Method and device for constructing fault diagnosis algorithm verification data of nuclear power plant

Abstract

The invention discloses a construction method and device for verification data of a fault diagnosis algorithm of a nuclear power plant. The method comprises the following steps: obtaining experiment operation data of a nuclear power plant system operation experiment bench; carrying out simulation modeling on the experiment bench, and obtaining corresponding simulation operation data; carrying out noise adding processing on the simulation operation data according to the experiment operation data; and based on the experimental operation data and the simulation operation data after noise addition processing, constructing verification data for a fault diagnosis algorithm. According to the method, simulation operation data can be closer to experiment operation data, on the other hand, data randomness can be increased by disturbing the data, and fault diagnosis algorithm training overfitting is prevented; based on the experiment operation data and the simulation operation data after noise adding processing, the verification data constructed together does not depend on the types of the fault diagnosis algorithms, the verification capability for various fault diagnosis algorithms is good, and the problems that in the prior art, the experiment data amount is small, and credibility is lacked can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent algorithm verification, and in particular to a method and device for constructing fault diagnosis algorithm verification data for a nuclear power plant. Background Art

[0002] At present, the nuclear power plant system can use intelligent algorithm models to realize health status assessment, minor fault warning, self-healing control, intelligent prediction and other functions. Before its engineering application, its algorithm model needs to be verified to ensure the accuracy and reliability of the algorithm and improve the technical maturity. The existing research on intelligent algorithms related to nuclear power plants focuses on algorithm optimization, data processing and feature extraction, and no clear solution is proposed for the algorithm verification process. The operation design system of nuclear power plants is complex and there are many types of equipment. In order to ensure its safety during operation, the current operating state is usually not changed at will. Therefore, the steady-state operation data of nuclear power plants is relatively easy to obtain. However, in the design process of the analysis algorithm, it is ideally necessary to obtain the transient operation data of the nuclear power plant under variable conditions and all accident conditions as training sample data. Due to the complexity of nuclear power plants and their environment, there are a large number of beyond-design basis accidents and unknown accident types. Moreover, for some known severe accidents, transient operation data cannot be directly obtained through experimental devices. The corresponding accident conditions can only be calculated through simulation to obtain transient operation data under simulation conditions. At the same time, the difficulties and high costs in obtaining experimental data also lead to a small amount of experimental data being obtained, which makes it unreliable to use it as a verification of intelligent algorithms.

[0003] The existing patent CN113821420B discloses a method for comparing and converting the performance of a CMS system of a wind turbine, which includes the following steps: Step S1: comparing the test performance of CMS systems of different manufacturers under the same test environment, including front-end acquisition hardware comparison and post-processing software output result comparison; in the front-end acquisition hardware comparison, data of different acceleration sensors under the same test environment are collected, and acceleration and speed are compared respectively; in the post-processing software output result comparison, vibration results output by CMS systems of different manufacturers under the same test environment are collected, and acceleration and speed are compared respectively; Step S2: based on the test performance comparison results of CMS systems of different manufacturers, a method for converting output indicators of CMS systems of different manufacturers is proposed.

[0004] The existing patent CN111881176B discloses a marine nuclear power anomaly detection method based on logical distance characterization of the safe operating domain. By analyzing the historical operating data of the nuclear power system, typical operating condition data samples of the system are retrieved; a standard operating condition sample library is constructed, and characteristic parameters characterizing the operating status are selected; a logical distance calculation function is constructed; and the logical distance threshold for judging system anomalies, i.e., the safety domain attribution threshold, is determined through training and learning of historical operating data, simulation calculation, or verification and verification of collected abnormal data samples, so as to detect system anomalies online or retrieve abnormal conditions in the historical operating data of the system.

[0005] In summary, the above two existing patents have not solved the problems faced by the intelligent algorithm model of nuclear power plants, such as the difficulty in obtaining experimental data and the small amount of data, which affect the accuracy and reliability of intelligent algorithm verification. Summary of the invention

[0006] Based on the above technical problems, the present invention proposes a method and device for constructing fault diagnosis algorithm verification data for nuclear power plants, which solves the problems faced by the intelligent algorithm model of nuclear power plants in the prior art, such as difficulty in obtaining experimental data and small amount of data, which affects the accuracy and reliability of intelligent algorithm verification.

[0007] A method for constructing verification data of a fault diagnosis algorithm of a nuclear power plant, comprising:

[0008] Obtain experimental operation data of nuclear power plant system operation test bench;

[0009] Conduct simulation modeling on the experimental bench and obtain the corresponding simulation operation data;

[0010] According to the experimental operation data, the simulation operation data is subjected to noise processing;

[0011] Based on the experimental operation data and the simulation operation data after noise processing, the verification data for the fault diagnosis algorithm is constructed.

[0012] Furthermore, the experimental operation data of the nuclear power plant system operation test bench is obtained, including:

[0013] Establish a test bench for nuclear power plant system operation;

[0014] Analyze fault conditions during the operation of nuclear power plant systems;

[0015] Use the test bench to run the fault condition and obtain the experimental operation data under the fault condition.

[0016] Furthermore, the fault conditions during the operation of the nuclear power plant system include: one or more of: a charging pipeline leakage fault, a heat exchanger scaling fault, and a steam generator heat transfer tube rupture fault.

[0017] Furthermore, the experimental operation data includes one or more of: pressurizer temperature, pressurizer liquid level, pressurizer pressure, inlet pipeline flow, and steam generator narrow range liquid level.

[0018] Furthermore, the test bench is used to run the fault condition to obtain the experimental operation data under the fault condition, including:

[0019] Obtain the sensor monitoring points set during the operation of the real system of the nuclear power plant;

[0020] Determining the test bench monitoring point based on the sensor monitoring point;

[0021] Acquire the experimental operation data corresponding to the monitoring point of the experimental bench under the fault condition.

