JP7329017B2 - Diagnostic device, diagnostic method, program and diagnostic system - Google Patents

Description

The present invention relates to a diagnostic device, diagnostic method, program and diagnostic system.

For example, Patent Literature 1 discloses a network device used in an IP relay network managed by a telecommunications carrier (carrier) and a technique for monitoring the communication path of the network device.

JP 2005-184638 A

When receiving a report from a user that communication is not possible, telecommunications carriers issue various commands to each communication device that makes up the communication network, check the response results, and perform fault diagnosis. there is However, in recent years, the number and types of communication devices that constitute a communication network have diversified in accordance with the expansion of the scale of communication networks and the expansion of services. Furthermore, since the diagnostic method is different for each communication device, the trouble diagnosis work performed by the communication carrier has become very complicated.

The present disclosure has been made in view of such circumstances, and aims to improve the efficiency of identifying failure types in communication networks.

A diagnostic apparatus according to an aspect of the present invention includes an identifying unit that identifies a predetermined communication device group accommodating a predetermined communication service from among a plurality of communication devices included in a communication network, and a collection unit for inputting a command to each of the communication devices included in the predetermined communication device group and collecting predetermined command responses indicating operation states of the communication devices included in the predetermined communication device group; and an output unit for outputting the failure type determined by the determination unit.

It is a figure which shows the system configuration example of the diagnostic system which concerns on this embodiment. It is a figure which shows the hardware structural example of a diagnostic apparatus and a communication apparatus. It is a figure which shows the functional block structural example of a diagnostic apparatus. It is a figure which shows the functional block structural example of a communication apparatus. FIG. 4 is a diagram for explaining feature vectors; FIG. FIG. 4 is a diagram showing an example of labels given to feature vectors; FIG. 5 is a flow chart showing an example of a processing procedure for determining a failure type; An example of NW configuration DB is shown. FIG. 4 is a diagram showing an example of commands and confirmation rules; 10 is a flowchart showing an example of a processing procedure for generating teacher data by performing clustering of feature vectors; FIG. 4 is a diagram showing an example of feature vectors labeled by clustering; It is a figure which shows the example which visualized the feature vector. It is a figure which shows an example of the trained model using a decision tree.

Embodiments of the present invention will be described with reference to the accompanying drawings. It should be noted that, in each figure, the same reference numerals have the same or similar configurations.

<System configuration>
FIG. 1 is a diagram showing a system configuration example of a diagnostic system 1 according to this embodiment. The diagnostic system 1 includes a diagnostic device 10, a plurality of communication networks N1-N4 and a user network U.

Communication networks N1 to N4 are communication networks managed by communication carriers. The communication networks N1 to N4 are composed of a plurality of communication devices 20-1 to 20-17 arranged in facilities managed by communication carriers. The communication device 20 is, for example, a network device such as a router, switch, hub, firewall, ONU (Optical Network Unit), or the like. In the following description, the multiple communication devices 20-1 to 20-17 are referred to as the communication device 20 when not distinguished. Also, there is no limit to the number of communication devices 20 forming the communication networks N1 to N4. In addition, in the example of FIG. 1, four communication networks N1 to N4 are shown as communication networks managed by a telecommunications carrier, but there is no limit to the number of communication networks managed by a telecommunications carrier. In the following description, the communication networks N1 to N4 are referred to as a communication network N when not distinguished. Although not shown, the communication network N (for example, communication network N4) may be connected to the Internet.

A user network U is a communication network managed by a user. The communication network managed by the user includes, for example, a wireless LAN communication environment in an individual user's home, a corporate user's network, and the like. Although only user network U is illustrated in the example of FIG. 1, the number of user networks is not limited.

Since the communication network N1 connects with the user network U, it may be called an access network (access network). On the other hand, the communication networks N2-N4 may be called core networks.

The diagnostic device 10 is a device that monitors the operating states of a plurality of communication devices 20 that constitute communication networks N1 to N4 managed by a communication carrier. More specifically, the diagnosis device 10 monitors the state of each interface of the communication device 20, the processing load of the main body of the communication device 20, and the like, based on the command responses received from each communication device 20. FIG. The interface provided in the communication device 20 means a physical interface for connecting cables and/or a logical interface created within the physical interface. A logical interface is sometimes called a virtual interface.

