Method, information processing device, and non-temporary storage medium
By constructing physical and statistical models to predict physical quantities, the method addresses the challenge of varying influencing factors, achieving accurate predictions for component deterioration in complex systems.
Patent Information
- Application Number
- JP2022196397
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2022-12-08
- Publication Date
- 2026-11-19
- Estimated Expiration
- 2042-12-08
AI Technical Summary
Existing methods for predicting the deterioration and thinning of components in complex systems like nuclear and thermal power plants fail to provide highly accurate results due to varying influencing factors over time and across different plants.
A method involving a processor that constructs physical and statistical models to predict physical quantities by formulating physical laws and statistically analyzing trends, combining prediction results from both models to estimate a final outcome.
This approach enables highly accurate prediction of physical quantities like wall thickness, accounting for unknown parameters and varying influencing factors, improving maintenance planning.
Smart Images

Figure 0007954050000002 
Figure 0007954050000003 
Figure 0007954050000004
Abstract
Description
[Technical Field]
[0001] This disclosure relates to a method, an information processing device, and a non-temporary storage medium. More specifically, it relates to a method, an information processing device, and a non-temporary storage medium for predicting physical quantities using statistical methods. [Background technology]
[0002] In complex systems such as nuclear power plants, thermal power plants, and waste treatment plants, it is known that phenomena that change over time, such as deterioration and thinning of pipes and other components, occur due to the fluid flow and conditions inside or outside the components. In such plants, in order to understand the changes in the state of the components, physical quantities (such as wall thickness and weight) corresponding to deterioration and thinning are measured periodically at many points on the components. By periodically measuring wall thickness and other parameters, future wall thickness is predicted from the thinning trend of past wall thickness data, and locations requiring repair or replacement are identified. For example, in Patent Document 1, the remaining lifespan of a pipe is predicted by taking spatial or temporal differences and predicting the lifespan from linear extrapolation of these differences. In this method, it is assumed that the rate of thinning of the pipe is constant, and the time to reach the amount of thinning that would occur if thinning continued at that rate is calculated. [Prior art documents] [Patent Documents]
[0003] [Patent Document 1] Japanese Patent Publication No. 3-289596 [Overview of the project] [Problems that the invention aims to solve]
[0004] However, phenomena such as deterioration and thinning of piping and other components are influenced by various factors, and do not progress uniformly in space or time, but rather differ from plant to plant. For this reason, even within the same plant, when the influencing factors change over time, or when considering other plants with different influencing factors, it was not possible to obtain highly accurate prediction results.
[0005] One of the objectives of the present invention is to provide a highly accurate physical quantity prediction formula in a method for predicting a physical quantity corresponding to deterioration and wall thinning of members constituting a plant. [MEANS FOR SOLVING THE PROBLEMS]
[0006] To solve the above-described problem, one aspect of the present invention is a method executed by a processor, the method including a first group of steps executed by the processor, wherein the first group of steps includes: a step of acquiring measured values of a physical quantity that changes over time; a step of constructing a physical model in which the physical quantity is formulated according to physical laws; a step of constructing a statistical model that statistically analyzes the trend of the measured values and expresses the physical quantity as a time function; a step of acquiring a prediction result of the physical quantity from the statistical model; a step of acquiring a prediction result of the physical quantity from the physical model; and a step of statistically estimating a final prediction result of the physical quantity based on the prediction result of the physical quantity calculated by the physical model and the prediction result of the physical quantity calculated by the statistical model.
[0007] Another aspect of the present invention is an information processing apparatus including a storage unit that stores instructions and a processor, wherein the processor executes the instructions stored in the storage unit to perform a first group of steps executed by the processor, and the first group of steps includes: a step of acquiring measured values of a physical quantity that changes over time; a step of constructing a physical model in which the physical quantity is formulated according to physical laws; and a step of statistically analyzing the trend of the measured values to convert the physical quantity into a time a step of constructing a statistical model expressed as an interval function; a step of acquiring a prediction result of the physical quantity from the statistical model; a step of acquiring a prediction result of the physical quantity from the physical model; and a step of statistically estimating a final prediction result of the physical quantity based on the prediction result of the physical quantity calculated by the physical model and the prediction result of the physical quantity calculated by the statistical model.
[0008] Another aspect of the present invention is a non-temporary recording medium on which a program executed by a processor of an information processing device is recorded, wherein the processor, by executing a first group of steps of the program, obtains measured values of a physical quantity that changes over time, constructs a physical model that formulates the physical quantity according to physical laws, constructs a statistical model that expresses the physical quantity as a function of time by statistically analyzing the trend of the measured values, obtains prediction results of the physical quantity from the statistical model, obtains prediction results of the physical quantity from the physical model, and statistically estimates the final prediction result of the physical quantity based on the prediction results of the physical quantity calculated by the physical model and the prediction results of the physical quantity calculated by the statistical model. [Brief explanation of the drawing]
[0009] [Figure 1] A schematic diagram of a waste-to-energy plant according to one embodiment of this disclosure is shown. [Figure 2] This disclosure outlines a system according to one embodiment. [Figure 3A] An example of a model structure for predicting a physical quantity to be predicted, according to one embodiment of this disclosure, is shown. [Figure 3B] An example of a model structure for predicting a physical quantity to be predicted according to one embodiment of this disclosure is shown. [Figure 3C] An example of a model structure for predicting a physical quantity to be predicted according to one embodiment of this disclosure is shown. [Figure 4] This diagram shows a conceptual model structure for predicting a physical quantity to be predicted, according to one embodiment of this disclosure. [Figure 5] An example of a processing flow for predicting a physical quantity according to one embodiment of this disclosure is provided below. [Figure 6] An example of a constructed physical model according to one embodiment of this disclosure is shown. [Figure 7] This figure conceptually illustrates the process of constructing a statistical model according to one embodiment of this disclosure. [Figure 8]This figure shows the probability density distribution of the oxide film thickness before resampling and the probability density distribution of the oxide film thickness after resampling is repeated multiple times, according to one embodiment of the present disclosure. [Figure 9] This figure shows the concept of a multiplication process according to one embodiment of the present disclosure. [Figure 10] A conceptual diagram of a model structure for predicting a physical quantity to be predicted, according to one embodiment of this disclosure, is shown. [Figure 11] An overview of a processing flow for predicting unknown physical parameters, performed by the processing unit 200 according to one embodiment of the present disclosure, is shown. [Figure 12] An example of the distribution of theoretical exhaust gas volume, theoretical air volume, and air ratio according to one embodiment of this disclosure is shown. [Modes for carrying out the invention]
[0010] [Description of Embodiments in this Disclosure] First, the contents of the embodiments of this disclosure will be listed and explained. One embodiment of this disclosure has the following configuration.
[0011] [Item 1] A method performed by a processor, the method being performed by the processor This includes a first set of steps, the first set of steps, The steps include obtaining measurements of a physical quantity that changes over time (for example, wall thickness W), The steps include: constructing a physical model in which the aforementioned physical quantities are formulated according to physical laws, The steps include constructing a statistical model that expresses the physical quantity as a time function by statistically analyzing the trend of the measured values, The steps include obtaining the prediction result of the physical quantity from the aforementioned statistical model, The steps include obtaining the prediction result of the physical quantity from the physical model, A method comprising the step of statistically estimating the final predicted result of the physical quantity based on the predicted result of the physical quantity calculated by the physical model and the predicted result of the physical quantity calculated by the statistical model.
[0012] [Item 2] The method according to item 1, wherein the step of estimating the final prediction result includes multiplying the prediction result of the physical quantity calculated by the physical model by the prediction result of the physical quantity calculated by the statistical model.
[0013] [Item 3] The aforementioned multiplication step is, The method according to item 2, comprising the step of multiplying the probability density value corresponding to the prediction result of the physical quantity calculated by the physical model by the probability density value corresponding to the prediction result of the physical quantity calculated by the statistical model.
[0014] [Item 4] The method according to any one of items 1 to 3, wherein the physical parameters of the physical model include a first physical parameter of unknown value (e.g., oxide film thickness L).
[0015] [Item 5] The step of estimating the final prediction result further includes: The method according to item 4, further comprising the step of resampling the probability distribution of the first physical parameter so that the prediction result of the physical quantity calculated by the physical model approaches the prediction result of the physical quantity calculated by the statistical model.
[0016] [Item 6] The step of performing the resampling process is as follows: The steps include discretizing the first physical parameter into particles, substituting each of the discretized particles into the physical model, and calculating the prediction result of the physical quantity. The process involves selecting high-likelihood particles from among the particles, generating new particles based on the selected particles, and repeatedly selecting high-likelihood particles from the generated particles, so that the predicted result of the physical quantity calculated by the physical model approaches the predicted result of the physical quantity calculated by the statistical model, thereby bringing the predicted result of the physical quantity calculated by the statistical model closer to the predicted result of the physical quantity calculated by the statistical model. The method described in item 5, including the method described in item 5.
