WO2024028974A1 - Performance inference model generation device, performance inference device, program, and performance inference model generation method - Google Patents

Performance inference model generation device, performance inference device, program, and performance inference model generation method Download PDF

Info

Publication number
WO2024028974A1
WO2024028974A1 PCT/JP2022/029632 JP2022029632W WO2024028974A1 WO 2024028974 A1 WO2024028974 A1 WO 2024028974A1 JP 2022029632 W JP2022029632 W JP 2022029632W WO 2024028974 A1 WO2024028974 A1 WO 2024028974A1
Authority
WO
WIPO (PCT)
Prior art keywords
performance
estimation model
information
performance estimation
item
Prior art date
Application number
PCT/JP2022/029632
Other languages
French (fr)
Japanese (ja)
Inventor
恭太 服部
智洋 郡川
智香子 高崎
英成 大和田
雅史 清水
Original Assignee
日本電信電話株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 日本電信電話株式会社 filed Critical 日本電信電話株式会社
Priority to PCT/JP2022/029632 priority Critical patent/WO2024028974A1/en
Publication of WO2024028974A1 publication Critical patent/WO2024028974A1/en

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/142—Network analysis or design using statistical or mathematical methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/16—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks using machine learning or artificial intelligence
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/50—Testing arrangements

Definitions

  • the present invention relates to a performance estimation model generation device, a performance estimation device, a program, and a performance estimation model generation method that generate a performance estimation model used for performance estimation of a communication device.
  • a program In conventional network simulation, a program is created that operates in the same way as an actual communication device, and the program is deployed on a simulator to perform the simulation. In such a simulation, a program is prepared, which requires a long development period and causes an increase in costs. Furthermore, when executing a simulation, the program operates on a packet-by-packet basis, so there is a problem in that the time required to complete the simulation increases.
  • the communication device's setting information, input traffic information, and communication data corresponding to the input traffic information are transmitted to the communication device to which the setting information has been applied.
  • a node modeling method has been proposed that predicts the performance of a communication device using a machine learning model that uses a set of performance information of the communication device obtained through measurement as learning data (see Non-Patent Document 1).
  • the present invention has been made in view of this background, and an object of the present invention is to make it possible to improve the estimation accuracy of performance information of a communication device.
  • a performance estimation model generation device is a machine learning model that estimates performance information of a communication device, and uses configuration information of the communication device and input traffic information as explanatory variables. , first generate a performance estimation model that is a machine learning model using the performance information of one item as an objective variable, and generate a performance estimation model that is a machine learning model using the performance information of one item as an objective variable, The model generation unit repeatedly generates a performance estimation model, which is a machine learning model, with performance information as an explanatory variable and performance information of one item different from the input performance information as an objective variable.
  • the explanatory variables of the first performance estimation model included in the performance estimation model and the repeatedly generated performance estimation model are the explanatory variables of the second performance estimation model generated immediately before the first performance estimation model and the explanatory variables of the second performance estimation model that are generated immediately before the first performance estimation model. This is performance information for items that are objective variables of the performance estimation model.
  • FIG. 2 is a diagram for explaining a performance estimation model according to the present embodiment.
  • FIG. 2 is a functional block diagram of a performance estimation model generation device according to the present embodiment.
  • FIG. 2 is a data configuration diagram of a real node information database according to the present embodiment.
  • 7 is a flowchart of performance estimation model generation processing according to the present embodiment. 7 is a flowchart of performance estimation processing according to the present embodiment. It is a functional block diagram of a performance estimating device concerning a modification of this embodiment.
  • FIG. 1 is a hardware configuration diagram showing an example of a computer that implements the functions of a performance estimation model generation device and a performance estimation device according to the present embodiment.
  • the performance estimation model generation device generates a performance estimation model of a communication device that estimates performance information based on setting information and input traffic information of the communication device. Examples of performance information include delay (processing delay), packet loss rate, and throughput.
  • the performance estimation model is not a single model that estimates all performance information (all items) based on configuration information and input traffic information, but multiple performance estimation models that sequentially estimate performance information (performance information items) one by one. This is a model for estimating.
  • FIG. 1 is a diagram for explaining performance estimation models 311 to 314 according to this embodiment. The explanatory variables and objective variables of the performance estimation models 311 to 314 will be explained below with reference to FIG.
  • the explanatory variables of the performance estimation model 311 are setting information 321 and input traffic information 322.
  • the objective variable of the performance estimation model 311 is performance information 331.
  • the explanatory variables of the performance estimation model 312 are setting information 321, input traffic information 322, and performance information 331.
  • the performance information 331 here is performance information 331 calculated using the performance estimation model 311.
  • the objective variable of the performance estimation model 312 is performance information 332.
  • the explanatory variables of the performance estimation model 313 are setting information 321, input traffic information 322, performance information 331, and performance information 332.
  • the performance information 331 here is performance information 331 calculated using the performance estimation model 311.
  • the performance information 332 is calculated using the performance estimation model 312.
  • the objective variable of the performance estimation model 313 is performance information 333.
  • the explanatory variables of the performance estimation model 314 are setting information 321, input traffic information 322, performance information 331, performance information 332, and performance information 333.
  • the performance information 331, 332, 333 is calculated using the performance estimation models 311, 312, 313, respectively.
  • the objective variable of the performance estimation model 314 is performance information 334.
  • the performance information 335 includes performance information 331 to 334.
  • the explanatory variables of the performance estimation model 340 are the setting information 321 and the input traffic information 322, and the objective variables are the performance information 331 to 334 (performance information 335).
  • the performance estimation model 340 the performance information of all items can be estimated based on the setting information 321 and the input traffic information 322.
  • each of the multiple performance estimation models estimates performance information one by one, and the performance information estimated by the previous performance estimation model is used as input (explanatory variable) to estimate the next performance information. This will increase the amount of training data for the model. As a result, the accuracy of performance estimation by the performance estimation model is improved.
  • the performance estimation models 311 to 314 are generated in this order (see FIG. 4).
  • One of the performance estimation models 312 to 314 will be referred to as a first performance estimation model, and the performance estimation model generated immediately before it will be referred to as a second performance estimation model.
  • the second performance estimation model is the performance estimation model 313.
  • the performance estimation model 312 is the first performance estimation model
  • the second performance estimation model is the performance estimation model 311.
  • the explanatory variables of the first performance estimation model are a combination of the explanatory variables of the second performance estimation model and the objective variable.
  • FIG. 2 is a functional block diagram of the performance estimation model generation device 100 according to this embodiment.
  • the performance estimation model generation device 100 is a computer and includes a control section 110, a storage section 120, and an input/output section 180.
  • a measurement device 250 is connected to the performance estimation model generation device 100.
  • the measuring device 250 measures performance information of the communication device 200. To explain in detail, the measurement device 250 sets setting information in the communication device 200, transmits input traffic corresponding to input traffic information to the communication device 200, and acquires and measures performance information. The setting information, input traffic information, and performance information are associated and sent to the performance estimation model generation device 100. Although there is one measuring device 250 in FIG. 2, there may be a plurality of devices that transmit and receive communication data to and from the communication device 200.
  • the input/output unit 180 includes a communication device and is capable of transmitting and receiving data including setting information, input traffic information, and performance information to and from the measuring device 250. Further, a media drive may be connected to the input/output unit 180, so that data can be exchanged using a recording medium.
  • the storage unit 120 includes storage devices such as ROM (Read Only Memory), RAM (Random Access Memory), and SSD (Solid State Drive).
  • the storage unit 120 stores a real node information database 130, a performance estimation model database 140, an estimation order list 121, and a program 128.
  • FIG. 3 is a data configuration diagram of the real node information database 130 according to this embodiment.
  • the real node information database 130 is, for example, tabular data, in which setting information, input traffic information, and performance information are stored in association with each other (as one row (record)) as information related to the communication device 200 of the real device.
  • Ru Examples of setting information include the number of CPUs (Central Processing Units), operating frequency, memory capacity, routing table size, and number of entries in ACL (Access Control List).
  • the input traffic information is traffic information transmitted to the communication device 200 when performance information, which will be described later, is acquired. Examples of performance information include delay, packet loss rate, and throughput. Note that the performance information may include statistical information such as the number of packets transmitted, the number of packets received, and CPU usage rate.
  • the performance estimation model database 140 stores a performance estimation model that is a machine learning model that predicts performance information of the communication device 200. As explained in FIG. 1, the performance estimation model estimates one piece of performance information.
  • the performance estimation model database 140 stores performance information and performance estimation models in association with each other.
  • the estimation order list 121 stores the order in which performance information is estimated. In the case of FIG. 1, the estimation order list 121 includes performance information 331, performance information 332, performance information 333, and performance information 334.
  • the program 128 includes a description of the procedure for performance estimation model generation processing (see FIG. 4, which will be described later).
  • the control unit 110 includes a CPU, and includes an information acquisition unit 111, a model generation unit 112, a performance estimation unit 113, and an input item selection unit 114.
  • the control unit 110 may further include a GPU (Graphics Processing Unit).
  • the information acquisition unit 111 acquires setting information, input traffic information, and performance information of the communication device 200 and stores them in the real node information database 130. To explain in detail, the information acquisition unit 111 acquires performance information of the communication device 200 from the measurement device 250 when a packet corresponding to the input traffic information is transmitted to the communication device 200 configured according to the setting information. , and stored in the real node information database 130.
  • Setting information, input traffic information, and performance information are learning data for a performance estimation model and data for evaluating estimation accuracy, and it is desirable that various setting information and performance information for input traffic be collected and obtained.
  • the model generation unit 112 generates a performance estimation model, which is a machine learning model for predicting performance information of the communication device 200, and stores it in the performance estimation model database 140.
  • Performance estimation unit 113 estimates performance information of communication device 200 using the performance estimation model.
  • the input item selection unit 114 identifies a performance estimation model with the best estimation accuracy among a plurality of performance estimation models having the same explanatory variable and different objective variables. In other words, it can be said that the input item selection unit 114 specifies the performance information (performance information item) with the best estimation accuracy.
  • Detailed processing contents of the model generation unit 112, performance estimation unit 113, and input item selection unit 114 will be described later with reference to FIG.
  • FIG. 4 is a flowchart of performance estimation model generation processing according to this embodiment.
  • the configuration information, input traffic information, and performance information have already been stored in the real node information database 130.
  • step S11 the model generation unit 112 uses the input item list, which is a variable, as setting information and input traffic information.
  • the performance information included in the input item list is also referred to as input performance information.
  • the input performance information is not included in the input item list in step S11, and is added one by one in step S19, which will be described later.
  • step S12 the model generating unit 112 sets the performance item list, which is a variable, as an item of performance information (hereinafter also referred to as a performance item). Examples of performance items include delay, packet loss rate, and throughput. In the following description, performance items are assumed to be delay, packet loss rate, and throughput.
  • step S13 the model generation unit 112 empties the estimated order list 121.
  • step S14 the model generation unit 112 starts a process of repeating steps S15 to S21 until there are no performance items in the performance item list.
  • step S19 in this repeated process, the number of performance items in the performance item list is decreased one by one, and the decreased performance item is added to the input item list.
  • step S15 the model generation unit 112 starts a process of repeating steps S16 to S18 for each performance item in the performance item list.
  • processing target performance items the performance items that are the targets of this repeated processing will be referred to as processing target performance items.
  • step S16 the model generation unit 112 generates a performance estimation model in which the explanatory variables are the input item list and the objective variables are the performance items to be processed.
  • the model generation unit 112 generates learning data in which the explanatory variable is an input item list and the objective variable is a performance item to be processed, based on some records stored in the real node information database 130.
  • a performance estimation model is generated (trained) using this learning data. Some of these records are for learning data generation, and the remaining records are for accuracy evaluation (see step S17, which will be described later).
  • the learning data for the explanatory variables that are performance information are calculated using the registered performance estimation model in step S20, which will be described later. Details will be described later using an example.
  • the performance estimation model generated in step S16 is also referred to as a performance estimation model candidate.
  • step S17 the performance estimation unit 113 calculates the accuracy of the performance estimation model candidate generated in step S16.
  • the performance estimating unit 113 receives the data of the input item list included in the accuracy evaluation record of the real node information database 130 (actual data) and estimates the performance item to be processed using a performance estimation model.
  • the value of the explanatory variable that is performance information is calculated using a registered performance estimation model in step S20, which will be described later. Details will be described later using an example.
  • the performance estimation unit 113 compares the value of the estimation result with the value of the performance item to be processed stored in the real node information database 130 to calculate accuracy.
  • step S18 the model generation unit 112 returns to step S16 and processes the next performance item in the performance item list. If there is no next performance item, the model generation unit 112 proceeds to step S19.
  • step S19 the input item selection unit 114 identifies the performance item to be processed that has the best accuracy among the accuracies calculated in step S17.
  • the input item selection unit 114 deletes the performance item to be processed from the performance item list and adds (moves) it to the input item list. Furthermore, the input item selection unit 114 adds the performance item to be processed to the end of the estimation order list 121.
  • step S20 the model generation unit 112 stores the performance estimation model candidate for estimating the performance item to be processed with the best accuracy identified in step S19 in the performance estimation model database 140 as a performance estimation model for estimating the performance item to be processed. Store.
  • step S21 the model generation unit 112 returns to step S15 and processes the performance item list with one less performance item (see step S19). If the performance item list is empty, the model generation unit 112 ends the performance estimation model generation process.
  • the performance item list in step S12 is assumed to be delay, packet loss rate, and throughput.
  • a performance estimation model in which the explanatory variables are configuration information and input traffic information and the objective variable is delay
  • a performance estimation model in which the explanatory variables are configuration information and input traffic information and the objective variable is packet loss rate.
  • a performance estimation model whose explanatory variables are setting information and input traffic information and whose objective variable is throughput are generated as performance estimation model candidates (see step S16).
