Fault prediction device, wireless communication system, and fault prediction program

Abstract

To improve the accuracy of predicting failures in wireless communication systems. [Solution] The fault prediction device comprises a correlation calculation unit and a prediction unit. The correlation calculation unit generates first correlation coefficient data representing the similarity between first communication status data representing the communication status of a first cell provided by a first base station and second communication status data representing the communication status of a second cell provided by a second base station. Alternatively, the correlation calculation unit generates second correlation coefficient data representing the similarity between the first communication status data and third communication status data representing the communication status within the communication area of ​​the first base station. The prediction unit determines the state of the first base station or the first cell based on at least one of the first correlation coefficient data or the second correlation coefficient data.

Description

Technical Field

[0001] The present invention relates to an apparatus, a method, and a program for predicting failures occurring in a wireless communication system.

Background Art

[0002] The fifth-generation mobile communication system (5G) has become widely used, and traffic has been rapidly increasing. However, in a wireless communication system with high traffic, if a failure occurs in network equipment such as a base station device, the damage is likely to be significant. Therefore, a technology for monitoring the operating status of a wireless communication system and predicting failures is required. For example, even if the wireless communication system appears to be operating normally, a method for detecting a silent failure such as a situation where some traffic processing cannot be performed is required. A silent failure does not immediately affect the service quality, but means a situation where the perceived quality of the end user deteriorates if left unattended.

[0003] Note that Patent Document 1 describes a method for monitoring virtual network functions. Patent Document 2 describes a method for monitoring a service using flow data collected from a plurality of network devices. Patent Document 3 describes a method for accurately detecting a failure of a network device.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Patent Document 3

Summary of the Invention

Problems to be Solved by the Invention

[0005] As mentioned above, technology for predicting failures in wireless communication systems is considered important. However, conventional technologies do not always provide sufficient prediction accuracy. Therefore, there is a need to improve the accuracy of predicting failures in wireless communication systems.

[0006] One aspect of the present invention is to improve the accuracy of predicting failures in wireless communication systems. [Means for solving the problem]

[0007] A fault prediction device according to one aspect of the present invention includes a correlation calculation unit that generates at least one of first correlation coefficient data representing the similarity between first communication status data representing the communication status of a first cell provided by a first base station and second communication status data representing the similarity between the first communication status data and third communication status data representing the communication status within the communication area of ​​the first base station, or second correlation coefficient data representing the similarity between the first communication status data and third communication status data representing the communication status within the communication area of ​​the first base station; and a prediction unit that determines the state of the first base station or the first cell based on at least one of the first correlation coefficient data or the second correlation coefficient data. [Effects of the Invention]

[0008] According to the above-described embodiment, the accuracy of predicting failures in wireless communication systems is improved. [Brief explanation of the drawing]

[0009] [Figure 1] This figure shows an example of a wireless communication system according to an embodiment of the present invention. [Figure 2] This figure shows an example of a fault prediction device according to an embodiment of the present invention. [Figure 3] This figure shows an example configuration of a base station that provides cells. [Figure 4] This figure shows another example configuration of a base station that provides cells. [Figure 5] This diagram shows an example of how to select a reference cell for a target cell. [Figure 6] It is a diagram showing an example of communication state data of a target cell and a reference cell. [Figure 7] It is a diagram showing an example of a monitoring period for monitoring the similarity between the communication state of a target cell and the communication state of a reference cell. [Figure 8] It is a diagram showing an example of a method for calculating the correlation coefficient between the communication state of a target cell and the communication state of a reference cell. [Figure 9] It is a diagram showing an example of the correlation coefficient calculated for each monitoring period. [Figure 10] It is a diagram for explaining the traffic within the communication area of a base station. [Figure 11] It is a diagram showing an example of the traffic volume calculated for a target base station. [Figure 12] It is a diagram showing an example of the operation of an autocorrelation calculation unit. [Figure 13] It is a diagram (part 1) showing an example of a process for generating feature amount data from correlation coefficient data. [Figure 14] It is a diagram (part 2) showing an example of a process for generating feature amount data from correlation coefficient data. [Figure 15] It is a diagram (part 3) showing an example of a process for generating feature amount data from correlation coefficient data. [Figure 16] It is a diagram showing an example of the learning of an AI model. [Figure 17] It is a diagram showing an example of a method for predicting a failure of a base station using a learned AI model. [Figure 18] It is a diagram showing an example of another method for predicting a failure of a base station. [Figure 19] It is a flowchart showing an example of the processing of a failure prediction device. [Figure 20] It is a diagram showing an example of the hardware configuration of a failure prediction device.

Embodiments for Carrying Out the Invention

[0010] In standardization organizations such as 3GPP (Registered Trademark) (3rd Generation Partnership Project) or O-RAN (Open RAN) Alliance, the standardization of the radio access network (RAN) is underway. Also, O-RAN provides an O-RAN architecture that introduces a RIC (RAN Intelligent Controller) based on 3GPP specifications.

[0011] FIG. 1 shows an example of a wireless communication system according to an embodiment of the present invention. The wireless communication system 1 according to the embodiment of the present invention is configured, for example, based on the O-RAN architecture. In this case, the wireless communication system 1 includes a RIC system and a base station system.

[0012] The RIC system provides services to each device or each function within the O-RAN architecture. Also, the RIC system includes a Non Real Time RIC (Non Real Time RIC) 10 and a Near Real Time RIC (Near Real Time RIC) 20. The Non Real Time RIC 10 is implemented within the SMO (Service Management Orchestration), although not particularly shown. The Near Real Time RIC 20 is provided outside the SMO in many cases. The Non Real Time RIC 10 and the Near Real Time RIC 20 are connected by an A1 interface. The A1 interface includes A1-P and A1-EI.

