KR102291615B1 - Apparatus for predicting failure of communication network and method thereof - Google Patents

Abstract

Disclosed are an apparatus and method for predicting a failure of a communication network based on a neural network. A communication network failure prediction apparatus according to an aspect includes: a collection module for collecting control data of a communication network; a first normalization module for classifying control messages from the collected control data and generating first normalized data obtained by normalizing a traffic pattern of each control message by a predetermined time unit; a traffic pattern prediction module for predicting a traffic pattern of a next period of each control message through a neural network based on the first normalized data; a second normalization module for generating second normalized data normalized by comparing the value of each item constituting the predicted next cycle traffic pattern with a threshold value; and a failure occurrence prediction module for predicting the probability of occurrence of a failure in the communication network through a neural network based on the second normalized data.

Description

Apparatus for predicting failure of communication network and method thereof

The present invention relates to a communication network failure prediction technology, and more particularly, to an apparatus and method for collecting communication network data and predicting a future communication network failure by using a neural network for the collected communication network data.

Deep Learning is a type of artificial intelligence technology that mathematically models a human neural network to solve a problem in a similar way to a neural network.

With the advent of the 4th industrial revolution, interest in and research on artificial intelligence is actively progressing. In particular, as the accuracy of artificial intelligence technology is dramatically improved due to the remarkable development of computing power and the advent of deep learning technology, research to utilize artificial intelligence technology in various fields is being actively conducted.

The mobile revolution represented by smart phones is possible only when a broadband mobile communication network is stably built and operated at the base. When a failure occurs in a mobile communication network, it causes great inconvenience in our lives. Currently, a variety of mobile communication networks, represented by LTE networks, are constructed and operated with a variety of devices, and when a failure occurs in one device, it immediately spreads to other devices, greatly affecting the overall service. Communication network operators are putting a lot of investment and effort into building a state-of-the-art operation management system to prevent network failures. However, in the LTE network, control data and network equipment data emitted by numerous devices are vast enough to reach several TB (Tera Bytes) per day, so even an operator with high skills can analyze this large amount of data in real time and It is almost impossible to predict.

Current technology is mainly applied to detecting failures in equipment and providing real-time notification and maintenance. The technology applied to predict the failure of the communication network has a disadvantage in that the accuracy of prediction is very low because the collected data is analyzed over time, the trend is analyzed and displayed in a graph, so that the operator can predict the failure by looking at the eyes. It may be possible for an experienced operator to predict a failure caused by some factors with the help of the currently applied network failure prediction technology, but it is very difficult to predict a failure that is complexly related to several factors.

Korean Patent Registration 10-0439674 (2004.07.12)

The present invention was created in recognition of the prior art as described above, collects communication network data, normalizes the collected communication network data to data necessary for prediction, and learns various associations that cause communication network failure with respect to the normalized data. An object of the present invention is to provide an apparatus and method for predicting the occurrence of a failure in a communication network through a neural network.

Another object of the present invention is to first predict network data to be generated in the future from the collected communication network data, and to secondarily predict failures that the predicted occurrence of network data may cause.

In addition, another object of the present invention is to selectively execute operator notification and automatic recovery as a precautionary measure in preparation for a predicted failure.

According to one aspect, an apparatus for predicting network failure based on a neural network includes: a collection module for collecting control data of the communication network at regular intervals; a first normalization module for classifying control messages from the control data collected for each period and generating first normalized data obtained by normalizing a traffic pattern of each control message to a predetermined time unit; a traffic pattern prediction module for predicting a traffic pattern of a next period of each control message through a neural network based on the first normalized data; a second normalization module for generating second normalized data normalized by comparing the value of each item constituting the predicted next cycle traffic pattern of each control message with a threshold value; and a failure occurrence prediction module for predicting the probability of occurrence of a failure of the communication network through a neural network based on the second normalized data.

The first normalization module counts and normalizes the number of occurrences, the number of failures, the delay time, and the number of timeouts of the control message constituting the traffic pattern for each predetermined time unit.

The second normalization module compares and normalizes the maximum value and the threshold value for each item constituting the predicted traffic pattern of each control message.

The second normalization module applies different weights to the threshold value and normalizes the value of each item by comparing it with the maximum value.

The second normalization module determines the value of each item as one of values of a type corresponding to a number obtained by adding 1 to the number of weights.

The second normalization module classifies the control messages for each device in the communication network, and normalizes the value of each item constituting the predicted traffic pattern of each classified control message.

The second normalization module applies a threshold value proportional to the value of the traffic pattern of each device.

The failure occurrence prediction module predicts the failure probability for each device.