  • step S17 accuracy is calculated (see step S17). As a result of comparing the accuracy, it is assumed that the accuracy of the packet loss rate is the best (see step S19). Then, the input item list becomes setting information, input traffic information, and packet loss rate, the performance item list becomes delay and throughput, and the estimated order list 121 becomes packet loss rate.
  • the performance estimation model for estimating the packet loss rate will be referred to as a packet loss rate estimation model.
  • a performance estimation model in which the explanatory variables are setting information, input traffic information, and packet loss rate and the objective variable is delay and a performance estimation model in which the explanatory variables are setting information, input traffic information, and packet loss rate.
  • a performance estimation model whose objective variable is throughput and loss rate is generated as a performance estimation model candidate (see step S16).
  • the packet loss rate included in the learning data as an explanatory variable is calculated based on the setting information and input traffic information using a packet loss rate estimation model.
  • step S17 accuracy is calculated (see step S17).
  • the packet loss rate is used as an explanatory variable when calculating the accuracy, and is calculated based on the setting information and input traffic information using a packet loss rate estimation model.
  • the accuracy of delay is the best (see step S19).
  • the input item list becomes setting information, input traffic information, packet loss rate, and delay
  • the performance item list becomes throughput
  • the estimated order list 121 becomes packet loss rate and delay.
  • a performance estimation model for estimating delay will be referred to as a delay estimation model.
  • a performance estimation model whose explanatory variables are setting information, input traffic information, packet loss rate, and delay and whose objective variable is throughput is generated as a performance estimation model candidate (step (See S16).
  • the packet loss rate included in the learning data as an explanatory variable is calculated based on the setting information and input traffic information using a packet loss rate estimation model.
  • the delay included in the learning data as an explanatory variable is calculated using the delay estimation model based on the setting information, input traffic information, and the packet loss rate calculated using the packet loss rate estimation model.
  • step S17 accuracy is calculated (see step S17). Since there is only one performance estimation model candidate, the throughput accuracy is the best, the input item list consists of setting information, input traffic information, packet loss rate, delay, and throughput, and the performance item list is empty. Furthermore, the estimated order list 121 includes packet loss rate, delay, and throughput. This completes the performance estimation model generation process. In the following, a performance estimation model for estimating throughput will be referred to as a throughput estimation model.
  • FIG. 5 is a flowchart of performance estimation processing according to this embodiment.
  • the process of estimating performance information based on setting information and input traffic information will be described with reference to FIG.
  • the performance estimation unit 113 uses the input data as setting information and input traffic information.
  • the performance estimation unit 113 repeats steps S33 to S35 in the order of the performance items in the estimation order list 121.
  • a performance item that is a target of repeated processing will be referred to as a processing target performance item.
  • step S33 the performance estimation unit 113 estimates the performance item to be processed.
  • the performance estimation unit 113 uses the performance estimation model corresponding to the performance item to be processed stored in the performance estimation model database 140, and uses the data in the input data as an explanatory variable to determine the performance item to be processed which is the objective variable. Estimate.
  • step S34 the performance estimation unit 113 adds the estimation result of step S33 to the input data.
  • step S35 the performance estimation unit 113 returns to step S33 and processes the next performance item in the estimation order list 121. If there is no next performance item, the performance estimation unit 113 ends the performance estimation process.
  • the input data includes the estimation results of all performance information. This estimation result becomes the estimation result of performance information for the setting information and input traffic information.
  • the estimated order list 121 includes packet loss rate, delay, and throughput.
  • the packet loss rate is estimated based on configuration information and input traffic information using a packet loss rate estimation model.
  • a delay performance estimation model is used to estimate the delay based on the configuration information, input traffic information, and the estimated packet loss rate.
  • throughput is estimated using a throughput estimation model based on configuration information, input traffic information, estimated packet loss rate, and estimated delay.
  • the performance estimation model generation device 100 repeatedly generates a performance estimation model for estimating one performance item, starting with explanatory variables from setting information and input traffic information.
  • the performance information estimated by the performance estimation model becomes an explanatory variable for the next performance estimation model.
  • the order in which the performance information is estimated is such that the estimation accuracy of the performance information is the best.
  • the performance estimation model generation device 100 generates a performance estimation model for estimating the performance information for each piece of performance information that has not been estimated (there is no performance estimation model to estimate) (see step S16 in FIG. 4).
  • the performance item with the best accuracy among the performance estimation models thus determined is set as the performance item to be estimated next (see step S19).
  • each of the multiple performance estimation models estimates performance information one by one, and the performance information estimated by the previous performance estimation model is used as input (explanatory variable) to estimate the next performance information. This will increase the amount of training data for the model. As a result, the accuracy of performance estimation by the performance estimation model is improved.
  • Performance estimation model generation process In the performance estimation model generation process (see Figure 4), regarding the order in which performance information is estimated, a performance estimation model is generated to estimate the performance information for each piece of performance information that has not been estimated, and The performance item with the best accuracy is selected as the next performance item to be estimated. On the other hand, performance estimation models may be generated for all estimation orders, and the estimation order that provides the best estimation result may be determined.
  • the performance information is delay, packet loss rate, and throughput.
  • the estimated order is delay ⁇ packet loss rate ⁇ throughput, delay ⁇ throughput ⁇ packet loss rate, packet loss rate ⁇ delay ⁇ throughput, packet loss rate ⁇ throughput ⁇ delay, throughput ⁇ delay ⁇ packet loss rate, and throughput ⁇ packet loss.
  • rate ⁇ delay There are six factors: rate ⁇ delay. Three performance estimation models are generated for each estimation order.
  • the delay, packet loss rate, and throughput are estimated in each estimation order based on the setting information and input traffic information, and the estimation order with the best estimation result is selected.
  • performance information may be weighted and compared. For example, an estimation order that provides the best throughput estimation accuracy may be adopted. By doing so, although the time required for the performance estimation model generation process becomes longer than in the above-described embodiment, it is possible to improve the estimation accuracy of the performance information that is considered important.
  • the estimation order may be changed for each piece of performance information, and estimation may be performed in the estimation order that provides the best estimation accuracy for each piece of performance information.
  • the performance estimation model generation device 100 estimates the performance information of the communication device 200 using a performance estimation model generated by itself, but another device may estimate the performance information.
  • FIG. 6 is a functional block diagram of a performance estimating device 100A according to a modification of this embodiment.
  • the performance estimation device 100A includes a performance estimation model generated by the performance estimation model generation device 100 (see performance estimation model database 140), an estimation order list 121, and a performance estimation unit 113.
  • the performance estimation device 100A estimates performance information based on performance information and input traffic information. The estimation procedure is as explained in FIG. 5.
  • the present invention can take various other embodiments, and furthermore, various changes such as omissions and substitutions can be made without departing from the gist of the present invention. These embodiments and their modifications are included within the scope and gist of the invention described in this specification and the like, as well as within the scope of the invention described in the claims and its equivalents.
  • FIG. 7 is a hardware configuration diagram showing an example of a computer 900 that implements the functions of the performance estimation model generation device 100 and the performance estimation device 100A.
  • the computer 900 includes a CPU 901, a ROM 902, a RAM 903, an SSD 904, an input/output interface 905 (described as an input/output I/F (Interface) in FIG. 7), a communication interface 906 (described as a communication I/F in FIG. 7), and a media interface. 907 (described as media I/F in FIG. 7).
  • the computer 900 may include an HDD (Hard Disc Drive) instead of the SSD 904, or may further include an HDD in addition to the SSD 904.
  • HDD Hard Disc Drive
  • the CPU 901 operates based on a program stored in the ROM 902 or the SSD 904, and is controlled by the control unit 110 in FIG.
  • the ROM 902 stores a boot program executed by the CPU 901 when the computer 900 is started, programs related to the hardware of the computer 900, and the like.
  • the CPU 901 controls an input device 910 such as a mouse and a keyboard, and an output device 911 such as a display and a printer via an input/output interface 905.
  • the CPU 901 obtains data from the input device 910 via the input/output interface 905 and outputs the generated data to the output device 911.
  • the SSD 904 stores programs executed by the CPU 901 and data used by the programs.
  • the communication interface 906 receives data from other devices (not shown) via a communication network and outputs it to the CPU 901, and also transmits data generated by the CPU 901 to other devices via the communication network.
  • Media interface 907 reads a program or data stored in recording medium 912 and outputs it to CPU 901 via RAM 903.
  • the CPU 901 loads a program from the recording medium 912 onto the RAM 903 via the media interface 907, and executes the loaded program.
  • the recording medium 912 is an optical recording medium such as a DVD (Digital Versatile Disk), a magneto-optical recording medium such as an MO (Magneto Optical disk), a magnetic recording medium, a conductive memory tape medium, a semiconductor memory, or the like.
  • the CPU 901 of the computer 900 executes the programs 128, 128A (see FIGS. 3 and 6) loaded on the RAM 903. , realizes the functions of the performance estimation model generation device 100 or the performance estimation device 100A.
  • the CPU 901 reads the program from the recording medium 912 and executes it.
  • the CPU 901 may read a program from another device via a communication network, or may install the program 128 from the recording medium 912 into the SSD 904 and execute it.
  • the performance estimation model generation device 100 is a machine learning model that estimates performance information of the communication device 200, and uses setting information of the communication device 200 and input traffic information as explanatory variables, and one item.
  • the model generation unit 112 is provided that first generates a performance estimation model that is a machine learning model using performance information of as an objective variable.
  • the model generation unit 112 uses setting information, input traffic information, and input performance information that is performance information of one or more items as explanatory variables, and uses performance information of one item different from the input performance information as an objective variable.
  • a performance estimation model which is a machine learning model, is repeatedly generated.
  • the explanatory variables of the first performance estimation model included in the initially generated performance estimation model and the repeatedly generated performance estimation model are the explanatory variables of the second performance estimation model generated immediately before the first performance estimation model.
  • This is performance information of variables and items that are objective variables of the second performance estimation model.
  • the first performance estimation model is the performance estimation model 313 shown in FIG. 1
  • the second performance estimation model is the performance estimation model 312
  • the explanatory variables of the first performance estimation model are the second performance estimation model ( These are setting information 321, input traffic information 322, performance information 331, which are explanatory variables of the performance estimation model 312), and performance information 332, which is the objective variable of the second performance estimation model (performance estimation model 312).
  • a performance estimation model for estimating one performance item is repeatedly generated starting from explanatory variables from setting information and input traffic information.
  • the performance information estimated by the performance estimation model becomes an explanatory variable for the next performance estimation model.
  • the number of learning data for the performance estimation model is increased.
  • the accuracy of performance estimation by the performance estimation model is improved.
  • the performance estimation model generation device 100 further includes a performance estimation section 113 and an input item selection section 114.
  • the model generation unit 112 When generating the first performance estimation model following the second performance estimation model, the model generation unit 112 generates items of performance information of the communication device 200 that are not included in the input performance information of the first performance estimation model.
  • One or more performance estimation models having the performance information of as an objective variable are generated as performance estimation model candidates for the first performance estimation model (see step S16 in FIG. 4).
  • the performance estimation unit 113 uses each of the performance estimation model candidates to determine an objective variable based on the actual data of the communication device 200 corresponding to the setting information, input traffic information, and input performance information of the first performance estimation model. Estimate performance information for items.
  • the performance estimation unit 113 calculates performance estimation based on the performance information of the estimated item and the actual data of the communication device 200 (data stored in the real node information database 130) corresponding to the performance information. , calculate the estimation accuracy of the performance information of the item that is the objective variable of the performance estimation model candidate.
  • the input item selection unit 114 identifies the performance information of the item with the best estimation accuracy among the performance information of the items to be the target variables of the performance estimation model candidate.
  • the model generation unit 112 selects, as the first performance estimation model, a performance estimation model candidate whose objective variable is the performance information of the item with the best estimation accuracy (see step S19).
  • the order of the estimated performance information is such that the estimation accuracy of the performance information is the best.
  • the performance estimation model generation device 100 generates a performance estimation model for estimating the performance information for each piece of performance information that has not been estimated (there is no performance estimation model to estimate) (see step S16 in FIG. 4).
  • the performance item with the best accuracy among the performance estimation models thus determined is set as the performance item to be estimated next (see step S19).
  • the performance estimation device 100A includes a storage unit 120 that stores one or more ordered performance estimation models (see performance estimation model database 140), which are machine learning models for estimating performance information of a communication device, and a performance estimation unit 113. Equipped with.
  • the explanatory variables of the first performance estimation model (see performance estimation model 311 in FIG. 1) in the ordered performance estimation models are setting information of the communication device 200 and input traffic information.
  • the objective variable of the first performance estimation model is first performance information that is performance information of one item.
  • the explanatory variables of the first performance estimation model included in the ordered performance estimation models after the first performance estimation model are the explanatory variables of the second performance estimation model immediately before the first performance estimation model and the explanatory variables of the second performance estimation model that precedes the first performance estimation model.
  • the objective variable of the first performance estimation model is performance information of one item that is not included in the explanatory variables of the first performance estimation model.
  • the performance estimation unit 113 estimates first performance information (see performance information 331) using the first performance estimation model (see performance estimation model 311) based on the setting information and input traffic information. Furthermore, the performance estimation unit 113 uses the first performance estimation model to determine the objective variable based on the explanatory variables of the second performance estimation model and the performance information of the items estimated using the second performance estimation model. Estimate performance information for items. For example, if the first performance estimation model is the performance estimation model 313, the second performance estimation model is the performance estimation model 312.
  • the performance estimation unit 113 includes explanatory variables (setting information 321, input traffic information 322, and performance information 331) of the second performance estimation model (performance estimation model 312) and the second performance estimation model (performance estimation model 312). Based on the performance information (performance information 332) of the item estimated using , the performance information (performance information 333) of the item that is the target variable is estimated using the first performance estimation model (performance estimation model 313).
  • performance information can be estimated using the performance estimation model generated by the performance estimation model generation device 100.
  • Performance estimation model generation device 100A Performance estimation device 110 Control unit 111 Information acquisition unit 112 Model generation unit 113 Performance estimation unit 114 Input item selection unit 120 Storage unit 121 Estimation order list 128 Program 130 Real node information database 140 Performance estimation model database 200 Communication device

