MODEL GENERATION DEVICE, MODEL GENERATION METHOD, AND MODEL GENERATION PROGRAM

Abstract

This model generation device (10) comprises a learning unit (23) and a generation unit (25). The learning unit learns a statistical quantity on the basis of a first secondary coefficient and a second secondary coefficient (22a), each of which is associated with a set of first and second bases (5) and which are different from one another. The generation unit generates a third secondary coefficient (25a) associated with the set of first and second bases on the basis of the learned statistical quantity. The first, second, and third secondary coefficients indicate the degree to which the set of first and second bases contributes to a plurality of bases including the first and second bases.

Description

[Technical field]

[0001] The embodiments relate to a model generation device, a model generation method, and a model generation program. [Background technology]

[0002] A communication network is formed by a number of base stations. When some of the base stations become unable to communicate due to a disaster or the like, some of the functions of the communication network are lost. A base station that becomes unable to communicate can be restored to a state where communication is possible by delivering resources such as power. However, particularly during a disaster, there are often restrictions on the delivery of resources.

[0003] For this reason, an optimization problem has been studied that maximizes the degree to which the restoration of base stations contributes to the restoration of the communication network (relief scale) under constraints. As a method for quantitatively solving this optimization problem, a method for quantifying the relief scale has been proposed. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Kimihiko Fukami and 3 others, "A study on a method to grasp the service impact caused by network failures," IEICE Technical Report, February 28, 2019, vol. 118, no. 483, ICM2018-51, pp. 13-18 Summary of the Invention [Problem to be solved by the invention]

[0005] However, the number of combinations of base stations to be restored increases exponentially with the number of base stations. For this reason, it is not realistic to use the proposed method to calculate the numerical relief scale for all combinations within the limited time when a disaster occurs. In other words, it is difficult to calculate the relief scale in a short time.

[0006] The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a means for calculating the relief scale in a short period of time. [Means for solving the problem]

[0007] A model generating device according to one aspect includes a learning unit and a generating unit. Among multiple sample rescue size models for multiple locations including the first and second locations, the first sample rescue size model First quadratic coefficient 、 and Included in the second sample rescue size model Based on the second quadratic coefficient, The third quadratic coefficient included in the approximate relief size model for the above multiple bases The first statistic is learned. The generating unit generates a first statistic based on the learned first statistic. , the above The first quadratic coefficient, the second quadratic coefficient, and the third quadratic coefficient are calculated by the above formula: Duplication This shows the contribution of the pair of the first and second locations to the number of locations. The first statistical amount includes a mean and a variance when the third quadratic coefficient is regarded as a probability density function. Effect of the Invention

[0008] According to the embodiment, it is possible to provide a means for calculating the relief scale in a short time. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a schematic diagram illustrating an example of the configuration of a delivery system according to an embodiment. [Diagram 2] FIG. 2 is a diagram illustrating a schematic of features of a relief scale model according to an embodiment. [Diagram 3] FIG. 3 is a diagram illustrating a schematic diagram of a feature of a relief scale model according to an embodiment. [Figure 4] FIG. 4 is a diagram illustrating a schematic diagram of a feature of a relief scale model according to an embodiment. [Diagram 5] FIG. 5 is a block diagram illustrating an example of a hardware configuration of a relief scale model generating device according to the embodiment. [Figure 6] FIG. 6 is a block diagram illustrating an example of a functional configuration of a relief scale model generating device according to the embodiment. [Figure 7]FIG. 7 is a flowchart showing an example of a remedy scale model generating method in the remedy scale model generating device according to the embodiment. [Figure 8] FIG. 8 is a flowchart showing an example of a learning operation in the relief scale model generating method in the relief scale model generating device according to the embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Hereinafter, the embodiments will be described with reference to the drawings. In the following description, components having the same function and configuration are given the same reference symbol. In addition, when multiple components having the same reference symbol are to be distinguished from each other, they are distinguished from each other by an additional reference symbol (e.g., a hyphen and a number such as "-1") that is added following the common reference symbol.

[0011] 1. Embodiment 1.1 Configuration The configuration of the delivery system according to the embodiment will be described.

