JP7198181B2 - COMMUNICATION QUALITY ANALYSIS SYSTEM, COMMUNICATION QUALITY ANALYSIS METHOD - Google Patents

Description

The present invention relates to a communication quality analysis device for communication equipment, a communication quality analysis system, and a communication quality analysis method.

The use of mobile communications is expanding. In order to realize mobile communication, operators who operate mobile networks install and operate wireless base stations. Patent Literature 1 discloses that deterioration in communication quality occurs in a radio communication section between a radio terminal and a radio base station. Further, Patent Literature 1 discloses that when communication quality deteriorates, information such as radio wave intensity and transmission/reception traffic is collected and analyzed in order to identify the cause.

JP 2010-239239 A JP 2017-142687 A

When communication quality deteriorates in a wireless communication section, the technique disclosed in Patent Document 2 can be used to analyze the data collected by the operator of the mobile network and determine whether or not the data exceeds a predetermined threshold. , the cause of deterioration can be estimated. However, even if the data collected by the operator of the mobile network reaches near the threshold but does not exceed the threshold, the analysis of the cause of degradation may be erroneous.

SUMMARY OF THE INVENTION An object of the present invention is to identify the cause of change (deterioration or improvement) in communication quality with high accuracy.

A representative example of the invention disclosed in the present application is as follows. That is, in a communication quality analysis system, at least one of communication equipment and equipment for measuring communication quality outputs a log data collection unit that acquires log data including data to be analyzed, and a statistical value of the data to be analyzed. A plurality of models for classifying causes of quality change of communication equipment are created from the calculated statistical values, and the classification probability calculated by each of the created models is multiplied by the weight of each model. a log data analysis unit for creating a composite model for classifying the cause of quality change of the communication equipment from the value, the log data analysis unit extracting the cause of quality change derived from the generated composite model and the operation management A combination of weights having a high degree of agreement with the cause of quality change specified by the operator is selected to create the composite model.

According to one aspect of the present invention, the cause of change in communication quality can be identified with high accuracy. Problems, configurations, and effects other than those described above will be clarified by the following description of the embodiments.

1 is a block diagram showing an example of the configuration of a radio communication system that is an analysis target of a log data analysis device; FIG. It is a block diagram which shows the example of a structure of a log-data analysis apparatus. It is a figure which shows the example of the operation|movement sequence of a log-data-analysis apparatus. It is a block diagram which shows the example of a structure of a log data analysis program. It is a figure which shows the example of the log data preserve|saved at a log-data accumulation area. FIG. 10 is a diagram showing an example of a calculation result of a correlation coefficient between a communication quality index and analysis target data; FIG. 10 is a diagram showing an example of calculation results of average values of analysis target data; FIG. 10 is a diagram showing an example of a change cause accumulation area; It is a figure which shows the example of a correlation coefficient model. It is a figure which shows the example of an average value model. FIG. 10 is a diagram showing examples of classification results and classification probabilities obtained by classifying causes of changes in communication quality using a correlation coefficient model; FIG. 10 is a diagram showing an example of classification results and classification probabilities obtained by classifying causes of changes in communication quality using an average value model; 4 is a flowchart of processing for calculating weights; FIG. 10 is a diagram showing examples of classification probabilities and classification results calculated by a composite model; FIG. 10 is a diagram showing an example of correct answer rate when changing the combination pattern of the weight of the correlation coefficient model and the weight of the mean value model. FIG. 10 is a diagram showing an example of a model display screen; FIG. 4 is a block diagram showing an example of the configuration of a change cause classification program; FIG. FIG. 10 is a diagram showing examples of classification probabilities and classification results calculated by a composite model; FIG. 11 shows an example of a change cause display screen; FIG.

An embodiment of the present invention will be described below with reference to the drawings.

(A1) System Configuration FIG. 1 is a block diagram showing an example of the configuration of a wireless communication system to be analyzed by the log data analysis device 50. As shown in FIG.

There are user terminals 10a, 10b, 10c, 10d, 10e that perform mobile communication with a wireless communication system. In the following, when a plurality of user terminals are not specified, the code "10" is used by omitting the lowercase letters of the alphabet, and the codes of the other components are similarly indicated. Here, the user terminal 10 may be a smart phone, a tablet PC (Personal Computer), or the like. The user terminal 10 wirelessly connects with wireless base stations 20a, 20b, and 20c installed by the operator of the mobile network, and transmits/receives data such as voice and video. The radio base station 20 is connected to the packet relay devices 30a and 30b by wire, divides data received from the user terminal 10 and log data created by the radio base station 20 into packets, and relays the packets. The packet relay devices 30a and 30b are connected to the packet relay device 30c by wire and relay packets. The packet relay device 30c transmits packets received from the packet relay devices 30a and 30b to the Internet 40, and transmits packets received from the Internet 40 to the packet relay devices 30a and 30b.

A log data analysis device 50, which is a cause classification system for classifying causes of changes in communication quality, is connected to the packet relay device 30 by wire, and analyzes log data output by one or both of the wireless base station 20 and equipment for measuring communication quality. is read and collected via the packet relay device 30 or from a recording medium. The log data analysis device 50 refers to the log data, and when detecting a quality change of the wireless base station 20, classifies the cause. A procedure for classifying the cause of communication quality change will be described later.

