Method for generating a pipeline deterioration prediction model, pipeline deterioration prediction method, program, and pipeline deterioration prediction device.

Abstract

This invention provides a method for generating a pipeline deterioration prediction model that enables more accurate prediction of pipeline deterioration. [Solution] The method for generating a pipeline deterioration prediction model comprises the steps of: extracting integrated pipeline data similar to the customer pipeline dataset as similar pipeline data from an integrated pipeline dataset which is obtained by integrating a customer pipeline dataset and a reference pipeline dataset; extracting customer pipeline data that is not similar pipeline data from the integrated pipeline dataset as first important pipeline data; creating a training pipeline dataset by combining the similar pipeline data and the first important pipeline data; and generating a pipeline deterioration prediction model using the training pipeline dataset.

Description

Technical Field

[0001] The present disclosure relates to a method for generating a pipeline aging degree prediction model, a pipeline aging degree prediction method, a program, and a pipeline aging degree prediction device.

Background Art

[0002] A pipe such as a water pipe is buried in the ground. While the pipe is used for a long time, the pipe corrodes. Japanese Patent Application Laid-Open No. 2007-107882 (Patent Document 1) discloses a pipe corrosion prediction method.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] An object of the present disclosure is to provide a method for generating a pipeline aging degree prediction model, a pipeline aging degree prediction method, a program, and a pipeline aging degree prediction device that enable more accurate prediction of the aging degree of a pipeline.

Means for Solving the Problems

[0005] The method for generating a pipeline deterioration prediction model according to this disclosure comprises the steps of creating a customer pipeline dataset and a reference pipeline dataset. The customer pipeline dataset includes multiple customer pipeline data. The multiple customer pipeline data includes the pipeline length, burial period, map data, and leakage incident history of multiple customer pipelines. The reference pipeline dataset includes multiple reference pipeline data. The multiple reference pipeline data includes the pipeline length, burial period, map data, and leakage incident history of multiple reference pipelines. The method for generating a pipeline deterioration prediction model according to this disclosure comprises the step of integrating the customer pipeline dataset and the reference pipeline dataset to create an integrated pipeline dataset consisting of multiple integrated pipeline data. The multiple integrated pipeline data includes multiple customer pipeline data and multiple reference pipeline data. The method for generating a pipeline deterioration prediction model according to this disclosure comprises the steps of: extracting integrated pipeline data from an integrated pipeline dataset that is similar to a customer pipeline dataset as similar pipeline data, and creating a similar pipeline dataset composed of the similar pipeline data; and extracting customer pipeline data from the integrated pipeline dataset that is not similar pipeline data as first important pipeline data, and creating a first important pipeline dataset composed of the first important pipeline data. The method for generating a pipeline deterioration prediction model according to this disclosure comprises the step of creating a training pipeline dataset. The step of creating a training pipeline dataset includes combining the similar pipeline dataset and the first important pipeline dataset. The method for generating a pipeline deterioration prediction model according to this disclosure comprises the step of generating a pipeline deterioration prediction model using the training pipeline dataset. The number of training pipeline data included in the training pipeline dataset is greater than the number of customer pipeline data.

[0006] The program of the first phase of this disclosure causes a processor to execute each step of the method for generating the pipeline deterioration prediction model of this disclosure.

[0007] The pipeline deterioration prediction method of this disclosure comprises the steps of creating an input pipeline dataset including pipeline length, burial period, and map data for multiple customer pipelines, and inputting the input pipeline dataset into a pipeline deterioration prediction model generated by the pipeline deterioration prediction model generation method of this disclosure to calculate the deterioration of multiple customer pipelines.

[0008] The program of the second phase of this disclosure causes a processor to execute each step of the pipeline deterioration prediction method of this disclosure.

[0009] The pipeline deterioration prediction device of this disclosure comprises an input pipeline dataset creation unit and a pipeline deterioration calculation unit that inputs the input pipeline dataset into a pipeline deterioration prediction model generated by the pipeline deterioration prediction model generation method of this disclosure and calculates the deterioration of multiple customer pipelines. The input pipeline dataset creation unit creates an input pipeline dataset that includes the pipeline length, burial period, and map data of multiple customer pipelines. The pipeline deterioration calculation unit inputs the input pipeline dataset into a pipeline deterioration prediction model generated by the pipeline deterioration prediction model generation method of this disclosure and calculates the deterioration of multiple customer pipelines. [Effects of the Invention]

[0010] According to the pipeline deterioration prediction model generation method, pipeline deterioration prediction method, program, and pipeline deterioration prediction device disclosed herein, it becomes possible to predict the degree of pipeline deterioration more accurately. [Brief explanation of the drawing]

[0011] [Figure 1] This figure shows the schematic configuration of the pipeline deterioration prediction system according to Embodiment 1 and Embodiment 2. [Figure 2] This is a schematic diagram showing the hardware configuration of the pipeline deterioration prediction model generation device of Embodiment 1 and Embodiment 2. [Figure 3] This is a block diagram illustrating the functional configuration of the pipeline deterioration prediction model generation device according to Embodiment 1. [Figure 4] This is a schematic diagram showing the storage unit of the pipeline deterioration prediction model generation device according to Embodiment 1. [Figure 5] FIG. is an example of the data structure of the original customer pipeline dataset. [Figure 6] FIG. is an example of the data structure of the original reference pipeline dataset. [Figure 7] FIG. is an example of the data structure of the learning pipeline dataset of the specified pipe material in Embodiment 1. [Figure 8] FIG. is an example of the data structure of the first customer pipeline dataset of the specified pipe material. [Figure 9] FIG. is an example of the data structure of the first reference pipeline dataset of the specified pipe material. [Figure 10] [[ID= seventeenth]]FIG. is an example of the data structure of the second customer pipeline dataset of the specified pipe material. [Figure 11] FIG. is an example of the data structure of the second reference pipeline dataset of the specified pipe material. [Figure 12] FIG. is an example of the data structure of the integrated pipeline dataset of the specified pipe material in Embodiment 1. [Figure 13] FIG. is a diagram for explaining the steps of machine learning of the pipeline leakage accident probability prediction model when the pipeline leakage accident probability prediction model is a neural network in Embodiment 1. [Figure 14] FIG. is a flowchart showing a method for generating a pipeline aging degree prediction model in Embodiment 1 and Embodiment 2. [Figure 15] [[ID=3二十]]FIG. is a flowchart showing the pipeline data preprocessing step in Embodiment 1. [Figure 16] FIG. is an example of an integrated pipeline dataset in a multi-dimensional space defined by the dimension of explanatory variables. [Figure 17] FIG. is an example of a flowchart of the similar pipeline data extraction step in Embodiment 1. <000009 forty]]FIG. is another example of a flowchart of the similar pipeline data extraction step in Embodiment 1. [Figure 19] FIG. is an example of a similar pipeline dataset in a multi-dimensional space defined by the dimension of explanatory variables. [Figure 20] [Figure 20]This figure shows another example of the flowchart for the critical pipeline data extraction step in Embodiment 1. [Figure 21] This figure shows an example of a first important similar pipeline dataset in a multidimensional space defined by the dimensions of the explanatory variables. [Figure 22] This figure shows another example of the flowchart for the second critical pipeline data extraction step in Embodiment 1. [Figure 23] This figure shows an example of the integrated pipeline data extracted in step S25. [Figure 24] This figure shows an example of a second important similar pipeline dataset in a multidimensional space defined by the dimensions of the explanatory variables. [Figure 25] This figure shows an example of a flowchart for the second important pipeline data extraction step in Embodiment 1. [Figure 26] This figure shows another example of the flowchart for the second critical pipeline data extraction step in Embodiment 1. [Figure 27] This figure shows an example of a training pipeline dataset in a multidimensional space defined by the dimensions of the explanatory variables. [Figure 28] This figure shows another example of the flowchart for the pipeline deterioration prediction model generation step in Embodiment 1. [Figure 29] This figure shows an example of the data structure for a leak accident prediction probability table. [Figure 30] This figure shows an example of the relationship between the predicted probability of a water leak accident calculated by a water leak accident prediction model, the length of the pipe, and the actual number of water leak accidents per unit time. [Figure 31] This figure shows an example of the relationship between the predicted probability of a water leak accident calculated by a pipeline water leak accident probability prediction model and the actual water leak accident rate. [Figure 32] This is a schematic diagram showing the hardware configuration of the pipeline deterioration prediction device of Embodiment 1 and Embodiment 2. [Figure 33] This is a schematic diagram showing the functional configuration of the pipeline deterioration prediction device according to Embodiment 1. [Figure 34] This is a schematic diagram showing the memory unit of the pipeline deterioration prediction device according to Embodiment 1. [Figure 35] This figure shows examples of pipeline deterioration prediction results for Embodiment 1 and Embodiment 2. [Figure 36] This figure shows another example of the pipeline deterioration prediction results for Embodiment 1 and Embodiment 2. [Figure 37] This diagram shows flowcharts of the pipeline deterioration prediction methods for Embodiment 1 and Embodiment 2. [Figure 38] This diagram shows a flowchart of the customer pipeline data processing step in Embodiment 1. [Figure 39] This is a block diagram illustrating the functional configuration of the pipeline deterioration prediction model generation device according to Embodiment 2. [Figure 40] This is a schematic diagram showing the storage unit of the pipeline deterioration prediction model generation device of Embodiment 2. [Figure 41] This is a schematic diagram showing an example of a pipeline deterioration prediction model according to Embodiment 2. [Figure 42] This is a schematic diagram showing an example of a general-purpose pipeline deterioration prediction model according to Embodiment 2. [Figure 43] This figure shows an example of the data structure of the learning pipeline dataset for specified pipe materials in Embodiment 2. [Figure 44] This figure shows an example of the data structure for the third customer pipeline dataset of specified pipe materials. [Figure 45] This figure shows an example of the data structure for the third pre-processed reference pipeline data of the specified pipe material. [Figure 46] This figure shows an example of the data structure of the integrated pipeline dataset for specified pipe materials in Embodiment 2. [Figure 47] This diagram shows a flowchart illustrating the method for generating a general-purpose pipeline deterioration prediction model according to Embodiment 2. [Figure 48] This diagram shows a flowchart of the reference pipeline data preprocessing step in Embodiment 2. [Figure 49] This figure shows another example of the flowchart for the general-purpose pipeline deterioration prediction model generation step in Embodiment 2. [Figure 50] This diagram shows a flowchart of the pipeline data preprocessing step in Embodiment 2. [Figure 51] This diagram shows a flowchart of the third pipeline data preprocessing step in Embodiment 2. [Figure 52] This diagram shows a flowchart of the third customer pipeline dataset creation step in Embodiment 2. [Figure 53] This diagram shows a flowchart of the third reference pipeline dataset creation step in Embodiment 2. [Figure 54] This figure illustrates the steps for machine learning a pipeline leak 0.2+ water accident probability prediction model in Embodiment 2, where the pipeline leak accident probability prediction model is a neural network. [Figure 55] This is a schematic diagram showing the functional configuration of the pipeline deterioration prediction device according to Embodiment 2. [Figure 56] This is a schematic diagram showing the memory unit of the pipeline deterioration prediction device of Embodiment 2. [Figure 57] This diagram shows a flowchart of the customer pipeline data processing step in Embodiment 2. [Figure 58] This diagram shows a flowchart of the steps for creating third customer pipeline data for specified pipe materials in the pipeline deterioration prediction method of Embodiment 2. [Modes for carrying out the invention]

[0012] Embodiments of the present disclosure will be described below. The same components will be given the same reference numerals, and their descriptions will not be repeated.

[0013] (Embodiment 1) <Pipeline Deterioration Prediction System 1> Referring to Figure 1, the pipeline deterioration prediction system 1 of this embodiment will be described. The pipeline deterioration prediction system 1 comprises a pipeline deterioration prediction model generation device 2 and a pipeline deterioration prediction device 3.

[0014] <Pipeline Deterioration Prediction Model Generation Device 2> Referring to Figure 1, the pipeline deterioration prediction model generation device 2 generates a pipeline deterioration prediction model 5 (see Figure 4) from the original customer pipeline dataset 31 (see Figure 5) and the original reference pipeline dataset 35 (see Figure 6).

[0015] <Hardware Configuration> Referring to Figure 2, the hardware configuration of the pipeline deterioration prediction model generation device 2 will be described. The pipeline deterioration prediction model generation device 2 includes an input device 201, a processor 202, memory 203, a display 204, a network controller 206, a storage medium drive 207, and storage 210.

[0016] The input device 201 accepts various input operations. The input device 201 is, for example, a keyboard, a mouse, or a touch panel.

[0017] The display 204 displays information necessary for processing in the pipeline deterioration prediction model generation device 2. The display 204 is, for example, an LCD (Liquid Crystal Display) or an organic EL (Electroluminescence) display.

[0018] The processor 202 executes the processing necessary to realize the functions of the pipeline aging prediction model generation device 2 by running the program described later. The processor 202 is composed of, for example, a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit).

[0019] Memory 203 provides a storage area for the processor 202 to temporarily store program code or work memory when executing a program described later. Memory 203 is, for example, a volatile memory device such as DRAM (Dynamic Random Access Memory) or SRAM (Static Random Access Memory).

[0020] The network controller 206 transmits and receives programs or data to and from any device, including the pipeline deterioration prediction device 3, via a communication network 4 (see Figure 1), such as the Internet or an intranet. For example, the network controller 206 transmits the pipeline deterioration prediction model 5 (see Figure 4) to the pipeline deterioration prediction device 3 (see Figure 1) via the communication network 4. The network controller 206 supports any communication method, such as Ethernet®, Wi-Fi (Local Area Network), or Bluetooth®.

[0021] The storage medium drive 207 is a device that reads programs or data stored in the storage medium 208. The storage medium drive 207 may also be a device that writes programs or data to the storage medium 208. The storage medium 208 is a non-transitory storage medium that stores programs or data non-volatilely. The storage medium 208 is, for example, an optical storage medium such as an optical disc (e.g., CD-ROM or DVD-ROM), a semiconductor storage medium such as flash memory or USB (Universal Serial Bus) memory, a magnetic storage medium such as a hard disk, FD (Flexible Disk) or storage tape, or a magneto-optical storage medium such as an MO (Magneto-Optical) disk.

[0022] Storage 210 stores the original customer pipeline dataset 31 (see Figure 5), the original reference pipeline dataset 35 (see Figure 6), the pipeline deterioration prediction model 5 (see Figure 4), and programs executed by the processor 202. This program includes the pipeline deterioration prediction model generation program 8 (see Figure 4). Storage 210 is, for example, a non-volatile memory device such as a hard disk or an SSD (Solid State Drive).

[0023] The program for realizing the functions of the pipeline aging prediction model generation device 2 may be stored and distributed on a non-transient storage medium 208 and installed on the storage 210. The program for realizing the functions of the pipeline aging prediction model generation device 2 may also be downloaded to the pipeline aging prediction model generation device 2 via a communication network 4 such as the Internet or an intranet. The program for realizing the functions of the pipeline aging prediction model generation device 2 includes the pipeline aging prediction model generation program 8 (see Figure 4).

[0024] In this embodiment, an example is shown in which a general-purpose computer (processor 202) implements the functions of the pipeline aging prediction model generation device 2 by executing a program. However, the embodiment is not limited to this, and all or part of the functions of the pipeline aging prediction model generation device 2 may be implemented using an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field-Programmable Gate Array).

[0025] <Functional Configuration> Referring to Figure 3, the functional configuration of the pipeline deterioration prediction model generation device 2 will be explained. The pipeline deterioration prediction model generation device 2 includes a pipeline data set receiving unit 10, a pipe material specification unit 11, a training data set creation unit 12, a pipeline deterioration prediction model generation unit 20, and a storage unit 25.

[0026] <Storage section 25> The memory unit 25 is implemented by, for example, at least one of a storage device 210 (see Figure 2) or a storage medium 208 (see Figure 2). Referring to Figure 4, the memory unit 25 stores the original customer pipeline dataset 31, the original reference pipeline dataset 35, the learning pipeline dataset 48, the map database 40, the pipeline deterioration prediction model 5, and the pipeline deterioration prediction model generation program 8.

[0027] Referring to Figure 5, the original customer pipeline dataset 31 is provided by the customer (water utility). The original customer pipeline dataset 31 is a set of multiple original customer pipeline data, and includes multiple original customer pipeline data. The multiple original customer pipeline data includes, for example, customer pipeline ID, data acquisition year, utility name, attribute data 31a, location 31p, and leakage accident history. The attribute data 31a of the customer pipeline includes, for example, the year of installation, diameter, pipe material, and pipeline length of the customer pipeline. The location 31p of the customer pipeline includes, for example, the latitude and longitude of the customer pipeline. The leakage accident history of the customer pipeline indicates whether or not there were any leakage accidents in the customer pipeline during a predetermined period including the data acquisition year (for example, the year of data acquisition, or the period from the data acquisition year to five years prior to the data acquisition year). If a leakage accident occurred in the customer pipeline during the predetermined period including the data acquisition year, "1" is entered in the customer pipeline leakage accident history column. If no water leakage incidents occurred in the customer's pipeline during the specified period including the year in which the data was acquired, "0" will be entered in the customer pipeline water leakage incident history field.

[0028] Referring to Figure 6, the original reference pipeline dataset 35 is a pipeline dataset collected from a wider area than the original customer pipeline dataset 31. The original reference pipeline dataset 35 is obtained, for example, from multiple entities throughout Japan. The original reference pipeline dataset 35 is a set of multiple original reference pipeline data, and contains multiple original reference pipeline data. The number of multiple original reference pipeline data included in the original reference pipeline dataset 35 is greater than the number of multiple original customer pipeline data included in the original customer pipeline dataset 31.

[0029] Multiple original reference pipeline data includes, for example, the reference pipeline ID, data acquisition year, business entity name, attribute data 35a, location 35p, and leak accident history. The attribute data 35a of the reference pipeline includes, for example, the year of installation, diameter, pipe material, and pipeline length. The location 35p of the reference pipeline includes, for example, the latitude and longitude of the reference pipeline. The location 35p of the reference pipeline is different from the location 31p of the customer pipeline. The leak accident history of the reference pipeline indicates whether or not there were any leak accidents in the reference pipeline during a predetermined period including the data acquisition year (for example, the year of data acquisition, or the period from the data acquisition year to five years prior to the data acquisition year). If a leak accident occurred in the reference pipeline during the predetermined period including the data acquisition year, "1" is entered in the reference pipeline leak accident history column. If no leak accidents occurred in the reference pipeline during the predetermined period including the data acquisition year, "0" is entered in the reference pipeline leak accident history column. The original reference pipeline dataset 35 is pre-stored in the storage unit 25 (see Figures 3 and 4).

[0030] Referring to Figure 7, the training pipeline dataset 48 is a set of multiple training pipeline data, and includes multiple training pipeline data. The multiple training pipeline data includes, for example, a training pipeline ID, a business entity name, attribute data 48a, location 48p, leakage accident history, and map data 48m. The pipeline attribute data 48a includes, for example, the pipeline diameter, pipe material, pipeline length, and burial period. The pipeline burial period is calculated, for example, by subtracting the pipeline laying year from the data acquisition year. The pipeline location 48p includes, for example, the pipeline latitude and longitude. The pipeline map data 48m includes, for example, the pipeline soil classification, as well as the annual average temperature and annual average precipitation at the pipeline location 48p. The number of multiple training pipeline data included in the training pipeline dataset 48 is greater than the number of multiple original customer pipeline data included in the original customer pipeline dataset 31.

[0031] Referring to Figure 4, the map database 40 is provided by public institutions, for example, through the Internet or storage medium 208, such as a Geographic Information System (GIS), and is pre-stored in the storage unit 25 (see Figures 3 and 4). The map database 40 includes, for example, a soil classification map 41, an annual average temperature map 42, and an annual average precipitation map 43. The soil classification map 41 includes location (latitude and longitude) and soil classification at that location. The annual average temperature map 42 includes location (latitude and longitude) and annual average temperature at that location. The annual average precipitation map 43 includes location (latitude and longitude) and annual average precipitation at that location.

[0032] Referring to Figure 4, the pipeline deterioration prediction model 5 includes pipeline deterioration prediction models 5a and 5b for each pipe material. For example, pipeline deterioration prediction model 5 includes pipeline deterioration prediction model 5a for rigid polyvinyl chloride (VP) pipes and pipeline deterioration prediction model 5b for ductile cast iron (DIP) pipes. Pipeline deterioration prediction model 5 outputs the degree of pipeline deterioration.

[0033] The degree of pipeline deterioration is not particularly limited, but for example, it is the pipeline leakage accident rate. The pipeline leakage accident rate means, for example, the number of leakage accidents per unit time per year and per unit pipeline length of 1 km. The unit of the pipeline leakage accident rate is, for example, accidents / year / km. The pipeline deterioration prediction model 5 includes, for example, pipeline leakage accident probability prediction models 6a,6b (see Figure 4) and conversion units 7a,7b (see Figure 4). For example, the pipeline deterioration prediction model 5a for VP pipes includes the VP pipe leakage accident probability prediction model 6a and the VP pipe conversion unit 7a. The pipeline deterioration prediction model 5b for DIP pipes includes the DIP pipe leakage accident probability prediction model 6b and the DIP pipe conversion unit 7b. The pipeline leakage accident probability prediction models 6a,6b output the predicted probability of pipeline leakage accidents. The predicted probability of a pipeline leak is the probability of a pipeline leak predicted by the pipeline leak probability prediction models 6a and 6b. The conversion units 7a and 7b convert the predicted probability of a pipeline leak into the pipeline leak rate.