[0022] Furthermore, based on the experimental operation data, the simulation operation data is subjected to noise processing, including:

[0023] The simulation operation data is divided into the first type of data and the second type of data, the first type of data is the data of the simulation test bench monitoring point corresponding to the test bench monitoring point, and the second type of data is the data of the monitoring point other than the test bench monitoring point;

[0024] Determine the standard deviation of the noise based on the experimental running data and the first type of data;

[0025] Based on the standard deviation of the noise, the first type of data is subjected to noise addition processing.

[0026] Furthermore, the standard deviation of the noise is determined based on the experimental operation data and the first type of data, including:

[0027] Obtaining the standard deviation of the experimental running data and the standard deviation of the first category of data;

[0028] Based on the standard deviation of the experimental running data and the standard deviation of the first type of data, the standard deviation of the noise is determined by formula 1, wherein: Among them, std exp is the standard deviation of the experimental run data, std sim is the standard deviation of the first type of data, std noisy is the standard deviation of the noise.

[0029] Furthermore, it also includes:

[0030] De-noising the experimental run data;

[0031] Based on the experimental running data after denoising, the simulation running data is denoised.

[0032] Furthermore, the experimental running data is subjected to denoising, including:

[0033] The empirical mode decomposition method is used to denoise the experimental operation data.

[0034] Furthermore, based on the experimental operation data and the simulation operation data after noise processing, verification data for the fault diagnosis algorithm is constructed, including:

[0035] Determine a first correlation coefficient between the first type of data after noise processing and the experimental running data;

[0036] Obtain the fault condition label of the second type of data;

[0037] Determine a second correlation coefficient between the second type of data and the fault condition label;

[0038] The first type of data is screened according to the first correlation coefficient;

[0039] The second type of data is screened according to the second correlation coefficient;

[0040] Based on the screened first-category data, the screened second-category data and the experimental running data, verification data for the fault diagnosis algorithm is constructed.

[0041] Furthermore, the first correlation coefficient and the second correlation coefficient are determined by Formula 2. Formula 2:

[0042]

[0043] Among them, r xy is the correlation coefficient between variables x and y, x i ,y i is the i-th monitored value of variables x and y, are the sample means of variables x and y.

[0044] Furthermore, the first type of data is screened according to the first correlation coefficient, including:

[0045] Determining whether the first correlation coefficient is greater than a preset first correlation coefficient threshold;

[0046] If the first correlation coefficient is less than or equal to a preset first correlation coefficient threshold, data corresponding to the first correlation coefficient is removed.

[0047] A device for constructing verification data of a fault diagnosis algorithm of a nuclear power plant, comprising:

[0048] An acquisition module, used to acquire experimental operation data of a nuclear power plant system operation test bench;

[0049] The simulation module is used to simulate and model the experimental bench and obtain the corresponding simulation operation data;

[0050] A processing module, used for performing noise processing on the simulation operation data according to the experimental operation data;

[0051] The building module is used to build verification data for the fault diagnosis algorithm based on the experimental operation data and the simulation operation data after noise processing.

[0052] Furthermore, a module is obtained for:

[0053] Establish a test bench for nuclear power plant system operation;

[0054] Analyze fault conditions during the operation of nuclear power plant systems;

[0055] Use the test bench to run the fault condition and obtain the experimental operation data under the fault condition.

[0056] Furthermore, the fault conditions during the operation of the nuclear power plant system include: one or more of: a charging pipeline leakage fault, a heat exchanger scaling fault, and a steam generator heat transfer tube rupture fault.

[0057] Furthermore, the experimental operation data includes one or more of: pressurizer temperature, pressurizer liquid level, pressurizer pressure, inlet pipeline flow, and steam generator narrow range liquid level.

[0058] Furthermore, the test bench is used to run the fault condition to obtain the experimental operation data under the fault condition, including:

[0059] Obtain the sensor monitoring points set during the operation of the real system of the nuclear power plant;

[0060] Determining the test bench monitoring point based on the sensor monitoring point;

[0061] Acquire the experimental operation data corresponding to the monitoring point of the experimental bench under the fault condition.

[0062] Furthermore, the processing module is used to:

[0063] The simulation operation data is divided into the first type of data and the second type of data, the first type of data is the data of the simulation test bench monitoring point corresponding to the test bench monitoring point, and the second type of data is the data of the monitoring point other than the test bench monitoring point;

[0064] Determine the standard deviation of the noise based on the experimental running data and the first type of data;

[0065] Based on the standard deviation of the noise, the first type of data is subjected to noise addition processing.

[0066] Furthermore, the standard deviation of the noise is determined based on the experimental operation data and the first type of data, including:

[0067] Obtaining the standard deviation of the experimental running data and the standard deviation of the first category of data;

[0068] Based on the standard deviation of the experimental running data and the standard deviation of the first type of data, the standard deviation of the noise is determined by formula 1, wherein: Among them, std exp is the standard deviation of the experimental run data, std sim is the standard deviation of the first type of data, std noisy is the standard deviation of the noise.

[0069] Furthermore, it also includes:

[0070] De-noising the experimental run data;

[0071] Based on the experimental running data after denoising, the simulation running data is denoised.

[0072] Furthermore, the experimental running data is subjected to denoising, including:

[0073] The empirical mode decomposition method is used to denoise the experimental operation data.

[0074] Furthermore, a module is constructed to:

[0075] Determine a first correlation coefficient between the first type of data after noise processing and the experimental running data;

[0076] Obtain the fault condition label of the second type of data;

[0077] Determine a second correlation coefficient between the second type of data and the fault condition label;

[0078] The first type of data is screened according to the first correlation coefficient;

[0079] The second type of data is screened according to the second correlation coefficient;

[0080] Based on the screened first-category data, the screened second-category data and the experimental running data, verification data for the fault diagnosis algorithm is constructed.