Note that the configuration, types, and numbers of the user terminal 10, the radio base station 20, the packet relay device 30, and the log data analysis device 50 are not limited to the example shown in FIG. .

(A2) Configuration of Log Data Analysis Device FIG. 2 is a block diagram showing an example of the configuration of the log data analysis device 50. As shown in FIG.

The log data analysis device 50 has a packet transmission/reception port 51 , a storage 52 , a memory 53 , and a CPU (Central Processing Unit) 54 . A part or all of the program of the software processing unit 55 is stored in the storage 52 . Here, the storage 52 is a storage device configured by a non-volatile memory such as a hard disk, semiconductor disk, flash memory, or the like. When the log data analysis device 50 operates, at least a program necessary for the operation is read into the memory 53 . The CPU 54 executes the processing of the log data analysis device 50 according to the program read into the memory 53 . The memory 53 may include a program execution environment such as an OS (Operating System) (not shown) other than the software processing unit 55, and may include data that the CPU 54 temporarily reads and writes. The CPU 54 may be composed of multiple CPUs.

The software processing unit 55 includes a log data collection program 56 , a log data analysis program 57 , a change cause classification program 58 , a model display program 59 and a classification result display program 61 . Each program is executed by the CPU 54 and exhibits the functions of the processing section. For example, the log data collection program 56 functions as a log data collection unit. Therefore, hereinafter, the description with the program as the subject may be replaced with the description with the CPU 54 that executes the program as the subject. The software processing unit 55 also includes a log data accumulation area 63 , an analysis result accumulation area 64 , a model accumulation area 65 , and a change cause accumulation area 66 .

The user interface program 62 acquires the data accumulated in the model accumulation area 65 and the cause of change accumulation area 66 and outputs the model and cause of change in communication quality to the display device 67 . It also has a function of accepting as an input the contents of the operation of the keyboard 68 and mouse 69 by the operator of the mobile network.

(A3) Operation Sequence of Log Data Analysis Device FIG. 3 is a diagram showing an example of an operation sequence of the log data analysis device 50. As shown in FIG.

The classification result display program 61 of the log data analysis device 50 receives an input of the change cause of the communication quality of each wireless base station from the mobile network operator 70 (1001), and writes it to the change cause storage area 66. When the change cause accumulation area 66 is updated, the log data analysis program 57 of the log data analysis device 50 reads the log data accumulated in the log data accumulation area 63 (1002) and analyzes the log data (1003). Then, the log data analysis program 57 writes the analysis result into the analysis result accumulation area 64 . The format of the data stored in the change cause storage area 66, the format of the log data stored in the log data storage area 63, the analysis procedure, and the format of the analysis results stored in the analysis result storage area 64 will be described later.

Next, the log data analysis program 57 of the log data analysis device 50 reads the data accumulated in the analysis result accumulation area 64 and change cause accumulation area 66 (1004), and classifies the cause of change in the communication quality of the radio base station. create a model for (1005). The log data analysis program 57 then writes the created model to the model accumulation area 65 . The procedure for creating a model and the format of the model stored in the model storage area 65 will be described later with reference to FIGS. 9 and 10. FIG.

Next, when the log data analysis program 57 of the log data analysis device 50 receives the log data from the wireless base station 20 (1006) or reads the log data from the recording medium, the log data analysis program 57 reads the log data The log data is shaped into the format of the accumulation area 63, and the log data is written. The log data analysis program 57 then reads the log data accumulated in the log data accumulation area 63 (1007) and analyzes the log data (1008). Furthermore, the log data analysis program 57 writes the analysis results into the analysis result accumulation area 64 .

Next, the change cause classification program 58 of the log data analysis device 50 reads the models accumulated in the model accumulation area 65 and the analysis results accumulated in the analysis result accumulation area 64 (1009), and uses them to The cause of change in communication quality of station 20 is classified (1010). Then, the change cause classification program 58 provides the classification result to the classification result display program 61 . The procedure for classifying causes of change will be described later.

Finally, the classification result display program 61 of the log data analysis device 50 displays the classification results provided from the change cause classification program 58 to the mobile network operator 70 (1011), prompting confirmation of the classification results. The procedure for displaying the classification results to the operator 70 of the mobile network will be described later. When the mobile network operator 70 confirms or corrects the cause of change in the communication quality of each radio base station, the classification result display program 61 writes the result as an input (1001) to the change cause storage area 66 . When the change cause accumulation area 66 is updated, the log data analysis device 50 recreates a new model and updates the previous model.

(A4) Procedure for Analyzing Log Data and Creating a Model FIG. 4 is a block diagram showing an example of the configuration of the log data analysis program 57. As shown in FIG.

The log data analysis program 57 includes a log data reading program 571, a correlation coefficient calculation program 572, an average value calculation program 573, a correlation coefficient model creation program 574, an average value model creation program 575, a weight calculation program 576, and the like.

In addition to the correlation coefficient calculation program 572 and the average value calculation program, the log data analysis program 57 may include a program for calculating other statistical values such as variance and probability. In addition to the correlation coefficient model creation program 574 and the mean value model creation program 575, a program for creating models for classifying causes by other statistical values such as variance and probability may be included.

