Method for generating deterioration prediction model, deterioration prediction method, deterioration prediction system, and deterioration prediction program

Abstract

An object of the present invention is to provide a novel technique capable of grasping the deterioration state of a pipeline that progresses over time. 【Solution means】 A deterioration prediction model generation method for generating a deterioration prediction model for predicting the deterioration of a pipeline over time, comprising: an acquisition step of acquiring pipeline information including a damage history regarding whether or not the pipeline leaked at a certain point in time t; a generation step of estimating parameters of the deterioration prediction model based on a prior distribution of the parameters of the deterioration prediction model, a likelihood function, and the pipeline information. The deterioration prediction model includes environmental variables X1 to X n and parameters β0 to β n , γ, ε, and is a model represented as JPEG0007692587000007.jpg575. Deterioration prediction model generation method.

Description

Technical Field

[0001] The present invention relates to a method for generating a degradation prediction model, a degradation prediction method, a degradation prediction system, and a degradation prediction program.

Background Art

[0002] In recent years, the aging of infrastructure has advanced, and accidents due to water leakage and the like have occurred frequently. In order to prevent such accidents, operators who manage pipelines that supply fluids such as water pipes and gas need to grasp the condition of the pipelines and replace the pipelines before an accident occurs. On the other hand, the number of pipelines that each operator should manage is very large, and it is very difficult to grasp the pipelines that should be replaced. In particular, since buried pipelines are buried underground, it is necessary to dig them up and conduct an investigation to grasp their condition. However, due to problems such as a shortage of funds and personnel, it is practically impossible for pipeline management operators to dig up all pipelines and confirm their condition.

[0003] In view of the above-described situation, technologies for predicting the degradation of pipelines from pipeline data have been developed in order to grasp the condition of the managed pipelines.

[0004] Patent Document 1 describes a pipeline degradation prediction system that predicts the degradation of a pipeline based on pipeline attribute information, which is information related to the pipeline.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0006] However, since pipelines that have not been laid for a long time are extremely unlikely to be damaged, even if the deterioration prediction is performed using the information of pipelines that have not been laid for a long time (for example, pipelines within 40 years from the laying) by utilizing the invention described in Patent Document 1, the breakage probability will be 0 or a value extremely close to it. Even for pipelines that have not been laid for a long time and are unlikely to be damaged, if even a little time has passed since the laying, deterioration due to the passage of time such as corrosion should have progressed to some extent. However, as described above, the state of such deterioration could not be grasped by the invention described in Patent Document 1.

[0007] In view of the above-described situation, an object of the present invention is to provide a novel technique capable of grasping the state of deterioration of a pipeline that progresses over time.

Means for Solving the Problems

[0008] [1] A deterioration prediction model generation method for generating a deterioration prediction model for predicting the deterioration of a pipeline over time, comprising: an acquisition step of acquiring pipeline information including a breakage history regarding whether or not the pipeline leaked at a certain time point t and a feature amount indicating the laying environment of the pipeline; a generation step of estimating parameters of the deterioration prediction model based on the pipeline information, wherein the deterioration prediction model is a model represented as JPEG0007692587000002.jpg575 by environmental variables X1 to X n and parameters β0 to β n , γ, ε, and in the generation step, by repeatedly performing processing related to sampling including proposal of samples of parameters of the deterioration prediction model, evaluation of the plausibility of the proposed samples based on the pipeline information, and determination of adoption of the samples based on the evaluation, a population of samples of parameters is output to estimate the posterior distribution of the parameters of the deterioration prediction model, and the parameters with the maximum posterior probability in the estimated posterior distribution are estimated as the parameters of the deterioration prediction model. ​Method for generating degradation prediction model. [2] In the generation step, the posterior distribution of the parameters of the degradation prediction model is estimated using Markov chain Monte Carlo. The degradation prediction model generation method according to [1]. [3] The prior distribution of the parameters of the degradation prediction model is a distribution determined based on the pipeline environment information included in the pipeline information. The degradation prediction model generation method according to [1] or [2]. [4] In the generation step, the corrosion degree D i approaches 1 as it increases, and the corrosion degree D i is used to evaluate the plausibility of the samples by using a model having a distribution that approaches 0 as it decreases as the likelihood function, and the parameters of the degradation prediction model are estimated. The degradation prediction model generation method according to any one of [1] to [3]. [5] A degradation prediction method including an acquisition step and a prediction step, in the acquisition step, pipeline information of a pipeline to be predicted with unknown presence or absence of damage is acquired, the pipeline information of the pipeline to be predicted includes environmental information regarding the installation environment and time information regarding the installation time, in the prediction step, the pipeline information of the pipeline to be predicted is input into the degradation prediction model generated by using the degradation prediction model generation method according to any one of [1] to [4], and the degradation prediction result of the pipeline to be predicted is output. Degradation prediction method. [6] A degradation prediction system including an acquisition unit and a prediction unit, the acquisition unit acquires pipeline information of a pipeline to be predicted with unknown presence or absence of damage, the pipeline information of the pipeline to be predicted includes environmental information regarding the installation environment and time information regarding the installation time, the prediction unit inputs the pipeline information of the pipeline to be predicted into the degradation prediction model generated by using the degradation prediction model generation method according to any one of [1] to [4], and outputs the degradation prediction result of the pipeline to be predicted. Degradation prediction system. [7] The prediction unit outputs the degree of corrosion as a deterioration prediction result. The deterioration prediction system according to [6]. [8] The prediction unit predicts the corrosion rate based on the time since the pipeline was laid included in the pipeline information and the degree of corrosion output using the deterioration prediction model. The deterioration prediction system according to [6] or [7]. [9] A deterioration prediction program that causes a computer to execute an acquisition step and a prediction step, In the acquisition step, pipeline information of a pipeline to be predicted whose breakage status is unknown is acquired. The pipeline information of the pipeline to be predicted includes environmental information regarding the installation environment and time information regarding the installation time. In the prediction step, by inputting the pipeline information of the pipeline to be predicted into the deterioration prediction model generated using the deterioration prediction model generation method according to any one of [1] to [4], the deterioration prediction result of the pipeline to be predicted is output. Deterioration prediction program.

