Method for issuing alarm for abnormality of nuclear power plant state monitoring model using independent component analysis and sliding window

Abstract

A method for issuing an alarm for an abnormality of a nuclear power plant state monitoring model using independent component analysis and sliding window comprises the steps of: setting a sliding window range and parameter of an individual signal for each model and performing independent component analysis of the individual signal for each model; determining as an individual signal of one model and reorganizing as an individual signal of another model on the basis of the independent component analysis; calculating an anomaly score of the individual signal on the basis of the sliding window range and parameter; calculating a performance index of the individual signal on the basis of the anomaly score of the individual signal; calculating a performance index of the plant state monitoring model on the basis of the performance index of the individual signal; and issuing an alarm for abnormalities of the nuclear power plant on the basis of the performance index of the plant state monitoring model.

Description

Anomaly alert method for nuclear power plant condition monitoring models using independent component analysis and sliding windows

[0001] This paper relates to an abnormality alarm method for a nuclear power plant status monitoring model using independent component analysis and sliding window.

[0002] In general, the plant condition monitoring model is a model that monitors the condition of a nuclear power plant by using signals received from sensors installed in the facilities included in the nuclear power plant.

[0003] The plant condition monitoring model monitors the condition of the nuclear power plant and alerts of any abnormalities in the nuclear power plant by using signals received from various sensors.

[0004] The abnormality alarm method of the conventional plant condition monitoring model learns the entire specific section in the form of a snapshot as a single point in time, calculates an abnormality score through new measurement information and calculations, and then alerts of abnormalities in the nuclear power plant.

[0005] The abnormality alarm method of the conventional plant status monitoring model has a problem in that the snapshot type is vulnerable to noise and has a high probability of false alarms due to the fact that a lot of noise occurs in the signals received from the nuclear power plant due to the various operating environments of the nuclear power plant.

[0006] One embodiment provides an abnormality alarm method of a nuclear power plant status monitoring model using independent component analysis and a sliding window, which efficiently alerts of abnormal signs for each system and facility included in a nuclear power plant even when a large amount of noise occurs in signals received from the nuclear power plant depending on various operating environments of the nuclear power plant.

[0007] One aspect provides a method for an abnormality alarm of a plant condition monitoring model using independent component analysis and a sliding window, including the steps of setting a sliding window range and parameters of individual signals for each model of a plant condition monitoring model of a nuclear power plant and performing independent component analysis on the individual signals for each model, determining highly correlated signals among the individual signals for each model based on the independent component analysis as individual signals of one model and reorganizing signals with low correlation as individual signals of another model, calculating an abnormality score of the individual signal based on the sliding window range and parameters, calculating a performance index of the individual signal based on the abnormality score of the individual signal, calculating a performance index of the plant condition monitoring model based on the performance index of the individual signal, and providing an abnormality alarm of the nuclear power plant based on the performance index of the plant condition monitoring model.

[0008] The step of performing the above independent component analysis can be performed by applying independent component analysis based on dimension reduction.

[0009] The above model can perform signal restoration prediction.

[0010] The step of calculating the abnormality score of the above individual signals can be performed by applying order statistics.

[0011] The step of calculating the performance index of the above individual signal can be performed by comparing the above abnormal score with a set normal range.

[0012] The step of calculating the performance indicators of the above plant condition monitoring model can be performed by classifying the model based on the physical classification system of the above nuclear power plant.

[0013] According to one embodiment, even if a large amount of noise occurs in signals received from a nuclear power plant depending on various operating environments of the nuclear power plant, an abnormality alarm method of a nuclear power plant status monitoring model using independent component analysis and a sliding window is provided, which efficiently alerts of abnormal signs for each system and facility included in the nuclear power plant.

[0014] Figure 1 is a flowchart illustrating an abnormality alarm method of a plant status monitoring model according to one embodiment.

[0015] FIG. 2 is a diagram showing an example of a sliding window used in an abnormality alert method of a plant status monitoring model according to one embodiment.

[0016] FIG. 3 is a graph showing an example of independent component analysis used in an abnormality alert method of a plant status monitoring model according to one embodiment.

