A method for temporal knowledge graph reasoning in cloud platform operation and maintenance

Abstract

The present invention discloses a method for reasoning a time series knowledge graph for cloud platform operation and maintenance, including establishing a time series knowledge graph reasoning model based on a graph convolutional neural network; constructing a reasoning graph according to the query time; weighted sampling the neighborhood of the reasoning graph node v; distinguishing different nodes from the same entity cloud platform component and cloud infrastructure in the reasoning graph; and integrating the relationship characteristics between the entity cloud platform component and the cloud infrastructure into the characteristics of the entity itself. The reasoning ability of the graph neural network is used to model and predict system operation and maintenance, thereby improving the efficiency and quality of the operation and maintenance of the cloud platform. The knowledge graph is constructed based on time series data, which can more comprehensively describe the operation of the system, and can better use historical data to predict the future. By using the reasoning ability of the graph neural network, the efficiency and quality of system operation and maintenance can be greatly improved, and it has great application potential.

Description

Technical Field

[0001] The present invention relates to the technical field of constructing knowledge graphs, and in particular to a method for reasoning about a time-series knowledge graph for cloud platform operation and maintenance. Background Art

[0002] Cloud computing platforms have become an important infrastructure for modern enterprises to build business architectures. However, as the scale of cloud platforms continues to expand, the difficulty of operation and maintenance has also increased exponentially. In order to better manage operation and maintenance data and improve operation and maintenance efficiency, cloud platform operation and maintenance time series knowledge graphs have been proposed and applied to the field of cloud platform operation and maintenance. However, the construction and reasoning of knowledge graphs are very time-consuming and difficult tasks, and there are even more challenges in traditional rule-based methods.

[0003] In recent years, knowledge graph construction and reasoning methods based on deep learning have received widespread attention and application. Among them, methods based on graph convolutional neural networks have achieved certain results in knowledge graph reasoning. Graph convolutional neural networks are widely used in fields such as image classification and social network analysis. They can learn complex relationships between nodes to better reason.

[0004] At present, the Chinese invention with application number CN 115277453A discloses a method for generating, an application method and a device for an abnormal knowledge graph in the field of operation and maintenance, which can automatically generate an abnormal knowledge graph in the field of operation and maintenance based on the abnormal data generated when an abnormal event occurs. However, in the process of constructing the knowledge graph, how to extract meaningful entities and relationships from a large number of time series data sources such as logs, monitoring and configuration of the cloud platform is a problem that needs to be solved; how to use information such as historical operation and maintenance data of the cloud platform to predict the status and trend of the cloud platform in the future is a problem that needs to be solved; how to capture the time series change characteristics of the knowledge graph model is a problem that needs to be solved. Summary of the invention

[0005] The technical problems solved by the present invention are: in the process of constructing a knowledge graph, how to extract meaningful entities and relationships from a large number of time-series data sources such as logs, monitoring and configurations of the cloud platform is a problem that needs to be solved; how to use information such as historical operation and maintenance data of the cloud platform to predict the status and trend of the cloud platform in the future is a problem that needs to be solved; how to capture the time-series change characteristics of the knowledge graph model is a problem that needs to be solved.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a cloud platform operation and maintenance time series knowledge graph reasoning method, comprising establishing a time series knowledge graph reasoning model based on a graph convolutional neural network; constructing a reasoning graph according to the query time; performing weighted sampling on the neighborhood of the reasoning graph node v; distinguishing different nodes from the same entity cloud platform components and cloud infrastructure in the reasoning graph; fusing the relationship features between the entity cloud platform components and the cloud infrastructure into the features of the entity itself; capturing the implicit time series features between the entity cloud platform components and the cloud infrastructure through GRU; normalizing the feature vectors output by the neighborhood feature aggregator; and obtaining the probabilities of all candidate entities and the final prediction results.

[0007] As a preferred solution of the cloud platform operation and maintenance time series knowledge graph reasoning method described in the present invention, the establishment of a time series knowledge graph reasoning model based on a graph convolutional neural network includes:

[0008] The temporal knowledge graph reasoning model is represented by a set G of quadruple (s, r, o, t), each quadruple G represents an event, where s represents the head entity cloud platform component, r represents the relationship, i.e., the fault information in the cloud platform, o represents the tail entity cloud infrastructure, and t represents time.

