JP6167052B2 - Communication traffic prediction apparatus and program - Google Patents

Description

  The present invention relates to a communication traffic prediction apparatus and program, and more particularly to a communication traffic prediction apparatus and program for predicting a future communication traffic amount based on the number of calls of a communication terminal.

  Patent Documents 1-4 disclose techniques for calculating future communication traffic based on actual values of traffic characteristics.

  In Patent Document 1, as a method for predicting the traffic volume of a communication area that fluctuates depending on the date and time, location registration information in a mobile / mobile communication network is monitored and measured, and the call status of a mobile terminal that has made a call is also monitored A technique for measuring and predicting the future date and time and the call situation for each region based on these past performance values is disclosed.

  In Patent Document 2, as a method for predicting traffic in a fluctuating communication region, traffic depending on each position (unit region) on a map is calculated from traffic in a telecommunication facility and a service providing region of the telecommunication facility. A technique for predicting future position-dependent traffic from actual values of position-dependent traffic is disclosed. The amount of traffic in each unit area is calculated based on the area ratio between the service providing area of the telecommunication equipment and the observation unit area as the traffic between the telecommunication equipment.

  In Patent Document 3, as a method for predicting the communication traffic volume of a fixed network such as a backbone line or an INET line, the time and the size of a packet that flow on the network are measured, and the traffic volume and its fluctuation are affected. A technique for managing external factors (calendar information, event status, etc.) and predicting the future traffic volume by applying a statistical analysis method to such information is disclosed.

  In Patent Document 4, as a method for predicting the amount of communication traffic that will occur when a new service is spread, the number of service subscribers affecting the traffic volume of the new service, and the trend of the traffic amount at the time when the service is spread in the past are disclosed.・ Technology is disclosed that manages and measures performance, etc., and predicts traffic volume when new services are spread in the future.

JP 2004-80315 A JP 2001-168985 A JP 2012-253445 JP JP 2012-182677 A

  Patent Document 1 predicts the number of calls by measuring signal information related to location registration and signaling, and predicts the amount of communication traffic based on the prediction result of the number of calls. Therefore, if the terminal user repeatedly moves in the boundary region of the location area related to location registration (region / area unit managed by location registration), location registration information is generated more than necessary, and the number of location registration information becomes the number of users. It is not proportional. As a result, the correlation between the number of location registration information and the traffic amount is also lowered, and the prediction accuracy of the traffic amount is lowered.

  Patent Documents 2-4 directly calculate the predicted traffic volume by applying a time-series prediction method to the actual value of the traffic volume. However, the accuracy of time-series prediction decreases as the total amount of communication traffic and the fluctuation range of the traffic amount per communication increase, so traffic that includes a large amount of traffic per communication and a small amount of communication like today. There was a technical problem that the prediction accuracy decreased under the environment.

  Furthermore, in Patent Document 2, although geographical-dependent traffic is measured for each unit area, the traffic amount of each unit area is the total traffic regardless of whether each area is a residential area, a shopping area, an office area, or the like. Calculated as area ratio to quantity. However, since the area ratio and traffic ratio of each unit region are not necessarily equivalent, the traffic volume cannot be predicted with high accuracy.

  SUMMARY OF THE INVENTION An object of the present invention is to solve the above-mentioned problems of the prior art and to consider a communication dependence prediction apparatus and program capable of accurately predicting the future total amount of communication traffic based on the number of calls in consideration of the geographical dependence inherent in the communication traffic. Is to provide.

  In order to achieve the above object, the present invention is characterized in that a communication traffic prediction apparatus for predicting the amount of communication traffic by a communication terminal has the following configuration.

  (1) Traffic characteristic measurement means for measuring traffic characteristics including the number of calls for each unit area obtained by dividing the traffic prediction target area according to geography, statistical analysis means for statistical analysis of the number of calls at a predetermined period, and statistical analysis The prediction model construction means for constructing a prediction model for predicting the future number of calls from the statistical analysis results for each unit area, and the statistical analysis results of the traffic characteristics, based on the results of the above and the actual number of calls thereafter Call number predicting means for predicting the future number of calls for each unit area by applying to the prediction model is provided.

