Pipeline deterioration prediction system, pipeline deterioration prediction method, and pipeline deterioration prediction program
The pipeline deterioration prediction system uses machine learning to generate models from pipeline attributes and damage data, facilitating effective maintenance by predicting deterioration by location, state, and cause, thus enhancing maintenance efficiency.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- KUBOTA CORP
- Filing Date
- 2024-12-06
- Publication Date
- 2026-06-18
AI Technical Summary
Existing technologies fail to predict pipeline deterioration efficiently, which hinders effective maintenance planning and resource allocation.
A pipeline deterioration prediction system using machine learning to generate models based on pipeline attributes, burial environment, and damage data, enabling prediction of deterioration by location, state, and cause, with weighted calculations for comprehensive evaluation.
Enables efficient maintenance planning by predicting pipeline deterioration accurately, allowing for targeted and timely maintenance actions.
Smart Images
Figure 2026098975000001_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a pipeline aging degree prediction system, a pipeline aging degree prediction method, and a pipeline aging degree prediction program.
Background Art
[0002] In Patent Document 1, a learning device is proposed that improves a technique for predicting damage to facilities caused by disasters in response to the problem that damage occurring during disasters cannot be quantitatively evaluated and the deterioration of facilities cannot be reflected in the prediction of damage.
[0003] The learning device includes a first model generation unit that generates a first model for calculating the degree of deterioration of the facility by machine learning using deterioration data including information on the facility, environmental conditions of the facility, and the degree of deterioration of the facility as training data, and a second model generation unit that generates a second model for calculating the degree of damage to the facility by machine learning using damage data including disaster conditions for the facility, the degree of damage caused by the disaster, and the degree of deterioration calculated by the first model.
[0004] And an estimation device capable of estimating damage is realized by including an estimation unit that applies target data including information on the facility of the target facility, environmental conditions of the target facility, and disaster conditions of the target facility to the first model and the second model and estimates the degree of damage to the target facility.
Prior Art Documents
Patent Documents
[0005]
Patent Document 1
Summary of the Invention
Problems to be Solved by the Invention
[0006] However, the learning device disclosed in Patent Document 1 was intended to predict damage that would occur during disasters such as earthquakes, and was not intended to contribute to the efficient pipeline maintenance work required on a daily basis.
[0007] The objective of the present invention is to provide a pipeline deterioration prediction system, a pipeline deterioration prediction method, and a pipeline deterioration prediction program that can predict the degree of pipeline deterioration in advance and enable efficient maintenance work. [Means for solving the problem]
[0008] To achieve the above objective, the first characteristic configuration of the pipeline deterioration prediction system according to the present invention is a pipeline deterioration prediction system that calculates a predicted value of the degree of deterioration of a pipeline buried underground, comprising: a learning data storage circuit that stores as learning data pipeline attribute data indicating pipeline attributes including the type of pipeline and the number of years buried; pipeline burial environment data indicating the burial environment including the soil in which the pipeline is buried; and damage type data indicating the damage type including the location and state of damage of the pipeline; and a system that reads the learning data stored in the learning data storage circuit, performs machine learning, and predicts the degree of deterioration The system comprises a machine learning processor that generates a model; a pipeline data storage circuit that stores pipeline attribute data, including the pipe type and buried years of the pipeline to be evaluated, and pipeline burial environment data, including the soil in which the pipeline to be evaluated is buried, as pipeline data to be evaluated; and a deterioration prediction value calculation processor that inputs the pipeline data read from the pipeline data storage circuit into the deterioration prediction model generated by the machine learning processor to calculate a predicted deterioration value for the pipeline to be evaluated.
[0009] The deterioration prediction model generated by the machine learning processor is obtained by machine learning using pipeline attribute data, pipeline burial environment data, and damage type data stored in the learning data storage circuit as training data. The pipeline attribute data indicates the attributes of the pipeline, including the pipe type and years buried, i.e., what kind of material and structure the pipeline is made of. The pipeline burial environment data indicates the burial environment, including the soil in which the pipeline is buried, i.e., the environment affecting the pipeline, such as corrosiveness and water pressure. The damage type data includes the location and state of damage. Therefore, an appropriate deterioration prediction model is generated by machine learning based on these datasets, which is performed by the machine learning processor.
[0010] The second characteristic configuration is that, in addition to the first characteristic configuration described above, the machine learning processor generates a deterioration prediction model for each damaged location based on damaged location learning data obtained by classifying the learning data according to the damaged location, and also generates a deterioration prediction model for each damaged state based on damaged state learning data obtained by classifying the learning data according to the damaged state.
[0011] Based on the learning data for damaged locations, a model for predicting deterioration by damaged location is generated, and based on the learning data for damaged conditions, a model for predicting deterioration by damaged condition is generated. Based on these models, predicted values focusing on damaged locations and predicted values focusing on damaged conditions are obtained as predicted values for the deterioration of the pipeline under evaluation. Therefore, effective maintenance parts can be selected for damaged locations and damaged conditions with high predicted values, and a maintenance work plan can be quickly formulated.