FIG. 5 is a diagram showing an example of log data read from the log data accumulation area 63 by the log data reading program 571 of the log data analysis program 57. As shown in FIG.

The log data stored in the log data storage area 63 includes, for each wireless base station 20, a time stamp 6301, a communication quality index 6302, each analysis target data 6303 collected by the operator of the mobile network, and the like. The analysis target data 6303 is, for example, an index indicating the operating state of the wireless base station 20, such as the number of users connected to the base station or traffic volume. For example, the entry 6304a indicates that at “10:00 on May 4, 2019”, the communication quality index is “100”%, the analysis target data A is “3000”, and the analysis target data B is “99”. indicate that there was When the log data reading program 571 reads this log data, the correlation coefficient calculation program 572 of the log data analysis program 57 calculates the correlation coefficient between the communication quality index 6302 and each analysis target data. Also, the average value calculation program 573 of the log data analysis program 57 calculates the average value of each piece of analysis target data.

FIG. 6 is a diagram showing an example of a correlation coefficient calculation result 6401 calculated by the correlation coefficient calculation program 572 .

The correlation coefficient calculation result 6401 includes a wireless base station name 6402, each analysis target data 6403 collected by the operator of the mobile network, and the like. For example, entry 6404a indicates that the correlation coefficient between the communication quality index of wireless base station “20a” and analysis target data A for a certain period of time is “0.2”, and the communication quality index of wireless base station “20a” and analysis target data A are “0.2”. It shows that the correlation coefficient for the fixed period of data B is "0.7". When the correlation coefficient calculation result 6401 is created, the correlation coefficient calculation program 572 writes it to the analysis result accumulation area 64 .

FIG. 7 is a diagram showing an example of the average value calculation result 6411 calculated by the average value calculation program 573 .

The average value calculation result 6411 includes a wireless base station name 6412, each analysis target data 6413 collected by the operator of the mobile network, and the like. For example, entry 6414a indicates that the average value of analysis target data A of radio base station “20a” over a certain period of time is “3000”, and the average value of analysis target data B of wireless base station “20a” over a certain period of time is “95”. ”. When the average calculation result 6411 is created, the average calculation program 573 writes it to the analysis result accumulation area 64 .

FIG. 8 is a diagram showing an example of the cause-of-change accumulation area 66. As shown in FIG.

The change cause storage area 66 includes a wireless base station name 6601a and a change cause 6602a. For example, the entry 6603a indicates that the communication quality of the wireless base station "20a" has changed due to "change cause 1". The cause-of-change storage area 66 is created by the operator of the mobile network by inputting, through the user interface program 62, the content of actually classifying the causes of change.

If the correlation coefficient calculation result 6401 and the change cause storage area 66 exist, the correlation coefficient model creation program 574 reads them, and uses a classification method such as a decision tree to determine the cause of change in the communication quality of the wireless base station and its cause. A characteristic pattern is found from the relationship with the correlation coefficient of time, and a model is created that classifies the cause of change based on the value of the correlation coefficient of the radio base station analysis target data. Finally, the created correlation coefficient model is written in the model storage area 65 . On the other hand, if the average value calculation result 6411 and the change cause storage area 66 exist, the average value model creation program 575 reads them, and uses a classification method such as a decision tree to determine the cause of change in the communication quality of the radio base station and its cause. A characteristic pattern is found from the relationship with the average value of time, and a model for classifying the cause of change based on the average value of the analysis target data of the radio base station is created. Finally, the created mean value model is written in the model accumulation area 65 .

FIG. 9 is a diagram showing an example of a correlation coefficient model stored in the model storage area 65 for classifying causes of changes in communication quality of radio base stations by correlation coefficients of analysis target data.

The correlation coefficient model includes nodes 641 and leaves 642 . When classifying the cause of change in the communication quality of the radio base station with this model, first refer to the judgment formula of the node 641a. For example, when the correlation coefficient of analysis target data A is less than "0.5", the process proceeds to the determination formula of node 641b. On the other hand, when the correlation coefficient of the analysis target data A is "0.5" or more, the process proceeds to the determination formula of node 641c. Proceeding to the node 641b, if the correlation coefficient of the analysis target data B is "0.6" or more, the process proceeds to the leaf 642a, where the change cause 1 is classified as the cause of the change in communication quality. On the other hand, when the correlation coefficient of the analysis target data B is less than "0.6", the process advances to the leaf 642b, and the change cause 2 is classified as the cause of the change in communication quality. Proceeding to the node 641c, if the correlation coefficient of the analysis object data C is "0.3" or more, the process proceeds to the leaf 642c, where the change cause 3 is classified as the cause of the change in communication quality. On the other hand, when the correlation coefficient of the analysis object data C is referred to and is less than "0.3", the process advances to the leaf 642d, and the change cause 4 is classified as the cause of the change in communication quality.

At this time, the classification probability by each leaf 642 can be calculated. The classification probability is an index that indicates the probability with which the radio base station causing each change is classified into each leaf when the calculation result of the correlation coefficient is actually input to the created model. For example, if 10 radio base stations are classified into the leaf 642a, only 9 radio base stations have cause of change 1, and the remaining 1 radio base station has cause of change 2. The classification probability of 1 is 90%, and the classification probability of change cause 2 is 10%.