[0009] According to the invention according to [1], a model for predicting the progress of pipeline corrosion can be generated even without data on the degree of pipeline corrosion.

[0010] According to the invention according to [2], suitable parameters can be estimated even when it is very difficult to obtain the posterior distribution by calculation.

[0011] According to the invention according to [3], a model capable of performing more accurate prediction using a distribution determined based on actual data can be generated.

[0012] According to the invention according to [4], a model capable of performing prediction close to actual corrosion can be generated.

[0013] According to the invention according to [5], even without data such as corrosion rate, deterioration of the pipeline over time can be predicted by assuming the degree of corrosion.

[0014] According to the invention according to [6], by using the information related to the pipeline such as the information related to the time since laying and the generated model

[0015] According to the invention according to [7], even without data on the degree of corrosion, the degree of corrosion can be predicted.

[0016] According to the invention according to [8], data such as the corrosion rate is not required, and the corrosion rate of the pipeline can be obtained.

[0017] According to the invention according to [9], the generated model can be used to predict deterioration.

Advantages of the Invention

[0018] The present invention can provide a novel technique capable of grasping the state of deterioration of a pipeline that progresses over time.

Brief Description of the Drawings

[0019]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Modes for Carrying Out the Invention

[0020] Hereinafter, the present invention will be described in more detail with reference to the accompanying drawings. The drawings show preferred embodiments. However, it can be implemented in many different forms and is not limited to the embodiments described in this specification.

[0021] For example, in the present embodiment, a method for generating a degradation prediction model, a degradation prediction method, etc. will be described, but the same effects can be achieved by a system, an apparatus, a computer program, etc. The program may be provided as a non-transitory computer-readable recording medium or may be provided so as to be downloadable from an external server.

[0022] Hereinafter, in the embodiment to be described, a degradation prediction system 0 that generates a degradation prediction model and performs prediction using pipeline information regarding a pipeline that supplies a fluid generates and performs prediction of a degradation prediction model using pipeline information of a pipeline that supplies a liquid such as a water supply pipe, but may also generate and perform prediction of a degradation prediction model using pipeline information of a pipeline that supplies other fluids such as a gas pipe.

[0023] <System Configuration> FIG. 1 is a block diagram showing the configuration of a system according to an embodiment. As shown in FIG. 1, the degradation prediction system 0 includes a degradation prediction device 1 that generates a degradation prediction model and performs degradation prediction using the generated degradation prediction model, and a pipeline information database DB that stores pipeline information.