The failure occurrence prediction module calculates the failure occurrence probability of the equipment with the possibility of failure.

The failure prediction result processing module further includes a failure prediction result processing module for determining the recovery possibility of the failure predicted by the failure occurrence prediction module by querying the failure knowledge database, and processing an operator notification or automatic recovery failure action according to the determination result.

The traffic pattern prediction module is based on a recurrent neural network, and the failure occurrence prediction module is based on a convolutional neural network.

According to another aspect, a communication network failure prediction method executed by a neural network-based communication network failure prediction apparatus includes: collecting control data of the communication network at regular intervals; classifying control messages from the control data collected for each period and generating first normalized data obtained by normalizing a traffic pattern of each control message to a predetermined time unit; predicting a traffic pattern of a next period of each control message through a neural network based on the first normalized data; generating second normalized data normalized by comparing the value of each item constituting the predicted next cycle traffic pattern of each control message with a threshold value; and predicting the probability of occurrence of a failure of the communication network through a neural network based on the second normalized data.

According to an aspect of the present invention, a control message processed by network equipment is predicted, a failure of network equipment that will occur due to the predicted control message is predicted, and recovery measures such as operator notification and command execution are executed for the predicted failure. Therefore, it prepares in advance for the occurrence of network failure.

According to another aspect of the present invention, the reliability of prediction is improved by first predicting control traffic of a control message to be generated in the future through a recurrent neural network, and second predicting the probability of occurrence of a failure for each device due to control traffic through a convolutional neural network. elevate

The following drawings attached to the present specification illustrate preferred embodiments of the present invention, and serve to further understand the technical spirit of the present invention together with the detailed description of the present invention, so that the present invention is described in such drawings. should not be construed as limited to
1 is a schematic configuration diagram of a communication network failure prediction apparatus according to an embodiment of the present invention.
FIG. 2 is an exemplary diagram of a control message classified by the apparatus for predicting network failure of FIG. 1 .
FIG. 3 is a schematic configuration diagram in which the apparatus for predicting network failure of FIG. 1 predicts a control message.
FIG. 4 is a graph comparing the traffic pattern predicted by the traffic pattern predictor of FIG. 3 and the actually generated traffic pattern.
5 is an exemplary diagram in which the traffic pattern prediction unit of FIG. 3 corrects a prediction error based on a recurrent neural network.
6 is a schematic configuration diagram in which the communication network failure prediction apparatus of FIG. 1 predicts the occurrence of a failure for each device.
7 is a flowchart illustrating a second normalization of the predicted traffic pattern value for each device by the second normalizer of FIG. 6 .
FIG. 8 is a schematic configuration diagram in which the apparatus for predicting network failure of FIG. 1 performs a failover in response to a predicted failure.
9 is a schematic flowchart of a failure prediction method according to an embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. Prior to this, the terms or words used in the present specification and claims should not be construed as being limited to their ordinary or dictionary meanings, and the inventor should properly understand the concept of the term in order to best describe his invention. Based on the principle that it can be defined, it should be interpreted as meaning and concept consistent with the technical idea of the present invention. Accordingly, the configuration shown in the embodiments and drawings described in the present specification is only the most preferred embodiment of the present invention and does not represent all of the technical idea of the present invention, so at the time of the present application, various It should be understood that there may be equivalents and variations.

1 is a schematic configuration diagram of a communication network failure prediction apparatus 100 according to an embodiment of the present invention.

The network system 150 to which the present invention is applied includes an eNB 151 (evolved NodeB) 151, an MME 152 (Mobility Management Entity) 152 constituting an Evolved Packet Core (EPC), Serving GateWay (SGW) (156) and Packet data network GateWay (PGW) (157), Equipment Identity Register (EIR) (153) connected to MME (152), Home Subscriber Server (HSS) (154) and Authentication Center (AuC) ) (155) and network equipment such as PCRF (Policy & Charging Rules Function) connected to the PGW (157).

In the network system 150, in the process of providing communication services to subscribers, each of the network devices 151 to 158 generates control data (eg, a control message) that controls the processing operation, and transmits the generated control data to other network devices. (151-158) communicates with them, and performs a processing operation according to the corresponding control data. Said generation, communication and processing of control data results in traffic of control processing. The control data corresponds to control traffic data causing control traffic. When the load of the communication service provided by the network system 150 to subscribers increases, control traffic increases in order to process the increased service. The increased control traffic may cause failure of each of the network devices 151 to 158 . When a failure occurs in the specific network equipment 151 to 158, the processing of the control traffic becomes impossible. And the inability leads to delay, inability, etc. of subscriber communication service.