The correlation coefficient tends to show a characteristic value when the communication quality index and the data to be analyzed change significantly with respect to each other. Therefore, by classifying the cause of change in the communication quality of the radio base station based on the correlation coefficient of each piece of analysis target data, it is effective to classify the cause when the communication quality index significantly deteriorates or improves.

FIG. 10 is a diagram showing an example of an average value model for classifying causes of changes in communication quality of radio base stations, stored in the model storage area 65, by average values of analysis target data.

The mean model includes nodes 641 and leaves 642 . When classifying the cause of change in the communication quality of the radio base station with this model, first refer to the determination formula of the node 641d. For example, when the average value of the analysis target data A is "2500" or more, the process advances to the determination formula of node 641e. On the other hand, when the average value of the analysis target data A is less than "2500", the process proceeds to the determination formula of node 641f. Proceeding to the node 641e, if the average value of the analysis target data B is "50" or more, the process proceeds to the leaf 642e, and the cause of the change in communication quality is classified as change cause 1. FIG. On the other hand, when the average value of the analysis object data B is less than "50", the process advances to the leaf 642f, and the change cause 2 is classified as the cause of the change in communication quality. Proceeding to the node 641f, if the average value of the analysis target data C is "80" or more, the process proceeds to the leaf 642g, and the cause of the change in communication quality is classified as change cause 3. FIG. On the other hand, if the average value of the analysis object data C is less than "80", the process advances to the leaf 642h, and the change cause 4 is classified as the cause of the change in communication quality.

In the mean value model, similarly to the correlation coefficient model, the classification probability by each leaf 642 can be calculated.

Even when the change in the communication quality index is not large, the average value tends to appear as a characteristic value in the analysis target data depending on the cause of the change. Therefore, by classifying the cause of change in the communication quality of the radio base station based on the average value of each piece of data to be analyzed, it is effective to classify the cause when the communication quality index continues to deteriorate or improve.

FIG. 11 is a diagram showing an example of classification results obtained by classifying causes of changes in communication quality of radio base stations using the correlation coefficient model shown in FIG.

The classification probability 6421 in the correlation coefficient model includes the radio base station name 6422, the classification result 6423 in this model, and the classification probability 6424 in the classified leaf. For example, entry 6425a indicates that radio base station “20a” is classified as a “cause of change 1” leaf and has a 90% probability of being cause of change 1 and a probability of being cause of change 2 of 10%. Entry 6425b indicates that radio base station “20b” is classified as a “cause of change 1” leaf with a 90% probability of being cause of change 1 and a 10% probability of being cause of change 2. Entry 6425c indicates that the radio base station “20c” is classified as a “cause of change 2” leaf and has a 35% probability of being cause of change 1, a 40% probability of being cause of change 2, and a 25% probability of being cause of change 3. %. Entry 6325d indicates that the radio base station “20d” is classified as a “cause of change 4” leaf and has a 100% probability of being a cause of change 4 .

Here, referring to the change cause 6602a shown in FIG. 8, the change causes of the radio base stations 20a, 20b, 20c, and 20d are respectively "change cause 1," "change cause 1," "change cause 3," and "change cause 4". Referring to the classification result 6423 shown in FIG. 11, the classification results of the change causes of the wireless base stations 20a, 20b, 20c, and 20d according to this model are "change cause 1," "change cause 1," and "change cause 2," respectively. , “change cause 4”, and three-fourths of the entries answered correctly.

FIG. 12 is a diagram showing an example of classification results obtained by classifying causes of changes in communication quality of radio base stations using the mean value model shown in FIG.

The classification probabilities 6431 in the mean value model include the radio base station name 6432, the classification results 6433 in this model, and the classification probabilities 6434 in the classified leaves. For example, entry 6335a indicates that radio base station '20a' is classified as a 'cause of change 3' leaf and has a 40% probability of being cause of change 2 and a 60% probability of being cause of change 3. Entry 6435b indicates that radio base station “20b” is classified as a “cause of change 1” leaf and has a 90% probability of being cause of change 1 and a probability of being cause of change 2 of 10%. Entry 6435c indicates that radio base station '20c' is classified as a 'cause of change 3' leaf with a 40% probability of being cause of change 2 and a 60% probability of being cause of change 3. Entry 6435d indicates that radio base station “20d” is classified as a leaf of “cause of change 4” and has a 100% probability of being cause of change 4 .

Here, referring to the change cause 6602a shown in FIG. 8, the change causes of the radio base stations 20a, 20b, 20c, and 20d are respectively "change cause 1," "change cause 1," "change cause 3," and "change cause 4". Referring to the classification result 6433 shown in FIG. 12, the classification results of the change causes of the radio base stations 20a, 20b, 20c, and 20d according to this model are "change cause 3," "change cause 1," and "change cause 3," respectively. , “change cause 4”, and three-fourths of the entries answered correctly.

