Apparatus and method for verifying validity of a sensor

Abstract

To provide a sensor validation device which can intuitively monitor the validity of a sensor by presenting control limit lines of a correlation between an analysis target sensor and a reference target sensor.SOLUTION: A sensor validation device is configured to: select historical data of an analysis target sensor and historical data of a reference target sensor; optimize initial values of parameters of a Bayesian model and select a degree of a polynomial regression model; infer a posterior distribution of a regression coefficient and an error term of a regression curve representing a relation between the historical data of the analysis target sensor and the historical data of the reference target sensor, thereby setting a credible interval based on validation of posterior distribution of the regression coefficient and the error term of the regression curve representing the relation between the historical data of the analysis target sensor and the historical data of the reference target sensor; set control limit lines of data of the analysis target sensor using the set credible interval; and validate the analysis target sensor based on current data of the analysis target sensor and the set control limit lines.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present disclosure relates to an apparatus and method for verifying the validity of a sensor. [Background technology]

[0002] Generally, a sensor may experience performance degradation or failure due to its limited life span or external impact, etc. If a problem occurs in a sensor, the accuracy of the measurement data decreases, which reduces the reliability of the measurement data, requiring replacement or repair of the sensor.

[0003] Also, a decrease in the reliability of sensor data may mean that the sensor's performance has decreased or a failure has occurred. Therefore, in order to ensure the continuity of equipment equipped with sensors, a process is required to verify whether the sensor itself is valid through real-time monitoring and analysis of sensor data. [Prior art documents] [Patent documents]

[0004] Patent Document 1: Korean Patent Registration No. 10-1866491 Summary of the Invention [Problem to be solved by the invention]

[0005] The present invention aims to provide a sensor validity verification device and a sensor validity verification method that can intuitively monitor the validity of a sensor by presenting control limits of the correlation between an analyzed sensor and a comparison sensor. [Means for solving the problem]

[0006] According to one embodiment of the present invention, a sensor validity verification device includes at least one processor, which optimizes initial values ​​of parameters of a Bayesian model and selects an order of a polynomial regression model based on past data of a sensor to be analyzed and past data of a sensor to be compared, and uses the Bayesian model optimized based on the past data of the sensor to be analyzed and the past data of the sensor to be compared and the polynomial regression model whose order has been selected to set a confidence interval based on a posterior distribution for a regression coefficient and an error term of a regression curve showing a relationship between the past data of the sensor to be analyzed and the past data of the sensor to be compared, sets a control limit line for the sensor data to be analyzed using the set confidence interval, and verifies the validity of the sensor to be analyzed based on the current data of the sensor to be analyzed and the set control limit line.

[0007] The at least one processor may also select the comparison target sensor data using distance correlation based on past data of the analysis target sensor and past data of a plurality of non-analysis target sensors.

[0008] In addition, the at least one processor may set a likelihood function by replacing a prior distribution set based on the multinomial regression model with a posterior distribution of a previous multinomial regression model.

[0009] The at least one processor can also verify, based on a predetermined method, the posterior distribution for the regression coefficients and error terms of the regression curve that show the relationship between the historical data of the analyzed sensor and the historical data of the compared sensor.

[0010] In addition, the at least one processor can set the confidence interval by applying a Highest Posterior Density (HPD) value having a preset percentage based on an expected value of a posterior distribution for the regression coefficient and error term of the regression curve, which indicates the relationship between the past data of the analysis target sensor and the past data of the comparison target sensor, and can set a control limit line of the analysis target sensor data by using boundary values ​​on both sides of the posterior distribution for the regression coefficient and error term of the regression curve, which indicates the relationship between the past data of the analysis target sensor and the past data of the comparison target sensor corresponding to the confidence interval.

[0011] According to another embodiment of the present invention, the method includes optimizing initial values ​​of parameters of a Bayesian model and selecting an order of a polynomial regression model based on past data of the analyzed sensor and past data of the compared sensor; inferring a posterior distribution for regression coefficients and error terms of a regression curve showing a relationship between the past data of the analyzed sensor and the past data of the compared sensor using the optimized Bayesian model and the polynomial regression model having an order selected based on the past data of the analyzed sensor and the past data of the compared sensor; setting a confidence interval based on the posterior distribution for regression coefficients and error terms of the regression curve showing a relationship between the past data of the analyzed sensor and the past data of the compared sensor, and setting a control limit line of the analyzed sensor data using the set confidence interval; and verifying the validity of the analyzed sensor based on the current data of the analyzed sensor and the set control limit line.

[0012] The method for validating the validity of the sensors may further include selecting the sensor data to be compared using a distance correlation based on past data of the sensor to be analyzed and past data of a plurality of non-analyzed sensors.

[0013] In addition, the step of inferring the posterior distribution for the regression coefficients and error terms of the regression curve, which indicates the relationship between the past data of the analysis target sensor and the past data of the comparison target sensor, using the optimized Bayesian model and the polynomial regression model with the selected order may include the steps of: replacing the prior distribution set based on the polynomial regression model with the posterior distribution of a previous model; and setting a likelihood function.

[0014] In addition, the step of inferring the posterior distribution for the regression coefficients and error terms of the regression curve, which indicate the relationship between the past data of the analyzed sensor and the past data of the compared sensor, using the optimized Bayesian model and the polynomial regression model with a selected order may include the step of verifying the posterior distribution for the regression coefficients and error terms of the regression curve, which indicate the relationship between the past data of the analyzed sensor and the past data of the compared sensor, based on a preset method.

[0015] In addition, the step of setting a control limit line of the sensor data to be analyzed using the set confidence interval may include the steps of: setting the confidence interval by applying a Highest Posterior Density (HPD) value having a preset percentage based on an expected value of a posterior distribution for a regression coefficient and an error term of the regression curve, which indicates a relationship between the past data of the sensor to be analyzed and the past data of the sensor to be compared; and setting a control limit line of the sensor data to be analyzed using boundary values ​​on both sides of a posterior distribution for a regression coefficient and an error term of the regression curve, which indicates a relationship between the past data of the sensor to be analyzed and the past data of the sensor to be compared, corresponding to the confidence interval. Effect of the Invention

[0016] The present invention advantageously provides an intuitive means for monitoring the effectiveness of sensors by providing control limits for the correlation between analyte and comparison sensors.

[0017] In addition, there is an advantage that the control limits can be tuned by setting the posterior distribution of the previous model as the prior distribution.

[0018] In addition, there is an advantage that the posterior distribution of the population of the sensor data to be analyzed can be verified. [Brief description of the drawings]

[0019] [Figure 1] 1 illustrates an apparatus for validating a sensor according to one embodiment of the present disclosure. [Diagram 2] 1 is a flow chart illustrating a method for validating a sensor, according to one embodiment of the present disclosure. [Diagram 3] 11 is a flow chart illustrating a method for validating a sensor according to another embodiment of the present disclosure. [Figure 4] 1 is a flow chart illustrating a method for optimizing initial values ​​of parameters of a polynomial regression model by a sensor validation device according to one embodiment of the present disclosure. [Figure 5a] 1 is a diagram showing an example of a safety region, a caution region, and a control limit line according to one embodiment of the present disclosure. [Figure 5b] 1 is a diagram showing an example of a safety region, a caution region, and a control limit line according to one embodiment of the present disclosure. [Figure 5c] 1 is a diagram showing an example of a safety region, a caution region, and a control limit line according to one embodiment of the present disclosure. [Figure 6] 1 is a diagram illustrating a sensor validation technique using a polynomial regression model according to one embodiment of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0020] In order to clearly describe the present invention, the description of parts that are not related to the present invention will be omitted, and the same reference numerals will be used throughout the specification to refer to the same or similar components.