[0012] The third characteristic configuration is that, in addition to the first characteristic configuration described above, the damage pattern data stored in the learning data storage circuit further includes the cause of damage as the damage pattern, and the machine learning processor generates a deterioration prediction model for each damage location based on the damage location learning data obtained by classifying the learning data for each damage location, generates a deterioration prediction model for each damage state based on the damage state learning data obtained by classifying the learning data for each damage state, and generates a deterioration prediction model for each cause of damage based on the cause of damage learning data obtained by classifying the learning data for each cause of damage.
[0013] Based on failure cause learning data, a deterioration prediction model categorized by failure cause is obtained, and a predicted value focusing on the failure cause is obtained as the predicted deterioration value of the pipeline under evaluation. Therefore, effective maintenance parts can be selected for failure causes with high predicted values, and a maintenance work plan can be quickly formulated.
[0014] The fourth characteristic configuration is that, in addition to the third characteristic configuration described above, the deterioration prediction value calculation processor calculates and outputs an overall evaluation value for the pipeline to be evaluated by weighting the deterioration prediction values calculated for each of the deterioration prediction models for each of the damaged location, the deterioration prediction model for each of the damaged state, and the deterioration prediction model for each of the damaged cause, based on their respective correlations.
[0015] An efficient maintenance work plan can be formulated based on an overall evaluation value calculated from the predicted deterioration values calculated for each deterioration prediction model. If there is a correlation between the deterioration values calculated from each deterioration prediction model, simply performing calculations such as addition may result in a biased overall evaluation value. However, by performing weighted calculations based on the respective correlations, an unbiased and objective overall evaluation value can be obtained.
[0016] The fifth characteristic configuration is that, in addition to the third characteristic configuration described above, the machine learning processor generates a comprehensive deterioration learning data by adding the predicted values for each deterioration level obtained by applying the learning data to the deterioration level prediction model by location of damage, the deterioration level prediction model by state of damage, and the deterioration level prediction model by cause of damage, to the learning data, and then generates a comprehensive deterioration level prediction model by machine learning the comprehensive deterioration learning data.
[0017] The overall deterioration prediction model generated by the machine learning processor provides an overall evaluation value for the pipeline under evaluation. The overall deterioration prediction model is generated by the machine learning processor applying the training data to each of the deterioration prediction models: a deterioration prediction model by location of damage, a deterioration prediction model by type of damage, and a deterioration prediction model by cause of damage. This training data is then used to generate the overall deterioration training data, which is added to the training data.
[0018] The sixth feature configuration is that, in addition to any of the first to fifth feature configurations described above, the machine learning processor generates a decision tree model obtained by ensemble learning the training data as the deterioration prediction model.
[0019] As a machine learning algorithm executed by a machine learning processor, a decision tree model generation algorithm obtained by ensemble learning of training data can be suitably utilized.
[0020] The first characteristic configuration of the pipeline deterioration prediction method according to the present invention is a pipeline deterioration prediction method for calculating a predicted value of the degree of deterioration of a pipeline buried underground, comprising: a learning data storage step in which pipeline attribute data indicating pipeline attributes including the type of pipeline and the number of years buried, pipeline burial environment data indicating the burial environment including the soil in which the pipeline is buried, and damage type data indicating the damage type including the location and state of damage of the pipeline are stored as learning data in a learning data storage circuit; and a deterioration prediction model step in which the learning data stored in the learning data storage circuit is read out and machine learning is performed by a machine learning processor to generate a deterioration prediction model. The system comprises: a pipeline generation step; a pipeline data storage step in which pipeline attribute data, including the type of pipe and the number of years buried, and pipeline burial environment data, including the soil in which the pipeline is buried, are stored as pipeline data in a pipeline data storage circuit; and a deterioration prediction value calculation step in which the pipeline data read from the pipeline data storage circuit is input into the deterioration prediction model generated by the machine learning processor, and the deterioration prediction value of the pipeline is calculated.
[0021] The second characteristic configuration is that, in addition to the first characteristic configuration described above, the deterioration prediction model generation step involves the machine learning processor generating a deterioration prediction model for each damaged location based on damaged location learning data obtained by classifying the learning data according to the damaged location, and generating a deterioration prediction model for each damaged state based on damaged state learning data obtained by classifying the learning data according to the damaged state.
[0022] The third characteristic configuration, in addition to the first characteristic configuration described above, is that the damage mode data stored in the learning data storage circuit further includes the cause of damage as the damage mode, and in the aging degree prediction model generation step, the machine learning processor generates an aging degree prediction model for each damaged part based on the damaged part learning data obtained by classifying the learning data for each damaged part, generates an aging degree prediction model for each damage state based on the damage state learning data obtained by classifying the learning data for each damage state, and generates an aging degree prediction model for each cause of damage based on the cause of damage learning data obtained by classifying the learning data for each cause of damage.