[0017] [Item 7] If the physical parameters of the physical model include a second physical parameter of unknown value (e.g., oxygen concentration), the method includes a second set of steps performed by the processor, the second set of steps being: The steps include: estimating the prediction results of the second physical parameter, Steps to update the physical model with the predicted results of the estimated second physical parameter. and, The method described in item 6, including the method described in item 6.
[0018] [Item 8] An information processing apparatus comprising a storage unit for storing instructions and a processor, wherein the processor includes a first group of steps executed by the processor by executing instructions stored in the storage unit, and the first group of steps is Steps include obtaining measurements of physical quantities that change over time, The steps include: constructing a physical model in which the aforementioned physical quantities are formulated according to physical laws, The steps include constructing a statistical model that expresses the physical quantity as a time function by statistically analyzing the trend of the measured values, The steps include obtaining the prediction result of the physical quantity from the aforementioned statistical model, The steps include obtaining the prediction result of the physical quantity from the physical model, An information processing apparatus comprising the step of statistically estimating the final prediction result of the physical quantity based on the prediction result of the physical quantity calculated by the physical model and the prediction result of the physical quantity calculated by the statistical model.
[0019] [Item 9] A non-temporary recording medium on which a program executed by the processor of an information processing device is recorded, The processor executes the first set of steps of the program, Obtain measurements of physical quantities that change over time. A physical model is constructed in which the aforementioned physical quantities are formulated according to physical laws. A statistical model is constructed that expresses the aforementioned physical quantity as a time function by statistically analyzing the trend of the measured values. From the aforementioned statistical model, the prediction results of the physical quantity are obtained, From the aforementioned physical model, the prediction results of the physical quantities are obtained, A non-temporary storage medium that statistically estimates the final predicted result of a physical quantity based on the predicted result of the physical quantity calculated by the physical model and the predicted result of the physical quantity calculated by the statistical model.
[0020] [Details of the embodiments of this disclosure] Embodiments of the present disclosure will be described below with reference to the drawings. Note that the following description is merely an example and is not intended to limit the technical scope of the present invention to these embodiments. In the drawings, identical or similar elements are denoted by identical or similar reference numerals, and redundant descriptions of identical or similar elements may be omitted in the description of each embodiment. Furthermore, features shown in each embodiment are applicable to other embodiments as long as they do not contradict each other. However, embodiments of the present disclosure are not necessarily limited to these forms. It will be apparent to those skilled in the art that embodiments of the present disclosure can take various forms that fall within the scope defined in the claims.
[0021] Hereinafter, as one embodiment of this disclosure, a method for predicting physical quantities to be predicted in complex systems such as various plants will be described with reference to the drawings. In this disclosure, plants include plants that produce energy (electricity, gas), plants that produce food, pharmaceuticals, etc., waste treatment plants, and biomass power plants. The physical quantities to be predicted are quantities that change over time due to influencing factors. In the example of a waste treatment plant, this may include the thickness, strength, and weight of components that make up the equipment used in the plant. The physical quantities to be predicted may also be physical quantities related to liquids or gases. In this case, the physical quantities to be predicted may be the concentrations of components contained in the liquid or gas, for example, certain components in exhaust gas (oxygen (O2), water vapor (H2O), hydrogen chloride (HCl), chlorine (Cl2)). Any concentration is acceptable. For example, in a waste treatment plant, the equipment includes boilers, incinerators, turbines, conveyors, etc., and the components include boiler water tubes, superheater tubes, heat transfer tubes, refractory materials inside the incinerator, conveyor scrapers and rails, blowers and turbines, pump impellers, etc. The materials of the components include all kinds of mechanical materials such as metals, inorganic materials, organic materials, and composite materials.
[0022] The components related to the physical quantities to be predicted in this disclosure are not limited to components used in large-scale facilities such as waste-to-energy plants, but may be any components that require inspection or replacement due to deterioration over time caused by corrosion or wear resulting from environmental or mechanical actions, and may include components used in building structures (e.g., structural materials such as steel frames, piping) or components inside vehicles (e.g., engine exhaust system components).
[0023] In the embodiments described below, we will explain an example of predicting a physical quantity (in this example, the wall thickness of piping such as heat transfer tubes) in a waste-to-energy plant.
[0024] Figure 1 shows a schematic diagram of a commonly used waste-to-energy plant 10. In Figure 1, 1 is the incinerator, 2 is the waste heat boiler, 3 is the pit for storing waste, 4 is the hopper, 5 is the crane for transferring waste from pit 3 to hopper 4, 6 is the dust feeder for pushing waste from the bottom of hopper 4 into the incinerator 1, and 9 is the heat transfer tube of the boiler. Also in Figure 1, 21 is the platform, from which waste collected by garbage trucks 22 is fed into the waste pit 3. The incinerator 1 includes a drying zone, combustion zone 1, combustion zone 2, and post-combustion zone from left to right in Figure 1. Waste is transported within the incinerator 1 by stokers 7a, 7b, 7c, and 7d provided in each zone, and air is blown up and supplied from below via air volume adjustment dampers Ba, Bb, Bc, and Bd provided at the bottom of each zone. Secondary air and recirculated exhaust gas are supplied to the top of the incinerator 1. The ash remaining after waste incineration is discharged through the ash discharge channel 8. 23 indicates the flow of fluids such as exhaust gas within the plant. Although not shown in the diagram, the waste-to-energy plant 10 also includes means for removing harmful gases such as soot and dioxins from the exhaust gas generated by incineration. The generator is also omitted from the diagram.
[0025] In the waste-to-energy plant 10, there are various physical quantities whose characteristics change over time (deterioration due to aging) due to influencing factors. One of the physical quantities that changes over time is the wall thickness of the piping inside the waste-to-energy plant 10, and the wall thickness of the piping gradually decreases due to deterioration over time. There are various factors that influence the changes in physical quantities over time, and one of these influencing factors is the properties of the waste fed into the incinerator 1 in the waste-to-energy plant 10. The properties of the waste vary depending on the local government, facility, season, or changes in the local government's waste disposal regulations. Depending on the properties of the waste, the composition and temperature of the exhaust gas generated when the waste is incinerated change, and as a result, the environment inside the waste-to-energy plant 10, such as the exhaust gas temperature, exhaust gas properties, and fly ash properties inside the boiler 2, change. Another influencing factor is the operating conditions of the waste-to-energy plant 10, and by changing the operating conditions, for example, the exhaust gas flow rate inside the boiler 2, the oxygen concentration in the exhaust gas, the moisture concentration in the exhaust gas, etc., change. Because the properties of the waste and the operating conditions of the plant differ from plant to plant, the exhaust gas temperature, exhaust gas composition, fly ash properties, exhaust gas flow rate, oxygen concentration in the exhaust gas, and moisture concentration in the exhaust gas in the boiler 2 of the waste-to-energy plant 10 change. For this reason, the tendency for corrosion and thinning of the piping of the equipment used in the waste-to-energy plant 10 changes from plant to plant, or in response to changes in operating conditions.
[0026] Next, an overview of a system 20 according to one embodiment of this disclosure will be described with reference to Figure 2. The processing device 200 can be implemented by an information processing device with a well-known configuration as described later in Figure 2.
[0027] System 20 comprises a database 220, a sensor 240, and a processing unit 200. The database 220, sensor 240, and processing unit 200 are connected to, for example, a LAN (Local Area Network), a WAN (Wide Area Network), the Internet, etc. They are connected to each other and can communicate with one another via the network shown in the diagram.
[0028] Database 220 is a collection of information stored in memory or an HDD (hard disk drive) and organized for retrieval or storage. Database 220 holds a large amount of data on measured physical quantities. The measured data is detected by sensor 240 and registered in database 220 in real time in a time series, or measured periodically or irregularly by maintenance personnel and registered in database 220 in a time series. The measured data is data associated with the physical quantity to be predicted. When the physical quantity to be predicted is the wall thickness of a pipe, the measured data may include historical data of the pipe wall thickness measured at multiple points, historical data of the temperature measured by a temperature sensor that senses the surface temperature of the pipe, historical data of the temperature measured by a temperature sensor that senses the temperature inside the incinerator 1 (shown in Figure 1), etc.
[0029] Database 220 may include not only measurement data but also data related to predetermined settings, coefficients, and conditions set by the maintenance administrator. For example, setting values include the fixed values of physical parameters that constitute the physical model, and coefficients include the coefficients of the physical model equation and the statistical model equation. Details of the physical model and the statistical model will be described later.