[0034] Referring to Figure 4, the pipeline deterioration prediction model generation program 8 is a program for generating a pipeline deterioration prediction model 5 (see Figure 4) using the original customer pipeline dataset 31 (see Figure 5) and the original reference pipeline dataset 35 (see Figure 6).

[0035] <Pipeline Data Set Reception Unit 10> Referring to Figure 3, the pipeline data set receiving unit 10 receives the original customer pipeline data set 31 (see Figure 5) from the customer (water utility). The original customer pipeline data set 31 is provided by the customer, for example, via a storage medium 208 (see Figure 2) or through a communication network 4 such as the Internet (see Figure 1). The pipeline data set receiving unit 10 outputs the original customer pipeline data set 31 to the storage unit 25 (see Figures 3 and 4). The original customer pipeline data set 31 is stored in the storage unit 25.

[0036] <Pipe material specification section 11> Referring to Figure 3, the pipe material specification unit 11 specifies the pipe material for which the pipeline deterioration prediction models 5a and 5b should be created. The pipe material specified by the pipe material specification unit 11 may hereafter be referred to as the "specified pipe material". The pipe material specification unit 11 may also be implemented by the input device 201 (see Figure 2).

[0037] <Training Dataset Creation Section 12> Referring to Figure 3, the training dataset creation unit 12 generates a training pipeline dataset 48 (see Figure 7) from the original customer pipeline dataset 31 (see Figure 5) and the original reference pipeline dataset 35 (see Figure 6). The training pipeline dataset 48 is used to generate the pipeline deterioration prediction model 5. The training dataset creation unit 12 includes a pipeline data preprocessing unit 13, a pipeline data integration unit 15, a similar pipeline data extraction unit 16, a critical pipeline data extraction unit 17, and a pipeline data merging unit 18.

[0038] <Pipeline data preprocessing unit 13> Referring to Figure 3, the pipeline data preprocessing unit 13 creates a preprocessed customer pipeline dataset for a specified pipe material from the original customer pipeline dataset 31 (see Figure 5), and also creates a preprocessed reference pipeline dataset for a specified pipe material from the original reference pipeline dataset 35 (see Figure 6). The preprocessed customer pipeline dataset is a set of multiple preprocessed customer pipeline data and contains multiple preprocessed customer pipeline data. The preprocessed reference pipeline dataset is a set of multiple preprocessed reference pipeline data and contains multiple preprocessed reference pipeline data. In this embodiment, the preprocessed customer pipeline dataset for a specified pipe material is the second customer pipeline dataset 33 (see Figure 10), and the preprocessed reference pipeline dataset for a specified pipe material is the second reference pipeline dataset 37 (see Figure 11). The pipeline data preprocessing unit 13 includes, for example, a first pipeline data preprocessing unit 13a and a second pipeline data preprocessing unit 13b.

[0039] <First pipeline data preprocessing unit 13a> Referring to Figure 3, the first pipeline data preprocessing unit 13a extracts the original customer pipeline data for the specified pipe material from the original customer pipeline dataset 31 (see Figure 5) to create the first customer pipeline dataset 32 ​​for the specified pipe material (see Figure 8). The specified pipe material is the pipe material specified by the pipe material specification unit 11. For example, if the specified pipe material is a VP pipe, the first pipeline data preprocessing unit 13a extracts the original customer pipeline data for the VP pipe from the original customer pipeline dataset 31 to create the first customer pipeline dataset 32 ​​for the VP pipe.

[0040] The first pipeline data preprocessing unit 13a extracts the original reference pipeline data for the specified pipe material from the original reference pipeline dataset 35 (see Figure 6) to create the first reference pipeline dataset 36 for the specified pipe material (see Figure 9). For example, if the specified pipe material is a VP pipe, the first pipeline data preprocessing unit 13a extracts the original reference pipeline data for the VP pipe from the original reference pipeline dataset 35 to create the first reference pipeline dataset 36 for the VP pipe.

[0041] <Second pipeline data preprocessing unit 13b> Referring to Figure 3, the second pipeline data preprocessing unit 13b (see Figure 3) creates a second customer pipeline dataset 33 (see Figure 10) for the specified pipe material from the first customer pipeline dataset 32 ​​(see Figure 8) for the specified pipe material. The second customer pipeline dataset 33 includes, for example, the customer pipeline ID, the business name, attribute data 31b, location 31p, leakage accident history, and map data 31m. The attribute data 31b of the customer pipeline includes, for example, the diameter, pipe material, pipeline length, and burial period of the customer pipeline. The map data 31m of the customer pipeline includes, for example, the soil classification of the customer pipeline, as well as the average annual temperature and average annual precipitation at the location 31p of the customer pipeline.

[0042] The second pipeline data preprocessing unit 13b (see Figure 3) creates a second reference pipeline dataset 37 (see Figure 11) for the specified pipe material from the first reference pipeline dataset 36 (see Figure 9) for the specified pipe material. The second reference pipeline dataset 37 includes, for example, the reference pipeline ID, the name of the utility company, attribute data 35b, location 35p, leakage accident history, and map data 35m. The attribute data 35b of the reference pipeline includes, for example, the diameter, pipe material, pipeline length, and burial period of the reference pipeline. The map data 35m of the reference pipeline includes, for example, the soil classification of the reference pipeline, as well as the average annual temperature and average annual precipitation at the location 35p of the reference pipeline.

[0043] <Pipeline Data Integration Unit 15> Referring to Figure 3, the pipeline data integration unit 15 integrates the pre-processed customer pipeline dataset and the pre-processed reference pipeline dataset for the specified pipe material to create an integrated pipeline dataset 38 for the specified pipe material (see Figure 12). In this embodiment, the pre-processed customer pipeline dataset for the specified pipe material is the second customer pipeline dataset 33 (see Figure 10), and the pre-processed reference pipeline dataset for the specified pipe material is the second reference pipeline dataset 37 (see Figure 11). The integrated pipeline dataset 38 includes multiple integrated pipeline data sets, each composed of multiple pre-processed customer pipeline data sets and multiple pre-processed reference pipeline data sets. The multiple integrated pipeline data sets include, for example, an integrated pipeline ID, a business entity name, attribute data 38b, a leak accident history, and map data 38m. The multiple integrated pipeline data sets may further include location data 38p. The pipeline data integration unit 15 assigns a new integrated pipeline ID to each integrated pipeline data set. The pipeline attribute data 38b includes, for example, the pipeline diameter, pipe material, pipeline length, and burial period. The pipeline map data 38m includes, for example, the soil classification of the pipeline, as well as the average annual temperature and average annual precipitation at the pipeline location 38p.

[0044] <Similar pipeline data extraction unit 16> Referring to Figure 3, the similar pipeline data extraction unit 16 extracts integrated pipeline data from the integrated pipeline dataset 38 (see Figure 12) of the specified pipe material that are similar to the pre-processed customer pipeline dataset of the specified pipe material, and creates a similar pipeline dataset composed of the similar pipeline data. In this embodiment, the pre-processed customer pipeline dataset of the specified pipe material is the second customer pipeline dataset 33 (see Figure 10). Two methods can be exemplified as methods for extracting similar pipeline data.

[0045] In the first example, the similar pipeline data extraction unit 16 calculates the data center of the pre-processed customer pipeline dataset included in the integrated pipeline dataset 38 (see Figure 12) in a multidimensional space defined by the dimensions of the explanatory variables (features) of the integrated pipeline dataset 38. In this specification, the data center of the pipeline dataset is given by the average value of the coordinates of all pipeline data included in the pipeline dataset in a multidimensional space defined by the dimensions (or number of explanatory variables) of the explanatory variables of the pipeline dataset. In this embodiment, the explanatory variables of the integrated pipeline dataset 38 are, for example, pipeline attribute data 38b and pipeline map data 38m. The similar pipeline data extraction unit 16 calculates the distance (Euclidean distance, Mahalanobis distance, Manhattan distance, or Chebyshev distance, etc.) between each integrated pipeline data and the data center of the pre-processed customer pipeline dataset in the multidimensional space. The similar pipeline data extraction unit 16 extracts integrated pipeline data for which this distance is less than or equal to a reference distance as similar pipeline data.

[0046] In the second example, the similar pipeline data extraction unit 16 calculates the data center of the pre-processed customer pipeline dataset included in the integrated pipeline dataset 38 (see Figure 12) in a multidimensional space defined by the dimensions of the explanatory variables of the integrated pipeline dataset 38. The similar pipeline data extraction unit 16 calculates the cosine similarity between each integrated pipeline data and the data center of the pre-processed customer pipeline dataset in the above multidimensional space. The similar pipeline data extraction unit 16 extracts integrated pipeline data whose cosine similarity is equal to or greater than a threshold value as similar pipeline data.

[0047] <Important pipeline data extraction unit 17> Referring to Figure 3, the critical pipeline data extraction unit 17 extracts critical pipeline data from the integrated pipeline dataset 38 (see Figure 12) to create a critical pipeline dataset composed of critical pipeline data. The critical pipeline data is integrated pipeline data from the integrated pipeline dataset 38 that is important for improving the prediction accuracy of the pipeline deterioration prediction model 5. The critical pipeline data extraction unit 17 includes, for example, a first critical pipeline data extraction unit 17a and a second critical pipeline data extraction unit 17b. The critical pipeline data includes first critical pipeline data extracted by the first critical pipeline data extraction unit 17a and second critical pipeline data extracted by the second critical pipeline data extraction unit 17b. The critical pipeline dataset includes a first critical pipeline dataset composed of first critical pipeline data and a second critical pipeline dataset composed of second critical pipeline data.

[0048] <First Important Pipeline Data Extraction Unit 17a> The first critical pipeline data extraction unit 17a extracts pre-processed customer pipeline data that was not extracted by the similar pipeline data extraction unit 16 from the integrated pipeline dataset 38 (see Figure 12) of the specified pipe material, and creates a first critical pipeline dataset consisting of the first critical pipeline data. The first critical pipeline data extraction unit 17a extracts the first critical pipeline data because the pre-processed customer pipeline data that was not extracted by the similar pipeline data extraction unit 16 is also important for improving the prediction accuracy of the pipeline deterioration prediction model 5. In this embodiment, the pre-processed customer pipeline dataset is the second customer pipeline dataset 33 (see Figure 10).

[0049] <Second important pipeline data extraction unit 17b> The second critical pipeline data extraction unit 17b extracts integrated pipeline data similar to the first critical pipeline data set from the integrated pipeline data set 38 (see Figure 12) of the specified pipe material as second critical pipeline data, and creates a second critical pipeline data set consisting of the second critical pipeline data. Since integrated pipeline data similar to the first critical pipeline data set from the integrated pipeline data set 38 is also important for improving the prediction accuracy of the pipeline deterioration prediction model 5, the second critical pipeline data extraction unit 17b extracts the second critical pipeline data.

[0050] For example, the second important pipeline data extraction unit 17b calculates the variability of each explanatory variable in the first important pipeline dataset. The second important pipeline data extraction unit 17b selects the explanatory variable with the least variability. The second important pipeline data extraction unit 17b calculates the mode of the selected explanatory variable. The second important pipeline data extraction unit 17b extracts integrated pipeline data from the integrated pipeline dataset 38 (see Figure 12) in which the selected explanatory variable is the mode. The extracted integrated pipeline data includes the first important pipeline data in which the selected explanatory variable is the mode, and pre-processed reference pipeline data in which the selected explanatory variable is the mode. From the extracted integrated pipeline data, the second important pipeline data extraction unit 17b extracts integrated pipeline data similar to the first important pipeline dataset in which the selected explanatory variable is the mode, as second important pipeline data, and creates a second important pipeline dataset composed of the second important pipeline data. Two methods can be exemplified for extracting data for the second most important pipeline from integrated pipeline data where the selected explanatory variable is the mode.

[0051] In the first example, the second important pipeline data extraction unit 17b calculates the data center of the first important pipeline dataset where the selected explanatory variable is the mode, in a multidimensional space defined by the dimensions of the explanatory variables (excluding the explanatory variable with the least variability) of the integrated pipeline dataset 38 (see Figure 12). The second important pipeline data extraction unit 17b calculates the distance (Euclidean distance, Mahalanobis distance, Manhattan distance, or Chebyshev distance, etc.) between each integrated pipeline data where the selected explanatory variable is the mode and the data center of the first important pipeline dataset where the selected explanatory variable is the mode, in the above multidimensional space. The second important pipeline data extraction unit 17b extracts integrated pipeline data where this distance is less than or equal to the reference distance as the second important pipeline data.

[0052] In the second example, the second important pipeline data extraction unit 17b calculates the data center of the first important pipeline dataset where the selected explanatory variable is the mode, in a multidimensional space defined by the dimensions of the explanatory variables (excluding the explanatory variable with the least variability) of the integrated pipeline dataset 38 (see Figure 12). The second important pipeline data extraction unit 17b calculates the cosine similarity between each of the integrated pipeline data where the selected explanatory variable is the mode and the data center of the first important pipeline dataset where the selected explanatory variable is the mode, in the above multidimensional space. The second important pipeline data extraction unit 17b extracts the integrated pipeline data where the cosine similarity is equal to or greater than a threshold value as the second important pipeline data.

[0053] <Pipeline data connection section 18> Referring to Figure 3, the pipeline data merging unit 18 combines the similar pipeline dataset extracted by the similar pipeline data extraction unit 16 with the important pipeline dataset extracted by the important pipeline data extraction unit 17 to create a learning pipeline dataset 48 for the specified pipe material (see Figure 7). The pipeline data merging unit 18 assigns a new pipeline ID (learning pipeline ID) to the learning pipeline data that constitutes the learning pipeline dataset 48.

[0054] <Pipeline Deterioration Prediction Model Generation Unit 20> Referring to Figure 3, the pipeline deterioration prediction model generation unit 20 generates a pipeline deterioration prediction model 5 (see Figure 4) using the training pipeline dataset 48 (see Figure 7). The pipeline deterioration prediction model generation unit 20 includes, for example, a pipeline leakage accident probability prediction model learning unit 21, a conversion unit generation unit 22, and a pipeline deterioration prediction model generation completion determination unit 23.

[0055] <Pipeline Leakage Accident Probability Prediction Model Learning Unit 21> Referring to Figure 3, the pipeline leakage accident probability prediction model learning unit 21 generates pipeline leakage accident probability prediction models 6a and 6b using machine learning with the learning pipeline dataset 48 (see Figure 7) as training data.

[0056] The leakage accident history is entered as either "0" or "1". It is rare for multiple leakage accidents to occur in a single pipeline during a predetermined period including the data acquisition year. Therefore, the pipeline leakage accident history can be considered as the probability of pipeline leakage accidents during a predetermined period including the data acquisition year. Accordingly, as shown in Figure 13, explanatory variables (pipe attribute data 48a (e.g., diameter, pipeline length, and burial period) and map data 48m (e.g., soil classification, average annual temperature, and average annual precipitation)) from the training pipeline dataset 48 (see Figure 7) for the specified pipe material are input into the prediction model. The prediction model outputs the predicted probability of pipeline leakage accidents. The prediction model is trained using machine learning to generate pipeline leakage accident probability prediction models 6a and 6b so that the predicted probability of pipeline leakage accidents approaches the pipeline leakage accident history included in the training pipeline dataset 48.

[0057] For example, the prediction model is a neural network model 50 as shown in Figure 13. The weights W and bias B of the neural network model 50 are determined by training the neural network using the training conduit dataset 48 (see Figure 7). The weights W of the neural network model 50 are a vector of the weights of each node in the neural network (w1, w2, w3, ...). The bias B of the neural network model 50 is the bias of each node in the neural network. The relationship between the input and output at each node of the neural network is expressed by the following equation (1). x j =σ(w1x1+w2x2+…+w n x n +B j ) …(1) Here, (x1, x2, ..., x n ) is the input to node j, and x j This is the output of node j, and B j σ is the bias of node j. σ is a function that represents the action of node j, such as the sigmoid function or the ReLU (Rectified Linear Unit) function.

[0058] Then, the weights W and bias B of the neural network model 50 are determined so as to minimize the error between the predicted probability of pipeline leaks output by the prediction model and the pipeline leak history included in the training pipeline dataset 48. In this way, pipeline leak probability prediction models 6a and 6b are generated as prediction models with optimized weights W and bias B.

[0059] In the example in Figure 13, the prediction model is shown as a three-layer neural network model 50, but it may be a neural network model with more layers. Furthermore, the prediction model is not limited to neural networks; it may be other machine learning models (e.g., random forests).

[0060] <Conversion Unit / Generation Unit 22> Referring to Figure 3, the conversion unit generation unit 22 generates conversion units 7a and 7b (see Figure 4). The conversion units 7a and 7b convert the predicted probability of pipeline leakage accidents calculated by the pipeline leakage accident probability prediction models 6a and 6b into the pipeline leakage accident rate.

[0061] <Pipeline deterioration prediction model generation completion determination unit 23> Referring to Figure 3, the pipeline deterioration prediction model generation completion determination unit 23 determines whether it has generated pipeline deterioration prediction models 5a and 5b for all pipe materials included in the original customer pipeline dataset 31 (see Figure 5). If the pipeline deterioration prediction model generation completion determination unit 23 determines that it has not generated pipeline deterioration prediction models 5a and 5b for all pipe materials included in the original customer pipeline dataset 31, the pipe material specification unit 11 (see Figure 3) specifies the pipe materials for which pipeline deterioration prediction models 5a and 5b have not yet been created among all pipe materials included in the original customer pipeline dataset 31. If the pipeline deterioration prediction model generation completion determination unit 23 determines that it has generated pipeline deterioration prediction models 5a and 5b for all pipe materials included in the original customer pipeline dataset 31, it terminates the generation of pipeline deterioration prediction model 5.

[0062] <Method for generating pipeline deterioration prediction model 5> Referring to Figures 14 to 31, the method for generating the pipeline deterioration prediction model 5 of this embodiment will be explained. Referring to Figure 14, the method for generating the pipeline deterioration prediction model 5 of this embodiment includes, for example, a customer pipeline dataset reception step S1, a pipe material specification step S2, a pipeline data preprocessing step S3, an integrated pipeline data creation step S4, a similar pipeline data extraction step S5, a critical pipeline data extraction step S6, a training pipeline dataset creation step S7, a pipeline deterioration prediction model generation step S8, and a model generation completion determination step S9.

[0063] Referring to Figure 14, in the original customer pipeline dataset reception step S1, the pipeline dataset reception unit 10 (see Figure 3) receives the original customer pipeline dataset 31 (see Figure 5) from the customer (water utility). The pipeline dataset reception unit 10 outputs the original customer pipeline dataset 31 to the storage unit 25 (see Figures 3 and 4). The original customer pipeline dataset 31 is stored in the storage unit 25.

[0064] Referring to Figure 14, in pipe material specification step S2, the pipe material specification unit 11 (see Figure 3) specifies the pipe material for which the pipeline deterioration prediction model 5 should be created. The pipe material specified in pipe material specification step S2 is one of the pipe materials included in the original customer pipeline dataset 31 (see Figure 5), and is called the specified pipe material. For example, VP pipe is specified as the pipe material for which the pipeline deterioration prediction model 5 should be created.

[0065] Referring to Figure 14, in pipeline data preprocessing step S3, the pipeline data preprocessing unit 13 creates a second customer pipeline dataset 33 for specified pipe materials (see Figure 10) and a second reference pipeline dataset 37 for specified pipe materials (see Figure 11) from the original customer pipeline dataset 31 (see Figure 5) and the original reference pipeline dataset 35 (see Figure 6). Referring to Figure 15, pipeline data preprocessing step S3 includes a first pipeline data preprocessing step S3a and a second pipeline data preprocessing step S3b.

[0066] In the first pipeline data preprocessing step S3a, the first pipeline data preprocessing unit 13a (see Figure 3) extracts the original customer pipeline data for the specified pipe material from the original customer pipeline dataset 31 (see Figure 5) to create the first customer pipeline dataset 32 ​​for the specified pipe material (see Figure 8). For example, if the specified pipe material is a VP pipe, the first pipeline data preprocessing unit 13a extracts the original customer pipeline data for VP pipes from the original customer pipeline dataset 31 to create the first customer pipeline dataset 32 ​​for VP pipes.

[0067] In the first pipeline data preprocessing step S3a, the first pipeline data preprocessing unit 13a (see Figure 3) extracts the original reference pipeline data for the specified pipe material from the original reference pipeline dataset 35 (see Figure 6) to create the first reference pipeline dataset 36 for the specified pipe material (see Figure 9). For example, if the specified pipe material is a VP pipe, the first pipeline data preprocessing unit 13a extracts the original reference pipeline data for the VP pipe from the original reference pipeline dataset 35 to create the first reference pipeline dataset 36 for the VP pipe.

[0068] In the second pipeline data preprocessing step S3b, the second pipeline data preprocessing unit 13b (see Figure 3) creates a second customer pipeline dataset 33 (see Figure 10) for the specified pipe material from the first customer pipeline dataset 32 ​​(see Figure 8) for the specified pipe material. The second customer pipeline dataset 33 includes, for example, the business name, customer pipeline ID, attribute data 31b, location 31p, leakage accident history, and map data 31m. The attribute data 31b of the customer pipeline includes, for example, the diameter, pipe material, pipeline length, and burial period of the customer pipeline. The map data 31m of the customer pipeline includes, for example, the soil classification of the customer pipeline, as well as the average annual temperature and average annual precipitation at the location 31p of the customer pipeline. In this embodiment, the second customer pipeline dataset 33 is a preprocessed customer pipeline dataset.