[0081] Furthermore, the first correlation coefficient and the second correlation coefficient are determined by Formula 2. Formula 2:

[0082]

[0083] Among them, r xy is the correlation coefficient between variables x and y, x i ,y i is the i-th monitored value of variables x and y, are the sample means of variables x and y.

[0084] Furthermore, the first type of data is screened according to the first correlation coefficient, including:

[0085] Determining whether the first correlation coefficient is less than or equal to a preset first correlation coefficient threshold;

[0086] If the first correlation coefficient is less than or equal to a preset first correlation coefficient threshold, data corresponding to the first correlation coefficient is removed.

[0087] Based on the above technical solution, the present invention has at least the following beneficial effects:

[0088] 1. The present invention uses a simulation model to obtain a large amount of operating data at a low cost, and performs noise processing on the simulation operating data based on the experimental operating data. On the one hand, it can make the simulation operating data closer to the experimental operating data, and on the other hand, it can increase the randomness of the data by perturbing the data to prevent overfitting of the fault diagnosis algorithm training. Finally, based on the experimental operating data and the simulation operating data after noise processing, verification data for the fault diagnosis algorithm is jointly constructed. The verification data does not depend on the type of the fault diagnosis algorithm, and has good verification capabilities for various fault diagnosis algorithms, which can overcome the problems of small amount of experimental data and lack of credibility in the prior art.

[0089] 2. The present invention screens the first and second types of data by respectively judging whether the first correlation coefficient between the first type of data after noise addition and the experimental operation data, and the second correlation coefficient between the second type of data and the fault condition label meet the corresponding first correlation coefficient threshold range and second correlation coefficient threshold range. This can not only enhance the consistency between the simulation data and the experimental data, but also accurately identify additional monitoring points with potential value for fault diagnosis, thereby further enriching the diversity and depth of the data. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0091] Figure 1 A flow chart of a method for constructing fault diagnosis algorithm verification data for a nuclear power plant according to an embodiment of the present invention;

[0092] Figure 2 A flowchart of performing noise processing on simulation operation data according to experimental operation data in one embodiment of the present invention;

[0093] Figure 3 is the flowchart of the noise adding algorithm;

[0094] Figure 4A schematic diagram of a device for constructing fault diagnosis algorithm verification data for a nuclear power plant according to an embodiment of the present invention. DETAILED DESCRIPTION

[0095] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0096] The present invention is further described in detail below in conjunction with specific embodiments. These embodiments should not be construed as limiting the scope of protection claimed by the present invention.

[0097] Example

[0098] In order to solve the problems faced by the intelligent algorithm model of nuclear power plants in the prior art, such as difficulty in obtaining experimental data and small amount of data, which affect the accuracy and reliability of intelligent algorithm verification, the present invention proposes a method and device for constructing fault diagnosis algorithm verification data for nuclear power plants.

[0099] like Figure 1 FIG. 4 is a flow chart showing a method for constructing fault diagnosis algorithm verification data for a nuclear power plant according to an embodiment of the present invention. Figure 1 As shown, the method comprises the following steps:

[0100] S1, obtaining experimental operation data of a nuclear power plant system operation test bench. Further, obtaining experimental operation data of a nuclear power plant system operation test bench includes:

[0101] S101, establish a test bench for nuclear power plant system operation.

[0102] This embodiment aims at the automatic startup process of a nuclear power plant, and the need to verify the fault diagnosis algorithm from the time when the pressurizer establishes the steam cavity to the normal operation stage. An experimental bench for the operation of the nuclear power plant system is established to analyze the possible faults that may occur in the nuclear power plant system during this process.

[0103] S102, analyzing fault conditions during operation of the nuclear power plant system.

[0104] Furthermore, the fault conditions during the operation of the nuclear power plant system include: one or more of: charging pipeline leakage fault, heat exchanger scaling fault, steam generator heat transfer tube rupture fault. It should be understood that different nuclear power plant system operations may correspond to different fault conditions.

[0105] S103, using the experimental bench to operate the fault condition and obtain experimental operation data under the fault condition.

[0106] This embodiment selects three fault conditions of upper filling pipeline leakage, heat exchanger scaling failure, and steam generator heat transfer tube rupture for simulation, so as to form an experimental bench operation experiment. Among them, the simulation method of the upper filling tube leakage fault condition is to make a part of the liquid in the upper filling pipeline flow to the collection water tank, and the simulation method is to adjust the valve opening to simulate the upper filling pipe leakage, and use temperature sensors, pressure sensors and other instruments to measure, and archive the measured data of each instrument. The simulation method of the heat exchanger scaling fault condition is to make a part of the liquid flowing into the regenerative heat exchanger in the lower discharge pipeline bypass, and the simulation method is to adjust the valve opening to simulate the heat exchanger scaling failure, and use temperature sensors, pressure sensors and other instruments to measure, and archive the measured data of each instrument. The simulation method of the steam generator heat transfer tube rupture fault condition is to bypass a part of the coolant flowing into the steam generator, and the simulation method of the steam generator heat transfer tube rupture is to adjust the valve opening, and use temperature sensors, pressure sensors and other instruments to measure, and archive the measured data of each instrument.

