Plant anomaly detection system, plant anomaly detection method and plant anomaly detection program

Abstract

To eliminate the need for storage of past normal data and anomalous data and to detect anomalies even in newly designed instruments or facilities.SOLUTION: A plant anomaly detection system 1 includes one or more computers 2 that function such that a probability density distribution of a parameter affecting a specific behavior in a plant is input to an analysis model for predicting a behavior of the plant and the analysis model is executed to calculate an occurrence probability, which is a probability of occurrence of the specific behavior in the plant and that are configured to: receive a monitoring value actually measured by a measuring instrument 8 provided in the plan; calculate an occurrence probability, which is a probability of occurrence of the monitoring value, from the probability density distribution; and detect an anomaly in the plant, based on the calculated occurrence probability of the monitoring value.SELECTED DRAWING: Figure 9

Description

[Technical Field]

[0001] FIELD An embodiment of the present invention relates to a plant abnormality detection technology. [Background technology]

[0002] A widely known method is to monitor the operational status of facilities such as power plants and chemical plants, and when unusual behavior is observed, to detect and report it as an anomaly. This method, called outlier detection, detects behavior that deviates from the measured data. A widely used outlier detection method evaluates outliers based on statistical probability theory. This method estimates the probability density distribution of monitored values ​​under normal conditions, and from this probability density distribution, estimates the probability of the observed monitored value occurring, and then quantifies the degree of anomaly based on the smallness of that probability. Furthermore, this method makes it possible to distinguish between normal and abnormal monitored values ​​by setting a threshold.

[0003] For example, there is the Hotelling T2 method, a classic method described in Non-Patent Document 1. This method calculates the mean value and covariance matrix of a multivariate normal distribution in advance from past combinations of normal monitored values, calculates the distance between these and a newly obtained combination of monitored values, and issues an alarm if the distance exceeds a threshold. This method is applied to applications such as monitoring semiconductor manufacturing processes.

[0004] Furthermore, with the evolution of recent machine learning methods, it is now possible to create libraries without assuming that the distribution of monitored values ​​under normal conditions is a multivariate normal distribution, as in Hotelling's T2 method. For example, Patent Document 1 proposes a technology for distinguishing normal data from abnormal data through machine learning by learning a latent variable model and a joint probability model in advance.

[0005] Meanwhile, among anomaly diagnosis methods that link the feature quantities of monitoring data with the tendency of anomaly causes, there is a method of reproducing anomalies through simulation, as in Patent Document 2. This method assumes an abnormal mode, provides its response through simulation, and evaluates the degree of conformance with the measured value. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Patent No. 6740247 [Patent Document 2] Patent No. 5768834 [Non-patent literature]

[0007] [Non-Patent Document 1] Tsuyoshi Ide and Masaru Sugiyama, "Anomaly Detection and Change Detection," Kodansha, 2015 Summary of the Invention [Problem to be solved by the invention]

[0008] Anomaly detection technology has been developed as an anomaly diagnosis technology that uses measured data to diagnose whether equipment is within the normal operating range or in an abnormal state. This technology collects large amounts of past normal data, analyzes its characteristics, and detects any measured data that deviates from this range as an anomaly. Anomaly detection technology has been developing in line with recent advances in machine learning, but in principle, it utilizes the property that observed values ​​with a high occurrence probability from past data have a low degree of anomaly, while observed values ​​with a low occurrence probability have a high degree of anomaly. Therefore, it is premised on evaluating the occurrence probability of observed values ​​based on accumulated past data. For example, with newly launched production equipment, there is insufficient accumulated observed values, making it impossible to properly evaluate the occurrence probability.

[0009] In addition, even in situations where observed values ​​change due to gradual changes in equipment characteristics, such as aging, detection becomes difficult because the probability of observed values ​​changing gradually as data accumulates. In particular, it is necessary to distinguish gradual changes in monitored values ​​due to aging from changes due to abnormalities.