  (2) The traffic characteristic measuring means further comprises means for measuring the traffic volume per call for each unit area, and the analyzing means statistically analyzes the time series of the traffic volume per call at a predetermined cycle. And a prediction model construction means constructs a prediction model Mvpc for each unit region that predicts the traffic volume per call and the traffic volume per call based on the analysis result. And the prediction means applies the analysis result of the traffic volume per call number measured in the prediction area to the prediction model Mvpc to which the traffic tendency is suitable, and calculates the traffic volume per call number in the future. And a means for predicting.

According to the present invention, the following effects are achieved.
(1) The number of calls is not affected by sudden or temporary fluctuations due to the difference or change in communication contents such as contents and services, but greatly influenced by the terminal penetration rate with high geographical dependence. According to the present invention, a prediction model for the number of calls is constructed based on the statistics of the number of calls for each unit area, and the actual number of calls observed in each unit area is applied to the prediction model to obtain the number of future calls. As a result, the total traffic amount is predicted, so that it is possible to accurately predict communication traffic reflecting the geographical dependency.

  (2) Since the traffic volume per call also shows the geographical dependence inherent in communication traffic, it is possible to establish not only the number of calls but also the traffic volume per call by constructing a prediction model for each unit area. Accurate traffic prediction that eliminates the geographical dependence inherent in traffic becomes possible.

It is the functional block diagram which showed the structure of 1st Embodiment of the communication traffic prediction apparatus with which this invention is applied. It is the flowchart which showed the operation | movement of 1st Embodiment of this invention. It is the functional block diagram which showed the structure of 2nd Embodiment of the communication traffic prediction apparatus with which this invention is applied. It is the flowchart which showed the operation | movement of 2nd Embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a functional block diagram showing the configuration of the main part of the communication traffic prediction apparatus according to the first embodiment of the present invention. Here, the configuration unnecessary for the description of the present invention is omitted.

  The communication traffic prediction apparatus of the present invention can be configured by mounting an application (program) for realizing each function on a general-purpose computer or server. Alternatively, a part of the application may be configured as a dedicated machine or a single-function machine in which hardware or ROM is implemented.

  The communication log collection unit 8 with position information was seen as the communication log of each communication terminal when time information, transmission source information, destination information, and each communication terminal communicated for each call (communication trial unit) (communication possible) Various information including communication delay times between the base station and each base station is collected, and further estimated by the positioning result using each communication delay time or the GPS function installed in each communication terminal. The recorded position is recorded as position information.

  The communication log DB1 with location information has a large number of geographically dependent unit areas (for example, 1 km × 1 km rectangles) by virtually dividing the communication service providing area into a mesh based on separately divided information. Each communication log is defined and linked to one of the unit areas based on the position information to create a database. The division information includes a prediction target area and information on a mesh standard / unit (size) defined by the Geographical Survey Institute or the Ministry of Internal Affairs and Communications, for example.

  The traffic characteristic measuring unit 2 measures the traffic characteristic including the number of calls C and the traffic amount per call for each unit area based on the communication log associated with each unit area.

The traffic tendency analysis unit 3 statistically processes the time series data of the number of calls C and calculates a traffic tendency index value. In the present embodiment, the traffic trend analysis unit 3 includes a primary differential value calculation unit 3a and a secondary differential value calculation unit 3b, and the primary differential value ΔC [every predetermined unit time τ with respect to the time series of the number of calls C. τi] and the second derivative Δ ² C [τi] are calculated.