[0024] The degradation prediction device 1 is a device such as a server device that generates a degradation prediction model for predicting the degradation of a pipeline and performs degradation prediction using the generated degradation prediction model, and is configured to be able to realize the functional configuration described later. In the example shown in FIG. 1, an example in which the degradation prediction device 1 performs both generation of a degradation prediction model and degradation prediction of a pipeline using the degradation prediction model is described, but a server device for generating a degradation prediction model and a server device for performing degradation prediction using the degradation prediction model may be separately arranged, and it may be configured using a plurality of computer devices. Further, a computer device such as a terminal of an operator who manages a pipeline may function as the degradation prediction device 1 by acquiring a degradation prediction model.

[0025] The pipeline information database DB is a database that stores pipeline information related to pipelines and is configured to be connectable to the deterioration prediction device 1. Note that the pipeline information database DB may be constituted by the deterioration prediction device 1 or may be constituted by one or a plurality of server devices etc. having a recording medium.

[0026] <Data configuration> Hereinafter, an example of the pipeline information stored in the pipeline information database DB will be described with reference to FIG. 3.

[0027] The pipeline information is information related to the installed pipelines. In the present embodiment, the pipeline information includes identification information (e.g., ID etc.) for uniquely identifying the pipeline, information related to the pipeline body, position information related to the position where the pipeline is installed, environmental information related to the environment where the pipeline is installed, and a damage history.

[0028] The information related to the pipeline body (body information) is information related to the pipeline body and includes, for example, the installation year of the pipeline, the pipe type, the joint type, the diameter, etc. In the process described later, the number of years since the installation of the pipeline is obtained from the installation year of the pipeline, but time information such as the number of years since the installation of the pipeline may be included in the body information. Note that the pipe type is the type of the pipeline and includes metal pipes such as ductile iron pipes (DIP) and ordinary cast iron pipes (CIP), and non-metal pipes such as PVC pipes and resin pipes.

[0029] The position information is information related to the position where the pipeline is installed. Note that the position information may be in any format such as an address, latitude and longitude, a code such as coordinates or an EPSG code, a hashed position information such as a Geohash or a Quadkey, a mesh code such as a standard area mesh, or may include information indicating a plurality of positions.

[0030] Environmental information is information about the environment where the pipeline is laid. In this embodiment, the environmental information includes information about the soil at the laying location (e.g., soil type, pH value, etc.), information about the terrain (e.g., terrain type, elevation, slope, etc.), average temperature at the laying location, average precipitation, distance from the main road, and other information. Further, as environmental information, information about the pipe type of adjacent pipelines and other information about adjacent pipelines (adjacent pipeline information) may also be included.

[0031] Furthermore, information about the flow rate, pressure, water supply inlet, flow velocity, etc. of the pipeline obtained by performing hydraulic calculations as environmental information may be included as environmental information.

[0032] Also, the deterioration prediction system 0 may use, as environmental information, information obtained from an external system that stores information about the environment, such as a weather information providing system that provides weather data or a map information providing system. At that time, based on the position information of the pipeline, the deterioration prediction system 0 acquires information about the environment (e.g., weather information, etc.) at the position where the pipeline is laid from an external system, and uses it for the processing described later as environmental information or stores it in the pipeline information database DB.

[0033] The damage history is information about the damage to the registered pipeline obtained and registered by conducting investigations such as leakage inspections on existing pipelines or the occurrence of sudden accidents (e.g., leakage accidents such as water pipe ruptures). In this embodiment, the damage history includes registration time information regarding the time when the pipeline was actually confirmed and the damage history was registered due to the implementation of the investigation or the occurrence of the accident, and the damage type. Note that the pipeline information of pipelines without the implementation of investigations or the occurrence of damage may not include the damage history.

[0034] The damage types shown in Fig. 2(b) are information indicating the causes of damage to the pipeline included in the damage history. For example, information regarding the causes of pipeline damage such as corrosion deterioration, electrolytic corrosion leakage, and vibration leakage is stored. Also, for pipelines that have been investigated but where leakage has not occurred, information such as "no damage" may be stored as the damage type. Note that in this embodiment, the damage history is information regarding water pipe leakage and damage. However, when the object of deterioration prediction is a gas pipe, it may be information regarding the damage history of the article to be predicted for deterioration, such as rupture or damage of the gas pipe.

[0035] In the generation of the deterioration prediction model in this embodiment, the model is generated using data with damage. Also, since the model generation is performed considering the damage history as the center, the processed pipeline information used for model generation is data including the characteristic quantities of the pipeline and one damage history (for example, in year XX, corrosion deterioration). Therefore, for pipelines that have had multiple damage histories, such as pipelines that were damaged and repaired in the past and then damaged again, a plurality of processed pipeline information corresponding to the number of damages is generated for the generation of the deterioration prediction model from the raw data of the pipeline information (pipeline information as shown in Fig. 3).

