CN115310558B

CN115310558B - Big data analysis method and AI analysis system for cloud service abnormity optimization

Info

Publication number: CN115310558B
Application number: CN202211118619.3A
Authority: CN
Inventors: 张宾; 陈翰卿
Original assignee: Beijing Euronet Alliance Technology Co ltd
Current assignee: Beijing Euronet Alliance Technology Co ltd
Priority date: 2022-09-15
Filing date: 2022-09-15
Publication date: 2023-04-18
Anticipated expiration: 2042-09-15
Also published as: CN115310558A; CN116150595A

Abstract

The embodiment of the application provides a big data analysis method and an AI analysis system aiming at cloud service anomaly optimization, wherein a first graph relation characteristic and a second graph relation characteristic of a target page operation node are determined from an anomaly relation attribute graph, a third graph relation characteristic of a target cloud service abnormal event is determined from an event linkage relation graph, and an anomaly optimization decision is performed according to the first graph relation characteristic, the second graph relation characteristic, the third graph relation characteristic and graph relation characteristic data of prior anomaly big data comprehensive dimensions of the target cloud service abnormal event, so that the anomaly optimization decision is performed through multi-dimensional graph relation characteristic data, the accuracy of anomaly optimization decision information can be improved, and the stability of subsequent cloud service page operation is further improved.

Description

Big data analysis method and AI analysis system for cloud service abnormity optimization

Technical Field

The invention relates to the technical field of big data, in particular to a big data analysis method and an AI analysis system aiming at cloud service abnormity optimization.

Background

Mobile internet is full of many new technologies, innovations and marketing opportunities. The cloud service can meet new requirements of users for sharing, accessing and exploring, stable and high-speed connection is brought in the mobile 5G era, good user experience is provided for the smart phone, the cloud service provides easier cross-platform capability, and the cloud service is proved to be a new emerging business creation field in the mobile internet. Therefore, for a cloud service abnormal event, abnormal information of the cloud service abnormal event in a corresponding target page operation node needs to be mined in time, and then abnormal optimization is performed in time, however, in the related technology, an abnormal optimization decision is usually performed based on single dimensional feature data, so that accuracy of the abnormal optimization decision information is poor, and stability of subsequent cloud service page operation is affected.

Disclosure of Invention

In order to overcome at least the above disadvantages in the prior art, the present invention provides a big data analysis method and an AI analysis system for cloud service anomaly optimization.

In a first aspect, the present application provides a big data analysis method for cloud service anomaly optimization, which is applied to an AI analysis system, where the AI analysis system is in communication connection with a plurality of cloud service software service systems, and the method includes:

performing feature extraction on an abnormal relation attribute graph containing a target cloud service abnormal event and a target page operation node, and determining a first graph relation feature of the target cloud service abnormal event and a second graph relation feature of the target page operation node;

performing feature extraction on an event linkage relation graph containing the target cloud service abnormal event, and determining a third graph relation feature of the target cloud service abnormal event;

performing an anomaly optimization decision according to the first graph relation characteristic, the second graph relation characteristic, the third graph relation characteristic and prior anomaly big data of the target cloud service anomaly event, and determining anomaly optimization decision information of the target cloud service anomaly event corresponding to the target page operating node;

and carrying out abnormal optimization on the target page operation node according to the abnormal optimization decision information of the target cloud service abnormal event corresponding to the target page operation node.

In a possible implementation manner of the first aspect, feature extraction is performed on an abnormal relationship attribute graph containing a target cloud service abnormal event and a target page running node, and a first graph relationship feature of the target cloud service abnormal event and a second graph relationship feature of the target page running node are determined;

In a possible implementation manner of the first aspect, the step of performing feature extraction on an abnormal relationship attribute graph including a target cloud service abnormal event and a target page operation node, and determining a first graph relationship feature of the target cloud service abnormal event and a second graph relationship feature of the target page operation node specifically includes:

determining a connection page operation node of the target cloud service abnormal event in the abnormal relation attribute graph and a connection cloud service abnormal event of the target page operation node in the abnormal relation attribute graph;

performing multilayer hidden layer feature extraction on the connection page operation node of the target cloud service abnormal event, and determining a plurality of cloud service abnormal event features of the target cloud service abnormal event;

fusing the plurality of cloud service abnormal event characteristics to determine a first graph relation characteristic of the target cloud service abnormal event;

performing multi-layer hidden layer feature extraction on the cloud service connection abnormal event of the target page operation node, and determining a plurality of page operation node features of the target page operation node;

and fusing the characteristics of the plurality of page operation nodes, and determining the second graph relation characteristic of the target page operation node.

In a possible implementation manner of the first aspect, the hidden layer feature extraction is implemented based on node relationship inference networks, and mapping interconnection relationships are configured among the node relationship inference networks;

the step of extracting multilayer hidden layer features of the running nodes of the connection page of the target cloud service abnormal event and determining the features of the plurality of cloud service abnormal events of the target cloud service abnormal event specifically comprises the following steps:

based on the first node relation reasoning network in the entity relation reasoning network in the mapping interconnection form, carrying out regularization conversion on the initial page operation node characteristics of the connection page operation nodes of the target cloud service abnormal events, and determining the cloud service abnormal event characteristics of the target cloud service abnormal events in the first node relation reasoning network;

loading the cloud service abnormal event characteristics of the target cloud service abnormal event in the first node relation reasoning network to a next mapping interconnected entity relation reasoning network, and continuously performing regularized conversion based on the next mapping interconnected entity relation reasoning network to determine the cloud service abnormal event characteristics of the target cloud service abnormal event in the next mapping interconnected entity relation reasoning network;

the step of extracting multilayer hidden layer features of the cloud service connection abnormal event of the target page operation node and determining a plurality of page operation node features of the target page operation node specifically comprises the following steps:

based on the first node relation reasoning network in the entity relation reasoning network in the mapping interconnection form, carrying out regularized conversion on the initial cloud service abnormal event characteristics of the target page operation node connected with the cloud service abnormal event, and determining the page operation node characteristics of the target page operation node in the first node relation reasoning network;

loading the page operation node characteristics of the target page operation node in the first node relation inference network to a next mapping interconnected entity relation inference network, and continuously performing regularized conversion based on the next mapping interconnected entity relation inference network to determine the page operation node characteristics of the target page operation node in the next mapping interconnected entity relation inference network;

the step of continuing to perform regularization conversion on the basis of the next mapping interconnected entity relationship inference network and determining the cloud service abnormal event characteristics of the target cloud service abnormal event in the next mapping interconnected entity relationship inference network specifically comprises the following steps:

the xth node relation inference network of the entity relation inference network based on the mapping interconnection form implements the following steps:

determining page operation node characteristics of a ith connection page operation node in a plurality of connection page operation nodes of the target cloud service abnormal event in an x-1 node relation inference network;

determining the connection number of the operation nodes of the ith connection page and the connection number of the abnormal events of the target cloud service;

performing regularized conversion on the page operation node characteristics of the ith connection page operation node in the xth node relation inference network according to the connection quantity of the yth connection page operation node and the connection quantity of the target cloud service abnormal event, and determining the regularized conversion characteristics of the yth connection page operation node;

and aggregating the regularized conversion characteristics respectively corresponding to the plurality of connection page operation nodes, and determining the cloud service abnormal event characteristics of the target cloud service abnormal event in the x-th node relation inference network.