[0009] As a preferred solution of the cloud platform operation and maintenance time series knowledge graph reasoning method described in the present invention, wherein: the time series knowledge graph reasoning model includes a neighborhood sampler, a time encoder, a neighborhood feature aggregator and a multi-layer perception decoder;

[0010] When a query P appears, p=(s p ,r p ,o p ,t p ).

[0011] As a preferred solution of the cloud platform operation and maintenance time series knowledge graph reasoning method described in the present invention, wherein: constructing a reasoning graph according to the query time includes:

[0012] The temporal knowledge graph reasoning model builds a reasoning graph G with reference to the temporal neighborhood where P is located. i ;

[0013] Define a node in the inference graph as v = (s, t), and its temporal neighborhood as M v ={(s',t')},t'<t;

[0014] Using the time encoder to perceive the learning process of physical cloud platform components and cloud infrastructure;

[0015] The neighborhood feature aggregator captures structural and temporal features within the neighborhood of physical cloud platform components and cloud infrastructure;

[0016] The multi-layer perceptual decoder calculates the probability of the entity cloud platform components and cloud infrastructure becoming the final prediction answer.

[0017] As a preferred solution of the cloud platform operation and maintenance time series knowledge graph reasoning method described in the present invention, distinguishing different nodes from the same entity cloud platform components and cloud infrastructure in the reasoning graph includes:

[0018] Neighborhood sampler pair (s p ,t p )’s temporal neighborhood is sampled;

[0019] Different relationship trends are represented by different time range information, and local time is used to represent relationship trends in a short period of time, while long-term time is used to represent relationship trends over a long period of time.

[0020] As a preferred solution of the cloud platform operation and maintenance time series knowledge graph reasoning method described in the present invention, weighted sampling of the neighborhood of the reasoning graph node v includes:

[0021] The learned time perception of each entity is represented by the time encoder, entity cloud platform components and cloud infrastructure i ∈E is composed of low-dimensional static vectors and time series vectors, which are expressed in the form of embedding of t. It distinguishes two nodes (s, t1) and (s, t2) from the same physical cloud platform components and cloud infrastructure but belonging to different times. The mathematical expression of time-aware physical cloud platform components and cloud infrastructure embedding is:

[0022]

[0023] in, Represents the characteristics of static entity cloud platform components and cloud infrastructure. Here, it represents the characteristics of entity cloud platform components and cloud infrastructure in the temporal knowledge graph that do not change over time. It refers to the temporal characteristics of physical cloud platform components and cloud infrastructure, that is, the characteristics of physical cloud platform components and cloud infrastructure changing over time. a[i], c[i], and b[i] represent the physical cloud platform components and cloud infrastructure s i The parameter matrix of σ represents the sine function, d s and d T Represent the dimensions of static embedding and temporal embedding respectively;

[0024] The node relationship in the reasoning graph is represented by r∈R. Since the relationship representation in the reasoning graph has the property of being time-invariant, its mathematical expression is:

[0025]

[0026] Among them, C0 represents a relationship embedding matrix, and r represents the relationship vector of the node.

[0027] As a preferred solution of the cloud platform operation and maintenance time series knowledge graph reasoning method described in the present invention, the integration of the relationship characteristics between the entity cloud platform components and the cloud infrastructure into the characteristics of the entity itself includes:

[0028] The neighborhood feature aggregator obtains the temporal features of the local entity by aggregating the information in the neighborhood;

[0029] Representing embedded vector relationships through physical cloud platform components and cloud infrastructure of temporal encoders;

[0030] The neighborhood feature aggregator is used to fuse the relational features of the vector relationship with the features of the physical cloud platform components and the cloud infrastructure using a Chebyshev polynomial approximation method;

[0031] Restrict the kernel η to be a polynomial of μ, that is where η∈R k represents the coefficient vector of the polynomial, K represents the size of the graph convolution kernel, K determines the maximum radius of the convolution from the central node, and the Chebyshev polynomial T k (x) is regarded as an approximate kernel, and its k-1-order truncated expansion is:

[0032]

[0033] in, Λ max represents the maximum eigenvalue of L, L represents the characteristic matrix of the graph convolution operator, and the mathematical expression of the graph convolution neural network is:

[0034]

[0035] in, represents a k-order Chebyshev polynomial, after scaling Laplace Calculation is performed at .

