 6.

Images