In a possible implementation manner of the first aspect, the step of performing feature extraction on an abnormal relationship attribute graph including a target cloud service abnormal event and a target page running node, and determining a first graph relationship feature of the target cloud service abnormal event and a second graph relationship feature of the target page running node specifically includes:

determining an inter-vertex relationship graph of the abnormal relationship attribute graph;

extracting multilayer hidden layer characteristics of the abnormal relation attribute graph according to the relationship graph between the vertexes, and determining a plurality of hidden layer characteristics of the abnormal relation attribute graph;

fusing the hidden layer characteristics to determine a target hidden layer characteristic of the abnormal relation attribute graph;

determining a first graph relation characteristic of the target cloud service abnormal event from the target hidden layer characteristic according to the characteristic point of the target cloud service abnormal event in the target hidden layer characteristic;

and determining a second graph relation characteristic of the target page operation node from the target hidden layer characteristics according to the characteristic point of the target page operation node in the target hidden layer characteristics.

the step of extracting the multilayer hidden layer features of the abnormal relationship attribute graph according to the relationship graph between the vertexes and determining the plurality of hidden layer features of the abnormal relationship attribute graph specifically includes:

based on the first node relationship inference network in the entity relationship inference network in the mapping interconnection form, carrying out regularized feature extraction on the initial hidden layer features of the abnormal relationship attribute graph and the relationship graph between the vertexes, and determining the hidden layer features of the abnormal relationship attribute graph in the first node relationship inference network;

loading the hidden layer characteristics of the abnormal relationship attribute graph in the first node relationship inference network to the next mapping interconnected entity relationship inference network, and implementing the following steps based on the z-th node relationship inference network of the mapping interconnected entity relationship inference network:

determining a connection number array of the abnormal relation attribute graph;

carrying out regularization conversion on the relationship graph between the vertexes according to the connection quantity array of the abnormal relationship attribute graph, and determining the relationship graph between the vertexes after regularization conversion;

and carrying out graph convolution feature extraction on the relationship graph between the vertexes after the regularization conversion and the hidden layer feature of the abnormal relationship attribute graph in the z-1 th node relationship inference network, and determining the hidden layer feature of the abnormal relationship attribute graph in the z th node relationship inference network.

In a possible implementation manner of the first aspect, the step of performing feature extraction on the event linkage relationship graph including the target cloud service abnormal event and determining a third graph relationship feature of the target cloud service abnormal event specifically includes:

determining a plurality of linkage connection cloud service abnormal events of the target cloud service abnormal events in the event linkage relation graph;

respectively extracting graph relation characteristics of the target cloud service abnormal event and each linkage connection cloud service abnormal event, and determining the event graph relation characteristics of the target cloud service abnormal event and the event graph relation characteristics of the linkage connection cloud service abnormal event;

carrying out nonlinear mapping processing on the event graph relation characteristics of the target cloud service abnormal events and the event graph relation characteristics of each linkage connection cloud service abnormal event, and determining the attention weight between the target cloud service abnormal events and each linkage connection cloud service abnormal event;

performing regularization conversion on the attention weight between the target cloud service abnormal event and each linkage connection cloud service abnormal event, and determining an attention influence value between the target cloud service abnormal event and each linkage connection cloud service abnormal event;

performing weighted fusion on the event graph relation characteristics of each linkage connection cloud service abnormal event according to the attention influence value, and determining the weighted fusion characteristics of the target cloud service abnormal event;

and mapping the weighted fusion characteristics of the target cloud service abnormal event, and determining a third graph relation characteristic of the target cloud service abnormal event.

In a possible implementation manner of the first aspect, the step of performing an anomaly optimization decision according to the first graph relation feature, the second graph relation feature, the third graph relation feature, and a priori anomaly big data of the target cloud service anomaly event, and determining anomaly optimization decision information of the target cloud service anomaly event corresponding to the target page operation node specifically includes:

acquiring prior abnormal big data of the target cloud service abnormal event;

performing characteristic linear mapping on the prior abnormal big data of the target cloud service abnormal event to determine the prior abnormal characteristic of the target cloud service abnormal event;

fusing the first graph relation feature, the second graph relation feature, the third graph relation feature and the prior anomaly feature to determine a fusion feature;

and performing an abnormal optimization decision on the fusion characteristics, and determining abnormal optimization decision information of the target cloud service abnormal event corresponding to the target page operation node.

In a possible implementation manner of the first aspect, the step of performing an anomaly optimization decision on the fusion feature and determining anomaly optimization decision information of the target cloud service anomaly event corresponding to the target page operation node includes:

acquiring a plurality of reference abnormal feature sequences as training sample data sequences, wherein each reference abnormal feature sequence corresponds to a priori abnormal knowledge point, and each reference abnormal feature sequence comprises a first reference graph relation feature, a second reference graph relation feature, a third reference graph relation feature and a reference priori abnormal feature, wherein the first reference graph relation feature is a graph relation feature of a reference cloud service abnormal event in a reference abnormal relation attribute graph, the second reference graph relation feature is a graph relation feature of a reference page operation node in the reference abnormal relation attribute graph, the third reference graph relation feature is an event linkage relation graph of the reference cloud service abnormal event, and the reference priori abnormal feature is a priori abnormal feature of reference priori abnormal big data of the reference cloud service abnormal event;

for each reference abnormal feature sequence in a plurality of reference abnormal feature sequences, performing space-time node feature random conversion according to the reference abnormal features in the reference abnormal feature sequence, and mixing at least two space-time node conversion reference features in space-time node conversion reference features obtained by performing space-time node feature random conversion on the reference abnormal features in the reference abnormal feature sequence to generate a mixed abnormal feature sequence corresponding to the reference abnormal feature sequence;

performing sequence forward selection on each space-time node conversion reference feature in a mixed abnormal feature sequence corresponding to the reference abnormal feature sequence, determining a reference redirection abnormal feature sequence, and loading the reference redirection abnormal feature sequence into the reference abnormal feature sequence to obtain a derived reference abnormal feature sequence of the reference abnormal feature sequence;

and training an abnormal knowledge point decision model for abnormal knowledge point decision based on a plurality of derived reference abnormal feature sequences of the plurality of reference abnormal feature sequences, wherein the trained abnormal knowledge point decision model carries out abnormal knowledge point decision on the target fusion features to obtain abnormal optimization decision information corresponding to the target cloud service abnormal event.

In a possible implementation manner of the first aspect, the generating a mixed abnormal feature sequence corresponding to the reference abnormal feature sequence by performing spatio-temporal node feature random transformation according to the reference abnormal feature in the reference abnormal feature sequence and mixing at least two spatio-temporal node transformation reference features in spatio-temporal node transformation reference features obtained by performing spatio-temporal node feature random transformation on the reference abnormal feature in the reference abnormal feature sequence includes:

performing space-time node characteristic random conversion on each reference abnormal characteristic in the reference abnormal characteristic sequence to generate a first space-time node conversion characteristic sequence;

scrambling the space-time node conversion reference features in the first space-time node conversion feature sequence based on a preset feature arrangement rule to generate a second space-time node conversion feature sequence, and mixing each space-time node conversion reference feature in the first space-time node conversion feature sequence with a corresponding space-time node conversion reference feature in the second space-time node conversion feature sequence to generate a third space-time node conversion feature sequence serving as a mixed abnormal feature sequence corresponding to the reference abnormal feature sequence; or

Scrambling processing is carried out on the reference abnormal features in the reference abnormal feature sequence based on a preset feature arrangement rule, and a scrambled abnormal feature sequence corresponding to the reference abnormal feature sequence is generated;

performing space-time node feature random conversion on each reference abnormal feature in the reference abnormal feature sequence and each reference abnormal feature in the scrambled abnormal feature sequence to respectively obtain a first space-time node conversion feature sequence and a second space-time node conversion feature sequence;

and mixing each space-time node conversion reference feature in the first space-time node conversion feature sequence with a corresponding space-time node conversion reference feature in the second space-time node conversion feature sequence to generate a third space-time node conversion feature sequence as a mixed abnormal feature sequence corresponding to the reference abnormal feature sequence.