[0021] In addition, throughout the specification, when a part is described as being "connected" to another part, this includes not only the case where the part is "directly connected" to another part, but also the case where the part is "electrically connected" to another part via another element in between. In addition, when a part is described as "comprising" a certain component, this does not mean that the other component is excluded, but that the part can further include the other component, unless otherwise specified.

[0022] Also, when a part is described as being "on" another part, this includes not only that the part is directly on top of the other part, but also that there are other parts intervening between them. In contrast, when a part is described as being "directly on" another part, this means that there are no other parts intervening between them.

[0023] Additionally, terms such as "first", "second" and "third" are used to describe various parts, components, regions, layers and / or sections, but are not limited thereto. These terms are used only to distinguish one part, one component, one region, one layer or one section from another part, one component, one region, one layer or one section. Thus, the "first part", "first component", "first region", "first layer" or "first section" described below may also be described as the "second part", "second component", "second region", "second layer" or "second section" without departing from the scope of the present invention.

[0024] In addition, the terminology used herein is merely for the purpose of describing a particular embodiment and is not intended to limit the present invention. The singular form used herein includes the plural form unless the phrase clearly indicates otherwise. And, the meaning of "comprise" as used in the specification is to embody a particular property, a particular region, a particular integer, a particular step, a particular operation, a particular element, and / or a particular component, and does not exclude the presence or addition of other properties, other regions, other integers, other steps, other operations, other elements, and / or other components.

[0025] Additionally, relative spatial terms, such as "below," "above," and the like, may be used to more easily understand and describe the relationship of one part to other parts as depicted in the drawings. Such terms are intended to include other meanings and operations of the device in use, as well as the intended meaning in the drawings. For example, imagine that the device in the drawings were turned upside down, and a part described as being "below" other parts could also be described as being "above" other parts. Thus, the exemplary term "below" includes both "above" and "below." The device could also be rotated 90 degrees or at other angles, and the relative spatial terms would be interpreted accordingly.

[0026] In addition, all terms, including technical and scientific terms, used herein have the same meaning as commonly understood by a person of ordinary skill in the art to which the present invention belongs. Terms commonly used and defined in dictionaries may additionally be interpreted as having a meaning consistent with the contents disclosed in the specification, and unless otherwise defined, are not interpreted in an ideal or very formal sense.

[0027] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will now be described in detail with reference to the accompanying drawings, in which: FIG. 1 is a block diagram of a semiconductor device according to an embodiment of the present invention;

[0028] FIG. 1 is a diagram showing the configuration of a sensor validity verification device according to an embodiment.

[0029] Referring to FIG. 1, the sensor validation device 100 according to one embodiment includes a processor 110, an input / output interface module 120, and a memory .

[0030] The processor 110, the input / output interface module 120 and the memory 130 included in the sensor validation device 100 are coupled to each other and can transmit data to each other.

[0031] According to various embodiments, the processor 110 may execute programs or commands stored in the memory 130. In this case, the memory 130 may store an operating program (e.g., an OS) for operating the sensor validation device 100.

[0032] Additionally, according to various embodiments, the processor 110 may execute a program for managing information related to the sensor validation device 100 .

[0033] Additionally, according to various embodiments, the processor 110 may execute programs for managing the operation of the sensor validation device 100 .

[0034] Additionally, according to various embodiments, the processor 110 may execute programs for managing the operation of the I / O interface module 120 .

[0035] i) Selection of sensor data to be compared

[0036] According to various embodiments, the processor 110 can obtain historical data from the analyte sensor and historical data from a number of non-analyte sensors through the input / output interface module 120 .

[0037] According to various embodiments, the processor 110 may set a prior distribution for the historical data of the analyzed sensor, where the prior distribution may be, but is not limited to, a prior distribution for regression coefficients and error terms of a regression curve showing a relationship between the historical data of the analyzed sensor and the historical data of a plurality of non-analyzed sensors.

[0038] According to various embodiments, the processor 110 may set a prior distribution for the past data of the non-analyzed sensors, where the prior distribution may be, but is not limited to, a prior distribution for the regression coefficients and error terms of the regression curve that indicates the relationship between the past data of the analyzed sensor and the past data of the non-analyzed sensors.

[0039] In addition, according to various embodiments, the processor 110 can use distance correlation to select data from the past data of the multiple sensors not being analyzed that has a high correlation with the past data of the sensor being analyzed, but the method of selecting data from the past data of the multiple sensors not being analyzed that has a high correlation with the past data of the sensor being analyzed is not limited to this.

[0040] Additionally, according to various embodiments, the processor 110 can select as the comparison sensor data data that is highly correlated with the historical data of the selected analysis target sensor.

[0041] According to various embodiments, the processor 110 may set a prior distribution for the past data of the comparison target sensor, where the prior distribution may be, but is not limited to, a prior distribution for the regression coefficients and error terms of the regression curve that indicates the relationship between the past data of the analysis target sensor and the past data of the comparison target sensor.

[0042] ii) Optimization of the initial parameters of the Bayesian model

[0043] 1. When only the historical data of the analysis target sensor and the comparison target sensor are obtained

[0044] According to various embodiments, the processor 110 can obtain historical data from the analyte sensor and historical data from the comparison sensor through the input / output interface module 120 .

[0045] Additionally, according to various embodiments, the processor 110 can optimize initial values ​​of parameters of the Bayesian model based on historical data of the analyzed sensor, a prior distribution on the historical data of the analyzed sensor, historical data of the comparison sensor, and a prior distribution on the historical data of the comparison sensor.

[0046] Additionally, according to various embodiments, the processor 110 can use the Bayesian model to obtain a posterior distribution for the historical data of the analyzed sensor and the historical data of the comparison sensor based on the historical data of the analyzed sensor, a prior distribution for the historical data of the analyzed sensor, and a prior distribution for the historical data of the comparison sensor and the historical data of the comparison sensor.

[0047] Additionally, according to various embodiments, the processor 110 can train the Bayesian model based on historical data of the analyzed sensor, a prior distribution on the historical data of the analyzed sensor, historical data of the comparison sensor, and a prior distribution on the historical data of the comparison sensor to optimize initial values ​​of parameters of the Bayesian model.

[0048] Additionally, according to various embodiments, the processor 110 can perform sampling from historical data of the analyzed sensor, a prior distribution for the historical data of the analyzed sensor, historical data of the comparison sensor, and a prior distribution for the historical data of the comparison sensor.

[0049] In addition, according to various embodiments, the processor 110 can sample polynomial coefficients and standard deviations corresponding to each data from each of the historical data of the analyzed sensor, the prior distribution for the historical data of the analyzed sensor, the historical data of the compared sensor, and the prior distribution for the historical data of the compared sensor, but sampling from each of the historical data of the analyzed sensor and the historical data of the compared sensor is not limited to this.

[0050] Also, according to various embodiments, the processor 110 may perform sampling from the historical data of the analyzed sensor, the prior distribution for the historical data of the analyzed sensor, the historical data of the comparison sensor, and the prior distribution for the historical data of the comparison sensor between 500 and 1000 times, but the number of samplings is not limited thereto.

[0051] In addition, according to various embodiments, the processor 110 can train the Bayesian model based on learning data generated by sampling from past data of the analyzed sensor, a prior distribution for the past data of the analyzed sensor, past data of the compared sensor, and a prior distribution for the past data of the compared sensor in order to optimize initial values ​​of parameters of the Bayesian model.