[0023] The fourth characteristic configuration, in addition to the third characteristic configuration described above, is that in the aging degree prediction value calculation step, the aging degree prediction value calculation processor weights and calculates the predicted values of the aging degree calculated for each of the aging degree prediction models for each damaged part, the aging degree prediction models for each damage state, and the aging degree prediction models for each cause of damage based on their respective correlation relationships, and calculates and outputs the comprehensive evaluation value of the pipeline to be evaluated.
[0024] The fifth characteristic configuration, in addition to the third characteristic configuration described above, is that in the aging degree prediction model generation step, the machine learning processor generates comprehensive aging degree learning data by adding the predicted values of the aging degree obtained by applying the learning data to each of the aging degree prediction models for each damaged part, the aging degree prediction models for each damage state, and the aging degree prediction models for each cause of damage to the learning data, and generates an aging degree prediction model by performing machine learning on the comprehensive aging degree learning data.
[0025] The characteristic configuration of the pipeline aging degree prediction program according to the present invention is that a computer is used to store, as learning data, pipeline attribute data indicating pipeline attributes including the type of pipeline and the number of years of burial of the pipeline buried underground, pipeline buried environment data indicating the buried environment including the buried soil of the pipeline, and damage mode data indicating the damage mode including the damage location and damage state of the pipeline, in a learning data storage circuit by a learning data storage processing unit, read out the learning data stored in the learning data storage circuit and perform machine learning by a machine learning processing unit to generate an aging degree prediction model, store, as evaluation target pipeline data, evaluation target pipeline attribute data indicating pipeline attributes including the type of pipeline and the number of years of burial of the evaluation target pipeline and evaluation target pipeline buried environment data indicating the buried environment including the buried soil of the evaluation target pipeline, in an evaluation target pipeline data storage circuit by an evaluation target pipeline data storage processing step, input the evaluation target pipeline data read out from the evaluation target pipeline data storage circuit into the aging degree prediction model generated by the machine learning processor, and calculate a predicted value of the aging degree of the evaluation target pipeline by an aging degree predicted value calculation processing unit.
Advantages of the Invention
[0026] As described above, according to the present invention, it has become possible to provide a pipeline aging degree prediction system, a pipeline aging degree prediction method, and a pipeline aging degree prediction program that can predict the aging degree of a pipeline in advance and enable efficient maintenance work.
Brief Description of the Drawings
[0027] [Figure 1] Explanation diagram of the configuration of a computer in which a pipeline aging degree prediction system is constructed [Figure 2] Explanation diagram of the functional blocks of a pipeline aging degree prediction system [Figure 3] Explanation diagram of learning data for generating an aging degree prediction model [Figure 4] (a) is an explanatory diagram of the damage location among the damage modes, (b) is an explanatory diagram of the damage state among the damage modes, and (c) is an explanatory diagram of the damage cause among the damage modes [Figure 5] Flowchart showing the procedure of a pipeline aging degree prediction method [Figure 6] Diagram illustrating the training data used to generate the comprehensive deterioration prediction model. [Figure 7] Diagram illustrating the decision tree model [Figure 8] Diagram illustrating the method for converting leakage probability to predicted leakage occurrences [per leak / year / km]. [Modes for carrying out the invention]
[0028] The pipeline deterioration prediction system, pipeline deterioration prediction method, and pipeline deterioration prediction program according to the present invention will be described below with reference to the drawings. [Pipeline Deterioration Prediction System] As shown in Figure 1, the pipeline deterioration prediction system 10 consists of a computer in which a motherboard 10A equipped with a processor such as a CPU and integrated circuits such as a chipset, a memory board 10B equipped with semiconductor memory such as ROM and RAM, and an I / O board 10C equipped with an input / output interface are connected by a communication bus 10D.
[0029] The computer is connected to storage devices 10E such as hard disk drives and flash memory devices, portable memory devices 10F such as USB memory, input devices 10G such as keyboards and mice, and output devices 10H such as printers and LCD displays, and is configured to connect to a cloud computer via a communication interface 10I such as Wi-Fi.
[0030] The pipeline deterioration prediction system 10 is realized when a pipeline deterioration prediction program, provided via a recording medium such as a storage device or portable memory device, is installed in the memory on the memory board, and the pipeline deterioration prediction program is executed by the CPU. Alternatively, the pipeline deterioration prediction program may be installed in the memory on the memory board from a cloud computer.
[0031] Furthermore, a pipeline deterioration prediction system 10 that links the computer and the cloud computer may be realized by executing an API (Application Programming Interface) corresponding to a pipeline deterioration prediction program executed on the cloud computer on the computer's CPU. Below, a pipeline deterioration prediction system 10 in which the pipeline deterioration prediction program is executed by the CPU of the computer will be described.
[0032] Figure 2 shows the functional block configuration of the pipeline deterioration prediction system 10. The pipeline deterioration prediction system 10 targets resin pipes and steel pipes as pipelines and includes a machine learning processor 11, a deterioration prediction value calculation processor 12, a learning data storage circuit 13, a deterioration prediction model storage circuit 14, an evaluation target pipeline data storage circuit 15, and a deterioration prediction value storage circuit 16. The machine learning processor 11 and the deterioration prediction value calculation processor 12 are implemented by processors provided on the motherboard 10A. The learning data storage circuit 13, the deterioration prediction model storage circuit 14, the evaluation target pipeline data storage circuit 15, and the deterioration prediction value storage circuit 16 are composed of semiconductor memory and storage devices 10E provided on the memory board 10B, and can also be composed of semiconductor memory and storage devices provided on a cloud computer.
