Estimation device, estimation method, and storage medium

Abstract

The estimation device includes: an observation system constraint determination unit that determines constraint data as data within a first constraint based on time-series data of each observation value, including a first observation value observed by a first observation system and a second observation value observed by a second observation system; a model estimation unit that estimates parameters of multiple models, including an observation model serving as a model of each observation system and a physical model serving as a model within a device equipped with each observation system, based on the constraint data; and a matching determination unit that determines the matching of the model based on the deviation between a first predicted observation value predicted from the second observation system based on the estimated parameters and the first observation value.

Description

Technical Field

[0001] This disclosure relates to estimation apparatus, estimation method, and storage medium.

[0002] This application claims priority to Japanese Patent Application No. 2020-130787, filed in Japan on July 31, 2020, the contents of which are incorporated herein by reference. Background Technology

[0003] In equipment such as large refrigeration units and submersible pumps, there are known technologies that monitor the state of the equipment by setting up observation systems such as sensors and measuring instruments.

[0004] For example, Patent Document 1 discloses the installation of sensors in a nuclear power plant.

[0005] Prior art literature

[0006] Patent documents

[0007] Patent Document 1: International Publication No. 2015 / 019499 Summary of the Invention

[0008] -The problem the invention aims to solve-

[0009] The device disclosed in Patent Document 1 includes an environmental sensor for monitoring environmental conditions where sensors are installed, and evaluates the integrity of the sensors.

[0010] However, in cases where the sensor indicates an anomaly regardless of environmental conditions, the matching of the sensor's observations cannot always be evaluated in the determination device disclosed in Patent Document 1.

[0011] Therefore, in the device disclosed in Patent Document 1, it is sometimes impossible to presume the matching of observations.

[0012] This disclosure was made to solve the above-mentioned problems, and its purpose is to provide an estimation device, estimation method and procedure for easily estimating the matching of observations.

[0013] -Methods for solving problems-

[0014] To address the aforementioned issues, the estimation apparatus disclosed herein comprises: an observation system constraint determination unit, which determines constraint data as data within a first constraint based on time-series data of each observation value of a first observation value observed by a first observation system and a second observation value observed by a second observation system; a model estimation unit, which estimates parameters of a plurality of models, including an observation model serving as a model of each observation system and a physical model serving as a model within a device equipped with each observation system, based on the constraint data; and a matching determination unit, which determines the matching of the model based on the deviation between a first predicted observation value predicted from the second observation system based on the estimated parameters and the first observation value.

[0015] The estimation method involved in this disclosure includes the following steps: determining constraint data as data within a first constraint based on time-series data of each observation value of a first observation value observed by a first observation system and a second observation value observed by a second observation system; estimating parameters of a plurality of models, including an observation model as a model of each observation system and a physical model as a model within a device equipped with each observation system, based on the constraint data; and determining the matching of the models based on the deviation between a first predicted observation value predicted from the second observation system based on the estimated parameters and the first observation value.

[0016] The procedure disclosed herein causes the computer of the estimation device to perform the following steps: determining constraint data as data within a first constraint based on time-series data of each observation value of a first observation value observed by a first observation system and a second observation value observed by a second observation system; estimating parameters of a plurality of models, including an observation model as a model of each observation system and a physical model as a model within a device in which each observation system is installed, based on the constraint data; and determining the matching of the models based on the deviation between a first predicted observation value predicted from the second observation system based on the estimated parameters and the first observation value.

[0017] -Invention Effects-

[0018] Based on the estimation apparatus, estimation method and procedure of this disclosure, the matching of observations can be easily estimated. Attached Figure Description

[0019] Figure 1 This is a block diagram of the observation device according to the first embodiment.

[0020] Figure 2 This is a diagram illustrating the function of the observation system constraint determination unit involved in the first embodiment.

[0021] Figure 3This diagram illustrates the function of the model estimation unit involved in the first embodiment.

[0022] Figure 4 This is a diagram illustrating the observation model involved in the first embodiment.

[0023] Figure 5 This is a graph illustrating the full-scale error.

[0024] Figure 6 This is a graph illustrating the error of the indicated value.

[0025] Figure 7 This is a graph illustrating Allen's variance.

[0026] Figure 8 This is a graph illustrating the relationship between Allen variance and time window.

[0027] Figure 9 This diagram illustrates the function of the model estimation unit involved in the first embodiment.

[0028] Figure 10 This diagram illustrates the function of the matching determination unit according to the first embodiment.

[0029] Figure 11 This diagram illustrates the function of the matching determination unit according to the first embodiment.

[0030] Figure 12 This diagram illustrates the function of the matching determination unit according to the first embodiment.

[0031] Figure 13 This diagram illustrates the function of the matching determination unit according to the first embodiment.

[0032] Figure 14 This is a flowchart of the estimation method involved in the first embodiment.

[0033] Figure 15 This is a block diagram of the observation device according to the second embodiment.

[0034] Figure 16 This diagram illustrates the function of the intervention unit involved in the second embodiment.

[0035] Figure 17 This is a flowchart of the observation method involved in the second embodiment.

[0036] Figure 18 This is a block diagram of the observation device according to the third embodiment.

[0037] Figure 19 This diagram illustrates the function of the intervention unit involved in the third embodiment.

[0038] Figure 20 This diagram illustrates an example of the hardware configuration of the computer included in the presumed device according to each embodiment. Detailed Implementation

[0039] The embodiments involved in this disclosure will now be described using the accompanying drawings. In all the drawings, the same or equivalent structures are labeled with the same reference numerals, and common descriptions are omitted.

[0040] <First Implementation>

[0041] (Overall structure)

[0042] Figure 1 This indicates the overall structure of the device 9 involved in the first embodiment.

[0043] Equipment 9 includes equipment 1, estimation device 2, first observation system 3 and second observation system 4.

[0044] Equipment 1 includes, for example, large-scale chillers, submersible pumps, and complete sets of equipment.

[0045] For example, device 1, estimation device 2, first observation system 3 and second observation system 4 can also be connected to each other via wired or wireless means and can communicate with each other.

[0046] (Structure of the first observation system)

[0047] The first observation system 3 is a system for observing the state of the observation device 1.