In the traffic characteristic DB 5, the time series of the number of calls C measured for each unit region and the primary differential value ΔC [τi] and the secondary differential value Δ ² C [τi] calculated for each unit time τi are history information. Registered as

The prediction model construction unit 6 is based on the time series of the number of calls C registered for each unit area in the traffic characteristic DB 5 and the primary differential value ΔC [τi] and the secondary differential value Δ ² C [τi]. Unit of prediction model Mc for predicting future call number C based on arbitrary call number C and primary differential value ΔC [τi] and secondary differential value Δ ² C [τi] calculated from call number C Build for each area.

The prediction model Mc includes a primary differential value ΔC [τi] and a secondary differential value Δ ² C [τi] for each unit time obtained from the time series of the number of calls C, and each unit after a predetermined period has elapsed from the unit time. It can be constructed by applying SVM (Support Vector Machine) and NN (Neural Network) using the combination with the number of calls at the time as teacher data. Alternatively, a prediction method such as a genetic algorithm or a machine learning algorithm may be applied, or a multiple regression analysis model or a clustering method may be used.

The primary differential value ΔC and the secondary differential value Δ ² C may be registered as normalized values divided by the peak value or average value of the number of calls C. In addition, the parameters used for the call number prediction together with the time series of the call number C are not limited to the above-mentioned primary differential value ΔC and secondary differential value Δ ² C, and calendar information such as date and day of the week is used together. You may do it.

The traffic prediction unit 4 includes a time series of the number of calls C measured in an arbitrary prediction region, and a primary differential value ΔCj [τi] and a secondary differential value Δ ² Cj calculated by the traffic tendency analysis unit 3. By applying [τi] to a registered prediction model Mc that matches (similar) the traffic tendency, the number of calls C (predicted call number Cf) at a future arbitrary time in the prediction region is calculated.

  The predicted traffic total amount calculation unit 7 calculates the future predicted traffic total amount Vf (= Cf × VpC) in the prediction region for each unit time τ by multiplying the predicted number of calls Cf by the traffic amount VpC per call. To do. The unit time τ may be a fixed time (for example, one hour cycle), or a unit region with a large amount of traffic is set to a shorter time cycle, while a unit region with a small amount of traffic is a longer time cycle. For example, it may be changed dynamically according to the traffic characteristics of each unit area. In addition, with regard to the traffic characteristics for each unit area, the daily fluctuation may be used as a time-series actual value.

  FIG. 2 is a flowchart showing the operation of the first embodiment of the present invention. Here, it is assumed that a large number of prediction models Mc are built for each unit region by the prediction model construction unit 6 and registered in advance. explain.

In step S1, the traffic characteristic measuring unit 2 measures the time series of the number of calls C in the prediction area based on the communication log stored in the communication log DB1 with position information, and further determines the traffic amount per call. VpC statistics are calculated. In step S2, the traffic trend analysis unit 3 calculates a primary differential value ΔC [τi] and a secondary differential value Δ ² C [τi] for each predetermined unit time τ for the time series of the number of calls C. Is done.

In step S3, the time series of the number of calls C measured in the prediction region and the primary differential value ΔC [τi thereof are selected from the large number of prediction models Mc that are previously constructed for each unit region by the prediction model construction unit 6. ] And the second derivative value Δ ² C [τi] are searched for one prediction model Mc having the highest degree of matching.

In step S4, it is determined whether or not a suitable prediction model has been searched. If it can be searched, the process proceeds to step S5. In step S5, the time series of the number of calls C measured in the prediction region and its first and second derivative values ΔC and Δ ² C are applied to the prediction model Mc, and the number of future calls in the prediction region. Cf is predicted.

On the other hand, if it is determined in step S4 that no suitable prediction model is registered, the process proceeds to step S6. In step S6, a general-purpose popular curve such as a logistic curve or a Gompertz curve, or a time series prediction model such as ARIMA, and the time series of the number of calls C and its first derivative value ΔC and second derivative value Δ ² C are applied. Thus, the future number of calls Cf in the prediction region is uniformly predicted.