Here, the flow of communication service failure response work in network operation work that is currently being performed will be described. First, a user who detects a communication service failure makes a report to the counter using the service ID assigned to each user as a key. Using the service ID as a key, the operator of the carrier confirms the communication network (for example, access network, relay network, etc.) accommodating the user's communication service and the communication device 20, remotely logs in to each communication device 20, Execute a command for failure diagnosis and identify the suspected failure point.

In addition, the failure response system is used to coordinate information such as worker arrangement, on-site recovery work instructions, normality confirmation results, etc., with terminals used by workers. Each command has a confirmation rule for fault diagnosis based on the response result. As for the confirmation rule, the contents of confirmation differ for each communication network, and the command to be executed by each communication network and each communication device 20 is also different accordingly. Therefore, in order to perform failure diagnosis such as identification of failure location and cause for a certain failure, it is necessary to check the command response results regarding all the communication devices 20 in which the user is accommodated, and to make a decision based on the results. There is In addition, there are a large number of commands to be executed for each communication device 20, and the text messages are complex text messages on the premise that the response results are visually determined. Therefore, the current failure diagnosis work is mainly manual work that depends on the expertise of the operator.

(Outline of diagnostic system)
When the diagnostic device 10 receives a report, for example, by telephone or e-mail, from a user who uses a service provided by a telecommunications carrier that communication is not possible, the diagnostic device 10 uses the reported service ID as a key to connect to the communication networks N1 to N3. A plurality of communication devices 20 that accommodate the communication service of the user are identified from among the plurality of communication devices 20 included in the . For example, in the example of FIG. 1, the route through which the user's communication service passes corresponds to the thick line in FIG. , 20-9 and 20-13. Next, the diagnostic device 10 remotely logs into the plurality of communication devices 20 that accommodate the specified user's communication service, and executes a command to check the operating state of each communication device 20 . , command responses indicating the operating status of each communication device 20 are collected. Next, the diagnostic device 10 determines the failure type based on the collected command response results.

In this embodiment, the "failure event" constitutes a feature vector used for learning, and the "failure type" is assigned to each failure event as a label and estimated. In other words, the "failure event" means the details of the "actually occurring" failure, and the "fault type" means the "estimated" details of the failure.

In this embodiment, the diagnostic device 10 provides at least one or more of the following three functions.
1. Fault type discrimination function using supervised learning: A trained model is prepared in advance using labeled feature vectors generated by assigning fault types as labels to feature vectors generated based on command response data. A function that determines the failure type by inputting the feature vector based on the command responses generated and collected by the user's report into the trained model. The labeled feature vector was previously generated by past fault diagnosis work.
2. Auxiliary function for creating teacher data using unsupervised learning: Feature vectors generated based on many command responses from each communication device 20 collected in the past are clustered using unsupervised learning to create a plurality of feature vectors. A function that uses a labeled feature vector generated by classifying faults into classes and assigning a label indicating the fault type to each class as training data.
The training data creation assistance function using unsupervised learning should be used when, for example, the number of labeled feature vectors generated in advance by past failure diagnosis work is not sufficient for model learning. I assume. Note that clustering refers to grouping a plurality of data into a plurality of groups based on the degree of similarity between the data. Further, it is assumed that each class and a specific failure event are associated in advance by an operator or the like of the telecommunications carrier.
3. Input command optimization function: Learning a decision tree using the labeled feature vectors, and using the learned decision tree, extracts commands that are effective for specifying the failure type from among the many commands provided in the communication device 20. By doing so, it is possible to efficiently collect command responses from the communication device 20 .

<Functional block configuration>
FIG. 2 is a diagram showing a hardware configuration example of the diagnostic device 10 and the communication device 20. As shown in FIG. The diagnostic device 10 and the communication device 20 include a processor 11 such as a CPU (Central Processing Unit) and a GPU (Graphical Processing Unit), a memory, a storage device 12 such as a HDD (Hard Disk Drive) and/or an SSD (Solid State Drive), It has a communication IF (Interface) 13 that performs wired or wireless communication, an input device 14 that receives input operations, and an output device 15 that outputs information. The input device 14 of the diagnostic apparatus 10 is, for example, a keyboard, touch panel, operation panel, input terminal, mouse and/or microphone. The output device 15 is, for example, a display, an output terminal, a touch panel and/or a speaker.