[0012] 1.1.1 Overall Configuration First, a configuration of a delivery system according to an embodiment will be described. Fig. 1 is a block diagram showing an example of the configuration of a delivery system according to an embodiment.

[0013] As shown in Fig. 1, the delivery system 1 is a system for supplying resources within a certain area using multiple delivery vehicles 4-1 and 4-2. The delivery system 1 includes a delivery support center 2, multiple delivery bases 3-1 and 3-2, and multiple base stations 5-1, 5-2, 5-3, 5-4, 5-5, and 5-6.

[0014] In the following, when there is no particular distinction between the multiple delivery bases 3-1 and 3-2, they may be referred to as "delivery bases 3." When there is no particular distinction between the multiple delivery vehicles 4-1 and 4-2, they may be referred to as "delivery vehicles 4." When there is no particular distinction between the multiple base stations 5-1 to 5-6, they may be referred to as "base stations 5."

[0015] The delivery support center 2 is a control center in the delivery system 1. The delivery support center 2 generates a delivery plan for comprehensively controlling the delivery of resources. The delivery support center 2 is configured to be wirelessly connected to the delivery base 3. As a result, the delivery support center 2 wirelessly transmits the generated delivery plan to the delivery base 3.

[0016] At least one delivery vehicle 4 is deployed at the delivery base 3. In the example of FIG. 1, delivery vehicles 4-1 and 4-2 are deployed at the delivery bases 3-1 and 3-2, respectively. In addition, resources to be delivered to the base station 5 are stored at the delivery base 3. Based on the delivery plan received from the delivery support center 2, the delivery base 3 instructs the delivery vehicle 4 to depart.

[0017] The delivery vehicle 4 delivers resources within the delivery base 3 where the delivery vehicle 4 is located to the base station 5 in accordance with various constraint conditions. The constraint conditions include, for example, the distance that the delivery vehicle 4 can travel, the time that the delivery vehicle 4 can travel, the amount of resources that can be carried, etc. The resources delivered by the delivery vehicle 4 include, for example, equipment and personnel for restoring a base station 5 that is in an uncommunicable state to a state where communication is possible.

[0018] The base station 5 is, for example, a base station of the communication network NW. The base station 5 provides communication services to users (not shown) located within a predetermined area centered on the base station itself. The base station 5 is also configured to communicate with other base stations 5 located within a predetermined area centered on the base station itself. In the example of FIG. 1, the base station 5-1 is placed in a position where it can communicate with the base stations 5-3 and 5-6. The base station 5-2 is placed in a position where it can communicate with the base stations 5-4 and 5-5. The base station 5-3 is placed in a position where it can communicate with the base stations 5-1 and 5-4. The base station 5-4 is placed in a position where it can communicate with the base stations 5-2 and 5-3. The base station 5-5 is placed in a position where it can communicate with the base stations 5-2 and 5-6. The base station 5-6 is placed in a position where it can communicate with the base stations 5-1 and 5-5. With the above configuration, a plurality of base stations 5 that can communicate with each other can form one huge communication network NW.

[0019] The base stations 5-1 to 5-6 include base stations that are not able to communicate and base stations that are able to communicate. In the example of FIG. 1, the base stations 5-1, 5-2, 5-3, and 5-4 are not able to communicate, and the base stations 5-5 and 5-6 are able to communicate. In this case, the communication network NW is limited to between the base stations 5-5 and 5-6 that are able to communicate. The base station 5 that is not able to communicate can be restored to a state where it is able to communicate by delivering resources. Therefore, the base station 5 that is not able to communicate is a candidate for a delivery destination to which resources are delivered by the delivery vehicle 4.

[0020] Next, a description will be given of the internal configuration of the delivery support center 2. The delivery support center 2 includes a relief scale model generating device 10 and a delivery plan generating device 30. The relief scale model generating device 10 and the delivery plan generating device 30 are, for example, computers.

[0021] The rescue scale model generating device 10 generates a rescue scale model. The rescue scale model is a model for quantifying the rescue scale achieved by delivering resources to the base station 5. The rescue scale can be regarded as, for example, the contribution or influence of the restored base station 5 to the communication network NW.

[0022] The delivery plan generating device 30 generates a delivery plan that can maximize the relief scale under constraint conditions, based on the relief scale model generated by the relief scale model generating device 10. The delivery plan generating device 30 transmits the generated delivery plan to the delivery base 3.