After calculating the classification probability of each deterioration cause by the correlation coefficient model and the classification probability of each deterioration cause by the mean value model, the weight calculation program 576 calculates the weight by which the classification probability of each model is multiplied. When the classification probability of each model is multiplied by the weight of each model and the calculated values are summed up, the cause of change with the highest classification probability is classified as the cause of change in communication quality. Then, a combination of weights to be multiplied by each model is carried out in a plurality of patterns, and a weight with a high degree of agreement with the cause of quality change specified by the operation manager, that is, a combination of weights with the highest percentage of correct answers is selected. As will be described later, the combination of weights may be selected by the log data analysis program 57 according to the operator's selection, or may be automatically selected by the log data analysis program 57 .

FIG. 13 is a flow chart of the weight calculation process performed by the weight calculation program 576 .

The weight calculation program 576 determines optimum weights for each of the correlation coefficient model and the mean value model in the composite model combining the correlation coefficient model and the mean value model. In this embodiment, two models are combined to create a composite model, but three or more models may be combined to create a composite model.

First, the weight [k] of the correlation coefficient model is set to 0.0. The weight [k] of the mean value model is calculated by 1-(weight [k] of the correlation coefficient model) (5761). That is, when the weight [k] of the correlation coefficient model is 0.0, the weight [k] of the mean value model is set to 1.0. Next, the classification probability in the correlation coefficient model is multiplied by the weight [k] of the correlation coefficient model. Also, the classification probability in the mean value model is multiplied by the weight [k] of the mean value model (5762). Then, those results are summed up (5763).

On the other hand, when the weight [k] of the correlation coefficient model is 0.0 and the weight [k] of the mean value model is 1.0, the classification probability in the correlation coefficient model and the classification probability in the mean value model are Multiplying 0.0, 1.0 and summing, the result is the same as the classification probability 6334 shown in FIG.

Next, for each entry, the “cause of change” with the highest classification probability is taken as the classification result of the composite model (5764). When the weight [k] of the correlation coefficient model is 0.0 and the weight [k] of the mean value model is 1.0, the classification result of the composite model is the same as the classification result 6433 shown in FIG.

Next, the number of correctly classified wireless base stations is calculated by comparing with the cause of change 6602a shown in FIG. is calculated (5765). If 3/4 of the entries are answered correctly, the correct answer rate [k] is 75%. Next, the weight [k+1] of the correlation coefficient model is increased by 0.1 and the same processing is repeated (5766).

FIG. 14 is a diagram showing examples of classification probabilities and classification results calculated by the composite model in order to classify causes of changes in communication quality of radio base stations.

The composite model 5771a includes a radio base station name 5772a, a classification result 5773a in the composite model, and a classification probability 5774. Classification probabilities 5774a to 5774d describe values calculated when the weight of the correlation coefficient model is 0.5 and the weight of the mean value model is 0.5. For example, entry 5775a indicates that radio base station “20a” is classified as “cause of change 1” with a probability of 45% for cause of change 1, a probability of 25% for cause of change 2, and a probability of 30% for cause of change 3. %. Entry 5775b indicates that radio base station “20b” is classified as “cause of change 1” with a 90% probability of being cause of change 1 and a probability of being cause of change 2 of 10%. Entry 5775c indicates that the radio base station “20c” is classified as “cause of change 3” and has a 17.5% probability of being cause of change 1, a 40% probability of being cause of change 2, and a 42% probability of being cause of change 3. .5%. Entry 5775d indicates that radio base station “20d” is classified as “cause of change 4” and has a probability of being cause of change 4 of 100%. Referring to FIG. 8, the causes of change of the radio base stations 20a, 20b, 20c, and 20d are "cause of change 1," "cause of change 1," "cause of change 3," and "cause of change 4," respectively. I understand. Referring to the classification result 5773a shown in FIG. 13, the results of classification of the wireless base stations 20a, 20b, 20c, and 20d as causes of change by the composite model are "cause of change 1," "cause of change 1," and "cause of change 3." ”, and “Cause of change 4”, and all four entries are correct.

FIG. 15 is a diagram showing an example of the correct answer rate 5781 when the combination pattern of the correlation coefficient model weight and the average value model weight is changed.

The correct answer rate 5781 for each weight combination pattern includes a correlation coefficient model weight 5782 , an average value model weight 5783 , and a correct answer rate 5784 . For example, entry 5785a indicates that the correct answer rate is 75% when the weight of the correlation coefficient model is 0.0 and the weight of the mean value model is 1.0. Entry 5785b indicates that the percentage of correct answers is 100% when the weight of the correlation coefficient model is 0.5 and the weight of the mean value model is 0.5.

When the weight calculation program 576 creates the correct answer rate 5781 for each weight combination pattern, the weight combination of the entry with the highest correct answer rate is written in the model accumulation area 65 (5767). In the case shown in FIG. 14, the correct answer rate is highest when the weight of the correlation coefficient model is 0.5 and the weight of the average value model is 0.5. A model weight of 0.5 is written in the model storage area 65 . Then, the model display program 59 displays the combination of the models forming the composite model and the calculated weights to the operator of the mobile network.

FIG. 16 is a diagram showing an example of the model display screen.

The model display screen 5901 includes a weight selection field 5902, weight display fields 5903a and 5903b, a composite model correct answer rate display field 5904, a model selection field 5905, a model ID display field 5906, a model name display field 5907, an OK button 5909, and the like. . For example, entry 5908a indicates that when the weight of the correlation coefficient model is "0.0" and the weight of the mean value model is "1.0", the correct answer rate of the composite model is "75%". Entry 5908b indicates that the correct answer rate of the composite model is "100%" when the weight of the correlation coefficient model is "0.5" and the weight of the mean value model is "0.5". Entry 5908c indicates that when the weight of the correlation coefficient model is "1.0" and the weight of the mean value model is "0.0", the correct answer rate of the composite model is "75%".