(Diagnostic device)
FIG. 3 is a diagram showing a functional block configuration example of the diagnostic device 10. As shown in FIG. The diagnostic device 10 includes a storage unit 100, a command generation unit 101, a collection unit 102, a preprocessing unit 103, a determination unit 104, an output unit 105, a reception unit 106, a generation unit 107, and a learning processing unit. 108 and a classifying unit 109 . The storage unit 100 can be implemented using the storage device 12 included in the diagnostic device 10 . Also, a command generation unit 101, a collection unit 102, a preprocessing unit 103, a determination unit 104, an output unit 105, a reception unit 106, a generation unit 107, a learning processing unit 108, and a class classification unit 109 can be realized by the processor 11 of the diagnostic device 10 executing a program stored in the storage device 12 . Also, the program can be stored in a storage medium. The storage medium storing the program may be a non-transitory computer readable medium. The non-temporary storage medium is not particularly limited, but may be a storage medium such as a USB memory or CD-ROM, for example.

The storage unit 100 stores a NW (Network) configuration DB (Data Base) 100a, a trained model 100b, a command response DB 100c, and a teacher data DB 100d.

The NW configuration DB 100a stores the connection relationship of each communication device 20 that constitutes the communication network N, the model of the communication device 20, the communication device 20 that accommodates the communication service of each user, and the communication service of each user in each communication device 20. Stores information such as identifiers of interfaces and ports.

The trained model 100b is a model that has been trained to have the ability to output a label indicating an estimated failure type when inputting a feature vector generated based on command responses collected from a plurality of communication devices 20. be. The trained model 100b includes feature vectors generated based on command responses collected from a plurality of communication devices 20 accommodating communication services in the communication network N, and labels indicating failure types corresponding to the feature vectors. It is generated by learning using the associated teacher data.

What kind of model is used in the trained model 100b, for example, a decision tree (for example, CART, XG Boost, etc.), SVM (support vector machine), neural network, multilayer neural network, random forest, logistic regression analysis, etc. good too.

The command response DB 100c stores command responses collected by transmitting commands to a plurality of communication devices 20 accommodating user communication services when an operator of a communication carrier or the like identifies a failure event. The command response DB 100c stores a large amount of command responses collected when fault isolation was performed in the past. The command response DB 100c is assumed to be used in the "teaching data creation assisting function using unsupervised learning".

The training data DB 100d stores training data in which feature vectors generated based on command responses obtained from a plurality of communication devices 20 accommodating user communication services are associated with labels indicating correct failure types. do. The learned model 100b described above may be learned using the teacher data DB 100e. The training data DB 100d may be generated in advance through past failure diagnosis work, or may be generated using the "teaching data creation assistance function using unsupervised learning".

The command generation unit 101 identifies a predetermined communication device group accommodating a predetermined communication service from among the plurality of communication devices 20 included in the communication network N, and generates a command to be executed by the identified predetermined communication device group. do. The predetermined communication service is a communication service for which the diagnosis device 10 investigates the failure type, and may be, for example, the communication service of the user who received the failure report.

The predetermined communication service includes any communication service. A communication service directed from a company's base A) to another user network U (for example, the same user company's base B) through a communication network N (or vice versa). Further, the predetermined communication device group means one or more communication devices 20 out of a large number of communication devices 20 existing in the communication network N through which data flowing through the communication service to be investigated passes.

The collection unit 102 executes the command generated by the command generation unit 101 in each of the communication devices 20 included in the predetermined communication device group identified by the command generation unit 101, thereby enabling the communication devices 20 included in the predetermined communication device group. Predetermined command responses indicative of the operational status of each communication device 20 received are collected.

The preprocessing unit 103 generates feature vectors based on the predetermined command responses collected by the collection unit 102 .

The determination unit 104 determines a failure type in a predetermined communication service based on the predetermined command responses collected by the collection unit 102 . More specifically, the determination unit 104 inputs the feature vectors generated by the preprocessing unit 103 based on the predetermined command responses collected by the collection unit 102 to the learned model 100b, and outputs the feature vectors from the learned model 100b. A failure type in a predetermined communication service may be determined by acquiring a label indicating the failure type to be used.

