Event early warning method and device based on nuclear power control logic

Abstract

The invention relates to the technical field of nuclear power station event early warning, aims to solve the problems of how to detect whether the current operation state is normal or not and how to predict the occurrence time of an abnormal event in the operation process of a unit, and provides an event early warning method and device based on nuclear power control logic. The method comprises the following steps: acquiring real-time data, calculating the real-time data, judging whether an abnormal event occurs or not, and when judging that the abnormal event occurs, predicting the change trend of the real-time data by adopting a key working condition parameter trend prediction algorithm based on a self-adaptive linear model so as to predict the occurrence time of the abnormal event. And outputting the occurrence time of the abnormal event and giving an alarm. According to the invention, abnormal events such as shutdown, shutdown and load shedding in the operation process can be monitored, analyzed and pre-warned, whether the current operation state is normal or not can be detected, the occurrence time of the abnormal events can be predicted, and thus auxiliary support is provided for ensuring stable and safe operation of a power plant.

Description

Technical Field

[0001] The present invention relates to the technical field of nuclear power plant event warning technology, and in particular to an event warning method and device based on nuclear power control logic. Background Art

[0002] During normal operation of a nuclear power plant, unexpected protection actions such as load shedding, shutdown, and reactor shutdown may occur due to equipment failure, instrumentation and control system failure, power grid failure, fire, insufficient personnel skills, and human error.

[0003] At present, the detection of abnormal conditions during the operation of the unit mainly relies on the alarm information of the DCS system. The corresponding alarm is triggered only when the threshold is reached, and there is a lack of trend analysis. In addition, although most of the typical abnormal events of the unit have different precursors before causing accidents, the DCS alarm cannot fully cover these events, nor can it prompt the operator in time according to the correlation fluctuations of parameters, so it is difficult to attract attention in the early stage of abnormal events.

[0004] Traditional monitoring methods are based on single-variable control charts, which are no longer able to reflect changes in complex industrial processes. Modern process monitoring is based on data-driven methods, and data-driven metrics are derived directly from process data. Although the big data of modern industrial systems brings rich information, engineers and operators cannot empirically draw an assessment of process operation based on this massive amount of data. Summary of the invention

[0005] The main purpose of the present invention is to provide an event warning method and device based on nuclear power control logic to solve the problem of how to detect whether the current operating state is normal during the operation of the unit.

[0006] Another object of the present invention is to provide an event warning method and device based on nuclear power control logic to solve the problem of how to predict the time of occurrence of the abnormal event.

[0007] In order to achieve the above object, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides an event early warning method based on nuclear power control logic, comprising:

[0009] S100, obtaining real-time data;

[0010] S200, calculating the real-time data to determine whether an abnormal event will occur;

[0011] S300, when it is determined that an abnormal event will occur, a key operating condition parameter trend prediction algorithm based on an adaptive linear model is used to predict the change trend of real-time data to predict the time when the abnormal event will occur;

[0012] S400: Output the time when the abnormal event occurs and generate an alarm.

[0013] In some embodiments, an abnormal monitoring algorithm based on multi-modal DPCA is used to monitor abnormal events in the operation of the unit, and the abnormal events include at least one of a shutdown event, a shutdown event, and a load rejection event.

[0014] In some embodiments, S200 includes:

[0015] S210, grouping the real-time data to obtain several groups of data;

[0016] S220, using an abnormal monitoring algorithm based on multi-modal DPCA to calculate the covariance matrix of several groups of data;

[0017] S230, based on the covariance matrix, using the anomaly monitoring algorithm based on multimodal DPCA to calculate the Hotelling statistic and the residual statistic of each group of data;

[0018] S240: judging whether an abnormal event will occur based on the Hotelling statistics and residual statistics of the plurality of data groups and the preset Hotelling statistics thresholds and residual statistics thresholds corresponding to the plurality of data groups.