For example, in one possible implementation manner of the first aspect, the method further includes:

determining whether the number of the reference abnormal features in the reference abnormal feature sequence is greater than a preset number and/or whether the number of the derived dimensions is greater than a preset dimension number;

when the reference abnormal feature sequence is determined to be not more than the preset number and/or the derived dimension number is not more than the preset dimension number, generating a mixed abnormal feature sequence corresponding to the reference abnormal feature sequence according to the random conversion of the spatio-temporal node features of the reference abnormal features in the reference abnormal feature sequence and the mixing of at least two spatio-temporal node conversion reference features in the spatio-temporal node conversion reference features obtained by the random conversion of the spatio-temporal node features of the reference abnormal features in the reference abnormal feature sequence, and generating the derived reference abnormal feature sequence of the reference abnormal feature sequence based on the reference abnormal feature sequence and the mixed abnormal feature sequence corresponding to the reference abnormal feature sequence.

For example, in a possible implementation manner of the first aspect, each reference abnormal feature in each reference abnormal feature sequence includes channel features of a plurality of channels, wherein performing spatio-temporal node feature random transformation on each reference abnormal feature in the reference abnormal feature sequence generates a first spatio-temporal node transformation feature sequence, including:

respectively carrying out space-time node feature random conversion on channel features of a plurality of channels included in the reference abnormal features aiming at each reference abnormal feature, and determining feature distribution including a plurality of space-time node conversion mapping features corresponding to the plurality of channels one by one, wherein the feature distribution is used as the space-time node conversion reference features corresponding to the reference abnormal features;

and taking the space-time node conversion reference characteristics corresponding to each reference abnormal characteristic in the reference abnormal characteristic sequence as a first space-time node conversion characteristic sequence of the reference abnormal characteristic sequence.

For example, in a possible implementation manner of the first aspect, the scrambling processing of spatio-temporal node transformation reference features in the first spatio-temporal node transformation feature sequence based on a preset feature arrangement rule to generate a second spatio-temporal node transformation feature sequence includes:

arranging the space-time node conversion reference characteristics of the first space-time node conversion characteristic sequence;

and extracting each space-time node conversion reference feature in the first space-time node conversion feature sequence based on the preset feature arrangement rule, and scrambling each space-time node conversion reference feature according to the preset feature arrangement rule to obtain a second space-time node conversion feature sequence.

For example, in one possible implementation of the first aspect, mixing each spatio-temporal node transformation reference feature of the first spatio-temporal node transformation signature sequence with a corresponding one of the spatio-temporal node transformation reference features of the second spatio-temporal node transformation signature sequence to generate a third spatio-temporal node transformation signature sequence comprises:

respectively mixing the feature distribution of the ith space-time node conversion reference feature in the first space-time node conversion feature sequence with the space-time node conversion mapping feature of the feature distribution of the ith space-time node conversion reference feature in the second space-time node conversion feature sequence according to channels, and determining new feature distribution corresponding to a serial number i, wherein i is more than or equal to 1 and less than or equal to N;

and taking each new characteristic distribution corresponding to each serial number as the third time-space node conversion characteristic sequence.

In a second aspect, an embodiment of the present application further provides a big data analysis system optimized for cloud service anomalies, where the big data analysis system optimized for cloud service anomalies includes an AI analysis system and a plurality of cloud service software service systems in communication connection with the AI analysis system;

the AI analysis system to:

In any aspect, the first graph relation characteristic and the second graph relation characteristic of the target page operation node are determined from the abnormal relation attribute graph, the third graph relation characteristic of the target cloud service abnormal event is determined from the event linkage relation graph, and abnormal optimization decision is performed according to the first graph relation characteristic, the second graph relation characteristic, the third graph relation characteristic and graph relation characteristic data of prior abnormal big data comprehensive dimensions of the target cloud service abnormal event, so that the abnormal optimization decision is performed through the multi-dimensional graph relation characteristic data, the accuracy of abnormal optimization decision information can be improved, and the operation stability of a subsequent cloud service page is further improved.

Drawings

Fig. 1 is a schematic flow chart of a big data analysis method for cloud service anomaly optimization according to an embodiment of the present invention.

Detailed Description

The following describes an architecture of a cloud service anomaly optimized big data analysis system 10 according to an embodiment of the present invention, where the cloud service anomaly optimized big data analysis system 10 may include an AI analysis system 100 and a cloud service software service system 200 communicatively connected to the AI analysis system 100. The AI analysis system 100 and the cloud service software service system 200 in the cloud service anomaly optimized big data analysis system 10 may be combined with a big data analysis method for cloud service anomaly optimization described in the following method embodiments, and the detailed description of the method embodiments may be referred to in the following steps of the AI analysis system 100 and the cloud service software service system 200.

The big data analysis method for cloud service anomaly optimization provided in this embodiment may be executed by the AI analysis system 100, and the details of the big data analysis method for cloud service anomaly optimization are described below with reference to fig. 1.

In the Process101, feature extraction is performed on an abnormal relationship attribute graph containing a target cloud service abnormal event and a target page operation node, and a first graph relationship feature of the target cloud service abnormal event and a second graph relationship feature of the target page operation node are determined.

The abnormal relationship attribute graph can be expressed in a form of a binary heterogeneous graph combined relationship attribute, and for example, the abnormal relationship attribute graph can include a relationship network graph of two types of entities, that is, the relationship network graph includes a cloud service abnormal event node and a page operation node entity, the cloud service abnormal event entity is an entity (that is, a cloud service abnormal event in an abnormal relationship attribute graph) for representing a cloud service abnormal event (that is, an operation event in an abnormal state during cloud service operation), the page operation node entity is an entity (that is, a page operation node in the abnormal relationship attribute graph) for representing a page operation node (that is, a software instance for supporting cloud service page operation service), the abnormal relationship attribute graph includes a plurality of cloud service abnormal event entities and a plurality of page operation node entities, wherein a target cloud service abnormal event corresponds to any cloud service abnormal event entity in the abnormal relationship attribute graph, and the target page operation node is any page operation node entity in the abnormal relationship attribute graph. Any cloud service abnormal event in the abnormal relationship attribute graph can be called a cloud service abnormal event entity, and any page operation node in the abnormal relationship attribute graph can be called a page operation node entity.

For example, an abnormal relationship attribute graph is constructed according to the session activity of the cloud service abnormal event and the page running node, and the abnormal relationship attribute graph is input into a graph neural network model to extract a first graph relationship feature of a target cloud service abnormal event (for example, a session activity graph feature of the cloud service abnormal event) and a second graph relationship feature of the target page running node (a session activity graph feature of the page running node), which cover the session activity of the cloud service abnormal event and the page running node.