The output unit 105 outputs the failure type determined by the determination unit 104 . For example, the output unit 105 may cause a display included in the diagnostic device 10 or a device connected to the diagnostic device 10 to display information indicating the failure type.

The reception unit 106 determines whether the failure type in the predetermined communication service determined by the determination unit 104 is correct or erroneous, and the correct failure type in the case where the failure type in the predetermined communication service is erroneous. Accepted from telecommunications carrier operators, etc.

When the reception unit 106 receives that the fault type in the predetermined communication service is correct, the generation unit 107 generates the feature vector generated based on the predetermined command response collected by the collection unit 102 and the determination unit 104 is generated in association with the label indicating the failure type in the predetermined communication service, and stored in the training data DB 100d. Further, when the reception unit 106 receives a correct failure type in a predetermined communication service, the generation unit 107 generates a feature vector generated based on the predetermined command response collected by the collection unit 102 and the reception unit 106 Teacher data associated with the received label indicating the correct failure type is generated and stored in the teacher data DB 100d.

The learning processing unit 108 generates a trained model 100b by learning the model using the teacher data DB 100e.

The class classification unit 109 clusters feature vectors generated based on a plurality of command responses collected from a plurality of communication devices 20 stored in the command response DB 100c into a plurality of classes using unsupervised learning. I do. Further, the class classification unit 109 assigns a label indicating a failure type corresponding to the class to which each feature vector belongs to each feature vector, and stores the label in the database to generate the teacher data DB 100d. The learning processing unit 108 learns the model using the feature vectors stored in the teacher data DB 100d and the labels corresponding to the clusters to which the feature vectors belong as teacher data, thereby generating the learned model 100b. good too.

Furthermore, the command generation unit 101 extracts specific commands used for branching of the decision tree in the learned model 100b trained using the decision tree. For example, the command generation unit 101 may extract all commands used for determining each branch in the learned decision tree as the specific commands.

Further, the collection unit 102 executes commands included in at least the specific commands extracted by the command generation unit 101 in each of the communication devices included in the predetermined communication device group, thereby obtaining commands included in the predetermined communication device group. Predetermined command responses that indicate the operational status of the communication device may be collected.

(Communication device)
FIG. 4 is a diagram showing a functional block configuration example of the communication device 20. As shown in FIG. Communication device 20 includes storage unit 200 , communication processing unit 201 , management unit 202 , input unit 203 and output unit 204 . The storage unit 200 can be implemented using the storage device 12 included in the communication device 20 . Also, the communication processing unit 201 can be realized using the communication IF 13 provided in the communication device 20 . Also, the management unit 202 , the input unit 203 , and the output unit 204 can be implemented by the processor 11 of the communication device 20 executing programs stored in the storage device 12 . Also, the program can be stored in a storage medium. The storage medium storing the program may be a computer-readable non-temporary storage medium. The non-temporary storage medium is not particularly limited, but may be a storage medium such as a USB memory or CD-ROM, for example.

The storage unit 200 stores various types of information necessary for the operation of the communication device 20, such as routing tables and various types of setting information. The communication processing unit 201 routes received data based on a routing table or the like. The management unit 202 manages the operating state of the communication device 20 (CPU usage rate, memory usage rate, status of each interface, data transmission/reception status, etc.). The input unit 203 receives inputs such as commands provided in the communication device 20 . The output unit 204 transmits a command response indicating the operating state of the communication device 20 to the diagnostic device 10 .

<Processing procedure>
(Composition of feature vector)
The diagnosis device 10 logs into each communication device 20 that accommodates the declared communication service, and executes a series of commands (hereinafter referred to as "command series") necessary for fault diagnosis. Further, the response results for all device commands are converted into feature vectors, and the feature vectors are input to the learned model 100b to estimate the failure type.

The diagnostic device 10 combines the verification results of the command responses obtained from each communication device 20 that accommodates the declared communication service and constructs one feature vector for each failure event. Here, the type of command executed for the communication device 20 that accommodates the communication service of each user differs depending on the communication device 20, and furthermore, the communication device 20 that accommodates the communication service of each user differs from user to user. . Therefore, in the present embodiment, all the communication devices 20 managed by the diagnostic device 10 and accommodating the communication services of each user are arranged so that the dimensions of the feature vectors are the same for all users. Let the number of commands be the number of dimensions. For example, if all device commands of all communication devices 20 accommodating communication services of each user are 350 commands in total, the feature vector becomes a 350-dimensional vector.