[0019] In some embodiments, S240 includes:

[0020] S241, obtaining a historical sample data set, and dividing the historical sample data set into several groups of initial data sets;

[0021] S242, calculating the covariance matrix of several groups of initial data sets;

[0022] S243, calculating the eigenvalues ​​and eigenvectors of several groups of covariance matrices;

[0023] S244, sorting several groups of eigenvalues, screening several groups of eigenvalues ​​using cumulative variance contribution rates, merging several groups of initial data sets, and generating multiple groups of intermediate data sets;

[0024] S245, modeling the multiple groups of intermediate data sets to obtain multiple groups of PCA models, and then calculating the residuals between the groups of PCA models;

[0025] S246, merging the multiple sets of intermediate data sets based on a preset threshold and the residuals between the PCA models to generate multiple sets of new data sets;

[0026] S247, repeat S242 to S246 until no new multiple sets of data sets are generated, and then execute S248;

[0027] S248. Using the PCA method, the Hotelling statistic threshold and the residual statistic threshold of each group of PCA models are finally calculated.

[0028] In some embodiments, the process of establishing a fault detection model includes:

[0029] (1) Denoted as the initial data set, the PCA model is constructed through the following iterative process;

[0030] (2) For each group PCA model The number of principal components retained was determined by cross-validation and the test data set was used as part of;

[0031] (3) The similarity between the two PCA models is compared, and the residual is used as the similarity measure:

[0032]

[0033] (4) If the residual is less than the threshold f, the corresponding data groups are merged and the newly generated data groups are expressed as Go to step (2), otherwise stop;

[0034] (5) Determine the detection threshold: The value of the threshold is the maximum residual of the corresponding modality training set itself.

[0035] It should be noted that steps (1)-(5) here describe how to build a fault detection model. This part of the model building is offline. The actual monitoring process is to input online data into the model in time sequence, and the monitoring itself is online. The following is a description of the online fault monitoring method.

[0036] Based on the above multimodal PCA model, the anomaly detection method based on multimodal PCA is as follows:

[0037] (1) When online data is received, the residual SPE between it and each group mode is calculated by formula (1);

[0038] (2) Detection logic: If at least one set of residuals is smaller than the threshold, it is a normal sample, otherwise it is a faulty sample.

[0039] In some embodiments, S300 includes:

[0040] S310, when it is determined that an abnormal event will occur, the empirical mode decomposition method is used to denoise the real-time data and extract trends to obtain preprocessed data;

[0041] S320, inputting the preprocessed data into a pre-built linear model to obtain an estimated value of the target variable;

[0042] S330: Calculate the time when the abnormal event occurs based on the alarm threshold of the target variable and the estimated value of the target variable.

[0043] In some embodiments, the key operating parameter trend prediction algorithm based on the adaptive index model includes:

[0044] (1) Collect t k At time s, the data z with window length s k , perform empirical mode decomposition on it and extract the IMF component containing the overall trend;

[0045] (2) Assign initial values ​​to the linear model parameters;

[0046] (3) At time k, based on t ks to k Data at the moment Train a linear model f(t i , φ, θ);

[0047] (4) Through the linear model f(t i ,φ,θ) estimate k s The predicted value at time k as follows:

[0048]

[0049] (5) Wait for the next moment data to arrive and jump to step (1);

[0050] At time k, the pair Solve and get the optimal solution (φ * ,θ * )k, and get the estimated value of the target variable at time k:

[0051] z k =φ * +θ * t k ;

[0052] The alarm threshold ξ of the known target variable satisfies:

[0053] ξ=φ * +θ * (t k +T k );

[0054] Calculate the remaining shutdown time:

[0055]

[0056] Where, T k represents the time when the abnormal event occurs, ξ represents the alarm threshold of the target variable, φ * ,θ * It represents the optimal solution of the random variable at time k, that is, the estimated value of the target variable at time k.

[0057] In a second aspect, the present invention provides an event warning device based on nuclear power control logic, comprising:

[0058] Data acquisition module, used to acquire real-time data;

[0059] The abnormality judgment module is used to calculate the real-time data using the abnormality monitoring algorithm based on multi-modal DPCA to determine whether an abnormal event will occur;

[0060] The abnormality prediction module is used to predict the change trend of real-time data and the time when the abnormal event will occur by using the key operating parameter trend prediction algorithm based on the adaptive linear model when it is judged that an abnormal event will occur;

[0061] The output module is used to output the time when the abnormal event occurs and to generate an alarm.