[0030] The sensor 240 may include a temperature sensor for detecting the surface temperature of the heat transfer tube 9, a temperature sensor for detecting the exhaust gas temperature, a temperature sensor for detecting the combustion temperature inside the incinerator 1, a flow rate sensor for detecting the exhaust gas flow velocity, a gas sensor for detecting specific components contained in the exhaust gas, and the like.
[0031] The processing unit 200 is a computer installed in the waste-to-energy plant 10, or a server computer located on the cloud. The processing unit 200 is composed of a computer having hardware resources such as a CPU (processor), ROM (Read Only Memory), RAM (Random Access Memory), and HDD (Hard Disk Drive). The CPU executes various programs stored in the memory unit. In this way, software-based information processing is realized using hardware resources. Furthermore, the prediction method of this embodiment is realized by having the computer's processor execute a program. Various programs can be stored in various types of memory such as ROM and RAM, and recording media such as HDDs.
[0032] <Model Structure> Figures 3A, 3B, and 3C illustrate, respectively, examples of the logical concepts of model structures 30A, 30B, and 30C for predicting physical quantities according to one embodiment of the present disclosure. Hereinafter, when it is not necessary to distinguish between model structures 30A, 30B, and 30C, they may be referred to as model structure 30. Model structure 30 comprises at least one physical model 32 and at least one statistical model 34. The processing unit 200 can estimate the final prediction result of a physical quantity based on the prediction results obtained from the physical model 32 and the prediction results obtained from the statistical model 34. Furthermore, the processing unit 200 can estimate the prediction result of the physical quantity to be predicted even if the physical model 32 includes parameters with unknown values.
[0033] Physical model 32 is a formalization of a phenomenon occurring in a certain physical quantity (e.g., thinning of pipe wall) according to physical laws. Physical model 32 and the model parameters constituting physical model 32 (hereinafter referred to as physical parameters) are selected in advance by the maintenance manager. In physical model 32, the physical relationships of each parameter (hereinafter referred to as physical parameters) are formalized. Physical model 32 is, in one example, a model that can reproduce the change in wall thickness due to thinning of pipe wall due to corrosion in piping within a waste-to-energy plant 10 (corresponding to the first embodiment described later). The physical parameters of physical model 32 may include known parameters for which there is directly measurable real data, and unknown parameters for which there is no real data for reasons such as not being directly measurable. In this disclosure, parameters with unknown values are those for which no actual data exists and whose values or possible ranges are unknown. On the other hand, known parameters are those for which actual data exists and whose possible ranges, values, or theoretical values are known.
[0034] The statistical model 34 is a model that statistically represents the trend of a certain physical quantity. For example, the statistical model 34 can statistically analyze the trends of the data based on the database 220 and express the physical quantity as a function of time or as a function of other physical quantities. The statistical model 34 and the model parameters that constitute the statistical model 34 (hereinafter referred to as statistical parameters) are selected in advance by the maintenance manager. Examples of statistical models include linear regression models, multiple regression models, polynomial regression models, logistic regression models, state-space models, hierarchical Bayesian models, and time series models. For example, the statistical model 34 is a linear regression model that represents the relationship between the wall thickness of piping in the waste-to-energy plant 10 and time (corresponding to the first embodiment described later).
[0035] (First model structure) Figure 3A shows an example of a first model structure 30A, which consists of one physical model 32 and one statistical model 34.
[0036] In Figure 3A, the physical model 32 is a model that formulates the physical quantity to be predicted according to physical laws. The physical model 32 consists of one or more physical parameters of unknown value and one or more known parameters. The statistical model 34 is a model that statistically represents the trend of change in the physical quantity to be predicted over time.
[0037] The processing unit 200 generates a physical model 32 by assigning an arbitrary probability distribution, such as a uniform distribution or a normal distribution, to one or more physical parameters whose values are unknown. Based on the prediction results of the physical quantities obtained from the physical model 32 and the prediction results of the physical quantities obtained from the statistical model 34, the processing unit 200 can estimate the probability distribution of the aforementioned physical parameters whose values are unknown and the final probability distribution (prediction result) of the physical quantities. Details of the process for obtaining the final prediction result (resampling, multiplication) will be described later. The example in Figure 3A corresponds to the first embodiment described later.
[0038] (Second model structure) Figure 3B shows an example of a model structure 30B consisting of a set of one physical model 32(1) and one statistical model 34(1) (Stage 1), and a set of one physical model 32(2) and one statistical model 34(2) (Stage 2). Hereinafter, when it is not necessary to distinguish between physical models 32(1) and (2), they will be referred to as physical model 32, and when it is not necessary to distinguish between statistical models 34(1) and (2), they will be referred to as statistical model 34. Stage 1 is configured to output the prediction results (final prediction results) of physical quantities. The physical model 32(1) in Stage 1 includes one or more first physical parameters with unknown values and one or more second physical parameters with unknown values, and Stage 2 is configured to output the prediction results (first prediction results) of these one or more second physical parameters to the physical model in Stage 1.
[0039] The Stage 1 physical model 32(1) is a model that represents the physical quantities to be predicted according to physical laws. The Stage 1 physical model 32(1) consists of one or more first physical parameters whose values are unknown, one or more second physical parameters which are the first prediction results output from Stage 2, and known physical parameters.
[0040] The Stage 2 physical model 32(2) is a model that expresses one or more second physical parameters according to physical laws. The Stage 2 physical model 32(2) consists of known physical parameters or a third physical parameter whose value is unknown.
[0041] The Stage 1 statistical model 34(1) is a model that statistically represents the trend of changes in the physical quantity to be predicted over time.
[0042] The Stage 2 statistical model 34(2) is a model that statistically represents the trend of one or more second physical parameters.
[0043] The following is an overview of the estimation process for the specific physical quantities to be predicted.
[0044] The processing unit 200 obtains the prediction results of the second physical parameter obtained from the physical model 32(2) in stage 2 and the prediction results of the second physical parameter obtained from the statistical model 34(2), and estimates the probability distribution of the second physical parameter (first prediction result) based on these.
[0045] Here, if the physical model 32(2) in Stage 2 includes one or more third physical parameters whose values are unknown, the processing unit 200 can construct the physical model 32(2) by assigning any probability distribution, such as a uniform distribution or a normal distribution, to the third physical parameter. The processing unit 200 may be configured to obtain the prediction results of the second physical parameter obtained from the constructed physical model 32(2) and the prediction results of the second physical parameter obtained from the statistical model 34(2), and to estimate the probability distribution of the third physical parameter and the probability distribution of the second physical parameter (first prediction result) based on these.
[0046] Next, the processing unit 200 updates the physical model 32(1) from stage 1 with the first prediction results of the second physical parameter estimated in stage 2. The processing unit 200 also constructs the physical model 32(1) by assigning arbitrary probability distributions, such as a uniform distribution or a normal distribution, to one or more first physical parameters whose values are unknown.
[0047] Next, in Stage 1, the processing unit 200 can estimate the probability distribution of the first physical parameter and the final probability distribution of the physical quantity based on the prediction results of the physical quantity obtained from the updated physical model 32(1) and the prediction results of the physical quantity obtained from the statistical model 34(1). The example in Figure 3B corresponds to the second embodiment described later.
[0048] (Third model structure) In the example in Figure 3A, the model structure 30A consists of one stage, and in the example in Figure 3B, the model structure 30B consists of two stages, but there may be three or more stages.
[0049] Figure 3C shows an example of a model structure 30C, which consists of a set of one physical model 32(1) and one statistical model 34(1) (Stage 1), a set of one physical model 32(2) and one statistical model 34(2) (Stage 2), and a set of one physical model 32(3) and one statistical model 34(3) (Stage 3). Here, Stage 1 is configured to output the prediction result of a physical quantity (final prediction result). The physical model 32(1) in Stage 1 includes one or more first physical parameters with unknown values and one or more second physical parameters with unknown values, and Stage 2 is configured to output the prediction result of this second physical parameter (second prediction result). The physical model 32(2) in Stage 2 includes one or more third physical parameters with unknown values, and Stage 3 is configured to output the prediction result of the third physical parameter (first prediction result).
[0050] The Stage 1 physical model 32(1) is a model that represents the physical quantities to be predicted according to physical laws. The Stage 1 physical model 32(1) consists of one or more first physical parameters whose values are unknown, one or more second physical parameters which are second prediction results output from Stage 2, and known physical parameters.
[0051] The Stage 2 physical model 32(2) is a model that expresses the second physical parameter according to the laws of physics. As shown in the figure, the Stage 2 physical model 32(2) consists of one or more third physical parameters of unknown value and known physical parameters, or it consists of one or more third physical parameters of unknown value, one or more fourth physical parameters of unknown value, and known physical parameters.