Next, the command response from the communication device 20 is composed of a complex text message that is assumed to be judged visually by humans. However, since the diagnosis device 10 uses the learned model 100b to determine the failure type, it is necessary to allow the learned model 100b to handle command responses collected from the communication device 20. FIG. Therefore, the diagnosis device 10 compares the command response result of each communication device 20 with a predefined confirmation rule, and converts the confirmation result for the response content of each command into a real number, thereby generating a feature vector. . The real number may be, for example, 0 (normal)/1 (abnormal), or may be further classified into a plurality of numerical values.

FIG. 5 is a diagram for explaining feature vectors. Although communication networks N1 to N4 are illustrated in the example of FIG. 5, this is for convenience of illustration, and the number of communication networks N is not limited. In FIG. 5, "communication network" is all the communication networks N that accommodate the communication services of each user, and "communication device" is installed in each communication device 20 and accommodates the communication service of each user. All communication devices 20 . "Command ID" is an identifier that uniquely identifies all device commands of all communication devices 20 that accommodate communication services of each user. For example, if all device commands of all communication devices 20 accommodating communication services of each user are 350 commands in total, then Cn=C349.

One feature vector is generated by transforming command responses from C1 to Cn to 0 or 1 according to predefined validation rules. For example, if transformed as C0=0, C1=0, C2=1, . , 0, 1, . . . , 0).

(Generation of learned model)
The learning processing unit 108 acquires teacher data (that is, labeled feature vectors) from the teacher data DB 100d and trains the model to generate a trained model 100b. FIG. 6 is a diagram showing an example of labels given to feature vectors. Each element in the table of FIG. 6 indicates nine types of label IDs and failure types. A labeled feature vector is, for example, a feature vector of the form (label ID, C0, C1, C2, . . . , C350)=(5, 0, 0, 1, .

The learning processing unit 108 sets various parameters that need to be set in advance in the model, and performs learning using teacher data. In addition, the learning processing unit 108 verifies whether or not the trained model 100b has been appropriately trained using verification data for verifying the trained model 100b. If the trained model 100b is not sufficiently trained, the parameter change, learning and verification are repeated. Part of the learning data stored in the trained model 100b may be used as verification data.

(Determination of failure type using supervised learning)
FIG. 7 is a flow chart showing an example of a processing procedure in which the diagnostic device 10 determines the failure type. In the example of FIG. 7, it is assumed that the learned model 100b has the ability to output labels indicating the fault types shown in FIG.

In step S10, when a failure report is received from the user, the command generation unit 101 refers to the NW configuration DB 100a and specifies a plurality of communication devices 20 accommodating communication services of the user who received the report.

FIG. 8 is a diagram showing an example of the NW configuration DB 100a. The example of FIG. 8 shows which communication device 20 and in what order the traffic of the communication service specified by each service ID passes through each communication network N1 to N4. For example, the traffic of the communication service with service ID1 passes through the communication devices 20 of A1, A2, B1, B2, C1, C2, D1 and D2 in the communication networks N1 to N4. Note that FIG. 8 is merely an example, and each user's communication service may pass through more communication networks N and communication devices 20 . Further, the communication devices 20 that construct each communication network N may include communication devices 20 of different models. Also, in the NW configuration DB 100a, physical port numbers (or virtual port numbers) and interface numbers that accommodate communication services specified by service IDs in each communication device 20 are stored in association with service IDs. good too.

In addition, the command generation unit 101 generates a command sequence required for failure diagnosis, which is executed by the specified plurality of communication devices 20 accommodating the communication service of the user. The command generation unit 101 may generate a command sequence according to rules predetermined for each communication device 20 . An example of commands and confirmation rules executed by the communication device 20 is shown in FIG.

In step S11, the collection unit 102 logs into each communication device 20 identified in step S10 and executes the generated command sequence to collect command responses.