The operator of the mobile network can confirm the model display screen 5901 , select a model in the model selection field 5905 , and select a combination of weights in the weight selection field 5902 . After selecting the model and weight, the OK button 5909 is pressed. When the OK button 5909 is operated, the model display program 59 updates or corrects the models and weights forming the composite model written in the model storage area 65 .

(A5) Procedure for Analyzing Log Data and Classifying Causes of Change FIG.

The change cause classification program 58 includes a log data analysis result reading program 581 , a model and weight reading program 582 , and a change cause calculation program 583 .

The log data collection program 56 of the log data analysis device 50 shown in FIG. 3 periodically receives log data from the radio base station (1006). Note that log data may be read from a recording medium. The log data collection program 56 processes the acquired log data into the format shown in FIG. 5 and writes it to the log data accumulation area 63 . When the log data reading program 571 of the log data analysis program 57 reads this log data, the correlation coefficient calculation program 572 of the log data analysis program 57 calculates the correlation coefficient between the communication quality index 6202 and each analysis target data. do. The correlation coefficient calculation program 572 writes the correlation coefficient calculation result 6401 in the analysis result storage area 64 in the format shown in FIG. Also, the average value calculation program 573 of the log data analysis program 57 calculates the average value of each piece of analysis target data. The average value calculation program 573 writes the average value calculation result 6411 into the analysis result accumulation area 64 in the format shown in FIG.

When the log data analysis result reading program 581 of the change cause classification program 58 reads the analysis result from the analysis result accumulation area 64, the model and weight reading program 582 of the change cause classification program 58 subsequently reads from the model accumulation area 65 as shown in FIG. And the model shown in FIG. 10 and the weights shown in FIG. 15 are read (1008).

The change cause calculation program 583 of the change cause classification program 58 inputs the analysis result in the format shown in FIG. 6 to the correlation coefficient model read from the model storage area 65, and outputs the classification probability in the format shown in FIG. 7 is input to the average value model read from the model storage area 65, and the classification probability in the format shown in FIG. 12 is output. Subsequently, the change cause calculation program 583 multiplies these classification probabilities by weights read from the model accumulation area 65 .

FIG. 18 is a diagram showing examples of classification probabilities and classification results calculated by a composite model in order to classify causes of changes in communication quality of radio base stations.

The result 5771b calculated by the composite model includes the radio base station name 5772b, the classification result 5773b in the composite model, and the classification probabilities 5774e to 5774h. The classification probability 5774 describes the value calculated when the weight of the correlation coefficient model is 0.5 and the weight of the mean value model is 0.5. For example, entry 5775e indicates that radio base station “20p” is classified as “cause of change 4” and has a probability of being cause of change 4 of 100%. Entry 5775f indicates that radio base station “20q” is classified as “cause of change 2” with a probability of 17.5% for cause of change 1, a probability of 50% for cause of change 2, and a probability of 12% for cause of change 3. .5%. Entry 5775g indicates that the radio base station "20r" is classified as "cause of change 1" with a 45% probability of being cause of change 1, a 25% probability of being cause of change 2, and a 30% probability of being cause of change 3. indicates that Entry 5775h indicates that the radio base station “20s” is classified as “cause of change 3” with a probability of 17.5% for cause of change 1, a probability of 40% for cause of change 2, and a probability of 42% for cause of change 3. .5%. The change cause calculation program 583 provides the classification result display program 61 with the calculated classification probabilities and classification results (FIG. 18).

(A6) Procedure for Displaying Classified Causes of Change When the classification result display program 61 of the log data analysis device 50 shown in FIG. to the mobile network operator 70 (1010).

FIG. 19 shows an example of a change cause display screen.

The change cause display screen 5911 is displayed to prompt the operator of the mobile network to confirm the cause of change in communication quality. The change cause display screen 5911 includes a confirmation priority column 5912 , a wireless base station name column 5913 , a display column 5914 for the difference between the classification probability 1st and the classification probability 2nd, a classification result column 5915 , and an OK button 5917 . The classification result display program 61 displays the classification results of the causes of change in ascending order of the difference 5914 between the first classification probability and the second classification probability. For example, entry 5916a indicates that the difference between the first classification probability (42.5% for change cause 3) and the second classification probability (40% for change cause 2) is 2.5% for radio base station “20s”. , indicates that it is classified as “change cause 3”. Entry 5916b indicates that for radio base station "20r", the difference between the first classification probability (45% of change cause 1) and the second classification probability (30% of change cause 3) is 15%, and "change cause 1" indicates that it is classified as Entry 5916c indicates that the difference between the first classification probability (50% of change cause 2) and the second classification probability (17.5% of change cause 1) is 32.5% for radio base station "20q", and " It indicates that it is classified as “change cause 2”. Entry 5916d indicates that for the radio base station "20p", the difference between the first classification probability (100% for change cause 4) and the second classification probability (0% for other change causes) is 100%, and "change cause 4 ” indicates that it is classified as

The operator of the mobile network confirms the classification result 5915 of the cause-of-change display screen 5911 and implements countermeasures for the radio base station according to the contents. If the deterioration of communication quality is resolved, the classification result 5915 is estimated to be correct. If the quality change persists, it is assumed that the classification result 5915 is incorrect. If the classification result 5916 is incorrect, the mobile network operator corrects the classification result 5916 and finally presses an OK button 5917 . When the OK button 5917 is operated, the classification result display program 61 writes the classification result 5915 to the change cause accumulation area 66 (1010).