[0069] The second pipeline data preprocessor 13b calculates the buried period of the customer pipeline by subtracting the laying year of the customer pipeline from the data acquisition year of the first customer pipeline dataset 32 ​​of the specified pipe material. The second pipeline data preprocessor 13b identifies the map data 31m of the customer pipeline by referring to the latitude and longitude of the customer pipeline and the map database 40 (see Figure 4). Specifically, the second pipeline data preprocessor 13b identifies the soil classification of the customer pipeline by referring to the latitude and longitude of the customer pipeline and the soil classification map 41 (see Figure 4). The second pipeline data preprocessor 13b identifies the annual average temperature at the customer pipeline location 31p by referring to the latitude and longitude of the customer pipeline and the annual average temperature map 42 (see Figure 4). The second pipeline data preprocessor 13b identifies the annual average precipitation at the customer pipeline location 31p by referring to the latitude and longitude of the customer pipeline and the annual average precipitation map 43 (see Figure 4). The second pipeline data preprocessing unit 13b combines the buried period and map data 31m of the customer pipeline with the first customer pipeline dataset 32 ​​of the specified pipe material. The second pipeline data preprocessing unit 13b deletes the data acquisition year and the laying year (see Figure 8). In this way, the second customer pipeline dataset 33 is obtained.

[0070] In the second pipeline data preprocessing step S3b, the second pipeline data preprocessing unit 13b (see Figure 3) creates a second reference pipeline dataset 37 (see Figure 11) for the specified pipe material from the first reference pipeline dataset 36 (see Figure 9) for the specified pipe material. The second reference pipeline dataset 37 is a set of multiple second reference pipeline data, and includes multiple second reference pipeline data. The multiple second reference pipeline data includes, for example, the name of the utility company, the reference pipeline ID, attribute data 35b, location 35p, leakage accident history, and map data 35m. The attribute data 35b of the reference pipeline includes, for example, the diameter of the reference pipeline, the pipe material, the pipeline length, and the burial period. The map data 35m of the reference pipeline includes, for example, the soil classification of the reference pipeline, as well as the average annual temperature and average annual precipitation at the location 35p of the reference pipeline. In this embodiment, the second reference pipeline dataset 37 is a preprocessed reference pipeline dataset.

[0071] The second pipeline data preprocessor 13b calculates the buried period of the reference pipeline by subtracting the laying year of the reference pipeline from the data acquisition year of the first reference pipeline dataset 36 of the specified pipe material. The second pipeline data preprocessor 13b identifies the map data 35m of the reference pipeline by referring to the latitude and longitude of the reference pipeline and the map database 40 (see Figure 4). Specifically, the second pipeline data preprocessor 13b identifies the soil classification of the reference pipeline by referring to the latitude and longitude of the reference pipeline and the soil classification map 41 (see Figure 4). The second pipeline data preprocessor 13b identifies the annual average temperature at the location 31p of the reference pipeline by referring to the latitude and longitude of the reference pipeline and the annual average temperature map 42 (see Figure 4). The second pipeline data preprocessor 13b identifies the annual average precipitation at the location 31p of the reference pipeline by referring to the latitude and longitude of the reference pipeline and the annual average precipitation map 43 (see Figure 4). The second pipeline data preprocessing unit 13b combines the burial period and map data of multiple reference pipelines with the first reference pipeline dataset 36 of the specified pipe material. The second pipeline data preprocessing unit 13b then removes the data acquisition year and the laying year (see Figure 9). In this way, the second reference pipeline dataset 37 is obtained.

[0072] Referring to Figure 14, in the integrated pipeline data creation step S4, the pipeline data integration unit 15 (see Figure 3) integrates the pre-processed customer pipeline dataset for the specified pipe material and the pre-processed reference pipeline dataset for the specified pipe material to create an integrated pipeline dataset 38 for the specified pipe material (see Figure 12). In this embodiment, the pre-processed customer pipeline dataset for the specified pipe material is the second customer pipeline dataset 33 (see Figure 10), and the pre-processed reference pipeline dataset for the specified pipe material is the second reference pipeline dataset 37 (see Figure 11). Figure 16 shows an example of the integrated pipeline dataset 38 in a multidimensional space defined by the dimensions of the explanatory variables. In Figure 16, for simplification, it is assumed that the explanatory variables are soil classification, average annual temperature, and average annual precipitation. The soil classification is also shown on a nominal scale.

[0073] Referring to Figure 14, in the similar pipeline data extraction step S5, the similar pipeline data extraction unit 16 (see Figure 3) extracts integrated pipeline data similar to the pre-processed customer pipeline dataset for the specified pipe material from the integrated pipeline dataset 38 (see Figure 12) for the specified pipe material, and creates a similar pipeline dataset composed of the similar pipeline data. In this embodiment, the pre-processed customer pipeline dataset for the specified pipe material is the second customer pipeline dataset 33 (see Figure 10). Figure 19 shows an example of a similar pipeline dataset in a multidimensional space defined by the dimensions of the explanatory variables. In Figure 19, for simplicity, it is assumed that the explanatory variables are soil classification, average annual temperature, and average annual precipitation. Also, the soil classification is shown on a nominal scale. Referring to Figures 17 and 18, two examples of the similar pipeline data extraction step S5 are described.

[0074] Referring to Figure 17, in an example of the similar pipeline data extraction step S5, the similar pipeline data extraction unit 16 calculates the data center of the pre-processed customer pipeline dataset (in this embodiment, the second reference pipeline dataset 37) included in the integrated pipeline dataset 38 (see Figure 12) in a multidimensional space defined by the dimensions of the explanatory variables of the integrated pipeline dataset 38 (see Figure 12) (step S11). The similar pipeline data extraction unit 16 calculates the distance (Euclidean distance, Mahalanobis distance, Manhattan distance, or Chebyshev distance, etc.) between the data center of each integrated pipeline data and the pre-processed customer pipeline dataset in the above multidimensional space (step S12). The similar pipeline data extraction unit 16 extracts integrated pipeline data for which this distance is less than or equal to the reference distance as similar pipeline data (step S13).

[0075] Referring to Figure 18, in another example of the similar pipeline data extraction step S5, the similar pipeline data extraction unit 16 calculates the data centers of the pre-processed customer pipeline dataset (in this embodiment, the second reference pipeline dataset 37) included in the integrated pipeline dataset 38 (see Figure 12) in a multidimensional space defined by the dimensions of the explanatory variables of the integrated pipeline dataset 38 (see Figure 12) (step S11). The similar pipeline data extraction unit 16 calculates the cosine similarity between each integrated pipeline data and the data center of the pre-processed customer pipeline dataset in the above multidimensional space (step S12b). The similar pipeline data extraction unit 16 extracts integrated pipeline data whose cosine similarity is equal to or greater than a threshold value as similar pipeline data (step S13b).

[0076] Referring to Figure 14, in critical pipeline data extraction step S6, the critical pipeline data extraction unit 17 extracts critical pipeline data from the integrated pipeline dataset 38 to create a critical pipeline dataset composed of critical pipeline data. The critical pipeline data is integrated pipeline data from the integrated pipeline dataset 38 that is important for improving the prediction accuracy of the pipeline deterioration prediction model 5. Referring to Figure 20, critical pipeline data extraction step S6 includes, for example, a first critical pipeline data extraction step S20 and a second critical pipeline data extraction step S21.

[0077] In the first critical pipeline data extraction step S20, the first critical pipeline data extraction unit 17a extracts pre-processed customer pipeline data that is not similar pipeline data (i.e., pre-processed customer pipeline data not extracted by the similar pipeline data extraction unit 16) from the integrated pipeline dataset 38 (see Figure 12) of the specified pipe material, and creates a first critical pipeline dataset composed of the first critical pipeline data. In this embodiment, the pre-processed customer pipeline data is the second customer pipeline dataset 33 (see Figure 10). Figure 21 shows an example of the first critical pipeline dataset in a multidimensional space defined by the dimensions of the explanatory variables. In Figure 21, for simplification, it is assumed that the explanatory variables are soil classification, average annual temperature, and average annual precipitation. Also, the soil classification is shown on a nominal scale.

[0078] In the second critical pipeline data extraction step S21, the second critical pipeline data extraction unit 17b extracts integrated pipeline data similar to the first critical pipeline data set from the integrated pipeline data set 38 (see Figure 12) of the specified pipe material as the second critical pipeline data, and creates a second critical pipeline data set consisting of the second critical pipeline data.

[0079] Specifically, referring to Figure 22, the second important pipeline data extraction unit 17b calculates the variability of each explanatory variable in the first important pipeline dataset (step S22). The second important pipeline data extraction unit 17b selects the explanatory variable with the least variability in the first important pipeline dataset (step S23). For example, in Figure 21, the explanatory variable with the least variability in the first important pipeline dataset is soil classification. The second important pipeline data extraction unit 17b calculates the mode of the explanatory variable selected in step S23 (step S24). For example, in Figure 21, the mode of soil classification is soil classification with a nominal scale of 4.

[0080] The second important pipeline data extraction unit 17b extracts integrated pipeline data from the integrated pipeline dataset 38 (see Figure 12) in which the selected explanatory variable is the mode (step S25). Figure 23 shows an example of the integrated pipeline data extracted in step S25 in a multidimensional space defined by the dimensions of the explanatory variables (excluding the explanatory variable with the least variability) of the integrated pipeline dataset 38 (see Figure 12). In Figure 23, integrated pipeline data with a nominal scale of 4 for soil classification is extracted from the integrated pipeline dataset 38 and plotted in a multidimensional space defined by the annual average temperature and annual average precipitation.

[0081] The second important pipeline data extraction unit 17b extracts integrated pipeline data similar to the first important pipeline dataset, where the selected explanatory variable is the mode, from the integrated pipeline data extracted in step S25, and creates a second important pipeline dataset consisting of the second important pipeline data (step S26). Figure 24 shows an example of a second important similar pipeline dataset. In Figure 24, integrated pipeline data similar to the first important pipeline dataset, where the nominal scale of the soil classification is 4, is extracted from the integrated pipeline data with a nominal scale of 4 extracted in step S25, and plotted in a multidimensional space defined by the average annual temperature and average annual precipitation. Two examples of step S26 will be explained with reference to Figures 25 and 26.

[0082] Referring to Figure 25, in an example of step S26, the second important pipeline data extraction unit 17b calculates the data center of the first important pipeline dataset where the selected explanatory variable is the mode in a multidimensional space defined by the dimensions of the explanatory variables (excluding the explanatory variable with the least variability) of the integrated pipeline dataset 38 (see Figure 12) (step S31). The second important pipeline data extraction unit 17b calculates the distance (Euclidean distance, Mahalanobis distance, Manhattan distance, or Chebyshev distance, etc.) between each of the integrated pipeline data where the selected explanatory variable is the mode in the above multidimensional space and the data center of the first important pipeline dataset where the selected explanatory variable is the mode (step S32). The second important pipeline data extraction unit 17b extracts integrated pipeline data where this distance is less than or equal to the reference distance as second important pipeline data (step S33). In Figure 24, in a multidimensional space defined by average annual temperature and average annual precipitation, integrated pipeline data where the Euclidean distance between each integrated pipeline data with a nominal scale of 4 for soil classification and the data center of the first important pipeline dataset, also with a nominal scale of 4 for soil classification, is less than or equal to the reference distance is extracted as the second important pipeline data.

[0083] Referring to Figure 26, in another example of step S26, the second important pipeline data extraction unit 17b calculates the data center of the first important pipeline dataset where the selected explanatory variable is the mode in a multidimensional space defined by the dimensions of the explanatory variables (excluding the explanatory variable with the least variability) of the integrated pipeline dataset 38 (see Figure 12) (step S31). The second important pipeline data extraction unit 17b calculates the cosine similarity between each of the integrated pipeline data where the selected explanatory variable is the mode in the above multidimensional space and the data center of the first important pipeline dataset where the selected explanatory variable is the mode (step S32b). The second important pipeline data extraction unit 17b extracts the integrated pipeline data where the cosine similarity is equal to or greater than a threshold value as the second important pipeline data (step S33b).

[0084] Referring to Figure 14, in the learning pipeline dataset creation step S7, the pipeline data merging unit 18 combines the similar pipeline dataset extracted by the similar pipeline data extraction unit 16 in the similar pipeline data extraction step S5 with the important pipeline dataset extracted by the important pipeline data extraction unit 17 in the important pipeline data extraction step S6 to create a learning pipeline dataset 48 for the specified pipe material (see Figure 7). The pipeline data merging unit 18 assigns a new pipeline ID (learning pipeline ID) to the learning pipeline data that constitutes the learning pipeline dataset 48. Figure 27 shows an example of the learning pipeline dataset 48 in a multidimensional space defined by the dimensions of the explanatory variables. In Figure 27, for simplification, it is assumed that the explanatory variables are soil classification, average annual temperature, and average annual precipitation. Also, the soil classification is shown on a nominal scale.

[0085] Referring to Figure 7, the learning pipeline dataset 48 is a set of multiple learning pipeline data, and includes multiple learning pipeline data. The learning pipeline dataset 48 for a specified pipe material includes all of the pre-processed customer pipeline datasets for the specified pipe material and a portion of the pre-processed reference pipeline datasets for the specified pipe material. In this embodiment, the pre-processed customer pipeline dataset for the specified pipe material is the second customer pipeline dataset 33 (see Figure 10), and the pre-processed reference pipeline dataset for the specified pipe material is the second reference pipeline dataset 37 (see Figure 11).

[0086] Referring to Figure 14, in the pipeline deterioration prediction model generation step S8, the pipeline deterioration prediction model generation unit 20 (see Figure 3) uses the training pipeline dataset 48 (see Figure 7) for the specified pipe material as training data to generate pipeline deterioration prediction models 5a and 5b for the specified pipe material. Referring to Figure 28, the pipeline deterioration prediction model generation step S8 includes the pipeline leakage accident probability prediction model generation step S36 and the conversion unit generation step S37.

[0087] In step S36, the pipeline leakage accident probability prediction model generation unit 21 (see Figure 3) generates pipeline leakage accident probability prediction models 6a and 6b using machine learning with the specified pipe material training pipeline dataset 48 (see Figure 7) as training data.

[0088] The leakage accident history is entered as either "0" or "1". It is rare for multiple leakage accidents to occur in a single pipeline during a predetermined period including the data acquisition year. Therefore, the pipeline leakage accident history can be considered as the probability of pipeline leakage accidents during a predetermined period including the data acquisition year. Accordingly, as shown in Figure 13, explanatory variables (pipe attribute data 48a (e.g., diameter, pipeline length, and burial period) and map data 48m (e.g., soil classification, average annual temperature, and average annual precipitation)) from the training pipeline dataset 48 (see Figure 7) for the specified pipe material are input into the prediction model (e.g., the neural network model 50 shown in Figure 13). The prediction model outputs the predicted probability of pipeline leakage accidents. The prediction model is trained using machine learning so that the predicted probability of pipeline leakage accidents approaches the pipeline leakage accident history included in the training pipeline dataset 48, and trained prediction models (pipe leakage accident probability prediction models 6a, 6b) are generated.

[0089] The pipeline leakage accident probability prediction model learning unit 21 outputs pipeline leakage accident probability prediction models 6a and 6b for the specified pipe material to the storage unit 25 (see Figures 3 and 4). The pipeline leakage accident probability prediction models 6a and 6b for the specified pipe material are stored in the storage unit 25.

[0090] Referring to Figure 28, in conversion unit generation step S37, the conversion unit generation unit 22 (see Figure 3) generates conversion units 7a and 7b. Conversion units 7a and 7b convert the pipeline leakage accident prediction probability calculated by the pipeline leakage accident probability prediction models 6a and 6b for the specified pipe material into a pipeline leakage accident rate. An example of the method for generating conversion units 7a and 7b is described below.

[0091] The conversion unit generation unit 22 (see Figure 3) inputs the learning pipeline dataset 48 (see Figure 7) into the pipeline leakage accident probability prediction models 6a and 6b for the specified pipe material to calculate the predicted probability of pipeline leakage accidents. As already mentioned, the pipeline leakage accident history indicates whether or not there were pipeline leakage accidents during a predetermined period including the data acquisition year. The pipeline leakage accident history can be considered as the number of pipeline leakage accidents during the predetermined period including the data acquisition year. Therefore, the value obtained by dividing the value in the leakage accident history column of the learning pipeline dataset 48 (see Figure 7) by the predetermined period (years) including the data acquisition year can be considered as the number of leakage accidents per year (accidents / year). The conversion unit generation unit 22 combines the predicted probability of pipeline leakage accidents with the learning pipeline ID, pipeline length, and number of leakage accidents per year included in the learning pipeline dataset 48 to create a leakage accident prediction probability table 53 (see Figure 29).

[0092] The conversion unit generation unit 22 (see Figure 3) calculates the actual water leakage accident rate for each water leakage accident prediction probability within a predetermined range from the water leakage accident prediction probability table 53 (see Figure 29).

[0093] Specifically, the conversion unit generation unit 22 (see Figure 3) groups the data included in the water leakage accident prediction probability table 53 (see Figure 29) according to predetermined ranges of water leakage accident prediction probabilities. For example, the conversion unit generation unit 22 groups the data included in the water leakage accident prediction probability table 53 according to a range of water leakage accident prediction probabilities of 0.2. Specifically, the conversion unit generation unit 22 divides the data included in the water leakage accident prediction probability table 53 into groups: a group with a water leakage accident prediction probability of 0 or more and less than 0.2; a group with a water leakage accident prediction probability of 0.2 or more and less than 0.4; a group with a water leakage accident prediction probability of 0.4 or more and less than 0.6; a group with a water leakage accident prediction probability of 0.6 or more and less than 0.8; and a group with a water leakage accident prediction probability of 0.8 or more and less than 1. Since the number of pipeline data in the water leakage accident prediction probability table 53 is equal to the number of pipeline data included in the training pipeline dataset 48, each group has a sufficient number of data to regress the relationship between the water leakage accident prediction probability and the actual water leakage accident rate.

[0094] The conversion unit generation unit 22 (see Figure 3) calculates the actual leakage accident rate (incidents / year / km) for each group. Specifically, referring to Figure 30, the conversion unit generation unit 22 calculates the total length of the pipelines included in each group as the total pipeline length for each group. The conversion unit generation unit 22 calculates the total number of leakage accidents per unit time included in each group as the actual number of leakage accidents per unit time for each group. The conversion unit generation unit 22 calculates the actual leakage accident rate for each group by dividing the actual number of leakage accidents per unit time for each group by the total pipeline length for each group. Referring to Figure 31, the actual leakage accident rate for each group is plotted on a graph against the center value of the range of predicted leakage accident probabilities for each group to obtain data on the relationship between the predicted leakage accident probability and the actual leakage accident rate (see the black dots in Figure 31).

[0095] The conversion unit generation unit 22 (see Figure 3) generates conversion units 7a and 7b by regressing data on the relationship between the predicted probability of a water leak accident and the actual water leak accident rate. For example, the actual water leak accident rate is considered to be proportional to the predicted probability of a water leak accident. Therefore, the conversion unit generation unit 22 generates conversion formulas that convert the predicted probability of a water leak accident into the water leak accident rate, as conversion units 7a and 7b, by performing linear regression on data on the relationship between the predicted probability of a water leak accident and the actual water leak accident rate.

[0096] As shown in Figure 30, the pipe length of each group decreases as the predicted probability of a water leak increases. This is because water leaks rarely occur in pipes, and as the predicted probability of a water leak increases, the number of pipe data points included in each group decreases. Therefore, as the predicted probability of a water leak increases, the reliability of the data relating the predicted probability of a water leak and the actual water leak rate (see black dots in Figure 31) decreases. For this reason, for water leak prediction probabilities below a predetermined value, a more reliable transformation unit 7a,7b may be generated by regression of the data relating the predicted probability of a water leak and the actual water leak rate. For example, the range of the predicted probability of a water leak to be regressioned may be, for example, less than 0.6.

[0097] Referring to Figure 14, the pipe material specification step S2 to the pipe deterioration prediction model generation step S8 is performed for all pipe materials included in the original customer pipe data set 31 (see Figure 5) to generate multiple pipe deterioration prediction models 5a and 5b (see Figure 4) for the specified pipe materials. The multiple pipe deterioration prediction models 5a and 5b for the specified pipe materials are stored in the storage unit 25 (see Figures 3 and 4). Each of the multiple pipe deterioration prediction models 5a and 5b for the specified pipe materials includes a pipe leakage accident probability prediction model 6a and 6b for the specified pipe material and a corresponding conversion unit 7a and 7b. In this way, a pipe deterioration prediction model 5 including the multiple pipe deterioration prediction models 5a and 5b for the specified pipe materials is generated.

[0098] Specifically, referring to Figure 14, in the model generation completion determination step S9, the pipeline deterioration prediction model generation completion determination unit 23 (see Figure 3) determines whether pipeline deterioration prediction models 5a and 5b have been generated for all pipe materials included in the original customer pipeline dataset 31 (see Figure 5). If the pipeline deterioration prediction model generation completion determination unit 23 determines that pipeline deterioration prediction models 5a and 5b have not been generated for all pipe materials included in the original customer pipeline dataset 31, the process returns to the pipe material specification step S2. In the pipe material specification step S2, the pipe material specification unit 11 (see Figure 3) specifies the pipe materials for which pipeline deterioration prediction models 5a and 5b have not yet been created among all pipe materials included in the original customer pipeline dataset 31. The process from the pipe material specification step S2 to the pipeline deterioration prediction model generation step S8 is repeatedly executed until pipeline deterioration prediction models 5a and 5b for all pipe materials included in the original customer pipeline dataset 31 have been generated. In this way, pipeline deterioration prediction models 5a and 5b are generated for all pipe materials included in the original customer pipeline dataset 31.