[0048] The first observation system 3 observes the first observation value OB1.

[0049] For example, the first observation system 3 can also be set in device 1.

[0050] For example, the first observation system 3 may also have a first sensor 31.

[0051] For example, the first observation system 3 can also observe the observation value A1 as the first observation value OB1.

[0052] For example, the first sensor 31 may also be located inside the device 1. In this case, the first sensor 31 may also be a pressure sensor that measures the internal pressure of the device 1, as the observed value A1.

[0053] (Structure of the second observation system)

[0054] The second observation system 4 is a system used to observe the status of the equipment 1.

[0055] The second observation system 4 observes the second observation value OB2.

[0056] For example, the second observation system 4 can also be another observation system that is independent of the first observation system 3.

[0057] For example, the second observation OB2 can also be an observation that is different from the first observation OB1.

[0058] For example, the second observation system 4 can also be set in device 1.

[0059] For example, the second observation system 4 may also have a second sensor 41, a third sensor 42, and a fourth sensor 43.

[0060] For example, the second observation system 4 can also observe the observation values ​​A2, A3, and A4 as the second observation value OB2.

[0061] The second sensor 41 can also be located at the outlet of device 1, for example. In this case, the second sensor 41 can also be a thermometer that measures the temperature of the fluid such as liquid or gas flowing out of device 1, i.e., the outlet temperature, as the observed value A2.

[0062] Alternatively, the third sensor 42 may be installed at the inlet of the device 1. In this case, the third sensor 42 may also be a thermometer that measures the temperature of the fluid such as liquid or gas flowing into the device 1, i.e., the inlet temperature, as the observed value A3.

[0063] Furthermore, the fourth sensor 43 may also be installed in the device 1, for example. In this case, the fourth sensor 43 may also be a flow meter that measures the flow rate of the fluid such as liquid or gas flowing in the device 1, i.e., the flow rate inside the device, as the observed value A4.

[0064] (Structure of the estimating device)

[0065] The estimation device 2 is a device used to estimate the parameters PR of each model of the estimation device 1, the first observation system 3 and the second observation system 4.

[0066] For example, the estimation device 2 can also be a device for carrying out asset management services based on the monitoring of device 1.

[0067] The estimation device 2 includes an observation system constraint determination unit 22, a model estimation unit 23, and a matching determination unit 24.

[0068] For example, the estimation device 2 may also include a physical quantity constraint determination unit 25 and a model constraint determination unit 26.

[0069] In addition, the estimation device 2 may also include an acquisition unit 21 and an output unit 27.

[0070] (Structure of the acquisition section)

[0071] The first observation OB1 was obtained from section 21.

[0072] For example, the acquisition unit 21 can also acquire the observation value A1 from the first observation system 3 as the first observation value OB1.

[0073] For example, the acquisition unit 21 can also acquire the internal pressure of the device 1 measured by the first sensor 31 as the observation value A1.

[0074] The second observation OB2 was obtained from section 21.

[0075] For example, the acquisition unit 21 can also acquire the observation values ​​A2, A3, and A4 from the second observation system 4 as the second observation value OB2.

[0076] For example, the acquisition unit 21 can also acquire the outlet temperature measured by the second sensor 41 as the observation value A2.

[0077] For example, the acquisition unit 21 can also acquire the inlet temperature measured by the third sensor 42 as the observation value A3.

[0078] For example, the acquisition unit 21 can also acquire the flow rate inside the device measured by the fourth sensor 43 as the observation value A4.

[0079] (Structure of the observation system's constraint determination unit)

[0080] The observation system constraint determination unit 22 determines the data within the first constraint R1, i.e. the constraint data DT2, based on the time series data DT1 of each observation value OB1 observed by the first observation system 3 and the second observation value OB2 observed by the second observation system 4.

[0081] By determining whether each observation system unit is suitable for the constraints, the observation system constraint determination unit 22 determines whether each observation value is data that represents an abnormal action performed by a single unit.

[0082] For example, such as Figure 2 As shown, the observation system constraint determination unit 22 can also determine the data DT2 within the constraint by observation system constraints such as constraints related to the maximum and minimum values, constraints related to the moving average rate of change, constraints related to the frequency of peak generation, and constraints related to the variance within the time window, which are the first constraint R1.

[0083] Therefore, the observation system constraint determination unit 22 determines the records of abnormal values, including those observed as individual observation values, as abnormal based on the measurement principle (physical law) of each observation system itself.

[0084] For example in Figure 2In the case of any observation system, the time series data of the original observations, i.e., the data DT3 in the time series data DT1, is considered as data whose frequency of peak generation of the observations is higher than the specified threshold, or data whose variance of the observations within the time window is higher than the specified threshold, and is therefore determined to be outside the constraint by the first constraint R1.

[0085] Similarly, data DT4 in the time series data DT1, which is either a data whose observation value is greater than the maximum value specified by the upper limit or less than the minimum value specified by the lower limit, is determined to be outside the constraint by the first constraint R1.

[0086] Similarly, data DT5 in the time series data DT1, being a data whose moving average rate of change exceeds a specified threshold, is determined to be outside the constraint by the first constraint R1.

[0087] On the other hand, data that is not outside the constraints is determined to be data within the constraints, DT2.

[0088] For example, if an out-of-constraint determination occurs in any of the observation systems, the observation system constraint determination unit 22 may also determine the data of other observation systems at the same time as out of constraint, and will not perform the subsequent processing in the estimation device 2.

[0089] (Structure of the model estimation section)

[0090] The model estimation unit 23 estimates the parameters PR of multiple models, including the observation model MLO, which serves as a model of each observation system, and the physical model MLP, which serves as a model within the device 1 where the observation system is installed, based on the constraint data DT2.

[0091] For example, the model estimation unit 23 may also include a network model NWM that contains multiple observation models MLO and multiple physical models MLP.

[0092] For example, each model of multiple observation models (MLO) and multiple physical models (MLP) may include parameters (PR).

[0093] For example, model estimation part 23 can also be used as an example of the network model NWM. Figure 3 The parameters PR of each model in the inferred observation model MLO and the physical model MLP in the network model NWM1 shown.