[0023] 1.1.2 Relief scale model Next, a relief scale model according to the embodiment will be described.

[0024] 2 is a diagram showing a typical feature of a rescue scale model according to an embodiment. In FIG. 2, as an example of a general feature of the rescue scale model, a manner in which the rescue scale changes with respect to the number of base stations 5 to be restored is shown.

[0025] As shown in FIG. 2, the rescue scale model is a concave function (or an upward convex function) with respect to the number of base stations 5 to be restored. Specifically, the rescue scale increases as the number of base stations 5 to be restored increases (i.e., as the number of base stations 5 in a communicable state increases). On the other hand, the increment in the rescue scale decreases as the number of base stations 5 to be restored increases. This characteristic indicates that when there are a large number of base stations 5 in a communicable state, the contribution of the base station 5 that has newly become communicable to the entire communication network NW is smaller than when there are a small number of base stations 5 in a communicable state.

[0026] In addition to the features shown in Fig. 2, the rescue scale model is required to be calculated according to the combination of base stations 5 to be restored. Figs. 3 and 4 are diagrams that typically show features of the rescue scale model according to the embodiment. Figs. 3 and 4 show the rescue scale when the base stations 5-1 to 5-6 are restored from a state in which only the base station 5-1 is capable of communication to a state in which two new base stations 5 are capable of communication.

[0027] As shown in FIG. 3, when the base stations 5-4 and 5-5 are restored, the number of base stations 5 in a communicable state increases to three, but the base stations 5-1, 5-4, and 5-5 cannot communicate with each other. On the other hand, as shown in FIG. 4, when the base stations 5-3 and 5-6 are restored, the number of base stations 5 in a communicable state is three, which is the same as in FIG. 3, but the base stations 5-1, 5-3, and 5-6 can communicate with each other. In this case, it is preferable that the rescue scale model is constructed so that the rescue scale corresponding to the case where the base stations 5-3 and 5-6 are restored is larger than the rescue scale corresponding to the case where the pair of base stations 5-4 and 5-5 is restored. In this way, it is preferable that the rescue scale model expresses the effect of the restoration of a certain base station 5 spreading to the surrounding base stations 5.

[0028] 1.1.3 Relief scale model generator Next, a configuration of the relief scale model generating device according to the embodiment will be described.

[0029] (Hardware configuration) 5 is a block diagram showing an example of a hardware configuration of a relief scale model generating device according to an embodiment. As shown in FIG. 5, the relief scale model generating device 10 includes a control circuit 11, a memory 12, a communication module 13, a user interface 14, and a drive 15.

[0030] The control circuit 11 is a circuit that overall controls each component of the relief scale model generating apparatus 10. The control circuit 11 includes a CPU (Central Processing Unit), a RAM (Random Access Memory), and a ROM (Read Only Memory).

[0031] The memory 12 is an auxiliary storage device of the relief scale model generating device 10. The memory 12 includes, for example, a hard disk drive (HDD), a solid state drive (SSD), a memory card, etc. The memory 12 stores various information used in the relief scale model generating operation and a relief scale model generating program. The relief scale model generating program can be stored in the memory 12 by being transmitted from outside the relief scale model generating device 10 via a network (not shown).

[0032] The relief scale model generation operation is a series of operations executed to generate a relief scale model in the relief scale model generator 10. The relief scale model generation program is a program for causing the relief scale model generator 10 to execute the relief scale model generation operation. The relief scale model generation operation will be described in detail later.

[0033] The communication module 13 is a circuit used for transmitting and receiving data via a network. The communication module 13 is constructed in accordance with, for example, Ethernet (registered trademark).

[0034] The user interface 14 is a circuit for communicating information between a user and the control circuit 11. The user interface 14 includes input devices and output devices. The input devices include, for example, a touch panel and an operation button. The output devices include, for example, an LCD (Liquid Crystal Display) and an EL (Electroluminescence) display, and a printer. The user interface 14 converts input from the user (user input) into an electrical signal and then transmits it to the control circuit 11. The user interface 14 outputs the execution result of the relief scale model generation program received from the control circuit 11 to the user.