In step S12, the preprocessing unit 103 generates feature vectors by preprocessing the command responses collected in step S11. Specifically, the preprocessing unit 103 generates a feature vector by converting the received command response group to 0 or 1 according to a predefined confirmation rule. Here, the preprocessing unit 103 sets "0 (normal)" for the dimension corresponding to the command ID not included in the received command response among the feature vectors (C0 to Cn). For example, when the feature vector has 350 dimensions from C0 to C349, the command responses include a total of 100 command responses with command IDs from 0 to 39, 60 to 99, and 200 to 219. and In this case, the preprocessing unit 103 sets “0 (normal)” for the dimensions corresponding to the command IDs of 40 to 59, 100 to 199, and 220 to 349 among the feature vectors (C0 to C349).

At step S13, the determination unit 104 inputs the feature vector generated at step S12 to the learned model 100b. Subsequently, the determination unit 104 acquires a label indicating the failure type output from the trained model 100b. Subsequently, the output unit 105 outputs information indicating the failure type corresponding to the label output from the trained model 100b (for example, the wording indicated by the failure type in FIG. 7) to a display or the like.

The reception unit 106 receives information indicating whether or not the failure type output in the processing procedure of step S13 is correct from the operator of the communication carrier who actually performed the failure analysis with reference to the failure type displayed on the display. , and if the fault type is incorrect, an input of a label indicating the correct fault type may be accepted. Further, the generation unit 107 associates the label value indicating the correct failure type with the feature vector generated in the processing procedure of step S12 to generate new labeled training data, and updates the training data DB 100d. may By using the teacher data DB 100d to which new teacher data has been added in this way, it is also possible to additionally learn the trained model 100b.

(Clustering processing in unsupervised learning)
FIG. 10 is a flowchart showing an example of a processing procedure for generating teacher data by clustering feature vectors.

At step S21, the preprocessing unit 103 generates a feature vector from each command response stored in the command response DB 100c. Note that the preprocessing unit 103 generates feature vectors by the same processing procedure as the processing procedure described in step S12 of FIG. Subsequently, the class classification unit 109 clusters the generated feature vectors using, for example, the k-means method. Here, it is assumed that a plurality of feature vectors are clustered into five classes, for example. The class classification unit 109 assigns a label (eg, 1 to 5) to each feature vector clustered into five classes.

In step S22, the reception unit 106 receives input of information indicating specific failure types for the five labeled classes from the operator of the telecommunications carrier. At this time, the receiving unit 106 may receive correction of the label given to each feature vector by the classifying unit 109 from an operator or the like who performs the clustering process.

FIG. 11 is a diagram showing an example of a database that stores feature vectors labeled by clustering. One row in A of FIG. 11 shows feature vectors and labels given to the feature vectors. Further, B of FIG. 11 shows an example in which the operator of the telecommunications carrier assigns specific failure types to five classes.

In step S23, the generation unit 107 stores in the storage unit 100 the teacher data DB 100d storing the labeled feature vectors obtained in the processing procedure of steps S21 and S22. After that, the learning processing unit 108 uses the teacher data DB 100d stored in the storage unit 100 to generate the learned model 100b.

(Optimization of input command)
In the present embodiment, in order to estimate the failure type with the minimum number of commands, the optimum command sequence to be executed by the communication device 20 is generated in advance by using the information of the decision tree used for failure type estimation. can be A decision tree is represented by a tree structure having a node as a criterion for classification of feature vectors and a leaf as a label by which each feature vector is finally classified. Each node in the decision tree represents which element of the feature vector has what value, that is, the confirmation result of the corresponding command. can figure out what to do. Therefore, by following the nodes in order from the root and extracting the commands used at each node, it is possible to obtain the optimum command sequence from the viewpoint of failure type estimation.

FIG. 12 is a diagram showing an example of visualization of feature vectors collected from each communication device 20 when a failure report is received. The vertical axis in FIG. 12 corresponds to the command type (that is, the number of dimensions of the feature vector, here 350), and the horizontal axis corresponds to each failure event.

Locations with dots in FIG. 12 indicate that the command response is abnormal. Here, as shown in FIG. 12, there is a bias in the commands indicating abnormality among the 350 commands. In other words, only some of the 350 commands are useful for identifying the failure type.

