WO2023245851A1

WO2023245851A1 - Deep inspection optimization method and system based on micro-service architecture

Info

Publication number: WO2023245851A1
Application number: PCT/CN2022/114210
Authority: WO
Inventors: 刘婷雯
Original assignee: 中电信数智科技有限公司
Priority date: 2022-06-20
Filing date: 2022-08-23
Publication date: 2023-12-28
Also published as: CN114996110A

Abstract

A deep inspection optimization method and system based on a micro-service architecture, relating to the technical field of artificial intelligence. The method comprises: constructing a health inspection model, and transversely layering a micro-service architecture system; inputting historical failure data into the health inspection model to obtain a probability of the occurrence of a failure of a physical device corresponding to each module of a transverse layer during the next period of inspection; preprocessing inspection data by means of quadratic interpolation, and constructing a deep inspection model; inputting the preprocessed inspection data into the deep inspection model to obtain an actual value; and comparing a fitted value with the actual value generated by operation of the deep inspection model, and evaluating data defects and hidden function failures possibly existing in the micro-service architecture system.

Description

An in-depth inspection optimization method and system based on microservice architecture

Technical field

The invention belongs to the field of artificial intelligence technology, and in particular relates to an in-depth inspection optimization method and system based on microservice architecture.

Background technique

With the rapid development of computer technology, information network has become an important guarantee for social development. In the microservice architecture system, in order to ensure the normal operation of the functions of each layer, inspection methods are usually used to monitor and perform regular inspections on each layer, so that the health problems of the physical device layer and application layer of the microservice architecture system are effectively solved. However, there is still a lot of room for improvement. Only by well solving the deeper dimensional problem of health inspection can we realize the operation and maintenance data center as the driving force, with AI algorithm as the core, covering infrastructure monitoring, precise fault location and intelligent processing, 3D Professional operation and maintenance service modules such as digital twins and management cockpit, and decompose huge monolithic applications into multiple services through the microservice architecture model.

With the large-scale application of big data and AI, how to ensure a stable operating environment for each layer of the microservice architecture system is a technical bottleneck faced today.

Contents of the invention

The main purpose of the present invention is to overcome the shortcomings and deficiencies of the existing technology, provide an in-depth inspection optimization method and system based on microservice architecture, and construct a health inspection model and in-depth inspection model by horizontally layering the microservice architecture system. The model can automatically implement in-depth inspection of each layer in the microservice architecture system. At the same time, by comparing the actual values generated by the operation of the in-depth inspection model with the fitted values calculated by the ridge regression method, the microservice system can be evaluated as a whole. Possible data defects and hidden functional failures in the architecture enable machines to make decisions instead of humans, thereby achieving complete automation and improving the efficiency of fault detection.

According to one aspect of the present invention, the present invention provides a deep inspection optimization method based on microservice architecture. The method includes the following steps:

S1: Construct a health inspection model and horizontally layer the microservice architecture system; obtain the physical device information corresponding to each module in the horizontal layer and the fault history data of the physical device, and input the fault history data into the health inspection model to obtain the probability of failure of the physical equipment corresponding to each module in the horizontal layer in the next period of inspection;

S2: Preprocess the inspection data through quadratic interpolation to construct a deep inspection model; input the preprocessed inspection data into the deep inspection model to obtain the actual value; calculate the fitting value through the ridge regression method ;

S3: Compare the fitting value with the actual value generated by the in-depth inspection model operation, and evaluate possible data defects and hidden functional failures in the microservice architecture system.

Preferably, building a health inspection model and horizontally layering the microservice architecture system includes:

The microservice architecture system is horizontally divided into seven layers: basic environment layer, supervision management layer, cloud platform, information resource layer, business layer, interface layer and UI display layer;

The health inspection model includes a Markov transfer matrix algorithm model: X(k+1)=X(k)×P

Among them: X(k) represents the state vector of the trend analysis and prediction object at time t=k, P represents the one-step transition probability matrix, and .

Preferably, the preprocessing of inspection data through quadratic interpolation includes:

To perform difference processing on data with uneven sampling at different nodes, the quadratic difference method is used to interpolate every 3 adjacent points to obtain quadratic interpolation. The formula is as follows:

Among them, x is the current value of the classification object; y is the three adjacent points of the classification object; i is the sequence number of the classification object.