[0010] Patent Document 1 points out that when an event is a rare event and multiple factors overlap, it is extremely difficult to set models, rules, or thresholds, and therefore a system with greater generalizability and adaptability is required. In other words, when the phenomena handled by devices and facilities are nonlinear, there is no guarantee that the probability density distribution of monitored values ​​will be normal, and it is necessary to set a reasonable probability density distribution.

[0011] The anomaly diagnosis method using an analytical model shown in Patent Document 2 can only handle anomalies that are predicted in advance. It is also difficult to detect anomalies under conditions where multiple factors overlap or when each factor has gradually deteriorated over time.

[0012] The embodiments of the present invention have been made in consideration of these circumstances, and aim to eliminate the need to store past normal and abnormal data, and to detect abnormalities even in newly designed devices or equipment. [Means for solving the problem]

[0013] A plant anomaly detection system according to an embodiment of the present invention includes one or more computers configured to: input a probability density distribution of parameters that affect specific behavior in a plant into an analytical model that predicts the behavior of the plant; execute the analytical model to calculate an occurrence probability, which is the probability that the specific behavior will occur in the plant; receive monitoring values ​​actually measured by measuring instruments installed in the plant; calculate the occurrence probability, which is the probability that the monitoring value will occur, from the probability density distribution; and detect an anomaly in the plant based on the calculated occurrence probability of the monitoring value. [Effects of the Invention]

[0014] According to the embodiment of the present invention, it is not necessary to store past normal data and abnormal data, and abnormalities can be detected even in newly designed devices or facilities. [Brief explanation of the drawings]

[0015] [Figure 1] FIG. 1 is a block diagram showing a plant anomaly detection system. [Figure 2] Block diagram showing the analytical model. [Figure 3] Block diagram showing the analytical model and model parameters. [Figure 4] FIG. 2 is a block diagram illustrating how model parameters are input into an analytical model. [Figure 5] 10 is a graph showing a probability density distribution assuming uncertainty in the monitored values. [Figure 6] Graph showing a uniform probability density distribution assuming uncertainty in monitored values. [Figure 7] 10 is a graph showing a probability density distribution assuming uncertainty in monitored values ​​taking into account changes over time. [Figure 8] 10 is a flowchart showing a process of calculating a probability density distribution of a monitoring value. [Figure 9] 10 is a flowchart showing an abnormality detection process. [Figure 10] Block diagram showing analytical model and measurements. [Figure 11] Graph showing the process of obtaining sample points for a pipe length. [Figure 12] Graph showing the process of obtaining sample points for the inside diameter of a pipe. [Figure 13] 6 is a graph showing the process of acquiring sample points for the pipe surface properties. [Figure 14] FIG. 10 is an explanatory diagram showing a process of performing analysis at sample points and obtaining analysis result points. [Figure 15] Histogram showing the process of estimating the probability density distribution of tank pressure. [Figure 16] 10 is a histogram showing a process of estimating the probability density distribution of the tank liquid transfer temperature. [Figure 17] 10 is a histogram showing the process of estimating the probability density distribution of the tank vapor temperature. [Figure 18] 10 is a graph showing a state in which an abnormality in tank pressure is detected. [Figure 19] 10 is a graph showing a state in which an abnormality in the tank liquid transfer temperature is detected. [Figure 20] 10 is a graph showing a state in which an abnormality in the tank gas phase temperature is detected. DETAILED DESCRIPTION OF THE INVENTION

[0016] Hereinafter, embodiments of a plant anomaly detection system, a plant anomaly detection method, and a plant anomaly detection program will be described in detail with reference to the drawings.

[0017] 1, reference numeral 1 denotes a plant anomaly detection system according to this embodiment. This plant anomaly detection system 1 is used to carry out a plant anomaly detection method.

[0018] The plant anomaly detection system 1 detects anomalies in a plant based on measuring instrument data obtained from the plant to be diagnosed. The plant may be, for example, a nuclear power plant, a thermal power plant, a factory facility, or a production facility. Note that detecting an anomaly also includes detecting its precursor.