The network system 150 is not particularly limited as long as a communication service is provided to subscribers through a control message between the respective network equipments 151 to 158. For example, the present invention is applied to 5G, next-generation networks, and Narrow Band-Internet of Things (NB-IoT). Hereinafter, it is assumed that the network system 150 provides an LTE service.

A process in which the network system 150 provides an LTE service to the UE 160 corresponding to the LTE terminal will be briefly described as follows. The UE 160 establishes a radio setting with the eNB 151 corresponding to the base station at the current location in order to receive the LTE service. When the UE 160 is connected to the EPC through the eNB 151 , the MME 152 processes the subscriber terminal authentication of the UE 160 through the EIR 153 . The EIR 153 is a DB for storing an International Mobile Equipment Identity (IMEI), which is a terminal unique number of the UE 160 opened by a service subscriber. The MME 152 registers the location of the UE 160 with the HSS 154 and receives a service profile. The MME 152 compares the information received from the UE 160 through the AuC 156 with the stored subscription information to authenticate as a legitimate subscriber. The SGW 156 connected to the MME 152 handles call setup management, packet data delivery, IP mobility management, and 3G network interworking. In addition, the PGW 157 connected to the SGW 156 allocates an IP address of the UE 151 and manages charge information according to data usage through the PCRF 158 .

The communication network failure prediction apparatus 100 according to an embodiment of the present invention includes a memory, a memory controller, one or more processors (CPU), a peripheral interface, an input/output (I/O) subsystem, a display device, an input device, and a communication circuit. can do.

The memory may include high-speed random access memory, and may also include one or more magnetic disk storage devices, non-volatile memories such as flash memory devices, or other non-volatile semiconductor memory devices. Access to the memory by other components, such as the processor and peripheral interfaces, may be controlled by the memory controller. The memory may store various kinds of information and program instructions, and the program is executed by the processor.

The peripheral interface connects an input/output peripheral device of the communication network failure prediction apparatus 100 with a processor and a memory. One or more processors execute various software programs and/or instruction sets stored in the memory to perform various functions for the communication network failure prediction apparatus 100 and process data. The I/O subsystem provides an interface between input/output peripherals such as display devices and input devices and the peripheral interface. The display device may use a liquid crystal display (LCD) technology or a light emitting polymer display (LED) technology.

The processor is a processor configured to perform an operation related to the communication network failure prediction apparatus 100 and perform instructions, for example, by using the instructions retrieved from the memory, input and output data between the components of the communication network failure prediction apparatus 100 . Reception and manipulation can be controlled. The communication circuit performs communication through an external port or communication by an RF signal. The communication circuitry converts electrical signals into RF signals and vice versa through which the RF signals can communicate with communication networks, other mobile gateway devices, and communication devices.

Referring to FIG. 1 , the communication network failure prediction apparatus 100 includes a collection unit 101 that collects control data from a network system 150 , classifies control messages from the collected control data, and normalizes traffic patterns of each control message. The first normalizer 102 for generating the first normalized data, the traffic pattern prediction unit 103 for predicting the traffic pattern of each control message to be generated in the future through the neural network based on the first normalized data, the predicted traffic pattern A second normalization unit 105 for generating normalized second normalized data by comparing the value of α with a threshold value, a failure occurrence prediction unit 106 for predicting a failure of a communication network through a neural network based on the second normalized data, and prediction It includes a failure prediction result processing unit 107 that processes failure recovery as a preliminary measure for the failure. These components may be implemented as software, stored in a memory, and executed by a processor, or may be implemented as a combination of software and hardware.

Here, the apparatus 100 for predicting network failure is a control message to be generated in the future based on the traffic pattern of the currently generated control message through the collection unit 101 , the first normalization unit 102 and the traffic pattern prediction unit 103 . predict traffic patterns. Then, the communication network failure prediction apparatus 100 through the second normalization unit 105, the failure occurrence prediction unit 106, and the failure prediction result processing unit 107, based on the traffic pattern of the predicted control message network failure Predict the likelihood and take proactive measures against predicted failures.

The collection unit 101 collects control data generated between each device 151 to 158 of the network system 150 at regular intervals to provide a data communication service to subscribers. The control data includes control messages (X2, S1, S5, S6a, S6a_AuC, S11, S13, Gx, etc.) interfaced between the devices 151 to 158 for processing the communication service. The collection unit 101 may store the collected control data in a DB.