[0107] When using the test bench to operate the fault condition and obtain the experimental operation data under the fault condition, the following steps are included: obtaining the sensor monitoring points set when the real system of the nuclear power plant is running; determining the monitoring points of the test bench based on the sensor monitoring points; and then obtaining the different time t under the fault condition from the monitoring points of the test bench. i Experimental run data X ij . The experimental operation data include: one or more of: pressurizer temperature, pressurizer liquid level, pressurizer pressure, inlet pipeline flow, steam generator narrow range liquid level. It should be understood that the experimental operation data is determined based on the sensor monitoring points set when the real system of the nuclear power plant is running, and the types of experimental operation data included in different embodiments may be different. As shown in Table 1, some experimental operation data under a certain fault condition obtained in this embodiment are shown.

[0108] Table 1 Part of the experimental operation data under a certain fault condition

[0109]

[0110] S2, simulate and model the experimental bench and obtain the corresponding simulation operation data.

[0111] Due to the difficulty of experimental cost control and procurement of experimental related equipment during the construction and use of the experimental bench, the sensors used in the experiment have the disadvantage of insufficient maximum acquisition frequency compared to the sensors used in the actual nuclear power plant, resulting in insufficient number of signal samples that can be collected per unit time. On the one hand, the representation of the signal waveform is not accurate enough, and on the other hand, the data volume is not enough for the verification of the intelligent algorithm. The present invention proposes to analyze the experimental bench built in the above step S1, use simulation software to simulate the basic physical characteristics, operating characteristics and working conditions of the experimental bench, obtain simulation operation data by simulating and modeling the experimental bench, and then supplement the experimental operation data with the above simulation operation data.

[0112] Furthermore, the entire system loop of the above-mentioned test bench is modeled through a simulation calculation program, specifically including the modeling of the core simulation body, main pump, steam generator simulation body, stabilizer simulation body, chemical volume system, various valves and pipelines. By debugging and modifying the model, it is ensured that the operating data of each device in the simulation model is close to the data under real working conditions when it is running stably. The basic modeling steps are as follows: (1) Analyze and record the geometric structure and other dimensional parameters of the structural components such as pipelines, core simulation body, main pump, steam generator simulation body, stabilizer simulation body, chemical volume system, etc. in the above-mentioned test bench; (2) Divide the control body of each test bench equipment and draw a node diagram, establish a model, and debug each part; (3) Connect each debugged equipment to complete the circuit simulation model of the test bench; (4) Run the constructed system simulation model to obtain and export the simulation results of pressure, temperature, etc. of each component of the test bench; (5) Draw the result curve and compare and analyze it with the experimental operation data.

[0113] The fault conditions introduced into the test bench in the above step S1 are: leakage of the upper charging pipeline, scaling of the heat exchanger heat transfer tube and rupture of the evaporator heat transfer tube. For subsequent work, it is necessary to select simulation data to facilitate fault diagnosis. For example, when simulating the leakage accident of the upper charging pipeline, the flow rate and temperature of the upper charging pipeline are extracted and fault diagnosis is performed; when simulating the rupture accident of the evaporator heat transfer tube, the data of the inlet and outlet flow rate and temperature of the steam generator are extracted; when simulating the structural failure of the heat exchanger heat transfer tube, the data of the inlet and outlet temperature of the hot section of the heat exchanger and the inlet and outlet temperature of the cold end are extracted and introduced into the fault diagnosis model.

[0114] The simulation operation data is obtained in the same format as the experimental operation data obtained from the test bench, including different time t i Data Y ij , an example of simulation running data is obtained as shown in Table 2.

[0115] Table 2 Simulation operation data example

[0116]

[0117] S3, performing noise processing on the simulation operation data according to the experimental operation data.

[0118] Since the noise in the real environment is often not caused by a single source, but a complex of noise from many different sources, the real noise is regarded as the sum of many random variables with different probability distributions, and each random variable is independent. Then, according to the central limit theorem, its normalized sum approaches the Gaussian distribution as the number of noise sources increases. In this complex situation, Gaussian white noise can achieve an approximate simulation when the real noise distribution is unknown. We choose to simulate the real noise of the experiment by adding Gaussian white noise to the simulation running data.

[0119] In order to make the simulation operation data closer to the actual experimental data after adding noise, and thus effectively supplement the experimental data set, this embodiment uses a simulation operation data noise addition algorithm. The algorithm is based on the standard deviation of the data characteristics, and through precise calculation and intelligent adjustment, it realizes the noise injection of the simulation operation data, ensures that the noise-added data is consistent with the experimental operation data in terms of fluctuation characteristics, enhances the diversity and authenticity of the data set, provides a richer and more reliable information basis for subsequent data analysis and fault diagnosis, and significantly improves the data consistency between the simulation environment and the actual experimental conditions.

[0120] Furthermore, if Figure 2 As shown, according to the experimental operation data, the simulation operation data is subjected to noise processing, including the following steps:

[0121] S301, dividing simulation operation data into first category data and second category data.

[0122] The first type of data is the data of the simulation test bench monitoring point corresponding to the test bench monitoring point, and the second type of data is the data of the monitoring point other than the test bench monitoring point. The first type of data needs to be denoised to simulate the random noise signal generated during the experiment, so that the simulation operation data is closer to the experimental data and the data authenticity is improved.

[0123] S302, determining a standard deviation of noise according to the experimental operation data and the first type of data.

[0124] In the process of injecting noise into the simulation operation data, this embodiment introduces Gaussian white noise, and the mean of the noise is set to 0 to keep the center position of the data unchanged. The calculation process of the standard deviation of the noise is as follows: obtain the standard deviation of the experimental operation data and the standard deviation of the first type of data; based on the standard deviation of the experimental operation data and the standard deviation of the first type of data, determine the standard deviation of the noise through formula 1, the formula 1, Among them, std exp is the standard deviation of the experimental run data, std sim is the standard deviation of the first type of data, std noisy is the standard deviation of the noise.

[0125] S303, performing noise processing on the first type of data based on the standard deviation of the noise.

