A network fault diagnosis method for power data center based on dependent network model

Abstract

The present invention discloses a method for diagnosing network faults of an electric data center based on a dependent network model, including: constructing a cascading failure model of the dependent network of the electric data center according to the log information of the terminal device accessing the electric data center; calculating the initial load and capacity for each edge in the network, simulating random network failures and updating the network status, calculating the current load of each edge in the updated network, and deleting the edge from the network if the current load is greater than the capacity of the edge; repeating status updates and load calculations and deletions until all edges in the electric data center network meet the condition that the current load is less than or equal to the capacity of the edge, or all nodes in the network are isolated nodes; calculating the current network model function value, and providing a quantitative judgment result. The present invention is oriented to complex electric data center scenarios, and captures the real-time changes of node status through dependent network modeling, effectively supporting status perception and fault analysis.

Description

Technical Field

[0001] The present invention relates to a method for diagnosing network faults in a power data center based on a dependent network model, and belongs to the technical field of power data centers. Background Art

[0002] With the completion of the construction of the two-level data center of power enterprises, the overall business data shows a trend of centralized storage. At present, power enterprises are in the digital transformation period, the construction and application of power grid informatization are gradually deepening, and various types of business data are growing rapidly. Power data shows the characteristics of multiple businesses, large scale and high value, which puts forward stronger requirements for the power data center network.

[0003] Traditional network fault diagnosis technology usually adopts a diagnostic mode of in-process monitoring and post-analysis. Based on the monitoring data, a fault propagation model is established to locate the fault and find the root cause of the network anomaly. However, the physical connection and business logic connection relationship in the power data center network often causes many related power equipment to issue alarms when a fault occurs, and even seriously affects the normal regulation and fault recovery of the power network. Therefore, it is necessary and urgent to study a fast and accurate power data center network fault diagnosis method. Summary of the invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a network fault diagnosis method based on a dependent network model to achieve network status analysis and fault diagnosis identification for a power data center.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A network fault diagnosis method based on a dependent network model comprises the following steps:

[0007] A cascading failure model of the dependent network of the power data center is constructed based on the log information of the terminal devices accessing the power data center. The two layers of networks in the dependent network are the user subject network and the behavior network. The edge between the two layers of networks is the user's access behavior. The nodes in the subject network are the terminal devices accessing the power data center. If there are two terminal devices accessing the same log, there is an edge between the nodes corresponding to the two terminal devices. The nodes in the behavior network are the short paths in the access path between the center data. If there is an access relationship between the two short paths, there is an edge between the corresponding two nodes.

[0008] Calculate the initial load and capacity for each edge in the network, simulate random network failures and update the network status, calculate the current load of each edge in the updated network, and delete the edge from the network if the current load is greater than the capacity of the edge;

[0009] Repeat the status update, load calculation, and deletion until all edges in the power data center network meet the requirement that the current load is less than or equal to the edge capacity, or all nodes in the network are isolated nodes;

[0010] Calculate the current network model function W = FoP-SoCF value, give the quantitative network model status evaluation result to be tested, and identify the network fault situation based on the analysis result, where FoP is the fault optimization rate and SoCF is the fault penetration.

[0011] Furthermore, constructing the cascading failure model of the power data center network includes: defining M = (N, E) to represent a multi-layer network, where N = {N α ,N β},N α represents the set of nodes in the main network, N β is the node set in the behavior network, and E is the edge in the dependency network, expressed as:

[0012] E={E αα ,E αβ ,E ββ}

[0013] In the formula, E αα represents the edge set in the main network, E αβ N is the main network node α and behavioral network node N β The connecting edge between them, E ββ is the edge set in the behavior network.

[0014] Furthermore, the fault penetration degree SOCF is defined as:

[0015]

[0016] In the formula, D represents the number of nodes in the network, V N Represents the number of nodes in the isolated connected subgraph that has been separated.

[0017] Furthermore, the failure optimization rate FoP is defined as:

[0018]

[0019] In the formula, define D S is the number of simulated cascading failures, D S (i) represents node v i The number of failures.

[0020] Furthermore, the initial load K of each edge in the network is ij The calculation is as follows:

[0021]

[0022] Where Y lb Represents node n l With n b The number of shortest paths between lb (e ij ) represents node n l With n b Through the edge e ij The number of shortest paths;

[0023] The capacity of the edge R ij Defined as:

[0024] R ij =(1+α)K ij

[0025] Wherein α>0 is a preset parameter.

[0026] Further, simulating random network failures and updating network status includes:

[0027] Randomly select a node in the network and its associated edges to delete and update the network status;

[0028] Randomly select the network nodes and their associated edges associated with the deleted node for deletion, and then update the network status.