[0036] As a preferred solution of the cloud platform operation and maintenance time series knowledge graph reasoning method described in the present invention, the implicit time series features between the physical cloud platform components and the cloud infrastructure are captured by GRU, including:

[0037] Before updating the embedding vectors of the physical cloud platform components and cloud infrastructure, the neighbor nodes in the reasoning graph are sorted in chronological order, and the time series features are used as the input of GRU to obtain the implicit time series features of the physical cloud platform components and cloud infrastructure, whose mathematical expression is:

[0038] hv =σ(X·GRU(h w,t-n ,h w,t-n+1 ,…,h w,t ))

[0039] Among them, X represents the learnable weight matrix, h w,t represents the fused features of the neighborhood node w of node v at time t in the inference graph, and σ represents the activation function ReLU of the activation weight matrix.

[0040] As a preferred solution of the cloud platform operation and maintenance time series knowledge graph reasoning method described in the present invention, the feature vector output by the neighborhood feature aggregator is normalized, including:

[0041] After the neighborhood feature aggregator outputs the feature vector, the multi-layer perceptual decoder takes the feature vector as input and normalizes it through the soft function to obtain the probability of the entity cloud platform components and cloud infrastructure. The mathematical expression is:

[0042] p(s,o|r,t)=soft(X u ·(h v ||e r ))

[0043] Among them, X u represents the trainable weight parameter, h v represents the embedding vectors of the entity cloud platform components and cloud infrastructure after aggregation, e r represents the embedded relation vector, and p(s,or,t) represents the probability of entity cloud platform components and cloud infrastructure.

[0044] As a preferred solution of the cloud platform operation and maintenance time series knowledge graph reasoning method described in the present invention, the probability of obtaining all candidate entities and the final prediction results are expressed as follows:

[0045] o t =argsoft(p(s,o|r,t))

[0046]

[0047] Among them, t represents the final prediction result, G represents the events in the training set, p(s k ,o k |r,t) represents the entity s under the condition of known fault relationship and time k Cloud platform components and entities k Probability of cloud infrastructure.

[0048] The beneficial effects of the present invention are as follows: by using graph convolutional neural networks to model graph data, it is possible to effectively capture the changing patterns of time series data and node feature information; by using time series data to construct knowledge graphs, the relationships and changes between different nodes are modeled, which can more comprehensively describe the operation of the system; based on the time series knowledge graph and graph convolutional neural network model, the reasoning ability of the graph neural network is used to model and predict system operation and maintenance, thereby improving the operation and maintenance efficiency and quality of the cloud platform; by constructing knowledge graphs based on time series data, it is possible to more comprehensively describe the operation of the system, and better use historical data to predict the future; by using the reasoning ability of the graph neural network, the efficiency and quality of system operation and maintenance can be greatly improved, and it has great application potential. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A basic flow chart of a cloud platform operation and maintenance time series knowledge graph reasoning method provided for an embodiment of the present invention.

[0050] Figure 2 A temporal knowledge graph reasoning model architecture diagram of a cloud platform operation and maintenance temporal knowledge graph reasoning method provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0052] Example 1

[0053] Reference Figure 1 and 2 , is an embodiment of the present invention, and provides a cloud platform operation and maintenance time series knowledge graph reasoning method, including:

[0054] Step 1: Establish a temporal knowledge graph reasoning model based on graph convolutional neural network;

[0055] The temporal knowledge graph reasoning model is represented by a set G of quadruple (s, r, o, t), each quadruple G represents an event, where s represents the head entity cloud platform component, r represents the relationship, i.e., the fault information in the cloud platform, o represents the tail entity cloud infrastructure, and t represents time.

[0056] Graph convolutional neural network is a learning method that can be used to reflect structural features. It mainly captures the structural dependencies between entities in the same period and the sequence information through graph neural network. Using graph convolutional neural network to model the graph can make full use of the relationship between nodes and improve modeling accuracy.