[0052] The Stage 3 physical model 32(3) is a model that expresses the third physical parameter according to the laws of physics. The Stage 3 physical model 32(3) may consist of only one or more known physical parameters, or it may be configured to include one or more fifth physical parameters of unknown value in addition to the known physical parameters.
[0053] The Stage 1 statistical model 34(1) is a model that statistically represents the trend of changes in the physical quantity to be predicted over time.
[0054] The Stage 2 statistical model 34(2) is a model that statistically represents the trend of the second physical parameter.
[0055] The Stage 3 statistical model 34(3) is a model that statistically represents the trend of the third physical parameter.
[0056] The following is an overview of the estimation process for the specific physical quantities to be predicted.
[0057] The processing unit 200 obtains the prediction results of the third physical parameter obtained from the physical model 32(3) in Stage 3 and the prediction results of the third physical parameter obtained from the statistical model 34(3), and estimates the probability distribution of the third physical parameter (first prediction result) based on these. Next, the processing unit 200 updates the physical model 32(2) in Stage 2 with the first prediction result of the third physical parameter estimated in Stage 3.
[0058] Here, if the physical model 32(3) in stage 3 includes one or more fifth physical parameters whose values are unknown, the processing unit 200 can construct the physical model 32(3) by assigning any probability distribution, such as a uniform distribution or a normal distribution, to the fifth physical parameter. The processing unit 200 may be configured to obtain the prediction results of the third physical parameter obtained from the constructed physical model 32(3) and the prediction results of the third physical parameter obtained from the statistical model 34(3), and to estimate the probability distribution of the fifth physical parameter and the probability distribution of the third physical parameter (first prediction result) based on these.
[0059] Next, in Stage 2, the processing unit 200 estimates the probability distribution of the second physical parameter (the second prediction result) based on the prediction results of the second physical parameter obtained from the updated physical model 32(2) and the prediction results of the second physical parameter obtained from the statistical model 34(2).
[0060] Here, if the physical model 32(2) in Stage 2 includes one or more fourth physical parameters whose values are unknown, the processing unit 200 can construct the physical model 32(2) by assigning an arbitrary probability distribution, such as a uniform distribution or a normal distribution, to the fourth physical parameter. The processing unit 200 may be configured to obtain the prediction results of the second physical parameter obtained from the constructed physical model 32(2) and the prediction results of the second physical parameter obtained from the statistical model 34(2), and to estimate the probability distribution of the fourth physical parameter and the probability distribution of the second physical parameter (second prediction result) based on these.
[0061] Next, the processing unit 200 updates the physical model 32(1) from Stage 1 with the second prediction result of the second physical parameter estimated in Stage 2. The processing unit 200 also constructs the physical model 32(1) by assigning an arbitrary probability distribution, such as a uniform distribution or a normal distribution, to the first physical parameter whose value is unknown.
[0062] Next, in Stage 1, the processing unit 200 can estimate the final probability distribution of the physical quantity based on the prediction results of the physical quantity obtained from the updated physical model 32(1) and the prediction results of the physical quantity obtained from the statistical model 34(1).
[0063] In this disclosure, even if the physical model 32 includes physical parameters with unknown values, by treating the physical parameters with unknown values as probability distributions rather than definite values and constructing the physical model 32, it is possible to obtain prediction results that take into account the errors that occur in the process of obtaining prediction results for physical quantities.
[0064] Furthermore, in this disclosure, even if the physical model 32 includes physical parameters whose values are unknown, the probability distribution of the physical quantities can be estimated by referring to the prediction results of the physical quantities output from the statistical model 34.
[0065] Furthermore, according to this disclosure, even if the physical parameters constituting the physical model 32 representing the physical quantity to be predicted include multiple parameters with unknown values, it is possible to estimate the physical quantity to be predicted.
[0066] <First Embodiment> Figure 4 shows a conceptual diagram of a model structure 40 for predicting a physical quantity to be predicted, according to one embodiment of the present disclosure. This embodiment corresponds to the model structure 30A in Figure 3A. In this embodiment, for the sake of explanation, the conditions necessary to predict the change in pipe wall thickness over time are simplified. In the following example, the pipe material is pure iron, and the pipe thins only due to oxidation. The pipe is also assumed to be in an atmosphere with an oxygen concentration that fluctuates within a known range. The pipe wall thickness is assumed to thin approximately linearly over time, and the thickness of the oxide film on the pipe is assumed to be constant everywhere. In this embodiment, the future distribution of pipe wall thickness is predicted when a pipe with an initial wall thickness of 4 mm is continuously exposed to an oxygen atmosphere for 10 years, and the wall thickness is measured at 50 points once a year. In this embodiment, for the sake of simplifying the model, the only influencing factor that affects the change in the phenomenon over time is the oxygen concentration. As the oxygen concentration, which is the influencing factor, changes, the diffusion rate of O2 in the oxide film on the pipe changes, and the amount of pipe thinning changes.
[0067] In the first embodiment, the oxygen concentration range is known, and the thickness of the oxide film is unknown.
[0068] Next, we will describe specific examples of the physical model 32 and statistical model 34 under the simplified conditions described above. The physical model 32 and statistical model 34 are assumed to be selected and prepared in advance.
[0069] (Physical model) Physical model 32 is a model that formalizes the physical laws governing the phenomenon, and a physical model is selected that can simply and accurately represent the mechanism of the phenomenon's occurrence. In this embodiment, physical model 32 is assumed to be an oxidative thinning model of a pipe in which the wall thickness of a pure iron pipe decreases over time due to oxidation and corrosion. The oxidation reaction equation representing this physical model 32 is 3Fe + 2O2 → Fe3O4.
[0070] The pipe oxidation thinning model (physical model 32) illustrates how the iron on the surface of a component (pipe) reacts with oxygen to produce triiron tetroxide (Fe3O4), a corrosion product, which causes the base material of the component to thin. Physical model 32 is expressed by the following equation (1).
[0071]
number
[0072] Equation (1) shows the wall thickness W at time t. Here, W is the wall thickness of the pipe (mm), W0 is the initial wall thickness of the member (mm), and O2 is the oxygen concentration (mol / m³). 3 ), where D is the diffusion constant of the triiron tetroxide (Fe3O4) layer, ρ is the density of the Fe3O4 layer, L is the oxide film thickness (mm) corresponding to the thickness of the triiron tetroxide (Fe3O4) layer, and t is time (years). The oxide film thickness L is a physical parameter whose value is unknown. Diffusion constant D (1.0 × 10⁻⁶) -9 (mm / s)), density ρ(2.23×10 4 (mol / m 3 )) are physical parameters with known values. Although the exact value of oxygen concentration O2 is unknown, its distribution (here, mean: 2.23 (mol / m³)) is unknown. 3 ), standard deviation: 0.5(mol / m 3 Assume that the distribution is normal. ) are known physical parameters.
[0073] (Statistical model) The statistical model is a model that statistically predicts the change in wall thickness over time in the target pipe in this embodiment. The statistical parameters are optimized to minimize the cost function using the change in measured values of the physical quantity (wall thickness) over time as input data. In this embodiment, a linear regression model is selected as the statistical model 34 based on the trend of the measured values of the pipe wall thickness over time. The statistical model 34 is expressed as shown in equation (2) below. W=W0-a×t Formula (2)
[0074] Here, W is the pipe wall thickness (mm), W0 is the initial wall thickness (mm), t is time (years), and a is the wall thinning rate (mm / year). Time t is the explanatory variable, and wall thickness W is the dependent variable. Initial wall thickness W0 and wall thinning rate a are statistical parameters, but in this embodiment 1, for simplicity, the initial wall thickness W0 is a fixed value that is known (=4mm), and only the wall thinning rate a is considered a statistical parameter.
[0075] Figure 5 shows an overview of the processing flow 500 for predicting physical quantities, which is performed by the processing unit 200 in the model structure 40 shown in Figure 4. Below, each process performed by the processing unit 200 to obtain the prediction results of physical quantities using the pre-prepared physical model 32 and statistical model 34 will be explained.
[0076] In step S502, the processing unit 200 obtains measurement data of the pipe wall thickness, which is the physical quantity to be predicted, from the database 220. The database 220 stores measurement data of the pipe wall thickness measured once a year at 50 locations on the pipe over a period of 10 years.
[0077] In step S504, the processing apparatus 200 obtains a physical model 32 that formulates the trend of changes in the physical quantity to be predicted over time based on physical laws, and physical parameters that serve as input to the physical model 32, and constructs the physical model 32 based on these. Equation (1) is set to 4 (mm) for the initial wall thickness W0, a uniform distribution for the oxide film thickness L since its value is unknown, and an average of 2.23 (mol / m³) for the oxygen concentration O2. 3 ), standard deviation 0.5 (mol / m 3) is given a normal distribution to construct the physical model 32. The thickness of the oxide film, which is a physical parameter with an unknown value By giving a uniform distribution to L, it is possible to obtain prediction results that take into account the distribution range of the oxide film thickness L.