  In step S7, the predicted traffic total amount calculation unit 4 calculates the predicted traffic total amount Vf by multiplying the predicted call count Cf by the traffic amount statistical value VpC per call. At this time, if the number of calls C is a predicted value in units of one hour, the predicted total traffic Vf is also calculated for each unit area and every hour. Further, regarding the statistical value VpC, an average value, a median value, a predetermined percentage value (% ile, for example, 99% ile) or the like may be used in order to remove abnormal values and noise components.

In step S8, the primary differential value ΔC and the secondary differential value Δ ² C of the prediction region are registered in the traffic characteristic DB 5, and are used for new construction of a prediction model Mc or relearning of an existing model.

  According to the present embodiment, the prediction model Mc of the number of calls is constructed Mc based on the statistical value of the number of calls for each unit region, and the prediction model Mc that best matches the actual measurement value and the statistical value of the number of calls measured in the prediction region. By applying this method to predicting the number of future calls and thus the total traffic volume, it becomes possible to accurately predict traffic traffic reflecting the geographical dependence.

Incidentally, in the first embodiment described above, predictive model Mc is built based on the time of the measured number of calls C per unit area sequence and its statistics (primary differential value ΔC and second derivative value delta ² C) However, according to the measurement results of the inventors, the traffic volume per call VpC also has a specific geographical dependency, and even if the areas where the number of calls C is the same are compared, the VpC may differ greatly. is there. Therefore, the total traffic amount that is the product of the unit traffic amount VpC may not be accurately predicted only by predicting the number of calls C.

  For example, VpC tends to be small because of the high use of Web-related content in stations and downtowns, whereas VpC tends to increase because of the large use of video and large-capacity content in residential areas. Therefore, predicting future traffic volume by applying the number of calls observed in a residential area to the prediction model Mc constructed based on the number of calls observed in a unit area near a station or downtown. As a result, a lower VpC than the original is applied, resulting in a lower prediction result.

  Also, in areas where high-function terminals that support high-speed communication, such as LTE, have become widespread, VpC increases moderately in areas where the use of web-related content is high, whereas large-capacity systems In areas where services are frequently used, VpC tends to rise sharply at the start of browsing files due to the increase in high-resolution image display and video browsing.

  In addition, in areas where communication quality is good and throughput is high, more communications and data transfer are possible per unit time, so VpC also increases. Even if they are the same, VpC may decrease due to congestion and congestion. Therefore, in the second embodiment described below, VpC is set as a prediction target by the prediction model in addition to the number of calls C.

  FIG. 3 is a functional block diagram showing the configuration of the second embodiment of the present invention. The same reference numerals as those described above represent the same or equivalent parts, and thus the description thereof is omitted.

In the traffic tendency analysis unit 3, the primary differential value calculation unit 3c calculates the primary differential value ΔVpC [τi] for each predetermined unit time τ with respect to the time series of the traffic volume VpC per call. The secondary differential value calculation unit 3d calculates a secondary differential value Δ ² VpC [τi] for each predetermined unit time τ with respect to the time series of VpC.

The prediction model construction unit 6 selects an arbitrary time based on the time series of VpC registered for each unit region in the traffic characteristic DB 5 and the primary differential value ΔVpC [τi] and the secondary differential value Δ ² VpC [τi]. A prediction model for predicting the future VpC of the region based on the VpC measured in the prediction region and the first derivative ΔVpC [τi] and the second derivative Δ ² VpC [τi] calculated from the number of calls VpC Build Mvpc for each unit area.

The traffic prediction unit 4 uses the analysis result of the time series of the VpC measured in the prediction region and the primary differential value ΔVpC [τi] and the secondary differential value Δ ² VpC [τi] to the prediction model Mvpc suitable for the traffic tendency. By applying this function, a function of calculating VpCf at a future arbitrary time in the prediction region is further provided.