The first normalizer 102 classifies a control message for each device from the control data collected in the current period, and normalizes the traffic pattern of the classified control message in a predetermined time unit. Here, the first normalizer 102 classifies the traffic pattern of the control message for each device and outputs the first normalized data obtained by first normalizing each item of the traffic pattern of the control message for each device in a predetermined time unit. The items of each traffic pattern include, for example, the number of attempts, the number of failures, the delay time, and the number of timeouts of the control message. The first normalization process is a process for generating input data for the prediction by analyzing the control message collected in the network system 150 with time and traffic patterns, and will be described in detail later with reference to FIG. 3 .

The traffic pattern prediction unit 103 predicts a traffic pattern of a control message to be generated for each device through a neural network using the first normalized data for each device that has been subjected to the first normalization process. The traffic pattern prediction unit 103 uses a recurrent neural network.

The second normalization unit 105 outputs second normalized data by performing a second normalization process on a traffic pattern of a control message predicted for each device in order to predict the possibility of a failure occurring in the future for each device. Here, the second normalization process normalizes the values of each item constituting the traffic pattern of the control message predicted for each device to a value that affects the failure of each device 151 to 158, and through FIG. It will be described later in detail.

The failure occurrence prediction unit 106 predicts the possibility of failure for each device 151 to 158 that will occur in the future through a neural network using the normalized second normalized data. The failure occurrence prediction unit 106 uses a convolutional neural network.

The failure prediction result processing unit 107 processes the result of the predicted failure probability for each equipment 151 to 158 and executes preliminary measures such as failure removal and failure recovery. If precautionary measures are taken, predicted failures are prevented in advance so that they do not actually occur.

FIG. 2 is an exemplary diagram of a control message classified by the apparatus 100 for predicting network failure of FIG. 1 .

The first normalizer 102 classifies control messages (eg, X2, S1, S5, S6a, S6a_AuC, S11, S13, Gx, etc.) from the control data collected from the network system 150 .

Here, X2 is a control message (eg, handover control) used between two adjacent eNBs 151 . S1 is a control message (eg, call setup) used between the eNB (151) and the MME (152). The S1 control message includes messages such as Attach, Service Request Mobile Originate (SRMO), SRMT, TAU, S1HO, Detach, ESRMO, and ESRMT as sub-messages. S5 is a control message used between SGW 156 and PGW 157 . The S5 control message includes messages such as Create Session Request (CSR), Modify Bearer Request (MBR), Delete Bearer Request (DBR), and Delete Session Request (DSR) as sub-messages. S6a is a control message used between the MME (152) and the HSS (154). S6a_AuC is a control message used between the MME (152) and the AuC (155). S11 is a control message used between the MME (152) and the SGW (156). In the S11 control message, there are messages such as CSR, MBR, DBR, and DSR as sub-messages. S13 is a control message used between the MME (152) and the EIR (153). Gx is a control message used between the PGW (157) and the PCRF (158).

3 is a schematic configuration diagram in which the communication network failure prediction apparatus 100 of FIG. 1 predicts a control message, and shows a control message processed by any one device.

Control data collected from the network system 150 at a predetermined cycle is generated whenever the state of the UE 160 or equipment 151 to 158 changes and is collected by the collection unit 101 in the form of a Call Detailed Record (CDR). Since the collected control data is not only large enough to accumulate several Tera bytes per day, but also the form of the data is diverse such as numbers, characters, strings, Booleans, etc., all data can be directly input into the neural network-based traffic pattern prediction unit 103. can't

Accordingly, the first normalization unit 102 classifies the control message as data necessary for failure prediction among the control data of the network system 150 collected by the collection unit 101 for each device, and the traffic of the control message classified for each device. By dividing the pattern, each item of the traffic pattern is normalized by a predetermined time unit.

Here, the traffic pattern is the number of attempts (Attempt count), the number of attempts to process the attempted control message failed (Fail Count), and the delay in the processing of the attempted control message in a unit time. It includes items of the average delay time (Delay) and the number of timeout times (Timeout count) timed out due to processing delay of the attempted control message.

In the analysis process, the first normalizer 102, for the control message classified by equipment, for each period (eg, t, 2t, 3t, ...) in a predetermined time unit, each item of the generated traffic pattern The first normalized data X ₁ to X _t are generated by counting the values of . The unit time is equal to or less than the collection period of the control data. The generated first normalized data may be expressed in the same data structure as the first normalized table 301 of FIG. 3 . The first normalization table 301 may be generated for each device, and consists of the number of occurrences of each item of the traffic pattern per unit time X ₁ to X _{t for each control message.}

When the traffic pattern of each control message generated in the current period t is analyzed as the first normalized data, the traffic pattern prediction unit 103 receives the first normalized data for each device generated by the first normalization unit 102 and receives , generate a traffic pattern of control messages predicted for each device in the next period (t+1 to 2t). The predicted value of the generated traffic pattern may be expressed in a data structure such as the prediction table 303 of FIG. 3 . _{The item values X 1} to X _t of the first normalization table 301 are values for each unit time of each item of the traffic pattern of each control message generated within the current period t, and the item value U _{1 of the prediction table 303} ~U _t is the value per unit time of each item of the traffic pattern of the control message that is predicted to occur in the next period (t+1~2t) using the _{value X 1} ~X _{t for each unit time of each item of the traffic pattern.} That is, the traffic pattern prediction unit 103 calculates the data of the network system 150 to be generated in the next period (t+1 to 2t) from the traffic pattern of the control message of the network system 150 generated in the current collection period (t). Predict the traffic pattern of control messages.