[0033] The pipeline deterioration prediction system 10 includes a machine learning unit comprising: a learning data storage circuit 13 that stores pipeline attribute data, pipeline burial environment data, and damage type data indicating the location, state, and cause of damage to the pipeline as learning data; a machine learning processor 11 that reads the learning data stored in the learning data storage circuit 13, performs machine learning on it, and generates a deterioration prediction model; and a deterioration prediction model storage circuit 14 that stores the generated deterioration prediction model. In addition, the system includes a deterioration prediction unit that calculates a deterioration prediction value based on the deterioration prediction model.
[0034] The deterioration prediction unit includes a target pipeline data storage circuit 15 that stores target pipeline attribute data indicating the pipeline attributes of the target pipeline and target pipeline buried environment data indicating the buried environment of the target pipeline as target pipeline data, and a deterioration prediction value calculation processor 12 that inputs the target pipeline data read from the target pipeline data storage circuit 15 into a deterioration prediction model stored in a deterioration prediction model storage circuit 14 to calculate a predicted deterioration value for the target pipeline.
[0035] Figure 3 shows the structure of the training data stored in the training data storage circuit 13. This training data is the training data input to the machine learning processor 11 that generates the deterioration prediction model. It is associated with pipeline IDs that identify individual pipelines that make up the existing pipeline network buried underground, and includes leakage accident history data (0 for no accident history, 1 for accident history), pipeline attribute data, pipeline burial environment data, and damage type data as explanatory variables. The evaluation data becomes the estimated deterioration value as the objective variable.
[0036] Pipeline attribute data is data that shows the specific configuration of the pipeline, and includes pipe type, diameter, pipe length, and year of installation. In this example, the pipe type is specified as either a resin pipe or a steel pipe, and if it is a resin pipe, the material (such as rigid polyvinyl chloride pipe or polyethylene pipe) is further specified.
[0037] Pipeline laying environment data includes data on the environment in which the pipeline is laid, and includes data on buried soil, topography, distance from water bodies, weather, and infrastructure. Buried soil data indicates the degree of impact on pipe corrosion and ground strength, and includes data on soil classification, surface geology, and whether or not land improvement has been performed. Topography data indicates the impact on drainage, ground strength, and water pressure inside the pipe, and includes data on topographic classification, microtopographic classification, and elevation. Distance from water bodies data indicates the impact on moisture in the surrounding soil, such as rivers, the degree of corrosiveness, and ground strength, and includes data on distance from rivers and distance from coastlines. Weather data indicates the impact on moisture due to soil moisture and ground temperature, and the degree of corrosiveness, and includes data on temperature and precipitation. Infrastructure data indicates the impact on water pressure fluctuations caused by sudden water usage and the magnitude of traffic volume, and includes data on population density, distance from road centerlines, and distance from railways. These are examples of pipeline burial environment data; the data on the soil, topography, distance from water bodies, weather, and infrastructure of the pipeline may consist of or include other data.
[0038] Damage pattern data shows the types of pipeline damage that have occurred in the past, and includes the location of the damage, the state of the damage, and the cause of the damage. Figure 4 shows a specific example of damage pattern data. Damage locations include data showing damaged locations such as water distribution joints, water distribution pipes, valves, fire hydrants, water intake joints, and water supply pipes. Water distribution refers to water distribution pipes, and water supply refers to water supply pipes that connect water distribution pipes to consumers. Damage states include data showing damage conditions such as detachment, cracks, breakage, and perforation. Causes of damage include data on factors that cause damage, such as corrosion, electrolytic corrosion, material deterioration, water hammer pressure, vibration load, and ground subsidence.
[0039] The machine learning processor 11 generates a decision tree model that serves as a deterioration prediction model by performing ensemble learning using the decision tree algorithm based on the training data described above.
[0040] A decision tree model is a knowledge representation that describes the procedure for separating features contained in training data using a branched tree structure. It determines the properties and classification categories of a pipeline being evaluated from multiple attribute data that characterize the pipeline. Attributes refer to the classification items that constitute the leakage accident history, pipeline attributes, pipeline burial environment, and damage type associated with the pipeline ID mentioned above. In the decision tree, branching progresses according to the answers to attribute-related questions, and ultimately, information estimating the degree of pipeline deterioration, specifically the probability of leakage, is obtained.