[0029] Compared with the prior art, the beneficial effects achieved by the present invention are: breaking the diagnostic analysis mode of post-judgment of traditional network failures, establishing a reasonable fault propagation model based on dependent networks, simulating the process of cascading failure of the power data center network, analyzing the intrinsic connection between the topological attributes of the power data center network nodes and the cascading failure of the network nodes, fitting and grasping the changes in network status, and realizing network fault diagnosis of the power data center. The present invention provides a simple and effective diagnostic idea for the acquisition and identification of network status of large-scale and high-traffic power data centers. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A flow chart of the network fault diagnosis method for the power data center provided by the present invention;

[0031] Figure 2 A diagram of the cascading failure model of the power data center network provided by the present invention;

[0032] Figure 3 It is a comparison diagram of network status analysis situations under different methods provided by the present invention. DETAILED DESCRIPTION

[0033] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0034] The present invention provides a method for diagnosing network faults in a power data center based on a dependent network model, which specifically includes the following contents:

[0035] Step 1: Construct a network cascading failure model for power data centers based on the dependent network model.

[0036] In the present invention, the two layers of the dependent network model are the user subject network and the behavior network, and the edge between the two layers of the network is the user's access behavior. The nodes in the subject network are terminal devices that access the power data middle station. If there are two terminal devices accessing the same log, there is an edge between the nodes corresponding to the two terminal devices. The nodes in the behavior network are short paths in the access path between the middle station data. If there is an access relationship between the two short paths, that is, there is an access path that passes through the two short paths respectively, the access path is the edge in the behavior network.

[0037] The field data of the log is symbolized as a log data model; the collected logs are represented as an ordered set L = {l1,l2,…,l m}; Each log message includes specific parameters, including: user IP (IPc), request size (Rs), request method (M), request path (U), status code (Sc) and date (D), as shown in Table 1.

[0038] Table 1 Network log composition

[0039]

[0040] Every time a user accesses the data center, a log message is generated. In the log message, the path of the network resource requested by the user is an important credential for determining the user's access intention.

[0041] Construct a cascading failure model for the power data center network and define M = (N, E) to represent a multi-layer network, where N = {N α ,N β},N α represents the set of nodes in the main network, N β is the node set in the behavior network, and E is the edge in the dependency network, expressed as:

[0042] E={E αα ,E αβ ,E ββ} (1)

[0043] In the formula, E αα represents the edge set in the main network, E αβ N is the main network node α and behavioral network nodes N βThe connecting edge between them, E ββ is the edge set in the behavioral network. At the same time, the connection between layers is defined as a one-to-one mode. Figure 2 The figure shows the cascading failure model diagram of the power data center network. The dependent network is the entire network including the subject network and the behavior network.

[0044] Step 2: Use fault penetration degree (SoCF) to quantify the impact of node failure on the communication quality of the data center.

[0045] In the present invention, the fault penetration degree SOCF is defined as:

[0046]

[0047] Where D represents the number of nodes in the dependent network, V N Represents the number of nodes in the isolated connected subgraph that has been separated.

[0048] Step 3: Introduce the Fault Optimization Rate (FoP) to quantify the ability of a node to work again after failure;

[0049] In the present invention, the failure optimization rate FoP is defined as:

[0050]

[0051] Where D S is the number of simulated cascading failures, D S (i) represents node v i The number of failures.

[0052] Step 4: Calculate the initial load K for each edge in the dependency network ij and capacity R ij :

[0053]

[0054] Where Y lb Represents node n l With n b The number of shortest paths between lb (e ij ) represents node n l With n b Through the edge e ij The number of shortest paths, with the center of the betweenness as the initial load of the edge, so that the capacity of the edge R ij It can be defined as:

[0055] R ij =(1+α)K ij (5)

[0056] Among them, α>0 is a parameter set in advance by humans;

[0057] Step 5: To simulate the randomness of network failures, randomly select a node in the network and its associated edges to delete and update the network status;

[0058] Step 6: Randomly select the network nodes and their associated edges associated with the deleted node for deletion, and calculate the load K' of each edge in the updated network ij ;

[0059] Step 7: If K' ij >R ij ; then delete edge e from the network ij ;

[0060] Step 8: Return to step 6 until all edges in the power data center network satisfy K' ij ≤R ij Or all nodes in the network are isolated nodes;

[0061] In the processing of steps 5 to 8, the deleted nodes and the nodes corresponding to the deleted edges constitute a separated isolated connected subgraph. The number of these nodes is an important indicator for measuring the severity of the fault in the entire network. Therefore, the present invention constructs a fault penetration index based on this number, see step 2.