[0036] <Hardware Configuration> Fig. 2 is a hardware configuration diagram. As shown in Fig. 2, the deterioration prediction device 1 has a processing unit 101, a storage unit 102, and a communication unit 103, which are used to exert the functions of each part and each process.

[0037] As the deterioration prediction device 1, one or more information processing devices such as general-purpose servers or computers can be used. Also, although the deterioration prediction device 1 has the functional configuration described later, a part of its functional configuration may be arranged for other information processing.

[0038] The processing unit 101 has a processor such as a CPU (Central Processing Unit) capable of executing an instruction set, and executes an OS (Operating System) as well as a deterioration prediction model generation program, a deterioration prediction program, etc. The storage unit 102 has a volatile memory such as a RAM (Random Access Memory) that can store an instruction set, and a non-volatile recording medium such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive) that can record an OS or the like. Also, the storage unit 102 stores various programs used in processes such as those of its parameters such as a degradation prediction model. The communication unit 103 has an interface for connecting to a network and performs communication with other devices or the like that execute communication control with the network.

[0039] <Functional configuration> As shown in FIG. 1, the degradation prediction device 1 includes an acquisition unit 11, a generation unit 12, and a prediction unit 13. Hereinafter, the functional configuration of the degradation prediction system 0 in the present embodiment will be described.

[0040] <Acquisition unit 11> The acquisition unit 11 acquires various necessary data such as pipeline information from a pipeline information database DB for learning a degradation prediction model and predicting degradation. Also, the acquisition unit 11 may perform a process of receiving necessary data such as input from a terminal of an operator who manages the pipeline.

[0041] Also, it is assumed that information regarding pipelines such as pipeline information may have different formats for each operator. The acquisition unit 11 may perform a process of arranging the data format, such as unifying the code indicating the pipe type based on a dictionary or aligning the data units, for the acquired pipeline information. Also, the acquisition unit 11 may perform a process for converting the raw pipeline information described above into processed pipeline information suitable for generating a degradation prediction model.

[0042] <Generation unit 12> The generation unit 12 generates a deterioration prediction model based on pipeline information including a damage history. Note that, in order to predict damages that occur over time, the generation unit 12 estimates the parameters of the deterioration prediction model based on pipeline information having a damage history of a damage type related to damages that occur over time, such as corrosion deterioration. Further, in the present embodiment, the parameters of the deterioration prediction model for each pipe type are estimated based on the pipeline information for each pipe type.

[0043] Hereinafter, in Equation 1, the deterioration prediction model in the present embodiment is shown. In the present embodiment, based on the knowledge that it is affected by characteristic quantities related to pipelines such as geology and soil, and the knowledge that the degree of corrosion progresses over time, a model as shown in Equation 1 is created. The deterioration prediction model shown below is a model showing the degree of corrosion D of a certain pipeline i at time t. i which is a model showing, and X1 to X n characteristic quantities related to the pipeline, and parameters β0 to β n , γ, ε, and is expressed by. The degree of corrosion D of the pipeline i cannot actually be observed, but by assuming it as a latent variable whose existence can be inferred from other data, a pipeline deterioration prediction model as shown in Equation 1 is constructed.

[0044]

Equation

[0045] Since the corrosion of the pipeline is affected by the environment in which the pipeline is laid, in the present embodiment, the characteristic quantities related to the pipeline shown as X1 to X n include characteristic quantities (environment information) related to the buried environment such as geology and soil included in the pipeline information. Note that, as long as it is information that affects the deterioration of the pipeline, other characteristic quantities other than the environment information, such as the diameter included in the body information, may be used as the characteristic quantities related to the pipeline. Further, in Equation 1 above, t is a variable of time (the time from burial at the time of prediction), and since the degree of corrosion progresses as time passes, γ>0 must be satisfied.

[0046] The following describes the estimation of the parameters of the degradation prediction model in this embodiment. In this embodiment, in order to generate a model by using the data (pipeline information) collected during the management of the pipeline and leveraging the knowledge obtained during the management of the pipeline, Bayes' theorem is applied to maximize the posterior probability based on the actually observed pipeline information, so as to find the D i parameters such as β and γ that can be obtained. The posterior distribution is the probability distribution (probability distribution) of the values that the parameters β, γ, etc. can take when a certain result (actual pipeline information) is obtained. The value at which the probability (posterior probability) in this probability distribution is maximized is determined as the D i value of the parameter.