[0057] The temporal knowledge graph reasoning model includes a neighborhood sampler, a temporal encoder, a neighborhood feature aggregator, and a multi-layer perceptual decoder;

[0058] When a query P appears, p=(s p ,r p ,o p ,t p ).

[0059] The quadruple G constructs an inference graph based on the query time, performs weighted sampling on the node neighborhood on the inference graph, distinguishes different nodes from the same entity cloud platform component or infrastructure in the inference graph, and integrates the relationship features between the entity cloud platform component and the infrastructure into the features of the entity itself. The implicit temporal features between the entity cloud platform components or infrastructure are captured through GRU. Finally, the feature vector output by the neighborhood feature aggregator is normalized to obtain the probability of the candidate entity and the final prediction result.

[0060] Step 2: Constructing the inference graph based on the query time includes:

[0061] The temporal knowledge graph reasoning model builds a reasoning graph G with reference to the temporal neighborhood where P is located. i ;

[0062] Define a node in the inference graph as v = (s, t), and its temporal neighborhood as M v ={(s',t')},t'<t; then the time encoder is used to perceive the learning process of the entity cloud platform components and cloud infrastructure, and the neighborhood feature aggregator captures the structural and temporal features within the neighborhood of the entity cloud platform components and cloud infrastructure. Finally, the multi-layer perception decoder can be used to calculate the probability that the entity cloud platform components and cloud infrastructure will become the final prediction answer.

[0063] Using the time encoder to perceive the learning process of physical cloud platform components and cloud infrastructure;

[0064] The neighborhood feature aggregator captures structural and temporal features within the neighborhood of physical cloud platform components and cloud infrastructure;

[0065] The multi-layer perceptual decoder calculates the probability of the entity cloud platform components and cloud infrastructure becoming the final prediction answer.

[0066] Meaningful entities and relationships are extracted using a large number of time-series data sources such as the cloud platform's logs, monitoring, and configurations. The cloud platform's historical operation and maintenance data and other information are used to predict the future status and trends of the cloud platform, thereby capturing the time-series change characteristics of the knowledge graph model and improving operation and maintenance efficiency and accuracy.

[0067] Step 3: Distinguish different nodes in the reasoning graph from the same entity cloud platform components and cloud infrastructure, including:

[0068] Neighborhood sampler pair (s p ,t p )’s temporal neighborhood is sampled;

[0069] Different relationship trends are represented by different time range information, and local time is used to represent relationship trends in a short period of time, while long-term time is used to represent relationship trends over a long period of time.

[0070] Different relationship trends are represented by different time range information, and local time is used to represent the relationship trend in a short period of time, and long-term time is used to represent the relationship trend in a long period of time. Time information should not be measured by the same standard, and different ranges will also affect time information. Therefore, in order to increase the time factor that affects the accuracy of the model.

[0071] Step 4: Weighted sampling of the neighborhood of node v in the inference graph includes:

[0072] The learned time perception of each entity is represented by the time encoder, entity cloud platform components and cloud infrastructure i ∈E is composed of low-dimensional static vectors and time series vectors, which are expressed in the form of embedding of t. It distinguishes two nodes (s, t1) and (s, t2) from the same physical cloud platform components and cloud infrastructure but belonging to different times. The mathematical expression of time-aware physical cloud platform components and cloud infrastructure embedding is:

[0073]

[0074] in, Represents the characteristics of static entity cloud platform components and cloud infrastructure. Here, it represents the characteristics of entity cloud platform components and cloud infrastructure in the temporal knowledge graph that do not change over time. It refers to the temporal characteristics of physical cloud platform components and cloud infrastructure, that is, the characteristics of physical cloud platform components and cloud infrastructure changing over time. a[i], c[i], and b[i] represent the physical cloud platform components and cloud infrastructure s i The parameter matrix of σ represents the sine function, d s and d TThey represent the dimensions of static embedding and temporal embedding respectively; the sine function can simulate the opening and closing of the state in a more diverse way, thereby combining the temporal characteristics with the control of weights. Compared with the static knowledge graph, the structure of all graphs in the temporal knowledge graph is not static, and the characteristics of the entity cloud platform components and cloud infrastructure will also change with the graph structure. The time encoder can represent the learning time perception of each entity.