[0013] The base station system comprises one or more E2 nodes 31 and multiple radio units (RUs) 32. The E2 node 31 includes a central unit (CU) and a distributed unit (DU). The E2 node 31 provides radio link control, media access control, and PHY-High functionality, and processes signals from the RUs 32 in the upper layer. Multiple RUs 32 may be connected to each DU. The E2 node 31 can also acquire configuration and statistical information of the radio access network and provide it to the RIC 10 in a non-real-time manner.

[0014] The RU32 is equipped with a wireless circuit and can accommodate multiple User Equipment (UEs). The RU32 connects to the non-real-time RIC10 via the O1 interface or an Open fronthaul interface (not shown). The RU32 also performs wireless communication according to instructions provided by the E2 node 31. Each RU32 can provide one or more wireless cells.

[0015] The non-real-time RIC10 can periodically acquire various data from the E2 node 31 via the O1 interface. Specifically, the non-real-time RIC10 collects PM (Performance Management) data and CM (Configuration Management) data. The non-real-time RIC10 also collects location data representing the location of each UE. Furthermore, the non-real-time RIC10 may acquire FM (Fault Management) data and TM (Trace Management) data.

[0016] Various application programs can be implemented in the non-real-time RIC10. For example, the non-real-time RIC10 includes a wireless network optimization unit 11 that uses an AI / ML model to determine optimal parameters according to the wireless environment and traffic demand, and provides them to the E2 node 31 via the O1 interface. The non-real-time RIC10 also generates policies related to the control of the wireless access network and notifies the near-real-time RIC20 via the A1 interface. In the following description, application programs operating in the non-real-time RIC10 may be referred to as "rApp". The wireless network optimization unit 11 is implemented as an rApp.

[0017] The non-real-time RIC10 includes a database (DB) 12. The database 12 stores PM data and CM data acquired by the non-real-time RIC10 from the E2 node 31.

[0018] The near real-time RIC20 collects and analyzes configuration and statistical information of the wireless access network from the E2 node 31 via the E2 interface. The near real-time RIC20 also controls the E2 node 31 according to policies communicated by the non-real-time RIC10. In the following description, application programs running on the near real-time RIC20 may be referred to as "xApps".

[0019] The A1 interface connects the non-real-time RIC10 and the near-real-time RIC20. The E2 interface connects the near-real-time RIC20 and the E2 node 31.

[0020] Figure 2 shows an example of a fault prediction device according to an embodiment of the present invention. The fault prediction device 40 according to an embodiment of the present invention comprises a data acquisition unit 50, a feature calculation unit 60, and a fault prediction unit 70. The fault prediction device 40 may further include other functions not shown in Figure 2. The fault prediction device 40 is also implemented, for example, by an rApp implemented in the non-real-time RIC 10 shown in Figure 1. In this case, the functions of the fault prediction device 40 may be part of the wireless network optimization unit 11.

[0021] The data collection unit 50 periodically collects communication status data representing the communication status of each cell provided by each base station. The communication status data is not particularly limited, but is collected at intervals of, for example, 5 to 15 minutes. In the configuration shown in Figure 1, the communication status data collected by the data collection unit 50 is stored in the database 12.

[0022] Cells are provided by a base station, which consists of an E2 node (CU / DU) 31 and RU 32. For example, in the case shown in Figure 3A, cell a is provided by a base station consisting of an E2 node 31a and RU 32a. Cell b is provided by a base station consisting of an E2 node 31b and RU 32b. However, as shown in Figure 3B, an E2 node 31 can accommodate multiple RU 32s. In this case, cell a is provided by a base station consisting of an E2 node 31 and RU 32a, and cell b is provided by a base station consisting of an E2 node 31 and RU 32b.

[0023] In the example shown in Figure 3, one RU32 forms one cell, but one RU32 may form multiple cells. For example, in the case shown in Figure 4, the base station consisting of an E2 node 31 and an RU32 provides three sectors (sectors 1 to 3). Here, the RU32 is equipped with three RU modules (RU1 to RU3) and processes communication for each sector. In this case, each "sector" corresponds to a "cell". Furthermore, sector 1 is provided by the base station consisting of the E2 node 31 and RU module 1, sector 2 is provided by the base station consisting of the E2 node 31 and RU module 2, and sector 3 is provided by the base station consisting of the E2 node 31 and RU module 3.

[0024] The communication status data collected by the data collection unit 50 includes the PM data described above. The PM data represents the traffic volume and the number of active users for each cell. The traffic volume of a cell is represented, for example, by the utilization rate of resource blocks. The number of active users represents the number of UE2s currently communicating.

[0025] Communication status data may include the CM data described above. CM data includes information representing the frequency band used within the cell, information representing the location where each RU32 is installed (e.g., latitude and longitude), and information representing the height of each RU32's antenna. However, since CM data does not change dynamically, it does not need to be collected periodically.

[0026] The communication status data may include location data representing the location of UE2. The location data may be GPS (Global Positioning System) data. The data collection unit 50 does not need to collect location data for all UE2s. For example, the data collection unit 50 collects location data for UE2s that have permitted the provision of location data.

[0027] The feature calculation unit 60 calculates feature quantities related to the communication status of each base station or cell based on the communication status data collected by the data acquisition unit 50. The feature calculation unit 60 includes a correlation calculation unit 61 and a conversion unit 62.