Preferably, the construction of an in-depth inspection model includes:

For the time dimension, PositionEncoding is used for timing encoding, and Attention is used to explore the feature correlation of the timing dimension. The formula is as follows:

PositionEncoding=cos2(pos/N)

For the spatial dimension, extract different multi-space dimension features and fuse the multi-space dimension features through MultiHead. The formula is as follows:

Headi＝Attention(Qi,Ki,Vi)

MultiHead(Q,K,V)＝Concat(head1,...,headh)*WO

Among them, d _k represents the dimension of K; N is the adjustable length; Q is the query feature map; K is the feature map to be matched; V is the monitoring data map; headi is the result obtained by time attention, Qi is the i-th group of queries Feature map; Ki is the i-th group of feature maps to be matched; Vi is the i-th group of monitoring data mapping; WO is the feature fusion matrix; Concat is the feature cascade fusion; pos is the data sequence number; PositionEncoding is the position sequence encoding; MultiHead is the multi-head Feature fusion.

Preferably, the fitted value calculated through the ridge regression method includes:

The model formula of the ridge regression method is: ||Xθ-y|| ² +||Γθ|| ²

The formula to prevent over-fitting is: θ(a)=(X ^T X+aI) ^-1 X ^T y

Among them, Find the value of θ if determined.

According to another aspect of the present invention, the present invention also provides an in-depth inspection optimization system based on microservice architecture. The system includes:

The first prediction module is used to build a health inspection model and horizontally layer the microservice architecture system; obtain the physical device information corresponding to each module in the horizontal layer and the fault history data of the physical device, and input the fault history data The health inspection model obtains the probability of failure of the physical equipment corresponding to each module in the horizontal layer in the next period of inspection;

The second prediction module is used to preprocess the inspection data through secondary interpolation to construct a deep inspection model; input the preprocessed inspection data into the deep inspection model to obtain the actual value; use the ridge regression method Calculate the fitting value;

The comparison and evaluation module is used to compare the fitting value with the actual value generated by the in-depth inspection model operation, and evaluate possible data defects and hidden functional failures in the microservice architecture system.

Preferably, the first prediction module builds a health inspection model, and horizontally layering the microservice architecture system includes:

Preferably, the second prediction module preprocessing the inspection data through quadratic interpolation includes:

Among them, x is the current value of the classification object, y is the three adjacent points of the classification object, and i is the sequence number of the classification object.

Preferably, the second prediction module constructing an in-depth inspection model includes:

PositionEncoding=cos2(pos/N)

Headi＝Attention(Qi,Ki,Vi)

MultiHead(Q,K,V)＝Concat(head1,...,headh)*WO

Preferably, the fitting value calculated by the second prediction module through ridge regression method includes:

The model formula of the ridge regression method is: ||Xθ-y|| ² +||Γθ|| ²

The formula to prevent over-fitting is: θ(a)=(X ^T X+aI) ^-1 X ^T y

Among them, Find the value of θ if determined.

Beneficial effects: The present invention conducts in-depth health inspections on the major categories and subcategories of the horizontal basic environment layer, supervision management layer, cloud platform, information resource layer, business layer, interface layer and UI display layer of the microservice architecture system; at the same time, it The combination technology of multiple algorithm models in the field of artificial intelligence makes up for the fact that the existing inspection technology places too much emphasis on physical equipment and application detection and lacks the simulation data to test the fitting values of each level and the calculation of each horizontal layer through the in-depth inspection model. From the perspective of actual value comparison, we can comprehensively evaluate the possible data defects and hidden functional failures of the microservice architecture, and the losses that may be caused to the application platform, so that machines can make decisions instead of people, thus realizing complete automation and intelligence in a true sense. become possible.

The features and advantages of the present invention will become apparent with reference to the following drawings and detailed description of specific embodiments of the invention.

Description of the drawings

Figure 1 is a flow chart of the in-depth inspection optimization method based on microservice architecture;

Figure 2 is a schematic diagram of an in-depth inspection and optimization system based on microservice architecture.

Detailed ways

The technical solutions in the embodiments of the present invention are clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

Example 1

Figure 1 is a flow chart of the in-depth inspection optimization method based on microservice architecture. As shown in Figure 1, the present invention provides an in-depth inspection optimization method based on microservice architecture. The method includes the following steps:

S1: Construct a health inspection model and horizontally layer the microservice architecture system; obtain the physical device information corresponding to each module in the horizontal layer and the fault history data of the physical device, and input the fault history data into the health inspection model to obtain the probability of failure of the physical equipment corresponding to each module in the horizontal layer in the next period of inspection.

Specifically, first, the microservice architecture system is horizontally divided into seven layers: basic environment layer, supervision management layer, cloud platform, information resource layer, business layer, interface layer and UI display layer to build an inspection model.

Secondly, the physical device information corresponding to the modules and sub-modules of each layer is obtained through the relational database of the information resource layer.