[0052] In addition, according to various embodiments, the processor 110 may compare the posterior distribution obtained using the Bayesian model with a preset criterion to verify the learning state of the Bayesian model. In this case, the preset criterion may be, but is not limited to, an acceptance rate and autocorrelation.

[0053] In addition, according to various embodiments, the processor 110 may decide to retrain the Bayesian model if the posterior distribution obtained using the Bayesian model does not satisfy both an acceptance rate (e.g., 0.2<=Acceptance rate<=0.5) and autocorrelation (e.g., 50% of the samples are within a 95% confidence interval).

[0054] In addition, according to various embodiments, the processor 110 can iteratively train the Bayesian model based on new learning data sampled from the past data of the analyzed sensor, the prior distribution for the past data of the analyzed sensor, the past data of the comparison sensor, and the prior distribution for the past data of the comparison sensor, until the posterior distribution obtained using the Bayesian model satisfies both an acceptance rate (e.g., 0.2<=Acceptance rate<=0.5) and autocorrelation (e.g., 50% of the samples are within a 95% confidence interval).

[0055] In addition, according to various embodiments, the processor 110 can stop training the Bayesian model if the posterior distribution obtained using the Bayesian model satisfies both an acceptance rate (e.g., 0.2<=Acceptance rate<=0.5) and autocorrelation (e.g., 50% of the samples are within a 95% confidence interval).

[0056] 2. When the prior distribution for the past data of the analysis target sensor and the past data of the comparison target sensor are also obtained

[0057] According to various embodiments, the processor 110 can obtain the historical data of the analysis target sensor, a prior distribution for the historical data of the analysis target sensor, the historical data of the comparison target sensor, and a prior distribution for the historical data of the comparison target sensor through the input / output interface module 120. Here, the prior distribution can be a prior distribution for the regression coefficients and error terms of the regression curve indicating the relationship between the historical data of the analysis target sensor and the historical data of the comparison target sensor, but is not limited thereto.

[0058] Additionally, according to various embodiments, the processor 110 can optimize initial values ​​of the parameters of the Bayesian model based on historical data of the analyzed sensor, a prior distribution on the historical data of the analyzed sensor, historical data of the comparison sensor, and a prior distribution on the historical data of the comparison sensor.

[0059] Additionally, according to various embodiments, the processor 110 can use the Bayesian model to obtain a posterior distribution for the historical data of the analyzed sensor and the historical data of the comparison sensor based on the historical data of the analyzed sensor, a prior distribution for the historical data of the analyzed sensor, the historical data of the comparison sensor, and a prior distribution for the historical data of the comparison sensor.

[0060] Additionally, according to various embodiments, the processor 110 can train the Bayesian model based on historical data of the analyzed sensor, a prior distribution on the historical data of the analyzed sensor, historical data of the comparison sensor, and a prior distribution on the historical data of the comparison sensor to optimize initial values ​​of parameters of the Bayesian model.

[0061] Additionally, according to various embodiments, the processor 110 can perform sampling from historical data of the analyzed sensor, a prior distribution for the historical data of the analyzed sensor, historical data of the comparison sensor, and a prior distribution for the historical data of the comparison sensor.

[0062] Further, according to various embodiments, the processor 110 can sample polynomial coefficients and standard deviations corresponding to each data from each of the historical data of the analyzed sensor, the prior distribution for the historical data of the analyzed sensor, the historical data of the compared sensor, and the prior distribution for the historical data of the compared sensor, but this is not limited to the fact that sampling can be performed from each of the historical data of the analyzed sensor, the prior distribution for the historical data of the analyzed sensor, the historical data of the compared sensor, and the prior distribution for the compared sensor.

[0063] Also, according to various embodiments, the processor 110 may perform sampling from the historical data of the analyzed sensor, the prior distribution for the historical data of the analyzed sensor, the historical data of the comparison sensor, and the prior distribution for the historical data of the comparison sensor between 500 and 1000 times, but the number of samplings is not limited thereto.

[0064] In addition, according to various embodiments, the processor 110 can train the Bayesian model based on learning data generated by sampling from past data of the analyzed sensor, a prior distribution for the past data of the analyzed sensor, past data of the compared sensor, and a prior distribution for the past data of the compared sensor in order to optimize initial values ​​of parameters of the Bayesian model.

[0065] In addition, according to various embodiments, the processor 110 may compare the posterior distribution obtained using the Bayesian model with a preset criterion to verify the learning state of the Bayesian model. In this case, the preset criterion may be, but is not limited to, an acceptance rate and autocorrelation.

[0066] In addition, according to various embodiments, the processor 110 may decide to retrain the Bayesian model if the posterior distribution obtained using the Bayesian model does not satisfy both an acceptance rate (e.g., 0.2<=Acceptance rate<=0.5) and autocorrelation (e.g., 50% of the samples are within a 95% confidence interval).

[0067] In addition, according to various embodiments, the processor 110 can iteratively train the Bayesian model based on new learning data sampled from the past data of the analyzed sensor, the prior distribution for the past data of the analyzed sensor, the past data of the comparison sensor, and the prior distribution for the past data of the comparison sensor, until the posterior distribution obtained using the Bayesian model satisfies both an acceptance rate (e.g., 0.2<=Acceptance rate<=0.5) and autocorrelation (e.g., 50% of the samples are within a 95% confidence interval).

[0068] In addition, according to various embodiments, the processor 110 can stop training the Bayesian model if the posterior distribution obtained using the Bayesian model satisfies both an acceptance rate (e.g., 0.2<=Acceptance rate<=0.5) and autocorrelation (e.g., 50% of the samples are within a 95% confidence interval).

[0069] iii) Selection of the order of the multinomial regression model

[0070] According to various embodiments, the processor 110 can separate the historical data of the analyzed sensor and the historical data of the comparison sensor into a training data set and a validation data set.

[0071] According to various embodiments, the processor 110 may set a degree range of the polynomial regression to be searched in order to determine the degree of the polynomial corresponding to the following "Equation 1." In this case, the degree range may be one of degrees 1 to 10, but is not limited thereto.

[0072]

number

[0073] Additionally, according to various embodiments, the processor 110 may model a polynomial regression model using training data for each order.

[0074] In addition, according to various embodiments, the processor 110 may calculate a root mean square error (RMSE) of the polynomial regression model modeled with training data for each degree using validation data.

[0075] Also, according to various embodiments, the processor 110 may store a difference between the RMSE of the previous order and the current order if the difference is less than '0'.

[0076] Also, according to various embodiments, the processor 110 may store "0" if the difference in RMSE between the previous order and the current order is greater than "0".

[0077] Additionally, according to various embodiments, the processor 110 can normalize each difference to a sum of all differences.

[0078] According to various embodiments, the processor 110 may select a cumulative sum equal to or greater than a preset threshold as an appropriate order of the polynomial regression model, where the threshold may be, but is not limited to, 0.7.

[0079] iv) Inference of the posterior distribution of the analyzed sensor data

[0080] According to various embodiments, the processor 110 may set a prior distribution of coefficients / standard deviations based on the selected order of the polynomial regression model.

[0081] Additionally, according to various embodiments, the processor 110 can set the regression coefficients of the initial polynomial regression equation to the mean of the prior distribution of the regression coefficients.

[0082] In addition, according to various embodiments, the processor 110 can input the set prior distribution, the historical data of the analyzed sensor, and the historical data of the compared sensor into the optimized Bayesian model, and infer posterior distributions for the regression coefficients and error terms of the regression curve that show the relationship between the historical data of the analyzed sensor and the historical data of the compared sensor.