[0028] The correlation calculation unit 61 comprises a cell / cell correlation calculation unit 61a, a cell / area correlation calculation unit 61b, and an autocorrelation calculation unit 61c. The cell / cell correlation calculation unit 61a calculates the correlation between the communication status of a target cell and the communication status of a reference cell. The "target cell" refers to the cell whose failure status should be predicted. The "reference cell" is one of the cells located around the target cell and is selected by a method described later. The cell / area correlation calculation unit 61b calculates the correlation between the communication status of the target base station providing the target cell and the communication status within the communication area of ​​that target base station. The autocorrelation calculation unit 61c calculates the autocorrelation of the communication status of the target cell at different time periods.

[0029] In this specification, "correlation" may be read as "similarity." Furthermore, the correlation calculation unit 61 does not need to include all of the cell / cell correlation calculation unit 61a, cell / area correlation calculation unit 61b, and autocorrelation calculation unit 61c. That is, the correlation calculation unit 61 only needs to include one or more of the cell / cell correlation calculation unit 61a, cell / area correlation calculation unit 61b, and autocorrelation calculation unit 61c. However, it is preferable that the correlation calculation unit 61 includes at least one of the cell / cell correlation calculation unit 61a or the cell / area correlation calculation unit 61b.

[0030] The conversion unit 62 converts the correlation coefficient calculated by the correlation calculation unit 61 into a predetermined feature quantity. The feature quantity represents, for example, the magnitude of the change in the correlation coefficient calculated by the correlation calculation unit 61. Alternatively, the feature quantity may represent the slope of the change in the correlation coefficient calculated by the correlation calculation unit 61. In other words, the feature quantity obtained by the conversion unit 62 represents the characteristics of the communication state of the target base station or target cell.

[0031] The failure prediction unit 70 determines the communication status of the target base station or target cell based on the features calculated by the feature calculation unit 60. For example, the failure prediction unit 70 predicts whether or not a failure will occur at the target base station. In this case, the failure prediction unit 70 may use a pre-created AI model to predict whether or not the base station will fail.

[0032] Figure 5 shows an example of how the cell / cell correlation calculation unit 61a selects a reference cell for a target cell. In this embodiment, the reference cell is selected based on the traffic of each cell.

[0033] The cell / cell correlation calculation unit 61a first extracts a number of surrounding cells located near the target cell. The position of each cell is represented, for example, by the latitude and longitude of the location where the RU32 forming the cell is located. In cases where a cell is formed in a specific direction relative to the RU32, the position of the cell is determined based on the position of the RU32 and the direction in which the cell is formed. In Figure 5, surrounding cells C1, C2, ... are extracted for the target cell.

[0034] The cell / cell correlation calculation unit 61a acquires traffic data representing the traffic volume of the target cell and each surrounding cell C1, C2, ... At this time, the cell / cell correlation calculation unit 61a may acquire traffic data for each cell according to the period in which the traffic volume fluctuates. For example, traffic data for one week for each cell is acquired. Then, the cell / cell correlation calculation unit 61a calculates the correlation coefficient between the traffic data of the target cell and the traffic data of each surrounding cell C1, C2, ... As a result, the surrounding cell with the highest similarity to the target cell is selected as the "reference cell". In the example shown in Figure 5, surrounding cell C1 is selected as the reference cell.

[0035] In the example shown in Figure 5, the reference cell is selected based on the amount of traffic, but the reference cell may also be selected using other communication conditions. For example, the cell / cell correlation calculation unit 61a may select the reference cell based on the similarity of the number of UE2s located within the cell.

[0036] In this way, a reference cell is selected from surrounding cells located near the target cell. Here, the communication state of the reference cell is similar to that of the target cell. In this embodiment of the present invention, it is assumed that if the base station providing the target cell is functioning normally, a high correlation coefficient between the communication state of the target cell and the communication state of the reference cell will be maintained. Therefore, the cell / cell correlation calculation unit 61a continuously monitors the correlation coefficient, which represents the similarity between the communication state of the target cell and the communication state of the reference cell.

[0037] Figure 6 shows an example of communication status data for the target cell and the reference cell. In this embodiment, traffic volume is used as the communication status data. In this embodiment, traffic volume is calculated on an hourly basis. The calculation results are stored, for example, in the database 12 shown in Figure 1.

[0038] Figure 7 shows an example of a monitoring period for monitoring the similarity between the communication status of a target cell and the communication status of a reference cell. The length of the monitoring period is not particularly limited, but for example, it is 24 hours. In addition, multiple monitoring periods are set with a predetermined time difference between them.

[0039] In the example shown in Figure 7, monitoring period (1) is set from time T1 to T11. The difference between time T1 and time T11 is 24 hours. Monitoring period (2) is set from time T2 to T12. Here, the time period in which monitoring period (2) is set is shifted by, for example, one hour relative to monitoring period (1). Subsequent monitoring periods are set in a similar manner.

[0040] The cell / cell correlation calculation unit 61a calculates a correlation coefficient representing the similarity between the communication state of the target cell and the communication state of the reference cell during each monitoring period. Specifically, the cell / cell correlation calculation unit 61a calculates a correlation coefficient representing the similarity between the fluctuation pattern of the communication state of the target cell and the fluctuation pattern of the communication state of the reference cell during each monitoring period. In this embodiment, the length of the monitoring period is 24 hours. Also, as shown in Figure 6, the traffic volume of the target cell and the reference cell is calculated in 1-hour increments. Therefore, each monitoring period includes 24 sampling points. This allows for obtaining the fluctuation pattern of the traffic volume for each monitoring period.