Then, the fault history data of the physical device corresponding to each sub-module at each horizontal level of the technical architecture is obtained from the historical fault database.

Finally, the data is involved in the operation of the health inspection model, and the operation result is the failure probability of the physical equipment corresponding to each module of the current horizontal layer in the next period of inspection.

Among them, the Markov transfer matrix algorithm model formula is: X(k+1)=X(k)×P

In the formula: X(k) represents the state vector of the trend analysis and prediction object at time t=k, P represents the one-step transition probability matrix, and vector.

There are currently 100 layer control indicators in this inspection, of which 30 are normal and 70 are abnormal. 60% of the normal monitoring indicators may continue to be normal and 40% may become abnormal [0.6, 0.4].

70% of the normal monitoring indicators in the current layer of this inspection may still be normal, and 30% may turn into abnormalities [0.3, 0.7]. The transition probability of abnormal indicators in the current layer of this inspection is [0.6, 0.4]. The layer normal indicator transition probability is [0.3, 0.7], the current layer fault occurrence probability during the next inspection = 30x0.6+30x0.7=39, and the normal occurrence probability = 30x0.4+70x0.7=61.

Through model calculation, it is concluded that the failure probability of each horizontal layer in the next period of inspection is 39%, and the normal occurrence probability is 61%. Repeat the model to obtain the probability of the next inspection of other horizontal layers.

S2: Preprocess the inspection data through quadratic interpolation to construct a deep inspection model; input the preprocessed inspection data into the deep inspection model to obtain the actual value; calculate the fitting value through the ridge regression method .

Specifically, an in-depth inspection model based on AI algorithms was conducted on the six horizontal layers except (the basic environment layer is mainly composed of physical equipment) and combined with the over-fitting algorithm to detect whether the core technology of each module has strong generalization capabilities. Generalizationability refers to the ability of a machine learning algorithm to adapt to fresh samples. The purpose of learning is to learn the rules hidden behind the data. For data outside the learning set with the same rules, the trained network can also give appropriate output. This ability is called generalization ability.

Transformer technology is used to build an in-depth inspection model. The secondary interpolation technology provided by Transformer technology can make the inspection data more accurate before calculation.

In order to adapt to model processing, difference processing is performed on data with uneven sampling at different nodes. The quadratic difference method is used to interpolate every 3 adjacent points to obtain the quadratic interpolation. That is, data optimized by artificial intelligence algorithms. Advantages of this method:

1. The intervals are even and more consistent with the transformer timing processing.

2. More realistically restore the missing data of the simulated real scenes at all horizontal levels during the inspection process (current module input data, vertical correlation module input data, current module historical monitoring normal collection data).

Preferably, the construction of an in-depth inspection model includes:

PositionEncoding=cos2(pos/N)

Headi＝Attention(Qi,Ki,Vi)

MultiHead(Q,K,V)＝Concat(head1,...,headh)*WO

Specifically, the accurate data after secondary interpolation is input into the deep inspection model for simulation operation, and the result is expressed as the predicted failure probability value after the deep inspection of the current horizontal layer module in the microservice technology architecture system.

The model formula of the ridge regression method is: ||Xθ-y|| ² +||Γθ|| ²

The formula to prevent over-fitting is: θ(a)=(X ^T X+aI) ^-1 X ^T y

Among them, Find the value of θ if determined.

Specifically, the ridge regression method is used to detect whether the functions of modules at each level are abnormal and whether they have strong generalization capabilities. The reasons for overfitting in machine learning may be as follows: (1) Noisy data (waste data); (2) Insufficient training data and limited training data; (3) Over-training the model makes the model very complex. Therefore, the ridge regression method is used to prevent model overfitting.

The traditional least squares method lacks stability and reliability. In order to solve the above problems, this embodiment transforms the ill-posed problem into a well-posed problem: adding a regularization term to the above loss function.

Specifically, the fitted value is compared with the actual business history inspection normal data of the current level -> sub-module to the total inspection ratio (hereinafter referred to as the actual value), that is, the fitted value - actual value = difference, if the difference If the value is within 10%, it means that the calculation result of the current model simulated inspection data and the actual business inspection data through the in-depth inspection model is smaller. If more than 10% is determined to be an overfitting state, relevant technical personnel need to be notified to conduct functional inspections on the level and module of the inspected object, including background inspection of data entry, analysis of historical data, and whether related other modules or levels of business are Data anomalies, etc.

Preferably, this embodiment also includes the following steps: generating an intelligent inspection report.