[0094] like Figure 3 As shown, in the network model NWM1, the model estimation unit 23 can estimate one of the multiple physical quantity estimation values ​​PV1 based on the observed value A1 via an observation model MLO.

[0095] On the other hand, in the network model NWM1, the model estimation unit 23 can also estimate the physical quantity estimation value PV1 based on the observation values ​​A2, A3 and A4, through multiple observation models MLO and multiple physical models MLP.

[0096] That is, in the network model NWM1, the model estimation unit 23 can estimate the common physical quantity estimation value PV1 from different systems.

[0097] In addition, the physical quantity estimate PV is a physical quantity within device 1 estimated based on parameter PR.

[0098] like Figure 3 As shown, for example, in the model estimation unit 23, the parameter PR of the observation model MLO related to the observation value A2 can also be estimated based on the observation value A2 and the observation value A2' before the step of observing the observation value A2 (the observation value A2' observed at the moment before the observation value A2 is about to be observed).

[0099] Similarly, in the model estimation section 23, the parameter PR of the observation model MLO related to the observation value A3 can also be estimated based on the observation value A3 and the observation value A3' before the step of observing the observation value A3 (the observation value A3' observed at the moment before the observation value A3 is about to be observed).

[0100] (Structure of the model)

[0101] For example, the observation model MLO and the physical model MLP can also be represented by nonlinear polynomials based on theoretical and experimental formulas, respectively. In this case, the coefficients of the polynomials are equivalent to the parameter PR, which becomes the parameter used to represent the deviation of the actual device 1.

[0102] Here, the physical model MLP is explained in detail.

[0103] For example, each physical model (MLP) can also be a model derived from physical phenomena based on known pure physical laws.

[0104] For example, Figure 3 The physical model MLPA shown can also be a model that can calculate the physical quantity estimate PVA representing the work done in equipment 1, such as the refrigeration capacity, based on the physical quantity estimate PV2 representing the outlet temperature, the physical quantity estimate PV3 representing the inlet temperature, and the physical quantity estimate PV4 representing the flow rate within the equipment.

[0105] For example, Figure 3 The physical model MLPB shown can also be a model that can calculate the physical quantity PVB representing the fluid saturation temperature of liquids, gases, etc. in equipment 1 based on the physical quantity PV2 representing the outlet temperature and the physical quantity PVA representing the work.

[0106] For example, Figure 3 The physical model MLPC shown can also be a model that can calculate the physical quantity PV1 representing the fluid pressure (saturated vapor pressure) of the liquid, gas, etc. in the device 1 based on the physical quantity PVB representing the saturated temperature of the fluid in the device 1.

[0107] Next, the observation model MLO will be explained in detail.

[0108] For example, each observation model MLO can also be Figure 4 The representative model shown is...

[0109] exist Figure 4 In this context, the error between the sensor's characteristics and the true value is represented as an intrinsic factor, while the error caused by the value of the observed object is represented as an extrinsic factor.

[0110] In addition to the intrinsic and extrinsic factors that can be modeled, the observed values ​​are also affected by random noise components.

[0111] In addition, external factors require any one of the following conditions: "small enough to be negligible", "external factors are measured using other sensors", or "the amount of data that can only be processed for noise".

[0112] In the observation model MLO, the accuracy of the observation system is determined by considering all errors, such as... Figure 5 The full-scale error shown (±0%FS), or Figure 6 The indicated value error (±0%RD) is specified by any one of the indicated values ​​shown, or by a combination of them.

[0113] In addition, these error ranges are always the accuracy under normal operating conditions. If the sensor's installation position is offset or it malfunctions, it will deviate from these ranges.

[0114] Furthermore, when the same value is measured continuously, the magnitude of the deviation between the results of comparing the average values ​​taken over a time window is called the Allen variance.

[0115] Typical Allen variance representation is as follows Figure 7 The variance is shown. Additionally... Figure 7 The σy(f) shown is a value related to the magnitude of the Allen variance.

[0116] like Figure 7 , Figure 8 As shown, during periods with short averaging time τ, the influence of noise is dominant, and the deviation between average values ​​gradually decreases. However, if the averaging time τ is extended, the opposite occurs, and the influence of long-term variations in the observed system parameters becomes significant, with the deviation gradually increasing.

[0117] Therefore, for example, the model estimation unit 23 can ideally estimate the parameter PR of the observed model MLO based on the recorded data with only the Allen variance minimized.

[0118] For example, the model estimation unit 23 can also use the data from the constraint data DT2, which contains no abnormal observations in the observation system constraint determination unit 22, and the data of the number of records required to estimate each parameter PR from the most recent observations. For the estimated value PV of the physical quantity that can be estimated from each system, the model estimation unit 23 accumulates the sum of the average value of the estimated value PV of the physical quantity of all systems and the sum of the squares of the errors of the estimated value PV of the physical quantity of each system in all records. At this time, the model estimation unit 23 can also determine the reward function for the accuracy of the estimation based on the magnitude of the predetermined penalty coefficient, based on the accumulated value and the magnitude of the correction term of each parameter PR, and estimate the parameter PR so that the value is minimized.

[0119] For example, the model estimation unit 23 can also estimate the parameter PR for observation systems that can derive a common physical quantity estimate PV1 by transforming the values ​​of each model based on the network model NWM1, so that the sum of the penalty for the deviation between the physical quantity estimate PV1 estimated from each system and the penalty for the correction term of each parameter PR is minimized.

[0120] Furthermore, even when the number of observation systems is usually very small relative to the number of parameters PR to be estimated, the model estimation unit 23 can estimate the physical quantity PV1 by using multiple records.

[0121] For example, the observation model MLO may not be a faithful model of the observation model, but rather a simplified observation model with appropriate range that takes into account the operating conditions of device 1 and the specifications of each sensor.

[0122] For example, in the observation model MLO associated with the observation A1, the penalty for the correction term for the bias, scaling factor (hereinafter also referred to as "SF") from the ideal state can also be determined based on the sensor accuracy.

[0123] The model estimation unit 23 may also be configured to determine the average of the squared errors of the differences between the average of the multiple physical quantity estimation values ​​PV1 and the estimation values ​​PV1 of each physical quantity estimated from each system in the network model NWM1.