[0035] The drive 15 is a device for reading a program stored in a storage medium 16. The drive 15 includes, for example, a CD (Compact Disk) drive and a DVD (Digital Versatile Disk) drive.

[0036] The storage medium 16 is a medium that stores information such as a program by electrical, magnetic, optical, mechanical, or chemical action. The storage medium 16 may store a relief scale model generation program.

[0037] (Functional configuration) FIG. 6 is a block diagram illustrating an example of a functional configuration of a relief scale model generating device according to the embodiment.

[0038] The CPU of the control circuit 11 loads the relief scale model generation program stored in the memory 12 or the storage medium 16 into the RAM. The CPU of the control circuit 11 then interprets and executes the relief scale model generation program loaded into the RAM to control the memory 12, the communication module 13, the user interface 14, the drive 15, and the storage medium 16. As a result, as shown in Fig. 6, the relief scale model generation device 10 functions as a computer including an input unit 21, a memory unit 22, a secondary coefficient learning unit 23, a primary coefficient estimating unit 24, a generation unit 25, and an output unit 26.

[0039] Based on an input from a user, the input unit 21 stores various information in the storage unit 22. The various information stored in the storage unit 22 includes, for example, a plurality of sample rescue scale models 22a.

[0040] The multiple sample rescue scale models 22a are rescue scale models constructed, for example, by an existing method when a past disaster occurred, so as to satisfy the above-mentioned requirements. That is, the multiple sample rescue scale models 22a are constructed based on the environment of the communication network NW when a past disaster occurred. The environment of the communication network NW includes the number of users who use each base station, and the number and arrangement of the base stations 5. That is, the multiple sample rescue scale models 22a do not necessarily match the current environment of the communication network NW (at the time when the rescue scale model generating device 10 should generate the rescue scale model). In addition, the multiple sample rescue scale models 22a may be different from each other.

[0041] A specific expression format of the multiple sample rescue scale model 22a will be described.

[0042] Explanatory variables for multiple sample rescue scale model 22a:=(x 1 ,…,x n ) T , x i When the number of base stations 5 is n, ∈{0,1},∀i is expressed as the following formula (1).

[0043]

number

[0044] The multiple sample rescue scale model 22a is expressed as the following formula (2) using the explanatory variable X described above as a function f(X).

[0045]

number

[0046] Here, n is an integer equal to or greater than 3. i, j, and k are integers equal to or greater than 1 and equal to or less than n. Furthermore, the first term on the right side of equation (2) indicates the sum of the contribution (relief scale) of each individual base station 5 to the communication network NW. The second term on the right side of equation (2) indicates the sum of the contribution of any combination of two of the base stations 5 to the communication network NW. The third term on the right side of equation (2) indicates the sum of the contribution of any combination of three of the base stations 5 to the communication network NW.

[0047] In addition, the linear coefficient α of the first term on the right side of the formula (2), the quadratic coefficient β of the second term on the right side, the cubic coefficient γ of the third term on the right side, ... (hereinafter simply referred to as "various coefficients") are parameters. That is, by determining the various coefficients, each of the multiple sample rescue scale models 22a becomes an explicit function. In other words, the memory unit 22 stores various coefficients corresponding to each of the multiple sample rescue scale models 22a.

[0048] The quadratic coefficient learning unit 23 learns the statistics of the quadratic coefficient β for each pair of any two stations of the base stations 5 based on a plurality of sample rescue scale models 22a in response to, for example, a user instruction. Specifically, the quadratic coefficient learning unit 23 regards the quadratic coefficient β for each pair of any two stations of the base stations 5 as a probability density function Pr as shown in the following formula (3).

[0049]

number

[0050] For example, the effect of the restoration of a pair of two stations spreading to the surrounding base stations 5 may depend on the distance on the communication network NW between the two stations to be restored. When the two stations to be restored are adjacent to each other on the communication network NW, there is no base station 5 that mediates communication between the two stations. Therefore, when the two stations to be restored are adjacent to each other on the communication network NW, the expected value (average) of the effect of the restoration of the pair of two stations spreading to the surrounding base stations 5 is relatively small. On the other hand, when the two stations to be restored are far from each other on the communication network NW, there is a base station 5 that can mediate communication between the two stations, so the effect of the restoration may spread to the base station 5 located between the two stations. Therefore, when the two stations to be restored are far from each other on the communication network NW, the expected value (average) of the effect of the restoration of the pair of two stations spreading to the surrounding base stations 5 is relatively large. In this way, the average contribution degree may change depending on the pair of two stations to be restored.