[0075] The node relationship in the reasoning graph is represented by r∈R. Since the relationship representation in the reasoning graph has the property of being time-invariant, its mathematical expression is:

[0076]

[0077] Among them, C0 represents a relationship embedding matrix, and r represents the relationship vector of the node. In the process of embedding the entity cloud platform components and cloud infrastructure in the reasoning graph, the embedding of the relationship between the entity cloud platform components and cloud infrastructure can also be represented.

[0078] Step 5: Integrate the relationship features between the entity cloud platform components and cloud infrastructure into the features of the entity itself, including:

[0079] The neighborhood feature aggregator mainly obtains the temporal features of local entities by aggregating information within the neighborhood. The input of the neighborhood feature aggregator can represent the relationship of the embedding vector through the entity cloud platform components and cloud infrastructure of the temporal encoder, and then the obtained relationship features can be fused with the features of the entity cloud platform components and cloud infrastructure, and finally the temporal features implicit in the entity cloud platform components and cloud infrastructure are captured through GRU.

[0080] The neighborhood feature aggregator obtains the temporal features of the local entity by aggregating the information in the neighborhood;

[0081] Representing embedded vector relationships through physical cloud platform components and cloud infrastructure of temporal encoders;

[0082] The neighborhood feature aggregator is used to fuse the relational features of the vector relationship with the features of the physical cloud platform components and the cloud infrastructure using a Chebyshev polynomial approximation method;

[0083] Restrict the kernel η to be a polynomial of μ, that is where η∈R k represents the coefficient vector of the polynomial, K represents the size of the graph convolution kernel, K determines the maximum radius of the convolution from the central node, and the Chebyshev polynomial T k (x) is regarded as an approximate kernel, and its k-1-order truncated expansion is:

[0084]

[0085] in, Λ max represents the maximum eigenvalue of L, L represents the characteristic matrix of the graph convolution operator, and the mathematical expression of the graph convolution neural network is:

[0086]

[0087] in, represents a k-order Chebyshev polynomial, after scaling Laplace Calculation is performed at .

[0088] During each reasoning process of the temporal knowledge graph reasoning model, the neighborhood sampler will synthesize the static knowledge graph sequence into an inference graph, in which all structural and temporal information can be expressed. The temporal and structural features in the temporal knowledge graph reasoning model will be captured by the message passing framework of the graph convolutional neural network.

[0089] Reasoning on the temporal knowledge graph means reasoning about future events based on past events. In order to obtain the interactive relationship between different time periods, the temporal knowledge graph reasoning model uses a time-aware neighborhood sampler to sample the neighborhood nodes of the physical cloud platform components and cloud infrastructure. At the same time, in order to solve the problem of difficulty in extracting temporal features during the reasoning process, a temporal encoder will be used to map the cloud platform components, cloud infrastructure and time in the quadruple into an embedding vector. The neighborhood feature aggregator can perceive the features of the neighborhood in time, which is conducive to refining the temporal features of the temporal knowledge graph reasoning model and enhancing the reasoning performance of the temporal knowledge graph.

[0090] Step 6: Capturing implicit temporal features between physical cloud platform components and cloud infrastructure through GRU includes:

[0091] Before updating the embedding vectors of the physical cloud platform components and cloud infrastructure, the neighbor nodes in the reasoning graph are sorted in chronological order, and the time series features are used as the input of GRU to obtain the implicit time series features of the physical cloud platform components and cloud infrastructure, whose mathematical expression is:

[0092] h v =σ(X·GRU(h w,t-n ,h w,t-n+1 ,…,h w,t ))

[0093] Among them, X represents the learnable weight matrix, h w,t represents the fused features of the neighborhood node w of node v at time t in the inference graph, and σ represents the activation function ReLU of the activation weight matrix.