[0041] Figure 8 shows an example of a method for calculating the correlation coefficient between the communication status of the target cell and the communication status of the reference cell. Figure 8A shows the case where the target cell is normal, and Figure 8B shows the case where an abnormality occurs in the target cell.

[0042] In the case shown in Figure 8A, the correlation between the communication status of the target cell and the communication status of the reference cell is calculated during the monitoring period (1). The cell / cell correlation calculation unit 61a plots the sampling pair values, which represent the traffic volume of the target cell and the traffic volume of the reference cell obtained at each of the 24 sampling points within the monitoring period (1), on the XY plane. At this time, the traffic volume of the target cell may be normalized with respect to the traffic volume of the reference cell. In this example, the X axis represents the traffic volume of the reference cell, and the Y axis represents the traffic volume of the target cell. Also, the "○" marks in Figure 8 represent pairs of traffic volume of the target cell and traffic volume of the reference cell obtained at one sampling point.

[0043] In cases where the target cell is functioning normally, a high correlation coefficient between the communication status of the target cell and the communication status of the reference cell is expected to be maintained. For example, if the traffic of the reference cell increases, the traffic of the target cell will also increase. In this case, as shown in Figure 8A, a high correlation coefficient is obtained between the traffic volume of the target cell and the traffic volume of the reference cell. That is, the correlation coefficient approaches 1.

[0044] In the case shown in Figure 8B, even if the traffic volume of the reference cell increases, the traffic volume of the target cell does not increase significantly. In this case, a high correlation coefficient cannot be obtained between the traffic volume of the target cell and the traffic volume of the reference cell. In other words, the correlation coefficient approaches zero.

[0045] The cell / cell correlation calculation unit 61a calculates a correlation coefficient between the traffic volume of the target cell and the traffic volume of the reference cell based on the distribution of 24 sampled pair values ​​for each monitoring period. The method for calculating the correlation coefficient between the X component and the Y component based on the distribution of multiple values ​​plotted on the XY plane is implemented using known techniques. The cell / cell correlation calculation unit 61a then stores the correlation coefficients calculated for each monitoring period in a time series, as shown in Figure 9.

[0046] In the embodiments shown in Figures 5 to 9, a correlation coefficient for traffic volume is calculated between the target cell and the reference cell, but correlation coefficients related to other communication states may also be calculated. For example, the cell / cell correlation calculation unit 61a may calculate the correlation coefficient for traffic volume and the correlation coefficient for the number of UE2s between the target cell and the reference cell for each monitoring period.

[0047] Next, the operation of the cell / area correlation calculation unit 61b will be described. The cell / area correlation calculation unit 61b calculates the correlation between the communication status of the target base station that provides the target cell and the communication status within the communication area of ​​that target base station.

[0048] Figure 10 is a diagram illustrating traffic within the communication area of ​​a base station. The cell / area correlation calculation unit 61b divides the communication area covered by the telecommunications carrier into multiple mesh blocks, for example, using map data. The shape of the mesh block is not particularly limited, but for example, it is a square. In this case, the size of the mesh block is, for example, 100 meters x 100 meters. In the following description, the base station that provides the target cell may be referred to as the "target base station." Also, the base stations that provide each surrounding cell may be referred to as "surrounding base stations."

[0049] The cell / area correlation calculation unit 61b identifies the communication area of ​​the target base station 81. First, the cell / area correlation calculation unit 61b refers to the PM data and the location data of each UE2 to detect the mesh block in which a UE2 connected to the target base station 81 existed. In the embodiment shown in Figure 10, the target base station 81 has been connected to by UE2s located within each mesh block m1 to m12. In this case, the area consisting of mesh blocks m1 to m12 is considered to be the communication area of ​​the target base station 81.

[0050] The cell / area correlation calculation unit 61b calculates the traffic volume of the target base station 81 by referring to PM data. The cell / area correlation calculation unit 61b also calculates the total traffic volume within the communication area of ​​the target base station 81 by referring to the PM data and the location data of each UE2. That is, the sum of the traffic volumes of UE2s located within mesh blocks m1 to m12 is calculated. However, in many cases, multiple cells provided by a wireless communication system partially overlap. Therefore, while most of the numerous UE2s located within mesh blocks m1 to m12 are connected to the target base station 81, some UE2s may be connected to other base stations (i.e., surrounding base stations). Thus, the cell / area correlation calculation unit 61b identifies the UE2s located within mesh blocks m1 to m12 and calculates the sum of the traffic volumes of the identified UE2s by referring to the PM data of the target cell and the PM data of surrounding cells. This calculates the total traffic volume within the communication area of ​​the target base station 81.

[0051] In this embodiment, some of the multiple UE2s located within mesh blocks m1 to m12 are connected to surrounding base stations 82 and 83. In this case, the sum of the traffic volume of each UE2 connected to the target base station 81, the traffic volume of each UE2 located within mesh blocks m1 to m12 and connected to surrounding base station 82, and the traffic volume of each UE2 located within mesh blocks m1 to m12 and connected to surrounding base station 83 is calculated.

[0052] Figure 11 shows an example of the traffic volume calculated for the target base station. In this example, the traffic volume for the target base station 81 and the total traffic volume within the communication area of ​​the target base station 81 are calculated every hour.

[0053] Here, the traffic of base station 81 is a portion of the total traffic within the communication area of ​​base station 81. Therefore, if the total traffic within the communication area of ​​base station 81 increases, the traffic volume of base station 81 will also increase, and if the total traffic within the communication area of ​​base station 81 decreases, the traffic volume of base station 81 will also decrease. In other words, when base station 81 is operating normally, the correlation coefficient between the traffic volume of base station 81 and the total traffic volume within the communication area of ​​base station 81 should be large.