Specifically, the inspection results of sub-modules at each layer are summarized and a report is generated, as shown in Table 1. This completes the entire process of intelligent in-depth inspection. Through intelligent reports, you can have a comprehensive understanding of the microservice technical architecture, from device monitoring indicators to comprehensive health monitoring of whether functions at each layer are running normally. It makes up for the existing inspection technology's overemphasis on physical equipment and application detection and lack of overall evaluation of microservice technology from the perspective of comparing the fitted values of each level through simulated data testing and the actual values generated by each horizontal layer through the operation of the in-depth inspection model. Possible data defects and hidden functional failures in the architecture may cause losses to the application platform.

Table 1 Inspection Report

This embodiment builds a health inspection model and a deep inspection model by horizontally layering the microservice architecture system, which can automatically implement in-depth inspections of each layer in the microservice architecture system. At the same time, the operation of the deep inspection model generates Comparing the actual value with the fitted value calculated by the ridge regression method can comprehensively evaluate the possible data defects and hidden functional failures of the microservice architecture, allowing the machine to make decisions instead of humans, thereby achieving complete automation and improving improves the efficiency of fault detection.

Example 2

Figure 2 is a schematic diagram of an in-depth inspection and optimization system based on microservice architecture. As shown in Figure 2, the present invention also provides an in-depth inspection optimization system based on microservice architecture. The system includes:

The first prediction module 201 is used to build a health inspection model and horizontally layer the microservice architecture system; obtain the physical device information corresponding to each module of the horizontal layer and the fault history data of the physical device, and combine the fault history data with Input the health inspection model to obtain the probability of failure of the physical equipment corresponding to each module in the horizontal layer in the next period of inspection;

The second prediction module 202 is used to preprocess the inspection data through secondary interpolation to construct a deep inspection model; input the preprocessed inspection data into the deep inspection model to obtain the actual value; use ridge regression to The method calculates the fitting value;

The comparison and evaluation module 203 is used to compare the fitting value with the actual value generated by the in-depth inspection model operation, and evaluate possible data defects and hidden functional failures in the microservice architecture system.

Preferably, the first prediction module 201 builds a health inspection model, and horizontally layering the microservice architecture system includes:

Preferably, the second prediction module 202 preprocesses the inspection data through quadratic interpolation including:

Preferably, the second prediction module 202 constructs an in-depth inspection model including:

PositionEncoding=cos2(pos/N)

Headi＝Attention(Qi,Ki,Vi)

MultiHead(Q,K,V)＝Concat(head1,...,headh)*WO

Preferably, the fitting value calculated by the second prediction module 202 through the ridge regression method includes:

The model formula of the ridge regression method is: ||Xθ-y|| ² +||Γθ|| ²

The formula to prevent over-fitting is: θ(a)=(X ^T X+aI) ^-1 X ^T y

Among them, Find the value of θ if determined.

The specific implementation process of the functions implemented by each module in Embodiment 2 is the same as the implementation process in Embodiment 1, and will not be described again here.

The above are only preferred embodiments of the present invention, and do not limit the patent scope of the present invention. Under the concept of the present invention, equivalent structural transformations can be made by using the contents of the description and drawings of the present invention, or directly/indirectly used in Other related technical fields are included in the patent protection scope of the present invention.

Claims

An in-depth inspection optimization method based on microservice architecture, characterized in that the method includes the following steps:

S1: Construct a health inspection model and horizontally layer the microservice architecture system; obtain the physical device information corresponding to each module in the horizontal layer and the fault history data of the physical device, and input the fault history data into the health inspection model to obtain the probability of failure of the physical equipment corresponding to each module in the horizontal layer in the next period of inspection;

S2: Preprocess the inspection data through quadratic interpolation to construct a deep inspection model; input the preprocessed inspection data into the deep inspection model to obtain the actual value; calculate the fitting value through the ridge regression method ;

S3: Compare the fitting value with the actual value generated by the in-depth inspection model operation, and evaluate possible data defects and hidden functional failures in the microservice architecture system.
The method according to claim 1, characterized in that said constructing a health inspection model and horizontally layering the microservice architecture system includes:

The microservice architecture system is horizontally divided into seven layers: basic environment layer, supervision management layer, cloud platform, information resource layer, business layer, interface layer and UI display layer;

The health inspection model includes a Markov transfer matrix algorithm model: X(k+1)=X(k)×P

Among them: X(k) represents the state vector of the trend analysis and prediction object at time t=k, P represents the one-step transition probability matrix, and .
The method according to claim 1, wherein preprocessing inspection data through quadratic interpolation includes:

To perform difference processing on data with uneven sampling at different nodes, the quadratic difference method is used to interpolate every 3 adjacent points to obtain quadratic interpolation. The formula is as follows:

Among them, x is the current value of the classification object; y is the three adjacent points of the classification object; i is the sequence number of the classification object.
The method according to claim 3, characterized in that said constructing an in-depth inspection model includes:

For the time dimension, PositionEncoding is used for timing encoding, and Attention is used to explore the feature correlation of the timing dimension. The formula is as follows:

PositionEncoding=cos2(pos/N)

For the spatial dimension, extract different multi-space dimension features and fuse the multi-space dimension features through MultiHead. The formula is as follows:

Headi＝Attention(Qi,Ki,Vi)

MultiHead(Q,K,V)＝Concat(head1,...,headh)*WO

Among them, d k represents the dimension of K; N is the adjustable length; Q is the query feature map; K is the feature map to be matched; V is the monitoring data map; headi is the result obtained by time attention, Qi is the i-th group of queries Feature map; Ki is the i-th group of feature maps to be matched; Vi is the i-th group of monitoring data mapping; WO is the feature fusion matrix; Concat is the feature cascade fusion; pos is the data sequence number; PositionEncoding is the position sequence encoding; MultiHead is the multi-head Feature fusion.
The method according to claim 4, wherein the fitting value calculated by the ridge regression method includes:

The model formula of the ridge regression method is: ||Xθ-y|| 2 +||Γθ|| 2

The formula to prevent over-fitting is: θ(a)=(X T X+aI) -1 X T y

Among them, Find the value of θ if determined.
An in-depth inspection and optimization system based on microservice architecture, characterized in that the system includes:

The first prediction module is used to build a health inspection model and horizontally layer the microservice architecture system; obtain the physical device information corresponding to each module in the horizontal layer and the fault history data of the physical device, and input the fault history data The health inspection model obtains the probability of failure of the physical equipment corresponding to each module in the horizontal layer in the next period of inspection;

The second prediction module is used to preprocess the inspection data through secondary interpolation to construct a deep inspection model; input the preprocessed inspection data into the deep inspection model to obtain the actual value; use the ridge regression method Calculate the fitting value;

The comparison and evaluation module is used to compare the fitting value with the actual value generated by the in-depth inspection model operation, and evaluate possible data defects and hidden functional failures in the microservice architecture system.
The system according to claim 6, characterized in that the first prediction module constructs a health inspection model, and horizontally layering the microservice architecture system includes:

The microservice architecture system is horizontally divided into seven layers: basic environment layer, supervision management layer, cloud platform, information resource layer, business layer, interface layer and UI display layer;

The health inspection model includes a Markov transfer matrix algorithm model: X(k+1)=X(k)×P

Among them: X(k) represents the state vector of the trend analysis and prediction object at time t=k, P represents the one-step transition probability matrix, and .
The system according to claim 6, characterized in that, the second prediction module preprocessing the inspection data through secondary interpolation includes:

To perform difference processing on data with uneven sampling at different nodes, the quadratic difference method is used to interpolate every 3 adjacent points to obtain quadratic interpolation. The formula is as follows:

Among them, x is the current value of the classification object; y is the three adjacent points of the classification object; i is the sequence number of the classification object.
The system according to claim 8, wherein the second prediction module constructing an in-depth inspection model includes:

For the time dimension, PositionEncoding is used for timing encoding, and Attention is used to explore the feature correlation of the timing dimension. The formula is as follows:

PositionEncoding=cos2(pos/N)

For the spatial dimension, extract different multi-space dimension features and fuse the multi-space dimension features through MultiHead. The formula is as follows:

Headi＝Attention(Qi,Ki,Vi)

MultiHead(Q,K,V)＝Concat(head1,...,headh)*WO

Among them, d k represents the dimension of K; N is the adjustable length; Q is the query feature map; K is the feature map to be matched; V is the monitoring data map; headi is the result obtained by time attention, Qi is the i-th group of queries Feature map; Ki is the i-th group of feature maps to be matched; Vi is the i-th group of monitoring data mapping; WO is the feature fusion matrix; Concat is the feature cascade fusion; pos is the data sequence number; PositionEncoding is the position sequence encoding; MultiHead is the multi-head Feature fusion.
The method according to claim 9, characterized in that the second prediction module calculates the fitting value through ridge regression method including:

The model formula of the ridge regression method is: ||Xθ-y|| 2 +||Γθ|| 2

The formula to prevent over-fitting is: θ(a)=(X T X+aI) -1 X T y

Among them, Find the value of θ if determined.