[0094] Step 7: Normalizing the feature vector output by the neighborhood feature aggregator includes:

[0095] After the neighborhood feature aggregator outputs the feature vector, the multi-layer perceptual decoder takes the feature vector as input and normalizes it through the soft function to obtain the probability of the entity cloud platform components and cloud infrastructure. The mathematical expression is:

[0096] p(s,o|r,t)=soft(X u ·(h v ||e r ))

[0097] Among them, X u represents the trainable weight parameter, h v represents the embedding vectors of the entity cloud platform components and cloud infrastructure after aggregation, e r represents the embedded relation vector, and p(s,or,t) represents the probability of entity cloud platform components and cloud infrastructure.

[0098] Step 8: Get the probabilities of all candidate entities and the final prediction results. The mathematical expression is:

[0099] o t =argsoft(p(s,o|r,t))

[0100]

[0101] Among them, t represents the final prediction result, G represents the events in the training set, p(s k ,o k |r,t) represents the entity s under the condition of known fault relationship and time k Cloud platform components and entities k Probability of cloud infrastructure.

[0102] Using graph convolutional neural networks to model graph data can effectively capture the changing patterns and node feature information of time series data. Using time series data to build a knowledge graph and model the relationships and changes between different nodes can more comprehensively describe the operation of the system. Based on the time series knowledge graph and graph convolutional neural network model, the reasoning ability of the graph neural network is used to model and predict system operation and maintenance, thereby improving the operation and maintenance efficiency and quality of the cloud platform. Building a knowledge graph based on time series data can more comprehensively describe the operation of the system and better use historical data to predict the future. By using the reasoning ability of the graph neural network, the efficiency and quality of system operation and maintenance can be greatly improved, and it has great application potential.

[0103] It should be appreciated that embodiments of the present invention may be implemented or enforced by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable memory. The method may be implemented in a computer program using standard programming techniques - including a non-transitory computer-readable storage medium configured with a computer program, wherein the storage medium so configured causes the computer to operate in a specific and predefined manner - according to the methods and drawings described in the specific embodiments. Each program may be implemented in a high-level procedural or object-oriented programming language to communicate with a computer system. However, if desired, the program may be implemented in an assembly or machine language. In any case, the language may be a compiled or interpreted language. In addition, the program may be run on a programmed ASIC for this purpose.