[0099] In the model generation completion determination step S9, if the pipeline deterioration prediction model generation completion determination unit 23 (see Figure 3) determines that it has generated pipeline deterioration prediction models 5a and 5b for all pipe materials included in the original customer pipeline dataset 31 (see Figure 5), the generation of pipeline deterioration prediction model 5 is terminated.

[0100] The pipeline deterioration prediction model generation unit 20 stores pipeline deterioration prediction models 5a and 5b for multiple specified pipe materials in the storage unit 25 (see Figures 39 and 40). Each of the pipeline deterioration prediction models 5a and 5b for multiple specified pipe materials includes a pipeline leakage accident probability prediction model 6a and 6b for the specified pipe material, and corresponding conversion units 7a and 7b. In this way, a pipeline deterioration prediction model 5 is generated that includes the pipeline deterioration prediction models 5a and 5b for multiple specified pipe materials.

[0101] The pipeline deterioration prediction model generation program 8 (see Figure 4) causes the processor 202 (see Figure 2) to execute the pipeline deterioration prediction model generation method of this embodiment. The pipeline deterioration prediction model generation program 8 of this embodiment may be recorded on a computer-readable storage medium (a non-transient computer-readable storage medium, for example, storage medium 208).

[0102] Referring to Figure 1, the pipeline deterioration prediction model generation device 2 transmits the pipeline deterioration prediction model 5 to the pipeline deterioration prediction device 3.

[0103] <Pipeline Deterioration Prediction Device 3> Referring to Figure 1, the pipeline deterioration prediction device 3 receives the pipeline deterioration prediction model 5 from the pipeline deterioration prediction model generation device 2. The pipeline deterioration prediction device 3 uses the pipeline deterioration prediction model 5 to calculate the degree of pipeline deterioration (for example, the leakage accident rate of the pipeline).

[0104] <Hardware Configuration> Referring to Figure 32, the hardware configuration of the pipeline deterioration prediction device 3 will be described. The pipeline deterioration prediction device 3 includes an input device 301, a processor 302, memory 303, a display 304, a network controller 306, a storage medium drive 307, and storage 310.

[0105] The input device 301 accepts various input operations. The input device 301 is, for example, a keyboard, a mouse, or a touch panel.

[0106] The display 304 displays information necessary for processing in the pipeline deterioration prediction device 3. For example, the display 304 displays the pipeline deterioration prediction result 74 (see Figures 35 and 36), which will be described later. The display 304 is, for example, an LCD or an organic EL display.

[0107] The processor 302 executes the processing necessary to realize the functions of the pipeline deterioration prediction device 3 by running a program described later. The processor 302 is composed of, for example, a CPU or a GPU.

[0108] Memory 303 provides a storage area for the processor 302 to temporarily store program code or work memory when executing a program. Memory 303 is, for example, a volatile memory device such as DRAM or SRAM.

[0109] The network controller 306 transmits and receives programs or data to and from any device, including the pipeline aging prediction model generation device 2, via a communication network 4 (see Figure 1), such as the Internet or an intranet. For example, the network controller 306 receives the pipeline aging prediction model 5 (see Figure 34) from the pipeline aging prediction model generation device 2 via the communication network 4. The network controller 306 supports any communication method, such as Ethernet®, wireless LAN, or Bluetooth®.

[0110] The storage medium drive 307 is a device that reads programs or data stored in the storage medium 308. The storage medium drive 307 may also be a device that writes programs or data to the storage medium 308. The storage medium 308 is a non-transitory storage medium that stores programs or data non-volatilely. The storage medium 308 is, for example, an optical storage medium such as an optical disc (e.g., CD-ROM or DVD-ROM), a semiconductor storage medium such as flash memory or USB memory, a magnetic storage medium such as a hard disk, floppy disk or storage tape, or a magneto-optical storage medium such as an MO disk.

[0111] The storage 310 stores the original customer pipeline dataset 31 (see Figures 5 and 34), the input pipeline dataset 72 (see Figures 10 and 34), the map database 40 (see Figure 34), the pipeline deterioration prediction model 5 (see Figure 34), and programs executed by the processor 302. These programs include the pipeline deterioration prediction program 75 (see Figure 34). The storage 310 is, for example, a non-volatile memory device such as a hard disk or SSD.

[0112] The program for implementing the functions of the pipeline aging prediction device 3 may be stored and distributed on a non-transient storage medium 308 and installed on the storage device 310. The program for implementing the functions of the pipeline aging prediction device 3 may also be downloaded to the pipeline aging prediction device 3 via a communication network 4 such as the Internet or an intranet.

[0113] In this embodiment, an example is shown in which a general-purpose computer (processor 302) implements the functions of the pipeline deterioration prediction device 3 by executing a program. However, the embodiment is not limited to this, and all or part of the functions of the pipeline deterioration prediction device 3 may be implemented using an integrated circuit such as an ASIC or FPGA.

[0114] <Functional Configuration> Referring to Figure 33, an example of the functional configuration of the pipeline deterioration prediction device 3 will be explained. The pipeline deterioration prediction device 3 comprises a pipeline data set receiving unit 60, a pipe material specification unit 61, an input pipeline data set creation unit 62, a model selection unit 66, a pipeline deterioration calculation unit 67, a pipeline deterioration calculation completion determination unit 68, a pipeline deterioration prediction result creation unit 69, and a storage unit 70.

[0115] <Storage section 70> The storage unit 70 is implemented by, for example, at least one of a storage device 310 (see Figure 32) or a storage medium 308 (see Figure 32). Referring to Figure 34, the storage unit 70 stores the original customer pipeline dataset 31 (see Figure 5), the input pipeline dataset 72 (see Figure 10), the map database 40, the pipeline deterioration prediction model 5, the pipeline deterioration prediction results 74 (see Figures 35 and 36), and the pipeline deterioration prediction program 75.

[0116] The original customer pipeline dataset 31 (see Figure 5) is provided by the customer (water utility). The original customer pipeline dataset 31 is a set of multiple original customer pipeline data and contains multiple original customer pipeline data. The original customer pipeline dataset 31 stored in the storage unit 70 has the same data structure as the original customer pipeline dataset 31 stored in the storage unit 25 (see Figure 4).

[0117] Referring to Figures 10 and 34, the input pipeline dataset 72 (see Figure 10) is a pipeline dataset input to the pipeline deterioration prediction model 5 (see Figure 34). The input pipeline dataset 72 has the same data structure as the second customer pipeline dataset 33 (see Figure 10). Specifically, the input pipeline dataset 72 includes, for example, the business name, customer pipeline ID, attribute data 31b, location 31p, leakage accident history, and map data 31m. The customer pipeline attribute data 31b includes, for example, the diameter, pipe material, pipeline length, and burial period of the customer pipeline. The customer pipeline map data 31m includes, for example, the soil classification of the customer pipeline, as well as the average annual temperature and average annual precipitation at the customer pipeline location 31p.

[0118] Referring to Figure 34, the map database 40 is provided by public institutions, for example, through the internet or storage medium 208, such as a geographic information system (GIS), and is pre-stored in the storage unit 70 (see Figures 33 and 34). The map database 40 stored in the storage unit 70 is the same as the map database 40 stored in the storage unit 25 (see Figure 4).

[0119] Referring to Figure 34, the pipeline deterioration prediction model 5 is generated by the pipeline deterioration prediction model generation device 2, transmitted from the pipeline deterioration prediction model generation device 2 to the pipeline deterioration prediction device 3, and stored in the memory unit 70 of the pipeline deterioration prediction device 3. The pipeline deterioration prediction model 5 includes pipeline deterioration prediction models 5a and 5b for each pipe material. For example, the pipeline deterioration prediction model 5 includes the pipeline deterioration prediction model 5a for rigid polyvinyl chloride (VP) pipes and the pipeline deterioration prediction model 5b for ductile cast iron (DIP) pipes. The pipeline deterioration prediction model 5 outputs the degree of pipeline deterioration.

[0120] The degree of pipeline deterioration is not particularly limited, but for example, it is the leakage accident rate of the pipeline. Pipeline deterioration prediction model 5 includes, for example, pipeline leakage accident probability prediction models 6a and 6b (see Figure 34) and conversion units 7a and 7b (see Figure 34). For example, the pipeline deterioration prediction model 5a for VP pipes includes the pipeline leakage accident probability prediction model 6a for VP pipes and the conversion unit 7a for VP pipes. The pipeline deterioration prediction model 5b for DIP pipes includes the pipeline leakage accident probability prediction model 6b for DIP pipes and the conversion unit 7b for DIP pipes. Pipeline leakage accident probability prediction models 6a and 6b output the predicted leakage accident probability of the pipeline. The conversion units 7a and 7b convert the predicted leakage accident probability of the pipeline into the pipeline leakage accident rate.

[0121] Referring to Figure 34, the pipeline deterioration prediction result 74 is, for example, at least one of the pipeline deterioration prediction table 74a (see Figure 35) or the pipeline deterioration prediction map 74b (see Figure 36). The pipeline deterioration prediction result creation unit 69 creates the pipeline deterioration prediction table 74a by associating the leakage accident rate of customer pipelines output from the pipeline deterioration prediction model 5 with the customer pipeline IDs in the original customer pipeline dataset 31 (see Figure 5) or the input pipeline dataset 72 (see Figure 10). The pipeline deterioration prediction result creation unit 69 creates the pipeline deterioration prediction map 74b by associating the leakage accident rate of customer pipelines output from the pipeline deterioration prediction model 5 with the customer pipeline locations 31p in the original customer pipeline dataset 31 or the input pipeline dataset 72.

[0122] The pipeline deterioration prediction program 75 is a program that creates an input pipeline dataset 72 (see Figures 10 and 34) from the original customer pipeline dataset 31 (see Figure 5) and outputs the pipeline deterioration prediction result 74 (see Figures 35 and 36) for customer pipelines.

[0123] <Pipeline Data Set Reception Unit 60> Referring to Figure 33, the pipeline data set receiving unit 60 receives the original customer pipeline data set 31 (see Figure 5) from the customer (water utility). The original customer pipeline data set 31 is provided by the customer, for example, via a storage medium 308 (see Figure 32) or through a communication network 4 such as the Internet (see Figure 1). The pipeline data set receiving unit 60 outputs the original customer pipeline data set 31 to the storage unit 70 (see Figures 33 and 34). The original customer pipeline data set 31 is stored in the storage unit 70.

[0124] <Pipe material specification section 61> Referring to Figure 33, the pipe material designation unit 61 designates the pipe material of the pipeline whose degree of deterioration should be predicted. The pipe material designation unit 61 may be implemented by the input device 301 (see Figure 32). The pipe material designated by the pipe material designation unit 61 is one of the pipe materials included in the original customer pipeline data set 31 (see Figure 5), and is called the designated pipe material.

[0125] <Input pipeline dataset creation unit 62> Referring to Figure 33, the input pipeline dataset creation unit 62 creates an input pipeline dataset 72 (see Figure 10) from the original customer pipeline dataset 31 (see Figure 5). The input pipeline dataset 72 has the same data structure as, for example, the second customer pipeline dataset 33 (see Figure 10). The input pipeline dataset creation unit 62 includes, for example, a first customer pipeline data processing unit 63 and a second customer pipeline data processing unit 64.

[0126] <First Customer Pipeline Data Processing Unit 63> Referring to Figure 33, the first customer pipeline data processing unit 63 (see Figure 33) has the same functions as the first pipeline data preprocessing unit 13a (see Figure 3). Specifically, the first customer pipeline data processing unit 63 extracts the original customer pipeline data for the specified pipe material from the original customer pipeline dataset 31 (see Figure 5) and creates the first customer pipeline dataset 32 ​​for the specified pipe material (see Figure 8). The specified pipe material is the pipe material specified by the pipe material specification unit 61. For example, if the specified pipe material is a VP pipe, the first customer pipeline data processing unit 63 extracts the original customer pipeline data for the VP pipe from the original customer pipeline dataset 31 and creates the first customer pipeline dataset 32 ​​for the VP pipe.

[0127] <Second Customer Pipeline Data Processing Unit 64> Referring to Figure 33, the second customer pipeline data processing unit 64 (see Figure 33) has the same functions as the second pipeline data preprocessing unit 13b (see Figure 3). Specifically, the second customer pipeline data processing unit 64 creates a second customer pipeline dataset 33 (see Figure 10) for a specified pipe material from a first customer pipeline dataset 32 ​​(see Figure 8) for a specified pipe material. In this embodiment, the second customer pipeline dataset 33 is the input pipeline dataset 72 (see Figure 10).

[0128] <Model Selection Section 66> Referring to Figure 33, the model selection unit 66 selects the pipeline deterioration prediction model 5a, 5b for the specified pipe material from among multiple pipeline deterioration prediction models 5a, 5b (see Figure 34) stored in the memory unit 70 (see Figures 33 and 34). For example, if the specified pipe material is a VP pipe, the model selection unit 66 selects the pipeline deterioration prediction model 5a for VP pipes from among the multiple pipeline deterioration prediction models 5a, 5b.

[0129] <Pipeline deterioration calculation unit 67> Referring to Figure 33, the pipeline deterioration calculation unit 67 (see Figure 33) inputs the input pipeline dataset 72 (see Figure 10) for the specified pipe material, created by the input pipeline dataset creation unit 62, into the pipeline deterioration prediction models 5a and 5b for the specified pipe material selected by the model selection unit 66, and calculates the deterioration of the customer pipeline. The deterioration of the customer pipeline is not particularly limited, but for example, it could be the leakage accident rate (incidents / year / km) of the customer pipeline.

[0130] Specifically, the pipeline deterioration calculation unit 67 inputs the explanatory variables (pipe attribute data 31b (e.g., diameter, pipe length, and burial period) and map data 31m (e.g., soil classification, average annual temperature, and average annual precipitation)) from the input pipeline dataset 72 (see Figures 10 and 34) for the specified pipe material to the pipeline deterioration prediction models 5a and 5b (see Figure 34) of the specified pipe material selected by the model selection unit 66, and calculates the predicted leakage accident probability of the customer pipeline. The pipeline deterioration calculation unit 67 inputs the predicted leakage accident probability of the customer pipeline to the conversion units 7a and 7b (see Figure 34) of the pipeline deterioration prediction models 5a and 5b (see Figure 34) of the specified pipe material selected by the model selection unit 66, and calculates the leakage accident rate of the customer pipeline.

[0131] <Pipeline deterioration calculation completion determination unit 68> Referring to Figure 33, the pipeline deterioration calculation completion determination unit 68 determines whether it has calculated the deterioration of all customer pipelines included in the original customer pipeline dataset 31 (see Figure 5). If the pipeline deterioration calculation completion determination unit 68 determines that it has not calculated the deterioration of all customer pipelines included in the original customer pipeline dataset 31, the pipe material specification unit 61 (see Figure 33) specifies the pipe materials from all the pipe materials included in the original customer pipeline dataset 31 for which the deterioration of customer pipelines has not yet been calculated. If the pipeline deterioration calculation completion determination unit 68 determines that it has calculated the deterioration of all customer pipelines included in the original customer pipeline dataset 31, it terminates the calculation of the deterioration of customer pipelines.

[0132] <Pipeline Deterioration Prediction Result Creation Section 69> Referring to Figure 33, the pipeline deterioration prediction result creation unit 69 creates a pipeline deterioration prediction result 74 (see Figures 35 and 36) using the deterioration of the customer pipeline calculated by the pipeline deterioration calculation unit 67 (see Figure 33). The pipeline deterioration prediction result 74 may be, for example, a pipeline deterioration prediction table 74a (see Figure 35) or a pipeline deterioration prediction map 74b (see Figure 36). The pipeline deterioration prediction result creation unit 69 outputs the pipeline deterioration prediction result 74 to the storage unit 70 (see Figures 33 and 34). The pipeline deterioration prediction result creation unit 69 may also output the pipeline deterioration prediction result 74 to at least one of the display 304, storage medium 308, or storage 310 shown in Figure 32.

[0133] <Method for predicting pipeline deterioration> The pipeline deterioration prediction method of this embodiment will be described with reference to Figures 37 and 38. The pipeline deterioration prediction method of this embodiment includes a customer pipeline dataset reception step S41, a pipe material specification step S42, a model selection step S43, an input pipeline dataset creation step S44, a customer pipeline deterioration calculation step S45, and a pipeline deterioration calculation completion determination step S46.

[0134] Referring to Figure 37, the original customer pipeline dataset reception step S41 is the same as the original customer pipeline dataset reception step S1 shown in Figure 14. Specifically, in the original customer pipeline dataset reception step S41, the pipeline dataset reception unit 60 (see Figure 33) receives the original customer pipeline dataset 31 (see Figure 5) from the customer (water utility). The pipeline dataset reception unit 60 outputs the original customer pipeline dataset 31 to the storage unit 70 (see Figures 33 and 34). The original customer pipeline dataset 31 is stored in the storage unit 70.

[0135] Referring to Figure 37, the pipe material specification step S42 is the same as the pipe material specification step S2 shown in Figure 14. Specifically, the pipe material specification section 61 (see Figure 33) specifies the pipe material of the pipeline whose degree of deterioration is to be predicted. The pipe material specified in pipe material specification step S42 is one of the pipe materials included in the original customer pipeline data set 31 (see Figure 5), and is called the specified pipe material.

[0136] Referring to Figure 37, in the model selection step S43, the model selection unit 66 (see Figure 33) selects the pipe deterioration prediction model 5a, 5b for the specified pipe material from among multiple pipe deterioration prediction models 5a, 5b (see Figure 34) stored in the memory unit 70 (see Figures 33 and 34). The specified pipe material is the pipe material specified in the pipe material specification step S42. For example, if the pipe material specified in step S42 is a VP pipe, the model selection unit 66 selects the pipe deterioration prediction model 5a for the VP pipe from among the multiple pipe deterioration prediction models 5a, 5b.

[0137] Referring to Figure 37, in the input pipeline dataset creation step S44, the input pipeline dataset creation unit 62 (see Figure 33) creates an input pipeline dataset 72 (see Figures 10 and 34) for a specified pipe material from the original customer pipeline dataset 31 (see Figure 5). In this embodiment, the input pipeline dataset 72 for a specified pipe material has the same data structure as the second customer pipeline dataset 33 (see Figure 10) for the specified pipe material. Referring to Figure 38, the input pipeline dataset creation step S44 includes, for example, a first customer pipeline data processing step S44a and a second customer pipeline data processing step S44b.

[0138] Step S44a of the first customer pipeline data processing is the same as step S3a of the first pipeline data preprocessing (see Figure 15). Specifically, the first customer pipeline data processing unit 63 (see Figure 33) extracts the original customer pipeline data for the pipe material specified in step S42 from the original customer pipeline dataset 31 (see Figure 5) to create the first customer pipeline dataset 32 ​​for the specified pipe material (see Figure 8). For example, if the specified pipe material is a VP pipe, the first customer pipeline data processing unit 63 extracts the original customer pipeline data for the VP pipe from the original customer pipeline dataset 31 to create the first customer pipeline dataset 32 ​​for the VP pipe.

[0139] The second customer pipeline data processing step S44b is the same as the second pipeline data preprocessing step S3b (see Figure 15). Specifically, the second customer pipeline data processing unit 64 (see Figure 33) creates a second customer pipeline dataset 33 (see Figure 10) for the specified pipe material from the first customer pipeline dataset 32 ​​(see Figure 8) for the specified pipe material. In this embodiment, the second customer pipeline dataset 33 is the input pipeline dataset 72 (see Figure 10).

[0140] Specifically, the second customer pipeline data processing unit 64 calculates the buried period of the customer pipeline by subtracting the year of installation of the customer pipeline from the data acquisition year of the first customer pipeline dataset 32 ​​for the specified pipe material. The second customer pipeline data processing unit 64 identifies the map data 31m of the customer pipeline by referring to the latitude and longitude of the customer pipeline and the map database 40 (see Figure 34). The second customer pipeline data processing unit 64 combines the buried period of the customer pipeline and the map data 31m with the first customer pipeline dataset 32 ​​for the specified pipe material. The second customer pipeline data processing unit 64 deletes the data acquisition year and the year of installation (see Figure 8).

[0141] Referring to Figure 37, in the customer pipeline deterioration calculation step S45, the pipeline deterioration calculation unit 67 (see Figure 33) inputs the input pipeline dataset 72 (see Figures 10 and 34) for the specified pipe material into the pipeline deterioration prediction models 5a and 5b selected in the model selection step S43, and calculates the deterioration of the customer pipeline. The deterioration of the customer pipeline is not particularly limited, but for example, it may be the leakage accident rate (incidents / year / km) of the customer pipeline.