[0078] Regarding the distribution range of the uniform distribution given to the oxide film thickness L, when there is no a priori information on the range of possible values of L, it is necessary to set a sufficiently wide distribution width. On the other hand, when some a priori information exists regarding the possible values of L, setting the distribution range as narrow as possible with reference to the a priori information is preferable because it can increase the calculation speed of the resampling process described later.
[0079] Figure 6 shows an example 610 of the constructed physical model 32. The initial wall thickness is 4 (mm), and the variation in wall thickness W increases as time t elapses.
[0080] Returning to Figure 5, in step S506, the processing device 200 constructs a statistical model 34 that represents the tendency of temporal change of physical quantities in a probabilistic and statistical manner. In the present embodiment, the distribution of the wall thinning rate a, which is a statistical parameter, is statistically estimated by the least squares method. The wall thinning rate a may be estimated by other statistical methods such as maximum likelihood estimation, maximum a posteriori probability estimation (MAP estimation), Bayesian estimation methods such as Markov Chain Monte Carlo (MCMC) method and Hamiltonian Monte Carlo method. The processing device 200 calculates the average value and standard deviation of the wall thinning rate a, assuming that the probability distribution of the wall thinning rate a follows a normal distribution. Further, the processing device 200 obtains the statistically estimated average value (in this example, -0.05 (mm / year)) and standard deviation (in this example, 8.06×10 -5 (mm / year)), generates a plurality of (for example, 1000) random numbers following a normal distribution, and substitutes each of them into Equation (2) of the statistical model 34 to construct the statistical model 34 (regression equation).
[0081] Figure 7 is a diagram conceptually illustrating an example of the construction process of the statistical model 34.
[0082] Figure 710 illustrates a graph showing the relationship between the measured pipe wall thickness W (mm) and time (t), where the initial wall thickness W0 was set to 4 mm, after measuring the pipe wall thickness at 50 locations once a year for 10 years. The measured pipe wall thickness W (mm) is read from database 220. From graph 710, it can be seen that the pipe wall thickness decreases linearly as time (t) progresses. For this reason, a linear regression model is selected as the statistical model 34. The processing unit 200 performs statistical analysis on the statistical model 34, for example, linear regression analysis using the least squares method, to statistically estimate the probability distribution of the thinning rate a. Note that a detailed explanation of linear regression analysis using the least squares method is omitted as it can be performed using known methods.
[0083] Figure 720 is a histogram showing the probability density distribution of the estimation results in the statistical model 34 of the wall thinning rate a according to one embodiment of the present disclosure. In this embodiment, the probability distribution of the wall thinning rate a in the regression equation representing the wall thickness of the pipe is assumed to be a normal distribution based on the shape of the histogram 720, but other distributions such as a uniform distribution may also be used.
[0084] Figure 730 shows a constructed statistical model 34 according to one embodiment of this disclosure. Figure 732 shows the forecast results of the statistical model 34 from year 10 onward.
[0085] Returning to Figure 5, in step S508, the processing unit 200 obtains prediction results from the physical model 32. By inputting the time to be predicted (20 years later) into equation (1) of the physical model 32, the processing unit 200 obtains the prediction result of the physical quantity (thickness) at the time to be predicted as a probability distribution. Specifically, first, the processing unit 200 discretizes the probability density distribution of the oxide film thickness L, which is a parameter with an unknown value, into a uniform distribution at equal intervals. The points represented when discretized are called particles. The processing unit 200 then processes these particles within a specified range. Substitute the values into equation (1) of physical model 32, which includes O2 and provides normally distributed random numbers within the given range, to obtain the probability distribution W of the pipe wall thickness when you want to predict it. p(Prediction result) is estimated. Probability distribution W of pipe wall thickness according to physical model 32. p The (prediction results) show variability, as shown in Figure 6, 610.
[0086] Next, in step S510, the processing unit 200 substitutes the time to be predicted (20 years later) into the statistical model 34, and the probability distribution W of the pipe wall thickness at the time to be predicted is obtained. s (Prediction result) can be estimated. In this embodiment, the probability distribution W of pipe wall thickness according to the statistical model 34 s The (prediction results) follow a normal distribution.
[0087] The prediction results from the physical model 32 shown at 610 in Figure 6 show more variability than the prediction results from the statistical model 34 shown at 730 in Figure 7.
[0088] Returning to Figure 5, in step S512, the processing unit 200 processes the predicted result W obtained by the statistical model 34. s And the prediction result W obtained from physical model 32. P The physical model 32 is updated by resampling to bring the two values closer together. Specifically, in step S512, the processing unit 200 performs resampling using a particle filter that selects particles with a high likelihood with a high probability based on the likelihood. The particle filter is also called a particle filter, Monte Carlo filter, or bootstrap filter.
[0089] More specifically, in the resampling process of step S512, the processing unit 200 processes the probability distribution W of the pipe wall thickness output from the physical model 32. p (Prediction result) and the probability distribution W of pipe wall thickness output from the statistical model. sTo bring the (predicted result) closer to the actual result, particles with a high likelihood are selected with a high probability. In resampling, for example, the particles selected by resampling are kept, and the unselected particles with a low likelihood are deleted. The selected particles are discretized to generate new particles. From the generated particles, particles with a higher likelihood are selected again, and particles with a lower likelihood are deleted. In resampling, the selection of particles is repeated. This results in the probability distribution W of wall thickness obtained from the physical model 32. p (Prediction result) and the probability distribution W of wall thickness obtained from the statistical model. s The process is carried out to bring the (predicted result) closer to the actual result. In other words, locations with a high probability density, i.e., locations where the actual wall thickness is likely to exist, can be updated. Increasing the number of times particle selection is repeated will update the probability distribution W of the pipe wall thickness in the physical model 32. p And the probability distribution W of pipe wall thickness in statistical model 34 s This brings us even closer. In step S512, after repeatedly selecting and discarding particles, the probability distribution W of the thickness after resampling is obtained. res The following is obtained: Probability distribution W of pipe wall thickness obtained from physical model 32 p And the probability distribution W of pipe wall thickness obtained from statistical model 34 s As these two values approach each other, if they coincide, the probability distribution W of the wall thickness obtained in step S512 is obtained. res The process can be terminated with the final predicted result for the physical quantity.
[0090] On the other hand, the probability distribution W of wall thickness obtained from the physical model 32 as a result of the resampling process in step S512 p (Prediction result) is the probability distribution W of wall thickness obtained from the statistical model. s If the result approaches the (prediction result) but does not match, the process proceeds to step S514.
[0091] In step S514, the processing device 200 determines that the probability distribution of the wall thickness is W resThe oxide film thickness L, a physical parameter whose value is unknown, is determined from equation (1) of physical model 32. The oxide film thickness L is treated as a probability distribution. In step S514, the processing apparatus 200 updates equation (1) of physical model 32 with the determined oxide film thickness distribution.
[0092] Figure 8 shows the probability density distribution 810 of the thickness L (mm) of the oxide film before resampling, according to one embodiment of the present disclosure, and the distribution when resampling is repeated multiple times (10 times in this example). This figure shows the probability density distribution of the oxide film thickness L (mm) in 820 data points. Resampling reveals that the probability distribution of the oxide film thickness L approaches a normal distribution with a mean of approximately 0.0001 (mm).
[0093] Returning to Figure 5, in step S516, the processing unit 200 multiplies the prediction result obtained from the physical model 32 updated in step S514 with the prediction result obtained from the statistical model 34 obtained in step S512. In this disclosure, multiplication refers to the probability distribution W of the wall thickness at a certain point in time obtained from the updated physical model 32. p_update The probability density value at a given time and the probability distribution W of the wall thickness at a given time obtained from statistical model 34. s This refers to the process of multiplying the probability density value in by the probability distribution W of the wall thickness obtained by multiplying the prediction results. c This is obtained as the estimated final prediction result (probability distribution). In other words, by multiplying the results, it is possible to accurately estimate the probability distribution of the likelihood of the physical quantity, wall thickness, existing at a given point in time.
[0094] Figure 9 shows the concept of multiplication according to one embodiment of the present disclosure. Figure 910 shows the probability density function of wall thickness obtained from the physical model 32 at year x, and Figure 920 shows the probability density function of wall thickness obtained from the statistical model 34 at year x. Figure 930 shows the probability density function of wall thickness at time x, obtained by multiplying the values of the probability density functions (probability density of wall thickness) of both models at year x. The probability distribution W of wall thickness obtained by multiplication.c This will be considered the estimated final prediction result.