[0062] Step 9: Calculate the value of the current network model function W=FoP-SoCF to quantitatively evaluate the state of the network model to be tested.

[0063] According to the W value calculated in step 9, a comprehensive evaluation result of the network model status can be obtained. Since the construction of the network model is based on the access relationship between the terminal device and the service, which implies the node topology of the data center, the fault situation in the complex network can be identified according to the evaluation result, so that targeted measures can be taken to deal with the fault. For example, a high degree of fault penetration indicates that there are a large number of faulty nodes in the data center network, and the focus can be on troubleshooting single-point failures of the equipment; a high fault optimization rate indicates that the network cascade failures are relatively more than the network node failures, indicating that there are deeper cascade effects between related devices, and attention should be paid to troubleshooting between functional modules; from the difference between the two and their changing trends over time, we can understand the changes in the equipment and the cascade relationship between devices.

[0064] In order to prove that the method of the present invention can realize the network status analysis of the power data center, the heuristic algorithm and the specific routing approximation algorithm are used as comparison methods, and the different methods are applied to the actual power data center network fault analysis experiment.

[0065] At the same time, the mean square error (MSE) is introduced to measure the gap between the predicted value and the true value of the indicator. The smaller the MSE indicator, the more accurate the prediction effect.

[0066]

[0067] Among them, p i For node m i The predicted value, d i is the predicted value.

[0068] The network state fitting under different methods is as follows Figure 3 shown.

[0069] like Figure 3 As shown, the network fault analysis method proposed by the present invention can better perform situational awareness and fault analysis on complex power data middle station networks, and is better for complex network analysis than other methods. This method uses fault penetration to quantify the impact of node failure on the communication quality of the data middle station, and introduces the fault optimization rate to quantify the ability of the node to work again after failure. It can better simulate the state change process of the data middle station network, analyze the topological attributes of the power data middle station network nodes and the intrinsic connection between the cascading failure of the network nodes, quantitatively evaluate the operating status of the network model to be tested, and realize the power data middle station network fault diagnosis.

[0070] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for diagnosing network faults in power data centers based on a dependent network model, characterized in that: The following steps are involved: A cascading failure model of the dependent network of the power data center is constructed based on the log information of the terminal devices accessing the power data center. The two layers of networks in the dependent network are the user subject network and the behavior network. The edge between the two layers of networks is the user's access behavior. The nodes in the user subject network are the terminal devices accessing the power data center. If there are two terminal devices accessing the same log, there is an edge between the nodes corresponding to the two terminal devices; if there is an access relationship between the nodes in the behavior network, there is an edge between the corresponding two nodes; Calculate the initial load and capacity for each edge in the network, simulate random network failures and update the network status, calculate the current load of each edge in the updated network, and delete the edge from the network if the current load is greater than the capacity of the edge; Repeat the status update, load calculation, and deletion until all edges in the power data center network meet the requirement that the current load is less than or equal to the edge capacity, or all nodes in the network are isolated nodes; Calculate the current network model function W = FoP-SoCF value, and give the quantitative judgment result of the network status in the power data according to the change trend of W value over time. According to the judgment result, identify the network fault situation, where FoP is the fault optimization rate and SoCF is the fault penetration degree; The failure optimization rate FoP is defined as: Where D S is the number of simulated cascading failures, D S (i) represents node v i The number of failures; The fault penetration degree SOCF is defined as: In the formula, D represents the number of nodes in the network, V N Represents the number of nodes in the isolated connected subgraph that has been separated.

2. The method according to claim 1, characterized in that Constructing the cascading failure model of the power data center network includes: defining M = (N, E) to represent a multi-layer network, where N = {N α ,N β },N α represents the node set in the user agent network, N β is the node set in the behavior network, and E is the edge in the dependency network, expressed as: And={And αα ,AND αβ ,AND ββ } In the formula, E αα represents the edge set in the user agent network, E αβ N is the user's main network node α and behavioral network node N β The connecting edge between them, E ββ is the edge set in the behavior network.

3. The method according to claim 1, characterized in that The initial load K of each edge in the network ij The calculation is as follows: Where Y lb Represents node n l With n b The number of shortest paths between lb (e ij ) represents node n l With n b Through the edge e ij The number of shortest paths; The capacity of the edge R ij Defined as: R ij =(1+α)K ij Wherein α>0 is a preset parameter.

4. The method according to claim 1, characterized in that: Simulating random network failures and updating the network status includes: Randomly select a node in the network and its associated edges to delete and update the network status; Randomly select the network nodes and their associated edges associated with the deleted node for deletion, and then update the network status.

Images