[0083] Additionally, according to various embodiments, the processor 110 can infer the regression coefficients of the polynomial regression model based on the regression coefficients of the regression curve, which indicates the relationship between the historical data of the analyzed sensor and the historical data of the compared sensor, and the posterior distribution for the error terms.

[0084] v) Tuning control limits

[0085] According to various embodiments, the processor 110 can update the control limits of the correlation due to changes in the conditions (maintenance, seasonal factors, aging, etc.) of the equipment from which the analyzed sensor collects data.

[0086] Additionally, according to various embodiments, the processor 110 can update the posterior distribution of the multinomial regression model to tune the control limits.

[0087] Also, according to various embodiments, the processor 110 may update the posterior distribution of the multinomial regression model by using the posterior distribution of a previous model as a prior distribution and using Bayesian sampling inference for the multinomial regression model. Here, the posterior distribution of the previous model may mean the posterior distribution for the regression coefficients and error terms (standard deviations) of the multinomial regression model that constitute the control limits, but the meaning of the posterior distribution of the previous model is not limited thereto.

[0088] Also, according to various embodiments, the processor 110 may set a likelihood function, which may be a function of normal distribution, but is not limited thereto.

[0089] Additionally, according to various embodiments, the processor 110 can set the likelihood function, which can be, but is not limited to, a function corresponding to past data of the sensor being analyzed.

[0090] In addition, according to various embodiments, the processor 110 can input the posterior distribution of the previous model and the set likelihood function, which replaces the prior distribution of the coefficients / standard deviations set based on the selected order of the multinomial regression model, into the optimized Bayesian model in order to replace the prior distribution of the coefficients / standard deviations set based on the selected order of the multinomial regression model.

[0091] vi) Verification of posterior distribution

[0092] According to various embodiments, the processor 110 can verify the posterior distribution for the regression coefficient and the error term of the regression curve, which indicates the relationship between the past data of the sensor to be analyzed and the past data of the comparison target sensor, based on a preset method.

[0093] Also, according to various embodiments, the processor 110 can verify the convergence based on the R-hat value (e.g., 0.95 < R-hat < 1.05) and the ESS (Effective Sample Size) value (e.g., ESS > 500) after Burn-in and Thinning.

[0094] Further, according to various embodiments, when the posterior distribution for the regression coefficient and the error term of the regression curve, which indicates the relationship between the past data of the sensor to be analyzed and the past data of the comparison target sensor, does not converge to the criteria (R-hat value (e.g., 0.95 < R-hat < 1.05) and ESS value (e.g., ESS > 500)), the processor 110 can increase the Draw size by an arithmetic difference by the amount of the initial Draw size, and then re-verify the posterior distribution for the regression coefficient and the error term of the regression curve, which indicates the relationship between the past data of the sensor to be analyzed and the past data of the comparison target sensor.

[0095] Also, according to various embodiments, when the Draw size increases, the processor 110 can discard the samples before Burn-in and continue the Markov chain from the samples immediately after the Burn-in period of the increased Draw size.

[0096] Also, according to various embodiments, when the posterior distributions for the regression coefficient and the error term of the regression curve, which indicate the relationship between the past data of the sensor to be analyzed and the past data of the comparison target sensor, converge to the criteria (R-hat value (e.g., 0.95 < R-hat < 1.05) and ESS value (e.g., ESS > 500)), the processor 110 can estimate the posterior distributions for the regression coefficient and the error term of the regression curve, which indicate the relationship between the past data of the sensor to be analyzed and the past data of the comparison target sensor, by means of the Markov chain with the lowest MCSE (Monte Carlo Standard Error) value.

[0097] vii) Setting of management limit line

[0098] According to various embodiments, the processor 110 can set a target credible interval for the posterior distributions for the regression coefficient and the error term of the regression curve, which indicate the relationship between the past data of the sensor to be analyzed, which has been inferred, and the past data of the comparison target sensor, and can set a management limit line for the sensor data to be analyzed by using the set credible interval.

[0099] Also, according to various embodiments, the processor 110 can set a specific percentage (e.g., 50%, 95%) of the area of the posterior distributions for the regression coefficient and the error term of the regression curve, which indicate the relationship between the past data of the sensor to be analyzed and the past data of the comparison target sensor, to the credible interval. Here, the specific percentage of the area of the posterior distributions for the regression coefficient and the error term of the regression curve, which indicate the relationship between the past data of the sensor to be analyzed and the past data of the comparison target sensor, is related to the HPD (Highest Posterior Density, hereinafter referred to as "HPD") value and can be changed according to the situation.

[0100] In addition, according to various embodiments, the processor 110 can set control limits for the analyzed sensor data using the regression coefficients of the regression curve and the boundary values ​​of the posterior distribution for the error terms, which indicate the relationship between the historical data of the analyzed sensor and the historical data of the comparison sensor, corresponding to the set confidence interval.

[0101] According to various embodiments, the processor 110 can set a probabilistic control limit line by using the posterior distribution and HPD statistics for the regression coefficients and error terms of the regression curve, which indicate the relationship between the historical data of the analysis target sensor and the historical data of the comparison target sensor. Here, since the posterior distribution for the regression coefficients and error terms of the regression curve, which indicate the relationship between the historical data of the analysis target sensor and the historical data of the comparison target sensor, includes a mean and a standard deviation, a specific percentage confidence interval can be set for each of the posterior distributions of the mean and the standard deviation, and the control limit line of the analysis target sensor data can be set by combining values ​​of the posterior distribution for the regression coefficients and error terms of the regression curve, which indicate the relationship between the historical data of the analysis target sensor and the historical data of the comparison target sensor, which correspond to both ends of each confidence interval.

[0102] In addition, according to various embodiments, the processor 110 can set a safe region, an attention region, and a control line using HPD values ​​having a preset percentage (e.g., 25%, 75%, 2.5%, 97.5%) based on the regression coefficients of the regression curve and the median value of the posterior distribution for the error terms, which indicate the relationship between the historical data of the analyzed sensor and the historical data of the compared sensor.

[0103] In addition, according to various embodiments, the processor 110 can set a safe region to a region where the mean and standard deviation of the posterior distribution for the regression coefficients and error terms of the regression curve, which indicate the relationship between the historical data of the analyzed sensor and the historical data of the compared sensor, satisfy the following "Equation 2" and "Equation 3", and where the posterior distribution for the regression coefficients and error terms of the regression curve, which indicate the relationship between the historical data of the analyzed sensor and the historical data of the compared sensor, is the safe region.

[0104]

number

[0105]

number

[0106] Here, μ ν cl50% means the 50% credible interval for the dependent variable in the regression, and X ν cl50% shows a new criterion created by combining HPD statistics.

[0107] In addition, according to various embodiments, the processor 110 can set an Attention Region to a region where the mean and standard deviation of the posterior distribution for the regression coefficients and error terms of the regression curve, which indicate the relationship between the past data of the analyzed sensor and the past data of the compared sensor, satisfy the following "Equation 4" and "Equation 5", and where the posterior distribution for the regression coefficients and error terms of the regression curve, which indicate the relationship between the past data of the analyzed sensor and the past data of the compared sensor, is the region.

[0108]

number

[0109]

number

[0110] Here, μ ν Cl95% and X ν Cl95% indicates the mean and standard deviation with 95% confidence interval, respectively.

[0111] In addition, according to various embodiments, the processor 110 can set a control line to the boundary value of the area of ​​the posterior distribution for the regression coefficients and error terms of the regression curve that satisfies the following "Equation 6" and indicates the relationship between the past data of the analysis target sensor and the past data of the comparison target sensor.