For example, when the cloud service exception event 1 has session activity with the page running node 3 and the page running node 4, a connection link between the cloud service exception event 1 and the page running node 3 and the page running node 4 in the exception relationship attribute graph is constructed, for example, when the page running node is an e-commerce interaction page node, the cloud service exception event 1 permeates the e-commerce interaction page node 3 and the e-commerce interaction page node 4.

In some exemplary design considerations, process101 may be implemented by Process1011-Process1015 as follows:

in the Process1011, determining a connection page running node of a target cloud service abnormal event in the abnormal relation attribute graph and a connection cloud service abnormal event of the target page running node in the abnormal relation attribute graph; in the Process1012, performing multi-layer hidden layer feature extraction on a connection page running node of a target cloud service abnormal event, and determining a plurality of cloud service abnormal event features of the target cloud service abnormal event; in the Process1013, the multiple cloud service abnormal event characteristics are fused to determine a first graph relation characteristic of a target cloud service abnormal event; in the Process1014, performing multi-layer hidden layer feature extraction on a connection cloud service abnormal event of a target page operation node, and determining a plurality of page operation node features of the target page operation node; in the Process1015, the characteristics of the multiple page operation nodes are fused, and the second graph relation characteristic of the target page operation node is determined.

The connection page operation node represents a page operation node adjacent to the target cloud service abnormal event in the abnormal relation attribute graph, and the connection cloud service abnormal event represents a cloud service abnormal event adjacent to the target page operation node in the abnormal relation attribute graph.

For example, by smoothing features on the abnormal relation attribute graph, entity representation (namely, the features of entities (page operation nodes or cloud service abnormal events) in the abnormal relation attribute graph) is accurately learned, for the purpose of being larger than the above purpose, the features of an iterative aggregation associated entity (entities adjacent to a target entity are called associated entities, when the target entity is a target cloud service abnormal event, the associated entities are connected page operation nodes, when the target entity is the target page operation node, the associated entities are connected cloud service abnormal events) are represented as new entities, for example, the connected page operation node of the target cloud service abnormal event is subjected to multiple times (more than or equal to 2 times) of hidden layer feature extraction, a plurality of cloud service abnormal event features of the target cloud service abnormal event are determined, and the plurality of cloud service abnormal event features are fused, and a first graph relation feature of the target cloud service abnormal event is determined (the first graph relation feature represents the embedded graph feature of the target cloud service abnormal event in the abnormal relation attribute graph), so that the iterative aggregation feature of the connected page operation node is used as the representation of the target cloud service abnormal event entity; the method comprises the steps of extracting multilayer hidden layer features of a connection cloud service abnormal event of a target page operation node, determining a plurality of page operation node features of the target page operation node, fusing the plurality of page operation node features, and determining a second graph relation feature of the target page operation node (the second graph relation feature represents an embedded graph feature of the target page operation node in an abnormal relation attribute graph), so that the features of the connection cloud service abnormal event are iteratively aggregated to be used as an entity representation of the target cloud service abnormal event.

In some exemplary design ideas, the hidden layer feature extraction is realized based on node relation reasoning networks, and mapping interconnection relations are configured among the node relation reasoning networks; the method comprises the steps of extracting multilayer hidden layer features of a connection page operation node of a target cloud service abnormal event and determining a plurality of cloud service abnormal event features of the target cloud service abnormal event, and specifically comprises the following steps: performing regularization conversion on initial page operation node characteristics of a connection page operation node of a target cloud service abnormal event through a first node relation inference network in an entity relation inference network in a mapping interconnection form, and determining cloud service abnormal event characteristics of the target cloud service abnormal event in the first node relation inference network; loading the cloud service abnormal event characteristics of the target cloud service abnormal event in the first node relation reasoning network to a next mapping interconnected entity relation reasoning network, and continuously performing regularized conversion through the next mapping interconnected entity relation reasoning network to determine the cloud service abnormal event characteristics of the target cloud service abnormal event in the next mapping interconnected entity relation reasoning network; the method comprises the steps of extracting multilayer hidden layer characteristics of a connection cloud service abnormal event of a target page operation node and determining a plurality of page operation node characteristics of the target page operation node, and specifically comprises the following steps: performing regularization conversion on initial cloud service abnormal event characteristics of a target page operation node connected with a cloud service abnormal event through a first node relation reasoning network in an entity relation reasoning network in a mapping interconnection form, and determining page operation node characteristics of the target page operation node in the first node relation reasoning network; and loading the page operation node characteristics of the target page operation node in the first node relation inference network to the next mapping interconnected entity relation inference network, and continuously performing regularized conversion through the next mapping interconnected entity relation inference network to determine the page operation node characteristics of the target page operation node in the next mapping interconnected entity relation inference network.

After the calculation of the entity relationship reasoning network of one layer, the analysis and the learning of the cloud service abnormal event characteristics or the page operation node characteristics can be further carried out, and the cloud service abnormal event characteristics or the page operation node characteristics can be gradually and accurately learned through the calculation of the multi-layer node relationship reasoning network. Through the processing of the mapping interconnection form, the cloud service abnormal event characteristics or the page operation node characteristics with gradually improved accuracy can be obtained.

For example, the connection page operation nodes (corresponding connection page operation node entities) of the target cloud service abnormal events are subjected to regularization conversion through a plurality of mapping interconnected entity relationship inference networks, and therefore the characteristics of the connection page operation nodes are iteratively aggregated. For example, the initial page operation node characteristics of the connection page operation nodes of the target cloud service abnormal events are subjected to regularization conversion through a first node relation inference network in the entity relation inference network in the mapping interconnection form, the cloud service abnormal event characteristics of the target cloud service abnormal events in the first node relation inference network are determined, the cloud service abnormal event characteristics of the target cloud service abnormal events in the first node relation inference network are loaded to a second node relation inference network, the page operation node characteristics of the connection page operation nodes of the target cloud service abnormal events in the first node relation inference network are subjected to regularization conversion, the cloud service abnormal event characteristics of the target cloud service abnormal events in the second node relation inference network are determined, and the plurality of mapping interconnection entity relation inference networks respectively execute the steps, so that a plurality of cloud service abnormal event characteristics of the target cloud service abnormal events are obtained. The initial page operation node features represent all page operation node entities in the abnormal relation attribute graph.

For example, the connection cloud service abnormal events (corresponding connection cloud service abnormal event entities) of the target page operation nodes are subjected to regularization conversion through a plurality of mapping interconnected entity relationship inference networks, and therefore the characteristics of the connection cloud service abnormal events are iteratively aggregated. For example, the initial cloud service abnormal event feature of a connection cloud service abnormal event of a target page operation node is regularly converted through a first node relation reasoning network in a mapping interconnection type entity relation reasoning network, the page operation node feature of the target page operation node in the first node relation reasoning network is determined, the page operation node feature of the target page operation node in the first node relation reasoning network is loaded to a second node relation reasoning network, the cloud service abnormal event feature of the connection cloud service abnormal event of the target page operation node in the first node relation reasoning network is regularly converted, the page operation node feature of the target page operation node in the second node relation reasoning network is determined, and the plurality of mapping interconnection entity relation reasoning networks sequentially execute the operations, so that the plurality of page operation node features of the target page operation node are obtained. The initial cloud service abnormal event feature represents all cloud service abnormal event entities in the abnormal relation attribute graph.