[0041] Figure 7 illustrates a decision tree model. For 100 pipelines and 4 leak incidents, considering the pipeline attributes, pipeline burial environment, and damage type, if the pipeline is buried for 40 years or more, the branching point is to the right, resulting in a pattern with a high probability of leakage: 20 pipelines, 3 leak incidents, and a leakage probability of 3 / 20. Furthermore, if the surface geology is sandy or gravelly, the branching point is to the right again, classifying the pipelines as a group with a high probability of leakage: 10 pipelines, 3 leaks, and a leakage probability of 3 / 10. If the pipeline is buried for less than 40 years, the branching point is to the left, resulting in a pattern with a low probability of leakage: 80 pipelines, 1 leak incident, and a leakage probability of 1 / 80. Furthermore, if the surface geology is not sandy or gravelly, the branching point is to the left again, classifying the pipelines as a group with a low probability of leakage: 10 pipelines, 0 leaks, and a leakage probability of 0. Here, leakage probability refers to the probability of an incident occurring, which indicates a predicted value of the degree of deterioration.
[0042] A decision tree model is generated by using the dataset shown in Figure 3 as training data and having the machine learning processor 11 perform machine learning based on the following decision tree algorithm. First, the training dataset is classified into subsets using pre-defined appropriate items from among pipeline attributes, pipeline burial environment, and damage type. If there are no attributes that can be used for classification, the algorithm terminates without completing the classification knowledge. For each subset, this algorithm is applied, the difference between the obtained predicted degree of deterioration and the actual degree of deterioration is calculated, and the process of creating a new decision tree (classification rule) to reduce the difference is repeated. At this time, instead of creating a single decision tree, multiple decision trees are created and ensemble learning is used, where the whole set of these is used as a single piece of knowledge, thereby generating a more appropriate decision tree.
[0043] For ensemble learning, parallel ensemble methods such as XGBoost, LightGBM, and Random Forest are suitable. For example, in the Random Forest method, multiple training datasets are created by randomly extracting data from a dataset like the one shown in Figure 3. A corresponding decision tree is created using each training dataset, and the overall result, i.e., damage estimation information, is obtained by using the average of the outputs of the multiple decision trees.
[0044] The LightGBM method has the following features, which enable it to generate models quickly: (1) It does not compute any further for elements (leaves) that no longer require branching (Leaf-wise tree growth); (2) When branching the decision tree, it creates a histogram and branches the values together instead of looking at all the values (Histogram-based); (3) It reduces the amount of training data by reducing data with small errors and keeping only data with large errors in order to prioritize learning elements that have not been learned (Gradient-based One-Side Sampling); and (4) It reduces computational complexity by combining features that do not seem to cause problems when combined into one (Exclusive Feature Bundling).
[0045] The machine learning processor 11 generates a deterioration prediction model for each damaged location based on damaged location learning data, which is obtained by classifying the learning data by damaged location, and also generates a deterioration prediction model for each damaged state based on damaged state learning data, which is obtained by classifying the learning data by damaged state.
[0046] The machine learning processor 11 generates a deterioration prediction model for each damaged location based on damaged location learning data obtained by classifying the learning data by damaged location, generates a deterioration prediction model for each damaged state based on damaged state learning data obtained by classifying the learning data by damaged state, and further generates a deterioration prediction model for each cause of damage based on damaged cause learning data obtained by classifying the learning data by cause of damage. For example, deterioration prediction models for each damaged location include "water distribution joint deterioration prediction model" and "water distribution direct section deterioration prediction model," deterioration prediction models for each damaged state include "crack deterioration prediction model" and "breakage deterioration prediction model," and deterioration prediction models for each cause of damage include "corrosion deterioration prediction model" and "material deterioration deterioration prediction model."
[0047] Based on the deterioration prediction models for each type of damage—location, condition, and cause—predictions of the deterioration of the pipeline under evaluation are obtained, focusing on the location of the damage, the condition of the damage, and the cause of the damage. Therefore, effective maintenance parts can be selected for each of the high predicted locations, conditions, and causes of damage, enabling the rapid formulation of maintenance work plans.
[0048] Figure 6 shows examples of deterioration prediction values obtained based on the deterioration prediction model by damage location, deterioration prediction model by damage condition, and deterioration prediction model by cause of damage. For example, for pipeline P1, the deterioration prediction value for each model is obtained as follows: 0.8 for the "water distribution joint deterioration prediction model", 0 for the "water distribution joint deterioration prediction model", and 0.3 for the "crack deterioration prediction model".
[0049] For each model, the predicted degree of deterioration provides predicted values for each individual type of damage to pipeline P1, i.e., the location of the damage, the state of the damage, and the cause of the damage. To evaluate pipeline P1 as a whole, it is preferable to obtain an overall evaluation value that evaluates the entire pipeline from the individual predicted degree of deterioration values. Therefore, it is preferable that the deterioration prediction value calculation processor 12 calculates an overall evaluation value for the pipeline to be evaluated by weighting the individual deterioration prediction values calculated for each of the deterioration prediction models, damage condition-based and damage cause-based, based on their respective correlations, and then stores this overall evaluation value in the deterioration prediction value storage circuit 16.