[0142] Specifically, the pipeline deterioration calculation unit 67 (see Figure 33) inputs the explanatory variables (pipe attribute data 31a (e.g., diameter, pipe length, and burial period) and map data 31m (e.g., soil classification, average annual temperature, and average annual precipitation)) from the input pipeline dataset 72 (see Figures 10 and 34) created in the input pipeline dataset creation step S44 for the specified pipe material to the pipeline deterioration prediction models 5a and 5b (see Figure 34) of the specified pipe material selected in the model selection step S43, and calculates the predicted leakage accident probability of the customer pipeline. The pipeline deterioration calculation unit 67 inputs the predicted leakage accident probability of the customer pipeline to the conversion units 7a and 7b (see Figure 34) of the pipeline deterioration prediction models 5a and 5b for the specified pipe material selected in the model selection step S43, and calculates the leakage accident rate of the customer pipeline.

[0143] Referring to Figure 37, the customer pipeline deterioration calculation step S45 is performed from the pipe material specification step S42 to the other specified pipe materials specified in the pipe material specification step S42, and the deterioration of all customer pipelines included in the original customer pipeline dataset 31 (see Figure 5) is calculated.

[0144] Specifically, in the pipeline deterioration calculation completion determination step S46, the pipeline deterioration calculation completion determination unit 68 (see Figure 33) determines whether the deterioration of all customer pipelines included in the original customer pipeline dataset 31 (see Figure 5) has been calculated. If the pipeline deterioration calculation completion determination unit 68 determines that the deterioration of all customer pipelines included in the original customer pipeline dataset 31 has not been calculated, the process returns to the pipe material specification step S42. In the pipe material specification step S42, the pipe material specification unit 61 (see Figure 33) specifies the pipe material from all the pipe materials included in the original customer pipeline dataset 31 for which the deterioration of customer pipelines has not yet been calculated. The process repeats from the pipe material specification step S42 to the customer pipeline deterioration calculation step S45 until the deterioration of all customer pipelines included in the original customer pipeline dataset 31 has been calculated. In this way, the deterioration of all customer pipelines included in the original customer pipeline dataset 31 is calculated.

[0145] In the pipeline deterioration calculation completion determination step S46, if the pipeline deterioration calculation completion determination unit 68 (see Figure 33) determines that it has calculated the deterioration of all customer pipelines included in the original customer pipeline data set 31 (see Figure 5), the calculation of the deterioration of customer pipelines is terminated.

[0146] Referring to Figure 37, in the pipeline deterioration prediction result creation step S47, the pipeline deterioration prediction result creation unit 69 (see Figure 33) creates a pipeline deterioration prediction result 74 (see Figures 35 and 36) using the deterioration of the customer pipeline calculated by the pipeline deterioration calculation unit 67 (see Figure 33). The pipeline deterioration prediction result 74 may be, for example, a pipeline deterioration prediction table 74a (see Figure 35) or a pipeline deterioration prediction map 74b (see Figure 36). The pipeline deterioration prediction result creation unit 69 outputs the pipeline deterioration prediction result 74 (see Figures 35 and 36) to the storage unit 70 (see Figures 33 and 34). The pipeline deterioration prediction result creation unit 69 may also output the pipeline deterioration prediction result 74 to at least one of the display 304, storage medium 308, or storage 310 shown in Figure 32.

[0147] The pipeline deterioration prediction program 75 (see Figure 34) causes the processor 302 (see Figure 32) to execute the pipeline deterioration prediction method of this embodiment. The pipeline deterioration prediction program 75 of this embodiment may be recorded on a computer-readable storage medium (a non-transient computer-readable storage medium, for example, storage medium 308).

[0148] In a modified version of this embodiment, the important pipeline data extraction unit 17 may not include the second important pipeline data extraction unit 17b. The pipeline data merging unit 18 may combine the similar pipeline dataset and the first important pipeline dataset to create a learning pipeline dataset 48. The important pipeline data extraction step S6 may not include the second important pipeline data extraction step S21. The learning pipeline dataset creation step S7 may combine the similar pipeline dataset and the first important pipeline dataset to create a learning pipeline dataset 48.

[0149] This document describes the method for generating the pipeline deterioration prediction model 5, the pipeline deterioration prediction method, the program, and the effects of the pipeline deterioration prediction device 3 according to this embodiment.

[0150] The method for generating the pipeline deterioration prediction model 5 of this embodiment includes a step (pipeline data preprocessing step S3) of creating a customer pipeline dataset (preprocessed customer pipeline dataset; in this embodiment, a second customer pipeline dataset 33) and a reference pipeline dataset (preprocessed reference pipeline dataset; in this embodiment, a second reference pipeline dataset 37). The customer pipeline dataset includes multiple customer pipeline data. The multiple customer pipeline data includes the pipeline length, burial period, map data, and leakage accident history of multiple customer pipelines. The reference pipeline dataset includes multiple reference pipeline data. The multiple reference pipeline data includes the pipeline length, burial period, map data, and leakage accident history of multiple reference pipelines. The method for generating the pipeline deterioration prediction model 5 of this embodiment includes a step (integrated pipeline data creation step S4) of integrating the customer pipeline dataset and the reference pipeline dataset to create an integrated pipeline dataset 38 consisting of multiple integrated pipeline data. The multiple integrated pipeline data includes multiple customer pipeline data and multiple reference pipeline data. The method for generating the pipeline deterioration prediction model 5 of this embodiment includes the steps of: extracting integrated pipeline data similar to the customer pipeline dataset from the integrated pipeline dataset 38 as similar pipeline data to create a similar pipeline dataset composed of similar pipeline data (similar pipeline data extraction step S5); and extracting multiple customer pipeline data that are not similar pipeline data from the integrated pipeline dataset 38 as first important pipeline data to create a first important pipeline dataset composed of first important pipeline data (first important pipeline data extraction step S20). The method for generating the pipeline deterioration prediction model 5 of this embodiment includes the step of creating a training pipeline dataset 48 (training pipeline dataset creation step S7). The step of creating the training pipeline dataset 48 includes combining the similar pipeline dataset and the first important pipeline dataset. The method for generating the pipeline deterioration prediction model 5 of this embodiment includes the step of generating the pipeline deterioration prediction model 5 using the training pipeline dataset 48 (pipeline deterioration prediction model generation step S8). The number of training pipeline data points included in the training pipeline dataset 48 is greater than the number of customer pipeline data points.

[0151] The training pipeline dataset 48 includes the customer pipeline dataset (pre-processed customer pipeline dataset; in this embodiment, the second customer pipeline dataset 33) and reference pipeline datasets (pre-processed reference pipeline dataset; in this embodiment, the second reference pipeline dataset 37) that are similar to the customer pipeline dataset. The number of training pipeline data in the training pipeline dataset 48 is greater than the number of customer pipeline data in the customer pipeline dataset. Furthermore, because the training pipeline dataset 48 includes similar pipeline datasets and the first important pipeline dataset, the characteristics of the customer pipeline dataset are reflected more strongly in the training pipeline dataset 48 than in the reference pipeline dataset. The pipeline deterioration prediction model 5 is then generated using this training pipeline dataset 48. As a result, a pipeline deterioration prediction model 5 is generated that enables more accurate prediction of the deterioration of customer pipelines.

[0152] The method for generating the pipeline deterioration prediction model 5 of this embodiment further includes the step of extracting integrated pipeline data similar to the first important pipeline data set from the integrated pipeline data set 38 as second important pipeline data, and creating a second important pipeline data set consisting of the second important pipeline data (second important pipeline data extraction step S21). The step of creating a learning pipeline data set 48 (learning pipeline data set creation step S7) includes combining the similar pipeline data set, the first important pipeline data set, and the second important pipeline data set.

[0153] Therefore, the number of training pipeline data points in the training pipeline dataset 48 increases further. Furthermore, the training pipeline dataset 48 more strongly reflects the characteristics of customer pipeline data than the reference pipeline dataset. The pipeline deterioration prediction model 5 is then generated using this training pipeline dataset 48. As a result, a pipeline deterioration prediction model 5 is generated that enables more accurate prediction of customer pipeline deterioration.

[0154] The program for the first phase of this embodiment (pipe deterioration prediction model generation program 8) causes the processor 202 to execute each step of the method for generating the pipe deterioration prediction model 5 of this embodiment.

[0155] Therefore, a pipeline deterioration prediction model 5 is generated, which enables more accurate prediction of the deterioration level of customer pipelines.

[0156] The pipeline deterioration prediction method of this embodiment comprises the steps of creating an input pipeline dataset 72 that includes the pipeline length, buried period, and map data of multiple customer pipelines (input pipeline dataset creation step S44), and inputting the input pipeline dataset 72 into the pipeline deterioration prediction model 5 generated by the pipeline deterioration prediction model 5 generation method of this embodiment to calculate the deterioration of multiple customer pipelines (customer pipeline deterioration calculation step S45).

[0157] By using the pipeline deterioration prediction model 5 of this embodiment to calculate the deterioration level of customer pipelines, it becomes possible to predict the deterioration level of customer pipelines more accurately.

[0158] The program for the second phase of this embodiment (pipe deterioration prediction program 75) causes the processor 302 to execute each step of the pipe deterioration prediction method of this embodiment.

[0159] By using the pipeline deterioration prediction model 5 of this embodiment to calculate the deterioration level of customer pipelines, it becomes possible to predict the deterioration level of customer pipelines more accurately.

[0160] The pipeline deterioration prediction device 3 of this embodiment includes an input pipeline data set creation unit 62 that creates an input pipeline data set 72 including the pipeline length, burial period, and map data of multiple customer pipelines, and a pipeline deterioration calculation unit 67 that inputs the input pipeline data set 72 into the pipeline deterioration prediction model 5 generated by the pipeline deterioration prediction model 5 generation method of this embodiment and calculates the deterioration of multiple customer pipelines.

[0161] By using the pipeline deterioration prediction model 5 of this embodiment to calculate the deterioration level of customer pipelines, it becomes possible to predict the deterioration level of customer pipelines more accurately.

[0162] (Embodiment 2) <Pipeline Deterioration Prediction System 1> Referring to Figure 1, the pipeline deterioration prediction system 1 of this embodiment will be described. The pipeline deterioration prediction system 1 of this embodiment has the same configuration as the pipeline deterioration prediction system 1 of Embodiment 1, but instead of the pipeline deterioration prediction model generation device 2 and pipeline deterioration prediction device 3 of Embodiment 1, it is equipped with the pipeline deterioration prediction model generation device 2b and pipeline deterioration prediction device 3b of this embodiment.

[0163] <Pipeline Deterioration Prediction Model Generator 2b> Referring to Figure 1, the pipeline deterioration prediction model generation device 2b generates pipeline deterioration prediction model 5 (see Figures 40 and 41) and general-purpose pipeline deterioration prediction model 76 (see Figures 40 and 42) from the original customer pipeline dataset 31 (see Figure 5) and the original reference pipeline dataset 35 (see Figure 6).

[0164] <Hardware Configuration> Referring to Figure 2, the hardware configuration of the pipeline deterioration prediction model generation device 2b in this embodiment is the same as the hardware configuration of the pipeline deterioration prediction model generation device 2b in Embodiment 1.

[0165] For example, the processor 202 executes the processing necessary to realize the functions of the pipeline deterioration prediction model generation device 2b by running a program described later. The processor 202 is composed of, for example, a CPU or a GPU.

[0166] Storage 210 stores the original customer pipeline dataset 31 (see Figure 5), the original reference pipeline dataset 35 (see Figure 6), the pipeline deterioration prediction model 5 (see Figures 40 and 41), the general-purpose pipeline deterioration prediction model 76 (see Figures 40 and 42), and programs executed by the processor 202. This program includes a pipeline deterioration prediction model generation program 8 (see Figure 40) and a general-purpose pipeline deterioration prediction model generation program 9 (see Figure 40). The pipeline deterioration prediction model generation program 8 is a program for generating the pipeline deterioration prediction model 5 from the original customer pipeline dataset 31 and the original reference pipeline dataset 35. The general-purpose pipeline deterioration prediction model generation program 9 is a program for generating the general-purpose pipeline deterioration prediction model 76 from the original reference pipeline dataset 35. Storage 210 is, for example, a non-volatile memory device such as a hard disk or SSD.

[0167] The program for realizing the functions of the pipeline aging prediction model generation device 2b (including the pipeline aging prediction model generation program 8 and the general-purpose pipeline aging prediction model generation program 9) may be stored and distributed on a non-transient storage medium 208 and installed on the storage device 210. The program for realizing the functions of the pipeline aging prediction model generation device 2b may also be downloaded to the pipeline aging prediction model generation device 2b via a communication network 4 such as the Internet or an intranet (see Figure 1).

[0168] In this embodiment, an example is shown in which a general-purpose computer (processor 202) implements the functions of the pipeline aging prediction model generation device 2b by executing a program. However, the embodiment is not limited to this, and all or part of the functions of the pipeline aging prediction model generation device 2b may be implemented using an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field-Programmable Gate Array).

[0169] <Functional Configuration> Referring to Figures 39 to 42, the functional configuration of the pipeline deterioration prediction model generation device 2b will be explained. The functional configuration of the pipeline deterioration prediction model generation device 2b in this embodiment is the same as that of the pipeline deterioration prediction model generation device 2 in Embodiment 1, but differs from the functional configuration of the pipeline deterioration prediction model generation device 2 in Embodiment 1 in the following respects.

[0170] <Storage section 25> Referring to Figure 40, the storage unit 25 of this embodiment has the same functions as the storage unit 25 of Embodiment 1, but the storage unit 25 of this embodiment also stores a general-purpose pipeline deterioration prediction model 76 (see Figures 40 and 42) and a general-purpose pipeline deterioration prediction model generation program 9 (see Figure 40).

[0171] Referring to Figure 42, the general-purpose pipeline deterioration prediction model 76 includes general-purpose pipeline deterioration prediction models 76a and 76b for each pipe material. For example, the general-purpose pipeline deterioration prediction model 76 includes the general-purpose pipeline deterioration prediction model 76a for rigid polyvinyl chloride (VP) pipes and the general-purpose pipeline deterioration prediction model 76b for ductile cast iron (DIP) pipes. The general-purpose pipeline deterioration prediction model 76 outputs the general deterioration of the pipeline. The general deterioration of the pipeline is not particularly limited, but for example, it is the leakage accident rate of the pipeline. The leakage accident rate of the pipeline means, for example, the number of leakage accidents per unit time per year and per unit pipeline length of 1 km. The unit of the pipeline leakage accident rate is, for example, accidents / year / km.

[0172] The general-purpose pipeline deterioration prediction model 76 includes, for example, general-purpose pipeline leakage accident probability prediction models 77a and 77b (see Figure 42) and conversion units 78a and 78b (see Figure 42). For example, the general-purpose pipeline deterioration prediction model 76a for VP pipes includes the general-purpose pipeline leakage accident probability prediction model 77a for VP pipes and the conversion unit 78a for VP pipes. The general-purpose pipeline deterioration prediction model 76b for DIP pipes includes the general-purpose pipeline leakage accident probability prediction model 77b for DIP pipes and the conversion unit 78b for DIP pipes. The general-purpose pipeline leakage accident probability prediction models 77a and 77b output the predicted leakage accident probability of the pipeline. The predicted leakage accident probability of the pipeline means the leakage accident probability of the pipeline predicted by the general-purpose pipeline leakage accident probability prediction models 77a and 77b. The conversion units 78a and 78b convert the predicted probability of a pipeline leak into the pipeline leak rate.

[0173] The general-purpose pipeline deterioration prediction model generation program 9 (see Figure 40) is a program for generating a general-purpose pipeline deterioration prediction model 76 (see Figure 42) from the original reference pipeline dataset 35 (see Figure 6).

[0174] Referring to Figure 43, the learning pipeline dataset 48 of this embodiment (see Figure 43) has a similar data structure to the learning pipeline dataset 48 of Embodiment 1 (see Figure 7), but further includes the general-purpose deterioration of pipelines calculated by the general-purpose pipeline deterioration prediction model 76 (see Figure 42).

[0175] The pipeline deterioration prediction model 5 of this embodiment differs from the pipeline deterioration prediction model 5 of Embodiment 1 in that it is generated using the learning pipeline dataset 48 of this embodiment (see Figure 43) rather than the learning pipeline dataset 48 of Embodiment 1 (see Figure 7).

[0176] <Pipe material specification section 11> Referring to Figure 39, the pipe material designation unit 11 of this embodiment has the same function as the pipe material designation unit 11 of Embodiment 1, but further designates the pipe material for which the general-purpose pipeline deterioration prediction models 76a and 76b should be created. When the pipe material designation unit 11 designates the pipe material for which the general-purpose pipeline deterioration prediction models 76a and 76b should be created, the pipe material designated by the pipe material designation unit 11 is one of the pipe materials included in the original reference pipeline dataset 35 (see Figure 6), and is also called the designated pipe material.

[0177] <Training Dataset Creation Section 12> Referring to Figure 39, the training dataset creation unit 12 of this embodiment has the same functions as the training dataset creation unit 12 of Embodiment 1, but instead of the pipeline data preprocessing unit 13 of Embodiment 1 (see Figure 3), it includes the pipeline data preprocessing unit 13 of this embodiment.

[0178] <Pipeline data preprocessing unit 13> Referring to Figure 39, the pipeline data preprocessing unit 13 of this embodiment creates a second reference pipeline dataset 37 (see Figure 11) from the original reference pipeline dataset 35 (see Figure 6), similar to the pipeline data preprocessing unit 13 of Embodiment 1 (see Figure 3). Furthermore, the pipeline data preprocessing unit 13 of this embodiment creates a pre-processed customer pipeline dataset for specified pipe materials (in this embodiment, a third customer pipeline dataset 34 (see Figure 44)) from the original customer pipeline dataset 31, and also creates a pre-processed reference pipeline dataset for specified pipe materials (in this embodiment, a third reference pipeline dataset 39 (see Figure 45)) from the original reference pipeline dataset 35.

[0179] The pipeline data preprocessing unit 13 includes, for example, a first pipeline data preprocessing unit 13a and a second pipeline data preprocessing unit 13b, as well as a third pipeline data preprocessing unit 13c. The first pipeline data preprocessing unit 13a and the second pipeline data preprocessing unit 13b in this embodiment have the same functions as the first pipeline data preprocessing unit 13a and the second pipeline data preprocessing unit 13b in Embodiment 1.

[0180] <Third pipeline data preprocessing unit 13c> Referring to Figure 39, the third pipeline data preprocessing unit 13c creates the third customer pipeline dataset 34 (see Figure 44) from the second customer pipeline dataset 33 (see Figure 10). The third customer pipeline dataset 34 further includes the general deterioration rate of customer pipelines in addition to the second customer pipeline dataset 33. The general deterioration rate of customers is obtained by inputting the second customer pipeline dataset 33 into the general pipeline deterioration rate prediction model 76 (see Figures 40 and 42). The third pipeline data preprocessing unit 13c creates the third reference pipeline dataset 39 (see Figure 45) from the second reference pipeline dataset 37 (see Figure 11). The third reference pipeline dataset 39 further includes the general pipeline deterioration rate of reference pipelines in addition to the second reference pipeline dataset 37 (see Figure 11). The general-purpose pipeline deterioration level of the reference pipeline is obtained by inputting the second reference pipeline dataset 37 into the general-purpose pipeline deterioration prediction model 76 (see Figures 40 and 42).

[0181] The third pipeline data preprocessing unit 13c includes, for example, a general-purpose model selection unit 81, a general-purpose pipeline deterioration calculation unit 83, and a general-purpose pipeline deterioration coupling unit 84.

[0182] <General-purpose model selection section 81> Referring to Figure 39, the general-purpose model selection unit 81 selects general-purpose pipeline deterioration prediction models 76a and 76b for the specified pipe material from the general-purpose pipeline deterioration prediction models 76 (see Figures 40 and 42) stored in the memory unit 25 (see Figures 39 and 40). The specified pipe material is the pipe material specified by the pipe material specification unit 11.

[0183] <General-purpose pipeline deterioration calculation unit 83> Referring to Figure 39, the general-purpose pipeline deterioration calculation unit 83 inputs the second customer pipeline dataset 33 (see Figure 10) of the specified pipe material, created by the second pipeline data preprocessing unit 13b, into the general-purpose pipeline deterioration prediction models 76a and 76b (see Figures 40 and 42) of the specified pipe material selected by the general-purpose model selection unit 81, and calculates the general-purpose deterioration of the customer pipeline. The general-purpose deterioration of the customer pipeline is the leakage accident rate (incidents / year / km) of the customer pipeline calculated by the general-purpose pipeline deterioration prediction models 76a and 76b.

[0184] The general-purpose pipeline deterioration calculation unit 83 inputs the second reference pipeline dataset 37 (see Figure 11) for the specified pipe material, created by the second pipeline data preprocessing unit 13b, into the general-purpose pipeline deterioration prediction models 76a and 76b (see Figures 40 and 42) for the specified pipe material selected by the general-purpose model selection unit 81, and calculates the general-purpose deterioration of the reference pipeline. The general-purpose deterioration of the reference pipeline is the leakage accident rate (incidents / year / km) of the reference pipeline calculated by the general-purpose pipeline deterioration prediction models 76a and 76b.

[0185] Specifically, the general-purpose pipeline deterioration calculation unit 83 inputs the explanatory variables (pipe attribute data 31a (e.g., diameter, pipeline length, and burial period) and map data 31m (e.g., soil classification, average annual temperature, and average annual precipitation)) of the second customer pipeline dataset 33 (see Figure 10) for the specified pipe material created by the second pipeline data preprocessing unit 13b into the general-purpose pipeline deterioration prediction models 76a and 76b (see Figure 42) of the general-purpose pipeline deterioration prediction models 76a and 76b for the specified pipe material selected by the general-purpose model selection unit 81, and calculates the general-purpose leakage accident prediction probability of the customer pipeline. The general-purpose pipeline deterioration calculation unit 83 inputs the general-purpose leakage accident prediction probability of the customer pipeline into the conversion units 78a and 78b (see Figure 42) of the general-purpose pipeline deterioration prediction models 76a and 76b for the specified pipe material selected by the general-purpose model selection unit 81, and calculates the general-purpose leakage accident rate of the customer pipeline.