[0051] The effect of the restoration of the pair of two stations spreading to the surrounding base stations 5 may depend on the status (damage status due to the disaster) of the surrounding base stations 5. As an example, a case where the pair of base stations 5-4 and 5-5 is restored in a situation where there is one base station in a communicable state will be described. For example, if the base station in a communicable state is base station 5-1, the three base stations 5-1, 5-4, and 5-5 in a communicable state after the base stations 5-4 and 5-5 are restored cannot communicate with each other. For example, if the base station in a communicable state is base station 5-3 or 5-6, the three base stations 5-3 or 5-6, 5-4, and 5-5 in a communicable state after the base stations 5-4 and 5-5 are restored can communicate with each other between the two stations. For example, if the base station in a communicable state is base station 5-2, the three base stations 5-2, 5-4, and 5-5 in a communicable state after the base stations 5-4 and 5-5 are restored can communicate with each other between the three stations. In this way, the contribution degree depends on the state of the surrounding base station 5, but the state of the surrounding base station 5 changes probabilistically. For this reason, the contribution degree may be distributed probabilistically according to the state of the surrounding base station 5.

[0052] Based on the above considerations, the quadratic coefficient learning unit 23 regards the quadratic coefficient β as a probability density function Pr. Then, the quadratic coefficient learning unit 23 learns the statistics of the probability density function Pr based on a plurality of sample rescue scale models 22a. The quadratic coefficient learning unit 23 transmits the statistics obtained by the learning operation to the generating unit 25.

[0053] Specifically, for example, the quadratic coefficient learning unit 23 assumes that the probability density function Pr is a normal distribution with a population mean μ and a population variance σ^2. The quadratic coefficient learning unit 23 calculates the sample mean β of the quadratic coefficient β for each pair of all two stations among the base stations 5 based on the multiple sample rescue scale models 22a. ave When the population variance σ^2 can be assumed to be known, the quadratic coefficient learning unit 23 calculates the posterior distribution (η post , τ post ) according to the following equations (4) and (5). The secondary coefficient learning unit 23 calculates the population variance σ^2 and the posterior distribution (η post , τ post ) is transmitted to the generating unit 25 as the statistics of the probability density function Pr.

[0054]

number

[0055] Here, (η pre , τ pre ) is the prior distribution of the population mean μ.

[0056] The primary coefficient estimation unit 24 estimates the primary coefficient α for all of the base stations 5 based on the current environment of the communication network NW. When estimating the primary coefficient α, the primary coefficient estimation unit 24 may use any existing method. The primary coefficient estimation unit 24 transmits the estimated primary coefficient α to the generation unit 25.

[0057] The generating unit 25 generates an approximate repair scale model 25 a based on the statistics of the probability density function Pr received from the secondary coefficient learning unit 23 and the primary coefficient α received from the primary coefficient estimating unit 24 .

[0058] Specifically, the generation unit 25 generates the approximate repair scale model 25a by considering up to the quadratic term of the function f(X). That is, the generation unit 25 generates the approximate repair scale model 25a by considering up to the quadratic term of the function f(X) shown in the following equation (6): - The first-order coefficient α' and the second-order coefficient β' of (X) are considered as the approximate relief size model 25a.

[0059]

number

[0060] Here, ε is a term that compensates for the difference with the function f(X). ε may be 0.

[0061] The generating unit 25 regards the primary coefficient α received from the primary coefficient estimating unit 24 as the primary coefficient α'. The generating unit 25 also estimates the secondary coefficient β' based on the statistics of the probability density function Pr received from the secondary coefficient learning unit 23. The statistics include the population variance σ^2 and the posterior distribution (η post , τ post ), the generation unit 25 may generate, for example, the posterior distribution η post is regarded as the quadratic coefficient β'. Regardless of the above example, the generation unit 25 may generate the population variance σ^2 and the posterior distribution (η post , τ post ) may be regarded as the secondary coefficient β'. The generating unit 25 transmits the primary coefficient α' and the secondary coefficient β' to the output unit 26 as an approximate repair scale model 25a.