[0050] For example, when damage occurs at the "direct connection point of the water supply," there is a high correlation with "cracks" but a low correlation with "detachment." Similarly, when damage occurs at the "water supply joint," there is a high correlation with "detachment" but a low correlation with "cracks." Therefore, the system can be configured to set a small weighting coefficient for each predicted degree of deterioration when the correlation is high, and a large weighting coefficient when the correlation is low, and then calculate an overall evaluation value by adding the product of each predicted degree of deterioration and the weighting coefficient. The weighting coefficients can be set to appropriate values in advance based on past water leakage accidents.
[0051] An efficient maintenance work plan can be formulated based on an overall evaluation value calculated from the predicted deterioration values calculated for each deterioration prediction model. If there is a correlation between the deterioration values calculated from each deterioration prediction model, simply performing calculations such as addition may result in a biased overall evaluation value. However, by performing weighted calculations based on the respective correlations, an unbiased and objective overall evaluation value can be obtained.
[0052] Instead of the aging degree prediction value calculation processor 12 calculating the overall evaluation value, the machine learning processor 11 may be configured to generate overall aging degree learning data by applying the learning data to each of the aging degree prediction models for each damaged location, aging degree prediction model for each damaged state, and aging degree prediction model for each damaged cause, and then generate an overall aging degree prediction model by machine learning the overall aging degree learning data using the machine learning algorithm described above.
[0053] Figure 5 shows a series of steps for the pipeline deterioration prediction method performed by the pipeline deterioration prediction system 10. Specifically, the pipeline deterioration prediction method consists of a learning data storage step (S1), a deterioration prediction model generation step (S2-S4) in which the learning data stored in the learning data storage circuit 13 is read out and machine learning is performed by the machine learning processor 11 to generate a deterioration prediction model, a comprehensive evaluation learning data storage step (S5), a comprehensive deterioration evaluation model generation step (S6-S7), a pipeline data storage step (S9), a deterioration prediction value calculation step (S10), and a maintenance plan generation step (S11).
[0054] The learning data storage step (S1) is a step in which pipeline attribute data, pipeline burial environment data, and damage type data indicating the type of damage, including the location of the damage, the state of the damage, and the cause of the damage, are stored as learning data in the learning data storage circuit 13.
[0055] The aging prediction model generation step (S2-S4) involves reading the training data stored in the training data storage circuit 13, performing machine learning on it using the machine learning processor 11, and generating an aging prediction model.
[0056] The deterioration prediction model generation steps (S2-S4) involve generating deterioration prediction models by damaged location based on damaged location learning data obtained by classifying the learning data by damaged location, generating deterioration prediction models by damaged state based on damaged state learning data obtained by classifying the learning data by damaged state, and generating deterioration prediction models by damaged cause based on damaged cause learning data obtained by classifying the learning data by damaged cause.
[0057] The comprehensive evaluation learning data storage step (S5) is a step in which the learning data is used to generate comprehensive deterioration learning data by adding the predicted values for each deterioration level obtained by applying the learning data to the deterioration level prediction model by location of damage, deterioration level prediction model by state of damage, and deterioration level prediction model by cause of damage, and storing this data in the learning data storage circuit 13.
[0058] The comprehensive deterioration assessment model generation step (S6-S7) is the step of generating a comprehensive deterioration assessment model based on comprehensive deterioration learning data and storing it in the deterioration prediction model memory circuit 14.
[0059] The evaluation target pipeline data storage step (S9) is a step of storing the evaluation target pipeline data in the evaluation target pipeline data storage circuit 15.
[0060] The deterioration prediction value calculation step (S10) is a step in which the deterioration prediction value calculation processor 12 reads the pipeline data to be evaluated from the pipeline data storage circuit 15 and applies it to the deterioration prediction model by damage location, deterioration prediction model by damage state, and deterioration prediction model by cause of damage to calculate the deterioration prediction values by damage location, deterioration prediction model by damage state, and deterioration prediction model by cause of damage. It also applies the pipeline data to be evaluated, to which the deterioration prediction values by damage location, deterioration prediction values by damage state, and deterioration prediction values by cause of damage have been added, to the overall deterioration prediction model to calculate the overall deterioration prediction value.
[0061] The maintenance plan generation step (S11) is a step in which a maintenance plan is created based on the predicted deterioration values for each damaged location, each damaged condition, each damaged cause, and the overall predicted deterioration value, which were calculated in the predicted deterioration value calculation step (S10). The maintenance plan is formulated prioritizing pipelines with high overall predicted deterioration values, and specific maintenance work such as parts replacement is formulated based on the predicted deterioration values for each damaged location, each damaged condition, and each damaged cause, which were calculated for the same pipeline. Maintenance includes pipeline renewal, leak repair, and pipeline rehabilitation.