[0104] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A cloud platform operation and maintenance time series knowledge graph reasoning method, characterized in that: include: Establish a temporal knowledge graph reasoning model based on graph convolutional neural network; Build an inference graph based on the query time; Perform weighted sampling on the neighborhood of node v in the inference graph; Distinguish different nodes in the reasoning graph from the same physical cloud platform components and cloud infrastructure; Incorporate the relationship characteristics between the entity cloud platform components and cloud infrastructure into the characteristics of the entity itself; Capture the implicit temporal features between physical cloud platform components and cloud infrastructure through GRU; Normalize the feature vector output by the neighborhood feature aggregator; Get the probabilities of all candidate entities and the final prediction results; The establishment of a temporal knowledge graph reasoning model based on graph convolutional neural network includes: The temporal knowledge graph reasoning model is represented by a set G of four-tuples (s, r, o, t), each of which represents an event, where s represents the head entity cloud platform component, r represents the relationship, i.e., the fault information in the cloud platform, o represents the tail entity cloud infrastructure, and t represents time; The temporal knowledge graph reasoning model includes a neighborhood sampler, a temporal encoder, a neighborhood feature aggregator, and a multi-layer perceptual decoder; When a query P appears, p=(s p ,r p ,o p ,t p ); Building an inference graph based on the query time includes: The temporal knowledge graph reasoning model builds a reasoning graph G with reference to the temporal neighborhood where P is located. i ; Define a node in the inference graph as v = (s, t), and its temporal neighborhood as M v ={(s',t')},t'<t; Using the time encoder to perceive the learning process of physical cloud platform components and cloud infrastructure; The neighborhood feature aggregator captures structural and temporal features within the neighborhood of physical cloud platform components and cloud infrastructure; The multi-layer perceptual decoder calculates the probability of the entity cloud platform component and the cloud infrastructure becoming the final prediction answer; Differentiating different nodes in the inference graph from the same entity cloud platform components and cloud infrastructure includes: Neighborhood sampler pair (s p ,t p )’s temporal neighborhood is sampled; Different relationship trends are represented by different time range information, and local time is used to represent short-term relationship trends, and long-term time is used to represent long-term relationship trends; Weighted sampling of the neighborhood of a node v in the inference graph includes: The learned time perception of each entity is represented by the time encoder, entity cloud platform components and cloud infrastructure i ∈E is composed of low-dimensional static vectors and time series vectors, which are expressed in the form of embedding of t. It distinguishes two nodes (s, t1) and (s, t2) from the same physical cloud platform components and cloud infrastructure but belonging to different times. The mathematical expression of time-aware physical cloud platform components and cloud infrastructure embedding is: in, Represents the characteristics of static entity cloud platform components and cloud infrastructure. Here, it represents the characteristics of entity cloud platform components and cloud infrastructure in the temporal knowledge graph that do not change over time. It refers to the temporal characteristics of physical cloud platform components and cloud infrastructure, that is, the characteristics of physical cloud platform components and cloud infrastructure changing over time. a[i], c[i], and b[i] represent the physical cloud platform components and cloud infrastructure s i The parameter matrix of σ represents the sine function, d s and d T Represent the dimensions of static embedding and temporal embedding respectively; The node relationship in the reasoning graph is represented by r∈R. Since the relationship representation in the reasoning graph has the property of being time-invariant, its mathematical expression is: Among them, C0 represents a relationship embedding matrix, r represents the relationship vector of the node; Integrating the relationship characteristics between the entity cloud platform components and cloud infrastructure into the characteristics of the entity itself includes: The neighborhood feature aggregator obtains the temporal features of the local entity by aggregating the information in the neighborhood; Representing embedded vector relationships through physical cloud platform components and cloud infrastructure of temporal encoders; The neighborhood feature aggregator is used to fuse the relational features of the vector relationship with the features of the physical cloud platform components and the cloud infrastructure using a Chebyshev polynomial approximation method; Restrict the kernel η to be a polynomial of μ, that is where η∈R k represents the coefficient vector of the polynomial, K represents the size of the graph convolution kernel, K determines the maximum radius of the convolution from the central node, and the Chebyshev polynomial T k (x) is regarded as an approximate kernel, and its k-1-order truncated expansion is: in, Λ max represents the maximum eigenvalue of L, L represents the characteristic matrix of the graph convolution operator, and the mathematical expression of the graph convolution neural network is: in, represents a k-order Chebyshev polynomial, after scaling Laplace Calculation is performed at The implicit temporal features between physical cloud platform components and cloud infrastructure captured by GRU include: Before updating the embedding vectors of the physical cloud platform components and cloud infrastructure, the neighbor nodes in the reasoning graph are sorted in chronological order, and the time series features are used as the input of GRU to obtain the implicit time series features of the physical cloud platform components and cloud infrastructure, whose mathematical expression is: h v =σ(X·GRU(h w,t-n ,h w,t-n+1 ,…,h w,t )) Among them, X represents the learnable weight matrix, h w,t represents the fused features of the neighborhood node w of node v at time t in the inference graph, and σ represents the activation function ReLU of the activation weight matrix.

2. The cloud platform operation and maintenance time series knowledge graph reasoning method according to claim 1, characterized in that: Normalizing the feature vector output by the neighborhood feature aggregator includes: After the neighborhood feature aggregator outputs the feature vector, the multi-layer perceptual decoder takes the feature vector as input and normalizes it through the soft function to obtain the probability of the entity cloud platform components and cloud infrastructure. The mathematical expression is: p(s,o|r,t)=soft(X u ·(h v ||e r )) Among them, X u represents the trainable weight parameter, h v represents the embedding vectors of the entity cloud platform components and cloud infrastructure after aggregation, e r represents the embedded relation vector, and p(s,o|r,t) represents the probability of entity cloud platform components and cloud infrastructure.

3. The cloud platform operation and maintenance time series knowledge graph reasoning method according to claim 2, characterized in that: Get the probability of all candidate entities and the final prediction results. The mathematical expression is: o t =argsoft(p(s,o|r,t)) Among them, t represents the final prediction result, G represents the events in the training set, p(s k ,o k |r,t) represents the entity s under the condition of known fault relationship and time k Cloud platform components and entities k Probability of cloud infrastructure.

Images