[0112]

number

[0113] Also, according to various embodiments, the processor 110 can determine that the analyzed sensor is normal if the analyzed sensor data is within a safe region and an attention region, which are regions between control lines.

[0114] In addition, according to various embodiments, the processor 110 can determine that the sensor being analyzed is abnormal (sensor abnormal) if the sensor data being analyzed is in an area higher than the upper limit of a control line or is in an area lower than the lower limit of a control line.

[0115] viii) Verification of the validity of the sensor to be analyzed

[0116] According to various embodiments, the processor 110 can obtain current data from the analyte sensor through the input / output interface module 120 .

[0117] Additionally, according to various embodiments, the processor 110 can compare the current data of the analyte sensor to a control line to verify the validity of the analyte sensor.

[0118] Also, according to various embodiments, the processor 110 can determine that the analyzed sensor is valid (normal) if the current data of the analyzed sensor is within a safe region and an attention region, which are areas between control lines.

[0119] In addition, according to various embodiments, the processor 110 can determine that the analyte sensor is invalid (sensor abnormality) if the current data of the analyte sensor is in an area higher than the upper limit of the control line or in an area lower than the lower limit of the control line.

[0120] The I / O interface module 120 can be connected to an external device (eg, a server) through a network.

[0121] Furthermore, the input / output interface module 120 can obtain data from an external device.

[0122] The input / output interface module 120 can also acquire past data from the sensor being analyzed.

[0123] Furthermore, the input / output interface module 120 can acquire past data of the comparison target sensor.

[0124] Furthermore, the input / output interface module 120 can obtain a prior distribution for past data of the sensor to be analyzed.

[0125] Furthermore, the input / output interface module 120 can obtain a prior distribution for the past data of the comparison target sensor.

[0126] Additionally, the input / output interface module 120 allows for user input.

[0127] The I / O interface module 120 can also output a validation result of the analytical target sensor.

[0128] Additionally, the input / output interface module 120 may be provided integrally with the sensor validation device 100 .

[0129] Additionally, the input / output interface module 120 may be provided separately from the sensor validation device 100 .

[0130] Also, the input / output interface module 120 may be a separate device communicatively connected to the sensor validation device 100 .

[0131] In addition, the input / output interface module 120 may include a port (eg, a USB port) for connection to an external device.

[0132] The input / output interface module 120 may include a monitor, a touch screen, a mouse, an electronic pen, a microphone, a keyboard, a speaker, an earphone, a headphone, or a touch pad, etc.

[0133] The memory 130 can store data obtained through the input / output interface module 120 .

[0134] Additionally, the memory 130 can store data acquired by the processor 110 .

[0135] In addition, the memory 130 can store a safety area set by the processor 110 .

[0136] Additionally, the memory 130 can store the attention area set by the processor 110 .

[0137] The memory 130 may also store control limits set by the processor 110 .

[0138] Additionally, the memory 130 can store the validation results of the analyte sensor.

[0139] FIG. 2 is a flow chart illustrating a method for validating a sensor, according to one embodiment.

[0140] Referring to FIG. 2, the method for validating the validity of the sensor includes a step S200 of selecting the sensor data to be compared, a step S210 of optimizing initial values ​​of parameters of the Bayesian model and selecting an order of the polynomial regression model, a step S220 of inferring a posterior distribution for regression coefficients and error terms of the regression curve indicating a relationship between the past data of the sensor to be analyzed and the past data of the sensor to be compared, a step S230 of verifying the posterior distribution, a step S240 of setting control limits of the sensor data to be analyzed, and a step S250 of verifying the validity of the sensor to be analyzed.

[0141] In step S200, the sensor validity verification device may obtain past data of the target sensor and past data of a plurality of non-target sensors.

[0142] In addition, in step S200, the sensor validity verification device may select data having a high correlation with the past data of the analysis target sensor from among the past data of the non-analysis target sensors using distance correlation, and select the selected data as the comparison target sensor data.

[0143] In step S200, the sensor validity verification device may set a prior distribution for the past data of the analysis target sensor, where the prior distribution may be, but is not limited to, a prior distribution for the regression coefficients and error terms of the regression curve indicating the relationship between the past data of the analysis target sensor and the past data of the plurality of non-analysis target sensors.

[0144] In step S200, the sensor validity verification device may set a prior distribution for the past data of the comparison target sensor, where the prior distribution may be a prior distribution for the regression coefficients and error terms of the regression curve indicating the relationship between the past data of the analysis target sensor and the past data of the comparison target sensor, but is not limited thereto.

[0145] In step S210, the sensor validity verification device may perform sampling from the past data of the analysis target sensor, a prior distribution for the past data of the analysis target sensor, the past data of the comparison target sensor, and a prior distribution for the past data of the comparison target sensor.

[0146] In addition, in step S210, the sensor validity verification device can train the Bayesian model based on learning data sampled from the past data of the sensor to be analyzed, the prior distribution for the past data of the sensor to be analyzed, the past data of the sensor to be compared, and the prior distribution for the past data of the sensor to be compared.

[0147] In addition, in step S210, the sensor validity verification device can repeatedly train the Bayesian model based on new learning data sampled from the past data of the analyzed sensor, the prior distribution for the past data of the analyzed sensor, the past data of the compared sensor, and the prior distribution for the past data of the compared sensor, until the posterior distribution obtained using the Bayesian model satisfies both the acceptance rate (e.g., 0.2<=acceptance rate<=0.5) and autocorrelation (e.g., 50% of the samples are within a 95% confidence interval).

[0148] In addition, in step S210, the sensor validity verification device can stop learning the Bayesian model if the posterior distribution obtained using the Bayesian model satisfies both acceptance rate (e.g., 0.2<=acceptance rate<=0.5) and autocorrelation (e.g., 50% of the samples are within a 95% confidence interval).

[0149] In step S210, the sensor validation device may set a range of orders of the polynomial regression to be searched in order to determine the order of the polynomial.

[0150] In addition, in step S210, the sensor validity verification device may model the polynomial regression model using learning data for each order.

[0151] In addition, in step S210, the sensor validity verification apparatus may calculate an RMSE of the polynomial regression model and select an appropriate order of the polynomial regression model using the calculated RMSE.

[0152] In step S220, the sensor validation device may set a prior distribution of coefficients / standard deviations based on the selected order of the polynomial regression model.

[0153] Also, in the step S220, the sensor effectiveness verification device can set the regression coefficient of the first multiple regression equation to the average value of the prior distribution of the regression coefficient.

[0154] Also, in the step S220, the sensor effectiveness verification device inputs the set prior distribution, the past data of the sensor to be analyzed, and the past data of the comparison target sensor into the optimized Bayesian model, and can infer the posterior distributions of the regression coefficient and the error term of the regression curve indicating the relationship between the past data of the sensor to be analyzed and the past data of the comparison target sensor.

[0155] Also, in the step S220, the sensor effectiveness verification device can infer the regression coefficient of the multiple regression model based on the posterior distributions of the regression coefficient and the error term of the regression curve indicating the relationship between the past data of the sensor to be analyzed and the past data of the comparison target sensor.

[0156] In the step S230, the sensor effectiveness verification device can verify the convergence based on the R-hat value (for example, 0.95 < R-hat < 1.05) and the ESS value (for example, ESS > 500) after Burn-in and Thinning.

[0157] Also, in the step S230, when the Draw size increases, the sensor effectiveness verification device can discard the samples before Burn-in and continue the Markov chain from the samples immediately after the Burn-in period of the increased Draw size.