[0019] 1, the plant anomaly detection system 1 includes a computer 2. The computer 2 includes a processing circuit 3, a storage unit 4, a communication unit 5, an input unit 6, and an output unit 7. The computer 2 is a device that executes various programs to realize software-based information processing using hardware resources. Furthermore, the plant anomaly detection method of this embodiment is realized by having the computer 2 execute the various programs.

[0020] It should be noted that each component of the plant anomaly detection system 1 does not necessarily have to be provided in one computer 2. For example, one plant anomaly detection system 1 may be realized by a plurality of computers 2 connected to each other via a network. For example, a function for storing an analytical model and a function for executing the same may each be installed in separate computers 2.

[0021] The processing circuit 3 is a circuit including, for example, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), or a dedicated or general-purpose processor. This processor realizes various functions by executing various programs stored in the storage unit 4. The processing circuit 3 may also be configured with hardware such as an FPGA (Field Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit). Various functions can also be realized by such hardware. The processing circuit 3 can also realize various functions by combining software processing by a processor and a program with hardware processing.

[0022] The storage unit 4 stores a predetermined program executed by the processing circuit 3. The storage unit 4 also stores various types of information required when executing the plant anomaly detection method.

[0023] The communication unit 5 receives measuring instrument data including monitoring values ​​measured by a plurality of measuring instruments 8 provided in various pieces of equipment that make up the plant. The computer 2 detects (determines) the presence or absence of an abnormality in the plant based on the various monitoring values ​​included in this measuring instrument data. The measuring instruments 8 of the equipment to be diagnosed are, for example, sensors such as pressure gauges or thermometers.

[0024] Furthermore, the communication unit 5 communicates with other information processing devices via a communication line such as the Internet. In this embodiment, the computer 2 and the other information processing devices are connected to each other via the Internet, but other aspects are also possible. For example, the computer 2 and the other information processing devices may be connected to each other via a LAN (Local Area Network), a WAN (Wide Area Network), or a mobile communication network. Furthermore, each device may be connected to each other via a bus.

[0025] Predetermined information is input to the input unit 6 in response to operations by a user of the computer 2. The input unit 6 includes input devices such as a mouse, keyboard, and touch panel. In other words, predetermined information is input to the computer 2 in response to operations of these input devices.

[0026] The output unit 7 outputs predetermined information. The output unit 7 includes a device for displaying images, such as a display that outputs the analysis results. The display may be separate from the main body of the computer 2, or may be integrated with it. Additionally or alternatively, a display provided on another computer 2 connected via a network may output the analysis results.

[0027] In this embodiment, a display is exemplified as the output unit 7, but other aspects may also be used. For example, the output unit 7 may be a head-mounted display or a projector. Furthermore, the output unit 7 may be a printer that prints information on a paper medium.

[0028] The computer 2 of this embodiment has a function of inputting the probability density distribution of parameters that affect specific behavior in the plant into an analytical model that predicts the behavior of the plant. Furthermore, the computer 2 has a function of calculating the occurrence probability, which is the probability that specific behavior will occur in the plant, by executing the analytical model. The computer 2 then receives monitored values ​​actually measured by measuring instruments 8 installed in the plant, calculates the occurrence probability, which is the probability that the monitored value will occur, from the probability density distribution, and detects an abnormality in the plant based on the calculated occurrence probability of the monitored value.

[0029] Figure 2 shows an example of a group of analytical models for a power plant that uses steam as a heat medium. The analytical models analyze the behavior of the plant based on a physical model that reproduces the physical state of the plant.

[0030] The analytical models include, for example, an inflow boundary model 11, a piping model 12, a valve model 13, a tank model 14, and an outflow boundary model 15. Each analytical model is a representative example, and other models that return a nonlinear response to an input can also be applied.

[0031] For example, the analytical model may be a 1DCAE model developed to evaluate transient behavior such as power plant startup or low-power operation. However, analytical models do not need to be limited to one dimension and may be two- or three-dimensional models.