[0047] The generation unit 12 estimates the parameters of the degradation prediction model D i described above. In this embodiment, the posterior distribution of the parameters is estimated, and the parameter with the maximum posterior probability is obtained from the estimated posterior distribution, thereby estimating the D i parameters. However, since it is difficult to obtain the posterior distribution by calculation, the shape of the posterior distribution is approximated by sampling using MCMC (Markov chain Monte-Carlo methods) to find the parameter with the maximum posterior probability.

[0048] MCMC is a general term for algorithms that sample a probability distribution by creating a Markov chain that has the probability distribution to be obtained as its equilibrium distribution. In this embodiment, the generation unit 12 uses the MCMC algorithm to perform sampling based on a predefined model such as the described likelihood function and prior distribution (initial knowledge of the parameters) and the observed data (actual pipeline information), so as to obtain a degradation prediction model D such that the observed values fit best. iEstimate the posterior distribution of the parameters (the probability distribution of the parameters when certain pipeline information is observed). As the MCMC algorithm, any MCMC algorithm such as Metropolis-Hastings, NUTS (No-U-Turn Sampler), etc. can be used. In MCMC, specifically, for the parameters to be estimated, a sample of new parameters is proposed from the current parameters, the newly proposed sample is evaluated, and based on this evaluation, a process of determining whether to accept the parameters is repeated to output a population of samples of plausible parameters that conform to the observed data, thereby estimating the shape of the posterior distribution of the parameters by sampling. In this embodiment, β0 to β n , a set of parameters of the degradation prediction model D i (for example, a combination such as "β0: 1,..., β n : 3, γ: 2, ε: 4"), the processing related to sampling such as sample proposal and evaluation is performed.

[0049] Also, modules such as PyMC are known that can execute MCMC based on prior assumptions and observed data to obtain the posterior distribution by sampling. In this embodiment, the generation unit 12 uses a program that can execute MCMC based on the assumptions and actual data (pipeline information) performed in Equation 1 to estimate the posterior distribution, and based on the degradation prediction model shown in Equation 1, the likelihood function shown in Equation 2, assumptions such as the prior distribution of the parameters, and pipeline information that is the information of the actual pipeline (in particular, body information, environmental information, damage history), by performing sample proposal and evaluation, etc., the posterior distribution of the parameters is estimated, and from the estimated posterior distribution, the degradation prediction model D i 's parameters (β0 to β n , γ, ε) are estimated.

[0050] Note that the parameters set in advance (β0 to β nThe prior distribution of (β0, γ, ε) is an arbitrary distribution determined in advance based on actual pipeline information, and it can be any distribution such as a uniform distribution or a normal distribution. Also, the prior distribution of the parameters may be set for each type of parameter. Here, the assumption of the prior distribution of the parameters set in advance affects the proposal of samples. In this embodiment, samples of parameters that follow the assumed distribution (the prior distribution of the parameters) are proposed. Note that when there is no previously observed data, the setting of the prior distribution does not have to be strict, such as setting a uniform distribution or a normal distribution. Note that D i The distribution of may be set in advance and used for the proposal of samples.

[0051] Degree of corrosion D i The likelihood function P indicating the occurrence Y of pipeline breakage when is as follows. The likelihood function is a function that regards a numerical value indicating the plausibility of inferring what the precondition was from the observation result when the result appears according to a certain precondition, with the precondition as a variable. Note that the likelihood function P shown in Equation 2 is a conditional probability indicating the probability of the occurrence of breakage (Y = 1) when takes a certain value, and is a function for evaluating whether the proposed sample of the parameters is a plausible value conforming to the actual data (pipeline information). In this embodiment, since pipeline information having a breakage history of "with breakage" (Y = 1) is used for generating the degradation prediction model, the proposed sample (β0 to β i , γ, ε) and the characteristic quantities (X1 to X n ) of the actual pipeline and a certain D obtained from Equation 1 n When the probability P of having breakage (Y = 1) is high due to the value of, it can be said that the proposed sample is a plausible parameter conforming to the actual data. The generation unit 12 obtains D from the proposed sample and the actual pipeline information i i ​Evaluate using the likelihood function shown in Equation 2, and accept the proposed sample as a plausible sample if the probability P is high (that is, if there is a high likelihood that there is a break (Y = 1), the same as the actual pipeline information). In this embodiment, P is expressed by a distribution as shown in Equation 2. However, in order to approximate the actual movement of corrosion progression such that the probability of breakage increases as the degree of corrosion increases, D i approaches 1 as the value of increases, and D i Any function having a distribution that approaches 0 as the value of decreases may be set as the function, such as a logistic distribution, a Cauchy distribution, or a Weibull distribution. In this embodiment, the generation unit 12 evaluates whether the proposed sample of parameters is plausible based on the likelihood function and the pipeline information as described above. However, any method may be used for evaluation as long as it is possible to evaluate the plausibility of the sample based on the pipeline information.