4 is a graph comparing the traffic pattern predicted by the traffic pattern prediction unit 103 of FIG. 3 and the actually generated traffic pattern. This is a graph comparing the actual values.

_{Traffic patterns U 1} to U _t of the control message predicted to be generated in the 2t period are represented by a prediction graph 401 . Thereafter, the traffic patterns X _t+1 to X _{2t that} are actually generated in the 2t period are displayed as an actual graph 403 . Then, the difference between the predicted traffic pattern U ₁ to U _t of the 2t period and the actual traffic pattern X _t+1 to X _2t of the 2t period is an error of prediction. The error of this prediction can be reduced through recurrent neural network learning.

5 is an exemplary diagram in which the traffic pattern prediction unit 103 of FIG. 3 corrects a prediction error based on a recurrent neural network.

The traffic pattern prediction unit 103 receives the traffic pattern X _T of the control message generated in the current period and outputs _{the traffic pattern U T} of the control message predicted through the recurrent neural network. Here, the predicted traffic pattern U _T is fed back to the error calculator 501 . The error calculation unit 501 calculates an error U _T - X _{T+1 for the predicted traffic pattern U T} and the input traffic pattern X _T+1 of the next period, and uses the calculated error as the traffic pattern prediction unit 103 . It is reflected in the training of the current neural network of , and the error of prediction is reduced.

6 is a schematic configuration diagram in which the communication network failure prediction apparatus 100 of FIG. 1 predicts the occurrence of a failure for each device.

_{The second normalizer 105 receives the traffic pattern U 1} ~U _t predicted for each control message of the prediction table 303 predicted by the traffic pattern prediction unit 103 for each device, and compares it with a threshold value. Normalization is performed to output second normalized data. Specifically, the second normalizer 105 compares the maximum value (the _{maximum value among U 1} to U _t ) with the threshold value for each item of the traffic pattern for each control message predicted for each device and normalizes the value of each item. do.

_{In the second normalization process, the value (U 1} ~ U _t ) of each item of the traffic pattern for each control message of the prediction table 303 is a quantitative value with a large difference in units from tens to tens of thousands for each item. When a value is predicted and calculated as it is, a large value is overreflected in the prediction and a small value is underreflected, resulting in a problem in the reliability of the prediction. For example, in a case where the distribution value of item A of the traffic pattern of the first control message is 1 to 10 and the predicted value of item A is 10 in any one device, and the distribution of item B of the traffic pattern of the second control message Assuming that is 1 to 100 and the predicted value of item B is 10, using the value 10 of item A and the value 10 of item B as it is as the calculated value of the prediction processing of the equipment will lead to an error in prediction. come. This is because item A has a maximum value of 10 in the distribution, so it should be used as a calculated value that affects much more significantly than the value of item B.

Therefore, the second normalization unit 105 performs a second normalization process on the quantitative value as a qualitative value in consideration of the mutual influence relationship between each device and the traffic pattern of the control message, so that the traffic pattern of the control message is actually determined by each network device (151 to 158). ) to reliably evaluate the degree of influence on

Therefore, the second normalizer 105 sets the maximum value U _{max of the values U 1} to U _t of each item for each item of traffic data for each control message for each equipment, and the threshold value T considering the distribution range of the traffic pattern and the equipment characteristics. and to generate second normalized data V _{ij .} Specifically, the second normalizer 105 gives three weights, for example, 1, 1.2, and 1.5, to _{the threshold value T, and assigns the maximum value U max} to each of the weighted threshold values T, 1.2T, Compared with 1.5T, the second normalized data V _ij is generated. For example, the maximum value U _max When <T, the second normalized data V _ij is 0, T<= U _max When <1.2T, the second normalized data V _ij is 1, 1.2T<= U _max When <1.5T, the second normalized data V _ij may be 2, and when 1.5T<= U _max , the second normalized data V _ij may be 3. The threshold value T and weight may be determined for each device, each control message type, and each traffic pattern item in consideration of the type of control message, distribution of each item, and device characteristics. _{The second normalized data V ij} for each item of the traffic pattern of the generated control message for each device may be expressed in the same data structure as the second normalization table 610 of FIG. 6 . The second normalization table 610 is obtained by second normalizing the prediction table 303 in consideration of the distribution value/threshold value of the traffic pattern and the characteristics of the equipment/control message.