In some exemplary design ideas, the step of continuing to perform regularization conversion through the next mapping interconnected entity relationship inference network to determine the cloud service abnormal event characteristics of the target cloud service abnormal event in the next mapping interconnected entity relationship inference network specifically includes: the method comprises the following steps of mapping an xth node relationship inference network of an interconnected entity relationship inference network to implement the following steps: the method comprises the following steps of aiming at the y-th connection page running node in a plurality of connection page running nodes of a target cloud service abnormal event: determining the page operation node characteristics of the operation node of the ith connection page in the (x-1) th node relation inference network; determining the connection number of the operation nodes of the y-th connection page and the connection number of the abnormal events of the target cloud service; performing regularized conversion on the page operation node characteristics of the operation node of the ith connection page in the x-th node relation inference network according to the connection quantity of the operation node of the ith connection page and the connection quantity of the target cloud service abnormal events, and determining the regularized conversion characteristics of the operation node of the ith connection page; aggregating the regularization conversion characteristics respectively corresponding to the plurality of connection page operation nodes, and determining the cloud service abnormal event characteristics of the target cloud service abnormal event in the xth node relation inference network; wherein x is an increasing natural number and has a value range of 1<x not more than J, J is the number of the node relation inference networks, J is a positive integer larger than 1, y is an increasing natural number and has a value range of 1 not less than y not more than M, M is the number of the operation nodes of the connection page, and M is a positive integer.

The regularization conversion process of the node relation inference network is shown as the following formula, wherein cloud service abnormal event characteristics (namely node representation) of a target cloud service abnormal event in an x-th node relation inference network (namely x-th graph convolution operation) are shown, page operation node characteristics of a y-th connection page operation node in an x-1-th node relation inference network are shown, the connection number of the target cloud service abnormal event (namely the number of connection page operation nodes of the target cloud service abnormal event) is shown, and the connection number of the y-th connection page operation node (namely the number of connection cloud service abnormal events of the y-th connection page operation node) is shown.

In some exemplary design ideas, the step of continuing to perform regularized transformation through the next mapping interconnected entity relationship inference network to determine the page operation node characteristics of the target page operation node in the next mapping interconnected entity relationship inference network specifically includes: the method comprises the following steps of mapping an xth node relationship inference network of an interconnected entity relationship inference network to implement the following steps: the method comprises the following steps of aiming at the y-th connection cloud service abnormal event in a plurality of connection cloud service abnormal events of a target page operation node: determining the cloud service abnormal event characteristics of the ith connection cloud service abnormal event in the x-1 node relation reasoning network; determining the connection quantity of the y-th connection cloud service abnormal event and the connection quantity of the target page operation node; according to the connection quantity of the y-th connection cloud service abnormal event and the connection quantity of the target page operation node, carrying out regularized conversion on the cloud service abnormal event characteristics of the y-th connection cloud service abnormal event in the x-th node relation inference network, and determining the regularized conversion characteristics of the y-th connection cloud service abnormal event; aggregating the regularization conversion characteristics respectively corresponding to the plurality of connected cloud service abnormal events, and determining the page operation node characteristics of the target page operation node in the x-th node relation inference network; wherein x is an increasing natural number and has a value range of 1<x not more than N, N is the number of the node relation inference networks, N is a positive integer greater than 1, y is an increasing natural number and has a value range of 1 not less than y not more than M, M is the number of the cloud service connection abnormal events, and M is a positive integer.

In some exemplary design ideas, the step of fusing the features of the multiple page operation nodes and determining the second graph relation feature of the target page operation node specifically includes: performing weighted fusion on the characteristics of the multiple page operation nodes, and determining second graph relation characteristics of the target page operation nodes; the method comprises the following steps of fusing the characteristics of a plurality of cloud service abnormal events and determining a first graph relation characteristic of a target cloud service abnormal event, wherein the steps specifically comprise: and performing weighted fusion on the plurality of cloud service abnormal event characteristics, and determining a first graph relation characteristic of a target page operation node.

In some exemplary design ideas, the steps of performing feature extraction on an abnormal relationship attribute graph including a target cloud service abnormal event and a target page operation node, and determining a first graph relationship feature of the target cloud service abnormal event and a second graph relationship feature of the target page operation node specifically include: determining a relationship graph between vertexes of the abnormal relationship attribute graph; extracting multilayer hidden layer characteristics of the abnormal relation attribute graph according to the relation graph between the vertexes, and determining a plurality of hidden layer characteristics of the abnormal relation attribute graph; fusing the plurality of hidden layer characteristics to determine a target hidden layer characteristic of the abnormal relation attribute graph; determining a first graph relation characteristic of the target cloud service abnormal event from the target hidden layer characteristic according to the characteristic point of the target cloud service abnormal event in the target hidden layer characteristic; and determining a second graph relation characteristic of the target page operation node from the target hidden layer characteristics according to the characteristic point of the target page operation node in the target hidden layer characteristics.

For example, in order to quickly calculate all node representations in the abnormal relationship attribute graph, calculation can be performed through an array form, and the inter-vertex relationship graph is an array form of the bipartite graph heterogeneous graph. Extracting multilayer hidden layer features of the abnormal relation attribute graph according to the relation graph between the vertexes, determining a plurality of hidden layer features of the abnormal relation attribute graph, fusing the hidden layer features, determining a target hidden layer feature (namely the last hidden layer feature) of the abnormal relation attribute graph, determining a first graph relation feature of a target cloud service abnormal event from the target hidden layer features according to a feature point of the target cloud service abnormal event on the target hidden layer feature, fusing the hidden layer features according to a feature point of a target page running node on the target hidden layer feature, and determining the target hidden layer feature of the abnormal relation attribute graph.

In some exemplary design ideas, the hidden layer feature extraction is realized based on node relation reasoning networks, and mapping interconnection relations are configured among the node relation reasoning networks; the method comprises the following steps of extracting multilayer hidden layer features of an abnormal relation attribute graph according to a relation graph between vertexes, and determining the plurality of hidden layer features of the abnormal relation attribute graph, wherein the steps specifically comprise: performing regularization feature extraction on the initial hidden layer features of the abnormal relationship attribute graph and the relationship graph between vertexes through a first node relationship inference network in the entity relationship inference network in a mapping interconnection form, and determining the hidden layer features of the abnormal relationship attribute graph in the first node relationship inference network; loading the hidden layer characteristics of the abnormal relationship attribute graph in the first node relationship inference network to the next mapping interconnected entity relationship inference network, continuously carrying out regularized characteristic extraction through the next mapping interconnected entity relationship inference network, and determining the hidden layer characteristics of the abnormal relationship attribute graph in the next mapping interconnected entity relationship inference network.

For example, the method comprises the steps of performing regularized feature extraction on an initial hidden layer feature and an inter-vertex relationship graph of an abnormal relationship attribute graph, determining a hidden layer feature of the abnormal relationship attribute graph in a first node relationship inference network, loading the hidden layer feature of the abnormal relationship attribute graph in the first node relationship inference network to a second node relationship inference network, performing regularized feature extraction on the hidden layer feature of the abnormal relationship attribute graph in the first node relationship inference network and the inter-vertex relationship graph, determining a hidden layer feature of the abnormal relationship attribute graph in the second node relationship inference network, and sequentially performing the operations by a plurality of entity relationship inference networks which are interconnected in a mapping manner, so as to obtain a plurality of hidden layer features of the abnormal relationship attribute graph.