[0052] In the example described above, the process for generating the degradation prediction model using pipeline information having a break history with a break (Y = 1) is described. However, the process for generating the degradation prediction model using pipeline information without a break (Y = 0) may be performed. When generating the degradation prediction model using pipeline information including a break history with no break (Y = 0), the generation unit 12 determines D obtained using the sample i The process may be performed to accept the proposed sample with a sample having a value of the likelihood function P close to 0 (the probability of breakage is close to 0 = the likelihood of no breakage is high) as a plausible sample.

[0053]

Equation

[0054] As a method for estimating the posterior distribution using MCMC in this embodiment, the generation unit 12 samples the parameters based on the assumptions set in Equations 1 and 2, the prior distribution of the parameters, and the pipeline information of the actually observed pipeline.

[0055] In FIG. 4, the flow of the process related to sampling in the present embodiment is shown. First, the generation unit 12 proposes a sample of new parameters based on assumptions such as the prior distribution of the parameters. When a sample of new parameters is proposed based on assumptions such as the prior distribution, the generation unit 12 evaluates the proposed sample based on the likelihood function and pipeline information, and determines whether to accept the sample of the proposed parameters based on an evaluation using pipeline information including the damage history. Note that the generation unit 12 determines whether to accept the proposed sample based on whether the likelihood of the proposed sample (in the present embodiment, the probability P obtained from the proposed parameters using the likelihood function shown in Equation 2) is greater than or less than a certain reference probability. By determining whether to accept the sample based on such an evaluation in this way, restrictions are imposed so that samples that do not match the actually collected pipeline information are not adopted into the population of samples used to estimate the shape of the posterior distribution. At this time, the generation unit 12 uses the current sample of parameters (β0 to β n , γ, ε) and the feature amounts (X1 to X n ) included in the actual pipeline information to generate the value of D i adopted in Equation 2 (the probability of occurrence of damage obtained from the sample of the proposed parameters), and may evaluate the sample of the proposed parameters based on whether it is close to the occurrence of damage Y (Y = 1; there is an occurrence of damage, Y = 0; there is no occurrence of damage) in the actual damage history, and determine whether to adopt the sample of the proposed parameters.

[0056] In this embodiment, the generation unit 12 repeats the processes related to sampling, namely, the proposal of the above-described parameters, the evaluation of the proposed parameters, and the determination of whether to adopt the parameters based on the evaluation, until the number of samples is sufficient, and outputs a population of plausible parameter samples suitable for reproducing the observed data (pipe line information). Then, the generation unit 12 estimates the shape of the posterior distribution of the parameters from the population of samples. At this time, the generation unit 12 may determine that the number of samples is sufficient when the proposed number of samples or the accepted number of samples reaches a certain value or more, or may determine that it is sufficient by determining that the values of the samples converge to a certain value. Also, when determining whether the samples are sufficient by determining the convergence of the samples, any convergence determination method may be used, such as a method using the Gelman-Rubin statistic or a determination method using the effective sample size (ESS). In the example of FIG. 4, an example of the case where only plausible samples (samples with a probability of P being a certain value or more) are adopted is described. However, samples with a low likelihood (probability of P being lower than a reference value) may be accepted with a certain probability, or the acceptance probability may be determined based on the probability of P and the proposal of samples may be accepted based on that probability.