The failure occurrence prediction unit 106 predicts a failure occurrence probability in the next cycle of each device through a convolutional neural network using the second normalized data of the second normalization table 610 . The failure occurrence prediction unit 106 may calculate the failure occurrence probability for each equipment.

7 is a flowchart for second normalizing the value of each item of the traffic pattern of the predicted control message for each device by the second normalizer 105 for each device. 7 is a flow chart for normalizing the values of any of a number of entries in the traffic patterns of any item of equipment, a second normalized data by normalizing the number of attempts of the entries in the traffic pattern of the control message 1 of PGW1 of Figure 6 (V ₁₁ ) to create a method.

Referring to FIG. 7 , in the second normalization process, the second normalization unit 105 _{initializes the second normalization data V 11} of the second normalization table 610 ( S701 ).

First, the second normalizer 105 obtains the maximum value U _{max from among the values of the number of attempts U 1} to U _t among the traffic pattern items of the control message 1 of the PGW1 corresponding to the input value. The reason to obtain the maximum value U _max is a maximum in order to predict the possibility of failure that are affected by the maximum value U _max of the traffic pattern to be generated during the period the art.

Next, the second normalizer 105 compares the maximum value U _max and the threshold value T and maintains the initial value "0" if it is less than the threshold value (S703). If the maximum value U _max is equal to or greater than the threshold value T and is less than 1.2 times the threshold value T, the second normalization unit 105 _{assigns the second normalized data V 11} to 1 ( S705 , S707 ). . In addition, if the maximum value U _max is 1.2 times or more of the threshold value T and less than 1.5 times the threshold value T, "2" is assigned, and if it is 1.5 times or more of the threshold value T, "3" is assigned (S709 to S713). Then, the number of attempts of the control message 1 predicted in the period of the PGW1 is generated as a single qualitative value.

FIG. 8 is a schematic configuration diagram in which the apparatus for predicting network failure of FIG. 1 performs a failover in response to a predicted failure.

Referring to FIG. 8 , the failure occurrence prediction unit 106 predicts the failure probability of each network equipment 151 to 158 through a convolutional neural network using the value of the second normalization table 610 . In the convolutional neural network, the value of the traffic pattern for each device of the second normalization table 610 considers the correlation between each device and the control messages to predict the occurrence of a failure. The failure occurrence prediction unit 106 generates failure prediction information including the equipment information, the failure code, and the failure occurrence probability (P) in which the failure occurrence is predicted.

When the occurrence of a failure is predicted through the convolutional neural network, the failure prediction processing unit 107 inquires the failure knowledge DB 810 shown in FIG. 8 with the generated failure code, and executes the recovery command with reference to the inquired failure knowledge information Auto-recovery, such as , or manual recovery, such as administrator notification, is implemented as a precautionary measure.

Here, the failure knowledge DB 810 shown in FIG. 8 includes data items such as failure codes, automatic or manual recovery, minimum probability of automatic recovery (P _MA : Minimum Probability for Auto recovery), and recovery commands by equipment type. includes The failure prediction processing unit 107 inquires the information of the failure knowledge DB 810 using the equipment type and failure code predicted by the failure occurrence prediction unit 106 as keys, and executes recovery processing according to the inquired information. Manual recovery is a case where the _{failure probability P i for each equipment is less than P MA,} and the failure details are notified to the operator for manual action. The recovery command by automatic recovery is executed by transmitting the recovery command to the Network Management System (NMS) when the _{failure probability P i for each device} is greater than or equal to P _MA. That is, the occurrence of a failure predicted by prior measures such as manual recovery and automatic recovery is blocked in advance, so that a failure does not actually occur.

9 is a schematic flowchart of a failure prediction method according to an embodiment of the present invention.

Referring to FIG. 9 , the apparatus 100 for predicting network failure collects control data processed between the respective network devices 151 to 158 from the network system 150 for one period t ( S901 ).