[0062] In a third aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the event warning method based on nuclear power control logic when executing the computer-readable instructions.

[0063] In a fourth aspect, the present invention provides a computer-readable storage medium having computer-readable instructions stored thereon, and the computer-readable instructions, when executed, implement the steps of the event warning method based on nuclear power control logic.

[0064] Compared with the prior art, the event warning method and device based on nuclear power control logic provided by the present invention has the following beneficial effects:

[0065] The present invention can monitor, analyze and warn of abnormal events such as shutdown, outage and load shedding during operation, detect whether the current operating status is normal, and predict the time when the abnormal event will occur, thereby providing auxiliary support for ensuring the smooth and safe operation of the power plant. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for the technical description.

[0067] Figure 1 A flow chart of an event early warning method based on nuclear power control logic provided by the present invention;

[0068] Figure 2 A schematic diagram of the initial grouping provided by the present invention;

[0069] Figure 3 A DPCA fault detection flow chart provided by the present invention;

[0070] Figure 4 The overall roadmap for trend prediction of key operating parameters provided by the present invention;

[0071] Figure 5 The EMD flow chart provided by the present invention;

[0072] Figure 6 A schematic diagram of an event warning device based on nuclear power control logic provided by the present invention. DETAILED DESCRIPTION

[0073] The following is further explained in detail through specific implementation methods.

[0074] like Figure 1 As shown, the present invention provides an event warning method based on nuclear power control logic, which is developed for major operating events of nuclear power plants, and can identify abnormal changes in equipment parameters related to various scenarios such as shutdown, outage, and load shedding, and can correctly predict possible events and the estimated time of occurrence. The warning method specifically includes the following steps:

[0075] S100, obtaining real-time data;

[0076] S200, using an abnormal monitoring algorithm based on multi-modal DPCA to calculate the real-time data to determine whether an abnormal event will occur, wherein the abnormal event is at least one of a reactor shutdown event, a shutdown event, and a load rejection event;

[0077] S300, when it is determined that an abnormal event will occur, a key operating condition parameter trend prediction algorithm based on an adaptive linear model is used to predict the change trend of real-time data to predict the time when the abnormal event will occur;

[0078] S400: Output the time when the abnormal event occurs and generate an alarm.

[0079] By performing anomaly detection on event-related parameters and dividing the input parameter range according to different events, data processing, feature extraction, multivariate statistical analysis, dynamic principal component analysis and other algorithms are used to carry out research on shutdown events and four scenarios. Then, trend prediction is introduced to determine how long it will take for an abnormal event to occur. It performs regression modeling based on historical power plant operation data and the correlation between parameters, predicts the operating trend of relevant parameters of specific events in the future, and finally merges the information and outputs it as early warning information, thereby providing auxiliary support for ensuring the smooth and safe operation of the power plant.

[0080] In one embodiment, in step S100, the real-time data is the real-time operation data of the unit or the twin system data of the product platform.

[0081] In one embodiment, in step S200, when an abnormal operating condition caused by equipment failure or failure occurs in a nuclear power plant, the online monitoring data will deviate from the behavior pattern under normal operating conditions. Therefore, the present invention monitors the real-time data to determine whether an abnormal event will occur. Since a nuclear reactor has multiple operating modes under normal operating conditions, the present invention uses an abnormal monitoring algorithm based on multi-modal DPCA to monitor abnormalities.

[0082] In one embodiment, in step S200, after the abnormality detection is performed by establishing a model offline, the model is then used for online detection. Specifically, in the offline modeling stage, a monitoring model of normal operating conditions is established using historical data of normal operating conditions, and in the online monitoring stage, the residual signal corresponding to the online data is generated using the above monitoring model to indicate the degree to which the online data deviates from the normal operating conditions, thereby determining an abnormal event alarm.