[0032] The 1DCAE model provides a response to parameters by solving a model based on empirical rules, empirical formulas, or a physical model. In addition to boundary conditions, it is possible to input physical information and parameters for each analytical model.

[0033] For example, in the analysis model, steam flows from an inflow boundary model 11 through a piping model 12. The flow rate of the steam is controlled by a valve model 13, and the steam is stored in a tank model 14 before leaving through an outflow boundary model 15.

[0034] 3 shows input parameters for a simplified analytical model, and the physical model and control model 21 set upstream of the model. The physical model includes, for example, pressure drop models 16, 19, and 23 and heat transfer models 17, 20, and 24.

[0035] The control model 21 is a behavior prediction model that reproduces the operation data of the plant. In other words, the analytical model analyzes the behavior of the plant based on a combination of the behavior prediction model and a physical model.

[0036] A pressure drop model 16 and a heat transfer model 17 are connected upstream of the piping model 12, and are configured to provide a pressure drop coefficient and heat flux to the piping. Model parameters 18 for these physical models are input from further upstream. The model parameters 18 are not a single parameter, but a parameter set that includes multiple parameters.

[0037] Like the piping model 12, the valve model 13 is connected to a pressure drop model 19, a heat transfer model 20, and a control model 21 on the upstream side, which provide the valve opening. Like the piping model 12, the tank model 14 is connected to a pressure drop model 23 and a heat transfer model 24. Furthermore, model parameters 22 and 25 of these physical models are input from further upstream.

[0038] Furthermore, a set of model parameters 26, 27, 28, 29, and 30 for controlling each physical model is input. These are used to input shape information or physical property information for each physical model. For example, model parameters 27 input to piping model 12 include the pipe diameter, pipe length, pipe wall thickness, number of pipe cells, initial pressure, initial temperature, and initial flow velocity.

[0039] FIG. 4 shows the flow of inputting a parameter set into the pressure drop model 16 and the heat transfer model 17. The input model parameters 18 include a plurality of parameters. For example, the pipe length, pipe inner diameter, pipe inner surface properties, pipe wall thickness, and pipe material thermal conductivity are set. The pressure drop coefficient and pipe heat conduction coefficient of the piping model 12 (FIG. 3) are determined, and these are input into the pressure drop model 16 and the heat transfer model 17.

[0040] The flow of anomaly detection will be described with reference to the flowcharts of Figures 8 and 9. Note that the aforementioned drawings may also be referenced. The following steps are at least a part of the processing included in the flow of anomaly detection, and other steps may also be included in the flow of anomaly detection.

[0041] 8 shows the process of estimating the probability density distribution of the monitored values. First, in step S1, the computer 2 (FIG. 1) creates an analysis model based on drawings or specifications of the equipment to be diagnosed.

[0042] In the next step S2, the computer 2 extracts model parameters that affect the monitored values ​​from the constructed analytical model. The model parameters are physical quantities used in the calculations of the analytical model. Examples of physical quantities include the length, inner diameter, surface properties, and heat transfer coefficient of the pipe.

[0043] In the next step S3, the computer 2 assumes a probability density distribution for each model parameter by assuming uncertainty based on at least one of the specifications and the device characteristics. Here, for model parameters that are not expected to change or deteriorate over time, the probability density distribution can be assumed from the device specifications or the device characteristics.

[0044] In addition, when taking into consideration the deterioration over time of the equipment to be diagnosed, the computer 2 assumes a probability density distribution of the model parameters based on at least one of the deterioration mode and the design tolerance for the model parameters when they are assumed to change or deteriorate.

[0045] Here, the design tolerance is the limit value that can be tolerated as a normal value. Specifically, by assuming the corrosion mode of the pipe, it is possible to determine the minimum pipe wall thickness from the viewpoint of pressure resistance, and it is also possible to assume the maximum pipe inner diameter that can be tolerated under normal conditions.