[0062] The output unit 26 outputs the approximate rescue scale model 25a received from the generation unit 25 to the delivery plan generation device 30 and the user.

[0063] With the above configuration, the delivery plan generating device 30 can generate a delivery plan based on the approximated delivery scale model 25a. Then, the user can check the generated approximated delivery scale model 25a.

[0064] 1.2. Operation Next, the operation of the relief scale model generating device according to the embodiment will be described.

[0065] 1.2.1 Relief scale model generation operation 7 is a flowchart showing an example of a repair scale model generating operation in the repair scale model generating device according to the embodiment. In the example of FIG. 7, it is assumed that a plurality of sample repair scale models 22a are stored in the memory 12 in advance by a user input.

[0066] 7, in response to a user instruction (start), the secondary coefficient learning unit 23 selects a combination of two base stations from the base stations 5 in the communication network NW (S10). That is, the secondary coefficient learning unit 23 selects one secondary coefficient β to be learned.

[0067] The quadratic coefficient learning unit 23 learns statistics corresponding to the combination selected in the process of S10 based on a plurality of sample rescue scale models 22a (S20).

[0068] The secondary coefficient learning unit 23 judges whether or not the statistics have been learned for all two-station combinations (S30). If there is a combination for which the statistics have not been learned (S30; no), the secondary coefficient learning unit 23 selects a combination for which the statistics have not been learned (S10). In this manner, the processes of S10 and S20 are repeated until the statistics have been learned for all two-station combinations.

[0069] When the learning of statistics has been completed for all combinations (S30; yes), the secondary coefficient learning unit 23 transmits all statistics obtained by the learning operation to the generation unit 25. The generation unit 25 estimates the secondary coefficient β' based on the statistics (S40).

[0070] The primary coefficient estimation unit 24 estimates the primary coefficients α′ by using an existing method (S50). The primary coefficient estimation unit 24 transmits all the estimated primary coefficients α′ to the generation unit 25.

[0071] The generating unit 25 generates an approximate repair scale model 25a based on the secondary coefficient β' estimated in the process of S40 and the primary coefficient α' estimated in the process of S50 (S60). The generating unit 25 transmits the generated approximate repair scale model 25a to the output unit 26.

[0072] The output unit 26 outputs the approximate repair scale model 25a generated in the process of S60 to the delivery plan generating device 30 and the user (S60).

[0073] When the process of S60 ends, the repair scale model generation operation ends (END).

[0074] 7, the process of S50 is executed after the process of S40, but this is not limiting. For example, the process of S50 may be executed in parallel with the processes of S10, S20, S30, and S40.

[0075] 1.2.2 Specific examples of learning behavior 8 is a flowchart showing an example of a learning operation of the repair scale model generating operation in the repair scale model generating device according to the embodiment. FIG 8 corresponds to a specific example of the process of S20 in FIG 7.

[0076] In the following, as a specific example, it is assumed that the combination of base stations 5-1 and 5-2 is selected. In this case, the probability density function Pr(x 1 ,x 2 ) is the statistic (μ 12 ,σ 12 ) is assumed to follow a normal distribution with the population variance σ 12 ^2 is known, and σ 12 =1.0.

[0077] When the combination of the two base stations 5-1 and 5-2 is selected (started) by the process of S10 described with reference to FIG. 7, the secondary coefficient learning unit 23 learns the secondary coefficient β corresponding to the selected combination based on a plurality of sample rescue scale models 22a. 12 The sample mean of β 12aveFor example, as a multiple sample rescue scale model 22a, three quadratic coefficients β 12 = 1.2, 1.5, and 0.6 are stored (N = 3), the sample mean β 12ave becomes (1.2+1.5+0.6) / 3=1.1.

[0078] The quadratic coefficient learning unit 23 learns the sample mean β 12ave The calculated quadratic coefficient β 12 The population mean μ in the probability density function Pr corresponding to 12 The prior distribution of (η 12pre , τ 12pre For example, the secondary coefficient learning unit 23 sets the prior distribution (η 12pre , τ 12pre )=(1.0,1.0).