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • Algebra (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Physics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Pure & Applied Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

A model generation unit (112) provided in a performance inference model generation device (100) first generates a performance inference model which is a machine learning model for inferring performance information pertaining to a communication device (200) and for which setting information pertaining to the communication device (200) and input traffic information are set as explanatory variables and one item of performance information is set as an objective variable, and then repeatedly generates a performance inference model for which the setting information, the input traffic information, and input performance information that is one or more items of performance information are set as explanatory variables and one item of performance information that is different from the input performance information is set as an objective variable. Explanatory variables for a performance inference model are explanatory variables and an item of performance information, which is an objective variable, for a performance inference model that has been generated immediately prior to that performance inference model.

Description

性能推定モデル生成装置、性能推定装置、プログラムおよび性能推定モデル生成方法Performance estimation model generation device, performance estimation device, program, and performance estimation model generation method
　本発明は、通信装置の性能推定に用いる性能推定モデルを生成する性能推定モデル生成装置、性能推定装置、プログラムおよび性能推定モデル生成方法に関する。 The present invention relates to a performance estimation model generation device, a performance estimation device, a program, and a performance estimation model generation method that generate a performance estimation model used for performance estimation of a communication device.
　従来のネットワークのシミュレーションでは、実機の通信装置と同じ動作をするプログラムを作成し、シミュレータ上に当該プログラムを配備してシミュレーションを行っている。このようなシミュレーションでは、プログラムを用意するので、開発期間を要しコスト増加の原因となっている。また、シミュレーションの実行時には、プログラムがパケット単位に動作することになるので、シミュレーション終了までの時間が増大するという問題がある。 In conventional network simulation, a program is created that operates in the same way as an actual communication device, and the program is deployed on a simulator to perform the simulation. In such a simulation, a program is prepared, which requires a long development period and causes an increase in costs. Furthermore, when executing a simulation, the program operates on a packet-by-packet basis, so there is a problem in that the time required to complete the simulation increases.
　このようなシミュレーションより簡易に通信装置のモデリングを行う手法として、通信装置の設定情報、入力トラヒック情報、および、当該設定情報が施された通信装置に当該入力トラヒック情報に対応する通信データを送信したときに測定して得られる通信装置の性能情報の組を学習データとする機械学習モデルを用いて、通信装置の性能を予測するノードモデリング手法が提案されている（非特許文献１参照）。 As a method for easily modeling a communication device using such simulation, the communication device's setting information, input traffic information, and communication data corresponding to the input traffic information are transmitted to the communication device to which the setting information has been applied. A node modeling method has been proposed that predicts the performance of a communication device using a machine learning model that uses a set of performance information of the communication device obtained through measurement as learning data (see Non-Patent Document 1).
　しかしながら、測定して得られる性能情報項目（メトリクス）によっては、通信装置の設定情報と入力トラヒック情報だけでは、性能を推定するのに十分な学習ができず、推定精度が低い場合があり、推定精度の向上が望まれている。
　本発明は、このような背景に鑑みてなされたものであり、通信装置の性能情報の推定精度向上を可能にすることを課題とする。 However, depending on the performance information items (metrics) obtained through measurement, it may not be possible to learn enough to estimate performance with only communication device setting information and input traffic information, and the estimation accuracy may be low. Improvement in accuracy is desired.
The present invention has been made in view of this background, and an object of the present invention is to make it possible to improve the estimation accuracy of performance information of a communication device.
　前記した課題を解決するため、本発明に係る性能推定モデル生成装置は、通信装置の性能情報を推定する機械学習モデルであって、前記通信装置の設定情報と、入力トラヒック情報とを説明変数とし、１つの項目の前記性能情報を目的変数とする機械学習モデルである性能推定モデルを最初に生成し、前記設定情報と、前記入力トラヒック情報と、１つ以上の項目の前記性能情報である入力性能情報とを説明変数とし、当該入力性能情報とは異なる１つの項目の前記性能情報を目的変数とする機械学習モデルである性能推定モデルを繰り返し生成するモデル生成部を備え、最初に生成された性能推定モデルおよび繰り返し生成された性能推定モデルに含まれる第１性能推定モデルの説明変数は、当該第１性能推定モデルの１つ前に生成された第２性能推定モデルの説明変数および当該第２性能推定モデルの目的変数である項目の性能情報である。 In order to solve the above problems, a performance estimation model generation device according to the present invention is a machine learning model that estimates performance information of a communication device, and uses configuration information of the communication device and input traffic information as explanatory variables. , first generate a performance estimation model that is a machine learning model using the performance information of one item as an objective variable, and generate a performance estimation model that is a machine learning model using the performance information of one item as an objective variable, The model generation unit repeatedly generates a performance estimation model, which is a machine learning model, with performance information as an explanatory variable and performance information of one item different from the input performance information as an objective variable. The explanatory variables of the first performance estimation model included in the performance estimation model and the repeatedly generated performance estimation model are the explanatory variables of the second performance estimation model generated immediately before the first performance estimation model and the explanatory variables of the second performance estimation model that are generated immediately before the first performance estimation model. This is performance information for items that are objective variables of the performance estimation model.
　本発明によれば、通信装置の性能情報の推定精度向上を可能にすることができる。 According to the present invention, it is possible to improve the estimation accuracy of performance information of a communication device.
本実施形態に係る性能推定モデルを説明するための図である。FIG. 2 is a diagram for explaining a performance estimation model according to the present embodiment. 本実施形態に係る性能推定モデル生成装置の機能ブロック図である。FIG. 2 is a functional block diagram of a performance estimation model generation device according to the present embodiment. 本実施形態に係る実ノード情報データベースのデータ構成図である。FIG. 2 is a data configuration diagram of a real node information database according to the present embodiment. 本実施形態に係る性能推定モデル生成処理のフローチャートである。7 is a flowchart of performance estimation model generation processing according to the present embodiment. 本実施形態に係る性能推定処理のフローチャートである。7 is a flowchart of performance estimation processing according to the present embodiment. 本実施形態の変形例に係る性能推定装置の機能ブロック図である。It is a functional block diagram of a performance estimating device concerning a modification of this embodiment. 本実施形態に係る性能推定モデル生成装置や性能推定装置の機能を実現するコンピュータの一例を示すハードウェア構成図である。FIG. 1 is a hardware configuration diagram showing an example of a computer that implements the functions of a performance estimation model generation device and a performance estimation device according to the present embodiment.
≪性能推定モデル生成装置の概要≫
　以下に本発明を実施するための形態（実施形態）における性能推定モデル生成装置を説明する。本実施形態に係る性能推定モデル生成装置は、通信装置の設定情報と入力トラヒック情報とを基に性能情報を推定する、通信装置の性能推定モデルを生成する。性能情報の例として、遅延（処理遅延）やパケットロス率、スループットなどがある。 ≪Overview of performance estimation model generation device≫
A performance estimation model generation device in a mode (embodiment) for carrying out the present invention will be described below. The performance estimation model generation device according to the present embodiment generates a performance estimation model of a communication device that estimates performance information based on setting information and input traffic information of the communication device. Examples of performance information include delay (processing delay), packet loss rate, and throughput.
　性能推定モデルは、設定情報と入力トラヒック情報を基に全て（全項目）の性能情報を推定する１つのモデルではなく、複数の性能推定モデルが順々に１つずつ性能情報（性能情報項目）を推定するモデルとなっている。図１は、本実施形態に係る性能推定モデル３１１～３１４を説明するための図である。以下、図１を参照しながら性能推定モデル３１１～３１４の説明変数と目的変数とを説明する。 The performance estimation model is not a single model that estimates all performance information (all items) based on configuration information and input traffic information, but multiple performance estimation models that sequentially estimate performance information (performance information items) one by one. This is a model for estimating. FIG. 1 is a diagram for explaining performance estimation models 311 to 314 according to this embodiment. The explanatory variables and objective variables of the performance estimation models 311 to 314 will be explained below with reference to FIG.
　性能推定モデル３１１の説明変数は、設定情報３２１、および入力トラヒック情報３２２である。性能推定モデル３１１の目的変数は、性能情報３３１である。
　性能推定モデル３１２の説明変数は、設定情報３２１、入力トラヒック情報３２２、および性能情報３３１である。ここで性能情報３３１は、性能推定モデル３１１を用いて算出された性能情報３３１である。性能推定モデル３１２の目的変数は、性能情報３３２である。 The explanatory variables of the performance estimation model 311 are setting information 321 and input traffic information 322. The objective variable of the performance estimation model 311 is performance information 331.
The explanatory variables of the performance estimation model 312 are setting information 321, input traffic information 322, and performance information 331. The performance information 331 here is performance information 331 calculated using the performance estimation model 311. The objective variable of the performance estimation model 312 is performance information 332.
　性能推定モデル３１３の説明変数は、設定情報３２１、入力トラヒック情報３２２、性能情報３３１、および性能情報３３２である。ここで性能情報３３１は、性能推定モデル３１１を用いて算出された性能情報３３１である。また性能情報３３２は、性能推定モデル３１２を用いて算出された性能情報３３２である。性能推定モデル３１３の目的変数は、性能情報３３３である。 The explanatory variables of the performance estimation model 313 are setting information 321, input traffic information 322, performance information 331, and performance information 332. The performance information 331 here is performance information 331 calculated using the performance estimation model 311. The performance information 332 is calculated using the performance estimation model 312. The objective variable of the performance estimation model 313 is performance information 333.
　性能推定モデル３１４の説明変数は、設定情報３２１、入力トラヒック情報３２２、性能情報３３１、性能情報３３２、および性能情報３３３である。ここで性能情報３３１，３３２，３３３は、それぞれ性能推定モデル３１１，３１２，３１３を用いて算出された性能情報３３１，３３２，３３３である。性能推定モデル３１４の目的変数は、性能情報３３４である。 The explanatory variables of the performance estimation model 314 are setting information 321, input traffic information 322, performance information 331, performance information 332, and performance information 333. Here, the performance information 331, 332, 333 is calculated using the performance estimation models 311, 312, 313, respectively. The objective variable of the performance estimation model 314 is performance information 334.
　性能情報３３５は、性能情報３３１～３３４を含む。性能推定モデル３１１～３１４を１つの性能推定モデル３４０と見なすと、性能推定モデル３４０の説明変数は設定情報３２１、入力トラヒック情報３２２であり、目的変数は性能情報３３１～３３４（性能情報３３５）となる。換言すれば、性能推定モデル３４０を用いることで、設定情報３２１および入力トラヒック情報３２２を基に、全項目の性能情報を推定することができる。 The performance information 335 includes performance information 331 to 334. Considering the performance estimation models 311 to 314 as one performance estimation model 340, the explanatory variables of the performance estimation model 340 are the setting information 321 and the input traffic information 322, and the objective variables are the performance information 331 to 334 (performance information 335). Become. In other words, by using the performance estimation model 340, the performance information of all items can be estimated based on the setting information 321 and the input traffic information 322.
　このように複数の性能推定モデルそれぞれが１つずつ性能情報を推定し、１つ前の性能推定モデルが推定した性能情報を入力（説明変数）として次の性能情報を推定することで、性能推定モデルの学習データ数を増やすことになる。延いては性能推定モデルによる性能推定の精度が向上する。 In this way, each of the multiple performance estimation models estimates performance information one by one, and the performance information estimated by the previous performance estimation model is used as input (explanatory variable) to estimate the next performance information. This will increase the amount of training data for the model. As a result, the accuracy of performance estimation by the performance estimation model is improved.
　なお後記するように、性能推定モデル３１１～３１４は、この順に生成される（図４参照）。性能推定モデル３１２～３１４のなかの１つを第１性能推定モデルと記し、その１つ前に生成された性能推定モデルを第２性能推定モデルと記す。例えば、性能推定モデル３１４が第１性能推定モデルとすると、第２性能推定モデルは性能推定モデル３１３となる。また、性能推定モデル３１２が第１性能推定モデルとすると、第２性能推定モデルは性能推定モデル３１１となる。また第１性能推定モデルの説明変数は、第２性能推定モデルの説明変数と目的変数とを合わせた変数となっている。 As will be described later, the performance estimation models 311 to 314 are generated in this order (see FIG. 4). One of the performance estimation models 312 to 314 will be referred to as a first performance estimation model, and the performance estimation model generated immediately before it will be referred to as a second performance estimation model. For example, if the performance estimation model 314 is the first performance estimation model, the second performance estimation model is the performance estimation model 313. Further, if the performance estimation model 312 is the first performance estimation model, the second performance estimation model is the performance estimation model 311. Further, the explanatory variables of the first performance estimation model are a combination of the explanatory variables of the second performance estimation model and the objective variable.
≪性能推定モデル生成装置の構成≫
　図２は、本実施形態に係る性能推定モデル生成装置１００の機能ブロック図である。性能推定モデル生成装置１００はコンピュータであり、制御部１１０、記憶部１２０、および入出力部１８０を備える。性能推定モデル生成装置１００には、測定装置２５０が接続される。 ≪Configuration of performance estimation model generation device≫
FIG. 2 is a functional block diagram of the performance estimation model generation device 100 according to this embodiment. The performance estimation model generation device 100 is a computer and includes a control section 110, a storage section 120, and an input/output section 180. A measurement device 250 is connected to the performance estimation model generation device 100.
　測定装置２５０は、通信装置２００の性能情報を測定する。詳しく説明すると測定装置２５０は、通信装置２００に設定情報を設定し、入力トラヒック情報に対応する入力トラヒックを通信装置２００に送信して、性能情報を取得して測定する。設定情報、入力トラヒック情報、および性能情報は、関連付けられて性能推定モデル生成装置１００に送られる。図２において測定装置２５０は１つであるが、通信装置２００と通信データを送受信する複数の装置であってもよい。 The measuring device 250 measures performance information of the communication device 200. To explain in detail, the measurement device 250 sets setting information in the communication device 200, transmits input traffic corresponding to input traffic information to the communication device 200, and acquires and measures performance information. The setting information, input traffic information, and performance information are associated and sent to the performance estimation model generation device 100. Although there is one measuring device 250 in FIG. 2, there may be a plurality of devices that transmit and receive communication data to and from the communication device 200.