The communication network failure prediction apparatus 100 classifies a control message for each device from the collected control data, classifies the traffic pattern of the control message classified for each device, and normalizes each item of the traffic pattern by a predetermined time unit to perform the first normalization process Data is generated (S902). Here, the traffic pattern is the number of attempts at which the control message is attempted (Attempt count), the number of times the attempted control message fails to be processed (Fail Count), and the average delay time at which the attempted control message is delayed (Delay) and an item of a timeout count timed out because the attempted control message is delayed in processing, and each item is counted in a predetermined time unit. The time unit is the same as or smaller than the time t. The first normalized data may be expressed in the same data structure as the first normalized table 301 of FIG. 3 . The first normalization table 301 is composed of the number of occurrences of each item of the traffic pattern per unit time for each control message X ₁ to X _t.

The communication network failure prediction apparatus 100 predicts a traffic pattern of a control message to be generated for each device in the future _{by using the first normalized data X 1} to X _{t for each device as input data of the neural network (S903).} That is, the traffic pattern of the control message collected in period t is predicted as the traffic pattern of the control message in the next period t+1 to 2t through recurrent neural network processing. Traffic patterns U ₁ to U _t of the predicted control message may be formed in the structure of the prediction table 303 of FIG. 3 . Here, the communication network failure prediction apparatus 100 _{calculates the difference between the predicted traffic patterns U 1} to U _{t and X t+1} to X _2t actually generated in the period of 2t as an error, and feeds back the calculated error to the recurrent It is reflected in the learning process of the neural network (S904).

When the traffic pattern of the control message for the next period of time t+1 to 2t is predicted for each device, the communication network failure prediction apparatus 100 for each device, for each item of traffic data of the control message, for reliable failure prediction, _{The second normalized data V ij} is generated by comparing the maximum value U _max of the values U ₁ to U _t of each item with a threshold value T in consideration of the distribution range of the traffic pattern and the equipment characteristics ( S905 ). Specifically, the communication network failure prediction apparatus 100 gives three weights, for example, 1, 1.2, 1.5, to _{the threshold value T, and assigns the maximum value U max} to each of the weighted threshold values T, 1.2T, Compared with 1.5T, the second normalized data V _ij is generated. _{The second normalized data V 11} to V _ij generated by the second normalization process may form the structure of the second normalization table 610 of FIG. 6 . _{When the second normalized data V ij} for each item of the traffic pattern is generated for each control message for each device, the network failure prediction apparatus 100 sets all the values (all [V _ij ]) to 0, each of the network devices 151 to 158 ), it is determined that there is no possibility of a failure occurring in the next period, so the subsequent failure prediction process is stopped and the process returns to the process of collecting data for the next period (t+1 to 2t) (S906). In this way, when all _Vij is 0, the load of the system can be greatly reduced by stopping the subsequent processing and moving on to the data collection step S901 of the next cycle.

If there is a non-zero value among _{V ij , the communication network failure prediction apparatus 100 inputs the second normalized data V ij} to the convolutional neural network, and the failure occurrence is predicted in the next period (t+1 to 2t). Prediction of failure information such as type, failure code and probability (P _{i ) (S907).}

If the failure information is not predicted, the communication network failure prediction apparatus 100 returns to the step (S901) of collecting control data of the next period (t+1 to 2t) (S908).

When the failure information is predicted, the communication network failure prediction apparatus 100 inquires the failure recovery information from the failure knowledge DB 810 using the predicted failure information as a key (S909). When the failure recovery information of manual recovery is inquired ( S910 ), the communication network failure prediction apparatus 100 notifies the operator of the predicted failure information for manual recovery ( S911 ). If automatic recovery is inquired, if the failure probability P _{i for each device} is less than the minimum failure probability P _MA , the predicted failure information is notified to the operator in order to prepare for manual recovery (S911). If P _i is greater than or equal to P _MA , the apparatus 100 for predicting network failure transmits an automatic recovery command to the network management system. After taking the failover in this way, the communication network failure prediction apparatus 100 returns to the step (S901) of collecting control data of the next period (t+1 to 2t).

Although the present invention has been described with reference to limited embodiments and drawings, the present invention is not limited thereto, and the technical idea of the present invention and claims to be described below by those skilled in the art to which the present invention pertains Of course, various modifications and variations are possible within the equivalent range of the range.

100: system 101: collection unit
102: first normalization unit 103: traffic pattern prediction unit
105: second normalization unit 106: failure occurrence prediction unit
107: failure prediction result processing unit 150: network system

Claims (22)