[0079] The secondary coefficient learning unit 23 learns the sample mean β 12ave and the prior distribution (η pre , τ pre ) based on the population mean μ in the probability density function Pr 12 The posterior distribution of (η 12post , τ 12post Specifically, the secondary coefficient learning unit 23 learns the posterior distribution (η 12post , τ 12post )=(1.075, 0.5). As a result, the secondary coefficient learning unit 23 calculates the population mean μ 12 We can learn that there is approximately a 67% probability that is in the range (1.075-0.5) or greater and (1.075+0.5) or less.

[0080] When the process of S23 ends, the learning operation for one combination ends (ends).

[0081] 1.3 Effects of the embodiment According to the embodiment, the secondary coefficient learning unit 23 is associated with a pair of the base stations 5-1 and 5-2 and learns a plurality of mutually different secondary coefficients β 12 Based on the probability density function Pr(x 1 ,x 2The generation unit 25 learns the statistics of the quadratic coefficient β' based on the learned statistics. 12 Generate multiple quadratic coefficients β 12 and β' 12 indicates the contribution of the pair of base stations 5-1 and 5-2 to a plurality of base stations 5 including the base stations 5-1 and 5-2. This allows the approximate repair scale model 25a to be generated without using the function f(X) based on the sample repair scale model 22a already generated using the function f(x) shown in equation (2). This makes it possible to suppress the calculation load required for generating the approximate repair scale model 25a. Therefore, the repair scale can be calculated in a short time.

[0082] Supplementally, the function f shown in equation (6) - (X) omits the third and higher terms for the function f(X). Therefore, the function f - The various coefficients α' and β' in (X) can be calculated with less computational load than the various coefficients α, β, γ, ... in the function f(X). In addition, while the function f(X) can calculate the relief scale with high accuracy, a relatively complicated mathematical model is used to calculate the various coefficients α, β, γ, .... In contrast, the function f - In particular, the quadratic coefficient β' of (X) is calculated as a statistic based on multiple quadratic coefficients β that have been calculated with high accuracy in the past. Therefore, the quadratic coefficient β' can be calculated using a mathematical model that is relatively simpler than the quadratic coefficient β. In addition, the accuracy of the statistic of the quadratic coefficient β' can be improved as the number of samples N increases. Therefore, the relief size can be calculated in a short time without excessively deteriorating the accuracy.

[0083] In addition, the secondary coefficient learning unit 23 executes the learning operation of the above-mentioned statistics for all combinations of two stations selected from the base stations 5 in the communication network NW. As a result, the function f - The above-mentioned effect can be obtained for all the secondary coefficients β' in (X), so that the relief scale can be calculated in a short time.

[0084] In addition, the quadratic coefficient learning unit 23 learns the population mean μ and / or population variance σ^2 of a probability density function with multiple quadratic coefficients β extracted from multiple sample rescue scale models 22a as samples as statistics. This allows existing methods such as Bayesian optimization to be applied to the learning operation. Therefore, by formulating the learning operation, it can be executed in a short time.

[0085] 2. Other It should be noted that various modifications can be applied to the above-described embodiment.

[0086] For example, in the above embodiment, the relief scale model generating device 10 and the delivery plan generating device 30 are configured by two computers, but this is not limited to the above. For example, the relief scale model generating device 10 and the delivery plan generating device 30 may be configured by one computer.

[0087] Also, for example, in the above-described embodiment, the case where the relief scale model generation program is executed by the relief scale model generation device 10 in the delivery support center 2 has been described, but this is not limited thereto. For example, the relief scale model generation program may be executed by a computing resource on the cloud.

[0088] Also, for example, in the above-mentioned embodiment, the case where n is an integer of 3 or more has been described, but this is not limited thereto. For example, n may be 2. In this case, the third term on the right side of the formula (2) is omitted.

[0089] Also, for example, in the above embodiment, the case where the base station 5 in the communication network NW is restored has been described, but the present invention is not limited to this. For example, the target of restoration may be a base in any network other than a communication network that is configured by multiple bases that mutually affect each other. Also, the base may be a building.