[0017] FIG. 4 is a diagram showing an example of calculating an abnormality score of an individual signal in an abnormality alarm method of a plant status monitoring model according to one embodiment.

[0018] FIG. 5 is a diagram showing an example of calculating performance indicators of individual signals of an abnormality alarm method of a plant status monitoring model according to one embodiment.

[0019] FIG. 6 is a diagram showing an example of calculating a performance index of a plant status monitoring model in an abnormality alarm method of a plant status monitoring model according to one embodiment.

[0020] FIG. 7 is a diagram showing an example of a plant status monitoring model classified based on the physical classification system of a nuclear power plant in an abnormality alarm method of a plant status monitoring model according to one embodiment.

[0021] Figure 8 is a diagram showing an example of a subsystem of the plant status monitoring model illustrated in Figure 7.

[0022] Fig. 9 is a flowchart showing an example of an abnormality alarm method of a plant status monitoring model according to one embodiment.

[0023] Hereinafter, with reference to the attached drawings, embodiments of the present invention will be described in detail so that those skilled in the art can easily implement the present invention. The present invention may be implemented in various different forms and is not limited to the embodiments described herein.

[0024] Additionally, throughout the specification, whenever a part is said to "include" a component, this does not mean that it excludes other components, but rather that it may include other components, unless otherwise specifically stated.

[0025] Hereinafter, an abnormality alarm method of a plant status monitoring model according to one embodiment will be described with reference to FIGS. 1 to 9.

[0026] An abnormality alarm method of a plant condition monitoring model according to one embodiment may include an abnormality alarm method of a nuclear power plant condition monitoring model using independent component analysis and a sliding window.

[0027] The abnormality alarm method of a plant status monitoring model according to one embodiment is such that the plant status monitoring model that monitors the status of a nuclear power plant can alert of abnormal signs of a nuclear power plant using independent component analysis and a sliding window, but is not limited thereto.

[0028] The abnormality alarm method of the plant condition monitoring model according to one embodiment may be performed by artificial intelligence driven by a computing device, but is not limited thereto.

[0029] Figure 1 is a flowchart illustrating an abnormality alarm method of a plant status monitoring model according to one embodiment.

[0030] Referring to Fig. 1, first, the sliding window range and parameters of individual signals for each model of the plant status monitoring model are set, and independent component analysis of individual signals for each model is performed (S100).

[0031] For example, the sliding window range and parameters of individual signals for each model of the plant condition monitoring model received from the nuclear power plant to the plant condition monitoring model are set.

[0032] FIG. 2 is a diagram showing an example of a sliding window used in an abnormality alert method of a plant status monitoring model according to one embodiment.

[0033] Referring to FIG. 2, the model-specific individual signals in the form of snapshots are set as model-specific individual signals in the form of a sliding window, thereby setting the sliding window range of the model-specific individual signals, and setting the parameters of the model-specific individual signals within this range. Here, the parameters of the model-specific individual signals may include various known parameters used in the plant status monitoring model among the parameters included in the model-specific individual signals.

[0034] Set the sliding window range and parameters of each signal for each model and perform independent component analysis (ICA) of each signal for each model.

[0035] For example, independent component analysis (ICA) is performed on individual signals for each model. ICA can include various known ICA methods. For example, various known ICA methods can be performed by artificial intelligence driven by a computing device.

[0036] For example, independent component analysis of individual signals for each model can be performed by applying independent component analysis based on dimension reduction.

[0037] FIG. 3 is a graph showing an example of independent component analysis used in an abnormality alert method of a plant status monitoring model according to one embodiment.

[0038] Referring to Figure 3, for example, AI can automatically classify the signal components of individual signals for each model by applying dimensionality reduction-based independent component analysis (ICA). AI can perform sophisticated predictions by grouping only strongly correlated signals and restoring the signal. ICA based on dimensionality reduction can produce superior prediction results compared to existing statistical techniques. AI can also automatically separate weakly correlated signals and perform separate predictions. ICA based on dimensionality reduction is more efficient and objective than the conventional manual separation by engineers.

[0039] Next, based on independent component analysis, signals with high correlation among individual signals for each model are confirmed as individual signals of one model, and signals with low correlation are reorganized as individual signals of another model (S200).