Therefore, the command generation unit 101 extracts specific commands used for branching in the decision tree from among the plurality of feature vectors input to the learned model 100b, which is the decision tree. FIG. 13 is a diagram showing an example of a trained model 100b using a decision tree. For example, it is assumed that the decision tree shown in FIG. 13 has learned to output a label (i) indicating the most probable failure type when commands C0 to C350 are input. In the example of FIG. 15, C4, C16, C70, C80, C100 and C200 are used for branching among the 350 commands forming 350 dimensions. Therefore, the command generator 101 may extract commands C4, C16, C70, C80, C100 and C200 as specific commands.

Note that, in realizing the input command optimization function, the learned model 100b for use in the failure type determination function and the decision tree analyzed by the input command optimization function may be the same model or different models. can be a model. For example, the learning processing unit 108 may create a decision tree model separately from the trained model 100b used for the failure type discrimination function in order to obtain an optimum command sequence.

In the processing procedure of step S11 in FIG. 7 described above, the command generation unit 101 generates a command sequence necessary for failure diagnosis to be executed by the plurality of specified communication devices 20. A command sequence may be generated within the range of the specified command. In other words, the commands automatically executed by the collection unit 102 are narrowed down to the minimum number of command executions that allow discrimination of the failure type based on the information of the decision tree. Since this eliminates the need to transmit unnecessary commands that do not affect the determination result of the failure type to each communication device 20, it is possible to determine the failure type more quickly.

<Summary>
According to the embodiment described above, the diagnostic apparatus 10 determines the failure type by inputting the feature vector to the trained model generated by supervised learning. This makes it possible to efficiently specify the failure type in the communication network. Further, the diagnostic device 10 clusters the command responses collected when a failure report is received by unsupervised learning, thereby assisting creation of teacher data. This makes it possible to generate a learned model even if the amount of teacher data created in past failure diagnosis work is not sufficient.

Further, when the operator determines that the determination result by the learned model 100b is different from the actual failure type, the diagnostic apparatus 10 accepts the correct failure type and also accepts the accepted correct failure type and the collected failure type. The feature vectors generated based on the command responses are associated with each other and saved as teacher data. This makes it possible to efficiently collect teacher data necessary for advancing supervised learning.

Further, the diagnosis device 10 uses the learned model 100b using the decision tree to extract specific commands used for branching in the decision tree. This eliminates the need to send unnecessary commands that are not used to determine the failure type to each communication device 20, so that command responses can be efficiently collected.

The embodiments described above are for facilitating understanding of the present invention, and are not intended to limit and interpret the present invention. Flowcharts, sequences, elements included in the embodiments, their arrangement, materials, conditions, shapes, sizes, and the like described in the embodiments are not limited to those illustrated and can be changed as appropriate. Also, it is possible to partially replace or combine the configurations shown in different embodiments.

Note that, in the present embodiment, the command generation unit 101 is an example of an identification unit and an extraction unit.

DESCRIPTION OF SYMBOLS 1... Diagnostic system, 10... Diagnostic apparatus, 11... Processor, 12... Storage device, 13... Communication IF, 14... Input device, 15... Output device, 20... Communication apparatus, 100... Storage part, 100a... NW configuration DB, 100b... Learned model, 100c... Command response DB, 100d... Teacher data DB, 101... Command generation unit, 102... Collection unit, 103... Preprocessing unit, 104... Discrimination unit, 105... Output unit, 106... Reception unit, 107...Generation unit 108...Learning processing unit 109...Class classification unit 200...Storage unit 201...Communication processing unit 202...Management unit 203...Input unit 204...Output unit

Claims (8)