[0054] Therefore, the cell / area correlation calculation unit 61b monitors a correlation coefficient that represents the similarity between the traffic volume of the target base station 81 and the total traffic volume within the communication area of ​​the target base station 81. The correlation coefficient is calculated, for example, using the method described with reference to Figures 7 and 8. At this time, the cell / area correlation calculation unit 61b calculates the correlation coefficient for each monitoring period. The cell / area correlation calculation unit 61b then stores the correlation coefficients calculated for each monitoring period in a time series.

[0055] Figure 12 shows an example of the operation of the autocorrelation calculation unit 61c. The autocorrelation calculation unit 61c calculates the autocorrelation coefficient of traffic data for each cell in order to monitor whether the behavior of traffic with a predetermined trend follows that trend. In this embodiment, attention is paid to the fact that weekday traffic and holiday traffic each have their own unique trends. Holidays include Saturdays, Sundays, and public holidays. In the following description, the autocorrelation coefficient of weekday traffic is calculated.

[0056] In this embodiment, traffic data for the target cell for the period from March 1st to March 11th is stored. The autocorrelation calculation unit 61c extracts traffic data for the target cell on weekdays (in this case, March 1st, March 4th to March 8th, and March 11th). Then, it sets a monitoring period for the extracted traffic data (hereinafter referred to as weekday traffic data). The monitoring period is 24 hours, as in the embodiment shown in Figure 7. Note that the weekday traffic data is an example of specific communication state data that satisfies predetermined conditions and is extracted from the traffic data of the target cell.

[0057] The autocorrelation calculation unit 61c sets a reference period for the monitoring period. The reference period is set immediately before the monitoring period in the weekday traffic data. Also, since there are 5 weekdays in a week, the length of the reference period is 5 × 24 hours. In this case, the reference period consists of 5 sub-reference periods R1 to R5. The length of each sub-reference period R1 to R5 is 24 hours. Then, the autocorrelation calculation unit 61c creates the average traffic data for the reference period by averaging the traffic data of each sub-reference period R1 to R5.

[0058] The autocorrelation calculation unit 61c calculates the correlation coefficient between the traffic data for the monitoring period and the average traffic data for the reference period. In other words, it obtains the autocorrelation coefficient for the traffic data of the target cell. Similarly, the autocorrelation calculation unit 61c calculates the autocorrelation coefficient for each monitoring period, shifting the monitoring period by one hour at a time. The autocorrelation calculation unit 61c then stores the autocorrelation coefficients calculated for each monitoring period in a time series.

[0059] Furthermore, when the target base station is operating normally, the traffic data during the monitoring period is expected to be similar to the average traffic data during the reference period. In other words, if the target base station is functioning normally, the autocorrelation coefficient between the traffic data during the monitoring period and the average traffic data during the reference period is expected to be large. Conversely, if a failure occurs at the target base station, the autocorrelation coefficient between the traffic data during the monitoring period and the average traffic data during the reference period is expected to decrease.

[0060] Furthermore, while the autocorrelation coefficient for weekday traffic data is calculated in the above embodiment, the autocorrelation coefficient for holiday traffic data may also be calculated. In this case, the length of the reference period is 2 × 24 hours.

[0061] Figures 13 to 15 show an example of the process of generating feature data from correlation coefficient data. The conversion unit 62 generates feature data from the correlation coefficient data calculated by the correlation calculation unit 61. The correlation coefficient data includes correlation coefficient data calculated by the cell / cell correlation calculation unit 61a, correlation coefficient data calculated by the cell / area correlation calculation unit 61b, and autocorrelation coefficient data calculated by the autocorrelation calculation unit 61c. However, in the descriptions relating to Figures 13 to 15, "correlation coefficient data" refers to the correlation coefficient data related to traffic volume calculated by the cell / cell correlation calculation unit 61a or the cell / area correlation calculation unit 61b.

[0062] In the case shown in Figure 13A, the target base station providing the target cell is operating normally. Furthermore, it is assumed that the surrounding base stations located near the target base station (in this case, the base stations providing the reference cell) are also operating normally. In this case, the correlation coefficient related to the traffic volume of the target cell stabilizes at a relatively high value. Note that in the graph representing the correlation coefficient shown in Figure 13A, each "○" represents the correlation coefficient calculated over one monitoring period. The same applies to Figures 13B, 14, and 15.

[0063] The conversion unit 62 generates differential data for each monitoring period. The differential data represents the difference between the correlation coefficient obtained in the target monitoring period and the correlation coefficient obtained in the monitoring period immediately preceding the target monitoring period. For example, as shown in Figure 7, when the monitoring periods are set with a one-hour shift, the difference is calculated in one-hour increments. In the embodiment shown in Figure 13A, the difference D represents the difference between the correlation coefficient C1 and the correlation coefficient C2. However, the conversion unit 62 may also calculate the difference between the correlation coefficient obtained in the target monitoring period and the correlation coefficient obtained in a monitoring period at a predetermined time away from the target monitoring period (for example, the monitoring period two hours before the action monitoring period).

[0064] When the target base station is operating normally, the correlation coefficient related to the traffic volume of the target cell is stable, so the difference calculated for each monitoring period stabilizes at a value close to zero. In other words, when the difference data is stable at a value close to zero, it can be assumed that the target base station is operating normally.