[0046] In the next step S4, the computer 2 acquires sample points P (FIGS. 11 to 13) according to the probability density distribution of the assumed model parameters. Here, the sample points P are points for a plurality of model parameters (physical quantities).

[0047] In the next step S5, the computer 2 executes the analytical model.

[0048] In the next step S6, the computer 2 determines whether the probability density distribution of the analytical value at the sample point P, which is a value corresponding to the monitored value, has converged. If the probability density distribution of the analytical value has converged (YES in step S6), the process proceeds to step S7. On the other hand, if the probability density distribution of the analytical value has not converged (NO in step S6), the process returns to step S4.

[0049] That is, the computer 2 acquires the analytical value at the sample point P by executing the analytical model. Then, the computer 2 moves the sample point P until the probability density distribution of the analytical value converges, and gradually converges the accumulation of the histogram of the monitoring values, thereby making it possible to acquire the probability density distribution of the analytical value.

[0050] In step S7, which is reached if the answer is YES in step S6, the computer 2 estimates the probability density distribution of the monitoring values ​​from the histogram of the monitoring values ​​obtained from the analytical model, and then the process is completed.

[0051] 9 shows a process for detecting an abnormality in a plant. First, in step S11, the computer 2 (FIG. 1) acquires monitored values ​​from the measuring instruments 8 (FIG. 1) provided in the equipment to be monitored in the plant to be diagnosed.

[0052] In the next step S12, the computer 2 estimates the occurrence probability of the monitoring value acquired from the measuring instrument 8 based on the probability density distribution of the monitoring value estimated in advance.

[0053] In the next step S13, the computer 2 determines whether the occurrence probability of the monitored value is equal to or greater than a preset value. If the occurrence probability of the monitored value is equal to or greater than the preset value (YES in step S13), the process returns to step S11. On the other hand, if the occurrence probability of the monitored value is less than the preset value (NO in step S13), that is, if the occurrence probability of the monitored value falls below the set value, the process proceeds to step S14. In other words, the computer 2 determines that an abnormality has occurred if the occurrence probability of the monitored value is below the preset value.

[0054] In the next step S14, the computer 2 outputs a notification of abnormality detection, and then the process is completed.

[0055] Next, we will explain the procedure for estimating the probability density distribution of a model parameter using the pressure loss coefficient as an example. The computer 2 (Fig. 1) estimates the probability density distribution based on at least one of the following values: a design value for which the occurrence probability of a monitored value is preset, a control value for managing the plant, and a design tolerance value that is allowable for the plant design.

[0056] The pressure loss coefficient is determined by the pipe length, pipe diameter, and pipe inner surface properties. The pipe length is assumed to have a design value (or control value), and uncertainty exists due to actual construction. In the following explanation, the design value may also be the control value.

[0057] The uncertainty is the length that can be tolerated in design (management), and can be determined by referring to information such as the specifications of the piping construction drawings. In addition to the methods mentioned above, the uncertainty can also be determined by measuring the actual piping length. For example, the probability density distribution can be estimated from the design tolerance.

[0058] Figure 5 shows an example of a probability density distribution of pipe length as a monitored value. It is assumed that the shape of the probability density distribution is a normal distribution. The probability density distribution can be revised by setting the median of the normal distribution as the design value 40 and the standard deviation as the design tolerance lower limit 41 and design tolerance upper limit 42.

[0059] Similarly, when the monitored value is the pipe diameter, when the monitored value is the pipe inner surface properties, or when the monitored value is the pipe surface roughness, a normal distribution is assumed, the median is the design value 40, and a distribution is assumed in which the standard deviation is the design allowable lower limit value 41 and the design allowable upper limit value 42.

[0060] Note that using the standard deviation of a normal distribution as the design tolerance is just one example, and the design tolerance can also be set to a certain interval within which the normal distribution exists, for example, the upper or lower limit of the 93% interval.