an identification unit that identifies a predetermined communication device group accommodating a predetermined communication service from among a plurality of communication devices included in a communication network;
By executing a command in each of the communication devices included in the predetermined communication device group identified by the identification unit, a predetermined command response indicating the operating state of each communication device included in the predetermined communication device group is collected. a collection unit that
Using teacher data in which feature vectors generated based on command responses collected from a plurality of communication devices accommodating communication services in the communication network are associated with labels indicating failure types corresponding to the feature vectors A trained model trained by
a determination unit that determines a failure type in the predetermined communication service based on the predetermined command response;
an output unit that outputs the failure type determined by the determination unit;
has
The determination unit inputs a predetermined feature vector generated based on the predetermined command response to the learned model, acquires a label indicating a failure type output from the learned model, and obtains the predetermined determine the failure type in the communication service of
diagnostic equipment. Clustering feature vectors generated based on a plurality of command responses collected from a plurality of communication devices accommodating communication services in the communication network into a plurality of classes, and associating a label indicating a failure type with each class. a storage unit for storing the obtained database;
a learning processing unit that generates the learned model by learning a model using feature vectors stored in the database and labels corresponding to clusters to which the feature vectors belong as teacher data;
A diagnostic device according to claim 1 . a receiving unit that receives whether the failure type in the predetermined communication service determined by the determining unit is correct or incorrect, and the correct failure type in the case where the failure type in the predetermined communication service is erroneous;
The feature vector generated based on the predetermined command response and the predetermined communication service determined by the determination unit when the reception unit receives that the failure type in the predetermined communication service is correct. feature vector generated based on the predetermined command response when the reception unit receives the correct failure type in the predetermined communication service and a generation unit that generates training data in which the label indicating the correct failure type received by the reception unit is associated with
A diagnostic device according to claim 2 . an extraction unit that extracts a specific command used for branching the learned model trained using a decision tree;
A diagnostic device according to claim 1 . The collecting unit indicates the operating state of the communication devices included in the predetermined communication device group by executing at least a command included in the specific command in each of the communication devices included in the predetermined communication device group. collecting the predetermined command response;
A diagnostic device according to claim 4 . identifying a predetermined communication device group accommodating a predetermined communication service from among a plurality of communication devices included in a communication network;
By inputting a command to each communication device included in the predetermined communication device group identified in the identifying step, a predetermined command response indicating the operating state of each communication device included in the predetermined communication device group is generated. a collecting step;
determining a failure type in the predetermined communication service based on the predetermined command response;
and outputting the failure type determined in the determining step ,
In the determining step, a predetermined feature vector generated based on the predetermined command response is compared with a feature vector generated based on command responses collected from a plurality of communication devices accommodating communication services in the communication network. and a label indicating the failure type corresponding to the feature vector are input to a trained model trained using teacher data in correspondence with each other, and a label indicating the failure type output from the trained model is obtained. By determining the failure type in the predetermined communication service,
A diagnostic method performed by a diagnostic device. to the computer,
identifying a predetermined communication device group accommodating a predetermined communication service from among a plurality of communication devices included in a communication network;
By executing a command in each communication device included in the predetermined communication device group identified in the identifying step, a predetermined command response indicating the operating state of each communication device included in the predetermined communication device group is generated. a collecting step;
determining a failure type in the predetermined communication service based on the predetermined command response;
a step of outputting the failure type determined in the determining step;
and
In the determining step, a predetermined feature vector generated based on the predetermined command response is compared with a feature vector generated based on command responses collected from a plurality of communication devices accommodating communication services in the communication network. and a label indicating the failure type corresponding to the feature vector are input to a trained model trained using teacher data in correspondence with each other, and a label indicating the failure type output from the trained model is obtained. By determining the failure type in the predetermined communication service,
program. A diagnostic system having a plurality of communication devices and diagnostic devices included in a communication network,
The communication device
a management unit that manages the operating state of the communication device;
a transmitting unit configured to transmit a command response indicating an operating state of the communication device to the diagnostic device;
The diagnostic device
an identifying unit that identifies, from among the plurality of communication devices, a predetermined communication device group that accommodates a predetermined communication service;
By executing a command in each communication device included in the predetermined communication device group identified by the identification unit, a predetermined command response indicating an operation state of each communication device included in the predetermined communication device group is generated. a collection unit that collects
Using teacher data in which feature vectors generated based on command responses collected from a plurality of communication devices accommodating communication services in the communication network are associated with labels indicating failure types corresponding to the feature vectors A trained model trained by
a determination unit that determines a failure type in the predetermined communication service based on the predetermined command response;
an output unit that outputs the failure type determined by the determination unit;
has
The determination unit inputs a predetermined feature vector generated based on the predetermined command response to the learned model, acquires a label indicating a failure type output from the learned model, and obtains the predetermined determine the failure type in the communication service of
diagnostic system.

Images