[0065] In the case shown in Figure 13B, a failure occurs at the target base station at time T1, and thereafter, the processing capacity of the target base station decreases. At this time, it is assumed that the surrounding base stations (in this case, the base stations providing the reference cell) are operating normally. In this case, the correlation coefficient related to the traffic volume of the target cell decreases. Furthermore, the absolute value of the difference calculated by the conversion unit 62 increases due to the occurrence of the failure. Therefore, when the absolute value of the difference calculated by the conversion unit 62 increases, it is estimated that a failure may have occurred at the target base station.

[0066] In the case shown in Figure 14, the performance of the target base station is gradually deteriorating. At this time, surrounding base stations (in this case, the base station providing the reference cell) are assumed to be operating normally. In this case, the correlation coefficient related to the traffic volume of the target cell gradually decreases. However, when the correlation coefficient changes slowly, the absolute value of the difference calculated by the conversion unit 62 is small. Therefore, in this case, it is difficult to estimate whether or not a failure has occurred at the target base station based on the difference data.

[0067] Therefore, the conversion unit 62 generates slope data in addition to the difference data. The slope data represents the trend of change in multiple correlation coefficients obtained within a predetermined period (i.e., the slope with respect to time). In the example shown in Figure 15, the slope G is calculated based on eight consecutive correlation coefficients C1 to C8. Note that, as shown in Figure 7, when the monitoring period is set in one-hour increments, the slope may be calculated based on, for example, 24 correlation coefficients obtained in the immediately preceding 24 hours.

[0068] When the performance of the target base station gradually deteriorates, the slope data remains shifted from zero to the negative direction over a long period of time, as shown in Figure 15. Therefore, when the slope data remains shifted from zero to the negative direction over a long period of time, it can be inferred that the performance of the target base station is gradually deteriorating.

[0069] Next, the operation of the failure prediction unit 70 will be described. The failure prediction unit 70 determines the state of the target base station or target cell based on the feature data (difference data, slope data) generated by the conversion unit 62. Alternatively, the failure prediction unit 70 may determine the state of the target base station or target cell based on the correlation coefficient data calculated by the correlation calculation unit 61 and the feature data generated by the conversion unit 62. In other words, the failure prediction unit 70 predicts a failure of the base station providing the target cell.

[0070] Here, the failure prediction unit 70 may predict base station failures based on the difference data and slope data of the correlation coefficient, as explained with reference to Figures 13 to 15. For example, when a difference greater than a predetermined threshold is detected, the failure prediction unit 70 may estimate that a failure has occurred at the target base station. Alternatively, when the slope data continues to shift from zero over a long period of time, the failure prediction unit 70 may estimate that the performance of the target base station is gradually deteriorating. Furthermore, the failure prediction unit 70 may use an AI model to predict the state of the base station.

[0071] AI models are not particularly limited, but for example, they can be implemented using a Gradient Boosting Decision Tree (GBDT). GBDT uses a decision tree, which is one of the machine learning algorithms. In addition, boosting, a type of ensemble learning, is used. Furthermore, gradient descent is used during boosting to minimize the error of previous predictions.

[0072] Figure 16 shows an example of AI model training. During the training of AI model 91, training information obtained from, for example, a telecommunications carrier is used. Training information consists of pairs of explanatory variables and target variables. The target variable represents the variable that we want to predict in machine learning. The explanatory variables represent variables that can explain the target variable.

[0073] When the AI ​​model 91 operates as the fault prediction unit 70, the target variable in the training information represents whether the base station is functioning normally or not. In this embodiment, "0" represents the state in which the base station is functioning normally, and "1" represents the state in which a fault has occurred at the base station. The explanatory variables are not particularly limited, but may include, for example, the following information for each cell. (1) Base station traffic volume (2) Number of UEs connected to the base station (3) Features related to similarity with reference cells (difference and slope) (4) Features related to the similarity with the total traffic volume within the communication area (difference and slope) (5) Deviation from the average weekday / holiday traffic pattern

[0074] Information (1) and (2) are obtained from PM data. Information (3) is generated based on the correlation coefficient obtained by the cell / cell correlation calculation unit 61a. Information (4) is generated based on the correlation coefficient obtained by the cell / area correlation calculation unit 61b. Information (5) is generated based on the autocorrelation coefficient obtained by the autocorrelation calculation unit 61c.

[0075] The AI ​​model 91 includes multiple parameters for calculating the target variable from the explanatory variables. These parameters are updated so that the target variable (i.e., the answer) can be obtained from the explanatory variables in the training information. For example, in the case shown in Figure 16, the AI ​​model 91 is updated so that the output value approaches "0" when the explanatory variable data EVD0001 is given, and so that the output value approaches "1" when the explanatory variable data EVD0005 is given.

[0076] Figure 17 shows an example of how to predict base station failures using a pre-trained AI model. The AI ​​model 91 is assumed to have been trained using the method shown in Figure 16.

[0077] The fault prediction unit 70 uses the updated AI model 91 to predict the state of the base station providing each cell. Specifically, for each cell, the fault prediction unit 70 provides the AI ​​model 91 with the above-mentioned information (1) to (5) at predetermined time intervals. The AI ​​model 91 then outputs a predicted value. Here, the AI ​​model 91 is trained so that the more likely a fault is to occur, the closer the predicted value approaches "1", and the less likely a fault is to occur, the closer the predicted value approaches "0". Therefore, the fault prediction unit 70 determines that a fault has occurred when the predicted value exceeds a predetermined threshold (for example, 0.5).

[0078] Note that the AI ​​models shown in Figures 16 and 17 require training information obtained from the telecommunications carrier. Therefore, if training information cannot be obtained from the telecommunications carrier, the fault prediction unit 70 predicts base station failures using other methods.