[0095] According to the embodiments of this disclosure, even when factors influencing the temporal changes of phenomena such as environmental conditions and operating conditions exist, the prediction accuracy is improved when influencing factors change over time in the same plant by referring to the predicted wall thickness values obtained by inputting them into the physical model. Furthermore, if there are unknown model parameters among the model parameters input into the physical model, such as when actual data does not exist, the probability distribution can be estimated by referring to the predicted wall thickness values output from the statistical model. Moreover, by treating the model parameters of the physical model as probability distributions rather than fixed values, it becomes possible to make predictions that take into account individual differences in components, measurement errors, and variations due to changes in environmental and operating conditions included in the actual data of the physical quantity to be predicted, thereby ensuring consistency between the prediction results and the actual data. As a result, it is possible to quickly identify areas with a high risk of wall thinning, evaluate the remaining life of high-risk areas, and revise and optimize the pipe wall thickness measurement plan and maintenance plan. In addition, this can contribute to shortening the periodic inspection period of the plant, preventing leakage events due to wall thinning, and thereby improving the plant's operating rate.
[0096] <Second Embodiment> Figure 10 shows a conceptual diagram of a model structure 100 for predicting a physical quantity to be predicted, according to one embodiment of the present disclosure. In this embodiment, for the sake of explanation, the conditions necessary for predicting the wall thickness of a pipe are simplified. In the following example, the model structure 100 is constructed under the same conditions as in the first embodiment. The second embodiment differs from the first embodiment in that both the oxygen concentration and the oxide film thickness are unknown, while only the oxide film thickness is unknown.
[0097] In this embodiment, both the oxide film thickness L (a first physical parameter with an unknown value) and the oxygen concentration O2 (a second physical parameter with an unknown value) are unknown. First, in Stage 2, the probability distribution of the oxygen concentration O2, which is the second physical parameter, is estimated (the first prediction result). Next, the processing device 200 updates the physical model 32(1) from Stage 1 with the estimated probability distribution of the oxygen concentration O2. The processing device 200 also constructs the physical model 32(1) from Stage 1 by assigning a uniform distribution to the oxide film thickness L (the first physical parameter) whose value is unknown. In Stage 1, the probability distribution of the wall thickness of the pipe to be predicted (the final prediction result) is obtained based on the updated physical model 32(1) and the statistical model 34(1). The details of the process are as described in the first embodiment, and will not be explained here.
[0098] First, we will explain specific examples of the physical model 32(2) and the statistical model 34(2) in Stage 2.
[0099] (Physical Model 32(2)) In this embodiment, the target piping (component) is assumed to be exposed to exhaust gas generated when a certain fuel is burned, and the physical model 32(2) is an oxygen concentration calculation model that represents the oxygen concentration contained in the exhaust gas generated when a unit mass of fuel is burned at a certain air ratio m.
[0100] The oxygen concentration O2 in the combustion exhaust gas when m≧1 is expressed by the following equation (3). O2 = (21 × (m-1) × A0) / (G 0W +(m-1)×A0) Equation (3)
[0101] Here, O2 is the oxygen concentration (v %), G 0W The theoretical exhaust gas volume (m³) is the amount of exhaust gas produced when a unit mass of fuel is burned at an air-to-air ratio of 1. 3 (NTP) / kg), A0 is the theoretical amount of air (m³) required when burning a unit mass of air at an air-to-air ratio of 1. 3 (NTP) / kg), where m is the air ratio.
[0102] When equation (3) is converted to molar concentration (mol / m3) in volume V (m3), it is expressed as follows: equation (3'). O2=(21×(m-1)×A0) / (G0W+(m-1)×A0)× 0.01 / 0.0224 Formula (3')
[0103] The oxygen concentration calculation model (physical model 32(2)) is ultimately expressed as shown in equation (4) below by dividing the numerator and denominator of the right-hand side of equation (3') by A0. O2=(21×(m-1)) / (R0+(m-1))×0.01 / 0.0224 Formula (4) Here, R0 is the ratio of the theoretical exhaust gas volume G0W to the theoretical air volume A0 = G0W / A0. Here, the theoretical exhaust gas volume G0W is the amount of exhaust gas emitted when a certain fuel is completely combusted, assuming that the air consists only of nitrogen and oxygen. The theoretical air volume A0 is the minimum amount of air required, determined from the chemical changes that occur when combustible components such as carbon, hydrogen, and sulfur contained in a unit amount of fuel are completely combusted.
[0104] In this second embodiment, the oxygen concentration O2 is the output of physical model 32(2). Also, the theoretical exhaust gas volume G 0W The ratio R0 of the theoretical air volume A0 is assumed to be a physical parameter whose probability distribution is unknown. The air ratio m is assumed to be a physical parameter whose probability distribution is known.
[0105] Figure 12 shows the theoretical exhaust gas volume G according to one embodiment of the present disclosure. 0W (m 3 (NTP) / kg) and theoretical air volume A0(m³ 3 The distribution of (NTP) / kg and air ratio m is shown. Figure 1202 represents the theoretical exhaust gas volume G. 0W The probability distribution (mean 10(m) 3 (NTP) / kg), standard deviation 0.1(m 3 (NTP) / kg) is a normal distribution, 1204 is the probability distribution of the theoretical air volume A0 (mean 5 (m 3 (NTP) / kg), standard deviation 0.05(m3 The normal distribution of (NTP) / kg)1206 shows the probability distribution of air ratio m (normal distribution with mean 1.6 and standard deviation 0.02). The physical model 32(2) does not take into account the operating conditions of each plant, and the prediction results from the physical model 32(2) may differ from the actual values.
[0106] (Statistical Model 34(2)) Statistical model 34(2) is a model that statistically predicts the oxygen concentration O2 in the exhaust gas to which the target piping (component) is exposed. Based on database 220, statistical model 34(2) statistically analyzes the trends of the data and expresses the oxygen concentration O2 as a function of the air ratio m. This is possible. Considering the relationship between the air ratio m and the oxygen concentration O2 in equation (4), the statistical model 34(2) can be expressed as shown in equation (5) below.
[0107] O2=c×(m-1) / (m+b) Equation (5) Here, c is a constant (=9.375), and b is a statistical parameter whose value is unknown.
[0108] Figure 11 shows an overview of the processing flow 500 performed by the processing device 200 to predict a physical quantity (oxygen concentration O2 in this embodiment) in the model structure 100 shown in Figure 10. Each process performed by the processing device 200 will be described below.
[0109] In step S1102, the processing device 200 obtains measurement data of the oxygen concentration O2 during incineration, which is a physical quantity, from the database 220.
[0110] In step S1104, the processing unit 200 obtains a physical model 32(2) which is formulated based on physical laws for the oxygen concentration O2 (second physical parameter), and the physical parameters that serve as input to the physical model 32(2), and constructs the physical model 32(2) based on these. Specifically, the processing unit 200 adds the theoretical exhaust gas amount G to equation (4) which represents the physical model 32(2). 0W (m 3(NTP) / kg) and theoretical air volume A0(m³ 3 Given the ratio R0 of (NTP) / kg and the air ratio m, we construct physical model 32(1).
[0111] Next, in step S1106, the processing device 200 constructs a statistical model 34(2) that probabilistically represents the trend of the second physical parameter, the oxygen concentration O2 in the exhaust gas.
[0112] In this embodiment, the distribution of the statistical parameter b is statistically estimated by nonlinear regression analysis using the least squares method. The statistical parameter b may also be estimated using other statistical methods such as maximum likelihood estimation, maximum posterior probability estimation (MAP estimation), Bayesian estimation methods such as Markov chain Monte Carlo (MCMC) method and Hamiltonian Monte Carlo method. The processing unit 200 calculates the mean and standard deviation of b, assuming that the probability distribution of the statistical parameter b follows a normal distribution. The processing unit 200 also calculates the mean and standard deviation of b, assuming that the probability distribution of b follows a normal distribution. Furthermore, the processing unit 200 generates multiple (e.g., 1000) random numbers following a normal distribution with a statistically estimated mean (0.992 in this example) and standard deviation (0.00308 in this example) for b, and substitutes each of these into equation (5) of statistical model 34(2) to construct statistical model 34(2) (regression equation).
[0113] In step S1108, the processing device 200 obtains the predicted result of the oxygen concentration O2 during incineration, which is the second physical parameter, from the physical model 32(2) of stage 2. 2_p1 Obtain it.
[0114] In step S1110, the processing device 200 obtains the predicted result of the second physical parameter, the oxygen concentration O2 during incineration, from the statistical model 34(2) of stage 2. 2_s1 Obtain it.