At this time, the entry at the top of the change cause display screen 5911 is an entry whose correct answer probability is not so high. The operator of the mobile network preferentially confirms the entries at the top of the cause-of-change display screen 5911, thereby correcting and preferentially accumulating the classification results of radio base stations that are presumed to have a low probability of correct answers. and improve the accuracy of cause classification. On the other hand, entries at the lower level of the change cause display screen 5911 are entries that are estimated to have a high probability of correct answers. If the operator of the mobile network does not have time, he/she may operate the OK button for the entries at the lower level of the cause-of-change display screen 5911 without performing countermeasures for the radio base station and verifying the classification results. As a result, a large amount of data can be accumulated in the change cause accumulation area 66 in a short period of time. When the data in the cause-of-change accumulation area 66 increases (for example, every time the data in the cause-of-change accumulation area 66 is updated, at predetermined time intervals, etc.), the correlation coefficient model and the average value model are updated to increase the accuracy of cause classification. improves. In addition, it is preferable to reconfirm the entry at the lower level of the cause-of-change display screen 5911 for which the OK button was operated before taking measures for the wireless base station, immediately after taking measures.

In the processing described above, a combined model obtained by multiplying and combining a plurality of models by the weight of each model is used to classify the cause of change in the communication quality of the radio base station. This makes it possible to classify the cause of change in communication quality of the radio base station with high accuracy. In addition, by urging mobile network operators to preferentially confirm the classification results of radio base stations that are estimated to have a low probability of correct answers, it is possible to identify radio base stations that are estimated to have a low probability of correct answers. The classification results are corrected and preferentially accumulated, the model is updated, and the accuracy of cause classification can be improved in a short time.

The present invention and/or the present embodiment can be configured as a log data analysis method or a computer program executed by the log data analysis device 50 in addition to the configuration of the log data analysis device 50 . The computer program is preferably recorded on a computer-readable recording medium. Various media such as floppy disks, CD-ROMs, DVD-ROMs, magneto-optical disks, memory cards, and hard disks can be used as recording media.

As described above, the communication quality analysis system of the embodiment of the present invention includes a log data collection unit ( A log data collection program 56), and a plurality of models (e.g., correlation coefficient model, mean value model) for calculating statistical values of the data to be analyzed and classifying causes of quality change in communication equipment from the calculated statistical values. and create a composite model for classifying the cause of quality change of communication equipment from the value obtained by multiplying the weight of each model by the classification probability calculated by each created model. Program 57), the log data analysis unit selects a weight combination with a high degree of matching between the cause of quality change derived from the created composite model and the cause of quality change specified by the operation manager, Since a composite model is created, it is possible to create a model that can identify the cause of quality change with high accuracy.

In addition, the log data analysis unit includes a correlation coefficient calculation unit (correlation coefficient calculation program 572) that calculates the correlation coefficient between the analysis target data and the communication quality index, and from the calculated correlation coefficient, the quality change of the communication equipment A correlation coefficient model creation unit (correlation coefficient model creation program 574) that creates a model for classifying causes, an average value calculation unit (average value calculation program 573) that calculates the average value of the communication quality index, and the calculated Since it has an average value model creation unit (average value model creation program 575) that creates a model for classifying the cause of quality change of communication equipment from the average value, the cause of quality change is identified by the communication quality index that is highly correlated with the cause of quality change. In addition to being able to specify with accuracy, by using the average value together, it is possible to grasp the change of the cause of quality change that cannot be grasped only by the correlation.

In addition, the log data analysis unit has a weight calculation unit (weight calculation program 576) that tries multiple patterns of combinations of weights for each model and selects a combination of weights that can classify the cause of quality change with the highest probability. Therefore, the optimal combination can be selected for all weight patterns. Also, the calculation is simple and the processing can be executed quickly.

In addition, since it has a change cause classification unit (change cause classification program 58) that classifies the cause of change in communication quality by applying the composite model created to the collected log data, the cause of change (deterioration or improvement) in communication quality can be identified. Can be specified with high accuracy.

In addition, since a classification result display unit (classification result display program 61) is provided for outputting the causes of change classified in ascending order of the classification probability calculated by the composite model, teacher data can be displayed from the cause of change in communication quality with a low classification probability. It is prepared and can improve the classification accuracy of the model at an early stage. That is, it is possible to reduce man-hours for creating training data.

It should be noted that the present invention is not limited to the embodiments described above, but includes various modifications and equivalent configurations within the scope of the appended claims. For example, the above-described embodiments have been described in detail for easy understanding of the present invention, and the present invention is not necessarily limited to those having all the described configurations. Also, part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Moreover, the configuration of another embodiment may be added to the configuration of one embodiment. Further, additions, deletions, and replacements of other configurations may be made for a part of the configuration of each embodiment.