[0186] The general-purpose pipeline deterioration calculation unit 83 inputs the explanatory variables (pipe attribute data 35b (e.g., diameter, pipeline length, and burial period) and map data 35m (e.g., soil classification, average annual temperature, and average annual precipitation)) of the second reference pipeline dataset 37 (see Figure 11) for the specified pipe material created by the second pipeline data preprocessing unit 13b into the general-purpose pipeline deterioration prediction models 76a and 76b (see Figure 42) for the specified pipe material selected by the general-purpose model selection unit 81, and calculates the general-purpose leakage accident prediction probability of the reference pipeline. The general-purpose pipeline deterioration calculation unit 83 inputs the general-purpose leakage accident prediction probability of the reference pipeline into the conversion units 78a and 78b (see Figure 42) of the general-purpose pipeline deterioration prediction models 76a and 76b for the specified pipe material selected by the general-purpose model selection unit 81, and calculates the general-purpose leakage accident rate of the reference pipeline.

[0187] <General-purpose pipeline aging joint 84> Referring to Figure 39, the general-purpose pipeline deterioration coupling unit 84 combines the general-purpose deterioration of the customer pipeline calculated by the general-purpose pipeline deterioration calculation unit 83 with the second customer pipeline dataset 33 (see Figure 10) of the specified pipe material, which was created by the second pipeline data preprocessing unit 13b. In this way, the third customer pipeline dataset 34 (see Figure 44) is created. The general-purpose pipeline deterioration coupling unit 84 combines the general-purpose deterioration of the reference pipeline calculated by the general-purpose pipeline deterioration calculation unit 83 with the second reference pipeline dataset 37 (see Figure 11) of the specified pipe material. In this way, the third reference pipeline dataset 39 (see Figure 45) is created.

[0188] <Pipeline Data Integration Unit 15> Referring to Figure 39, the pipeline data integration unit 15 integrates the pre-processed customer pipeline dataset for the specified pipe material and the pre-processed reference pipeline dataset for the specified pipe material to create an integrated pipeline dataset 38 for the specified pipe material (see Figure 46). In this embodiment, the pre-processed customer pipeline dataset for the specified pipe material is the third customer pipeline dataset 34 (see Figure 44), and the pre-processed reference pipeline dataset for the specified pipe material is the third reference pipeline dataset 39 (see Figure 45). The integrated pipeline dataset 38 in this embodiment is similar to the integrated pipeline dataset 38 in Embodiment 1 (Figure 12), but further includes the general deterioration degree of the pipeline as an explanatory variable.

[0189] The integrated pipeline dataset 38 includes multiple integrated pipeline data sets, each composed of multiple pre-processed customer pipeline data sets and multiple pre-processed reference pipeline data sets. The multiple integrated pipeline data sets include, for example, the entity name, integrated pipeline ID, attribute data 38b, location 38p, leakage accident history, map data 38m, and general deterioration level. The pipeline data integration unit 15 assigns a new integrated pipeline ID to each integrated pipeline data set. The attribute data 38b of the integrated pipeline includes, for example, the pipeline diameter, pipe material, pipeline length, and burial period. The map data 38m of the pipeline includes, for example, the soil classification of the pipeline, as well as the average annual temperature and average annual precipitation at the pipeline location 38p.

[0190] <Similar pipeline data extraction unit 16> Referring to Figure 39, the similar pipeline data extraction unit 16 extracts integrated pipeline data from the integrated pipeline dataset 38 (see Figure 46) of the specified pipe material that are similar to the pre-processed customer pipeline dataset of the specified pipe material, and creates a similar pipeline dataset composed of the similar pipeline data. In this embodiment, the pre-processed customer pipeline dataset of the specified pipe material is the third customer pipeline dataset 34 (see Figure 44). Integrated pipeline data similar to the pre-processed customer pipeline dataset of the specified pipe material is extracted, as in Embodiment 1, based on the distance between the data center of each integrated pipeline data and the data center of the pre-processed customer pipeline dataset in a multidimensional space defined by the dimensions of the explanatory variables of the integrated pipeline dataset 38 (see Figure 46) (see Figure 17), or the cosine similarity between the data center of each integrated pipeline data and the data center of the pre-processed customer pipeline dataset (see Figure 18). The integrated pipeline dataset 38 in this embodiment further includes the general deterioration degree of the pipeline as an explanatory variable.

[0191] <Important pipeline data extraction unit 17> Referring to Figure 39, the critical pipeline data extraction unit 17 extracts critical pipeline data from the integrated pipeline dataset 38 (see Figure 46) to create a critical pipeline dataset composed of critical pipeline data. The critical pipeline data is integrated pipeline data from the integrated pipeline dataset 38 that is important for improving the prediction accuracy of the pipeline deterioration prediction model 5 in this embodiment. The critical pipeline data extraction unit 17 in this embodiment operates similarly to the critical pipeline data extraction unit 17 in Embodiment 1. The critical pipeline data extraction unit 17 includes, for example, a first critical pipeline data extraction unit 17a and a second critical pipeline data extraction unit 17b.

[0192] The critical pipeline data includes first critical pipeline data extracted by the first critical pipeline data extraction unit 17a and second critical pipeline data extracted by the second critical pipeline data extraction unit 17b. The critical pipeline dataset includes a first critical pipeline dataset composed of the first critical pipeline data and a second critical pipeline dataset composed of the second critical pipeline data. The critical pipeline dataset, first critical pipeline dataset, and second critical pipeline dataset of this embodiment are the same as the first critical pipeline dataset and second critical pipeline dataset of Embodiment 1, but further include the general aging degree of the pipeline as an explanatory variable.

[0193] <First Important Pipeline Data Extraction Unit 17a> The first critical pipeline data extraction unit 17a (see Figure 39) in this embodiment, similar to the first critical pipeline data extraction unit 17a in Embodiment 1 (see Figure 3), extracts pre-processed customer pipeline data that was not extracted by the similar pipeline data extraction unit 16 from the integrated pipeline dataset 38 (see Figure 46) of the specified pipe material, as first critical pipeline data, and creates a first critical pipeline dataset composed of the first critical pipeline data. The first critical pipeline data extraction unit 17a extracts the first critical pipeline data because the pre-processed customer pipeline data that was not extracted by the similar pipeline data extraction unit 16 is also important for improving the prediction accuracy of the pipeline deterioration prediction model 5. In this embodiment, the pre-processed customer pipeline dataset is the third customer pipeline dataset 34 (see Figure 44).

[0194] <Second important pipeline data extraction unit 17b> The second critical pipeline data extraction unit 17b in this embodiment (see Figure 39), similar to the second critical pipeline data extraction unit 17b in Embodiment 1 (see Figure 3), extracts integrated pipeline data similar to the first critical pipeline data set from the integrated pipeline data set 38 (see Figure 46) of the specified pipe material as second critical pipeline data, and creates a second critical pipeline data set composed of the second critical pipeline data. Since integrated pipeline data similar to the first critical pipeline data set from the integrated pipeline data set 38 is also important for improving the prediction accuracy of the pipeline deterioration prediction model 5, the second critical pipeline data extraction unit 17b extracts the second critical pipeline data.

[0195] <Pipeline data connection section 18> Referring to Figure 39, the pipeline data merging unit 18 combines the similar pipeline dataset extracted by the similar pipeline data extraction unit 16 with the important pipeline dataset extracted by the important pipeline data extraction unit 17 to create a learning pipeline dataset 48 for the specified pipe material (see Figure 43). The pipeline data merging unit 18 assigns a new pipeline ID (learning pipeline ID) to the learning pipeline data that constitutes the learning pipeline dataset 48. The learning pipeline dataset 48 in this embodiment is similar to the learning pipeline dataset 48 in Embodiment 1 (see Figure 7), but further includes the general deterioration degree of the pipeline as an explanatory variable.

[0196] <Pipeline Deterioration Prediction Model Generation Unit 20> Referring to Figure 39, the pipeline deterioration prediction model generation unit 20 generates a pipeline deterioration prediction model 5 (see Figures 40 and 41) using the learning pipeline dataset 48 of this embodiment (see Figure 43) instead of the learning pipeline dataset 48 of Embodiment 1 (see Figure 7). The pipeline deterioration prediction model generation unit 20 generates a general-purpose pipeline deterioration prediction model 76 (see Figures 40 and 42) using the second reference pipeline dataset 37 (see Figure 11) instead of the learning pipeline dataset 48 of Embodiment 1. The pipeline deterioration prediction model generation unit 20 of this embodiment includes a pipeline leakage accident probability prediction model learning unit 21, a conversion unit generation unit 22, and a pipeline deterioration prediction model generation completion determination unit 23, similar to the pipeline deterioration prediction model generation unit 20 of Embodiment 1.

[0197] <Pipeline Leakage Accident Probability Prediction Model Learning Unit 21> Referring to Figure 39, the pipeline leakage accident probability prediction model learning unit 21 generates general-purpose pipeline leakage accident probability prediction models 77a and 77b by machine learning using the second reference pipeline dataset 37 (see Figure 11) as training data. The machine learning for generating the general-purpose pipeline leakage accident probability prediction models 77a and 77b in this embodiment is the same as the machine learning for generating the pipeline leakage accident probability prediction models 6a and 6b in Embodiment 1, but differs from the machine learning for generating the pipeline leakage accident probability prediction models 6a and 6b in Embodiment 1 in that the training data used for machine learning is the second reference pipeline dataset 37 (see Figure 11).

[0198] Referring to Figure 39, the pipeline leakage accident probability prediction model learning unit 21 generates pipeline leakage accident probability prediction models 6a and 6b by machine learning using the learning pipeline dataset 48 (see Figure 43) as training data. The machine learning for generating the pipeline leakage accident probability prediction models 6a and 6b in this embodiment is the same as the machine learning for generating the pipeline leakage accident probability prediction models 6a and 6b in Embodiment 1, but differs from the machine learning for generating the pipeline leakage accident probability prediction models 6a and 6b in Embodiment 1 in that the training data used for machine learning is the learning pipeline dataset 48 (see Figure 43).

[0199] <Conversion Unit / Generation Unit 22> Referring to Figure 39, the conversion unit generation unit 22 of this embodiment has the same functions as the conversion unit generation unit 22 of Embodiment 1, but further generates conversion units 78a and 78b (see Figure 42) of the general-purpose pipeline deterioration prediction model 76. The conversion units 78a and 78b convert the pipeline leakage accident prediction probability calculated by the general-purpose pipeline leakage accident probability prediction models 77a and 77b into a pipeline leakage accident rate. In addition to the conversion units 7a and 7b, the conversion unit generation unit 22 outputs the conversion units 78a and 78b to the storage unit 25 (see Figures 39 and 40). In addition to the conversion units 7a and 7b, the conversion units 78a and 78b are stored in the storage unit 25.

[0200] <Pipeline deterioration prediction model generation completion determination unit 23> Referring to Figure 39, the pipeline deterioration prediction model generation completion determination unit 23 of this embodiment has the same functions as the pipeline deterioration prediction model generation completion determination unit 23 of Embodiment 1. However, the pipeline deterioration prediction model generation completion determination unit 23 of this embodiment further determines whether it has generated general-purpose pipeline deterioration prediction models 76a and 76b for all pipe materials included in the original reference pipeline dataset 35 (see Figure 6). If the pipeline deterioration prediction model generation completion determination unit 23 determines that it has not generated general-purpose pipeline deterioration prediction models 76a and 76b for all pipe materials included in the original reference pipeline dataset 35, the pipe material specification unit 11 (see Figure 39) specifies the pipe materials for which general-purpose pipeline deterioration prediction models 76a and 76b have not yet been created among all pipe materials included in the original reference pipeline dataset 35. If the pipeline deterioration prediction model generation completion determination unit 23 determines that it has generated general-purpose pipeline deterioration prediction models 76a and 76b for all pipe materials included in the original reference pipeline dataset 35, it terminates the generation of the general-purpose pipeline deterioration prediction model 76.

[0201] <Method for generating the general-purpose pipeline deterioration prediction model 76> The method for generating the general-purpose pipeline deterioration prediction model 76 (see Figures 40 and 42) is the same as the method for generating the pipeline deterioration prediction model in Embodiment 1 (see Figures 14 and 15), but the general-purpose pipeline deterioration prediction model 76 is generated using the second reference pipeline dataset 37 (see Figure 11) instead of the training pipeline dataset 48 (see Figure 7). The method for generating the general-purpose pipeline deterioration prediction model 76 will be explained with reference to Figures 47 to 49.

[0202] Referring to Figure 47, the method for generating the general-purpose pipeline deterioration prediction model 76 includes a pipe material specification step S51, a reference pipeline data preprocessing step S52, a general-purpose pipeline deterioration prediction model generation step S55, and a general-purpose model generation completion determination step S58.

[0203] Referring to Figure 47, the pipe material specification step S51 is the same as the pipe material specification step S2 in Embodiment 1 (see Figure 14). Specifically, the pipe material specification section 11 (see Figure 39) specifies the pipe material for which the general-purpose pipeline deterioration prediction model 76 should be created. The pipe material specified in the pipe material specification step S51 is one of the pipe materials included in the original reference pipeline dataset 35 (see Figure 6), and is called the specified pipe material. For example, a VP pipe is specified as the pipe material for which the general-purpose pipeline deterioration prediction model 76 should be created.

[0204] Referring to Figure 47, the reference pipeline data preprocessing step S52 is the same as the processing of the original reference pipeline dataset 35 (see Figure 6) in the pipeline data preprocessing step S3 (see Figure 14) of Embodiment 1. Specifically, the pipeline data preprocessing unit 13 preprocesses the original reference pipeline dataset 35 to create a second reference pipeline dataset 37 (see Figure 11) for the specified pipe material. Referring to Figure 48, the reference pipeline data preprocessing step S52 includes, for example, a first reference pipeline data preprocessing step S53 and a second reference pipeline data preprocessing step S54.

[0205] The first reference pipeline data preprocessing step S53 is the same as the processing of the original reference pipeline dataset 35 (see Figure 6) in the first pipeline data preprocessing step S3a (see Figure 15) of Embodiment 1. Specifically, the first pipeline data preprocessing unit 13a (see Figure 39) extracts the original reference pipeline data of the pipe material specified in the pipe material specification step S51 from the original reference pipeline dataset 35 to create the first reference pipeline dataset 36 (see Figure 9) for the specified pipe material.

[0206] The second reference pipeline data preprocessing step S54 in this embodiment is the same as the processing of the first reference pipeline dataset 36 (see Figure 9) in the second pipeline data preprocessing step S3b (see Figure 15) of Embodiment 1. Specifically, the second pipeline data preprocessing unit 13b (see Figure 39) creates a second reference pipeline dataset 37 (see Figure 11) for the specified pipe material from the first reference pipeline dataset 36 for the specified pipe material.

[0207] Referring to Figure 47, in the general-purpose pipeline deterioration prediction model generation step S55, the pipeline deterioration prediction model generation unit 20 (see Figure 39) generates general-purpose pipeline deterioration prediction models 76a and 76b for the specified pipe material using the second reference pipeline dataset 37 (see Figure 11) for the specified pipe material. The general-purpose pipeline deterioration prediction model generation step S55 is similar to the pipeline deterioration prediction model generation step S8 (see Figure 14) of Embodiment 1, but differs from the pipeline deterioration prediction model generation step S8 of Embodiment 1 in that the training data is the second reference pipeline dataset 37 for the specified pipe material. Referring to Figure 49, the general-purpose pipeline deterioration prediction model generation step S55 includes, for example, the general-purpose pipeline leakage accident probability prediction model generation step S56 and the conversion unit generation step S57.

[0208] Referring to Figure 49, in the general-purpose pipeline leak accident probability prediction model generation step S56, the pipeline leak accident probability prediction model learning unit 21 (see Figure 39) generates general-purpose pipeline leak accident probability prediction models 77a and 77b for the specified pipe material by machine learning using the second reference pipeline dataset 37 (see Figure 11) for the specified pipe material as training data. The general-purpose pipeline leak accident probability prediction model generation step S56 is similar to the pipeline leak accident probability prediction model generation step S36 (see Figure 28) of Embodiment 1, but differs from the pipeline leak accident probability prediction model generation step S36 of Embodiment 1 in that the training data is the second reference pipeline dataset 37 (see Figure 11) for the specified pipe material.

[0209] The prediction model (for example, the neural network model 50 shown in Figure 13) is trained using machine learning so that the general-purpose leak accident prediction probability of a reference pipeline output by the prediction model approaches the leak accident history of the reference pipeline included in the second reference pipeline dataset 37, thereby generating general-purpose pipeline leak accident probability prediction models 77a and 77b. The pipeline leak accident probability prediction model learning unit 21 outputs the general-purpose pipeline leak accident probability prediction models 77a and 77b for the specified pipe material to the storage unit 25 (see Figures 40 and 42). The general-purpose pipeline leak accident probability prediction models 77a and 77b for the specified pipe material are stored in the storage unit 25.

[0210] Referring to Figure 49, in conversion unit generation step S57, the conversion unit generation unit 22 (see Figure 39) generates conversion units 78a and 78b. The conversion units 78a and 78b convert the pipeline leakage accident prediction probability calculated by the general-purpose pipeline leakage accident probability prediction models 77a and 77b for the specified pipe material into a pipeline leakage accident rate. The conversion unit generation step S57 in this embodiment is the same as the conversion unit generation step S37 in Embodiment 1 (see Figure 28). The conversion unit generation unit 22 outputs the conversion units 78a and 78b to the storage unit 25 (see Figures 40 and 42). The conversion units 78a and 78b are stored in the storage unit 25. Thus, the general-purpose pipeline deterioration prediction models 76a and 76b for the specified pipe material (see Figures 40 and 42), which include the general-purpose pipeline leakage accident probability prediction models 77a and 77b for the specified pipe material and the corresponding conversion units 78a and 78b, are stored in the storage unit 25.

[0211] Referring to Figure 47, the general-purpose pipeline deterioration prediction model generation step S55 is performed from the pipe material specification step S51 to other specified pipe materials specified in the pipe material specification step S51, generating general-purpose pipeline deterioration prediction models 76a and 76b for multiple specified pipe materials (see Figures 40 and 42). The general-purpose pipeline deterioration prediction models 76a and 76b for multiple specified pipe materials are stored in the storage unit 25 (see Figures 3 and 4). Each of the general-purpose pipeline deterioration prediction models 76a and 76b for multiple specified pipe materials includes a general-purpose pipeline leakage accident probability prediction model 77a and 77b for the specified pipe material, and a corresponding conversion unit 78a and 78b. In this way, a general-purpose pipeline deterioration prediction model 76 is generated that includes the general-purpose pipeline deterioration prediction models 76a and 76b for multiple specified pipe materials.

[0212] Specifically, referring to Figure 47, in the general-purpose model generation completion determination step S58, the pipeline deterioration prediction model generation completion determination unit 23 (see Figure 39) determines whether it has generated general-purpose pipeline deterioration prediction models 76a and 76b for all pipe materials included in the original reference pipeline dataset 35 (see Figure 6). The general-purpose model generation completion determination step S58 in this embodiment is the same as the model generation completion determination step S9 (see Figure 14) in Embodiment 1.

[0213] In the general-purpose model generation completion determination step S58, if the pipeline deterioration prediction model generation completion determination unit 23 determines that it has not generated general-purpose pipeline deterioration prediction models 76a and 76b for all pipe materials included in the original reference pipeline dataset 35 (see Figure 6), the process returns to the pipe material specification step S51. In the pipe material specification step S51, the pipe material specification unit 11 (see Figure 39) specifies the pipe materials for which general-purpose pipeline deterioration prediction models 76a and 76b have not yet been created, out of all pipe materials included in the original reference pipeline dataset 35. The process from the pipe material specification step S51 to the general-purpose pipeline deterioration prediction model generation step S55 is repeatedly executed until general-purpose pipeline deterioration prediction models 76a and 76b for all pipe materials included in the original reference pipeline dataset 35 are generated. In this way, general-purpose pipeline deterioration prediction models 76a and 76b for all pipe materials included in the original reference pipeline dataset 35 are generated.

[0214] In the general-purpose model generation completion determination step S58, if the pipeline deterioration prediction model generation completion determination unit 23 (see Figure 4) determines that it has generated general-purpose pipeline deterioration prediction models 76a and 76b for all pipe materials included in the original reference pipeline dataset 35 (see Figure 6), it terminates the generation of the general-purpose pipeline deterioration prediction model 76.

[0215] The general-purpose pipeline deterioration prediction model generation program 9 (see Figure 40) causes the processor 202 (see Figure 2) to execute the method for generating the general-purpose pipeline deterioration prediction model 76 of this embodiment. The general-purpose pipeline deterioration prediction model generation program 9 of this embodiment may be recorded on a computer-readable storage medium (a non-transient computer-readable storage medium, for example, storage medium 208).

[0216] <Method for generating pipeline deterioration prediction model 5> The method for generating the pipeline deterioration prediction model 5 (Figures 40 and 41) in this embodiment is the same as the method for generating the pipeline deterioration prediction model in Embodiment 1 (see Figures 14 and 15), however, the pipeline deterioration prediction model 5 in this embodiment is generated using the training pipeline dataset 48 (see Figure 43) instead of the training pipeline dataset 48 (see Figure 7). The method for generating the pipeline deterioration prediction model 5 in this embodiment will be explained with reference to Figures 14 and 50 to 53.