[0115] In step S1112, the processing unit 200 performs a resampling process so that the prediction result obtained from the statistical model 34(2) in stage 2 and the prediction result obtained from the physical model 32(2) in stage 2 become closer, and the probability distribution of the second physical parameter (oxygen concentration O2) after resampling is obtained when the probability distribution of the oxygen concentration O2 obtained from the physical model 32(2) and the probability distribution of the oxygen concentration O2 obtained from the statistical model 34(2) become closer. 2_res1 We seek.
[0116] Figure 12 shows the distribution of the probability density of oxygen concentration O2 before resampling and the distribution of the probability density of oxygen concentration O2 when resampling is repeated multiple times, according to one embodiment of the present disclosure. Cloth O 2_res1 This is a diagram showing 1208.
[0117] Returning to Figure 11, in step S1114, the processing apparatus 200 calculates the probability distribution of the estimated oxygen concentration O2. 2_res1 Then update equation (4) of physical model 32(2).
[0118] In step S1116, the processing unit 200 obtains the prediction result O from the physical model 32(2) updated in step S1114. 2_p_update1 And the prediction result O obtained from the statistical model 34(2) obtained in step S1112. 2_s1 By multiplying these two factors, we obtain the predicted result for the second physical parameter (oxygen concentration O2) (the first predicted result).
[0119] In step S1118, the processing device 200 calculates the probability distribution W of the second physical parameter (oxygen concentration O2) obtained as a result of the multiplication. c1 This result is substituted into the second physical parameter of the Stage 1 physical model 32(1) as the prediction result of the second physical parameter (first prediction result). The subsequent processing in Stage 1 is as shown in Figure 5.
[0120] <Other Embodiments> Two embodiments of the present invention have been described above. However, the present invention can be used by appropriately setting the number of stages, the physical model and statistical model for each stage, and their parameters, depending on the process and physical quantity to be predicted.
[0121] For example, when predicting the thinning of boiler heat transfer tubes in a waste-to-energy plant, the thinning rate is mainly governed by the properties of corrosion products formed on the surface of the heat transfer tubes and their formation / desorption rates. Here, the properties and formation rate of corrosion products are mainly governed by the concentration of corrosive gases in the outer layer of corrosion products and the diffusion phenomenon of said corrosive gases in the corrosion product layer. Furthermore, the concentration of the corrosive gases is mainly governed by the properties of the adhering ash that adheres to or condenses on the surface of the heat transfer tubes or the outer layer of corrosion products, specifically the concentration of chlorides and sulfides in the adhering ash and the reaction of corrosive gas formation due to the catalytic action of these chlorides and sulfides. Furthermore, the concentration of chlorides and sulfides in the adhering ash is mainly governed by the exhaust gas temperature around the heat transfer tubes and the concentration of chlorides and sulfides in the exhaust gas, and the resulting adhesion or condensation phenomenon of chlorides and sulfides on the surface of the heat transfer tubes.
[0122] Based on the above phenomena, the following configuration can be adopted as an example of the number of stages to set, the physical model of each stage, and its parameters. • Thinning model based on the growth / detachment and diffusion phenomena of corrosion products (Stage 1) • Chemical reaction model for the generation of corrosive gases (Stage 2) • Model of attachment / condensation of attached ash (chloride and sulfide) (Stage 3)
[0123] (Physical model in Stage 1) In this embodiment, the physical model in Stage 1 can be a diffusion model that represents the properties of corrosion products generated on the surface of the heat transfer tube, their generation / desorption phenomena, and the diffusion phenomenon of corrosive gases in the outer layer of the corrosion products within the corrosion product layer.
[0124] Furthermore, the physical parameters set for the above physical model may include the heat transfer tube surface temperature, exhaust gas temperature, the concentration of corrosive gases in the outer layer of the corrosion product (e.g., concentrations of oxygen, hydrogen chloride, chlorine, sulfur oxides, hydrogen sulfide, etc.), the diffusion coefficient of the corrosive gases in the corrosion product, and the density of the heat transfer tube base material and the corrosion product. With such a configuration, it is possible to predict the thickness of the corrosion product at any location and the change in the wall thickness of the heat transfer tube at any location over time (final predicted result).
[0125] Furthermore, there may be two or more types of corrosion products generated, and the diffusion coefficient of the corrosive gas in the corrosion products and the density of the corrosion products may be separated for each type of corrosion product. .
[0126] Furthermore, in waste-to-energy plant boilers, operating procedures are performed to periodically remove ash adhering to the heat transfer tube surface using devices such as soot blowers. In addition, due to fluctuations in operating temperature and other factors, there may be an upper limit to the thickness of corrosion products formed on the heat transfer tube surface. In such cases, including the maximum thickness of corrosion products as a physical parameter allows for more accurate prediction.
[0127] (Statistical model in Stage 1) On the other hand, a piecewise linear model can be used as a statistical model in Stage 1. In the case of a boiler in a waste-to-energy plant, the rate of wall thinning of heat transfer tubes may increase after a latency period of several months to several years following the start of operation. In such cases, by adopting a piecewise linear model that represents the change (increase) in wall thinning rate after a certain period of time, the change in wall thickness of the heat transfer tubes over time can be accurately represented. When adopting a piecewise linear model as the statistical model, the parameters of the statistical model can include the elapsed time until the increase in wall thinning rate occurs, and the wall thinning rate before and after the increase in wall thinning rate. Furthermore, the parameters of the statistical model may also include the initial wall thickness, which is the wall thickness of the material at the time of manufacture, and the measurement error of the wall thickness that occurs in the measurement result. This makes it possible to accurately predict the wall thinning trend by taking into account the variability in the initial wall thickness of the material and the effects of measurement errors of the measuring instrument.
[0128] In the configuration of Stage 1 described above, if there is one or fewer physical parameters with unknown values, the first model structure of the present invention can be adopted. For example, the concentration of corrosive gases in the outer layer of corrosion products (e.g., concentrations of oxygen, hydrogen chloride, chlorine, sulfur oxides, hydrogen sulfide, etc.) is an important parameter that governs the phenomenon in the physical model of Stage 1, but it is difficult to measure its concentration in actual equipment during operation, and its value is often unknown. In this case, by adopting the first model structure, the change in wall thickness of the heat transfer tube over time can be accurately predicted by referring to the prediction results from the physical model and the prediction results from the statistical model, and performing resampling or multiplication.
[0129] On the other hand, in addition to the above-described configuration of Stage 1, the following Stage 2 can be added to create a configuration in which the corrosive gas concentration, which is a physical parameter whose value is unknown in Stage 1, is predicted in Stage 2.
[0130] (Physical model in Stage 2) As a physical model for Stage 2, a thermodynamic equilibrium model can be adopted, which expresses the phenomena related to the reactions that generate corrosive gases in the outer layer of the corrosion product (e.g., 4HCl + O2 → 2Cl2 + 2H2O, 2NaCl + O2 + SO2 → Cl2 + Na2SO4, 2KCl + O2 + SO2 → Cl2 + K2SO4, PbCl2 + O2 + SO2 → Cl2 + PbSO4, etc.) based on thermodynamic equilibrium theory.
[0131] Furthermore, the physical parameters set for the above physical model may include the surface temperature of the ash adhering to the heat transfer tube surface or the outer layer of corrosion products at any given location, the exhaust gas concentration (e.g., HCl, O2, SO2), the properties of the condensed adhering ash, specifically the concentrations of chlorides (e.g., NaCl, KCl, PbCl2) and sulfides (Na2SO4, K2SO4, PbSO4) in the adhering ash, the equilibrium constant, and the equilibrium conversion rate. With such a configuration, it is possible to predict the concentration of corrosive gases (e.g., Cl2) at any given location.
[0132] (Statistical model in Stage 2) As a statistical model in Stage 2, for example, a polynomial approximation model can be adopted in which the exhaust gas concentration and the chloride and sulfide concentrations in the attached ash are the explanatory variables, and the concentration of the generated corrosive gas is the dependent variable. Here, in the boiler of a waste-to-energy plant, a certain exhaust gas The concentration of corrosive gases generated can vary depending on the gas concentration and the chloride and sulfide concentrations in the attached ash. In such cases, adopting a polynomial approximation model constructed based on experimental and actual measurement results can accurately represent the changes in corrosive gas concentration due to exhaust gas concentration and chloride and sulfide concentrations in the attached ash. When adopting a polynomial approximation model as a statistical model, the parameters of the statistical model can be the exhaust gas concentration and the chloride and sulfide concentrations in the attached ash.
[0133] In the configuration of Stage 2 described above, if there is one or fewer physical parameters with unknown values, the second model structure of the present invention can be adopted. For example, the properties of the condensing adhering ash (e.g., the concentration of chlorides and sulfides in the adhering ash) are important parameters that govern the phenomenon in the physical model of Stage 2, but it is difficult to precisely measure the concentration at all parts of an actual facility in operation, and the value is often unknown. In this case, by adopting the second model structure, the concentration of corrosive gas at any part can be accurately predicted by referring to the prediction results from the physical model and the prediction results from the statistical model, and performing resampling or multiplication.