[0061] Also, as shown in Figure 6, it is possible to assume a uniform distribution, set that distribution as the design value 43, and assume a uniform probability density distribution between the design allowable lower limit value 44 and the design allowable upper limit value 45 as the design allowable value. Furthermore, it is possible to set it by assuming a distribution shape such as a beta distribution. Of course, it is also possible to select an appropriate distribution based on the phenomenon, including aging degradation.

[0062] The above explanation assumes that there is uncertainty in the design values ​​due to construction or manufacturing, but there are cases where equipment changes and deteriorates over time. For example, this explains how to assume a probability density distribution when pipe diameter thinning is considered a monitored value. Assume the maximum pipe inner diameter that is acceptable as normal even if the pipe thins.

[0063] The maximum allowable normal pipe inner diameter can be theoretically calculated from the pressure resistance of the pipe wall thickness and the corresponding safety factor. In addition to thinning, the conditions under which deposits accumulate inside the pipe and cause blockages must also be considered. It is also possible to estimate the minimum allowable pipe inner diameter as a change in pipe diameter due to deposits. The minimum pipe inner diameter can be determined by setting conditions such as allowing blockages of up to 10% of the pipe cross-sectional area.

[0064] The method for determining the minimum pipe inner diameter is not limited to the above-mentioned method, and may be calculated from the upper limit of the pipe flow rate. For example, there is a method for calculating the minimum pipe inner diameter by determining the upper limit of the pressure loss considering the volume state as an orifice.

[0065] As shown in Figure 6, the probability density distribution of the pipe inner diameter as a monitored value can be assumed to be a uniform distribution, with the maximum pipe inner diameter that is normal and acceptable as the upper limit and the minimum pipe inner diameter as the lower limit. Note that the probability density distribution is not limited to a uniform distribution, and can also be assumed to be a normal distribution or a beta distribution.

[0066] Next, we will show a method for estimating the probability density distribution when the surface quality of the inner surface of a pipe changes due to corrosion or deposits, which are the cause of pipe deterioration. When considering the surface quality as a property, as with the inner diameter of a pipe, for example, an allowable surface roughness can be assumed. Specifically, by taking into account the state of rust inside the pipe and assuming that the surface condition is such that pitting corrosion does not occur, the maximum allowable surface roughness can be determined. Furthermore, if we assume that the surface roughness of the pipe will not decrease, we can assume that the probability density distribution follows a chi-square distribution.

[0067] Figure 7 shows the probability density distribution of pipe surface roughness taking into account the effects of corrosion. The probability density distribution can be assumed by setting the design value 46 as the origin and the maximum surface roughness for which the standard deviation is acceptable as the design tolerance upper limit 47. Note that the shape of the probability density distribution is not limited to a chi-square distribution, and a gamma distribution or log-normal distribution can also be assumed. Of course, there are no restrictions on the distribution shape assumed above.

[0068] As shown in Figure 10, after setting the probability density distribution of the parameters, the probability density distribution of the monitored values ​​can be estimated according to the analytical model. Here, the tank pressure 50, tank liquid temperature 51, and tank vapor temperature 52 can be calculated as monitored values. However, the number of monitored values ​​does not need to be limited to three. As an example, the Monte Carlo method may be used. Other methods that can be applied include the Markov chain Monte Carlo method, Latin hypercube, and Sobol' permutation.

[0069] 11, 12, and 13, sample points P can be randomly acquired for parameters that determine the probability density distribution. Here, the sample points P are generated so as to follow the probability density distribution.

[0070] 14, it is possible to estimate the monitoring value when an analysis is performed on a sample point P. An analysis is performed on each sample point P, and an analysis result point R can be obtained.

[0071] As shown in Figures 15, 16, and 17, the computer 2 creates a histogram by collecting multiple analysis result points R. Specifically, the histogram is created by dividing the tank pressure 50 into intervals and counting which intervals the analysis result points R are in. The computer 2 increases the number of sample points P from several thousand to several tens of thousands in order to estimate the probability density distribution of the monitoring points.