In the neural network-based communication network failure prediction device,
a collection module for collecting control data of the communication network at regular intervals;
a first normalization module for classifying control messages from the control data collected for each period and generating first normalized data obtained by normalizing a traffic pattern of each control message to a predetermined time unit;
a traffic pattern prediction module for predicting a traffic pattern of a next period of each control message through a neural network based on the first normalized data;
a second normalization module for generating second normalized data normalized by comparing the value of each item constituting the predicted next cycle traffic pattern of each control message with a threshold value; and
and a failure prediction module for predicting the probability of occurrence of a failure of the communication network through a neural network based on the second normalized data,
The threshold value is determined for each item in consideration of the distribution of the values of each item. The method of claim 1,
The first normalization module,
A communication network failure prediction apparatus, characterized in that for each predetermined time unit, the number of occurrences for each item constituting the traffic pattern is counted and normalized. The method of claim 1,
The second normalization module,
Network failure prediction apparatus, characterized in that for each item constituting the predicted traffic pattern of each control message, the maximum value and the threshold value are compared and normalized. 4. The method of claim 3,
The second normalization module,
The apparatus for predicting network failure, characterized in that different weights are applied to the threshold value and the value of each item is normalized by comparing it with the maximum value. 5. The method of claim 4,
The second normalization module,
The apparatus for predicting network failure, characterized in that the value of each item is determined as one of values corresponding to a number obtained by adding 1 to the number of weights. 5. The method of claim 4,
The second normalization module,
The apparatus for predicting network failure, characterized in that the control messages are classified for each device in the communication network, and values of each item constituting the predicted traffic pattern of each classified control message are normalized. 7. The method of claim 6,
The second normalization module,
Network failure prediction apparatus, characterized in that applying the threshold value proportional to the value of the traffic pattern of each device. 7. The method of claim 6,
The failure occurrence prediction module,
A communication network failure prediction device, characterized in that predicting the possibility of failure by equipment. 9. The method of claim 8,
The failure occurrence prediction module,
Communication network failure prediction device, characterized in that for calculating the failure occurrence probability of the equipment with the possibility of failure. The method of claim 1,
The failure prediction result processing module further comprises a failure prediction result processing module for determining the recovery possibility of the failure predicted by the failure occurrence prediction module by querying the failure knowledge database, and processing the failure action of the operator notification or automatic recovery according to the determination result Network failure prediction device. The method of claim 1,
The traffic pattern prediction module,
Based on the recurrent neural network,
The failure occurrence prediction module,
A communication network failure prediction device, characterized in that based on a convolutional neural network. In the communication network failure prediction method executed by the neural network-based communication network failure prediction apparatus,
collecting control data of the communication network at regular intervals;
classifying control messages from the control data collected for each period and generating first normalized data obtained by normalizing a traffic pattern of each control message to a predetermined time unit;
predicting a traffic pattern of a next period of each control message through a neural network based on the first normalized data;
generating second normalized data normalized by comparing the value of each item constituting the predicted next cycle traffic pattern of each control message with a threshold value; and
Predicting the probability of occurrence of a failure of the communication network through a neural network based on the second normalized data,
The threshold value is a communication network failure prediction method that is determined for each item in consideration of a distribution of values of the respective items. 13. The method of claim 12,
The step of generating the first normalized data comprises:
The method for predicting network failure, characterized in that the counting and normalizing the number of occurrences for each item constituting the traffic pattern for each predetermined time unit. 13. The method of claim 12,
The step of generating the second normalized data comprises:
The method for predicting a network failure, characterized in that for each item constituting the predicted traffic pattern of each control message, comparing and normalizing the maximum value with the threshold value. 15. The method of claim 14,
The step of generating the second normalized data comprises:
and applying a different weight to the threshold value and normalizing the value of each item by comparing it with the maximum value. 16. The method of claim 15,
The step of generating the second normalized data comprises:
and determining the value of each item as one of values corresponding to a number obtained by adding 1 to the number of weights. 16. The method of claim 15,
The step of generating the second normalized data comprises:
Classifying control messages for each device in the communication network, and normalizing the values of each item constituting the predicted traffic pattern of each classified control message. 18. The method of claim 17,
The step of generating the second normalized data comprises:
A method for predicting network failure, characterized in that applying the threshold value proportional to the value of the traffic pattern of each device. 18. The method of claim 17,
The step of predicting the probability of occurrence of the failure,
A communication network failure prediction method, characterized in that predicting the probability of failure by equipment. 20. The method of claim 19,
The step of predicting the probability of occurrence of the failure,
Network failure prediction method, characterized in that calculating the failure probability of the equipment with the possibility of failure. 13. The method of claim 12,
After predicting the probability of occurrence of the failure,
The method for predicting a network failure, further comprising: querying the failure knowledge database to determine the recovery possibility of the predicted failure, and processing an operator notification or automatic recovery failover according to the determination result. 13. The method of claim 12,
Predicting the traffic pattern comprises:
Based on the recurrent neural network,
The step of predicting the probability of occurrence of the failure,
A communication network failure prediction method, characterized in that based on a convolutional neural network.

Images