[0124] For example, such as Figure 9As shown, when the estimated value PV of each physical quantity can be estimated from three or more different systems in a network model NWM that includes multiple models, the model estimation unit 23 can also determine the reward function for the average of the squared errors of the difference between the average value of the multiple estimated values ​​PV and the estimated value PV of each physical quantity, as the deviation between the estimated values ​​PV of each physical quantity estimated from three or more systems.

[0125] exist Figure 9 Similarly, in the case shown, the model estimation unit 23 can also estimate the parameter PR so that the sum of the penalty for the deviation between the estimated values ​​PV of the physical quantities estimated from three or more systems and the penalty for the correction term of each parameter PR is minimized.

[0126] (Structure of the matching determination unit)

[0127] The matching determination unit 24 determines the matching of the model based on the deviation between the first predicted observation PA1 and the first observation OB1, wherein the first predicted observation PA1 is based on the parameters PR of multiple models estimated by the model estimation unit 23 and predicted from the second observation OB2.

[0128] That is, the matching determination unit 24 can regress a certain observation from other observations based on the estimated parameter PR, and evaluate the magnitude of the deviation.

[0129] In addition, the first predicted observation PA1 is a value derived by transforming the values ​​of each model based on the network model NWM1, and is the value equivalent to the first observation OB1 in the network model NWM1.

[0130] For example, the matching determination unit 24 can also compare the regression result predicted based on the parameters PR of each model estimated by the model estimation unit 23, i.e., the first predicted observation value PA1, with the first observation value OB1 as the measured value.

[0131] For example, as a model matching, the matching determination unit 24 can also determine whether the deviation between the first predicted observation value PA1 predicted based on the second observation value OB2 and the first observation value OB1 is within the specification range of the first observation system 3.

[0132] For example, such as Figure 10 As shown, as a model matching, the matching determination unit 24 can also determine whether the deviation between the first predicted observation value PA1 predicted based on the observation values ​​A2, A3 and A4 and the observation value A1 is within the specification range of the first observation system 3.

[0133] For example, the matching determination unit 24 may also predict a first predicted observation value PA1 based on the observation value A2, observation value A3 and observation value A4 by importing multiple observation models MLO and multiple physical models MLP that have been estimated by the model estimation unit 23, and compare the predicted first predicted observation value PA1 with the observation value A1.

[0134] For example, the matching determination unit 24 can also obtain the difference between the first predicted observation value PA1 and the first observation value OB1 as the deviation between the first predicted observation value PA1 and the first observation value OB1. In this case, as a determination of the matching of the models, the matching determination unit 24 can also determine that the multiple observation models MLO and multiple physical models MLP with parameter PR are matched if the obtained difference is below a predetermined threshold.

[0135] For example, the matching determination unit 24 can also perform mutual regression to evaluate the matching of all observation systems.

[0136] For example, it can also be in addition to Figure 10 In addition to the evaluations shown, such as Figure 11 As shown, the matching determination unit 24 imports multiple observation models MLO and multiple physical models MLP with parameters PR estimated by the model estimation unit 23, predicts a second predicted observation value PA2 based on the observation value A1, observation value A3 and observation value A4, and further compares the predicted second predicted observation value PA2 with the observation value A2.

[0137] Furthermore, with Figure 10 The evaluation related to the observed value A1 shown below Figure 11 Similarly, the matching determination unit 24 can perform an evaluation related to the observation value A3 based on the observation values ​​A1, A2, and A4, and it can also perform an evaluation related to the observation value A4 based on the observation values ​​A1, A2, and A3.

[0138] For example, such as Figure 12 , Figure 13 As shown, when multiple systems independently exist that regress the same observation value, the matching determination unit 24 can also calculate the deviation of each system that performs the regression, and take the minimum value of the multiple calculated deviations as the evaluation result of the observation value.

[0139] (Structure of the physical quantity constraint determination unit)

[0140] The physical quantity constraint determination unit 25 determines whether the estimated value PV of the physical quantity based on the parameter PR is within the second constraint R2.

[0141] For example, the physical quantity constraint determination unit 25 can also determine whether the physical quantity estimated value PV is within the second constraint R2, wherein the physical quantity estimated value PV is estimated based on the physical model MLP containing the parameter PR when it is determined to be matched in the matching determination unit 24 and the observed value of the constraint data DT2.

[0142] For example, the second constraint R2 can also be a specified numerical range of physical quantities within the design level (specification) of equipment 1.

[0143] For example, if the estimated value PV of the physical quantity is within the second constraint R2, the physical quantity constraint determination unit 25 determines it to be normal; if it is not within the second constraint R2, the physical quantity constraint determination unit 25 determines it to be abnormal. In the case of being determined to be normal, the physical quantity constraint determination unit 25 can consider the parameter PR imported into each model to be appropriate.

[0144] For example, if the physical quantity estimate value PV is determined to be abnormal, the physical quantity constraint determination unit 25 can also determine that the equipment 1 is abnormal.

[0145] That is, if based on the assumption that a physical quantity can be estimated from the constraint data DT2, which is determined to be normal, then if the estimated value PV of the physical quantity is not within the constraints of the design specifications, it can be regarded as an abnormality of equipment 1.

[0146] For example, when the observations of the constraint data DT2 are discrete in time by applying the first constraint R1, the physical quantity constraint determination unit 25 can also perform time interpolation on the discrete observations and estimate the physical quantity estimate value PV based on the observations of the constraint data DT2 after time interpolation.

[0147] For example, the physical quantity constraint determination unit 25 can also determine whether the estimated value PV of the physical quantity estimated in each time step is within the second constraint R2.

[0148] For example, the physical quantity constraint determination unit 25 can also compare the estimated value PV of each physical quantity with the design level (specification) and make a determination such as within the specification, above the specification, or below the specification, thereby converting it into discretized data of about 2 to 7 stages.

[0149] (Structure of the model constraint decision unit)

[0150] The model constraint determination unit 26 determines whether each parameter PR is within the third constraint R3.