[0126] like Figure 3 The flowchart of the noise adding algorithm is shown in , which is intended to determine whether the simulation running data needs to be noised by comparing the standard deviation of the simulation running data and the experimental running data. First, the standard deviation of the experimental running data and the first type of data is calculated respectively; the standard deviation std_simulation of the simulation running data is compared with the standard deviation std_experiment of the experimental running data; if the volatility of the simulation running data is less than the experimental running data (that is, std_simulation is less than std_experiment), it is necessary to add noise; if the volatility of the simulation running data is greater than or equal to the experimental running data, no noise is added; after determining that the simulation running data needs to be noised, the standard deviation std_noisy of the noise is calculated, and its value is the difference between the standard deviation of the experimental running data and the standard deviation of the simulation running data. Then, according to std_noisy, a Gaussian white noise x_gaussnoisy with a corresponding standard deviation is generated, and this noise is added to the simulation running data to obtain the noised data x_noised. Table 3 shows the simulation running data after some noise addition in this embodiment.

[0127] Table 3. Examples of adding noise to simulation operation data

[0128]

[0129]

[0130] By generating Gaussian white noise sequences with this specific standard deviation and integrating these noise sequences into the simulation data at an appropriate time length, it is possible to add random fluctuations similar to the experimental operation data to the simulation operation data without changing the overall trend of the simulation operation data, thereby improving the authenticity of the data. The subsequent use of the noisy simulation operation data as a supplement to the experimental operation data is more credible.

[0131] In another embodiment of the present invention, step S3 also includes the following processes: performing denoising processing on the experimental operation data; and performing denoising processing on the simulation operation data based on the denoised experimental operation data.

[0132] Since the data obtained from the experiment contains various noises due to factors such as the experimental environment and equipment, some useful information is submerged. Directly using the experimental operation data for the verification of the intelligent verification algorithm has low credibility. It is necessary to use statistical methods to verify the experimental operation data X ij De-noising is performed. Common methods for data denoising include standard deviation denoising, binning denoising, wavelet denoising, filter denoising, clustering algorithm denoising, empirical mode decomposition (EMD), etc. The principles for selecting data denoising methods based on the characteristics of experimental data are as follows: For continuous data, binning denoising is used to discretize continuous variables. When establishing a classification model, discretization features can make the model more stable and reduce the risk of overfitting. For smooth data, mean filtering or median filtering is used. By selecting the filter window size, calculating the mean or median in the window, replacing the central sample value and repeating it on the entire signal, periodic noise is reduced. For nonlinear and non-stationary signals, the empirical mode decomposition (EMD) method is used. The original signal can be directly decomposed without presetting the basis function, and the intrinsic mode function (IMF) component and the final residual term can be obtained by decomposition to achieve the purpose of filtering and denoising. Data with other characteristics are denoised using clustering algorithms, including DBSCAN algorithm based on density, k-means algorithm based on distance, and Z-score standardization and box plot denoising based on statistics.

[0133] Since the experimental operation data is nonlinear and non-steady, the empirical mode decomposition (EMD) method is used to denoise the experimental operation data. The specific steps are as follows:

[0134] (1) Finding extreme points: Obtain all the maximum and minimum values ​​of the signal sequence (experimental running data) through the Find Peaks algorithm; (2) Fitting envelope curves: Obtain two smooth peak / trough fitting curves, namely the upper envelope and lower envelope of the signal, through the maximum and minimum value groups of the signal sequence through cubic spline interpolation; (3) Mean envelope curve: Averaging the two extreme value curves to obtain the mean envelope curve; (4) Intermediate signal: Subtract the mean envelope curve from the original signal to obtain the intermediate signal; (5) Determine the intrinsic mode function (IMF): The IMF needs to meet two conditions: ① In the entire data segment, the number of extreme points and the number of zero crossings must be equal or the difference cannot exceed one at most. ② At any time, the average value of the upper envelope formed by the local maximum points and the lower envelope formed by the local minimum points is zero, that is, the upper and lower envelopes are locally symmetric with respect to the time axis; (6) IMF1: The first intermediate signal that satisfies the IMF condition is the first intrinsic mode function component IMF1 of the original signal (the new data is obtained by subtracting the envelope average from the original data. If there are still negative local maxima and positive local minima, it means that this is not yet an intrinsic mode function and needs to be further "screened"); (7) After obtaining the first IMF using the above method, subtract IMF1 from the original signal as the new original signal, and then through the above screening analysis, IMF2 can be obtained, and so on to complete the EMD decomposition.

[0135] Finally, the experimental running data after denoising is obtained, which eliminates the relatively obvious noise in the data and makes the experimental running data smoother. 01 As shown in Table 4, in step S302, the simulation operation data may be subjected to noise addition processing based on the experimental operation data after noise removal processing.

[0136] Table 4 Experimental run data denoising example

[0137]

[0138] S4, constructing verification data for the fault diagnosis algorithm based on the experimental operation data and the simulation operation data after noise processing.

[0139] Based on the experimental operation data and the simulation operation data after noise processing, the verification data for the fault diagnosis algorithm is constructed, including:

[0140] S401, determining a first correlation coefficient between the first type of data after noise processing and the experimental operation data.

[0141] Furthermore, in this embodiment, the first correlation coefficient and the second correlation coefficient are determined by Formula 2. Formula 2:

[0142]

[0143] Among them, r xy is the correlation coefficient between variables x and y, x i ,y i is the i-th monitored value of variables x and y, are the sample means of variables x and y.

[0144] It should be understood that in other embodiments, the first correlation coefficient and the second correlation coefficient may also be determined by other methods other than Formula 2.

[0145] This step can ensure that the first type of data has a high correlation with the experimental operation data, as shown in Table 5, which shows an example of the calculation result of the first correlation coefficient corresponding to each data at a certain time.