[0217] Referring to Figures 14 and 50, the pipeline data preprocessing step S3 of this embodiment further includes a third pipeline data preprocessing step S3c. In the third pipeline data preprocessing step S3c, the third pipeline data preprocessing unit 13c creates a third customer pipeline dataset 34 (see Figure 44) for the specified pipe material from the second customer pipeline dataset 33 (see Figure 10) for the specified pipe material as a preprocessed customer pipeline dataset for the specified pipe material. The third pipeline data preprocessing unit 13c creates a third reference pipeline dataset 39 (see Figure 45) for the specified pipe material from the second reference pipeline dataset 37 (see Figure 11) for the specified pipe material as a preprocessed reference pipeline dataset for the specified pipe material.

[0218] Referring to Figure 51, the third pipeline data preprocessing step S3c includes a general-purpose model selection step S61, a third customer pipeline dataset creation step S62, and a third reference pipeline dataset creation step S65.

[0219] In the general-purpose model selection step S61, the general-purpose model selection unit 81 (see Figure 39) selects general-purpose pipeline deterioration prediction models 76a and 76b for the specified pipe material from the general-purpose pipeline deterioration prediction models 76 (see Figures 40 and 42) stored in the memory unit 25 (see Figures 39 and 40). The specified pipe material is the pipe material specified in the pipe material specification step S2 (see Figure 14).

[0220] Referring to Figure 52, the third customer pipeline dataset creation step S62 includes a customer pipeline general deterioration calculation step S63 and a customer pipeline general deterioration merging step S64.

[0221] In the customer pipeline general deterioration calculation step S63, the general pipeline deterioration calculation unit 83 (see Figure 39) inputs the second customer pipeline dataset 33 (see Figure 10) of the specified pipe material, created in the second pipeline data preprocessing step S3b, into the general pipeline deterioration prediction models 76a and 76b of the specified pipe material to calculate the general deterioration of the customer pipeline. The general deterioration of the customer pipeline is, for example, the leakage accident rate (incidents / year / km) of the customer pipeline calculated by the general pipeline deterioration prediction models 76a and 76b. The customer pipeline general deterioration calculation step S63 is similar to the customer pipeline deterioration calculation step S45 (see Figure 37) of Embodiment 1, but differs from the customer pipeline deterioration calculation step S45 of Embodiment 1 mainly in that the general pipeline deterioration prediction models 76a and 76b are used instead of the pipeline deterioration prediction models 5a and 5b of Embodiment 1.

[0222] Specifically, the general-purpose pipeline deterioration calculation unit 83 inputs the explanatory variables (pipe attribute data 31a (e.g., diameter, pipeline length, and burial period) and map data 31m (e.g., soil classification, average annual temperature, and average annual precipitation)) of the second customer pipeline dataset 33 (see Figure 10) for the specified pipe material created by the second pipeline data preprocessing unit 13b into the general-purpose pipeline deterioration prediction models 76a and 76b (see Figure 42) of the general-purpose pipeline deterioration prediction models 76a and 76b for the specified pipe material selected by the general-purpose model selection unit 81, and calculates the general-purpose leakage accident prediction probability of the customer pipeline. The general-purpose pipeline deterioration calculation unit 83 inputs the general-purpose leakage accident prediction probability of the customer pipeline into the conversion units 78a and 78b (see Figure 42) of the general-purpose pipeline deterioration prediction models 76a and 76b for the specified pipe material selected by the general-purpose model selection unit 81, and calculates the general-purpose leakage accident rate of the customer pipeline.

[0223] In the customer pipeline general-purpose deterioration merging step S64, the general-purpose pipeline deterioration merging unit 84 (see Figure 39) merges the general-purpose deterioration of the customer pipeline with the second customer pipeline dataset 33 (see Figure 10) for the specified pipe material. In this way, the third customer pipeline dataset 34 (see Figure 44) for the specified pipe material is created as a pre-processed customer pipeline dataset.

[0224] Referring to Figure 53, the third reference pipeline dataset creation step S65 includes the general reference pipeline deterioration calculation step S66 and the general reference pipeline deterioration merging step S67.

[0225] In the general-purpose deterioration calculation step S66 for the reference pipeline, the general-purpose pipeline deterioration calculation unit 83 (see Figure 39) inputs the second reference pipeline dataset 37 for the specified pipe material (see Figure 11) into the general-purpose pipeline deterioration prediction models 76a and 76b for the specified pipe material to calculate the general-purpose deterioration of the reference pipeline. The general-purpose deterioration of the reference pipeline is, for example, the leakage accident rate (incidents / year / km) of the reference pipeline calculated by the general-purpose pipeline deterioration prediction models 76a and 76b. The general-purpose deterioration calculation step S66 for the reference pipeline is similar to the customer pipeline deterioration calculation step S45 (see Figure 37) of Embodiment 1, but differs from the customer pipeline deterioration calculation step S45 of Embodiment 1 mainly in that the general-purpose pipeline deterioration prediction models 76a and 76b and the second reference pipeline dataset 37 are used.

[0226] Specifically, the general-purpose pipeline deterioration calculation unit 83 inputs the explanatory variables (pipe attribute data 35b (e.g., diameter, pipeline length, and burial period) and map data 35m (e.g., soil classification, average annual temperature, and average annual precipitation)) of the second reference pipeline dataset 37 (see Figure 11) for the specified pipe material created by the second pipeline data preprocessing unit 13b into the general-purpose pipeline deterioration prediction models 76a and 76b (see Figure 42) for the specified pipe material selected by the general-purpose model selection unit 81, and calculates the general-purpose leakage accident prediction probability of the reference pipeline. The general-purpose pipeline deterioration calculation unit 83 inputs the general-purpose leakage accident prediction probability of the reference pipeline into the conversion units 78a and 78b (see Figure 42) of the general-purpose pipeline deterioration prediction models 76a and 76b for the specified pipe material selected by the general-purpose model selection unit 81, and calculates the general-purpose leakage accident rate of the reference pipeline.

[0227] Referring to Figure 53, in the general-purpose deterioration coupling step S67 of the reference pipeline, the general-purpose pipeline deterioration coupling section 84 (see Figure 39) couples the general-purpose deterioration of the reference pipeline with the second reference pipeline dataset 37 of the specified pipe material (see Figure 11). In this way, the third reference pipeline dataset 39 of the specified pipe material (see Figure 45) is created as the pre-processed reference pipeline dataset.

[0228] Referring to Figure 14, in the integrated pipeline data creation step S4, the pipeline data integration unit 15 (see Figure 39) integrates the pre-processed customer pipeline dataset for the specified pipe material and the pre-processed reference pipeline dataset for the specified pipe material to create an integrated pipeline dataset 38 for the specified pipe material (see Figure 46). In this embodiment, the pre-processed customer pipeline dataset for the specified pipe material is the third customer pipeline dataset 34 (see Figure 44), and the pre-processed reference pipeline dataset for the specified pipe material is the third reference pipeline dataset 39 (see Figure 45). The integrated pipeline dataset 38 in this embodiment is similar to the integrated pipeline dataset 38 in Embodiment 1 (Figure 12), but further includes the general deterioration degree of the pipeline as an explanatory variable.

[0229] Referring to FIG. 14, in the similar pipeline data extraction step S5, the similar pipeline data extraction unit 16 (see FIG. 39) extracts integrated pipeline data similar to the preprocessed customer pipeline data set of the designated pipe material from the integrated pipeline data set 38 (see FIG. 46) of the designated pipe material. In the present embodiment, the preprocessed customer pipeline data of the designated pipe material is the third customer pipeline data set 34 (see FIG. 44). In the similar pipeline data extraction step S5 of the present embodiment, similar to the similar pipeline data extraction step S5 of the first embodiment, the integrated pipeline data similar to the preprocessed customer pipeline data set of the designated pipe material is the distance (see FIG. 17) between each integrated pipeline data and the data center of the preprocessed customer pipeline data set, or the cosine similarity (see FIG. 18) between each integrated pipeline data and the data center of the preprocessed customer pipeline data set in the multi-dimensional space defined by the dimension of the explanatory variables of the integrated pipeline data set 38. The integrated pipeline data set 38 of the present embodiment further includes the general aging degree of the pipeline as an explanatory variable. The dimension of the explanatory variables of the integrated pipeline data set 38 of the present embodiment is one more than the dimension of the explanatory variables of the integrated pipeline data set 38 (see FIG. 12) of the first embodiment.

[0230] Referring to FIG. 14, in the important pipeline data extraction step S6, the important pipeline data extraction unit 17 extracts important pipeline data from the integrated pipeline data set 38 (see FIG. 46) and creates an important pipeline data set composed of the important pipeline data. The important pipeline data is the integrated pipeline data important for improving the prediction accuracy of the pipeline aging prediction model 5 in the integrated pipeline data set 38. Referring to FIG. 20, the important pipeline data extraction step S6 includes, for example, a first important pipeline data extraction step S20 and a second important pipeline data extraction step S21. The important pipeline data includes the first important pipeline data extracted in the first important pipeline data extraction step S20 and the second important pipeline data extracted in the second important pipeline data extraction step S21.

[0231] In the first important pipeline data extraction step S20, the first important pipeline data extraction unit 17a extracts, as the first important pipeline data, the preprocessed customer pipeline data that is not similar pipeline data from the integrated pipeline data set 38 (see FIG. 46) of the specified pipe material, and creates a first important pipeline data set composed of the first important pipeline data. In the present embodiment, the preprocessed customer pipeline data is the third customer pipeline data set 34 (see FIG. 44).

[0232] In the second important pipeline data extraction step S21, the second important pipeline data extraction unit 17b extracts, as the second important pipeline data, the integrated pipeline data similar to the first important pipeline data set from the integrated pipeline data set 38 (see FIG. 46) of the specified pipe material, and creates a second important pipeline data set composed of the second important pipeline data.

[0233] Referring to FIG. 22, the second important pipeline data extraction step S21 of the present embodiment includes steps S22 to S26 in the same manner as the second important pipeline data extraction step S21 of Embodiment 1.

[0234] Specifically, the second important pipeline data extraction unit 17b calculates the variation of each explanatory variable in the first important pipeline data set (step S22). In the present embodiment, the first important pipeline data further includes the general aging degree of the pipeline as an explanatory variable. The second important pipeline data extraction unit 17b selects the explanatory variable with the least variation (step S23). The second important pipeline data extraction unit 17b calculates the most frequent value of the selected explanatory variable selected in step S23 (step S24).

[0235] The second critical pipeline data extraction unit 17b extracts integrated pipeline data from the integrated pipeline dataset 38 (see Figure 46) in which the selected explanatory variable has the most mode (step S25). From the integrated pipeline data extracted in step S25, the second critical pipeline data extraction unit 17b extracts integrated pipeline data similar to the first critical pipeline dataset in which the selected explanatory variable has the most mode, as the second critical pipeline data, and creates a second critical pipeline dataset consisting of the second critical pipeline data (step S26). A first example of step S26 includes steps S31 to S33, as shown in Figure 25. A second example of step S26 includes steps S31 to S33b, as shown in Figure 26. In this embodiment, the second critical pipeline data further includes the general deterioration degree of the pipeline as an explanatory variable.

[0236] Referring to Figure 14, in the learning pipeline dataset creation step S7, the pipeline data merging unit 18 (see Figure 39) combines the similar pipeline dataset extracted in the similar pipeline data extraction step S5 and the important pipeline dataset extracted in the important pipeline data extraction step S6 to create a learning pipeline dataset 48 (see Figure 43) for the specified pipe material. The pipeline data merging unit 18 assigns a new pipeline ID (learning pipeline ID) to the learning pipeline data that constitutes the learning pipeline dataset 48. The learning pipeline dataset 48 in this embodiment is similar to the learning pipeline dataset 48 in Embodiment 1 (Figure 7), but further includes the general deterioration rate of the pipeline.

[0237] Referring to Figure 14, in the pipeline deterioration prediction model generation step S8, the pipeline deterioration prediction model generation unit 20 (see Figure 39) generates pipeline deterioration prediction models 5a and 5b for the specified pipe material using the learning pipeline dataset 48 (see Figure 43) for the specified pipe material. The pipeline deterioration prediction model generation step S8 in this embodiment is the same as the pipeline deterioration prediction model generation step S8 in Embodiment 1, but differs from the pipeline deterioration prediction model generation step S8 in Embodiment 1 in that the learning pipeline dataset 48 (see Figure 43) further includes general pipeline deterioration.

[0238] For example, referring to Figure 28, in the pipeline leakage accident probability prediction model generation step S36, the pipeline leakage accident probability prediction model learning unit 21 (see Figure 39) uses the learning pipeline dataset 48 for the specified pipe material (see Figure 43) as learning data to generate pipeline leakage accident probability prediction models 6a and 6b for the specified pipe material.

[0239] Specifically, explanatory variables (pipe attribute data 48a (e.g., diameter, pipe length, and burial period), map data 48m (e.g., soil classification, average annual temperature, and average annual precipitation), and general deterioration level) from the training pipeline dataset 48 (see Figure 43) for specified pipe materials are input to a prediction model (e.g., the neural network model 50 shown in Figure 54). The prediction model outputs the predicted probability of a pipeline leak. The prediction model is trained using machine learning so that the predicted probability of a pipeline leak approaches the pipeline leak history included in the training pipeline dataset 48, and trained prediction models (pipe leak probability prediction models 6a, 6b (see Figure 41)) are generated.

[0240] Referring to Figure 28, in conversion unit generation step S37, the conversion unit generation unit 22 (see Figure 3) generates conversion units 7a and 7b (see Figure 41). The conversion unit generation step S37 in this embodiment is the same as the conversion unit generation step S37 in Embodiment 1.

[0241] Referring to Figure 14, the model generation completion determination step S9 in this embodiment is the same as the model generation completion determination step S9 in Embodiment 1. The pipe material specification step S2 to the pipeline deterioration prediction model generation step S8 is executed for all specified pipe materials included in the original customer pipeline dataset 31 (see Figure 5) to generate pipeline deterioration prediction models 5a and 5b (see Figures 40 and 41) for multiple specified pipe materials.

[0242] The pipeline deterioration prediction model generation unit 20 stores pipeline deterioration prediction models 5a and 5b for multiple specified pipe materials in the storage unit 25 (see Figures 39 and 40). Each of the pipeline deterioration prediction models 5a and 5b for multiple specified pipe materials includes a pipeline leakage accident probability prediction model 6a and 6b for the specified pipe material, and corresponding conversion units 7a and 7b. In this way, a pipeline deterioration prediction model 5 is generated that includes the pipeline deterioration prediction models 5a and 5b for multiple specified pipe materials.

[0243] The pipeline deterioration prediction model generation program 8 (see Figure 40) causes the processor 202 (see Figure 2) to execute the method for generating the pipeline deterioration prediction model 5 of this embodiment. The pipeline deterioration prediction model generation program 8 of this embodiment may be recorded on a computer-readable storage medium (a non-transient computer-readable storage medium, for example, storage medium 208) of this embodiment.

[0244] Referring to Figure 1, the pipeline deterioration prediction model generation device 2b transmits the pipeline deterioration prediction model 5 and the general-purpose pipeline deterioration prediction model 76 to the pipeline deterioration prediction device 3b.

[0245] <Pipeline deterioration prediction device 3b> Referring to Figure 1, the pipeline deterioration prediction device 3b receives the pipeline deterioration prediction model 5 (see Figures 40 and 41) and the general-purpose pipeline deterioration prediction model 76 (see Figures 40 and 42) from the pipeline deterioration prediction model generation device 2b. The pipeline deterioration prediction device 3b uses the pipeline deterioration prediction model 5 to calculate the deterioration of the pipeline (for example, the leakage accident rate of the pipeline) and uses the general-purpose pipeline deterioration prediction model 76 to calculate the general deterioration of the pipeline (for example, the leakage accident rate of the pipeline).

[0246] <Hardware Configuration> Referring to Figure 32, the hardware configuration of the pipeline deterioration prediction device 3b in this embodiment is the same as the hardware configuration of the pipeline deterioration prediction device 3 in Embodiment 1.

[0247] <Functional Configuration> Referring to Figures 55 and 56, the functional configuration of the pipeline deterioration prediction device 3b in this embodiment is the same as that of the pipeline deterioration prediction device 3 in Embodiment 1, but differs from the functional configuration of the pipeline deterioration prediction device 3 in Embodiment 1 mainly in the following points.

[0248] <Storage section 70> Referring to Figures 55 and 56, the storage unit 70 stores the pipeline deterioration prediction model 5 of this embodiment and the general-purpose pipeline deterioration prediction model 76 received from the pipeline deterioration prediction model generation device 2b of this embodiment.

[0249] The memory unit 70 stores the input pipeline dataset 72 of this embodiment (see Figures 44 and 56). The input pipeline dataset 72 is a pipeline dataset that is input to the pipeline deterioration prediction model 5 (see Figure 56). The input pipeline dataset 72 of this embodiment has the same data structure as the third customer pipeline dataset 34 (see Figure 44). The input pipeline dataset 72 of this embodiment is similar to the input pipeline dataset 72 of Embodiment 1 (see Figure 10), but further includes general pipeline deterioration.

[0250] <Input pipeline dataset creation unit 62> Referring to Figure 55, the input pipeline data set creation unit 62 creates an input pipeline data set 72 (see Figures 44 and 56) from the original customer pipeline data set 31 (see Figure 5). In addition to the first customer pipeline data processing unit 63 and the second customer pipeline data processing unit 64, the input pipeline data set creation unit 62 further includes a third customer pipeline data processing unit 85. The first customer pipeline data processing unit 63 and the second customer pipeline data processing unit 64 in this embodiment have the same functions as the first customer pipeline data processing unit 63 (see Figure 33) and the second customer pipeline data processing unit 64 (see Figure 33) in Embodiment 1.

[0251] <Third Customer Pipeline Data Processing Unit 85> Referring to FIG. 55, the third customer pipeline data processing unit 85 creates a third customer pipeline data set 34 (refer to FIG. 44) from the second customer pipeline data set 33 (refer to FIG. 10). In this embodiment, the third customer pipeline data set 34 is the input pipeline data set 72 (refer to FIGS. 44 and 56). The input pipeline data set 72 further includes the general aging degree of the customer pipeline in addition to the second customer pipeline data set 33 (refer to FIG. 10). The general aging degree of the customer pipeline is obtained by inputting the second customer pipeline data set 33 into the general pipeline aging prediction model 76 (refer to FIGS. 40 and 42). The third customer pipeline data processing unit 85 includes, for example, a general model selection unit 86, a general pipeline aging calculation unit 88, and a general pipeline aging combination unit 89.

[0252] <General model selection unit 86> Referring to FIG. 55, the general model selection unit 86 functions in the same manner as the general model selection unit 81 (refer to FIG. 39). Specifically, the general model selection unit 86 selects the general pipeline aging prediction models 76a and 76b for the specified pipe material from the general pipeline aging prediction model 76 (refer to FIGS. 56 and 42) stored in the storage unit 70 (refer to FIGS. 55 and 56). The specified pipe material is the pipe material specified by the pipe material specifying unit 61.

[0253] <General pipeline aging calculation unit 88> Referring to FIG. 55, the general pipeline aging calculation unit 88 functions in the same manner as the general pipeline aging calculation unit 83 (refer to FIG. 39). Specifically, the general pipeline aging calculation unit 88 inputs the second customer pipeline data set 33 (refer to FIG. 10) of the specified pipe material created by the second customer pipeline data processing unit 64 into the general pipeline aging prediction models 76a and 76b (refer to FIGS. 56 and 42) of the specified pipe material selected by the general model selection unit 86, and calculates the general aging degree of the customer pipeline. The general aging degree of the customer pipeline is, for example, the leakage accident rate (number of cases / year / km) of the customer pipeline calculated by the general pipeline aging prediction models 76a and 76b.

[0254] Specifically, the general-purpose pipeline deterioration calculation unit 88 inputs the explanatory variables (pipe attribute data 31b (e.g., diameter, pipeline length, and burial period) and map data 31m (e.g., soil classification, average annual temperature, and average annual precipitation)) of the second customer pipeline dataset 33 (see Figure 10) for the specified pipe material created by the second customer pipeline data processing unit 64 into the general-purpose pipeline deterioration prediction models 76a and 76b (see Figure 42) of the general-purpose pipeline deterioration prediction models 76a and 76b for the specified pipe material selected by the general-purpose model selection unit 86, and calculates the general-purpose leakage accident prediction probability of the customer pipeline. The general-purpose pipeline deterioration calculation unit 88 inputs the general-purpose leakage accident prediction probability of the customer pipeline into the conversion units 78a and 78b (see Figure 42) of the general-purpose pipeline deterioration prediction models 76a and 76b for the specified pipe material selected by the general-purpose model selection unit 86, and calculates the general-purpose leakage accident rate of the customer pipeline.

[0255] <General-purpose pipeline aging joint 89> Referring to Figure 55, the general-purpose pipeline deterioration coupling unit 89 functions similarly to the general-purpose pipeline deterioration coupling unit 84 (see Figure 39). Specifically, the general-purpose pipeline deterioration coupling unit 89 combines the general-purpose deterioration of the customer pipeline calculated by the general-purpose pipeline deterioration calculation unit 88 with the second customer pipeline dataset 33 (see Figure 10) of the specified pipe material, created by the second customer pipeline data processing unit 64. In this way, the input pipeline dataset 72 (see Figures 44 and 56) is created.