[0072] 18, 19, and 20 show an example of an anomaly detection method using the probability density distribution of monitored values. When monitored values ​​are obtained from an actual device, the occurrence probability of the monitored value is estimated from the probability density distribution.

[0073] For example, if the occurrence probability of tank pressure 50 is calculated from the probability density distribution and it is within the 93% interval of the probability density distribution, it can be determined to be normal.

[0074] The interval of the probability density distribution that is determined to be normal may be changed depending on the detection accuracy, and even minor abnormalities can be detected by setting the 66% interval of the probability density distribution. Also, by setting the 99% interval of the probability density distribution, it is possible to detect only clear abnormalities and reduce the influence of outliers or noise in the measurement values.

[0075] Similarly, the occurrence probability of the monitored value for the tank liquid-phase temperature 51 can be estimated, and if it is within the normal range, it can be determined to be normal.

[0076] Furthermore, the occurrence probability of the monitored value for the tank vapor temperature 52 can be estimated from the probability density distribution. Since the occurrence probability is outside the set interval, the computer 2 determines that something is abnormal and notifies the occurrence of an abnormality.

[0077] The anomaly detection method is not limited to the method of evaluating the existence interval of the probability density distribution. For example, a method of determining that a monitored value that is a certain distance or more from the median of the probability density distribution is an abnormal value can be applied.

[0078] In the above-described embodiment, the determination of an arbitrary value (occurrence probability) using a reference value (set value) may be a determination of "whether the arbitrary value is equal to or greater than the reference value." This determination may also be a determination of "whether the arbitrary value exceeds the reference value." This determination may also be a determination of "whether the arbitrary value is equal to or less than the reference value." This determination may also be a determination of "whether the arbitrary value is less than the reference value." The reference value may not be fixed, but may be variable. Therefore, instead of the reference value, a value within a predetermined range may be used to determine whether the arbitrary value falls within the predetermined range. Furthermore, errors occurring in the device may be analyzed in advance, and a predetermined range including the error range centered on the reference value may be used for the determination.

[0079] In the flowcharts of the above-described embodiments, steps are illustrated as being executed serially, but the order of steps is not necessarily fixed, and some steps may be executed in reverse order. Also, some steps may be executed in parallel with other steps.

[0080] The plant anomaly detection system 1 of the above-described embodiment includes a control device, a storage device, an output device, an input device, and a communication interface. Here, the control device includes a highly integrated processor such as a central processing unit (CPU), a graphics processing unit (GPU), a field programmable gate array (FPGA), or a dedicated chip. The storage device includes a read-only memory (ROM), a random access memory (RAM), a hard disk drive (HDD), a solid-state drive (SSD), etc. The output device includes a display panel, a head-mounted display, a projector, a printer, etc. The input device includes a mouse, a keyboard, a touch panel, etc. This plant anomaly detection system 1 can be realized with a hardware configuration using a typical computer 2.

[0081] The program executed by the plant anomaly detection system 1 of the above-described embodiment is provided by being pre-installed in a ROM or the like. Additionally or alternatively, the program is provided by being stored in a non-transitory storage medium readable by the computer 2 as a file in an installable or executable format. This storage medium includes a CD-ROM, a CD-R, a memory card, a DVD, a flexible disk (FD), and the like.

[0082] The program executed by this plant anomaly detection system 1 may be stored in a computer 2 connected to a network such as the Internet and provided by downloading it via the network. That is, the program may be provided from cloud computing resources. Alternatively, a server on the cloud may execute the program, and only the processing results may be provided via the cloud. This system may also be configured by combining separate modules that independently perform the functions of the components and are interconnected via a network or dedicated lines.

[0083] According to the embodiment described above, the occurrence probability, which is the probability that a monitored value will occur, is calculated from the probability density distribution, and an abnormality in the plant is detected based on the calculated occurrence probability of the monitored value. This eliminates the need to store past normal data and abnormal data, and makes it possible to detect abnormalities even in newly designed devices or equipment.