[0146] Table 5 Example of the first correlation coefficient

[0147]

[0148] S402, obtaining a fault condition label of the second type of data.

[0149] This embodiment marks the fault condition label according to the type of the fault condition, wherein the fault condition label 0 represents the normal condition, 1 represents the charging pipeline leakage fault condition, 2 represents the heat exchanger scaling fault condition, and 3 represents the steam generator heat transfer tube rupture fault condition.

[0150] S403, determining a second correlation coefficient between the second type of data and the fault condition label.

[0151] Table 6 shows an example of calculating the second correlation coefficient corresponding to each data at a certain time.

[0152] Table 6 Second correlation coefficient example

[0153]

[0154] S404, screening the first category of data according to the first correlation coefficient.

[0155] Furthermore, the first type of data is screened according to the first correlation coefficient, including the following sub-steps:

[0156] S4041, determining whether the first correlation coefficient is less than or equal to a preset first correlation coefficient threshold.

[0157] S4042: If the first correlation coefficient is less than or equal to a preset first correlation coefficient threshold, remove the data corresponding to the first correlation coefficient.

[0158] Taking the second correlation coefficient calculated in Table 6 as an example, when the second correlation coefficient threshold in this embodiment is 0.5, the correlation coefficient value of the "RCV downflow temperature" in the example is 0.84651, and the simulation operation data corresponding to the correlation coefficient should be retained. The other three data items, the pressurizer steam side temperature, the pressurizer spray temperature, and the heat pipe average temperature, are all less than the preset second correlation coefficient threshold, and the corresponding simulation operation data should be removed. The data screening can be completed according to the above steps S4041 to S4042.

[0159] S405, screening the second category of data according to the second correlation coefficient.

[0160] The method of screening the second category of data based on the second correlation coefficient is the same as the method of screening the first category of data in the above step S404, specifically including: determining whether the second correlation coefficient is less than or equal to a preset second correlation coefficient threshold; if the second correlation coefficient is less than or equal to the preset second correlation coefficient threshold, removing the data corresponding to the second correlation coefficient.

[0161] S405 , constructing verification data for the fault diagnosis algorithm based on the screened first category data, the screened second category data and the experimental operation data.

[0162] The above-mentioned first-category data after screening, the second-category data after screening, and the experimental operation data obtained in step S1 will be used as a verification data set for feature and classification label segmentation, and the training set and test set will be divided by cross-validation and other methods. Subsequently, the divided training set and test set can be used to test and verify the written fault diagnosis algorithm. When the accuracy of the simulation model meets the requirements of each model, the verification is reliable to ensure that the algorithm module and the software program can meet the requirements. This construction method can overcome the problems of small data volume and high data acquisition difficulty in the experimental operation data. On the basis of the experimental operation data, the simulation operation data is added to supplement the experimental operation data, which can ensure sufficient verification data volume and complete the verification of the intelligent algorithm. The saved fault data can also be used to develop the state monitoring and fault diagnosis algorithm offline.

[0163] According to whether the fault condition label is present, the verification data is divided into verification data without label and verification data with label. Specifically, the verification data sets formed in this embodiment are shown in Table 7 and Table 8, respectively. Table 7 is an example of verification data with label, and Table 8 is an example of verification data without label.

[0164] Table 7. Examples of unlabeled validation data

[0165]

[0166] Table 8: Example of labeled validation data

[0167]

[0168] like Figure 4 Schematic diagram of a device for constructing fault diagnosis algorithm verification data of a nuclear power plant according to an embodiment of the present invention is shown in FIG. , and the device includes an acquisition module 51 , a simulation module 52 , a processing module 53 and a construction module 54 .

[0169] The acquisition module 51 is used to acquire the experimental operation data of the nuclear power plant system operation test bench.

[0170] Furthermore, the acquisition module 51 is used to:

[0171] Establish a test bench for nuclear power plant system operation.

[0172] Analyze fault conditions during operation of nuclear power plant systems.

[0173] Use the test bench to run the fault condition and obtain the experimental operation data under the fault condition.

[0174] Furthermore, the fault conditions during the operation of the nuclear power plant system include: one or more of: a charging pipeline leakage fault, a heat exchanger scaling fault, and a steam generator heat transfer tube rupture fault.

[0175] Furthermore, the experimental operation data includes one or more of: pressurizer temperature, pressurizer liquid level, pressurizer pressure, inlet pipeline flow, and steam generator narrow range liquid level.

[0176] Furthermore, the test bench is used to run the fault condition to obtain the experimental operation data under the fault condition, including:

[0177] Obtain the sensor monitoring points set during the operation of the real system of the nuclear power plant;

[0178] Determining the test bench monitoring point based on the sensor monitoring point;

[0179] Acquire the experimental operation data corresponding to the monitoring point of the experimental bench under the fault condition.

[0180] The simulation module 52 is used to simulate and model the experimental bench and obtain corresponding simulation operation data.

[0181] The processing module 53 is used to perform noise processing on the simulation operation data according to the experimental operation data.

[0182] Furthermore, the processing module 53 is used to:

[0183] The simulation operation data are divided into first category data and second category data. The first category data are data of the simulation test bench monitoring points corresponding to the test bench monitoring points, and the second category data are data of monitoring points other than the test bench monitoring points.

[0184] The standard deviation of the noise is determined based on the experimental run data and the first type of data.

[0185] Based on the standard deviation of the noise, the first type of data is subjected to noise addition processing.