[0256] <Method for predicting pipeline deterioration> The pipeline deterioration prediction method of this embodiment will be described with reference to Figures 37, 57, and 58. The pipeline deterioration prediction method of this embodiment is similar to the pipeline deterioration prediction method of Embodiment 1 shown in Figure 37, but differs from the pipeline deterioration prediction method of Embodiment 1 mainly in that the pipeline deterioration prediction model 5 of this embodiment (see Figures 41 and 56) and the input pipeline dataset 72 (see Figures 44 and 56) are used.

[0257] Referring to Figure 37, the original customer pipeline data set reception step S41 and pipe material specification step S42 in this embodiment are the same as the original customer pipeline data set reception step S41 and pipe material specification step S42 in Embodiment 1 (see Figure 37).

[0258] Referring to Figure 37, in the model selection step S43, the model selection unit 66 (see Figure 55) selects the pipeline deterioration prediction model 5a, 5b for the specified pipe material from a plurality of pipeline deterioration prediction models 5a, 5b (see Figures 41 and 56) stored in the memory unit 70 (see Figures 55 and 56). The model selection step S43 in this embodiment is the same as the model selection step S43 in Embodiment 1 (see Figure 37), however, the pipeline deterioration prediction models 5a, 5b selected in the model selection step S43 of this embodiment are the pipeline deterioration prediction models 5a, 5b (see Figures 40 and 41) generated by the pipeline deterioration prediction model 5 generation method of this embodiment (see Figures 47 to 53).

[0259] Referring to Figure 37, in the input pipeline dataset creation step S44, the input pipeline dataset creation unit 62 (see Figure 55) creates the input pipeline dataset 72 (see Figures 44 and 56) from the original customer pipeline dataset 31 (see Figure 5). The input pipeline dataset 72 in this embodiment has the same data structure as the third customer pipeline dataset 34 (see Figure 44). The input pipeline dataset creation step S44 in this embodiment is similar to the input pipeline dataset creation step S44 in Embodiment 1 (see Figure 37), but as shown in Figure 57, it further includes the third customer pipeline data processing step S44c in addition to the first customer pipeline data processing step S44a and the second customer pipeline data processing step S44b.

[0260] The first customer pipeline data processing step S44a and the second customer pipeline data processing step S44b in this embodiment are the same as the first customer pipeline data processing step S44a and the second customer pipeline data processing step S44b in Embodiment 1 (see Figure 38).

[0261] Referring to Figure 58, the third customer pipeline data processing step S44c of this embodiment includes a general-purpose model selection step S71, a general-purpose customer pipeline deterioration calculation step S72, and a general-purpose customer pipeline deterioration merging step S73.

[0262] In the general-purpose model selection step S71, the general-purpose model selection unit 86 (see Figure 55) selects general-purpose pipeline deterioration prediction models 76a and 76b for the specified pipe material from the general-purpose pipeline deterioration prediction models 76 (see Figures 56 and 42) stored in the memory unit 70 (see Figures 55 and 56). The specified pipe material is the pipe material specified in the pipe material specification step S42 (see Figure 37). The general-purpose model selection step S71 in this embodiment is the same as the general-purpose model selection step S61 (see Figure 51) in this embodiment.

[0263] In the general-purpose deterioration calculation step S72 for customer pipelines, the general-purpose pipeline deterioration calculation unit 88 (see Figure 55) inputs the second customer pipeline dataset 33 (see Figure 10) for the specified pipe material, created in the second customer pipeline data processing step S44b, into the general-purpose pipeline deterioration prediction models 76a and 76b for the specified pipe material, and calculates the general deterioration of the customer pipelines included in the second customer pipeline dataset 33 for the specified pipe material. The general-purpose deterioration calculation step S72 for customer pipelines in this embodiment is the same as the general-purpose deterioration calculation step S63 (see Figure 52) for customer pipelines in this embodiment.

[0264] In the customer pipeline general-purpose deterioration coupling step S73, the general-purpose pipeline deterioration coupling unit 89 (see Figure 55) couples the general-purpose deterioration of the customer pipeline with the second customer pipeline dataset 33 (see Figure 10) of the specified pipe material. The customer pipeline general-purpose deterioration coupling step S73 in this embodiment is the same as the customer pipeline general-purpose deterioration coupling step S64 (see Figure 52) in this embodiment. In this way, the input pipeline dataset 72 (see Figures 44 and 56) is created.

[0265] Referring to Figure 37, in the customer pipeline deterioration calculation step S45, the pipeline deterioration calculation unit 67 (see Figure 55) inputs the input pipeline dataset 72 for the specified pipe material created in the input pipeline dataset creation step S44 (see Figures 44 and 56) into the pipeline deterioration prediction models 5a and 5b for the specified pipe material selected in the model selection step S43, and calculates the deterioration of the customer pipeline. The customer pipeline deterioration calculation step S45 in this embodiment is the same as the customer pipeline deterioration calculation step S45 in Embodiment 1 (see Figure 37).

[0266] Specifically, the pipeline deterioration calculation unit 67 (see Figure 55) inputs the explanatory variables (pipe attribute data 31b (e.g., diameter, pipe length, and burial period), map data 31m (e.g., soil classification, average annual temperature, and average annual precipitation, etc.), and general deterioration) from the input pipeline dataset 72 (see Figures 44 and 56) created in the input pipeline dataset creation step S44 for the specified pipe material to the pipeline deterioration prediction models 5a and 5b (see Figures 41 and 56) of the specified pipe material selected in the model selection step S43, and calculates the predicted leakage accident probability of the customer pipeline. The pipeline deterioration calculation unit 67 inputs the predicted leakage accident probability of the customer pipeline to the conversion units 7a and 7b (see Figures 41 and 56) of the pipeline deterioration prediction models 5a and 5b for the specified pipe material selected in the model selection step S43, and calculates the leakage accident rate of the customer pipeline.

[0267] Referring to Figure 37, the pipeline deterioration calculation completion determination step S46 in this embodiment is the same as the pipeline deterioration calculation completion determination step S46 in Embodiment 1. The pipe material specification step S42 to the customer pipeline deterioration calculation step S45 described above are also executed for other specified pipe materials specified in the pipe material specification step S42 to calculate the deterioration of all customer pipelines included in the original customer pipeline data set 31 (see Figure 5). The pipeline deterioration prediction result creation step S47 in this embodiment is the same as the pipeline deterioration prediction result creation step S47 in Embodiment 1.

[0268] The pipeline deterioration prediction program 75 (see Figure 56) causes the processor 302 (see Figure 32) to execute the pipeline deterioration prediction method of this embodiment. The computer-readable storage medium of this embodiment (a non-transient computer-readable storage medium, for example, storage medium 308) may store the pipeline deterioration prediction program 75 of this embodiment.

[0269] In a modified version of this embodiment, the important pipeline data extraction unit 17 may not include the second important pipeline data extraction unit 17b. The pipeline data merging unit 18 may combine the similar pipeline dataset and the first important pipeline dataset to create a learning pipeline dataset 48. The important pipeline data extraction step S6 may not include the second important pipeline data extraction step S21. The learning pipeline dataset creation step S7 may combine the similar pipeline dataset and the first important pipeline dataset to create a learning pipeline dataset 48.

[0270] This document describes the method for generating the pipeline deterioration prediction model 5, the pipeline deterioration prediction method, the program, and the effects of the pipeline deterioration prediction device 3b according to this embodiment.

[0271] In the pipeline deterioration prediction model 5 generation method of this embodiment, the multiple customer pipeline data further includes the general deterioration of multiple customer pipelines. The multiple reference pipeline data further includes the general deterioration of multiple reference pipelines. The step of creating a customer pipeline dataset (preprocessed customer pipeline dataset; in this embodiment, the third customer pipeline dataset 34) and a reference pipeline dataset (preprocessed customer pipeline dataset; in this embodiment, the third reference pipeline dataset 39) (pipeline data preprocessing step S3) involves inputting the general customer pipeline dataset (in this embodiment, the second customer pipeline dataset 33) into the general pipeline deterioration prediction model 76 generated using the general reference pipeline dataset (in this embodiment, the second reference pipeline dataset 37) to obtain multiple The process includes the steps of: calculating the general deterioration level of customer pipelines (customer pipeline general deterioration level calculation step S63); combining the general deterioration levels of multiple customer pipelines into a general customer pipeline dataset (customer pipeline general deterioration level combining step S64); inputting a general reference pipeline dataset into a general pipeline deterioration level prediction model 76 to calculate the general deterioration level of multiple reference pipelines (reference pipeline general deterioration level calculation step S66); and combining the general deterioration levels of multiple reference pipelines into a general reference pipeline dataset (reference pipeline general deterioration level combining step S67). The general customer pipeline dataset includes the pipeline length, burial period, map data, and leakage accident history of multiple customer pipelines, but does not include the general deterioration level of multiple customer pipelines. The general reference pipeline dataset includes the pipeline length, burial period, map data, and leakage accident history of multiple reference pipelines, but does not include the general deterioration level of multiple reference pipelines.

[0272] The training pipeline dataset 48 (see Figure 43) includes the general deterioration rate of customer pipelines and the general deterioration rate of reference pipelines. The general deterioration rate of customer pipelines approximates the deterioration rate of customer pipelines relatively well. The general deterioration rate of reference pipelines approximates the deterioration rate of reference pipelines relatively well. Therefore, a pipeline deterioration prediction model 5 is generated that enables more accurate prediction of the deterioration rate of customer pipelines.

[0273] The program for the first phase of this embodiment (pipe deterioration prediction model generation program 8) causes the processor 202 to execute each step of the method for generating the pipe deterioration prediction model of this embodiment.

[0274] The general-purpose deterioration rate of customer pipelines approximates the deterioration rate of customer pipelines relatively well. The general-purpose deterioration rate of reference pipelines approximates the deterioration rate of reference pipelines relatively well. Therefore, a pipeline deterioration prediction model 5 is generated that enables a more accurate prediction of the deterioration rate of customer pipelines.

[0275] The pipeline deterioration prediction method of this embodiment includes the steps of creating an input pipeline dataset 72 that includes the pipeline length, burial period, map data, and general deterioration level of multiple customer pipelines (input pipeline dataset creation step S44), and inputting the input pipeline dataset 72 into the pipeline deterioration prediction model 5 generated by the pipeline deterioration prediction model 5 generation method of this embodiment to calculate the deterioration level of multiple customer pipelines (customer pipeline deterioration calculation step S45). The step of creating the input pipeline dataset 72 includes the steps of inputting a general customer pipeline dataset (in this embodiment, a second customer pipeline dataset 33) into a general pipeline deterioration prediction model 76 to calculate the general deterioration level of multiple customer pipelines (customer pipeline general deterioration level calculation step S72), and combining the general deterioration levels of multiple customer pipelines into the general customer pipeline dataset (customer pipeline general deterioration level combination step S73).

[0276] The general-purpose deterioration rate of customer pipelines provides a relatively good approximation of the actual deterioration rate of those pipelines. Therefore, it becomes possible to predict the deterioration rate of customer pipelines more accurately.

[0277] The program for the second phase of this embodiment (pipe deterioration prediction program 75) causes the processor 302 to execute each step of the pipe deterioration prediction method of this embodiment.

[0278] The general-purpose deterioration rate of customer pipelines provides a relatively good approximation of the actual deterioration rate of those pipelines. Therefore, it becomes possible to predict the deterioration rate of customer pipelines more accurately.

[0279] The pipeline deterioration prediction device 3b of this embodiment comprises an input pipeline dataset creation unit 62 and a pipeline deterioration calculation unit. The input pipeline dataset creation unit 62 creates an input pipeline dataset 72 that includes the pipeline length, burial period, map data, and general deterioration level of multiple customer pipelines. The input pipeline dataset creation unit 62 inputs the general customer pipeline dataset (in this embodiment, the second customer pipeline dataset 33) into the general pipeline deterioration prediction model 76 to calculate the general deterioration level of multiple customer pipelines. The pipeline deterioration calculation unit inputs the input pipeline dataset 72 into the pipeline deterioration prediction model generated by the pipeline deterioration prediction model generation method of this embodiment to calculate the deterioration level of multiple customer pipelines and combines the general deterioration levels of the multiple customer pipelines into the general customer pipeline dataset.

[0280] The general-purpose deterioration rate of customer pipelines provides a relatively good approximation of the actual deterioration rate of those pipelines. Therefore, it becomes possible to predict the deterioration rate of customer pipelines more accurately.

[0281] Embodiments 1 and 2 disclosed herein should be considered in all respects to be illustrative and not restrictive. The scope of this disclosure is indicated by the claims rather than the foregoing description and is intended to include all modifications in the sense and scope equivalent to the claims. [Explanation of symbols]

[0282] 1 Pipeline deterioration prediction system, 2,2b Pipeline deterioration prediction model generation device, 3,3b Pipeline deterioration prediction device, 4 Communication network, 5,5a,5b Pipeline deterioration prediction model, 6a,6b Pipeline water leakage accident probability prediction model, 7a,7b,78a,78b Conversion unit, 8 Pipeline deterioration prediction model generation program, 9 General-purpose pipeline deterioration prediction model generation program, 10 Pipeline data set reception unit, 11 Pipe material specification unit, 12 Training data set creation unit, 13 Pipeline data preprocessing unit, 13a First pipeline data preprocessing unit, 13b Second pipeline data preprocessing unit, 13c Third pipeline data preprocessing unit, 15 Pipeline data integration unit, 16 Similar pipeline data extraction unit, 17 Important pipeline data extraction unit, 17a First important pipeline data extraction unit, 17b Second important pipeline data extraction unit, 18 20 Pipeline data merging unit, 20 Pipeline deterioration prediction model generation unit, 21 Pipeline water leakage accident probability prediction model learning unit, 22 Conversion unit generation unit, 23 Pipeline deterioration prediction model generation completion determination unit, 25, 70 Storage unit, 31 Original customer pipeline dataset, 31a, 31b, 35a, 35b, 38b, 48a Attribute data, 31m, 35m, 38m, 48m Map data, 31p, 35p, 38p, 48p Location, 32 First customer pipeline dataset, 33 Second customer pipeline dataset, 34 Third customer pipeline dataset, 35 Original reference pipeline dataset, 36 First reference pipeline dataset, 37 Second reference pipeline dataset, 38 Integrated pipeline dataset, 39 Third reference pipeline dataset, 40 Map database, 41 Soil classification map, 42 Annual average temperature map, 43 Annual average precipitation map, 48 50. Learning pipeline dataset, 50. Neural network model, 53. Leakage accident prediction probability table, 60. Pipeline dataset reception unit, 61. Pipe material specification unit, 62. Input pipeline dataset creation unit, 63. First customer pipeline data processing unit, 64. Second customer pipeline data processing unit, 66. Model selection unit, 67. Pipeline deterioration calculation unit, 68. Pipeline deterioration calculation completion determination unit, 69. Pipeline deterioration prediction result creation unit, 72. Input pipeline dataset, 74. Pipeline deterioration prediction result, 74a. Pipeline deterioration prediction table, 74b. Pipeline deterioration prediction map, 75. Pipeline deterioration prediction program, 76, 76a, 76b. General-purpose pipeline deterioration prediction model, 77a,77b General-purpose pipeline leak accident probability prediction model, 81 General-purpose model selection unit, 83 General-purpose pipeline deterioration calculation unit, 84,89 General-purpose pipeline deterioration coupling unit, 85 Third customer pipeline data processing unit, 86 General-purpose model selection unit, 88 General-purpose pipeline deterioration calculation unit, 201,301 Input device, 202,302 Processor, 203,303 Memory, 204,304 Display, 206,306 Network controller, 207,307 Storage media drive, 208,308 Storage media, 210,310 Storage.

Claims

1. The process includes the steps of creating a customer pipeline dataset and a reference pipeline dataset, wherein the customer pipeline dataset includes multiple customer pipeline data, each including the pipeline length, burial period, map data, and leakage accident history of the multiple customer pipelines, and the reference pipeline dataset includes multiple reference pipeline data, each including the pipeline length, burial period, map data, and leakage accident history of the multiple reference pipelines. The process includes the step of integrating the customer pipeline dataset and the reference pipeline dataset to create an integrated pipeline dataset consisting of multiple integrated pipeline data sets, wherein the multiple integrated pipeline data sets include the multiple customer pipeline data sets and the multiple reference pipeline data sets. The steps include: extracting integrated pipeline data from the integrated pipeline dataset that is similar to the customer pipeline dataset as similar pipeline data, and creating a similar pipeline dataset composed of the similar pipeline data; The steps include: extracting customer pipeline data that is not similar pipeline data from the integrated pipeline dataset as first important pipeline data, and creating a first important pipeline dataset composed of the first important pipeline data; The method comprises the step of creating a learning pipeline dataset, wherein the step of creating the learning pipeline dataset includes combining the similar pipeline dataset and the first important pipeline dataset. The method includes the step of generating a pipeline deterioration prediction model using the aforementioned training pipeline dataset. A method for generating a pipeline deterioration prediction model, wherein the number of training pipeline data points included in the training pipeline dataset is greater than the number of customer pipeline data points.

2. The further step involves extracting integrated pipeline data similar to the first important pipeline data set from the integrated pipeline data set as second important pipeline data, and creating a second important pipeline data set composed of the second important pipeline data. The method for generating a pipeline deterioration prediction model according to claim 1, wherein the step of creating the learning pipeline dataset includes combining the similar pipeline dataset, the first important pipeline dataset, and the second important pipeline dataset.

3. The aforementioned customer pipeline data further includes the general deterioration status of the aforementioned customer pipelines, The aforementioned multiple reference pipeline data further includes the general deterioration status of the aforementioned multiple reference pipelines, The steps of creating the customer pipeline dataset and the reference pipeline dataset include: inputting the general-purpose customer pipeline dataset into a general-purpose pipeline deterioration prediction model generated using the general-purpose reference pipeline dataset to calculate the general deterioration of the plurality of customer pipelines; combining the general deterioration of the plurality of customer pipelines into the general-purpose customer pipeline dataset; inputting the general-purpose reference pipeline dataset into the general-purpose pipeline deterioration prediction model to calculate the general deterioration of the plurality of reference pipelines; and combining the general deterioration of the plurality of reference pipelines into the general-purpose reference pipeline dataset. The aforementioned general-purpose customer pipeline dataset includes the pipeline length, the buried period, the map data, and the leakage accident history of the plurality of customer pipelines, but does not include the general-purpose deterioration level of the plurality of customer pipelines. A method for generating a pipeline deterioration prediction model according to claim 1 or claim 2, wherein the general-purpose reference pipeline dataset includes the pipeline length, burial period, map data, and leakage accident history of the plurality of reference pipelines, and does not include the general deterioration degree of the plurality of reference pipelines.

4. A program that causes a processor to perform each step of the method for generating the pipeline deterioration prediction model according to claim 1 or claim 2.

5. A program that causes a processor to execute each step of the method for generating the pipeline deterioration prediction model described in claim 3.

6. The steps include creating an input pipeline dataset that includes pipeline length, burial period, and map data for multiple customer pipelines, and A method for predicting pipeline deterioration, comprising the steps of inputting the input pipeline dataset into the pipeline deterioration prediction model generated by the method for generating the pipeline deterioration prediction model according to claim 1 or claim 2, and calculating the deterioration of the plurality of customer pipelines.

7. The steps include creating an input pipeline dataset that includes pipeline length, burial period, map data, and general deterioration status for multiple customer pipelines, and The method for generating the pipeline deterioration prediction model described in claim 3 comprises the step of inputting the input pipeline dataset into the pipeline deterioration prediction model generated by the method for generating the pipeline deterioration prediction model described in claim 3, and calculating the deterioration of the plurality of customer pipelines. A pipeline deterioration prediction method comprising the steps of: creating the input pipeline dataset, inputting the general-purpose customer pipeline dataset into the general-purpose pipeline deterioration prediction model to calculate the general deterioration of the plurality of customer pipelines; and combining the general deterioration of the plurality of customer pipelines with the general-purpose customer pipeline dataset.

8. A program that causes a processor to execute each step of the pipeline deterioration prediction method described in claim 6.

9. A program that causes a processor to perform each step of the pipeline deterioration prediction method described in claim 7.

10. An input pipeline dataset creation unit creates an input pipeline dataset that includes pipeline length, burial period, and map data for multiple customer pipelines. A pipeline deterioration prediction device comprising: a pipeline deterioration calculation unit that inputs the input pipeline dataset into the pipeline deterioration prediction model generated by the pipeline deterioration prediction model generation method described in claim 1 or claim 2, and calculates the deterioration of the plurality of customer pipelines.

11. An input pipeline dataset creation unit creates an input pipeline dataset that includes the pipeline length, buried period, map data, and general deterioration level of multiple customer pipelines. The system comprises a pipeline deterioration calculation unit that inputs the input pipeline dataset into the pipeline deterioration prediction model generated by the pipeline deterioration prediction model generation method described in claim 3, and calculates the deterioration of the plurality of customer pipelines, The input pipeline dataset creation unit inputs the general-purpose customer pipeline dataset into the general-purpose pipeline deterioration prediction model to calculate the general-purpose deterioration of the multiple customer pipelines, and combines the general-purpose deterioration of the multiple customer pipelines into the general-purpose customer pipeline dataset, thereby forming a pipeline deterioration prediction device.

Images