[0057] In FIG. 5, an example of the posterior distribution of the parameters estimated by sampling is shown. FIG. 5 is a graph showing the posterior distribution of a certain parameter in D i where the horizontal axis represents the value of the parameter and the vertical axis represents the probability. In this embodiment, the generation unit 12 estimates the posterior distribution (probability distribution showing the probability that the parameter takes each value when actual pipe line information is obtained) as shown in FIG. 5 from the population of adopted plausible samples through the processes related to sampling including the proposal of samples and the evaluation of the proposed samples. In the posterior distribution of the parameters as shown in FIG. 5 obtained by the above process, the generation unit 12 uses the value of the parameter with the maximum probability (parameter with the maximum posterior probability) as the plausible parameter, D iThe parameters are adopted. In the example of the posterior distribution shown in FIG. 5, since the probability increases as the value of the parameter approaches 0, the generation unit 12 estimates 0 or a value close thereto as the parameter with the maximum posterior probability. In the present embodiment, the parameter with the maximum posterior probability is determined by a statistic obtained from a population of samples, such as the mode of the population of samples generated to approximate the shape of the posterior distribution (that is, the value with the maximum probability in the sample population). The process related to the estimation of this posterior probability is D i For each of the parameters set to D (β0 to β n , γ, ε), the generation unit 12 estimates a likely value of the parameter and generates a degradation prediction model D i based on the estimated parameter.

[0058] In the present embodiment, in order to predict degradation that occurs over time, such as corrosion, for model generation, only pipeline information having a failure history related to a failure that occurs over time, such as the failure type being "corrosion degradation" in the failure history, is used for model generation. Also, a model may be generated for each failure type, or for metal pipes, only pipeline information having a failure history due to a cause corresponding to the pipe type, such as "corrosion degradation" for metal pipes and "chemical degradation" for non-metal pipes, may be used for generating a degradation prediction model.

[0059] Also, the method for generating the degradation prediction model described above is an explanation of the flow of generating a degradation prediction model for predicting degradation due to corrosion with a metal pipe as the prediction target, but a degradation prediction model is generated in the same flow for each pipe type even for pipelines other than metal pipes, such as non-metal pipes. For example, for a degradation prediction model for predicting degradation of a pipeline such as a non-metal pipe, a degradation prediction model D i indicating the degree of chemical degradation may be generated and the parameters may be estimated.

[0060] <Prediction unit 13> The prediction unit 13 predicts the deterioration of the pipeline over time using the deterioration prediction model generated by the generation unit 12 and the pipeline information acquired by the acquisition unit 11. Note that the pipeline to be predicted may be a pipeline without a damage history, or a pipeline with a damage history such as a pipeline that has been damaged in the past but has been repaired and is still in use.

[0061] The prediction unit 13 inputs various feature quantities (X0 to X n ) such as environmental information included in the pipeline information of the pipeline to be predicted and the number of years since the pipeline was laid into the deterioration prediction model (Equation 1) to output the degree of corrosion of the pipeline as the result of the deterioration prediction.

[0062] Further, the prediction unit 13 may output the corrosion rate as the result of the deterioration prediction by further using the degree of corrosion generated by the deterioration prediction model by referring to the equation related to corrosion provided by ANSI (American National Standards Institute).

[0063] Hereinafter, the equation related to corrosion provided by ANSI in Equation 3 will be shown. The following equation is an equation for obtaining the corrosion rate (depth), indicating that the corrosion rate of a substance can be obtained based on a constant for each material, a time-dependent exponent, time (time since installation), etc.

[0064]

Equation

[0065] In the present embodiment, the prediction unit 13 can use the degree of corrosion D obtained by using the deterioration prediction model generated for each pipe type as the material constant K in the equation for the corrosion rate (depth). i By using this, the corrosion rate is obtained based on the pipeline information and output as the result of the deterioration prediction. At this time, the prediction unit 13 uses the deterioration prediction model expressed by the estimated parameters to calculate the degree of corrosion D iObtain it, and then, using the time-dependent exponent n (any exponent that depends on time) and the time t (time since installation), obtain the corrosion rate. Note that the time-dependent exponent n in the formula provided by ANSI is close to 1, but a value obtained using other data such as the measured value of the corrosion rate (depth) may be used as the time-dependent exponent n.

[0066] Note that in this embodiment, the prediction unit 13 obtains the degree of corrosion, corrosion rate, etc. at the time of predicting deterioration. However, for example, by using the embedding period (time since installation) when assuming that the pipeline is installed until that year for predicting deterioration at a certain future time, the degree of deterioration, deterioration rate, etc. of the pipeline at a certain future time may be predicted.