　入出力部１８０には、ディスプレイやキーボード、マウスなどのユーザインターフェイス機器が接続される。入出力部１８０は通信デバイスを備え、測定装置２５０と設定情報や入力トラヒック情報、性能情報を含むデータの送受信が可能である。また入出力部１８０にメディアドライブが接続され、記録媒体を用いたデータのやり取りが可能であってもよい。 User interface devices such as a display, keyboard, and mouse are connected to the input/output unit 180. The input/output unit 180 includes a communication device and is capable of transmitting and receiving data including setting information, input traffic information, and performance information to and from the measuring device 250. Further, a media drive may be connected to the input/output unit 180, so that data can be exchanged using a recording medium.
≪記憶部≫
　記憶部１２０は、ＲＯＭ（Read Only Memory）やＲＡＭ（Random Access Memory）、ＳＳＤ（Solid State Drive）などの記憶機器を含んで構成される。記憶部１２０には、実ノード情報データベース１３０、性能推定モデルデータベース１４０、推定順序リスト１２１、およびプログラム１２８が記憶される。 ≪Storage section≫
The storage unit 120 includes storage devices such as ROM (Read Only Memory), RAM (Random Access Memory), and SSD (Solid State Drive). The storage unit 120 stores a real node information database 130, a performance estimation model database 140, an estimation order list 121, and a program 128.
　図３は、本実施形態に係る実ノード情報データベース１３０のデータ構成図である。実ノード情報データベース１３０は、例えば表形式のデータであって、実機の通信装置２００に係る情報として設定情報と入力トラヒック情報と性能情報とが関連付けられて（１つの行（レコード）として）格納される。設定情報の例として、ＣＰＵ（Central Processing Unit）の数や動作周波数、メモリ容量、ルーティングテーブルのサイズ、ＡＣＬ（Access Control List）のエントリ数などがある。入力トラヒック情報は、後記する性能情報を取得したときにおける通信装置２００へ送信されたトラヒック情報である。性能情報の例として、遅延やパケットロス率、スループットなどがある。なお性能情報として、パケット送信数やパケット受信数、ＣＰＵ使用率などの統計情報を含めてもよい。 FIG. 3 is a data configuration diagram of the real node information database 130 according to this embodiment. The real node information database 130 is, for example, tabular data, in which setting information, input traffic information, and performance information are stored in association with each other (as one row (record)) as information related to the communication device 200 of the real device. Ru. Examples of setting information include the number of CPUs (Central Processing Units), operating frequency, memory capacity, routing table size, and number of entries in ACL (Access Control List). The input traffic information is traffic information transmitted to the communication device 200 when performance information, which will be described later, is acquired. Examples of performance information include delay, packet loss rate, and throughput. Note that the performance information may include statistical information such as the number of packets transmitted, the number of packets received, and CPU usage rate.
　性能推定モデルデータベース１４０には、通信装置２００の性能情報を予測する機械学習モデルである性能推定モデルが格納される。図１で説明したように、性能推定モデルは１つの性能情報を推定する。性能推定モデルデータベース１４０には、性能情報と性能推定モデルとが関連付けられて記憶される。
　推定順序リスト１２１には、性能情報が推定される順序が記憶される。図１の場合では、推定順序リスト１２１は、性能情報３３１、性能情報３３２、性能情報３３３、性能情報３３４となる。プログラム１２８は、性能推定モデル生成処理（後記する図４参照）の手順の記述を含む。 The performance estimation model database 140 stores a performance estimation model that is a machine learning model that predicts performance information of the communication device 200. As explained in FIG. 1, the performance estimation model estimates one piece of performance information. The performance estimation model database 140 stores performance information and performance estimation models in association with each other.
The estimation order list 121 stores the order in which performance information is estimated. In the case of FIG. 1, the estimation order list 121 includes performance information 331, performance information 332, performance information 333, and performance information 334. The program 128 includes a description of the procedure for performance estimation model generation processing (see FIG. 4, which will be described later).
≪制御部≫
　制御部１１０は、ＣＰＵを含んで構成され、情報取得部１１１、モデル生成部１１２、性能推定部１１３、および入力項目選定部１１４が備わる。制御部１１０は、さらにＧＰＵ（Graphics Processing Unit）を含んで構成されてもよい。情報取得部１１１は、通信装置２００の設定情報や入力トラヒック情報、性能情報を取得して実ノード情報データベース１３０に格納する。詳しく説明すると情報取得部１１１は、設定情報に対応する設定がされた通信装置２００に、入力トラヒック情報に対応するパケットを送信したときの通信装置２００の性能情報を、測定装置２５０から取得して、実ノード情報データベース１３０に格納する。設定情報、入力トラヒック情報、および性能情報は性能推定モデルの学習データや推定精度評価用のデータであって、様々な設定情報、入力トラヒックにおける性能情報が収集され、取得されることが望ましい。 ≪Control unit≫
The control unit 110 includes a CPU, and includes an information acquisition unit 111, a model generation unit 112, a performance estimation unit 113, and an input item selection unit 114. The control unit 110 may further include a GPU (Graphics Processing Unit). The information acquisition unit 111 acquires setting information, input traffic information, and performance information of the communication device 200 and stores them in the real node information database 130. To explain in detail, the information acquisition unit 111 acquires performance information of the communication device 200 from the measurement device 250 when a packet corresponding to the input traffic information is transmitted to the communication device 200 configured according to the setting information. , and stored in the real node information database 130. Setting information, input traffic information, and performance information are learning data for a performance estimation model and data for evaluating estimation accuracy, and it is desirable that various setting information and performance information for input traffic be collected and obtained.
　モデル生成部１１２は、通信装置２００の性能情報を予測する機械学習モデルである性能推定モデルを生成して、性能推定モデルデータベース１４０に格納する。
　性能推定部１１３は、性能推定モデルを用いて、通信装置２００の性能情報を推定する。
　入力項目選定部１１４は、説明変数が同じであり目的変数が異なる複数の性能推定モデルのなかで、推定精度が最良となる性能推定モデルを特定する。換言すれば入力項目選定部１１４は、推定精度が最良となる性能情報（性能情報項目）を特定するともいえる。
　モデル生成部１１２、性能推定部１１３、および入力項目選定部１１４の詳しい処理内容は、図４を参照しながら後記する。 The model generation unit 112 generates a performance estimation model, which is a machine learning model for predicting performance information of the communication device 200, and stores it in the performance estimation model database 140.
Performance estimation unit 113 estimates performance information of communication device 200 using the performance estimation model.
The input item selection unit 114 identifies a performance estimation model with the best estimation accuracy among a plurality of performance estimation models having the same explanatory variable and different objective variables. In other words, it can be said that the input item selection unit 114 specifies the performance information (performance information item) with the best estimation accuracy.
Detailed processing contents of the model generation unit 112, performance estimation unit 113, and input item selection unit 114 will be described later with reference to FIG.
≪性能推定モデル生成処理≫
　図４は、本実施形態に係る性能推定モデル生成処理のフローチャートである。性能推定モデル生成処理の開始時点において、設定情報、入力トラヒック情報、および性能情報は実ノード情報データベース１３０に格納済みである。 ≪Performance estimation model generation process≫
FIG. 4 is a flowchart of performance estimation model generation processing according to this embodiment. At the start of the performance estimation model generation process, the configuration information, input traffic information, and performance information have already been stored in the real node information database 130.
　ステップＳ１１においてモデル生成部１１２は、変数である入力項目リストを設定情報、入力トラヒック情報とする。入力項目リストに含まれる性能情報を入力性能情報とも記す。ステップＳ１１において入力項目リストに入力性能情報は含まれず、後記するステップＳ１９において１つずつ追加される。
　ステップＳ１２においてモデル生成部１１２は、変数である性能項目リストを性能情報の項目（以下、性能項目とも記す）とする。性能項目としては、例えば遅延、パケットロス率、スループットなどがある。以下の説明において性能項目は、遅延、パケットロス率およびスループットであるとする。
　ステップＳ１３においてモデル生成部１１２は、推定順序リスト１２１を空にする。 In step S11, the model generation unit 112 uses the input item list, which is a variable, as setting information and input traffic information. The performance information included in the input item list is also referred to as input performance information. The input performance information is not included in the input item list in step S11, and is added one by one in step S19, which will be described later.
In step S12, the model generating unit 112 sets the performance item list, which is a variable, as an item of performance information (hereinafter also referred to as a performance item). Examples of performance items include delay, packet loss rate, and throughput. In the following description, performance items are assumed to be delay, packet loss rate, and throughput.
In step S13, the model generation unit 112 empties the estimated order list 121.
　ステップＳ１４においてモデル生成部１１２は、性能項目リストにある性能項目がなくなるまでステップＳ１５～Ｓ２１を繰り返す処理を開始する。なおステップＳ１９で説明するように、この繰り返す処理において、性能項目リストにある性能項目は１つずつ減り、減った性能項目が入力項目リストに加わる。
　ステップＳ１５においてモデル生成部１１２は、性能項目リストにある性能項目ごとにステップＳ１６～Ｓ１８を繰り返す処理を開始する。以下、この繰り返す処理の対象となる性能項目を処理対象性能項目と記す。 In step S14, the model generation unit 112 starts a process of repeating steps S15 to S21 until there are no performance items in the performance item list. As described in step S19, in this repeated process, the number of performance items in the performance item list is decreased one by one, and the decreased performance item is added to the input item list.
In step S15, the model generation unit 112 starts a process of repeating steps S16 to S18 for each performance item in the performance item list. Hereinafter, the performance items that are the targets of this repeated processing will be referred to as processing target performance items.
　ステップＳ１６においてモデル生成部１１２は、説明変数が入力項目リスト、目的変数が処理対象性能項目である性能推定モデルを生成する。詳しく説明するとモデル生成部１１２は、実ノード情報データベース１３０に格納される一部のレコードを基に、説明変数が入力項目リストであり目的変数が処理対象性能項目である学習データを生成して、この学習データを用いて性能推定モデルを生成（訓練）する。この一部のレコードは学習データ生成用のレコードであって、残りのレコードは精度評価用のレコードとなる（後記するステップＳ１７参照）。
　ここで入力項目リストにある説明変数のなかで性能情報である説明変数の学習データは、後記するステップＳ２０で登録済みの性能推定モデルを用いて算出される。詳細は例を用いて後記する。このステップＳ１６で生成される性能推定モデルを性能推定モデル候補とも記す。 In step S16, the model generation unit 112 generates a performance estimation model in which the explanatory variables are the input item list and the objective variables are the performance items to be processed. To explain in detail, the model generation unit 112 generates learning data in which the explanatory variable is an input item list and the objective variable is a performance item to be processed, based on some records stored in the real node information database 130. A performance estimation model is generated (trained) using this learning data. Some of these records are for learning data generation, and the remaining records are for accuracy evaluation (see step S17, which will be described later).
Among the explanatory variables in the input item list, the learning data for the explanatory variables that are performance information are calculated using the registered performance estimation model in step S20, which will be described later. Details will be described later using an example. The performance estimation model generated in step S16 is also referred to as a performance estimation model candidate.
　ステップＳ１７において性能推定部１１３は、ステップＳ１６で生成された性能推定モデル候補の精度を算出する。詳しく説明すると性能推定部１１３は、実ノード情報データベース１３０（実データ）の精度評価用のレコードに含まれる入力項目リストのデータを入力として、性能推定モデルを用いて処理対象性能項目を推定する。ここで入力項目リストにある説明変数のなかで性能情報である説明変数の値は、後記するステップＳ２０で登録済みの性能推定モデルを用いて算出される。詳細は例を用いて後記する。次に性能推定部１１３は、推定結果の値と、実ノード情報データベース１３０に格納される処理対象性能項目の値とを比較して、精度を算出する。
　ステップＳ１８においてモデル生成部１１２は、ステップＳ１６に戻って性能項目リストにある次の性能項目を処理する。次の性能項目がない場合にモデル生成部１１２は、ステップＳ１９に進む。 In step S17, the performance estimation unit 113 calculates the accuracy of the performance estimation model candidate generated in step S16. To explain in detail, the performance estimating unit 113 receives the data of the input item list included in the accuracy evaluation record of the real node information database 130 (actual data) and estimates the performance item to be processed using a performance estimation model. Here, among the explanatory variables in the input item list, the value of the explanatory variable that is performance information is calculated using a registered performance estimation model in step S20, which will be described later. Details will be described later using an example. Next, the performance estimation unit 113 compares the value of the estimation result with the value of the performance item to be processed stored in the real node information database 130 to calculate accuracy.
In step S18, the model generation unit 112 returns to step S16 and processes the next performance item in the performance item list. If there is no next performance item, the model generation unit 112 proceeds to step S19.
　ステップＳ１９において入力項目選定部１１４は、ステップＳ１７で算出した精度のなかで精度が最良となる処理対象性能項目を特定する。次に入力項目選定部１１４は、性能項目リストから当該処理対象性能項目を削除し、入力項目リストに加える（移す）。さらに入力項目選定部１１４は、当該処理対象性能項目を推定順序リスト１２１の最後に追加する。
　ステップＳ２０においてモデル生成部１１２は、ステップＳ１９で特定した精度が最良となる処理対象性能項目を推定する性能推定モデル候補を、当該処理対象性能項目を推定する性能推定モデルとして性能推定モデルデータベース１４０に格納する。
　ステップＳ２１においてモデル生成部１１２は、ステップＳ１５に戻って、性能項目が１つ減った（ステップＳ１９参照）性能項目リストに対して処理する。性能項目リストが空ならばモデル生成部１１２は、性能推定モデル生成処理を終える。 In step S19, the input item selection unit 114 identifies the performance item to be processed that has the best accuracy among the accuracies calculated in step S17. Next, the input item selection unit 114 deletes the performance item to be processed from the performance item list and adds (moves) it to the input item list. Furthermore, the input item selection unit 114 adds the performance item to be processed to the end of the estimation order list 121.
In step S20, the model generation unit 112 stores the performance estimation model candidate for estimating the performance item to be processed with the best accuracy identified in step S19 in the performance estimation model database 140 as a performance estimation model for estimating the performance item to be processed. Store.
In step S21, the model generation unit 112 returns to step S15 and processes the performance item list with one less performance item (see step S19). If the performance item list is empty, the model generation unit 112 ends the performance estimation model generation process.
≪性能推定モデル生成処理の例≫
　以下に処理の例を説明する。ステップＳ１２における性能項目リストを遅延、パケットロス率およびスループットとする。最初のステップＳ１５～Ｓ２１の繰り返し処理において、説明変数が設定情報および入力トラヒック情報であり目的変数が遅延となる性能推定モデル、説明変数が設定情報および入力トラヒック情報であり目的変数がパケットロス率となる性能推定モデル、並びに、説明変数が設定情報および入力トラヒック情報であり目的変数がスループットとなる性能推定モデルが、性能推定モデル候補として生成される（ステップＳ１６参照）。 ≪Example of performance estimation model generation process≫
An example of processing will be explained below. The performance item list in step S12 is assumed to be delay, packet loss rate, and throughput. In the first iterative process of steps S15 to S21, a performance estimation model in which the explanatory variables are configuration information and input traffic information and the objective variable is delay, and a performance estimation model in which the explanatory variables are configuration information and input traffic information and the objective variable is packet loss rate. and a performance estimation model whose explanatory variables are setting information and input traffic information and whose objective variable is throughput are generated as performance estimation model candidates (see step S16).