[0151] For example, the model constraint determination unit 26 can also determine whether each parameter PR, which is determined to be a match in the matching determination unit 24, is within the third constraint R3.

[0152] For example, the third constraint R3 for each parameter PR in the physical model MLP can also be a specified numerical range of parameters within the design level (specification) of device 1.

[0153] For example, the third constraint R3 for each parameter PR in the observation model MLO can also be a specified numerical range of the parameter within the design level (specification) for each observation.

[0154] For example, if each parameter PR is within the third constraint R3, the model constraint determination unit 26 determines it as normal; if it is not within the third constraint R3, the model constraint determination unit 26 determines it as abnormal. In the case of a normal determination, the model constraint determination unit 26 considers each parameter PR imported into each model to be appropriate.

[0155] For example, the parameters PR determined in the model constraint determination unit 26 can also be parameters estimated by each moving window or batch window that has undergone mutual regression.

[0156] For example, the model constraint determination unit 26 can also compare the estimated parameters PR with the design level (specification) and make a determination such as within the specification, above the specification, or below the specification, thereby converting it into discretized data of about 2 to 7 stages.

[0157] For example, if the parameter PR of the physical model MLP is determined to be abnormal, the model constraint determination unit 26 can also determine that the device 1 is abnormal.

[0158] That is, if based on the assumption that the parameter estimation can be carried out correctly, then if the parameter PR of the physical model MLP is not within the constraints of the design specifications, it can be regarded as an anomaly of device 1.

[0159] For example, if the parameter PR of the observation model MLO is determined to be abnormal, the model constraint determination unit 26 can also determine that the observation system is abnormal. Furthermore, the model constraint determination unit 26 can also determine that the sensor associated with the observation model MLO that has the parameter PR determined to be abnormal is abnormal.

[0160] That is, if based on the assumption that the parameter estimation can be carried out correctly, then if the parameter PR of the observation model MLO is not within the constraints of the design specifications, it can be regarded as an anomaly of the observation system.

[0161] (Structure of the output section)

[0162] The output unit 27 outputs the parameter PR that is determined to be normal and the estimated value of the physical quantity PV based on the parameter PR.

[0163] In addition, the output unit 27 outputs the parameter PR that is determined to be abnormal and the estimated value of the physical quantity PV based on the parameter PR.

[0164] For example, if the parameter PR that is determined to be abnormal is the parameter PR of the observation model MLO, the output unit 27 can also output the meaning of sensor abnormality related to the observation model MLO that has the parameter PR that is determined to be abnormal.

[0165] For example, if the parameter PR that is determined to be abnormal is the parameter PR of the physical model MLP, the output unit 27 can also output the meaning of abnormal device 1.

[0166] (action)

[0167] The operation of the estimation device 2 in this embodiment will be explained.

[0168] The operation of the estimation device 2 is equivalent to the estimation method of this embodiment.

[0169] The operation of the estimation device 2 can also be, for example, as follows: Figure 14 Implement it as shown.

[0170] First, the acquisition unit 21 acquires the first observation value OB1 observed by the first observation system 3 and the second observation value OB2 observed by the second observation system 4 (ST01: acquisition step).

[0171] After ST01 is implemented, the observation system constraint determination unit 22 determines the constraint data DT2 as the data within the first constraint R1 based on the time series data DT1 of each observation value of the first observation value OB1 and the second observation value OB2 (ST02: observation system constraint determination step).

[0172] After ST02 is implemented, the model estimation unit 23 estimates the parameters PR of multiple models, including the observation model MLO, which is the model of each observation system, and the physical model MLP, which is the model in the device 1 where the observation system is installed, based on the constraint data DT2 (ST03: model estimation step).

[0173] After ST03 is implemented, the matching determination unit 24 determines the matching of the model based on the deviation between the first predicted observation PA1 and the first observation OB1 (ST04: matching determination step), wherein the first predicted observation PA1 is based on the parameters PR of multiple models estimated by the model estimation unit 23 and predicted based on the second observation OB2.

[0174] For example, in ST04, the matching determination unit 24 can also determine that there is an anomaly in the observation model or the physical model if the deviation is greater than a specified value (ST04A). In this case, the matching determination unit 24 can also delete records that deviate from the model even after optimization.

[0175] For example, after implementing ST04A, the matching determination unit 24 can also determine the mismatched data in the constraint data DT2 and exclude it from the constraint data DT2 as abnormal data (ST04B).

[0176] For example, after ST04 is implemented, the physical quantity constraint determination unit 25 can also determine whether the estimated value PV of the physical quantity based on the parameter PR is within the second constraint R2 (ST05: physical quantity constraint determination step).

[0177] For example, in parallel with ST05, the model constraint determination unit 26 determines whether each parameter PR is within the third constraint R3 (ST06: model constraint determination step).

[0178] For example, after implementing ST04 and ST05, the output unit 27 can also output the parameter PR that is determined to be normal and the estimated value of the physical quantity PV based on the parameter PR, and output the parameter PR that is determined to be abnormal and the estimated value of the physical quantity PV based on the parameter PR (ST07: output step).

[0179] (Function and effect)

[0180] According to this embodiment, the estimation device 2 can compare observations obtained from different observation systems by estimating the parameters PR of multiple models, including the observation model MLO and the physical model MLP.

[0181] Therefore, estimation device 2 can evaluate the matching of observations.

[0182] Therefore, estimation device 2 can easily estimate the matching of observations.

[0183] Furthermore, according to this embodiment, the matching determination unit 24 determines the matching of the estimated model.

[0184] The model derived by the model estimation unit 23 is a best effort with parameters PR adjusted to minimize bias, etc., and it is unclear whether it is a model that can estimate the correct observations.

[0185] Therefore, in this embodiment, since the matching determination unit 24 determines the matching of the estimated model, it is possible to determine whether the estimated observation is within, for example, the accuracy of the compensation object.

[0186] Furthermore, according to one embodiment of this invention, the estimation device 2 also includes a physical quantity constraint determination unit 25, which can determine whether the physical quantity in the device 1 is normal.

[0187] Therefore, based on the estimation device 2, the user can identify the abnormality of the device 1 itself.