[0134] On the other hand, in addition to the above-described configuration of Stage 2, a Stage 3 can be added to create a configuration in which the properties of the condensing adhering ash, which are physical parameters whose values are unknown in Stage 2, are predicted in Stage 3.
[0135] (Physical model in Stage 3) As a physical model for Stage 3, a heat transfer model and a mass transfer model can be adopted, which express the phenomenon of chlorides and sulfides adhering to or condensing on the surface of heat transfer tubes at any given location, based on the theories of heat transfer and mass transfer phenomena.
[0136] Furthermore, the physical parameters set for the above physical model can include, in addition to the exhaust gas temperature, exhaust gas flow rate, exhaust gas thermal characteristics (e.g., thermal conductivity, specific heat, viscosity coefficient, heat transfer coefficient) at any given location, the heat transfer tube surface temperature, ash surface temperature, thickness of the deposited ash, the amount of fly ash at the boiler inlet and its properties (e.g., the content of each element constituting the chloride or sulfide of interest), the thermal conductivity in the ash, the boundary film mass transfer coefficient, the diffusion constant of the vapor of each chloride in the air, and the saturated vapor concentration of each chloride at any given exhaust gas temperature. With such a configuration, it is possible to predict the properties of the ash deposited on the heat transfer tube surface at any given location, specifically the concentration (condensation amount) of condensed chlorides and sulfides.
[0137] (Statistical model in Stage 3) As a statistical model in Stage 3, for example, a rectangular model or trapezoidal model can be adopted, which represents the amount (condensation rate) of chlorides and sulfides condensed with respect to exhaust gas temperature as a rectangular or trapezoidal polyline. In the case of a waste-to-energy plant boiler, the amount of chlorides and sulfides condensed may increase in a certain exhaust gas temperature range. In such cases, by adopting a rectangular model or trapezoidal model that represents the change (increase) in the amount of chlorides and sulfides condensed in a certain exhaust gas temperature range, the amount of chlorides and sulfides condensed, which changes with exhaust gas temperature, can be accurately represented. When adopting a rectangular model or trapezoidal model as a statistical model, the parameters of the statistical model can include information on multiple points that define the shape of the rectangular or trapezoidal polyline (a combination of the exhaust gas temperature at a point where the amount of condensation changes significantly and the condensation rate of chlorides and sulfides at that exhaust gas temperature) on a coordinate plane with exhaust gas temperature on the horizontal axis and the amount (condensation rate) of chlorides and sulfides condensed on the vertical axis.
[0138] In the configuration of Stage 3 described above, if there is one or fewer physical parameters with unknown values, the third model structure of the present invention can be adopted. For example, the amount of fly ash at the boiler inlet and its properties (e.g., the content of each element constituting the chloride or sulfide of interest) can be used in this Stage 3 configuration. In the physical model of D3, this is an important parameter that governs the phenomenon, but it is difficult to measure its properties precisely at all parts of an actual facility in operation, and its value is often unknown. In this case, by adopting a third model structure, the prediction results from the physical model and the prediction results from the statistical model can be referenced, and by resampling or multiplying them, the properties of ash adhering to the surface of heat transfer tubes at any given location, specifically the concentration of condensed chlorides and sulfides, can be accurately predicted. Yeah
[0139] While embodiments of the present invention have been described above, the embodiments of the invention described above are for the purpose of facilitating understanding of the present invention and do not limit it. The present invention can be modified and improved without departing from its spirit, and of course, the present invention includes equivalents thereof. Furthermore, any combination of embodiments and modifications is possible to the extent that at least a part of the above-mentioned problems can be solved or at least a part of the effects can be achieved, and any combination or omission of each component described in the claims and specification is possible. [Explanation of Symbols]
[0140] 10: Waste-to-energy plant 32(1), (2), (3): Physical models 34(1), (2), (3): Statistical models 40: Model Structure 100: Model Structure 200: Processing unit 220: Database 240: Sensor O2: Probability distribution of oxygen concentration L: Probability distribution of oxide film thickness W: Thick W0: Initial wall thickness W c Probability distribution of wall thickness obtained by multiplication W c1 Probability distribution of wall thickness obtained by multiplication W p : Probability distribution of wall thickness obtained by physical model W s : Probability distribution of wall thickness obtained by statistical model W res : Probability distribution of wall thickness obtained by resampling W res1 : Probability distribution of wall thickness obtained by resampling
Claims
1. A method performed by a processor, the method comprising a first set of steps performed by the processor, the first set of steps is Steps include obtaining measurements of physical quantities that change over time, The steps include: constructing a physical model in which the aforementioned physical quantities are formulated according to physical laws, The steps include constructing a statistical model that expresses the physical quantity as a time function by statistically analyzing the trend of the measured values, The steps include obtaining the prediction result of the physical quantity from the aforementioned statistical model, The steps include obtaining the prediction result of the physical quantity from the physical model, A method comprising the step of statistically estimating the final prediction result of the physical quantity based on the prediction result of the physical quantity calculated by the physical model and the prediction result of the physical quantity calculated by the statistical model.
2. The method according to claim 1, wherein the prediction result obtained from the statistical model and the prediction result obtained from the physical model are prediction results at the same prediction time.
3. The method according to claim 1, wherein the step of estimating the final prediction result includes multiplying the prediction result of the physical quantity calculated by the physical model by the prediction result of the physical quantity calculated by the statistical model.
4. The aforementioned multiplication step is, The method according to claim 3, further comprising the step of multiplying the probability density value corresponding to the prediction result of the physical quantity calculated by the physical model by the probability density value corresponding to the prediction result of the physical quantity calculated by the statistical model.
5. The method according to any one of claims 1 to 4, wherein the physical parameters of the physical model include a first physical parameter whose value is unknown.
6. The step of estimating the final prediction result further includes: The method according to claim 5, further comprising the step of resampling the probability distribution of the first physical parameter so that the prediction result of the physical quantity calculated by the physical model approaches the prediction result of the physical quantity calculated by the statistical model.
7. The step of performing the resampling process is as follows: The steps include discretizing the first physical parameter into particles, substituting each of the discretized particles into the physical model, and calculating the prediction result of the physical quantity. The process of selecting high-likelihood particles from among the particles, generating new particles based on the selected particles, and selecting high-likelihood particles from the generated particles, is repeated, thereby bringing the predicted result of the physical quantity calculated by the physical model closer to the predicted result of the physical quantity calculated by the statistical model. The method according to claim 6, including the method described in claim 6.
8. If the physical parameters of the physical model include a second physical parameter whose value is unknown, the method includes a second set of steps performed by the processor, the second set of steps is: The steps include: estimating the prediction result of the second physical parameter, The steps include updating the physical model with the predicted results of the estimated second physical parameter, The method according to claim 7, including the method described in claim 7.
9. An information processing apparatus comprising a storage unit for storing instructions and a processor, wherein the processor includes a first group of steps executed by the processor by executing instructions stored in the storage unit, and the first group of steps is Steps include obtaining measurements of physical quantities that change over time, The steps include: constructing a physical model in which the aforementioned physical quantities are formulated according to physical laws, The steps include constructing a statistical model that expresses the physical quantity as a time function by statistically analyzing the trend of the measured values, The steps include obtaining the prediction result of the physical quantity from the aforementioned statistical model, The steps include obtaining the prediction result of the physical quantity from the physical model, An information processing apparatus comprising the step of statistically estimating the final prediction result of the physical quantity based on the prediction result of the physical quantity calculated by the physical model and the prediction result of the physical quantity calculated by the statistical model.
10. A non-temporary recording medium on which a program executed by the processor of an information processing device is recorded, The processor executes the first set of steps of the program, Obtain measurements of physical quantities that change over time. A physical model is constructed in which the aforementioned physical quantities are formulated according to physical laws. A statistical model is constructed that expresses the aforementioned physical quantity as a time function by statistically analyzing the trend of the measured values. From the aforementioned statistical model, the prediction results of the physical quantity are obtained, From the aforementioned physical model, the prediction results of the physical quantities are obtained, A non-temporary storage medium that statistically estimates the final predicted result of a physical quantity based on the predicted result of the physical quantity calculated by the physical model and the predicted result of the physical quantity calculated by the statistical model.
Citation Information
Patent Citations
Evaluation system for pipe thinning management
JP1991289596A
Facility evaluation device, facility evaluation method and computer program
JP2016177682A
Acoustic model optimization device and computer program therefor
JP2018013722A
Information processing device
JP2021060292A
Industrial Plant Optimization
JP2023505591A