[0084] In particular, even for equipment or facilities that handle nonlinear phenomena, the use of analytical models makes it possible to provide a numerically valid probability density distribution. Furthermore, it is possible to provide a probability density distribution based on the analytical model. In addition, it is possible to predict changes in monitored values ​​that take into account aging of the equipment or facilities from the analytical model, and it is also possible to detect changes in the probability density distribution of monitored values ​​due to aging or deterioration.

[0085] Although several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, modifications, and combinations can be made without departing from the spirit of the invention. These embodiments and modifications thereof are intended to be included within the scope and spirit of the invention, as well as within the scope of the invention described in the claims and their equivalents. [Explanation of symbols]

[0086] 1...Plant anomaly detection system, 2...Computer, 3...Processing circuit, 4...Memory unit, 5...Communication unit, 6...Input unit, 7...Output unit, 8...Measuring instrument, 11...Inflow boundary model, 12...Piping model, 13...Valve model, 14...Tank model, 15...Outflow boundary model, 16...Pressure loss model, 17...Heat transfer model, 18...Model parameters, 19...Pressure loss model, 20...Heat transfer model, 21...Control model, 22...Model parameter, 23...pressure drop model, 24...heat transfer model, 25,26,27,28,29,30...model parameters, 40...design value, 41...design allowable lower limit, 42...design allowable upper limit, 43...design value, 44...design allowable lower limit, 45...design allowable upper limit, 46...design value, 47...design allowable upper limit, 50...tank pressure, 51...tank liquid temperature, 52...tank vapor temperature, P...sample point, R...analysis result point.

Claims

1. a function of inputting a probability density distribution of parameters that affect a specific behavior in a plant into an analytical model that predicts the behavior of the plant, and executing the analytical model to calculate an occurrence probability, which is the probability that the specific behavior will occur in the plant; receiving a monitoring value actually measured by a measuring instrument installed in the plant; calculating the occurrence probability, which is the probability that the monitored value occurs, from the probability density distribution; detecting an abnormality in the plant based on the calculated occurrence probability of the monitored value; one or more computers configured to Plant abnormality detection system.

2. the analytical model analyzes the behavior of the plant based on a physical model that reproduces the physical state of the plant; The plant anomaly detection system according to claim 1 .

3. the analysis model analyzes the behavior of the plant based on a combination of a behavior prediction model that reproduces operation data of the plant and the physical model; The plant abnormality detection system according to claim 2 .

4. determining that an abnormality has occurred when the occurrence probability of the monitoring value is lower than a preset value; 3. The plant anomaly detection system according to claim 1 or 2.

5. The probability density distribution of the occurrence probability of the monitored value is estimated based on at least one of a preset design value, a control value for managing the plant, and a design tolerance value allowed in design of the plant.

3. The plant anomaly detection system according to claim 1 or 2.

6. one or more computers having a function of inputting a probability density distribution of parameters that affect a specific behavior in a plant into an analytical model that predicts the behavior of the plant and executing the analytical model to calculate an occurrence probability, which is the probability that the specific behavior will occur in the plant; receiving a monitoring value actually measured by a measuring instrument installed in the plant; calculating the occurrence probability, which is the probability that the monitored value occurs, from the probability density distribution; detecting an abnormality in the plant based on the calculated occurrence probability of the monitored value; Execute the process, Plant anomaly detection method.

7. a probability density distribution of parameters that affect a specific behavior in a plant is input into an analytical model that predicts the behavior of the plant, and the analytical model is executed to provide one or more computers with a function of calculating an occurrence probability, which is the probability that the specific behavior will occur in the plant; receiving a monitoring value actually measured by a measuring instrument installed in the plant; calculating the occurrence probability, which is the probability that the monitored value occurs, from the probability density distribution; detecting an abnormality in the plant based on the calculated occurrence probability of the monitored value; causing the computer to execute a process; Plant abnormality detection program.

Images