[0188] Furthermore, according to one example of this embodiment, the estimation device 2 also includes a model constraint determination unit 26, which can determine whether each parameter PR is normal.

[0189] Therefore, based on the estimation device 2, the user can identify anomalies in the device 1 itself and anomalies caused by the observation system.

[0190] Furthermore, by estimating the physical quantities inside device 1 that cannot be estimated by ordinary observations alone, as well as the parameters PR of the observation model MLO and the physical model MLP, a more detailed diagnosis of device 1 can be performed.

[0191] As a comparative example, we cite the case where anomalies in the observation system are detected by setting up 2 to 3 sensors for redundant measurements of the same observation values.

[0192] In this situation, multiple sensors need to be installed for the same observations. For new equipment, this increases manufacturing costs, and for existing equipment, it sometimes results in sensor installation costs and the inability to apply services.

[0193] Furthermore, the same method only guarantees that the observed values ​​are not abnormal, and sometimes cannot provide internal state quantities for inferring the cause of the abnormality of the device itself.

[0194] In contrast, the estimation device 2 according to one example of this embodiment can determine the anomaly of the observation system based on different observation values, and therefore can be configured, for example, to not install multiple sensors for the same observation value.

[0195] Furthermore, in one embodiment of this embodiment, the estimation device 2 can provide the internal state of the device 1 itself, so as to identify the abnormalities of the device 1 itself and the abnormalities caused by the observation system.

[0196] Furthermore, according to one example of this embodiment, the model estimation unit 23 calculates the deviations between the estimated values ​​of physical quantities estimated from each of three or more systems in the network model NWM, which includes multiple models, so the estimation device 2 can compare the estimated values ​​of three or more physical quantities with each other.

[0197] Therefore, estimation device 2 is able to estimate a more likely model.

[0198] <Second Implementation>

[0199] The estimation device 2 according to the second embodiment will be described with reference to the accompanying drawings.

[0200] The structure of the estimation device 2 in this embodiment is the same as that in the first embodiment, except for the points described below.

[0201] (structure)

[0202] For example, such as Figure 15 As shown, the estimation device 2 may also include an intervention implementation unit 28.

[0203] like Figure 16 As shown, the intervention unit 28 can change the parameter PR by intervening in the physical model MLP. At this time, the estimation device 2 can use the observed value as input condition to estimate the true value of each physical quantity that affects the performance of the device 1, i.e., the estimated value PV of each physical quantity, based on the intervened physical model MLP.

[0204] For example, the output unit 27 may also output parameter PR, which includes the modified parameter PR, and the estimated value of physical quantity PV estimated based on the modified parameter PR.

[0205] For example, the intervention unit 28 can change the parameter PR based on the parameters input by the user.

[0206] In addition, in this embodiment, the estimation device 2 may or may not include a physical quantity constraint determination unit 25 and a model constraint determination unit 26.

[0207] (action)

[0208] The operation of the estimation device 2 in this embodiment will be explained.

[0209] The operation of the estimation device 2 is equivalent to the estimation method of this embodiment.

[0210] The operation of the estimation device 2 can also be, for example, as follows: Figure 17 Implement it as shown.

[0211] First, the estimation device 2 is implemented in the same way as in the first embodiment, ST01 to ST04.

[0212] For example, after implementing ST04, the intervention unit 28 can change the parameter PR through the intervention physical model MLP (ST11: intervention step). At this time, the output unit 27 can also output the parameter PR including the changed parameter PR and the estimated value PV of the physical quantity estimated based on the changed parameter PR (ST12: input step).

[0213] In addition, in this embodiment, the estimation device 2 may or may not implement ST05 to ST07 of the first embodiment.

[0214] (Function and effect)

[0215] In this embodiment, it has the same function and effect as in the first embodiment.

[0216] Furthermore, according to one example of this embodiment, the estimation device 2 is able to simulate the impact of changes in parameter PR in the physical model MLP on the performance of device 1.

[0217] Therefore, based on the estimation device 2, the user can evaluate the performance of the device 1.

[0218] Furthermore, according to one example of this embodiment, the estimation device 2 can quantitatively calculate the impact on the performance of the device 1 when the parameter PR is improved to a shape close to the specifications by intervention, based on the parameters of the physical model MLP estimated based on the current state of the device 1.

[0219] Therefore, users can determine the appropriate timing for intervention.

[0220] <Third Implementation Method>

[0221] The estimation device 2 according to the third embodiment will be described with reference to the accompanying drawings.

[0222] The structure of the estimation device 2 in this embodiment is the same as that in the second embodiment, except for the points described below.

[0223] (constitute)

[0224] For example, such as Figure 18 As shown, the intervention unit 28 may include a feedback control unit 281.

[0225] like Figure 19 As shown, the feedback control unit 281 feeds back the first observation value OB1 to the physical model MLP to change the parameter PR.

[0226] For example, the feedback control unit 281 can also determine the command value based on the first observation value OB1 and through control logic, and feed it back to the physical model MLP via the observation model MLO to change the parameter PR.

[0227] In addition, in this embodiment, the estimation device 2 may or may not include a physical quantity constraint determination unit 25 and a model constraint determination unit 26.

[0228] (action)

[0229] The operation of the estimation device 2 in this embodiment is equivalent to the estimation method in this embodiment.

[0230] The operation of the estimation device 2 in this embodiment can, for example, be the same as in the second embodiment, such as... Figure 17 Implement it as shown.

[0231] In addition, in this embodiment, the estimation device 2 may or may not implement ST05 to ST07 of the first embodiment.

[0232] (Function and effect)

[0233] In this embodiment, it has the same function and effect as in the second embodiment.

[0234] Furthermore, according to one example of this embodiment, the intervention unit 28 can feed back the first observation value OB1 to the parameter PR in the physical model MLP.

[0235] Therefore, the user can simulate the control parameters used for feedback control of the observation-based device 1.

[0236] Therefore, based on the estimation device 2, the user can evaluate the control parameters.

[0237] Furthermore, according to one example of this embodiment, the estimation device 2 can estimate the internal parameters of the physical model MLP and the observation model MLO based on the current state of the device 1 through the operation up to ST04.

[0238] Therefore, the estimation device 2 can accurately simulate how the observed values ​​represent the response when the input conditions are fed back for control.