[0040] For example, based on independent component analysis (ICA), highly correlated signals within each model are identified as individual signals of one model. Based on ICA, signals with low correlations within each model are reorganized as individual signals of another model.

[0041] One model, where highly correlated signals among the individual signals of each model are identified, can perform signal restoration prediction. Another model, where less correlated signals among the individual signals of each model are reorganized, can perform Approximate Nearest Neighbor Search (ANNS) prediction based on the physical classification system of the nuclear power plant. The ANNS can include various known ANNS methods. For example, various known ANNS prediction methods can be performed by artificial intelligence driven by a computing device.

[0042] Next, the abnormality score of the individual signal is calculated based on the sliding window range and parameters (S300).

[0043] For example, anomaly scores for individual signals are calculated based on the sliding window range and parameters of each signal, which may be included in the aforementioned model or another model. Calculating the anomaly scores for individual signals can be performed by applying order statistics.

[0044] FIG. 4 is a diagram showing an example of calculating an abnormality score of an individual signal in an abnormality alarm method of a plant status monitoring model according to one embodiment.

[0045] Referring to Fig. 4, the calculation of the anomaly score (AS) of an individual signal based on the set sliding window is performed by considering the model learning parameters or the characteristics of the signal, and the window length M and the normal setting value of the residual г T , and the calculated anomaly score (AS) can be calculated based on the residual of the signal within the window period. Here, in order to exclude false alarms due to momentary noise of individual signals, etc., the corrected residual that excludes outliers can be used by applying order statistics. The calculation formula shown in Fig. 3 is an example, and calculating the anomaly score of an individual signal may not be limited to the calculation formula shown in Fig. 3.

[0046] Next, the performance index of the individual signal is calculated based on the abnormality score of the individual signal (S400).

[0047] For example, a performance indicator for an individual signal is calculated based on the anomaly score of the individual signal. This calculation can be performed by comparing the anomaly score with a set normal range.

[0048] FIG. 5 is a diagram showing an example of calculating performance indicators of individual signals of an abnormality alarm method of a plant status monitoring model according to one embodiment.

[0049] Referring to Figure 5, the calculation of the performance indicator (KPI) of an individual signal is performed by setting the abnormality score (AS) of an individual signal to the normal setpoint of the residual to help the user's intuitive judgment. T A normalized performance indicator (KPI) can be calculated. The KPI can approach 100% as the residual within the window approaches normality, and decrease to 0% as it deviates from the normal range. The calculation formula shown in Fig. 4 is an example, and calculating the performance indicator of an individual signal may not be limited to the calculation formula shown in Fig. 4.

[0050] Next, the performance indicators of the plant status monitoring model are calculated based on the performance indicators of the individual signals (S500).

[0051] For example, performance indicators for a plant condition monitoring model can be calculated based on the performance indicators of individual signals. This calculation can be accomplished by classifying the model based on the physical classification system of nuclear power plants.

[0052] FIG. 6 is a diagram showing an example of calculating a performance index of a plant status monitoring model in an abnormality alarm method of a plant status monitoring model according to one embodiment.

[0053] Referring to Fig. 6, the performance indicator (KPI) of the plant condition monitoring model can be calculated by synthesizing individual signals included in each model of the plant condition monitoring model. In order to compensate for the fact that the individual influence is reduced when the number of performance indicators (KPIs) is large, an amplification function that amplifies the influence according to the size or number of performance indicators (KPIs) can be used. The results using the amplification function can be weighted and averaged according to importance, and then the inverse function of the amplification function can be used to calculate the overall performance indicator (KPI). The calculation formula illustrated in Fig. 5 is an example, and the calculation of the performance indicator of the plant condition monitoring model may not be limited to the calculation formula illustrated in Fig. 5.

[0054] The calculation of the performance indicators (KPIs) of the system structure included in the plant condition monitoring model can be classified by using or based on the physical breakdown structure (BPS) of the nuclear power plant. The comprehensive performance indicators (KPIs) for each classified system can be calculated, and the calculation method can be the same as the calculation method for the model-specific performance indicators (KPIs) described above. However, depending on the importance of the model or facility, the amplification function result y i can be calculated by taking a weighted average.