  The predicted traffic total amount calculation unit 7 multiplies the predicted number of calls Cf and the predicted traffic amount VpCf per call for each unit time, thereby calculating the future predicted traffic total amount Vf (= Cf × VpCf) for each unit time τ. To calculate.

  FIG. 4 is a flowchart showing the operation of the present embodiment. Here, description will be made assuming that each of the prediction models Mc and Mvpc has been previously built and registered for each unit area by the prediction model construction unit 6.

  In step S21, the traffic characteristic measuring unit 2 measures the actual value of the number of calls C for each unit area based on the communication log stored in the communication log DB1 with position information, and further, the traffic per call. A statistical value of the quantity VpC is calculated.

In step S22, the traffic trend analysis unit 3 calculates a primary differential value ΔC [τi] and a secondary differential value Δ ² C [τi] for each predetermined unit time τ for the time series of the number of calls C. Is done. Further, a primary differential value ΔVpC [τi] and a secondary differential value Δ ² VpC [τi] are calculated for each predetermined unit time τ for the time series of the traffic volume VpC per call.

In step S23, the time series of the number of calls C used for the construction of each prediction model Mc and the primary differential value ΔC [τi] and the secondary differential value Δ ² C, the time series of the number of calls C to be predicted and the time series thereof. The degree of conformity between the primary differential value ΔC [τi] and the secondary differential value Δ ² C [τi] is appropriately calculated. Furthermore, the time series of VpC used for the construction of each prediction model Mvpc and its first derivative ΔVpC [τi] and second derivative Δ ² VpC, the time series of VpC to be predicted and its first derivative ΔVpC [τi ] And the second derivative Δ ² VpC [τi] are calculated.

  In step S24, it is determined whether or not there are prediction models Mc and Mvpc whose fitness exceeds a predetermined threshold. If it exists, it will progress to step S25 and the prediction model pair (Mc, Mvpc) with the highest fitness will be determined.

  The prediction model pair (Mc, Mvpc) may be determined based only on the fitness, irrespective of the unit region in which each prediction model Mc, Mvpc is calculated. In this case, the unit region in which the prediction model Mc is constructed and the unit region in which the prediction model Mvpc is constructed are not necessarily the same.

  Alternatively, the prediction model pair (Mc, Mvpc) is constructed by calculating the fitness of each prediction model Mc, Mvpc for each unit region and constructing the unit region where the representative value such as a simple arithmetic average value or weighted average value has the highest value. ) May be determined. In this case, the constructed unit regions of the respective prediction models Mc and Mvpc are the same, but each of the determined prediction models Mc and Mvpc has the best fitness among all the Mc prediction models or all the Mvpc prediction models, respectively. Not necessarily high.

  On the other hand, if it is determined in step S24 that there is no prediction model Mc or Mvpc whose fitness exceeds the threshold, the process proceeds to step S29. In step S29, the time series prediction model such as logistic curve or Gompertz curve, time series prediction model such as ARIMA, the time series of the number of calls C and VpC, and the first and second derivative values thereof are applied. Future calls C and VpC are predicted uniformly.

In step S27, the predicted traffic total amount calculation unit 4 calculates the predicted traffic total amount Vf by multiplying the predicted call number Cf by the predicted traffic amount VpCf per call. In step S28, the primary differential values ΔC and ΔVpC and the secondary differential values Δ ² C and Δ ² VpC are registered in the traffic characteristic DB 5 and used for new construction of a prediction model or relearning of an existing model.

  According to the present embodiment, since the prediction model is constructed and predicted for each unit area with respect to the traffic volume per number of calls indicating the geographical dependence specific to the communication traffic, the geographical dependence specific to the communication traffic is established. This makes it possible to accurately predict traffic without the need for sex.