　次に精度が算出される（ステップＳ１７参照）。精度を比較した結果、パケットロス率の精度が最良だったとする（ステップＳ１９参照）。すると、入力項目リストは設定情報、入力トラヒック情報およびパケットロス率になり、性能項目リストは遅延およびスループットとなって、推定順序リスト１２１はパケットロス率となる。以下では、パケットロス率を推定する性能推定モデルを、パケットロス率推定モデルと記す。 Next, accuracy is calculated (see step S17). As a result of comparing the accuracy, it is assumed that the accuracy of the packet loss rate is the best (see step S19). Then, the input item list becomes setting information, input traffic information, and packet loss rate, the performance item list becomes delay and throughput, and the estimated order list 121 becomes packet loss rate. In the following, the performance estimation model for estimating the packet loss rate will be referred to as a packet loss rate estimation model.
　２回目のステップＳ１５～Ｓ２１の繰り返し処理において、説明変数が設定情報、入力トラヒック情報およびパケットロス率であり目的変数が遅延となる性能推定モデル、並びに、説明変数が設定情報、入力トラヒック情報およびパケットロス率であり目的変数がスループットとなる性能推定モデルが、性能推定モデル候補として生成される（ステップＳ１６参照）。ここで学習データに説明変数として含まれるパケットロス率は、パケットロス率推定モデルを用いて、設定情報および入力トラヒック情報を基に算出される。 In the second iteration of steps S15 to S21, a performance estimation model in which the explanatory variables are setting information, input traffic information, and packet loss rate and the objective variable is delay, and a performance estimation model in which the explanatory variables are setting information, input traffic information, and packet loss rate. A performance estimation model whose objective variable is throughput and loss rate is generated as a performance estimation model candidate (see step S16). Here, the packet loss rate included in the learning data as an explanatory variable is calculated based on the setting information and input traffic information using a packet loss rate estimation model.
　次に精度が算出される（ステップＳ１７参照）。ここで精度が算出されるときの説明変数としてパケットロス率は、パケットロス率推定モデルを用いて、設定情報および入力トラヒック情報を基に算出される。精度を比較した結果、遅延の精度が最良だったとする（ステップＳ１９参照）。すると、入力項目リストは設定情報、入力トラヒック情報、パケットロス率および遅延になり、性能項目リストはスループットとなって、推定順序リスト１２１はパケットロス率、遅延となる。以下では、遅延を推定する性能推定モデルを、遅延推定モデルと記す。 Next, accuracy is calculated (see step S17). Here, the packet loss rate is used as an explanatory variable when calculating the accuracy, and is calculated based on the setting information and input traffic information using a packet loss rate estimation model. As a result of comparing the accuracy, it is assumed that the accuracy of delay is the best (see step S19). Then, the input item list becomes setting information, input traffic information, packet loss rate, and delay, the performance item list becomes throughput, and the estimated order list 121 becomes packet loss rate and delay. In the following, a performance estimation model for estimating delay will be referred to as a delay estimation model.
　３回目のステップＳ１５～Ｓ２１の繰り返し処理において、説明変数が設定情報、入力トラヒック情報、パケットロス率および遅延であり目的変数がスループットとなる性能推定モデルが、性能推定モデル候補として生成される（ステップＳ１６参照）。ここで学習データに説明変数として含まれるパケットロス率は、パケットロス率推定モデルを用いて、設定情報および入力トラヒック情報を基に算出される。また学習データに説明変数として含まれる遅延は、遅延推定モデルを用いて、設定情報、入力トラヒック情報、およびパケットロス率推定モデルを用いて算出されたパケットロス率を基に算出される。 In the third iteration of steps S15 to S21, a performance estimation model whose explanatory variables are setting information, input traffic information, packet loss rate, and delay and whose objective variable is throughput is generated as a performance estimation model candidate (step (See S16). Here, the packet loss rate included in the learning data as an explanatory variable is calculated based on the setting information and input traffic information using a packet loss rate estimation model. Further, the delay included in the learning data as an explanatory variable is calculated using the delay estimation model based on the setting information, input traffic information, and the packet loss rate calculated using the packet loss rate estimation model.
　次に精度が算出される（ステップＳ１７参照）。性能推定モデル候補は１つしかないので、スループットの精度が最良となり、入力項目リストは設定情報、入力トラヒック情報、パケットロス率、遅延およびスループットになって、性能項目リストは空となる。また推定順序リスト１２１はパケットロス率、遅延、スループットとなる。これで性能推定モデル生成処理が終了する。以下では、スループットを推定する性能推定モデルを、スループット推定モデルと記す。 Next, accuracy is calculated (see step S17). Since there is only one performance estimation model candidate, the throughput accuracy is the best, the input item list consists of setting information, input traffic information, packet loss rate, delay, and throughput, and the performance item list is empty. Furthermore, the estimated order list 121 includes packet loss rate, delay, and throughput. This completes the performance estimation model generation process. In the following, a performance estimation model for estimating throughput will be referred to as a throughput estimation model.
≪性能推定処理≫
　図５は、本実施形態に係る性能推定処理のフローチャートである。図５を参照しながら、設定情報と入力トラヒック情報を基に性能情報を推定する処理を説明する。
　ステップＳ３１において性能推定部１１３は、入力データを設定情報および入力トラヒック情報とする。
　ステップＳ３２において性能推定部１１３は、推定順序リスト１２１にある性能項目の順にステップＳ３３～Ｓ３５を繰り返す。以下では繰り返し処理の対象となる性能項目を処理対象性能項目と記す。 ≪Performance estimation processing≫
FIG. 5 is a flowchart of performance estimation processing according to this embodiment. The process of estimating performance information based on setting information and input traffic information will be described with reference to FIG.
In step S31, the performance estimation unit 113 uses the input data as setting information and input traffic information.
In step S32, the performance estimation unit 113 repeats steps S33 to S35 in the order of the performance items in the estimation order list 121. In the following, a performance item that is a target of repeated processing will be referred to as a processing target performance item.
　ステップＳ３３において性能推定部１１３は、処理対象性能項目を推定する。詳しく説明すると性能推定部１１３は、性能推定モデルデータベース１４０に記憶される処理対象性能項目に対応する性能推定モデルを用いて、入力データにあるデータを説明変数として、目的変数である処理対象性能項目を推定する。
　ステップＳ３４において性能推定部１１３は、ステップＳ３３の推定結果を入力データに追加する。 In step S33, the performance estimation unit 113 estimates the performance item to be processed. To explain in detail, the performance estimation unit 113 uses the performance estimation model corresponding to the performance item to be processed stored in the performance estimation model database 140, and uses the data in the input data as an explanatory variable to determine the performance item to be processed which is the objective variable. Estimate.
In step S34, the performance estimation unit 113 adds the estimation result of step S33 to the input data.
　ステップＳ３５において性能推定部１１３は、ステップＳ３３に戻って推定順序リスト１２１にある次の性能項目を処理する。次の性能項目がない場合に性能推定部１１３は、性能推定処理を終える。
　性能推定処理が終了した時点において、入力データには全ての性能情報の推定結果が含まれる。この推定結果が、設定情報と入力トラヒック情報に対する性能情報の推定結果となる。 In step S35, the performance estimation unit 113 returns to step S33 and processes the next performance item in the estimation order list 121. If there is no next performance item, the performance estimation unit 113 ends the performance estimation process.
At the time when the performance estimation process is completed, the input data includes the estimation results of all performance information. This estimation result becomes the estimation result of performance information for the setting information and input traffic information.
≪性能推定処理の例≫
　以下に処理の例を説明する。推定順序リスト１２１はパケットロス率、遅延、スループットとする。最初に、パケットロス率推定モデルを用いて設定情報および入力トラヒック情報を基にパケットロス率が推定される。次に遅延性能推定モデルを用いて設定情報、入力トラヒック情報、および推定されたパケットロス率を基に遅延が推定される。最後にスループット推定モデルを用いて設定情報、入力トラヒック情報、推定されたパケットロス率、および推定された遅延を基にスループットが推定される。 ≪Example of performance estimation processing≫
An example of processing will be explained below. The estimated order list 121 includes packet loss rate, delay, and throughput. First, the packet loss rate is estimated based on configuration information and input traffic information using a packet loss rate estimation model. Next, a delay performance estimation model is used to estimate the delay based on the configuration information, input traffic information, and the estimated packet loss rate. Finally, throughput is estimated using a throughput estimation model based on configuration information, input traffic information, estimated packet loss rate, and estimated delay.
≪性能推定モデル生成装置の特徴≫
　以上に説明したように性能推定モデル生成装置１００は、説明変数を設定情報および入力トラヒック情報から始めて、１つの性能項目を推定する性能推定モデルを繰り返し生成する。性能推定モデルが推定した性能情報は、次の性能推定モデルの説明変数となる。推定される性能情報の順序は、性能情報の推定精度が最良になるような順序である。詳しく説明すると性能推定モデル生成装置１００は、推定されていない（推定する性能推定モデルがない）性能情報それぞれについて当該性能情報を推定する性能推定モデルを生成し（図４のステップＳ１６参照）、生成された性能推定モデルのなかで精度が最良となる性能項目を次に推定する性能項目とする（ステップＳ１９参照）。 ≪Characteristics of performance estimation model generation device≫
As described above, the performance estimation model generation device 100 repeatedly generates a performance estimation model for estimating one performance item, starting with explanatory variables from setting information and input traffic information. The performance information estimated by the performance estimation model becomes an explanatory variable for the next performance estimation model. The order in which the performance information is estimated is such that the estimation accuracy of the performance information is the best. To explain in detail, the performance estimation model generation device 100 generates a performance estimation model for estimating the performance information for each piece of performance information that has not been estimated (there is no performance estimation model to estimate) (see step S16 in FIG. 4). The performance item with the best accuracy among the performance estimation models thus determined is set as the performance item to be estimated next (see step S19).
　このように複数の性能推定モデルそれぞれが１つずつ性能情報を推定し、１つ前の性能推定モデルが推定した性能情報を入力（説明変数）として次の性能情報を推定することで、性能推定モデルの学習データ数を増やすことになる。延いては性能推定モデルによる性能推定の精度が向上する。 In this way, each of the multiple performance estimation models estimates performance information one by one, and the performance information estimated by the previous performance estimation model is used as input (explanatory variable) to estimate the next performance information. This will increase the amount of training data for the model. As a result, the accuracy of performance estimation by the performance estimation model is improved.
≪変形例：性能推定モデル生成処理≫
　性能推定モデル生成処理（図４参照）では、性能情報を推定する順序について、推定されていない性能情報それぞれについて当該性能情報を推定する性能推定モデルを生成して、生成された性能推定モデルのなかで精度が最良となる性能項目を次に推定する性能項目としている。これに対して全ての推定順序について、それぞれの性能推定モデルを生成して、推定結果が最良となる推定順序を求めるようにしてもよい。 ≪Modified example: Performance estimation model generation process≫
In the performance estimation model generation process (see Figure 4), regarding the order in which performance information is estimated, a performance estimation model is generated to estimate the performance information for each piece of performance information that has not been estimated, and The performance item with the best accuracy is selected as the next performance item to be estimated. On the other hand, performance estimation models may be generated for all estimation orders, and the estimation order that provides the best estimation result may be determined.
　例えば性能情報が、遅延、パケットロス率およびスループットとする。すると推定順序は、遅延→パケットロス率→スループット、遅延→スループット→パケットロス率、パケットロス率→遅延→スループット、パケットロス率→スループット→遅延、スループット→遅延→パケットロス率、およびスループット→パケットロス率→遅延の６つとなる。このそれぞれの推定順序について３つの性能推定モデルを生成する。 For example, assume that the performance information is delay, packet loss rate, and throughput. Then, the estimated order is delay → packet loss rate → throughput, delay → throughput → packet loss rate, packet loss rate → delay → throughput, packet loss rate → throughput → delay, throughput → delay → packet loss rate, and throughput → packet loss. There are six factors: rate → delay. Three performance estimation models are generated for each estimation order.
　次に設定情報と入力トラヒック情報を基にそれぞれの推定順序で遅延、パケットロス率およびスループットを推定して、推定結果が最良となる推定順序を選択する。推定結果を比較する際に、性能情報に重み付けを付与して比較するようにしてもよい。例えばスループットの推定精度が最良となる推定順序を採用するようにしてもよい。このようにすることで、上記する実施形態と比較して性能推定モデル生成処理の時間は長くなるが、重要視する性能情報の推定精度を上げることができる。または、性能情報ごとに推定順序を変えて、性能情報ごとに推定精度が最良となる推定順序で推定するようにしてもよい。 Next, the delay, packet loss rate, and throughput are estimated in each estimation order based on the setting information and input traffic information, and the estimation order with the best estimation result is selected. When comparing the estimation results, performance information may be weighted and compared. For example, an estimation order that provides the best throughput estimation accuracy may be adopted. By doing so, although the time required for the performance estimation model generation process becomes longer than in the above-described embodiment, it is possible to improve the estimation accuracy of the performance information that is considered important. Alternatively, the estimation order may be changed for each piece of performance information, and estimation may be performed in the estimation order that provides the best estimation accuracy for each piece of performance information.
≪その他の変形例≫
　以上、本発明のいくつかの実施形態について説明したが、これらの実施形態は、例示に過ぎず、本発明の技術的範囲を限定するものではない。例えば、性能推定モデル生成装置１００は、自身が生成した性能推定モデルを用いて通信装置２００の性能情報を推定するが、別の装置が推定してもよい。図６は、本実施形態の変形例に係る性能推定装置１００Ａの機能ブロック図である。性能推定装置１００Ａは、性能推定モデル生成装置１００が生成した性能推定モデル（性能推定モデルデータベース１４０参照）、推定順序リスト１２１、および性能推定部１１３を備える。性能推定装置１００Ａは、性能情報と入力トラヒック情報とを基に性能情報を推定する。推定の手順は、図５で説明したとおりである。 ≪Other variations≫
Although several embodiments of the present invention have been described above, these embodiments are merely illustrative and do not limit the technical scope of the present invention. For example, the performance estimation model generation device 100 estimates the performance information of the communication device 200 using a performance estimation model generated by itself, but another device may estimate the performance information. FIG. 6 is a functional block diagram of a performance estimating device 100A according to a modification of this embodiment. The performance estimation device 100A includes a performance estimation model generated by the performance estimation model generation device 100 (see performance estimation model database 140), an estimation order list 121, and a performance estimation unit 113. The performance estimation device 100A estimates performance information based on performance information and input traffic information. The estimation procedure is as explained in FIG. 5.
　本発明はその他の様々な実施形態を取ることが可能であり、さらに、本発明の要旨を逸脱しない範囲で、省略や置換等種々の変更を行うことができる。これら実施形態やその変形は、本明細書等に記載された発明の範囲や要旨に含まれるとともに、特許請求の範囲に記載された発明とその均等の範囲に含まれる。 The present invention can take various other embodiments, and furthermore, various changes such as omissions and substitutions can be made without departing from the gist of the present invention. These embodiments and their modifications are included within the scope and gist of the invention described in this specification and the like, as well as within the scope of the invention described in the claims and its equivalents.
≪ハードウェア構成≫
　上記した実施形態に係る性能推定モデル生成装置１００、および変形例に係る性能推定装置１００Ａは、例えば図７に示すような構成のコンピュータ９００によって実現される。図７は、性能推定モデル生成装置１００や性能推定装置１００Ａの機能を実現するコンピュータ９００の一例を示すハードウェア構成図である。コンピュータ９００は、ＣＰＵ９０１、ＲＯＭ９０２、ＲＡＭ９０３、ＳＳＤ９０４、入出力インターフェイス９０５（図７では入出力Ｉ／Ｆ（Interface）と記載）、通信インターフェイス９０６（図７では通信Ｉ／Ｆと記載）、およびメディアインターフェイス９０７（図７ではメディアＩ／Ｆと記載）を備える。コンピュータ９００は、ＳＳＤ９０４に替わりにＨＤＤ（Hard Disc Drive）を備えてもよいし、ＳＳＤ９０４に加えて、さらにＨＤＤを備えてもよい。 ≪Hardware configuration≫
The performance estimation model generation device 100 according to the embodiment described above and the performance estimation device 100A according to the modified example are realized by, for example, a computer 900 having a configuration as shown in FIG. 7. FIG. 7 is a hardware configuration diagram showing an example of a computer 900 that implements the functions of the performance estimation model generation device 100 and the performance estimation device 100A. The computer 900 includes a CPU 901, a ROM 902, a RAM 903, an SSD 904, an input/output interface 905 (described as an input/output I/F (Interface) in FIG. 7), a communication interface 906 (described as a communication I/F in FIG. 7), and a media interface. 907 (described as media I/F in FIG. 7). The computer 900 may include an HDD (Hard Disc Drive) instead of the SSD 904, or may further include an HDD in addition to the SSD 904.