[0079] Figure 18 shows an example of another method for predicting base station failures. In this embodiment, the failure prediction unit 70 uses principal component analysis to predict base station failures. In principal component analysis, several principal components are created by aggregating data with a large number of variables. This method is equivalent to unsupervised learning, which does not require supervised information.

[0080] In this embodiment, two principal components (X component and Y component) are created from the information (1) to (5) described with reference to Figures 16 to 17. Principal component analysis, which creates the main components from a large number of elements (i.e., reduces dimensionality), is performed using known techniques.

[0081] The failure prediction unit 70 plots the principal component data (X component and Y component) obtained by principal component analysis on a two-dimensional coordinate system for each time period. In Figure 18, each circle ("○") represents the principal component data calculated for one cell.

[0082] Here, the principal component data values ​​of normal cells are considered to be approximate to each other. That is, when the principal component data of normal cells is plotted on a 2D coordinate system, it is expected to appear around a specific coordinate. Therefore, the probability of an anomaly occurring can be estimated according to the distance from the center of the distribution of the plotted principal component data.

[0083] For example, the cells plotted inside region E shown in Figure 18 are presumed to be normal cells. In contrast, the two cells plotted inside region E are presumed to potentially be abnormal.

[0084] Figure 19 is a flowchart illustrating an example of the processing performed by the fault prediction device 40. In the wireless communication system 1 shown in Figure 1, the processing in this flowchart is performed by the non-real-time RIC 10. This processing is also performed for each base station (or cell) provided by the telecommunications carrier. In the following description, the cell on which the procedure shown in Figure 19 is performed will be referred to as the "target cell," and the base station that provides the target cell will be referred to as the "target base station."

[0085] In S1, the data acquisition unit 50 collects CM data, PM data, and UE data from the base station system. In S2, the feature calculation unit 60 selects a reference cell for the target cell based on the CM data and PM data. In S3, the feature calculation unit 60 identifies the communication area of ​​the target base station based on the CM data, PM data, and UE data. After this, the fault prediction device 40 periodically repeats the processes in S4 to S9.

[0086] In S4, the data collection unit 50 acquires PM data and UE data from the base station. In S5, the feature calculation unit 60 calculates the correlation coefficient between the communication status of the target cell and the communication status of the reference cell. The communication status data used to calculate the correlation coefficient represents the traffic volume and / or the number of UEs. In S6, the feature calculation unit 60 calculates the correlation coefficient between the traffic volume of the target base station and the total traffic volume within the communication area identified in S3. In S7, the feature calculation unit 60 calculates the autocorrelation coefficient of the traffic volume of the target cell for both weekdays and holidays.

[0087] In S8, the feature calculation unit 60 calculates features for each of the correlation coefficients calculated in S5 to S7. These features correspond to the difference data explained with reference to Figure 13 and the slope data explained with reference to Figure 15. In S9, the failure prediction unit 70 predicts the state of the target base station or target cell based on the features calculated by the feature calculation unit 60 in S8. If, as a result, there is a possibility that the target base station is malfunctioning, the failure prediction device 40 may output an alarm.

[0088] As described above, the fault prediction device 40 according to the embodiment of the present invention predicts faults at each base station. However, simply monitoring the communication status of each base station individually makes it difficult to distinguish between a state in which the requested traffic volume has decreased and a state in which the base station's performance has deteriorated due to a fault or the like. In contrast, the fault prediction device 40 according to the embodiment of the present invention predicts a fault at a target base station based on the similarity between the communication status of the target base station and the communication status of surrounding base stations, and / or the fluctuation in the similarity between the communication status of the target base station and the communication status within the communication area of ​​the target base station. Therefore, according to the embodiment of the present invention, it is possible to distinguish between a state in which the requested traffic volume has decreased and a state in which the base station's performance has deteriorated due to a fault or the like, thus increasing the accuracy of fault prediction.

[0089] The fault prediction device 40 determines the status of the target base station or the target cell. For example, if the target base station is functioning normally, but the radio wave environment within the target cell deteriorates, the fault prediction device 40 can detect that a fault has occurred in the target cell.

[0090] Furthermore, in the above-described embodiment, a correlation coefficient is calculated between the communication state of the target cell and the communication state of one reference cell, but embodiments of the present invention are not limited to this configuration. For example, the fault prediction device 40 may calculate a correlation coefficient between the communication state of the target cell and the communication states of multiple surrounding cells. In this case, for example, the communication state of the target base station or target cell is determined based on the similarity between the communication state of the target cell and the average of the communication states of multiple surrounding cells.

[0091] <Hardware Configuration> Figure 20 shows an example of the hardware configuration of the fault prediction device 40. The fault prediction device 40 is implemented by a computer system 100 which includes a processor system 101, memory 102, storage device 103, input / output device 104, recording medium reader 105, and communication interface 106.

[0092] The processor system 101 can execute the failure prediction program and other programs stored in the storage device 103. The failure prediction program is executed as rApp, for example, in the non-real-time RIC 10 shown in Figure 1. The processor system 101 is implemented by one or more processors. When the processor system 101 includes multiple processors, the multiple processors may be connected to each other via a network. The execution of the failure prediction program by the processor system 101 provides the functions of the data acquisition unit 50, the feature calculation unit 60, and the failure prediction unit 70 shown in Figure 2. Memory 102 is used as the working area of ​​the processor system 101. Storage device 103 stores the failure prediction program and other programs mentioned above. The database 12 shown in Figure 1 is implemented by memory 102 and / or storage device 103.