[0067] <Other Embodiments> Patent Document 1 describes a model that learns by inputting a feature quantity related to a pipeline and outputting the presence or absence of damage to the pipeline, and outputs the probability of pipeline deterioration. However, for such a model that calculates the probability of damage, the corrosion rate, etc. output by the prediction unit 13 in the present invention may be further used as a feature quantity of the pipeline to calculate the probability of damage. Further, the prediction unit 13 may i calculate the probability of damage occurring due to the passage of time such as corrosion based on the value of D. In this case, the prediction unit 13 may calculate the probability of damage due to the passage of time obtained using the value of D and the probability of damage output from a model that calculates the probability of deterioration by inputting the feature quantity of the pipeline, and calculate the average value or a value weighted by an arbitrary weight as the final probability of damage occurring. i

[0068] Also, the result of the pipeline deterioration prediction obtained by the prediction unit 13 may be registered in the pipeline information database DB in association with the pipeline information of the pipeline to be predicted. Further, at this time, if the deterioration prediction system 0 includes a display unit, a map of the pipeline color-coded for each degree and speed of deterioration at the time of deterioration prediction may be displayed on the map based on the registered pipeline information and the result of the deterioration prediction.

Explanation of Reference Numerals

[0069] 0 Degradation prediction system 1 Degradation prediction device 11 Acquisition unit 12 Generation unit 13 Prediction unit

Claims

1. A deterioration prediction model generation method for generating a deterioration prediction model for predicting deterioration of a pipeline over time, comprising: an acquisition step of acquiring pipeline information including a damage history regarding whether or not the pipeline has leaked at a certain time t and a feature quantity indicating the pipeline installation environment; A generation step of estimating parameters of the deterioration prediction model based on the pipeline information, The deterioration prediction model is an environmental variable X 1 ~X n and parameter β ０ ~β ｎ , γ, ε, A model represented as: In the generation step, a sampling process including proposing samples of parameters of the deterioration prediction model, evaluating the likelihood of the proposed samples based on the pipeline information, and deciding to adopt the samples based on the evaluation is repeated to output a group of parameter samples, thereby estimating a posterior distribution of the parameters of the deterioration prediction model, and estimating a parameter with a maximum posterior probability in the estimated posterior distribution as the parameter of the deterioration prediction model. A method for generating a deterioration prediction model.

2. In the generation step, a posterior distribution of parameters of the deterioration prediction model is estimated using a Markov chain Monte Carlo method. The degradation prediction model generating method according to claim 1 .

3. a prior distribution of parameters of the deterioration prediction model is a distribution determined based on environmental information of the pipeline included in the pipeline information; The degradation prediction model generating method according to claim 2 .

4. In the generating step, the corrosion degree D ｉ As the value of D increases, it approaches 1. ｉ A model having a distribution approaching 0 when is small is used as a likelihood function to evaluate the likelihood of the proposed sample, and parameters of the deterioration prediction model are estimated. The degradation prediction model generating method according to claim 1 .

5. A deterioration prediction method including an acquisition step and a prediction step, In the acquiring step, pipeline information of a pipeline to be predicted, the presence or absence of which is unknown, is acquired; The pipeline information of the pipeline to be predicted includes environmental information related to an installation environment and time information related to an installation time, In the prediction step, pipeline information of the pipeline to be predicted is input to the deterioration prediction model generated using the deterioration prediction model generation method according to claim 1, thereby outputting a deterioration prediction result of the pipeline to be predicted. Deterioration prediction methods.

6. A deterioration prediction system including an acquisition unit and a prediction unit, The acquisition unit acquires pipeline information of a pipeline to be predicted, the presence or absence of which is unknown, The pipeline information of the pipeline to be predicted includes environmental information related to an installation environment and time information related to an installation time, the prediction unit inputs pipeline information of the pipeline to be predicted into the deterioration prediction model generated using the deterioration prediction model generation method according to claim 1, thereby outputting a deterioration prediction result of the pipeline to be predicted. Deterioration prediction system.

7. The prediction unit outputs a corrosion degree as a deterioration prediction result. The deterioration prediction system according to claim 6 .

8. the prediction unit predicts a corrosion rate based on a time since the installation of the pipeline included in the pipeline information and a corrosion degree output using the deterioration prediction model. The deterioration prediction system according to claim 6 .

9. A deterioration prediction program for causing a computer to execute an acquisition step and a prediction step, In the acquiring step, pipeline information of a pipeline to be predicted, the presence or absence of which is unknown, is acquired; The pipeline information of the pipeline to be predicted includes environmental information related to an installation environment and time information related to an installation time, In the prediction step, pipeline information of the pipeline to be predicted is input to the deterioration prediction model generated using the deterioration prediction model generation method according to claim 1, thereby outputting a deterioration prediction result of the pipeline to be predicted. Deterioration prediction program.

Images