　ＣＰＵ９０１は、ＲＯＭ９０２またはＳＳＤ９０４に記憶されたプログラムに基づいて作動し、図２の制御部１１０による制御を行う。ＲＯＭ９０２は、コンピュータ９００の起動時にＣＰＵ９０１により実行されるブートプログラムや、コンピュータ９００のハードウェアに係るプログラムなどを記憶する。
　ＣＰＵ９０１は、入出力インターフェイス９０５を介して、マウスやキーボードなどの入力装置９１０、およびディスプレイやプリンタなどの出力装置９１１を制御する。ＣＰＵ９０１は、入出力インターフェイス９０５を介して、入力装置９１０からデータを取得するとともに、生成したデータを出力装置９１１へ出力する。 The CPU 901 operates based on a program stored in the ROM 902 or the SSD 904, and is controlled by the control unit 110 in FIG. The ROM 902 stores a boot program executed by the CPU 901 when the computer 900 is started, programs related to the hardware of the computer 900, and the like.
The CPU 901 controls an input device 910 such as a mouse and a keyboard, and an output device 911 such as a display and a printer via an input/output interface 905. The CPU 901 obtains data from the input device 910 via the input/output interface 905 and outputs the generated data to the output device 911.
　ＳＳＤ９０４は、ＣＰＵ９０１により実行されるプログラムおよび当該プログラムによって使用されるデータなどを記憶する。通信インターフェイス９０６は、通信網を介して不図示の他の装置からデータを受信してＣＰＵ９０１へ出力し、また、ＣＰＵ９０１が生成したデータを、通信網を介して他の装置へ送信する。
　メディアインターフェイス９０７は、記録媒体９１２に格納されたプログラムまたはデータを読み取り、ＲＡＭ９０３を介してＣＰＵ９０１へ出力する。ＣＰＵ９０１は、プログラムを、メディアインターフェイス９０７を介して記録媒体９１２からＲＡＭ９０３上にロードし、ロードしたプログラムを実行する。記録媒体９１２は、ＤＶＤ（Digital Versatile Disk）などの光学記録媒体、ＭＯ（Magneto Optical disk）などの光磁気記録媒体、磁気記録媒体、導体メモリテープ媒体または半導体メモリなどである。 The SSD 904 stores programs executed by the CPU 901 and data used by the programs. The communication interface 906 receives data from other devices (not shown) via a communication network and outputs it to the CPU 901, and also transmits data generated by the CPU 901 to other devices via the communication network.
Media interface 907 reads a program or data stored in recording medium 912 and outputs it to CPU 901 via RAM 903. The CPU 901 loads a program from the recording medium 912 onto the RAM 903 via the media interface 907, and executes the loaded program. The recording medium 912 is an optical recording medium such as a DVD (Digital Versatile Disk), a magneto-optical recording medium such as an MO (Magneto Optical disk), a magnetic recording medium, a conductive memory tape medium, a semiconductor memory, or the like.
　例えば、コンピュータ９００が性能推定モデル生成装置１００または性能推定装置１００Ａとして機能する場合、コンピュータ９００のＣＰＵ９０１は、ＲＡＭ９０３上にロードされたプログラム１２８，１２８Ａ（図３、図６参照）を実行することにより、性能推定モデル生成装置１００または性能推定装置１００Ａの機能を実現する。ＣＰＵ９０１は、プログラムを記録媒体９１２から読み取って実行する。この他、ＣＰＵ９０１は、他の装置から通信網を介してプログラムを読み込んでもよいし、記録媒体９１２からＳＳＤ９０４にプログラム１２８をインストールして実行してもよい。 For example, when the computer 900 functions as the performance estimation model generation device 100 or the performance estimation device 100A, the CPU 901 of the computer 900 executes the programs 128, 128A (see FIGS. 3 and 6) loaded on the RAM 903. , realizes the functions of the performance estimation model generation device 100 or the performance estimation device 100A. The CPU 901 reads the program from the recording medium 912 and executes it. In addition, the CPU 901 may read a program from another device via a communication network, or may install the program 128 from the recording medium 912 into the SSD 904 and execute it.
≪効果≫
　以下に、性能推定モデル生成装置１００、および性能推定装置１００Ａの効果を説明する。 ≪Effect≫
The effects of the performance estimation model generation device 100 and the performance estimation device 100A will be explained below.
　上記した実施形態に係る性能推定モデル生成装置１００は、通信装置２００の性能情報を推定する機械学習モデルであって、通信装置２００の設定情報と、入力トラヒック情報とを説明変数とし、１つの項目の性能情報を目的変数とする機械学習モデルである性能推定モデルを最初に生成するモデル生成部１１２を備える。
　モデル生成部１１２は、設定情報と、入力トラヒック情報と、１つ以上の項目の性能情報である入力性能情報とを説明変数とし、当該入力性能情報とは異なる１つの項目の性能情報を目的変数とする機械学習モデルである性能推定モデルを繰り返し生成する。
　最初に生成された性能推定モデルおよび繰り返し生成された性能推定モデルに含まれる第１性能推定モデルの説明変数は、当該第１性能推定モデルの１つ前に生成された第２性能推定モデルの説明変数および当該第２性能推定モデルの目的変数である項目の性能情報である。補足すると、ここで第１性能推定モデルを図１記載の性能推定モデル３１３だとすると、第２性能推定モデルは性能推定モデル３１２であり、第１性能推定モデルの説明変数は、第２性能推定モデル（性能推定モデル３１２）の説明変数である設定情報３２１、入力トラヒック情報３２２、性能情報３３１、および、第２性能推定モデル（性能推定モデル３１２）の目的変数である性能情報３３２である。 The performance estimation model generation device 100 according to the embodiment described above is a machine learning model that estimates performance information of the communication device 200, and uses setting information of the communication device 200 and input traffic information as explanatory variables, and one item. The model generation unit 112 is provided that first generates a performance estimation model that is a machine learning model using performance information of as an objective variable.
The model generation unit 112 uses setting information, input traffic information, and input performance information that is performance information of one or more items as explanatory variables, and uses performance information of one item different from the input performance information as an objective variable. A performance estimation model, which is a machine learning model, is repeatedly generated.
The explanatory variables of the first performance estimation model included in the initially generated performance estimation model and the repeatedly generated performance estimation model are the explanatory variables of the second performance estimation model generated immediately before the first performance estimation model. This is performance information of variables and items that are objective variables of the second performance estimation model. As a supplement, if the first performance estimation model is the performance estimation model 313 shown in FIG. 1, the second performance estimation model is the performance estimation model 312, and the explanatory variables of the first performance estimation model are the second performance estimation model ( These are setting information 321, input traffic information 322, performance information 331, which are explanatory variables of the performance estimation model 312), and performance information 332, which is the objective variable of the second performance estimation model (performance estimation model 312).
　このような性能推定モデル生成装置１００によれば、説明変数を設定情報および入力トラヒック情報から始めて、１つの性能項目を推定する性能推定モデルが繰り返し生成される。性能推定モデルが推定した性能情報は、次の性能推定モデルの説明変数となる。このようにすることで、性能推定モデルの学習データ数を増やすことになる。延いては性能推定モデルによる性能推定の精度が向上する。 According to such a performance estimation model generation device 100, a performance estimation model for estimating one performance item is repeatedly generated starting from explanatory variables from setting information and input traffic information. The performance information estimated by the performance estimation model becomes an explanatory variable for the next performance estimation model. By doing so, the number of learning data for the performance estimation model is increased. As a result, the accuracy of performance estimation by the performance estimation model is improved.
　性能推定モデル生成装置１００は、さらに性能推定部１１３と、入力項目選定部１１４を備える。
　モデル生成部１１２は、第２性能推定モデルに続いて第１性能推定モデルを生成するときに、通信装置２００の性能情報の項目であって第１性能推定モデルの入力性能情報に含まれない項目の性能情報を目的変数とする１つ以上の性能推定モデルを第１性能推定モデルの性能推定モデル候補として生成する（図４のステップＳ１６参照）。
　性能推定部１１３は、性能推定モデル候補それぞれを用いて、設定情報と入力トラヒック情報と第１性能推定モデルの入力性能情報とに対応する通信装置２００の実データとを基に、目的変数である項目の性能情報を推定する。
　性能推定部１１３は、性能推定モデル候補それぞれについて、推定された項目の性能情報と、当該性能情報に対応する通信装置２００の実データ（実ノード情報データベース１３０に格納されるデータ）とを基に、当該性能推定モデル候補の目的変数である項目の性能情報の推定精度を算出する。
　入力項目選定部１１４は、性能推定モデル候補の目的変数となる項目の性能情報のなかで推定精度が最良となる項目の性能情報を特定する。
　モデル生成部１１２は、推定精度が最良となる項目の性能情報を目的変数とする性能推定モデル候補を第１性能推定モデルとして選択する（ステップＳ１９参照）。 The performance estimation model generation device 100 further includes a performance estimation section 113 and an input item selection section 114.
When generating the first performance estimation model following the second performance estimation model, the model generation unit 112 generates items of performance information of the communication device 200 that are not included in the input performance information of the first performance estimation model. One or more performance estimation models having the performance information of as an objective variable are generated as performance estimation model candidates for the first performance estimation model (see step S16 in FIG. 4).
The performance estimation unit 113 uses each of the performance estimation model candidates to determine an objective variable based on the actual data of the communication device 200 corresponding to the setting information, input traffic information, and input performance information of the first performance estimation model. Estimate performance information for items.
For each performance estimation model candidate, the performance estimation unit 113 calculates performance estimation based on the performance information of the estimated item and the actual data of the communication device 200 (data stored in the real node information database 130) corresponding to the performance information. , calculate the estimation accuracy of the performance information of the item that is the objective variable of the performance estimation model candidate.
The input item selection unit 114 identifies the performance information of the item with the best estimation accuracy among the performance information of the items to be the target variables of the performance estimation model candidate.
The model generation unit 112 selects, as the first performance estimation model, a performance estimation model candidate whose objective variable is the performance information of the item with the best estimation accuracy (see step S19).
　このような性能推定モデル生成装置１００によれば、推定される性能情報の順序は、性能情報の推定精度が最良になるような順序である。詳しく説明すると性能推定モデル生成装置１００は、推定されていない（推定する性能推定モデルがない）性能情報それぞれについて当該性能情報を推定する性能推定モデルを生成して（図４のステップＳ１６参照）生成された性能推定モデルのなかで精度が最良となる性能項目を次に推定する性能項目とする（ステップＳ１９参照）。 According to the performance estimation model generation device 100, the order of the estimated performance information is such that the estimation accuracy of the performance information is the best. To explain in detail, the performance estimation model generation device 100 generates a performance estimation model for estimating the performance information for each piece of performance information that has not been estimated (there is no performance estimation model to estimate) (see step S16 in FIG. 4). The performance item with the best accuracy among the performance estimation models thus determined is set as the performance item to be estimated next (see step S19).
　性能推定装置１００Ａは、通信装置の性能情報を推定する機械学習モデルである、１つ以上の順序付けられた性能推定モデル（性能推定モデルデータベース１４０参照）を記憶する記憶部１２０と、性能推定部１１３とを備える。
　順序付けられた性能推定モデルにおける最初の性能推定モデル（図１の性能推定モデル３１１参照）の説明変数は、通信装置２００の設定情報と、入力トラヒック情報とである。
　最初の性能推定モデルの目的変数は、１つの項目の性能情報である第１性能情報である。
　最初の性能推定モデル以降の順序付けられた性能推定モデルに含まれる第１性能推定モデルの説明変数は、当該第１性能推定モデルの１つ前の第２性能推定モデルの説明変数および当該第２性能推定モデルの目的変数である項目の性能情報である。
　第１性能推定モデルの目的変数は、当該第１性能推定モデルの説明変数には含まれない１つの項目の性能情報である。
　性能推定部１１３は、設定情報と入力トラヒック情報とを基に、最初の性能推定モデル（性能推定モデル３１１参照）を用いて第１性能情報（性能情報３３１参照）を推定する。
　また性能推定部１１３は、第２性能推定モデルの説明変数、および、当該第２性能推定モデルを用いて推定された項目の性能情報を基に、第１性能推定モデルを用いて目的変数である項目の性能情報を推定する。例えば第１性能推定モデルを性能推定モデル３１３とすると、第２性能推定モデルは性能推定モデル３１２となる。性能推定部１１３は、第２性能推定モデル（性能推定モデル３１２）の説明変数（設定情報３２１、入力トラヒック情報３２２、および、性能情報３３１）および、当該第２性能推定モデル（性能推定モデル３１２）を用いて推定された項目の性能情報（性能情報３３２）を基に、第１性能推定モデル（性能推定モデル３１３）を用いて目的変数である項目の性能情報（性能情報３３３）を推定する。 The performance estimation device 100A includes a storage unit 120 that stores one or more ordered performance estimation models (see performance estimation model database 140), which are machine learning models for estimating performance information of a communication device, and a performance estimation unit 113. Equipped with.
The explanatory variables of the first performance estimation model (see performance estimation model 311 in FIG. 1) in the ordered performance estimation models are setting information of the communication device 200 and input traffic information.
The objective variable of the first performance estimation model is first performance information that is performance information of one item.
The explanatory variables of the first performance estimation model included in the ordered performance estimation models after the first performance estimation model are the explanatory variables of the second performance estimation model immediately before the first performance estimation model and the explanatory variables of the second performance estimation model that precedes the first performance estimation model. This is performance information for items that are objective variables of the estimation model.
The objective variable of the first performance estimation model is performance information of one item that is not included in the explanatory variables of the first performance estimation model.
The performance estimation unit 113 estimates first performance information (see performance information 331) using the first performance estimation model (see performance estimation model 311) based on the setting information and input traffic information.
Furthermore, the performance estimation unit 113 uses the first performance estimation model to determine the objective variable based on the explanatory variables of the second performance estimation model and the performance information of the items estimated using the second performance estimation model. Estimate performance information for items. For example, if the first performance estimation model is the performance estimation model 313, the second performance estimation model is the performance estimation model 312. The performance estimation unit 113 includes explanatory variables (setting information 321, input traffic information 322, and performance information 331) of the second performance estimation model (performance estimation model 312) and the second performance estimation model (performance estimation model 312). Based on the performance information (performance information 332) of the item estimated using , the performance information (performance information 333) of the item that is the target variable is estimated using the first performance estimation model (performance estimation model 313).
　このような性能推定装置１００Ａによれば、性能推定モデル生成装置１００が生成した性能推定モデルを用いて性能情報を推定することができる。 According to such a performance estimation device 100A, performance information can be estimated using the performance estimation model generated by the performance estimation model generation device 100.
１００　性能推定モデル生成装置
１００Ａ　性能推定装置
１１０　制御部
１１１　情報取得部
１１２　モデル生成部
１１３　性能推定部
１１４　入力項目選定部
１２０　記憶部
１２１　推定順序リスト
１２８　プログラム
１３０　実ノード情報データベース
１４０　性能推定モデルデータベース
２００　通信装置 100 Performance estimation model generation device 100A Performance estimation device 110 Control unit 111 Information acquisition unit 112 Model generation unit 113 Performance estimation unit 114 Input item selection unit 120 Storage unit 121 Estimation order list 128 Program 130 Real node information database 140 Performance estimation model database 200 Communication device