[0083] In one embodiment, step S200 specifically includes:

[0084] S210, grouping the real-time data to obtain several groups of data;

[0085] S220, using an abnormal monitoring algorithm based on multi-modal DPCA to calculate the covariance matrix of several groups of data;

[0086] S230, based on the covariance matrix, using the anomaly monitoring algorithm based on multimodal DPCA to calculate the Hotelling statistic and the residual statistic of each group of data;

[0087] S240: judging whether an abnormal event will occur based on the Hotelling statistics and residual statistics of the plurality of data groups and the preset Hotelling statistics thresholds and residual statistics thresholds corresponding to the plurality of data groups.

[0088] In one embodiment, a method for calculating Hotelling statistic thresholds and residual statistic thresholds for a plurality of sets of data includes:

[0089] S241, obtaining a historical sample data set, and dividing the historical sample data set into several groups of initial data sets;

[0090] S242, calculating the covariance matrix of several groups of initial data sets;

[0091] S243, calculating the eigenvalues ​​and eigenvectors of several groups of covariance matrices;

[0092] S244, sorting the plurality of groups of eigenvalues, screening the plurality of groups of eigenvalues ​​by using the cumulative variance contribution rate, and merging the plurality of groups of initial data sets to generate a plurality of groups of intermediate data sets;

[0093] S245, modeling the multiple groups of intermediate data sets to obtain multiple groups of PCA models, and then calculating the residuals between the groups of PCA models;

[0094] S246, merging the multiple sets of intermediate data sets based on a preset threshold and the residuals between the PCA models to generate multiple sets of new data sets;

[0095] S247, repeat S242 to S246 until no new multiple sets of data sets are generated, and then execute step S248;

[0096] S248. Using the PCA method, the Hotelling statistic threshold and the residual statistic threshold of each group of PCA models are finally calculated.

[0097] In one embodiment, during the offline detection phase, the training data is first grouped, and then the groups of data are merged according to the residual of each group of data. Finally, each group of data is modeled into a mode, and the maximum residual of each mode is saved as a threshold.

[0098] Specifically, in the initial grouping stage, Figure 2 As shown in the figure, the historical data is integrated, the minimum group size H is set, and the massive data is automatically divided into groups. At the same time, the number of eigenvalues ​​of the group is calculated. If the contribution rate is greater than 99% (default value), the number of principal components is still equal to the number of variables input. It is considered that the amount of data in this group is too small, and it continues to be merged with the next group. At the same time, the contribution rate and the number of principal components of the newly merged group are calculated until the conditions for being called an independent group are met. After the initial grouping is completed, the first group after the grouping is modeled, the training set residual is calculated, and a threshold is automatically set according to the residual, such as the mean of the residual as the threshold. The data of other groups are input into the model established by the first group, and the residual is calculated. If the residual is less than the threshold, it is merged with the first group, and all the merged new first groups are modeled again until no new groups are generated. At this time, the first group of data is modeled, the model is stored in the modal library, and the first group of data is removed. The second group of data is modeled, and the above operation is repeated until no new model is generated.

[0099] In order to better understand the detection process, the following Figure 3 The fault detection process (offline modeling) is described in detail. The following is the process of establishing the fault monitoring model:

[0100] 1. Represented as the initial data group, where the superscript i represents the i-th grouping and the subscript n iRepresents the number of grouping groups, and the PCA model is constructed through the following iterative process;

[0101] 2. For each group PCA model The number of principal components retained can be determined by cross-validation and the test data set is used as part of;

[0102] 3. Compare the similarity between the two PCA models, and use the residual as the similarity measure:

[0103]

[0104] 4. If the residual is less than the threshold f, the corresponding data groups are merged and the newly generated data groups are expressed as Go to step 2, otherwise stop.

[0105] 5. Determine the detection threshold: The value of the threshold is the maximum residual of the corresponding modality training set itself.

[0106] Based on the above multimodal PCA model, the anomaly detection method based on multimodal PCA is as follows:

[0107] 1. When online data is received, the residual SPE between it and each group mode is calculated by formula (1);

[0108] 2. Detection logic: If at least one set of residuals is smaller than the threshold, it is a normal sample, otherwise it is a faulty sample.