In some exemplary design ideas, the step of continuously performing regularized feature extraction through the next mapping interconnected entity relationship inference network to determine hidden layer features of the abnormal relationship attribute graph in the next mapping interconnected entity relationship inference network specifically includes: the following steps are implemented by the z-th node relation inference network of the entity relation inference network in the form of mapping interconnection: determining a connection quantity array of the abnormal relation attribute graph; carrying out regularization conversion on the relationship graph between the vertexes according to the connection quantity array of the abnormal relationship attribute graph, and determining the relationship graph between the vertexes after regularization conversion; performing graph convolution feature extraction on the relationship graph between vertexes after regularization conversion and the hidden layer feature of the abnormal relationship attribute graph in the z-1 th node relationship inference network, and determining the hidden layer feature of the abnormal relationship attribute graph in the z th node relationship inference network; wherein z is an increasing natural number and the value range is 1<z which is not more than K, K is the number of the node relation inference networks, and K is a positive integer greater than 1.

In some exemplary design ideas, the step of fusing a plurality of hidden layer features and determining a target hidden layer feature of an abnormal relationship attribute graph specifically includes: and performing weighted fusion on the plurality of hidden layer characteristics, and determining the target hidden layer characteristics of the abnormal relation attribute graph.

In the Process102, feature extraction is performed on the event linkage relation graph containing the target cloud service abnormal event, and third graph relation features of the target cloud service abnormal event are determined.

The event linkage relationship graph (homogeneous graph) refers to a graph network comprising one type of entity, namely comprises cloud service abnormal event (such as cloud service abnormal event) entities, the cloud service abnormal event entities are entities used for representing the cloud service abnormal events, the two homogeneous graphs comprise a plurality of cloud service abnormal event entities, and the target cloud service abnormal events correspond to any cloud service abnormal event entity in the homogeneous graphs.

For example, a homogeneity graph is constructed according to a linkage triggering relation chain among the cloud service abnormal events, and the homogeneity graph is input into a graph attention network to extract a third graph relation characteristic of the target cloud service abnormal event aggregated with linkage triggering relation information.

As shown in fig. 7, when the cloud service abnormal event 1 has a linkage triggering relationship with the cloud service abnormal event 5 and the cloud service abnormal event 7, an edge between the cloud service abnormal event 1 and the cloud service abnormal event 5 and the cloud service abnormal event 7 in the homogenous graph is constructed.

Process102 may be implemented by Process1021-Process 1023:

in the Process1021, a plurality of linkage connection cloud service abnormal events of the target cloud service abnormal events in the event linkage relation graph are determined; in the Process1022, graph relation features of the target cloud service abnormal event and each linkage connection cloud service abnormal event are extracted respectively, and an event graph relation feature of the target cloud service abnormal event and an event graph relation feature of the linkage connection cloud service abnormal event are determined; in the Process1023, the event graph relation characteristics of the target cloud service abnormal event and the event graph relation characteristics of each linkage connection cloud service abnormal event are subjected to penalty item-based characteristic selection processing, and the third graph relation characteristics of the target cloud service abnormal event are determined.

In some exemplary design ideas, the step of performing feature selection processing based on penalty terms on the event graph relationship features of the target cloud service abnormal event and the event graph relationship features of each linkage connection cloud service abnormal event, and determining the third graph relationship features of the target cloud service abnormal event specifically includes: carrying out feature selection decision based on penalty items on the event graph relationship features of the target cloud service abnormal events and the event graph relationship features of each linkage connection cloud service abnormal event, and determining the attention influence value between the target cloud service abnormal events and each linkage connection cloud service abnormal event; and generating a third graph relation characteristic of the target cloud service abnormal event according to the event graph relation characteristic and the attention influence value of each linkage connection cloud service abnormal event.

In some exemplary design ideas, a feature selection decision based on a penalty term is performed on the event graph relationship feature of the target cloud service abnormal event and the event graph relationship feature of each linkage connection cloud service abnormal event, and a focus influence value between the target cloud service abnormal event and each linkage connection cloud service abnormal event is determined, specifically including: carrying out nonlinear mapping processing on the event graph relation characteristics of the target cloud service abnormal event and the event graph relation characteristics of each linkage connection cloud service abnormal event, and determining the attention weight between the target cloud service abnormal event and each linkage connection cloud service abnormal event; and carrying out regularization conversion on the attention weight between the target cloud service abnormal event and each linkage connection cloud service abnormal event, and determining the attention influence value between the target cloud service abnormal event and each linkage connection cloud service abnormal event.

In some exemplary design ideas, the step of generating a third graph relation feature of the target cloud service abnormal event according to the event graph relation feature and the attention influence value of each linkage connection cloud service abnormal event specifically includes: carrying out weighted fusion on the event graph relation characteristics of each linkage connection cloud service abnormal event according to the attention influence value, and determining the weighted fusion characteristics of the target cloud service abnormal event; and mapping the weighted fusion characteristics of the target cloud service abnormal event, and determining the third graph relation characteristics of the target cloud service abnormal event.

The mapping processing can be realized by activating a function, and the third graph relation characteristic of the target cloud service abnormal event, namely the cloud service abnormal event, is represented by embedding a graph of a linkage triggering relation chain.

In the Process103, an anomaly optimization decision is performed according to the first graph relation characteristic, the second graph relation characteristic, the third graph relation characteristic and the prior anomaly big data of the target cloud service anomaly event, and anomaly optimization decision information of the target page operation node corresponding to the target cloud service anomaly event is determined.

For example, an anomaly optimization decision is performed according to the first graph relation characteristic, the second graph relation characteristic, the third graph relation characteristic and the graph relation characteristic data of the prior anomaly big data comprehensive dimension of the target cloud service anomaly event, so that the anomaly optimization decision is performed through the multi-dimensional graph relation characteristic data, the accuracy of anomaly optimization decision information can be improved, and the running stability of a subsequent cloud service page is further improved.

In some exemplary design ideas, performing an anomaly optimization decision according to the first graph relation feature, the second graph relation feature, the third graph relation feature and prior anomaly big data of the target cloud service anomaly event, and determining anomaly optimization decision information of the target page operation node corresponding to the target cloud service anomaly event specifically includes: acquiring prior abnormal big data of a target cloud service abnormal event; performing characteristic linear mapping on the prior abnormal big data of the target cloud service abnormal event to determine the prior abnormal characteristic of the target cloud service abnormal event; fusing the first graph relation characteristic, the second graph relation characteristic, the third graph relation characteristic and the prior anomaly characteristic to determine a fusion characteristic; and performing an abnormal optimization decision on the fusion characteristics, and determining abnormal optimization decision information of the target page operation node corresponding to the target cloud service abnormal event.

The prior abnormal big data of the target cloud service abnormal event can be a recorded data set formed by historical abnormal state data of the target cloud service abnormal event. The method comprises the steps of carrying out feature linear mapping on prior abnormal big data of a target cloud service abnormal event through a multilayer perceptron constructed by a deep neural network, and determining prior abnormal features (feature representation of the prior abnormal big data) of the target cloud service abnormal event. After the first graph relation characteristic, the second graph relation characteristic, the third graph relation characteristic and the prior anomaly characteristic are fused, anomaly optimization decision is carried out on the fused characteristic through an anomaly knowledge point positioning model, and anomaly optimization decision information of a target page operation node corresponding to a target cloud service anomaly event is determined.

In the Process104, performing exception optimization on the target page operation node according to the exception optimization decision information of the target cloud service exception corresponding to the target page operation node.