[0093] The input / output device 104 may include input devices such as a keyboard, mouse, touch panel, and microphone. The input / output device 104 may also include output devices such as a display device and speaker. The recording medium reader 105 can acquire data and information recorded on the recording medium 110. The recording medium 110 is a removable recording medium that can be attached to and detached from the computer system 100. The recording medium 110 can be implemented, for example, by semiconductor memory, a medium that records signals by optical action, or a medium that records signals by magnetic action. The failure prediction program may be provided from the recording medium 110 to the computer system 100. The communication interface 106 provides the function of connecting to a network. When the failure prediction program is stored in the program server 120, the computer system 100 may acquire the failure prediction program from the program server 120. [Explanation of symbols]

[0094] 1. Wireless communication system 2. Terminal device (UE) 10 Non-real-time RIC 11 Wireless Network Optimization Unit 12. Database (DB) 20 Near Real-Time RIC 31 E2 nodes 32 Wireless Units (RUs) 40. Fault prediction device 50 Data Acquisition Unit 60 Feature Calculation Unit 61 Correlation Calculation Unit 61a Cell / Cell Correlation Calculation Unit 61b Cell / Area Correlation Calculation Unit 61c Autocorrelation Calculation Unit 62 Conversion section 70 Failure Prediction Unit 81 Target base stations 82, 83 surrounding base stations 91 AI Models 100 Computer Systems 101 Processor System

Claims

1. At least one of the following is generated: first correlation coefficient data representing the similarity between first communication status data representing the communication status of a first cell provided by a first base station and second communication status data representing the communication status of a second cell provided by a second base station, or second correlation coefficient data representing the similarity between the first communication status data and third communication status data representing the communication status within the communication area of ​​the first base station. The state of the first base station or the first cell is determined based on at least one of the first correlation coefficient data or the second correlation coefficient data. A failure prediction program that instructs a computer to perform a process.

2. The first correlation coefficient data is generated by calculating a first correlation coefficient representing the similarity between the first communication status data and the second communication status data at predetermined monitoring periods. The second correlation coefficient data is generated by calculating a second correlation coefficient representing the similarity between the first communication status data and the third communication status data at predetermined monitoring periods. The failure prediction program according to feature 1.

3. When the first correlation coefficient data is generated, first difference data representing the difference of the first correlation coefficient calculated for two different monitoring periods, and first slope data representing the slope of the first correlation coefficient with respect to time are generated. Based on the first difference data and the first tilt data, the state of the first base station or the first cell is determined. The failure prediction program according to claim 2, which causes the computer to perform the processing.

4. The first communication status data represents the traffic volume of the first cell, The second communication status data represents the traffic volume of the second cell, The first correlation coefficient, calculated for each monitoring period, represents the similarity between the traffic volume of the first cell and the traffic volume of the second cell. The failure prediction program according to feature 2.

5. The first communication status data represents the number of terminals connected to the first base station, The second communication status data represents the number of terminals connected to the second base station, The first correlation coefficient, calculated for each monitoring period, represents the degree of similarity between the number of terminals connected to the first base station and the number of terminals connected to the second base station. The failure prediction program according to feature 2.

6. When the second correlation coefficient data is generated, a second difference data representing the difference between the second correlation coefficients calculated for two different monitoring periods, and a second slope data representing the slope of the second correlation coefficient with respect to time are generated. Based on the second difference data and the second slope data, the state of the first base station or the first cell is determined. The failure prediction program according to claim 2, which causes the computer to perform the processing.

7. The first communication status data represents the traffic volume of the first base station, The third communication status data represents the total traffic volume within the communication area of ​​the first base station, The second correlation coefficient, calculated for each monitoring period, represents the similarity between the traffic volume of the first base station and the total traffic volume within the communication area of ​​the first base station. The failure prediction program according to feature 2.

8. From the first communication status data, specific communication status data that satisfies predetermined conditions is extracted, Autocorrelation coefficient data is generated by calculating the autocorrelation coefficient of the specific communication status data at predetermined monitoring intervals. Based on the autocorrelation coefficient data, the state of the first base station or the first cell is determined. A failure prediction program according to claim 1, which causes the computer to perform the processing.

9. Among a plurality of peripheral cells located near the first cell, the peripheral cell having the communication state most similar to the communication state of the first cell is selected as the second cell. A failure prediction program according to claim 1, which causes the computer to perform the processing.

10. A correlation calculation unit that generates at least one of the following: first correlation coefficient data representing the similarity between first communication status data representing the communication status of a first cell provided by a first base station and second communication status data representing the communication status of a second cell provided by a second base station, or second correlation coefficient data representing the similarity between the first communication status data and third communication status data representing the communication status within the communication area of ​​the first base station; A prediction unit that determines the state of the first base station or the first cell based on at least one of the first correlation coefficient data or the second correlation coefficient data, A fault prediction device equipped with the following features.

11. A base station system that provides multiple cells using multiple base stations, The system comprises a control system for controlling the plurality of base stations, The control system includes a fault prediction device that determines the status of the plurality of base stations or the plurality of cells, The fault prediction device, A correlation calculation unit that generates at least one of the following: first correlation coefficient data representing the similarity between first communication status data representing the communication status of a first cell provided by a first base station and second communication status data representing the communication status of a second cell provided by a second base station, or second correlation coefficient data representing the similarity between the first communication status data and third communication status data representing the communication status within the communication area of ​​the first base station; The system includes a prediction unit that determines the state of the first base station or the first cell based on at least one of the first correlation coefficient data or the second correlation coefficient data. A wireless communication system characterized by the following features.

Images