[0158] Also, in the step S230, when the posterior distributions of the regression coefficient and the error term of the regression curve, which indicate the relationship between the past data of the sensor to be analyzed and the past data of the comparison target sensor, converge to the criteria (R-hat value (e.g., 0.95 < R-hat < 1.05) and ESS value (e.g., ESS > 500)), the posterior distribution of the parameter of the sensor data to be analyzed can be estimated by the Markov chain with the lowest MCSE value.

[0159] In the step S240, the sensor effectiveness verification device can set a specific percentage (e.g., 50%, 95%) of the posterior distribution region of the regression coefficient and the error term of the regression curve, which indicate the relationship between the past data of the sensor to be analyzed and the past data of the comparison target sensor, as a confidence interval.

[0160] Also, in the step S240, the sensor effectiveness verification device can set the management limit line of the sensor data to be analyzed by using the boundary values of the posterior distribution of the regression coefficient and the error term of the regression curve, which indicate the relationship between the past data of the sensor to be analyzed and the past data of the comparison target sensor, corresponding to the set confidence interval.

[0161] In the step S250, when the current data of the sensor to be analyzed exists in the safe region and the attention region, which are regions between the control lines, the sensor to be analyzed can be determined to be effective (normal).

[0162] Also, in the step S250, when the current data of the sensor to be analyzed exists in a region higher than the upper limit of the control line or in a region lower than the lower limit of the control line, the sensor to be analyzed can be determined to be ineffective (sensor abnormality).

[0163] FIG. 3 is a flow chart illustrating a method for validating a sensor, according to another embodiment.

[0164] Referring to FIG. 3, the method for validating the validity of the sensor includes a step S300 of selecting sensor data to be compared, a step S310 of optimizing initial values ​​of parameters of a Bayesian model and selecting an order of a polynomial regression model, a step S320 of tuning control limits, a step S330 of inferring a posterior distribution for regression coefficients and error terms of a regression curve showing a relationship between past data of a sensor to be analyzed and past data of the sensor to be compared, a step S340 of verifying the posterior distribution, a step S350 of setting control limits of the sensor data to be analyzed, and a step S360 of verifying the validity of the sensor to be analyzed.

[0165] In step S300, the sensor validity verification device may obtain past data of the target sensor and past data of a plurality of non-target sensors.

[0166] In addition, in step S300, the sensor validity verification device may select data having a high correlation with the past data of the analysis target sensor from among the past data of the non-analysis target sensors using distance correlation, and select the selected data as the comparison target sensor data.

[0167] In step S300, the sensor validity verification device may set a prior distribution for the past data of the analysis target sensor, where the prior distribution may be, but is not limited to, a prior distribution for the regression coefficients and error terms of the regression curve indicating the relationship between the past data of the analysis target sensor and the past data of the plurality of non-analysis target sensors.

[0168] In addition, in step S300, the sensor validity verification device may set a prior distribution for the past data of the comparison target sensor, where the prior distribution may be a prior distribution for the regression coefficients and error terms of the regression curve indicating the relationship between the past data of the analysis target sensor and the past data of the comparison target sensor, but is not limited thereto.

[0169] In step S310, the sensor validity verification device may perform sampling from the past data of the analysis target sensor, a prior distribution for the past data of the analysis target sensor, the past data of the comparison target sensor, and a prior distribution for the past data of the comparison target sensor.

[0170] In addition, in step S310, the sensor validity verification device can train the Bayesian model based on learning data sampled from the past data of the sensor to be analyzed, the prior distribution for the past data of the sensor to be analyzed, the past data of the sensor to be compared, and the prior distribution for the past data of the sensor to be compared.

[0171] In addition, in step S310, the sensor validity verification device can repeatedly train the Bayesian model based on new learning data sampled from the past data of the analyzed sensor, the prior distribution for the past data of the analyzed sensor, the past data of the compared sensor, and the prior distribution for the past data of the compared sensor, until the posterior distribution obtained using the Bayesian model satisfies both the acceptance rate (e.g., 0.2<=acceptance rate<=0.5) and autocorrelation (e.g., 50% of the samples are within a 95% confidence interval).

[0172] In addition, in step S310, the sensor validity verification device can stop learning the Bayesian model if the posterior distribution obtained using the Bayesian model satisfies both acceptance rate (e.g., 0.2<=acceptance rate<=0.5) and autocorrelation (e.g., 50% of the samples are within a 95% confidence interval).

[0173] In step S310, the sensor validation device may set a range of orders of the polynomial regression to be searched in order to determine the order of the polynomial.

[0174] In addition, in step S310, the sensor validity verification device may model the polynomial regression model using learning data for each order.

[0175] In addition, in step S310, the sensor validity verification apparatus may calculate an RMSE of the polynomial regression model and select an appropriate order of the polynomial regression model using the calculated RMSE.

[0176] In step S320, the sensor validity verification device can update the control limits of the correlation as the conditions (maintenance, seasonal factors, aging, etc.) of the equipment from which the analyzed sensor obtains data change.

[0177] Also, in step S320, the sensor validation device can update the posterior distribution of the polynomial regression model to tune the control limits.

[0178] In step S320, the sensor validity verification device may update the posterior distribution of the polynomial regression model by using a Bayesian sampling inference for the polynomial regression model as a prior distribution, where the posterior distribution of the polynomial regression model may mean a posterior distribution for the regression coefficients and error terms (standard deviations) of the polynomial regression model that constitute the control limits, but the meaning of the posterior distribution of the previous model is not limited thereto.

[0179] In addition, in step S320, the sensor validity verification device may set a likelihood function, which may be a function of normal distribution, but is not limited thereto.

[0180] In addition, in step S320, the sensor validity verification device may input the posterior distribution of the previous model and the set likelihood function, which replace the prior distribution of the coefficients / standard deviations set based on the selected order of the polynomial regression model, to the optimized Bayesian model in order to replace the prior distribution of the coefficients / standard deviations set based on the selected order of the polynomial regression model.

[0181] In step S330, the sensor validation device may set a prior distribution of coefficients / standard deviations based on the selected order of the polynomial regression model.

[0182] In addition, in step S330, the sensor validity verification device inputs the set prior distribution, the past data of the analysis target sensor, and the past data of the comparison target sensor into the optimized Bayesian model, and can infer a posterior distribution for the regression coefficients and error terms of the regression curve that indicate the relationship between the past data of the analysis target sensor and the past data of the comparison target sensor.

[0183] Also, in the step S330, the sensor effectiveness verification device can infer the regression coefficient of the multiple regression model based on the posterior distribution of the regression coefficient and the error term of the regression curve, which shows the relationship between the past data of the sensor to be analyzed and the past data of the comparison target sensor.

[0184] In the step S340, the sensor effectiveness verification device can verify the convergence based on the R-hat value (e.g., 0.95 < R-hat < 1.05) and the ESS value (e.g., ESS > 500) after Burn-in and Thinning.

[0185] Also, in the step S340, when the Draw size increases, the sensor effectiveness verification device can discard the samples before Burn-in and continue the Markov chain from the samples immediately after the Burn-in period of the increased Draw size.

[0186] Also, in the step S340, when the posterior distribution of the regression coefficient and the error term of the regression curve, which shows the relationship between the past data of the sensor to be analyzed and the past data of the comparison target sensor, converges to the criteria (R-hat value (e.g., 0.95 < R-hat < 1.05) and ESS value (e.g., ESS > 500)), the sensor effectiveness verification device can estimate the posterior distribution of the population parameter of the sensor data to be analyzed with the Markov chain having the lowest MCSE value.