The abnormal optimization decision information can represent one or more abnormal knowledge points of the target page running node corresponding to the target cloud service abnormal event, and the abnormal knowledge points can be running defect (BUG) points existing in a software running program corresponding to the target page running node, so that repair patch data corresponding to each abnormal knowledge point can be obtained from a cloud repair library, and the target page running node is subjected to abnormal optimization based on the repair patch data.

The following describes an artificial intelligence based anomaly optimization decision method provided in another embodiment of the present application, including the following steps.

Step S110, obtaining a plurality of reference abnormal feature sequences as training sample data sequences, where each reference abnormal feature sequence corresponds to a priori abnormal knowledge point, and each reference abnormal feature sequence includes a first reference graph relation feature, a second reference graph relation feature, a third reference graph relation feature, and a reference priori abnormal feature, where the first reference graph relation feature is a graph relation feature of a reference cloud service abnormal event in a reference abnormal relation attribute graph, the second reference graph relation feature is a graph relation feature of a reference page operation node in a reference abnormal relation attribute graph, the third reference graph relation feature is an event linkage relation graph of the reference cloud service abnormal event, and the reference priori abnormal feature is a priori abnormal feature of reference prior abnormal big data of the reference cloud service abnormal event;

step S120, for each reference abnormal feature sequence in a plurality of reference abnormal feature sequences, performing spatio-temporal node feature random conversion according to the reference abnormal features in the reference abnormal feature sequences, and mixing at least two spatio-temporal node conversion reference features in spatio-temporal node conversion reference features obtained by performing spatio-temporal node feature random conversion on the reference abnormal features in the reference abnormal feature sequences to generate a mixed abnormal feature sequence corresponding to the reference abnormal feature sequence;

step S130, performing sequence forward selection on each space-time node conversion reference feature in the mixed abnormal feature sequence corresponding to the reference abnormal feature sequence, determining a reference redirection abnormal feature sequence, and loading the reference redirection abnormal feature sequence into the reference abnormal feature sequence to obtain a derivative reference abnormal feature sequence of the reference abnormal feature sequence;

step S140, training an abnormal knowledge point decision model for abnormal knowledge point decision based on a plurality of derived reference abnormal feature sequences of a plurality of reference abnormal feature sequences, wherein the trained abnormal knowledge point decision model performs abnormal knowledge point decision on the target fusion features to obtain abnormal optimization decision information corresponding to the target cloud service abnormal event.

In one possible implementation manner, in step S120, a spatio-temporal node feature random transformation may be performed on each reference abnormal feature in the reference abnormal feature sequence to generate a first spatio-temporal node transformation feature sequence, a second spatio-temporal node transformation feature sequence is generated by scrambling spatio-temporal node transformation reference features in the first spatio-temporal node transformation feature sequence based on a preset feature arrangement rule, and each spatio-temporal node transformation reference feature in the first spatio-temporal node transformation feature sequence is mixed with a corresponding one of the spatio-temporal node transformation reference features in the second spatio-temporal node transformation feature sequence to generate a third spatio-temporal node transformation feature sequence as a mixed abnormal feature sequence corresponding to the reference abnormal feature sequence.

Or in another possible embodiment, in step S120, a scrambling process based on a preset feature arrangement rule may be performed on reference abnormal features in the reference abnormal feature sequence to generate a scrambled abnormal feature sequence corresponding to the reference abnormal feature sequence, then space-time node feature random conversion is performed on each reference abnormal feature in the reference abnormal feature sequence and each reference abnormal feature in the scrambled abnormal feature sequence to obtain a first space-time node conversion feature sequence and a second space-time node conversion feature sequence, and finally each space-time node conversion reference feature in the first space-time node conversion feature sequence is mixed with a corresponding space-time node conversion reference feature in the second space-time node conversion feature sequence to generate a third space-time node conversion feature sequence as a mixed abnormal feature sequence corresponding to the reference abnormal feature sequence.

In a possible implementation manner, this embodiment may further determine whether the number of reference abnormal features in the reference abnormal feature sequence is greater than a preset number and/or the number of derived dimensions is greater than a preset number, and when it is determined that the number of reference abnormal features in the reference abnormal feature sequence is not greater than the preset number and/or the number of derived dimensions is not greater than the preset number of dimensions, generate a mixed abnormal feature sequence corresponding to the reference abnormal feature sequence according to a mixture between spatio-temporal node feature random conversion of the reference abnormal features in the reference abnormal feature sequence and at least two spatio-temporal node conversion reference features in spatio-temporal node conversion reference features obtained by subjecting the reference abnormal features in the reference abnormal feature sequence to spatio-temporal node feature random conversion, and generate a derived reference abnormal feature sequence of the reference abnormal feature sequence based on the reference abnormal feature sequence and the mixed abnormal feature sequence corresponding to the reference abnormal feature sequence.

In one possible implementation, each reference abnormal feature in each reference abnormal feature sequence includes channel features of a plurality of channels, wherein each reference abnormal feature in the reference abnormal feature sequence is subjected to space-time node feature random conversion, and a first space-time node conversion feature sequence is generated, and the method includes: respectively carrying out space-time node feature random conversion on channel features of a plurality of channels included in the reference abnormal features aiming at each reference abnormal feature, and determining feature distribution including a plurality of space-time node conversion mapping features corresponding to the plurality of channels one by one, wherein the feature distribution is used as the space-time node conversion reference features corresponding to the reference abnormal features; and taking the space-time node conversion reference characteristics corresponding to each reference abnormal characteristic in the reference abnormal characteristic sequence as a first space-time node conversion characteristic sequence of the reference abnormal characteristic sequence.

In a possible implementation manner, the scrambling processing of the spatio-temporal node conversion reference features in the first spatio-temporal node conversion feature sequence based on a preset feature arrangement rule to generate a second spatio-temporal node conversion feature sequence includes: arranging the space-time node conversion reference characteristics of the first space-time node conversion characteristic sequence; and extracting each time-space node conversion reference feature in the first time-space node conversion feature sequence based on the preset feature arrangement rule, and scrambling each time-space node conversion reference feature according to the preset feature arrangement rule to obtain a second time-space node conversion feature sequence.

In one possible embodiment, mixing each spatio-temporal node transformation reference feature in the first spatio-temporal node transformation signature sequence with a corresponding one spatio-temporal node transformation reference feature in the second spatio-temporal node transformation signature sequence to generate a third spatio-temporal node transformation signature sequence, comprises: respectively mixing the feature distribution of the ith space-time node conversion reference feature in the first space-time node conversion feature sequence with the space-time node conversion mapping feature of the feature distribution of the ith space-time node conversion reference feature in the second space-time node conversion feature sequence according to channels, and determining new feature distribution corresponding to a serial number i, wherein i is more than or equal to 1 and less than or equal to N; and taking each new characteristic distribution corresponding to each serial number as the third time-space node conversion characteristic sequence.

For some possible implementations, the AI analysis system 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

The processor 110 may perform various appropriate actions and processes by a program stored in the machine-readable storage medium 120, such as program instructions related to the big data analysis method optimized for cloud traffic service anomaly described in the foregoing embodiments. The processor 110, the machine-readable storage medium 120, and the communication unit 140 perform signal transmission through the bus 130.

In particular, the processes described in the above exemplary flow diagrams may be implemented as computer software programs, according to embodiments of the present invention. For example, embodiments of the invention include a computer program product comprising a computer program embodied on a computer-readable medium, the computer program comprising program code for performing the method illustrated in the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network via the communication unit 140, and when executed by the processor 110, performs the above-described functions defined in the methods of the embodiments of the present invention.