Claims (6)

  1. 　通信装置の性能情報を推定する機械学習モデルであって、前記通信装置の設定情報と、入力トラヒック情報とを説明変数とし、１つの項目の前記性能情報を目的変数とする機械学習モデルである性能推定モデルを最初に生成し、
    　前記設定情報と、前記入力トラヒック情報と、１つ以上の項目の前記性能情報である入力性能情報とを説明変数とし、当該入力性能情報とは異なる１つの項目の前記性能情報を目的変数とする機械学習モデルである性能推定モデルを繰り返し生成するモデル生成部を備え、
    　最初に生成された性能推定モデルおよび繰り返し生成された性能推定モデルに含まれる第１性能推定モデルの説明変数は、当該第１性能推定モデルの１つ前に生成された第２性能推定モデルの説明変数および当該第２性能推定モデルの目的変数である項目の性能情報である
    　性能推定モデル生成装置。     A machine learning model for estimating performance information of a communication device, wherein setting information of the communication device and input traffic information are used as explanatory variables, and the performance information of one item is used as an objective variable. First generate the estimation model,
    The setting information, the input traffic information, and the input performance information that is the performance information of one or more items are used as explanatory variables, and the performance information of one item different from the input performance information is used as the objective variable. Equipped with a model generation unit that repeatedly generates a performance estimation model that is a machine learning model,
    The explanatory variables of the first performance estimation model included in the initially generated performance estimation model and the repeatedly generated performance estimation model are the explanatory variables of the second performance estimation model generated immediately before the first performance estimation model. A performance estimation model generation device that is performance information of variables and items that are objective variables of the second performance estimation model.
  2. 　性能推定部と、入力項目選定部とをさらに備え、
    　前記モデル生成部は、
    　前記第２性能推定モデルに続いて前記第１性能推定モデルを生成するときに、前記通信装置の性能情報の項目であって前記第１性能推定モデルの入力性能情報に含まれない項目の性能情報を目的変数とする１つ以上の性能推定モデルを前記第１性能推定モデルの性能推定モデル候補として生成し、
    　前記性能推定部は、
    　前記性能推定モデル候補それぞれを用いて、前記設定情報と前記入力トラヒック情報と前記第１性能推定モデルの入力性能情報とに対応する前記通信装置の実データとを基に、目的変数である項目の性能情報を推定し、
    　前記性能推定モデル候補それぞれについて、推定された項目の性能情報と、当該性能情報に対応する前記通信装置の実データとを基に、当該性能推定モデル候補の目的変数である項目の性能情報の推定精度を算出し、
    　前記入力項目選定部は、
    　前記性能推定モデル候補の目的変数となる項目の性能情報のなかで前記推定精度が最良となる項目の性能情報を特定し、
    　前記モデル生成部は、
    　前記推定精度が最良となる項目の性能情報を目的変数とする性能推定モデル候補を前記第１性能推定モデルとして選択する
    　請求項１に記載の性能推定モデル生成装置。     further comprising a performance estimation section and an input item selection section,
    The model generation unit is
    When generating the first performance estimation model following the second performance estimation model, performance information of items of performance information of the communication device that are not included in the input performance information of the first performance estimation model. Generate one or more performance estimation models with objective variables as performance estimation model candidates for the first performance estimation model,
    The performance estimator includes:
    Using each of the performance estimation model candidates, calculate the item that is the objective variable based on the setting information, the input traffic information, and the actual data of the communication device corresponding to the input performance information of the first performance estimation model. Estimate performance information,
    For each of the performance estimation model candidates, the performance information of the item that is the objective variable of the performance estimation model candidate is estimated based on the estimated performance information of the item and the actual data of the communication device corresponding to the performance information. Calculate the accuracy,
    The input item selection section is
    identifying the performance information of the item with the best estimation accuracy among the performance information of the item serving as the objective variable of the performance estimation model candidate;
    The model generation unit is
    The performance estimation model generation device according to claim 1, wherein a performance estimation model candidate whose objective variable is performance information of an item with the best estimation accuracy is selected as the first performance estimation model.
  3. 　通信装置の性能情報を推定する機械学習モデルである、１つ以上の順序付けられた性能推定モデルを記憶する記憶部と、性能推定部とを備え、
    　前記順序付けられた性能推定モデルにおける最初の性能推定モデルの説明変数は、前記通信装置の設定情報と、入力トラヒック情報とであり、
    　前記最初の性能推定モデルの目的変数は、１つの項目の前記性能情報である第１性能情報であり、
    　前記最初の性能推定モデル以降の前記順序付けられた性能推定モデルに含まれる第１性能推定モデルの説明変数は、当該第１性能推定モデルの１つ前の第２性能推定モデルの説明変数および当該第２性能推定モデルの目的変数である項目の性能情報であり、
    　前記第１性能推定モデルの目的変数は、当該第１性能推定モデルの説明変数には含まれない１つの項目の前記性能情報であり、
    　前記性能推定部は、
    　前記設定情報と前記入力トラヒック情報とを基に、前記最初の性能推定モデルを用いて前記第１性能情報を推定し、
    　前記第２性能推定モデルの説明変数、および、当該第２性能推定モデルを用いて推定された項目の性能情報を基に、前記第１性能推定モデルを用いて目的変数である項目の性能情報を推定する
    　性能推定装置。     comprising a storage unit storing one or more ordered performance estimation models, which are machine learning models for estimating performance information of the communication device, and a performance estimation unit,
    Explanatory variables of the first performance estimation model in the ordered performance estimation model are setting information of the communication device and input traffic information,
    The objective variable of the first performance estimation model is first performance information that is the performance information of one item,
    The explanatory variables of the first performance estimation model included in the ordered performance estimation models after the first performance estimation model are the explanatory variables of the second performance estimation model immediately preceding the first performance estimation model, and the explanatory variables of the second performance estimation model that precedes the first performance estimation model. 2 Performance information of the item that is the objective variable of the performance estimation model,
    The objective variable of the first performance estimation model is the performance information of one item that is not included in the explanatory variables of the first performance estimation model,
    The performance estimator includes:
    estimating the first performance information using the first performance estimation model based on the setting information and the input traffic information;
    Based on the explanatory variables of the second performance estimation model and the performance information of the item estimated using the second performance estimation model, the performance information of the item that is the target variable is calculated using the first performance estimation model. Estimating performance estimation device.
  4. 　コンピュータを、請求項１に記載の性能推定モデル生成装置として機能させるためのプログラム。 A program for causing a computer to function as the performance estimation model generation device according to claim 1.
  5. 　コンピュータを、請求項３に記載の性能推定装置として機能させるためのプログラム。 A program for causing a computer to function as the performance estimating device according to claim 3.
  6. 　性能推定モデル生成装置が、
    　通信装置の性能情報を推定する機械学習モデルであって、前記通信装置の設定情報と、入力トラヒック情報とを説明変数とし、１つの項目の前記性能情報を目的変数とする機械学習モデルである性能推定モデルを最初に生成し、
    　前記設定情報と、前記入力トラヒック情報と、１つ以上の項目の前記性能情報である入力性能情報とを説明変数とし、当該入力性能情報とは異なる１つの項目の前記性能情報を目的変数とする機械学習モデルである性能推定モデルを繰り返し生成するステップを実行し、
    　最初に生成された性能推定モデルおよび繰り返し生成された性能推定モデルに含まれる第１性能推定モデルの説明変数は、当該第１性能推定モデルの１つ前に生成された第２性能推定モデルの説明変数および当該第２性能推定モデルの目的変数である項目の性能情報である
    　性能推定モデル生成方法。     The performance estimation model generator is
    A machine learning model for estimating performance information of a communication device, wherein setting information of the communication device and input traffic information are used as explanatory variables, and the performance information of one item is used as an objective variable. First generate the estimation model,
    The setting information, the input traffic information, and the input performance information that is the performance information of one or more items are used as explanatory variables, and the performance information of one item different from the input performance information is used as the objective variable. Execute the step of iteratively generating a performance estimation model that is a machine learning model,
    The explanatory variables of the first performance estimation model included in the initially generated performance estimation model and the repeatedly generated performance estimation model are the explanatory variables of the second performance estimation model generated immediately before the first performance estimation model. A method for generating a performance estimation model, which is performance information of a variable and an item that is an objective variable of the second performance estimation model.
PCT/JP2022/029632 2022-08-02 2022-08-02 Performance inference model generation device, performance inference device, program, and performance inference model generation method WO2024028974A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/JP2022/029632 WO2024028974A1 (en) 2022-08-02 2022-08-02 Performance inference model generation device, performance inference device, program, and performance inference model generation method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2022/029632 WO2024028974A1 (en) 2022-08-02 2022-08-02 Performance inference model generation device, performance inference device, program, and performance inference model generation method