[0186] Furthermore, the standard deviation of the noise is determined based on the experimental operation data and the first type of data, including:

[0187] Obtaining the standard deviation of the experimental running data and the standard deviation of the first category of data;

[0188] Based on the standard deviation of the experimental running data and the standard deviation of the first type of data, the standard deviation of the noise is determined by formula 1, wherein: Among them, std exp is the standard deviation of the experimental run data, std sim is the standard deviation of the first type of data, std noisy is the standard deviation of the noise.

[0189] Furthermore, it also includes:

[0190] De-noising the experimental run data.

[0191] Based on the experimental running data after denoising, the simulation running data is denoised.

[0192] Furthermore, the experimental running data is subjected to denoising, including:

[0193] The empirical mode decomposition method is used to denoise the experimental operation data.

[0194] The construction module 54 is used to construct verification data for the fault diagnosis algorithm based on the experimental operation data and the simulation operation data after the noise processing.

[0195] Further, a module 54 is constructed to:

[0196] Determine the first correlation coefficient between the first type of data after noise processing and the experimental running data.

[0197] Get the fault condition label of the second type of data.

[0198] A second correlation coefficient between the second type of data and the fault condition label is determined.

[0199] The first type of data is screened according to the first correlation coefficient.

[0200] The second type of data is screened according to the second correlation coefficient.

[0201] Based on the screened first-category data, the screened second-category data and the experimental running data, verification data for the fault diagnosis algorithm is constructed.

[0202] Furthermore, the first correlation coefficient and the second correlation coefficient are determined by Formula 2. Formula 2:

[0203]

[0204] Among them, r xy is the correlation coefficient between variables x and y, x i ,y i is the i-th monitored value of variables x and y, are the sample means of variables x and y.

[0205] Furthermore, the first type of data is screened according to the first correlation coefficient, including:

[0206] It is determined whether the first correlation coefficient is less than or equal to a preset first correlation coefficient threshold.

[0207] If the first correlation coefficient is less than or equal to a preset first correlation coefficient threshold, data corresponding to the first correlation coefficient is removed.

[0208] It should be understood that the description of a device for constructing verification data of a fault diagnosis algorithm of a nuclear power plant is consistent with the description of a corresponding method for constructing verification data of a fault diagnosis algorithm of a nuclear power plant, so it will not be repeated in this embodiment.

[0209] In summary, it can be seen from the above description that the above embodiments of the present invention achieve the following technical effects:

[0210] 1. The present invention uses a simulation model to obtain a large amount of operating data at a low cost, and performs noise processing on the simulation operating data based on the experimental operating data. On the one hand, it can make the simulation operating data closer to the experimental operating data, and on the other hand, it can increase the randomness of the data by perturbing the data to prevent overfitting of the fault diagnosis algorithm training. Finally, based on the experimental operating data and the simulation operating data after noise processing, verification data for the fault diagnosis algorithm is jointly constructed. The verification data does not depend on the type of the fault diagnosis algorithm, and has good verification capabilities for various fault diagnosis algorithms, which can overcome the problems of small amount of experimental data and lack of credibility in the prior art.

[0211] 2. The present invention screens the first and second types of data by respectively judging whether the first correlation coefficient between the first type of data after noise addition and the experimental operation data, and the second correlation coefficient between the second type of data and the fault condition label meet the corresponding first correlation coefficient threshold range and second correlation coefficient threshold range. This can not only enhance the consistency between the simulation data and the experimental data, but also accurately identify additional monitoring points with potential value for fault diagnosis, thereby further enriching the diversity and depth of the data.

[0212] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0213] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0214] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, which can be embodied in any computer-readable medium for use by an instruction execution system, apparatus or device (such as a computer-based system, a system including a processor or other system that can fetch instructions from an instruction execution system, apparatus or device and execute instructions), or used in combination with these instruction execution systems, apparatuses or devices.

[0215] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0216] It should be noted that, in the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

Claims

1. A method for constructing fault diagnosis algorithm verification data for a nuclear power plant, characterized in that: include: Obtain experimental operation data of nuclear power plant system operation test bench; Performing simulation modeling on the experimental bench and obtaining corresponding simulation operation data; According to the experimental operation data, performing noise processing on the simulation operation data; Based on the experimental operation data and the simulation operation data after noise processing, verification data for a fault diagnosis algorithm is constructed.

2. The method according to claim 1, characterized in that: Obtain experimental operation data of the nuclear power plant system operation test bench, including: Establishing a test bench for the operation of the nuclear power plant system; Analyzing fault conditions during operation of the nuclear power plant system; The fault condition is run using the test bench to obtain experimental operation data under the fault condition.

3. The method according to claim 2, characterized in that The fault conditions during the operation of the nuclear power plant system include: one or more of: a charging pipeline leakage fault, a heat exchanger scaling fault, and a steam generator heat transfer tube rupture fault.

4. The method according to claim 2, characterized in that: The experimental operation data includes: One or more of pressurizer temperature, pressurizer liquid level, pressurizer pressure, inlet line flow, steam generator narrow range liquid level.

5. The method according to claim 2, characterized in that: Using the test bench to run the fault condition and obtain experimental operation data under the fault condition, includes: Obtain the sensor monitoring points set during the operation of the real system of the nuclear power plant; Determining the test bench monitoring point based on the sensor monitoring point; Acquire the experimental operation data corresponding to the monitoring point of the experimental bench under the fault condition.

6. The method according to claim 5, characterized in that According to the experimental operation data, the simulation operation data is subjected to noise adding processing, including: Dividing the simulation operation data into a first type of data and a second type of data, wherein the first type of data is data of a simulation test bench monitoring point corresponding to the test bench monitoring point, and the second type of data is data of a monitoring point other than the test bench monitoring point; Determining a standard deviation of noise based on the experimental operation data and the first type of data; The first type of data is subjected to noise adding processing based on the standard deviation of the noise.