The invention further provides a computer-readable storage medium, in which computer-executable instructions are stored, and when the computer-executable instructions are executed by a processor, the computer-readable storage medium is used for implementing the big data analysis method for cloud business service anomaly optimization according to any one of the above embodiments.

Yet another embodiment of the present invention further provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the method for analyzing big data for cloud business service anomaly optimization according to any one of the above embodiments is implemented.

The foregoing is only an alternative implementation of some implementation scenarios in this application, and it should be noted that a person having ordinary skill in the art can also use other similar implementation means based on the technical idea of this application without departing from the technical idea of the present application, and the scope of protection of the embodiments of this application also belongs to this application.

Claims

1. A big data analysis method aiming at cloud business service abnormity optimization is characterized by comprising the following steps:

2. The big data analysis method aiming at cloud service anomaly optimization according to claim 1, wherein the step of performing feature extraction on an anomaly relationship attribute graph containing a target cloud service anomaly event and a target page running node, and determining a first graph relationship feature of the target cloud service anomaly event and a second graph relationship feature of the target page running node specifically comprises:

performing multilayer hidden layer feature extraction on a connection page running node of the target cloud service abnormal event, and determining a plurality of cloud service abnormal event features of the target cloud service abnormal event;

fusing the cloud service abnormal event characteristics to determine a first graph relation characteristic of the target cloud service abnormal event;

performing multilayer hidden layer feature extraction on the cloud service connection abnormal event of the target page operation node, and determining a plurality of page operation node features of the target page operation node;

3. The big data analysis method aiming at cloud service anomaly optimization according to claim 2, wherein the hidden layer feature extraction is realized based on node relation inference networks, and mapping interconnection relations are configured among the node relation inference networks;

the step of extracting multilayer hidden layer features of the connection page operation node of the target cloud service abnormal event and determining a plurality of cloud service abnormal event features of the target cloud service abnormal event specifically comprises the following steps:

based on a first node relation reasoning network in an entity relation reasoning network in a mapping interconnection form, carrying out regularization conversion on initial page operation node characteristics of a connection page operation node of the target cloud service abnormal event, and determining the cloud service abnormal event characteristics of the target cloud service abnormal event in the first node relation reasoning network;

the step of extracting multilayer hidden layer characteristics of the cloud service connection abnormal event of the target page running node and determining a plurality of page running node characteristics of the target page running node specifically comprises the following steps:

the step of continuing to perform regularization conversion based on the next mapping interconnected entity relationship inference network and determining the cloud service abnormal event characteristics of the target cloud service abnormal event in the next mapping interconnected entity relationship inference network specifically includes:

and aggregating the regularized conversion characteristics corresponding to the plurality of connection page operation nodes respectively, and determining the cloud service abnormal event characteristics of the target cloud service abnormal event in the xth node relation inference network.

4. The big data analysis method for cloud service anomaly optimization according to claim 1, wherein the step of performing feature extraction on an anomaly relationship attribute graph including a target cloud service anomaly event and a target page operation node, and determining a first graph relationship feature of the target cloud service anomaly event and a second graph relationship feature of the target page operation node specifically includes:

determining a relationship graph between vertexes of the abnormal relationship attribute graph;

and determining a second graph relation characteristic of the target page operation node from the target hidden layer characteristic according to the characteristic point of the target page operation node in the target hidden layer characteristic.

5. The big data analysis method aiming at cloud service anomaly optimization according to claim 4, wherein the hidden layer feature extraction is realized based on node relation inference networks, and mapping interconnection relations are configured among the node relation inference networks;

the step of extracting the multilayer hidden layer features of the abnormal relation attribute graph according to the relationship graph between the vertexes and determining the multiple hidden layer features of the abnormal relation attribute graph specifically comprises the following steps:

based on a first node relation inference network in an entity relation inference network in a mapping interconnection form, carrying out regularized feature extraction on the initial hidden layer features of the abnormal relation attribute graph and the relation graph between the vertexes, and determining the hidden layer features of the abnormal relation attribute graph in the first node relation inference network;

determining a connection number array of the abnormal relation attribute graph;

6. The big data analysis method for cloud service anomaly optimization according to claim 1, wherein the step of performing feature extraction on the event linkage relationship graph containing the target cloud service anomaly event and determining a third graph relationship feature of the target cloud service anomaly event specifically comprises:

and mapping the weighted fusion characteristics of the target cloud service abnormal event, and determining the third graph relation characteristics of the target cloud service abnormal event.

7. The big data analysis method aiming at cloud service anomaly optimization according to claim 1, wherein the step of performing anomaly optimization decision according to the first graph relation feature, the second graph relation feature, the third graph relation feature and a priori anomaly big data of the target cloud service anomaly event to determine anomaly optimization decision information of the target cloud service anomaly event corresponding to the target page operation node specifically comprises:

acquiring prior abnormal big data of the target cloud service abnormal event;

8. The big data analysis method aiming at cloud service anomaly optimization according to claim 7, wherein the step of performing anomaly optimization decision on the fusion features and determining anomaly optimization decision information of the target cloud service anomaly event corresponding to the target page operation node comprises the following steps:

for each reference abnormal feature sequence in a plurality of reference abnormal feature sequences, performing spatio-temporal node feature random conversion according to the reference abnormal features in the reference abnormal feature sequences, and generating a mixed abnormal feature sequence corresponding to the reference abnormal feature sequence by mixing at least two spatio-temporal node conversion reference features in spatio-temporal node conversion reference features obtained by performing spatio-temporal node feature random conversion on the reference abnormal features in the reference abnormal feature sequences;

performing sequence forward selection on each space-time node conversion reference feature in a mixed abnormal feature sequence corresponding to the reference abnormal feature sequence, determining a reference redirection abnormal feature sequence, and loading the reference redirection abnormal feature sequence into the reference abnormal feature sequence to obtain a derivative reference abnormal feature sequence of the reference abnormal feature sequence;

9. The big data analysis method for cloud service anomaly optimization according to claim 8, wherein a mixed anomaly feature sequence corresponding to the reference anomaly feature sequence is generated by performing spatio-temporal node feature random conversion according to the reference anomaly feature in the reference anomaly feature sequence and mixing at least two spatio-temporal node conversion reference features in spatio-temporal node conversion reference features obtained by performing spatio-temporal node feature random conversion on the reference anomaly feature in the reference anomaly feature sequence, and the method comprises:

scrambling the space-time node conversion reference features in the first space-time node conversion feature sequence based on a preset feature arrangement rule to generate a second space-time node conversion feature sequence, and mixing each space-time node conversion reference feature in the first space-time node conversion feature sequence with a corresponding space-time node conversion reference feature in the second space-time node conversion feature sequence to generate a third space-time node conversion feature sequence serving as a mixed abnormal feature sequence corresponding to the reference abnormal feature sequence; or alternatively

performing space-time node characteristic random conversion on each reference abnormal characteristic in the reference abnormal characteristic sequence and each reference abnormal characteristic in the scrambled abnormal characteristic sequence to respectively obtain a first space-time node conversion characteristic sequence and a second space-time node conversion characteristic sequence;

10. An AI analysis system comprising a processor and a memory for storing a computer program executable on the processor, the processor being configured to execute the big data analysis method optimized for cloud traffic service anomalies of any one of claims 1-9 when executing the computer program.