[0187] In the step S350, the sensor effectiveness verification device can set a specific percentage (e.g., 50%, 95%) of the region of the posterior distribution of the regression coefficient and the error term of the regression curve, which shows the relationship between the past data of the sensor to be analyzed and the past data of the comparison target sensor, as a confidence interval.

[0188] In addition, in step S350, the sensor validity verification device can set a control limit line for the analysis target sensor data by using a regression coefficient of the regression curve and a boundary value of a posterior distribution for an error term, which corresponds to the set confidence interval and indicates a relationship between the past data of the analysis target sensor and the past data of the comparison target sensor.

[0189] In step S360, the sensor validity verification device can determine that the sensor to be analyzed is valid (normal) if the current data of the sensor to be analyzed is within a safe region and an attention region, which are regions between control lines.

[0190] In addition, in step S360, the sensor validity verification device can determine that the sensor to be analyzed is invalid (sensor abnormality) if the current data of the sensor to be analyzed is in an area higher than the upper limit of the control line or is in an area lower than the lower limit of the control line.

[0191] FIG. 4 is a flow chart illustrating how the sensor validation device optimizes initial values ​​of the parameters of the polynomial regression model, according to one embodiment.

[0192] Referring to FIG. 4, the sensor validity verification device can obtain past data 401 of the analysis target sensor and past data 402 of the comparison target sensor.

[0193] According to various embodiments, the sensor validity verification device may set a prior distribution for the past data of the analyzed sensor in step S410. Here, the prior distribution may be, but is not limited to, a prior distribution for the regression coefficients and error terms of the regression curve indicating the relationship between the past data of the analyzed sensor and the past data of the plurality of non-analyzed sensors.

[0194] According to various embodiments, the sensor validity verification device can set a prior distribution for the past data of the comparison target sensor, where the prior distribution can be, but is not limited to, a prior distribution for the regression coefficients and error terms of the regression curve indicating the relationship between the past data of the analysis target sensor and the past data of the comparison target sensor.

[0195] In addition, according to various embodiments, in step S410, the sensor validity verification device may perform sampling from past data of the analysis target sensor, a prior distribution for the past data of the analysis target sensor, past data of the comparison target sensor, and a prior distribution for the past data of the comparison target sensor to generate learning data.

[0196] According to various embodiments, the sensor validity verification device may calculate a likelihood using past data of the analysis target sensor in step S410. The method of calculating the likelihood is a well-known method, and therefore a detailed description thereof will be omitted here.

[0197] According to various embodiments, the sensor validity verification device may calculate a likelihood using past data of the comparison sensor in step S410. The method of calculating the likelihood is a well-known method, and therefore a detailed description thereof will be omitted here.

[0198] Furthermore, according to various embodiments, in step S420, the sensor validity verification device can obtain posterior distributions for regression coefficients and error terms of the regression curve, which indicate the relationship between the past data of the analysis target sensor and the past data of the comparison target sensor, based on the learning data generated by sampling.

[0199] Furthermore, according to various embodiments, the posterior distribution of the regression coefficients and error terms of the regression curve, which indicates the relationship between the past data of the analysis target sensor and the past data of the comparison target sensor, acquired by the sensor validity verification device, can be expressed using the following "Number 7" based on the learning data, the likelihood of the past data of the analysis target sensor, the likelihood of the past data of the comparison target sensor, the set prior distribution for the past data of the analysis target sensor, and the set prior distribution for the past data of the comparison target sensor.

[0200]

number

[0201] where P(H|D) is the posterior distribution, P(D|H) is the likelihood, P(H) is the prior distribution over the data, and P(D) is the probability over the data.

[0202] According to various embodiments, the sensor validity verification device may compare the posterior distribution obtained using the Bayesian model with a preset criterion in step S430 to verify the learning state of the Bayesian model. In this case, the preset criterion may be, but is not limited to, an acceptance rate and an autocorrelation.

[0203] Also, according to various embodiments, in step S430, the sensor validation device may decide to retrain the Bayesian model if the posterior distribution obtained using the Bayesian model does not satisfy both the acceptance rate (e.g., 0.2<=Acceptance rate<=0.5) and autocorrelation (e.g., 50% of the samples are within a 95% confidence interval).

[0204] Furthermore, according to various embodiments, in step S430, the sensor validity verification device can iteratively train the Bayesian model based on new learning data sampled from the past data of the analyzed sensor, the prior distribution for the past data of the analyzed sensor, the past data of the comparison sensor, and the prior distribution for the past data of the comparison sensor, until the posterior distribution obtained using the Bayesian model satisfies both an acceptance rate (e.g., 0.2<=acceptance rate<=0.5) and autocorrelation (e.g., 50% of the samples are within a 95% confidence interval).

[0205] In addition, according to various embodiments, in step S430, the sensor validation verification device can stop learning the Bayesian model if the posterior distribution obtained using the Bayesian model satisfies both an acceptance rate (e.g., 0.2<=Acceptance rate<=0.5) and autocorrelation (e.g., 50% of the samples are within a 95% confidence interval).

[0206] 5a-5c are exemplary diagrams illustrating safety zones, caution zones and control limits according to one embodiment.

[0207] Referring to FIG. 5a, the upper and lower control limits are shown in FIG.

[0208] The central plot of FIG. 5a shows the expected value of the posterior distribution relative to the mean in FIG. 5b.

[0209] In FIG. 5a, line 500 represents the upper limit of the control limit line μ ν avg +3σ ν avg The line 510 indicates the lower limit of the control limit line in the following "Equation 8", μ ν avg -3σ ν avg Shows.

[0210]

number

[0211] FIG. 5b is a diagram illustrating an example of a graph of the posterior distribution of the regression coefficients and error terms of the regression curve. That is, Y=w0+w r0 x+w rl x 2 Regression coefficients of each regression curve at +σ (w0,w r0 ,w rl ) and the error term (σ).

[0212] Referring to FIG. 5b, the left column shows the posterior distribution of the regression coefficients and error terms of each regression curve, and the right column shows examples of samples obtained by Bayesian sampling inference.

[0213] Figure 5c shows the region from "Number 9" to "Number 13" below.

[0214]

number

[0215]

number

[0216]

number

[0217]

number

[0218]

number

[0219] In the top right diagram, which is an enlarged view of the upper portion 530 of FIG. 5c, 531 shows the upper limit of the control limit (μ ν avg +3σ ν avg ) and the upper limit of the standard deviation of the attention region (μ ν 97.5% +3σ ν hpd97.5% ) where 532 is the upper limit of the standard deviation of the safety region (μ ν hpd75% +3σ ν hpd75% ) is shown below.

[0220] In the figure in the center of the right side, which is an enlarged view of the central part 540 of FIG. 5c, 541 shows the upper limit of the standard deviation of the safe region (μ ν hpd75% +3σ ν hpd75% ), and 542 and 544 show the upper limit of the average of the attention region (μ ν hpd97.5% ) 542 and the lower limit of the mean (μ ν hpd2.5% ) 544. Also, 543 indicates "number 8", and 545 indicates the lower limit of the standard deviation of the safety area (μ ν hpd2.5%- 3σ ν hpd97.5% ) is shown below.

[0221] In the bottom right diagram, which is an enlarged view of the lower part 550 of FIG. 5c, 551 shows the lower limit of the standard deviation of the safe region (μ ν hpd25%- 3σ ν hpd75% ) is shown. In the figure at the bottom right, 552 is the lower limit of the control limit (μ ν avg- 3σ ν avg ) and the lower limit of the standard deviation of the attention region (μ ν hpd2.5%- 3σ ν hpd97.5% ) is shown below.