[0062] Each deterioration prediction value may be a leakage probability expressed as a number between "0" and "1" calculated by the deterioration prediction value calculation processor 12, or it may be the unit length of the pipeline and the predicted number of leakage occurrences per year [incidents / year / km]. Figure 8 illustrates a method for converting leakage probability to predicted leakage occurrences [incidents / year / km] using rigid polyvinyl chloride pipes as an example. In this example, training data and validation data are aggregated to ensure sufficient data for both pipe length and leakage occurrences. Since the leakage probabilities calculated by the learning model of the present invention are concentrated between 0 and 0.1, the graph above shows the aggregated results of pipe length transformed into a normal distribution using logarithmic transformation, with the horizontal axis representing Log. 10 (Probability of leakage), the vertical axis represents the pipeline length [km] of each pipeline and the actual number of leakage incidents [incidents / year], with the actual number of leakage incidents plotted. The horizontal axis of the graph below is Log 10 (Probability of water leakage), the vertical axis represents the predicted number of water leakage incidents [incidents / year / km], and the probability of water leakage is converted to the predicted number of water leakage incidents based on the following formula. y = A exp(Bx), where A and B are coefficients calculated for each model.
[0063] The pipeline deterioration prediction program, which is installed in the memory on the memory board, is preferably provided via a program storage medium such as a CD-ROM or a stick memory.
[0064] The pipeline deterioration prediction program includes a learning data storage processing unit that stores the following as learning data in a learning data storage circuit: pipeline attribute data indicating pipeline attributes including the type of pipe and the number of years buried in the underground pipeline; pipeline burial environment data indicating the burial environment including the soil in which the pipeline is buried; and damage type data indicating the damage type including the location and condition of damage in the pipeline; a machine learning processing unit that reads the learning data stored in the learning data storage circuit, performs machine learning, and generates a deterioration prediction model; and the type of pipe and the number of years buried of the pipeline to be evaluated. This pipeline deterioration prediction program functions as follows: an evaluation target pipeline data storage processing step that stores evaluation target pipeline attribute data, which indicates pipeline attributes including the number, and evaluation target pipeline burial environment data, which indicates the burial environment including the soil in which the evaluation target pipeline is buried, in an evaluation target pipeline data storage circuit as evaluation target pipeline data; and a deterioration prediction value calculation processing unit that inputs the evaluation target pipeline data read from the evaluation target pipeline data storage circuit into a deterioration prediction model generated by a machine learning processor to calculate a predicted deterioration value for the evaluation target pipeline.
[0065] In the embodiments described above, an example was explained in which the damage pattern data included the location of the damage, the state of the damage, and the cause of the damage. However, it is sufficient for the damage pattern data to include the location of the damage and the state of the damage, and the cause of the damage may be omitted.
[0066] In the embodiments described above, a pipeline deterioration prediction system 10 was described that targeted resin pipes and steel pipes as pipelines. However, the target pipelines are not limited to these, and ductile iron pipes and cast iron pipes may also be targeted. In this case, the specific items for each of the damage location, damage state, and damage cause can be appropriately selected as damage type data.
[0067] The embodiments described above represent one aspect of the present invention, and the technical scope of the present invention is not limited based on this description. It goes without saying that the design can be modified as appropriate within the scope in which the effects of the present invention are achieved. [Explanation of symbols]
[0068] 10: Pipeline Deterioration Prediction System 11: Machine Learning Processor 12: Deterioration degree prediction value calculation processor 13: Learning Data Storage Circuit 14: Deterioration Prediction Model Memory Circuit 15: Data storage circuit for pipelines to be evaluated 16: Deterioration degree prediction value storage unit
Claims
1. A pipeline deterioration prediction system that calculates a predicted value for the degree of deterioration of pipelines buried underground, A learning data storage circuit stores the following as learning data: pipeline attribute data indicating pipeline attributes including the type of pipe and the number of years buried of the pipeline; pipeline burial environment data indicating the burial environment including the soil in which the pipeline is buried; and damage type data indicating the damage type including the location and condition of the damage to the pipeline. A machine learning processor that reads the learning data stored in the learning data storage circuit, performs machine learning, and generates a deterioration prediction model. A pipeline data storage circuit stores the following as pipeline data: pipeline attribute data indicating pipeline attributes including the pipe type and years buried of the pipeline to be evaluated, and pipeline burial environment data indicating the burial environment including the soil in which the pipeline to be evaluated is buried. A deterioration prediction value calculation processor that inputs the pipeline data to be evaluated, read from the pipeline data storage circuit, into the deterioration prediction model generated by the machine learning processor, and calculates a predicted value for the deterioration of the pipeline to be evaluated. A pipeline deterioration prediction system equipped with this feature.
2. The pipeline deterioration prediction system according to claim 1, wherein the machine learning processor generates a deterioration prediction model for each damaged location based on damaged location learning data obtained by classifying the learning data according to each damaged location, and generates a deterioration prediction model for each damaged state based on damaged state learning data obtained by classifying the learning data according to each damaged state.
3. The damage pattern data stored in the learning data storage circuit further includes the cause of damage as the damage pattern, The pipeline deterioration prediction system according to claim 1, wherein the machine learning processor generates a deterioration prediction model for each damaged location based on damaged location learning data obtained by classifying the learning data according to each damaged location, generates a deterioration prediction model for each damaged state based on damaged state learning data obtained by classifying the learning data according to each damaged state, and generates a deterioration prediction model for each damaged cause based on damaged cause learning data obtained by classifying the learning data according to each damaged cause.