[0239] Therefore, based on the estimation device 2, by optimizing control parameters such as PID gain within the control logic through a simulation-based model and the operational performance of device 1, effective efficiency improvement can be achieved.

[0240] <Variation Example>

[0241] In the above embodiments, the model estimation unit 23 includes the network model NWM, but it can be configured arbitrarily as long as the parameter PR can be estimated.

[0242] For example, the model estimation unit 23 can also estimate the parameters PR of the network model NWM stored outside the estimation device 2 by communicating with the outside of the estimation device 2.

[0243] <Computer Hardware Architecture>

[0244] Furthermore, in the above embodiments, programs for implementing various functions of the estimation device 2 are recorded on a computer-readable recording medium. The programs recorded on this recording medium are then read into a computer system such as a microcomputer and executed, thereby performing various processes. Here, the various processing procedures of the computer system's CPU are stored in the form of programs on the computer-readable recording medium. The computer reads and executes these programs to perform the aforementioned various processes. Furthermore, computer-readable recording media include disks, optical disks, CD-ROMs, DVD-ROMs, semiconductor memories, etc. Additionally, the computer program can be distributed to a computer via a communication line, and the receiving computer executes the program.

[0245] In the above embodiments, examples of the hardware structure of the computer that executes the program for implementing the various functions of the estimation device 2 will be described.

[0246] like Figure 20 As shown, the computer 29 of the estimation device 2 includes a CPU 291, a memory 292, a storage / playback device 293, an input / output interface (hereinafter referred to as "IO I / F") 294, and a communication interface (hereinafter referred to as "communication I / F") 295.

[0247] The memory 292 is a medium such as random access memory (hereinafter referred to as "RAM") that temporarily stores data and other data used in the program executed by the presumed device 2.

[0248] Storage / playback device 293 is a device for storing data to or reproducing data from external media such as CD-ROM, DVD, and flash memory.

[0249] IO I / F294 is an interface used for inputting and outputting information, etc., between the estimation device 2 and other devices.

[0250] The communication I / F295 is an interface for communication between the estimated device 2 and other devices via communication lines such as the Internet or dedicated communication lines.

[0251] <Other Implementation Methods>

[0252] The embodiments of this disclosure have been described above, but these embodiments are provided as examples and are not intended to limit the scope of the disclosure. These embodiments can be implemented in various other ways, and various omissions, substitutions, and modifications can be made without departing from the spirit of the disclosure. These embodiments and their variations are included within the scope and spirit of the disclosure.

[0253] <Postscript>

[0254] The estimation device, estimation method, and procedure described in the above embodiments are as follows.

[0255] (1) The estimation device 2 of the first method includes: an observation system constraint determination unit 22, which determines constraint data DT2 as data within the first constraint R1 based on the time series data DT1 of each observation value OB1 observed by the first observation system 3 and the second observation value OB2 observed by the second observation system 4; a model estimation unit 23, which estimates the parameters PR of multiple models including the observation model MLO as the model of each observation system and the physical model MLP as the model in the device 1 where each observation system is installed, based on the constraint data DT2; and a matching determination unit 24, which determines the matching of the model based on the deviation between the first predicted observation value PA1 predicted from the second observation system 4 based on the estimated parameters PR and the first observation value OB1.

[0256] According to this method, the estimation device 2 can compare observations obtained from different observation systems by estimating the parameters PR of multiple models, including the observation model MLO and the physical model MLP.

[0257] Therefore, estimation device 2 can evaluate the matching of observations.

[0258] Therefore, estimation device 2 can easily estimate the matching of observations.

[0259] (2) The estimation device 2 of the second method is based on the estimation device 2 of (1), and it also has a physical quantity constraint determination unit 25, which determines whether the physical quantity estimated in the device 1 based on the parameter PR, namely the physical quantity estimation value PV, is within the second constraint R2.

[0260] According to this method, the estimation device 2 can determine whether the physical quantity in the equipment 1 is normal.

[0261] Therefore, based on the estimation device 2, the user can identify the abnormality of the device 1 itself.

[0262] (3) The third-party estimation device 2, based on the estimation device 2 of (1) or (2), also has: a model constraint determination unit, which determines whether each parameter PR is within the third constraint R3.

[0263] According to this method, the estimation device 2 can determine whether each parameter PR is normal.

[0264] Therefore, based on the estimation device 2, the user can identify anomalies in the device 1 itself and anomalies caused by the observation system.

[0265] (4) The estimation device 2 of the fourth method is based on the estimation device 2 of (1). The model estimation unit 23 calculates the deviation between the physical quantities in the device 1 estimated based on the parameters, namely the physical quantity estimation values ​​PV, from three or more systems in the network model NWM that includes multiple models.

[0266] According to this method, the estimation device 2 can compare the estimated values ​​of three or more physical quantities with each other.

[0267] Therefore, estimation device 2 is able to estimate a more likely model.

[0268] (5) The fifth method estimation device 2, based on any one of the estimation devices 2 in (1) to (4), further includes: an intervention action unit 28, which can change the parameter PR by intervening in the physical model MLP.

[0269] According to this method, the estimation device 2 can simulate the impact of changes in parameter PR in the physical model MLP on the performance of device 1.

[0270] Therefore, based on the estimation device 2, the user can evaluate the performance of the device 1.

[0271] (6) The sixth method of estimation device 2 is based on the estimation device 2 in (5), and the intervention action unit 28 is equipped with: feedback control unit 281, which feeds back the first observation value OB1 to the physical model MLP to change the parameter PR.

[0272] According to this method, the intervention unit 28 can feed back the first observation OB1 to the parameter PR in the physical model MLP.

[0273] Therefore, the user can simulate the control parameters used for feedback control of the observation-based device 1.

[0274] Therefore, based on the estimation device 2, the user can evaluate the control parameters.