[0055] FIG. 7 is a diagram illustrating an example of a plant condition monitoring model classified based on the physical classification system of a nuclear power plant in an abnormality alert method of a plant condition monitoring model according to one embodiment. FIG. 8 is a diagram illustrating an example of a subsystem of the plant condition monitoring model illustrated in FIG. 7.

[0056] Referring to FIGS. 7 and 8, for example, the performance indicators (KPIs) of the plant status monitoring model can be displayed in the plant status monitoring model classified by system and facility model based on or using the physical classification system of the nuclear power plant.

[0057] Next, an abnormality in the nuclear power plant is alerted based on the performance indicators of the plant status monitoring model (S600).

[0058] For example, a nuclear power plant's abnormalities can be alerted based on the performance indicators of the plant condition monitoring model. For example, an abnormality in a model with a near-0% level can be alerted based on the performance indicators of the system and equipment models of the plant condition monitoring model illustrated in Figures 7 and 8.

[0059] Fig. 9 is a flowchart showing an example of an abnormality alarm method of a plant status monitoring model according to one embodiment.

[0060] Referring to Fig. 9, as an example of an abnormality alert method for a plant condition monitoring model, necessary monitoring signals can be selected based on plant operation information, and a plant condition monitoring prediction model can be constructed based on a physical classification system. Normal operation patterns are extracted, and independent component analysis is performed by setting window ranges and parameters for each model. Principal component signals with strong correlations are confirmed as signal models to perform signal restoration prediction, and signal model reorganization can be performed if there are signals with weak correlations. Anomaly scores for individual signals can be calculated using the predicted residuals of individual signals, and performance indices for individual signals can be calculated through normalization. The performance indices of each signal are synthesized based on the upper system and facility models to calculate performance indices for the model and the system / facility, thereby enabling monitoring of both individual signal anomaly scores and the overall anomaly score of the system / facility model. If a model classified in the plant condition monitoring model falls outside the performance indicator range set by the model, an anomaly sign can be extracted and an alarm can be issued.

[0061] In this way, even if a lot of noise occurs in the signals received from the nuclear power plant depending on the various operating environments of the nuclear power plant, an abnormality alarm method of a nuclear power plant status monitoring model using independent component analysis and a sliding window is provided, which efficiently alerts of abnormal signs for each system and facility included in the nuclear power plant by using a sliding window.

[0062] Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements made by those skilled in the art using the basic concept of the present invention defined in the following claims also fall within the scope of the present invention.

Claims

1. A step of setting the sliding window range and parameters of individual signals for each model of a plant status monitoring model of a nuclear power plant and performing independent component analysis of the individual signals for each model; A step of determining, based on the independent component analysis, signals with high correlation among the individual signals for each model as individual signals of one model and reorganizing signals with low correlation as individual signals of another model; A step of calculating an anomaly score of the individual signal based on the sliding window range and parameters; A step of calculating a performance index of the individual signal based on the abnormality score of the individual signal; A step of calculating a performance indicator of the plant condition monitoring model based on the performance indicator of the individual signal; and A step of alerting an abnormality in the nuclear power plant based on the performance indicators of the plant status monitoring model. An abnormality alarm method for a plant condition monitoring model using independent component analysis and sliding window including .

2. In paragraph 1, The step of performing the above independent component analysis is an independent component analysis performed by applying independent component analysis based on dimension reduction and an abnormality alarm method of a plant status monitoring model using a sliding window.

3. In paragraph 1, The above model is an independent component analysis that performs signal restoration prediction and an abnormality alarm method of a plant condition monitoring model using a sliding window.

4. In paragraph 1, The step of calculating the abnormality score of the above individual signals is an independent component analysis performed by applying order statistics and an abnormality alarm method of a plant condition monitoring model using a sliding window.

5. In paragraph 1, The step of calculating the performance index of the above individual signal is performed by comparing the above abnormal score with a set normal range, and the method for abnormal alarm of a plant status monitoring model using independent component analysis and sliding window.

6. In paragraph 1, The step of calculating the performance indicator of the above plant condition monitoring model is an independent component analysis and an abnormality alarm method of the plant condition monitoring model using a sliding window, which is performed by classifying the model based on the physical classification system of the above nuclear power plant.

Images