[0222] FIG. 6 is a drawing for explaining a technique for validating the effectiveness of a sensor to which the multiple regression model is applied according to an embodiment.

[0223] Referring to FIG. 6, the sensor effectiveness verification device can obtain the past data 601 of the analysis target sensor and the past data 602 of the comparison target sensor, and optimize the initial values of the parameters of the Bayesian model (step S610).

[0224] According to various embodiments, the sensor effectiveness verification device can obtain the past data 601 of the analysis target sensor and the past data 602 of the comparison target sensor, and set the degree of the multiple regression model (step S620).

[0225] Also, according to various embodiments, the sensor effectiveness verification device inputs the prior distribution of the coefficient / standard deviation set based on the degree of the multiple regression model, the past data 601 of the analysis target sensor, and the past data 602 of the comparison target sensor into the optimized Bayesian model 631, and can infer (632) the posterior distribution for the regression coefficient and the error term of the regression curve indicating the relationship between the past data of the analysis target sensor and the past data of the comparison target sensor (step S630).

[0226] Also, according to various embodiments, the sensor effectiveness verification device can replace the prior distribution of the coefficient / standard deviation set based on the degree of the multiple regression model with the posterior distribution of the previous model (641), set the likelihood function (642), and tune the management limit (step S640).

[0227] Also, according to various embodiments, the sensor effectiveness verification device can verify the convergence of the posterior distribution for the regression coefficient and the error term of the regression curve indicating the relationship between the past data of the analysis target sensor and the past data of the comparison target sensor based on the R-hat value (for example, 0.95 < R-hat < 1.05) and the ESS value (for example, ESS > 500) after Burn-in and Thinning (step S650).

[0228] In addition, according to various embodiments, the sensor validity verification device can set a target credible interval for the posterior distribution of the regression coefficients and error terms of the regression curve, which indicates the inferred relationship between the past data of the analyzed sensor and the past data of the comparison sensor, and set a control limit line of the analyzed sensor data using the set credible interval (step S660).

[0229] Also, according to various embodiments, the sensor validity verification device can verify the validity of the analyte sensor by comparing the current data 680 of the analyte sensor with the set control limit line (step S670).

[0230] Although the present disclosure has been described with reference to the embodiments shown in the drawings, these are merely illustrative, and a person having ordinary skill in the art to which the present invention pertains will understand that various modifications and equivalent embodiments are possible. Therefore, the true technical scope of protection of the present disclosure should be determined by the technical ideas of the appended claims. [Explanation of symbols]

[0231] 100...Sensor validity verification device

Claims

1. In the sensor validity verification device, at least one processor; The at least one processor: Optimizing initial values ​​of parameters of a Bayesian model and selecting the order of a polynomial regression model based on the past data of the analysis target sensor and the past data of the comparison target sensor; Inferring a posterior distribution for regression coefficients and error terms of a regression curve showing a relationship between the historical data of the analysis target sensor and the historical data of the comparison target sensor, using the optimized Bayesian model and the polynomial regression model with a selected order based on the historical data of the analysis target sensor and the historical data of the comparison target sensor; setting a confidence interval based on a posterior distribution for the regression coefficient and error term of the regression curve, which indicates a relationship between the historical data of the analysis target sensor and the historical data of the comparison target sensor, and setting a control limit line for the data of the analysis target sensor using the set confidence interval; Validating the validity of the analyte sensor based on the current data of the analyte sensor and the established control limits; Sensor validity verification device.

2. The at least one processor: selecting the data of the comparison sensor using distance correlation based on the past data of the analysis target sensor and the past data of a plurality of non-analysis target sensors; The sensor validity verification device according to claim 1 .

3. The at least one processor: replacing the prior distribution set based on the multinomial regression model with a posterior distribution for the regression coefficients and error terms of the multinomial regression model constituting the control limits of the data of the sensor to be analyzed; Set the likelihood function, The sensor validity verification device according to claim 1 .

4. The at least one processor: verifying a posterior distribution for the regression coefficients and error terms of the regression curve, which indicates a relationship between the past data of the analysis target sensor and the past data of the comparison target sensor, based on a preset method; The sensor validity verification device according to claim 1 .

5. The at least one processor: setting the confidence interval by applying a Highest Posterior Density (HPD) value having a preset percentage based on an expected value of a posterior distribution for a regression coefficient and an error term of the regression curve indicating a relationship between the past data of the analysis target sensor and the past data of the comparison target sensor; setting a control limit line for the data of the target sensor using boundary values ​​on both sides of a posterior distribution for the regression coefficient and error term of the regression curve, which indicates a relationship between the historical data of the target sensor and the historical data of the comparison sensor and corresponds to the confidence interval; The sensor validity verification device according to claim 1 .

6. A computer-implemented method for validating a sensor, comprising: optimizing initial values ​​of parameters of the Bayesian model and selecting the order of the polynomial regression model based on the historical data of the analysis target sensor and the historical data of the comparison target sensor; inferring posterior distributions for regression coefficients and error terms of a regression curve that indicates a relationship between the historical data of the analysis target sensor and the historical data of the comparison target sensor, using the Bayesian model optimized based on the historical data of the analysis target sensor and the historical data of the comparison target sensor and the polynomial regression model whose order has been selected; setting a confidence interval based on a posterior distribution for the regression coefficients and error terms of the regression curve, which indicates the relationship between the historical data of the analyzed sensor and the historical data of the comparison sensor, and setting a control limit line for the data of the analyzed sensor using the set confidence interval; and verifying the validity of the analyte sensor based on current data of the analyte sensor and the established control limits; How to verify the validity of sensors.

7. The method for validating a sensor comprises: selecting the data of the comparison sensor using distance correlation based on the past data of the analysis target sensor and the past data of a plurality of non-analysis target sensors; The method of claim 6 .

8. Inferring a posterior distribution for regression coefficients and error terms of the regression curve, which indicates a relationship between the historical data of the analysis target sensor and the historical data of the comparison target sensor, using the optimized Bayesian model and the polynomial regression model with the selected order, includes: A step of replacing the prior distribution set based on the multinomial regression model with a posterior distribution for the regression coefficients and error terms of the multinomial regression model constituting the control limits of the data of the sensor to be analyzed; and establishing a likelihood function; The method of claim 6 .

9. Inferring a posterior distribution for regression coefficients and error terms of the regression curve, which indicates a relationship between the historical data of the analysis target sensor and the historical data of the comparison target sensor, using the optimized Bayesian model and the polynomial regression model with the selected order, includes: and verifying a posterior distribution for the regression coefficients and error terms of the regression curve, which indicates a relationship between the historical data of the analysis target sensor and the historical data of the comparison target sensor, based on a preset method. The method of claim 6 .

10. The step of setting a control limit line for the data of the sensor to be analyzed using the set confidence interval includes: setting the confidence interval by applying a Highest Posterior Density (HPD) value having a preset percentage based on the regression coefficient of the regression curve indicating the relationship between the past data of the analysis target sensor and the past data of the comparison target sensor and the expected value of the posterior distribution for the error term; and setting control limits for the data of the target sensor using boundary values ​​on both sides of a posterior distribution for the regression coefficients and error terms of the regression curve, which indicate the relationship between the historical data of the target sensor and the historical data of the comparison sensor, corresponding to the confidence interval; The method of claim 6 .

11. A program for causing a computer to execute the method for verifying the validity of a sensor according to any one of claims 6 to 10.

Images