7. The method according to claim 6, characterized in that Determining a standard deviation of noise according to the experimental operation data and the first type of data includes: Obtaining the standard deviation of the experimental running data and the standard deviation of the first category of data; Based on the standard deviation of the experimental running data and the standard deviation of the first type of data, the standard deviation of the noise is determined by formula 1, wherein: Among them, std exp is the standard deviation of the experimental run data, std sim is the standard deviation of the first type of data, std noisy is the standard deviation of the noise.

8. The method according to claim 6, characterized in that Also includes: Performing denoising on the experimental running data; According to the experimental operation data after denoising, the simulation operation data is subjected to denoising.

9. The method according to claim 8, characterized in that The experimental running data is subjected to denoising processing, including: The experimental running data is denoised using an empirical mode decomposition method.

10. The method according to any one of claims 6 to 9, characterized in that: Based on the experimental operation data and the simulation operation data after noise processing, construct verification data for the fault diagnosis algorithm, including: Determine a first correlation coefficient between the first type of data after noise processing and the experimental running data; Obtaining a fault condition label of the second type of data; Determining a second correlation coefficient between the second type of data and the fault condition label; screening the first category of data according to the first correlation coefficient; screening the second type of data according to the second correlation coefficient; Based on the screened first-category data, the screened second-category data and the experimental operation data, verification data for a fault diagnosis algorithm is constructed.

11. The method according to claim 10, characterized in that The first correlation coefficient and the second correlation coefficient are determined by formula 2, wherein: Among them, r xy is the correlation coefficient between variables x and y, x i ,y i is the i-th monitored value of variables x and y, are the sample means of variables x and y.

12. The method according to claim 10, characterized in that Screening the first category of data according to the first correlation coefficient includes: Determining whether the first correlation coefficient is less than or equal to a preset first correlation coefficient threshold; If the first correlation coefficient is less than or equal to the preset first correlation coefficient threshold, data corresponding to the first correlation coefficient is removed.

13. A device for constructing fault diagnosis algorithm verification data for a nuclear power plant, characterized in that: include: An acquisition module, used to acquire experimental operation data of a nuclear power plant system operation test bench; A simulation module, used to simulate and model the experimental bench and obtain corresponding simulation operation data; A processing module, used for performing noise addition processing on the simulation operation data according to the experimental operation data; A construction module is used to construct verification data for a fault diagnosis algorithm based on the experimental operation data and the simulation operation data after noise processing.

14. The device according to claim 13, characterized in that The acquisition module is used to: Establishing a test bench for the operation of the nuclear power plant system; Analyzing fault conditions during operation of the nuclear power plant system; The fault condition is run using the test bench to obtain experimental operation data under the fault condition.

15. The device according to claim 14, characterized in that The fault conditions during the operation of the nuclear power plant system include: one or more of: a charging pipeline leakage fault, a heat exchanger scaling fault, and a steam generator heat transfer tube rupture fault.

16. The device according to claim 14, characterized in that The experimental operation data include: one or more of: pressurizer temperature, pressurizer liquid level, pressurizer pressure, inlet pipeline flow, and steam generator narrow range liquid level.

17. The device according to claim 14, characterized in that Using the test bench to run the fault condition and obtain experimental operation data under the fault condition, includes: Obtain the sensor monitoring points set during the operation of the real system of the nuclear power plant; Determining the test bench monitoring point based on the sensor monitoring point; Acquire the experimental operation data corresponding to the monitoring point of the experimental bench under the fault condition.

18. The device according to claim 17, characterized in that The processing module is used to: Dividing the simulation operation data into a first type of data and a second type of data, wherein the first type of data is data of a simulation test bench monitoring point corresponding to the test bench monitoring point, and the second type of data is data of a monitoring point other than the test bench monitoring point; Determining a standard deviation of noise based on the experimental operation data and the first type of data; The first type of data is subjected to noise adding processing based on the standard deviation of the noise.

19. The device according to claim 18, characterized in that Determining a standard deviation of noise according to the experimental operation data and the first type of data includes: Obtaining the standard deviation of the experimental running data and the standard deviation of the first category of data; Based on the standard deviation of the experimental running data and the standard deviation of the first type of data, the standard deviation of the noise is determined by formula 1, wherein: Among them, std exp is the standard deviation of the experimental run data, std sim is the standard deviation of the first type of data, std noisy is the standard deviation of the noise.

20. The device according to claim 18, characterized in that Also includes: Performing denoising on the experimental running data; According to the experimental operation data after denoising, the simulation operation data is subjected to denoising.

21. The device according to claim 20, characterized in that The experimental running data is subjected to denoising processing, including: The experimental running data is denoised using an empirical mode decomposition method.

22. The device according to any one of claims 18 to 21, characterized in that The building blocks are used to: Determine a first correlation coefficient between the first type of data after noise processing and the experimental running data; Obtaining a fault condition label of the second type of data; Determining a second correlation coefficient between the second type of data and the fault condition label; screening the first category of data according to the first correlation coefficient; screening the second type of data according to the second correlation coefficient; Based on the screened first-category data, the screened second-category data and the experimental operation data, verification data for a fault diagnosis algorithm is constructed.

23. The device according to claim 22, characterized in that The first correlation coefficient and the second correlation coefficient are determined by formula 2, wherein: Among them, r xy is the correlation coefficient between variables x and y, x i ,y i is the i-th monitored value of variables x and y, are the sample means of variables x and y.

24. The device according to claim 22, characterized in that Screening the first category of data according to the first correlation coefficient includes: Determining whether the first correlation coefficient is less than or equal to a preset first correlation coefficient threshold; If the first correlation coefficient is less than or equal to the preset first correlation coefficient threshold, data corresponding to the first correlation coefficient is removed.

Images