[0275] (7) The estimation method of the seventh method includes the following steps: determining the constraint data DT2, which is the data within the first constraint R1, based on the time series data DT1 of each observation value OB1 observed by the first observation system 3 and the second observation value OB2 observed by the second observation system 4; estimating the parameters PR of multiple models, including the observation model MLO, which is the model of each observation system, and the physical model MLP, which is the model within the device 1 equipped with each observation system, based on the constraint data DT2; and determining the model matching based on the deviation between the first predicted observation value PA1 predicted from the second observation system 4 based on the estimated parameters PR and the first observation value OB1.

[0276] According to this method, the estimation method can compare observations obtained from different observation systems by estimating the parameters PR of multiple models, including the observation model MLO and the physical model MLP.

[0277] Therefore, the estimation method can evaluate the matching of observations.

[0278] Therefore, the estimation method can easily estimate the matching of observations.

[0279] (8) The procedure of the eighth method causes the computer of the estimation device 2 to perform the following steps: determine the constraint data DT2, which is the data within the first constraint R1, based on the time series data DT1 of each observation value OB1 observed by the first observation system 3 and the second observation value OB2 observed by the second observation system 4; estimate the parameters PR of multiple models, including the observation model MLO, which is the model of each observation system, and the physical model MLP, which is the model within the device 1 where each observation system is set, based on the constraint data DT2; and determine the model matching based on the deviation between the first predicted observation value PA1 predicted from the second observation system 4 based on the estimated parameters PR and the first observation value OB1.

[0280] According to this method, the estimation device 2, which executes the program, can compare observations obtained from different observation systems by estimating the parameters PR of multiple models, including the observation model MLO and the physical model MLP.

[0281] Therefore, the estimation device 2, which has executed the procedure, is able to evaluate the matching of the observations.

[0282] Therefore, the estimation device 2, which has executed the program, can easily estimate the matching of the observations.

[0283] Industrial availability

[0284] Based on one of the methods described above, it is easy to infer the matching of the observations.

[0285] -Explanation of Figure Markers-

[0286] 1 Equipment

[0287] 2. Estimation device

[0288] 3 First Observation System

[0289] 4 Second Observation System

[0290] 21 Acquisition Department

[0291] 22. Observation System Constraint Determination Department

[0292] 23 Model Estimation Section

[0293] 24. Matching Determination Department

[0294] 25 Physical Quantity Constraint Judgment Department

[0295] 26 Model Constraint Decision Department

[0296] 27 Output Section

[0297] 28. Intervention and Policy Implementation Department

[0298] 29 Computers

[0299] 31 First Sensor

[0300] 41 Second Sensor

[0301] 42 Third Sensor

[0302] 43. Fourth sensor

[0303] 281 Feedback Control Department

[0304] 291 CPU

[0305] 292 Memory

[0306] 293 Storage / Replay Device

[0307] 294 IO I / F

[0308] 295 Communication I / F

[0309] A1 Observation

[0310] A2 Observations

[0311] A2' Observation

[0312] A3 Observations

[0313] A3' Observation

[0314] A4 Observations

[0315] DT1 time series data

[0316] DT2 Constraint Data

[0317] DT3 data

[0318] DT4 data

[0319] DT5 data

[0320] MLO observation model

[0321] MLP physical model

[0322] MLPA physical model

[0323] MLPB physical model

[0324] MLPC physical model

[0325] NWM network model

[0326] NWM1 network model

[0327] OB1 First observation

[0328] OB2 Second observation

[0329] PA1 First Predicted Observation

[0330] PA2 Second Predicted Observation

[0331] PR parameters

[0332] Estimated value of PV physical quantity

[0333] Estimated value of physical quantity PV1

[0334] Estimated value of physical quantity PV2

[0335] Estimated value of physical quantity PV3

[0336] Estimated value of physical quantity PV4

[0337] PVA physical quantity estimated value

[0338] PVB physical quantity estimated value

[0339] R1 First Constraint

[0340] R2 Second Constraint

[0341] R3 Third Constraint

Claims

1. A estimation device, characterized in that, have: The observation system constraint determination unit determines the constraint data that is the data within the first constraint based on the time series data of each observation value of the first observation value observed by the first observation system and the second observation value observed by the second observation system; The model estimation unit estimates parameters of multiple models, including observation models that serve as models of each observation system and physical models that serve as models within devices equipped with each observation system, based on the constraint data. as well as The matching determination unit determines the matching of the model based on the deviation between the first predicted observation value predicted from the second observation system based on the estimated parameters and the first observation value.

2. The estimation device according to claim 1, wherein, The estimation device further includes: a physical quantity constraint determination unit, which determines whether the physical quantity in the device estimated based on the parameters, i.e., the physical quantity estimation value, is within the second constraint.

3. The estimation device according to claim 1 or 2, wherein, The estimation device further includes: a model constraint determination unit, which determines whether each parameter is within the third constraint.

4. The estimation device according to claim 1, wherein, The model estimation unit calculates the deviations between the physical quantities within the device, i.e., the estimated values ​​of the physical quantities, estimated based on the parameters, from three or more systems in the network model that includes the multiple models.

5. The estimating device according to claim 1 or 2, wherein, The estimation device further includes an intervention unit capable of changing the parameters by intervening in the physical model.

6. The estimation device according to claim 5, wherein, The intervention unit includes a feedback control unit that feeds back the first observation value to the physical model to change the parameters.

7. A method of estimation, characterized in that, Includes the following steps: Based on the time series data of the first observation value observed by the first observation system and the second observation value observed by the second observation system, the data within the constraints that are considered as data within the first constraints are determined. Based on the constrained data, parameters of multiple models are estimated, including observation models that serve as models of each observation system and physical models that serve as models within devices equipped with each observation system; and The model's matching ability is determined based on the deviation between the first predicted observation value predicted from the second observation system based on the inferred parameters and the first observation value.

8. A non-transitory computer-readable storage medium for storing a program, characterized in that, The procedure causes the computer of the estimation device to perform the following steps: Based on the time series data of the first observation value observed by the first observation system and the second observation value observed by the second observation system, the data within the constraints that are considered as data within the first constraints are determined. Based on the constrained data, parameters of multiple models are estimated, including observation models that serve as models of each observation system and physical models that serve as models within devices equipped with each observation system. as well as The model's matching ability is determined based on the deviation between the first predicted observation value predicted from the second observation system based on the inferred parameters and the first observation value.

Images