[0090] The present invention is not limited to the above-mentioned embodiment, and can be modified in various ways without departing from the gist of the present invention. The embodiments may be combined as appropriate, and in that case, the combined effect can be obtained. Furthermore, the above-mentioned embodiment includes various inventions, and various inventions can be extracted by combinations selected from the multiple components disclosed. For example, if the problem can be solved and the effect can be obtained even if some components are deleted from all the components shown in the embodiment, the configuration from which the components are deleted can be extracted as an invention. [Explanation of symbols]

[0091] 1. Delivery system 2. Delivery Support Center 3-1, 3-2…Distribution base 4-1,4-2…Delivery vehicle 5-1,5-2,5-3,5-4,5-5,5-6…Base station 10…Relief scale model generator 11...Control circuit 12…Memory 13...Communication module 14...User Interface 15…Drive 16...Storage medium 21...Input section 22...Storage section 22a…Sample rescue scale model 23...Quadratic coefficient learning section 24... Primary coefficient estimator 25…Generation part 25a…Approximate relief size model 26...Output section 30...Distribution plan generation device

Claims

1. A learning unit that learns a first statistic of a third quadratic coefficient included in an approximate rescue scale model for a plurality of locations, including a first location and a second location, based on a first quadratic coefficient included in a first sample rescue scale model and a second quadratic coefficient included in a second sample rescue scale model; A generating unit that generates the third secondary coefficient based on the learned first statistics; Equipped with the first secondary coefficient, the second secondary coefficient, and the third secondary coefficient indicate a contribution degree of a pair of the first location and the second location to the plurality of locations, The first statistic includes a mean and a variance when the third quadratic coefficient is regarded as a probability density function. Model generation device.

2. The generation unit generates a model including the third secondary coefficient, a first primary coefficient indicating a contribution degree of only the first location to the plurality of locations, and a second primary coefficient indicating a contribution degree of only the second location to the plurality of locations.

2. The model generating apparatus of claim 1.

3. The plurality of bases further includes a third base, The learning unit is learning a second statistic of an eighth quadratic coefficient included in the approximate rescue scale model based on a fourth quadratic coefficient included in the first sample rescue scale model and a fifth quadratic coefficient included in the second sample rescue scale model; A third statistic of a ninth quadratic coefficient included in the approximated rescue scale model is learned based on a sixth quadratic coefficient included in the first sample rescue scale model and a seventh quadratic coefficient included in the second sample rescue scale model; The generation unit is generating the eighth quadratic coefficient based on the learned second statistic; generating the ninth quadratic coefficient based on the learned third statistic; the fourth quadratic coefficient, the fifth quadratic coefficient, and the eighth quadratic coefficient indicate contributions of a pair of the first location and the third location to the plurality of locations, the sixth quadratic coefficient, the seventh quadratic coefficient, and the ninth quadratic coefficient indicate contributions of a pair of the second location and the third location to the plurality of locations, The second statistic includes a variance and a mean when the eighth quadratic coefficient is regarded as a probability density function, The third statistic includes a variance and a mean when the ninth quadratic coefficient is regarded as a probability density function.

3. The model generating apparatus according to claim 2.

4. The generation unit generates the model further including the eighth secondary coefficient, the ninth secondary coefficient, and a third primary coefficient indicating a contribution degree of only the third location to the plurality of locations.

4. The model generating apparatus according to claim 3.

5. The degree of contribution indicates the magnitude of the influence on the function of the communication network constituted by the plurality of bases.

2. The model generating apparatus of claim 1.

6. A model generation method in a model generation device, comprising: Among a plurality of sample rescue scale models for a plurality of locations including a first location and a second location, based on a first quadratic coefficient included in a first sample rescue scale model and a second quadratic coefficient included in a second sample rescue scale model, a first statistic of a third quadratic coefficient included in an approximate rescue scale model for the plurality of locations is learned; generating the third quadratic coefficients based on the learned first statistics; Equipped with the first secondary coefficient, the second secondary coefficient, and the third secondary coefficient indicate a contribution degree of a pair of the first location and the second location to the plurality of locations, The first statistic includes a mean and a variance when the third quadratic coefficient is regarded as a probability density function. Model generation method.

7. A model generating program for causing a computer to function as each unit of the model generating device according to any one of claims 1 to 5.

Images