4. The pipeline deterioration prediction system according to claim 3, wherein the deterioration prediction value calculation processor calculates and outputs an overall evaluation value for the pipeline to be evaluated by weighting the deterioration prediction values calculated for each of the deterioration prediction models for each of the damaged locations, deterioration prediction models for each of the damaged conditions, and deterioration prediction models for each of the damaged cause, based on their respective correlations.
5. The pipeline deterioration prediction system according to claim 3, wherein the machine learning processor generates comprehensive deterioration learning data by adding the predicted values of deterioration obtained by applying the learning data to each of the deterioration prediction models by location of damage, deterioration prediction model by state of damage, and deterioration prediction model by cause of damage, to the learning data, and generates a comprehensive deterioration prediction model by performing machine learning on the comprehensive deterioration learning data.
6. The pipeline deterioration prediction system according to any one of claims 1 to 5, wherein the machine learning processor generates a decision tree model obtained by ensemble learning the training data as the deterioration prediction model.
7. A method for predicting the degree of deterioration of underground pipelines, which calculates a predicted value for the degree of deterioration of underground pipelines, A learning data storage step involves storing the following as learning data in a learning data storage circuit: pipeline attribute data indicating pipeline attributes including the type of pipe and the number of years buried; pipeline burial environment data indicating the burial environment including the soil in which the pipeline is buried; and damage type data indicating the damage type including the location and condition of the damage to the pipeline. A deterioration prediction model generation step involves reading the learning data stored in the learning data storage circuit, performing machine learning on it using a machine learning processor, and generating a deterioration prediction model; A pipeline data storage step involves storing the following in a pipeline data storage circuit as pipeline data to be evaluated: pipeline attribute data indicating pipeline attributes including the pipe type and years buried of the pipeline to be evaluated, and pipeline burial environment data indicating the burial environment including the soil in which the pipeline to be evaluated is buried; A deterioration prediction value calculation step involves inputting the pipeline data to be evaluated, read from the pipeline data storage circuit, into the deterioration prediction model generated by the machine learning processor, using a deterioration prediction value calculation processor, to calculate a predicted deterioration value for the pipeline to be evaluated. A pipeline deterioration prediction method that includes the following features.
8. The pipeline deterioration prediction method according to claim 7, wherein the deterioration prediction model generation step involves generating a deterioration prediction model for each damaged location based on damaged location learning data obtained by classifying the learning data for each damaged location using the machine learning processor, and generating a deterioration prediction model for each damaged state based on damaged state learning data obtained by classifying the learning data for each damaged state.
9. The damage pattern data stored in the learning data storage circuit further includes the cause of damage as the damage pattern, The pipeline deterioration prediction method according to claim 7, wherein the deterioration prediction model generation step involves generating a deterioration prediction model for each damaged location based on damaged location learning data obtained by classifying the learning data for each damaged location using the machine learning processor, generating a deterioration prediction model for each damaged state based on damaged state learning data obtained by classifying the learning data for each damaged state, and generating a deterioration prediction model for each damaged cause based on damaged cause learning data obtained by classifying the learning data for each damaged cause.
10. The pipeline deterioration prediction method according to claim 9, wherein the deterioration prediction value calculation step involves the deterioration prediction value calculation processor calculating and outputting an overall evaluation value for the pipeline to be evaluated by weighting the deterioration prediction values calculated for each of the deterioration prediction models for each of the damaged locations, the deterioration prediction models for each of the damaged conditions, and the deterioration prediction models for each of the damaged causes, based on their respective correlations.
11. The pipeline deterioration prediction method according to claim 9, wherein the deterioration prediction model generation step involves applying the learning data to each of the deterioration prediction models by location of damage, deterioration prediction model by state of damage, and deterioration prediction model by cause of damage, to generate comprehensive deterioration learning data by adding the predicted deterioration values obtained to the learning data, and then generating a comprehensive deterioration prediction model by machine learning the comprehensive deterioration learning data using the machine learning processor.
12. Computers, A learning data storage processing unit stores the following as learning data in a learning data storage circuit: pipeline attribute data indicating pipeline attributes including the type of pipe and the number of years buried in the ground; pipeline burial environment data indicating the burial environment including the soil in which the pipeline is buried; and damage type data indicating the damage type including the location and condition of the damage to the pipeline. A machine learning processing unit reads the learning data stored in the learning data storage circuit, performs machine learning, and generates a deterioration prediction model, A pipeline data storage processing step involves storing, as pipeline data, pipeline attribute data indicating pipeline attributes including the pipe type and years buried of the pipeline to be evaluated, and pipeline burial environment data indicating the burial environment including the soil in which the pipeline to be evaluated is buried, in a pipeline data storage circuit. A deterioration prediction value calculation processing unit that inputs the pipeline data to be evaluated, read from the pipeline data storage circuit, into the deterioration prediction model generated by the machine learning processor, and calculates a predicted value for the deterioration of the pipeline to be evaluated. A pipeline deterioration prediction program designed to enable this function.