  DESCRIPTION OF SYMBOLS 1 ... Communication log DB with position information, 2 ... Traffic characteristic measurement part, 3 ... Traffic tendency analysis part, 4 ... Traffic prediction part, 5 ... Traffic characteristic DB, 6 ... Prediction model construction part, 7 ... Predictive traffic total amount calculation part, 8 ... Communication log collection part with location information

Claims (10)

In a communication traffic prediction device that predicts the traffic volume of a communication terminal,
A traffic characteristic measuring means for measuring traffic characteristics including the number of calls for each unit area obtained by dividing a communication service providing area according to geography;
An analysis means for statistically analyzing the time series of the number of calls at a predetermined period;
Prediction model construction means for constructing a prediction model Mc for predicting the number of calls in the future for each unit area based on the number of calls and the analysis result thereof;
A communication traffic prediction apparatus comprising prediction means for predicting the number of calls in the future by applying the analysis result of the number of calls measured in the prediction area to a prediction model Mc to which a traffic tendency is adapted.   The communication traffic prediction according to claim 1, further comprising a predicted traffic total amount calculation unit that calculates a predicted traffic total amount of the prediction area based on the prediction result of the number of calls and the traffic amount per call. apparatus. The traffic characteristic measuring means further comprises means for measuring the traffic amount per call for each unit area,
The analysis means further comprises means for statistically analyzing a time series of the traffic volume per number of calls at a predetermined period,
The prediction model construction means further comprises means for constructing, for each unit region, a prediction model Mvpc for predicting the traffic volume per call number in the future based on the traffic volume per call number and the analysis result. ,
Means for predicting the traffic volume per call in the future by applying the analysis result of the traffic volume per call measured in the prediction area to a prediction model Mvpc that matches the traffic tendency; The communication traffic prediction apparatus according to claim 1 or 2, further comprising:   4. The communication traffic prediction apparatus according to claim 3, further comprising: a predicted traffic total amount calculation unit that calculates a predicted traffic total amount based on the prediction result of the number of calls and the prediction result of the traffic amount per call. . The prediction means uses a prediction model Mc to which the analysis result of the number of calls is applied and a prediction model Mvpc to which the analysis result of the traffic amount per call is applied based on traffic characteristics measured in the same unit region. 5. The communication traffic prediction apparatus according to claim 3, wherein the prediction model pair is limited to a prediction model pair constructed in the above manner. The prediction means uses a prediction model Mc to which the analysis result of the number of calls is applied and a prediction model Mvpc to which the analysis result of the traffic amount per call is applied based on traffic characteristics measured in the same unit region. 5. The communication traffic prediction apparatus according to claim 3, wherein the prediction model pair is not limited to the prediction model pair constructed in the above manner.   The communication traffic prediction apparatus according to claim 1, wherein the analysis unit calculates a primary differential value and a secondary differential value for each unit time of the number of calls.   The communication traffic prediction apparatus according to claim 3, wherein the analysis unit calculates a primary differential value and a secondary differential value for each unit time of the traffic amount per call. In a communication traffic prediction program for predicting the traffic volume of a communication terminal, a procedure for measuring traffic characteristics including the number of calls for each unit area obtained by dividing a communication service provision area according to geography,
A procedure for statistically analyzing the time series of the number of calls at a predetermined period;
A procedure for building a prediction model Mc for predicting the number of calls in the future based on the number of calls and the analysis result for each unit region;
A communication traffic prediction program that causes a computer to execute a procedure for predicting the number of calls in the future by applying the analysis result of the number of calls measured in the prediction area to a prediction model Mc that matches a traffic tendency. A procedure for measuring the traffic volume per call for each unit area;
A procedure for statistically analyzing a time series of the traffic volume per number of calls at a predetermined period;
A procedure for constructing a prediction model Mvpc for each unit region for predicting a future traffic volume per number of calls based on the traffic volume per call number and an analysis result thereof;
And applying a result of analyzing the traffic volume per call measured in the prediction area to a prediction model Mvpc that matches the traffic tendency to predict the traffic volume per call in the future. The communication traffic prediction program according to claim 9.