Publications (1)

Publication Number Publication Date
WO2024028974A1 true WO2024028974A1 (en) 2024-02-08

Family

ID=89848629

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2022/029632 WO2024028974A1 (en) 2022-08-02 2022-08-02 Performance inference model generation device, performance inference device, program, and performance inference model generation method

Country Status (1)

Country Link
WO (1) WO2024028974A1 (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2016038857A1 (en) * 2014-09-09 2016-03-17 日本電気株式会社 Number-of-scales estimation apparatus, number-of-scales management system, number-of-scales estimation method, number-of-scales management method, and storage medium
JP2019008554A (en) * 2017-06-23 2019-01-17 富士通株式会社 Information processing device, information processing system, information processing program, and information processing method

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2016038857A1 (en) * 2014-09-09 2016-03-17 日本電気株式会社 Number-of-scales estimation apparatus, number-of-scales management system, number-of-scales estimation method, number-of-scales management method, and storage medium
JP2019008554A (en) * 2017-06-23 2019-01-17 富士通株式会社 Information processing device, information processing system, information processing program, and information processing method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
KYOTA HATTORI, TOMOHIRO KORIKAWA, CHIKAKO TAKASAKI, HIDENARI OOWADA, MASAFUMI SHIMIZU, NAOKI TAKAYA: "Proposal of Network Digital Replica", IEICE TECHNICAL REPORT, NS, IEICE, JP, vol. 121, no. 301 (NS2021-104), 9 December 2021 (2021-12-09), JP, pages 41 - 46, XP009552556 *

Similar Documents

Publication Publication Date Title
Horng et al. Evolutionary algorithm for stochastic job shop scheduling with random processing time
EP3779731A1 (en) Distributed graph embedding method, apparatus and system, and device
US7483864B2 (en) Active learning method and system
EP3761562A1 (en) Indicator determination method and related device thereto
Isaila et al. Collective I/O tuning using analytical and machine learning models
CN113822456A (en) Service combination optimization deployment method based on deep reinforcement learning in cloud and mist mixed environment
JP5673473B2 (en) Distributed computer system and method for controlling distributed computer system
US10248462B2 (en) Management server which constructs a request load model for an object system, load estimation method thereof and storage medium for storing program
Yu et al. A hybrid GP-Tabu approach to QoS-aware data intensive Web service composition
WO2022091408A1 (en) Solution finding method selection device and method
WO2024028974A1 (en) Performance inference model generation device, performance inference device, program, and performance inference model generation method
US20230222385A1 (en) Evaluation method, evaluation apparatus, and non-transitory computer-readable recording medium storing evaluation program
Darir et al. Mleflow: Learning from history to improve load balancing in tor
Hu et al. Proactive planning of bandwidth resource using simulation-based what-if predictions for Web services in the cloud
CN111598390B (en) Method, device, equipment and readable storage medium for evaluating high availability of server
JP2019008554A (en) Information processing device, information processing system, information processing program, and information processing method
CN112882917A (en) Virtual machine service quality dynamic prediction method based on Bayesian network migration
Chatziprimou et al. Surrogate-assisted online optimisation of cloud iaas configurations
WO2022097302A1 (en) Generation program, generation method, and information processing device
US20230379737A1 (en) Devices and methods for predicting a mobile network experience
Hristov Performance models for analysis and control of it systems
CN116798103B (en) Artificial intelligence-based face image processing method and system
Begin et al. Towards an automatic modeling tool for observed system behavior
JP7421391B2 (en) Learning methods and programs
Hong et al. Topological Experience Replay for Fast Q-Learning

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22953963

Country of ref document: EP

Kind code of ref document: A1