In addition, each configuration, function, processing unit, processing means, etc. described above may be realized by hardware, for example, by designing a part or all of them with an integrated circuit, and the processor realizes each function. It may be realized by software by interpreting and executing a program to execute.

Information such as programs, tables, and files that implement each function can be stored in storage devices such as memories, hard disks, SSDs (Solid State Drives), or recording media such as IC cards, SD cards, and DVDs.

In addition, the control lines and information lines indicate those considered necessary for explanation, and do not necessarily indicate all the control lines and information lines necessary for mounting. In practice, it can be considered that almost all configurations are interconnected.

10 User terminal 20 Radio base station 30 Packet relay device 40 Internet 50 Log data analysis device 51 Packet transmission/reception port 52 Storage 53 Memory 54 CPU
55 Software processing unit 56 Log data collection program 57 Log data analysis program 58 Change cause classification program 59 Model display program 61 Classification result display program 62 User interface program 63 Log data accumulation area 64 Analysis result accumulation area 65 Model accumulation area 66 Change cause accumulation Area 67 Display Device 68 Keyboard 69 Mouse 70 Operator

Claims (12)

A communication quality analysis system,
a log data collection unit that acquires log data including analysis target data output by at least one of communication equipment and equipment for measuring communication quality;
Calculate statistical values of the data to be analyzed, create a plurality of models for classifying causes of quality change of communication equipment from the calculated statistical values, and classify the probability calculated by each of the created models, A log data analysis unit that creates a composite model for classifying the cause of quality change of the communication equipment from the value obtained by multiplying the weight of each model,
The log data analysis unit selects a weight combination with a high degree of matching between the quality change cause derived from the created composite model and the quality change cause identified by the operation manager, and creates the composite model. A communication quality analysis system characterized by: The communication quality analysis system according to claim 1,
The log data analysis unit
a correlation coefficient calculation unit that calculates a correlation coefficient between the analysis target data and the communication quality index;
a correlation coefficient model creation unit that creates a model for classifying causes of quality change of communication equipment from the calculated correlation coefficient;
an average value calculation unit that calculates an average value of the communication quality index;
A communication quality analysis system, comprising: an average value model creation unit that creates a model for classifying causes of quality change of communication equipment from the calculated average value. The communication quality analysis system according to claim 1,
A communication quality analysis system, wherein the communication equipment is a radio base station. The communication quality analysis system according to claim 1,
The log data analysis unit is characterized by having a weight calculation unit that tries combinations of weights of the models in a plurality of patterns and selects a combination of weights that can classify the cause of quality change with the highest probability. Communication quality analysis system. The communication quality analysis system according to claim 1,
A communication quality analysis system, comprising: a change cause classification unit that classifies causes of communication quality change by applying the created composite model to the collected log data. The communication quality analysis system according to claim 1,
A communication quality analysis system, comprising: a classification result display unit that outputs causes of change classified in ascending order of difference between the first classification probability and the second classification probability calculated by the composite model. A communication quality analysis method in which a computer analyzes the communication quality of a communication system,
The computer has an arithmetic unit that executes predetermined arithmetic processing and a storage device that can be accessed by the arithmetic unit,
The communication quality analysis method includes:
A log data collection procedure in which the arithmetic device acquires log data output by at least one of communication equipment and equipment for measuring communication quality and including analysis target data;
The computing device calculates statistical values of the data to be analyzed, creates a plurality of models for classifying causes of quality change of communication equipment from the calculated statistical values, and calculates the values of each of the created models. a log data analysis procedure for creating a composite model for classifying the cause of quality change of the communication equipment from the value obtained by multiplying the classification probability by the weight of each model;
In the log data analysis procedure, the computing device selects a weight combination with a high degree of matching between the cause of quality change derived from the created composite model and the cause of quality change identified by the operation manager, A communication quality analysis method characterized by creating a composite model. The communication quality analysis method according to claim 7,
The log data analysis procedure includes:
a correlation coefficient calculation procedure in which the calculation device calculates a correlation coefficient between the analysis target data and the communication quality index;
a correlation coefficient model creation procedure in which the computing device creates a model for classifying causes of quality change of communication equipment from the calculated correlation coefficient;
an average value calculation procedure in which the arithmetic device calculates an average value of the communication quality index;
A communication quality analysis method, wherein the arithmetic unit creates a model for classifying the cause of quality change of the communication equipment from the calculated average value. The communication quality analysis method according to claim 7,
A communication quality analysis method, wherein the communication equipment is a radio base station. The communication quality analysis method according to claim 7,
The log data analysis procedure includes a weight calculation procedure in which the computing device tries combinations of weights of the models in a plurality of patterns, and selects a combination of weights that can classify the cause of quality change with the highest probability. A communication quality analysis method characterized by: The communication quality analysis method according to claim 7,
A communication quality analysis method, wherein the computing device applies the created composite model to the collected log data to classify causes of communication quality change. The communication quality analysis method according to claim 7,
The communication quality analysis method, wherein the computing device includes a classification result display procedure for outputting the classified causes of change in ascending order of difference between the first classification probability and the second classification probability calculated by the composite model.

Images