[0109] In one embodiment, step S300 specifically includes:

[0110] S310, when it is determined that an abnormal event will occur, the empirical mode decomposition method is used to denoise the real-time data and extract trends to obtain pre-processed data;

[0111] S320, inputting the preprocessed data into a pre-built linear model to obtain an estimated value of the target variable;

[0112] S330: Calculate the time when the abnormal event occurs based on the alarm threshold of the target variable and the estimated value of the target variable.

[0113] In one embodiment, the linear model is:

[0114] z i =φ+θt i =f(t i , φ, θ);

[0115] In the formula, φ and θ are random variables, z iis the estimated value of the target variable at time i.

[0116] In one embodiment, the objective function of the linear model is:

[0117]

[0118] In the formula, y i is the true value of the target variable at time i, N is the window length, f(t i , φ, θ) are the estimated values ​​of the target variable.

[0119] After detecting an abnormal operating condition, the changing trend of key operating condition parameters is predicted, and then the occurrence time of events such as automatic shutdown and pump shutdown is predicted, which has important reference significance for timely maintenance after abnormal operating condition alarm. The difficulty in predicting abnormal operating condition trends lies in the fact that abnormal events are extremely rare in historical data, and it is also difficult to verify whether the abnormal operating condition data generated by the simulator can accurately reflect the actual situation. To this end, the present invention uses online data after abnormal operating condition alarm to realize online adaptive modeling and prediction of key operating condition parameter trend changes in the absence of abnormal operating condition historical data. Figure 4 As shown, the main steps include S310 trend extraction, S320 online modeling and S330 prediction of remaining unshutdown time (including the last two modules starting with prediction).

[0120] Since the data of the target parameters are approximately linear under fault conditions and in transitional conditions to normal conditions, the present invention adopts a linear model to make the solved parameters closer to the actual situation, thereby making the prediction of the remaining outage time more accurate.

[0121] Furthermore, since the noise and local oscillation contained in the real-time data itself will interfere with trend prediction, the present invention adopts the empirical mode decomposition (EMD) method to denoise the data and extract trends.

[0122] Specifically, the Hilbert-Huang transform (HHT) based on EMD is a method for analyzing nonlinear and non-stationary signals. HHT consists of two parts, namely EMD and Hilbert spectrum analysis (HAS). Its characteristic is that it uses EMD to linearize and stabilize nonlinear and non-stationary data to obtain the intrinsic mode function (IMF-Intrinsic mode function) component, and retains the characteristics of the data itself during the decomposition process. This decomposition method decomposes the signal based on the scale characteristics of the signal itself, so it can analyze the characteristics of the signal as quickly as possible and has good local adaptability.

[0123] The IMF extracted from the original signal using the EMD method highlights the local characteristic information of the original signal. IMF has the following characteristics:

[0124] In the entire time history, the number of zero crossings of the waveform is equal to the number of extreme value points, or the difference between the two is at most 1;

[0125] At any point on the signal, the mean of the upper envelope defined by the local maximum and the lower envelope defined by the local minimum is 0.

[0126] Of the two constraints mentioned above, the first one is actually very similar to the narrowband requirement of the commonly used stationary Gaussian process. The second one is a new constraint, which changes the commonly used global signal condition to a partial signal condition. This condition allows the signal to be decomposed and the relevant results to be obtained. At the same time, the decomposed multi-layer IMF is also a result with physical significance.

[0127] However, not all signals meet the requirements of the two IMFs mentioned above, because at any time a signal generally includes more than one mode.

[0128] The EMD method decomposes signals based on the following three assumptions:

[0129] The signal has at least one maximum point and one minimum point;

[0130] The characteristic time scale is set as the time interval between adjacent extreme value points;

[0131] If the signal has no extreme points but only inflection points, then before decomposing it, first differentiate the original signal once or multiple times to obtain the extreme points, and then integrate the results to get the corresponding components.

[0132] EMD is specifically broken down through a process called "sieving". Figure 5 , the specific operation process is as follows:

[0133] The first step of processing is to find all the maximum and minimum points of the original signal s(t), then connect all the maximum points with a cubic spline curve to form the upper envelope of s(t), and use the same method to connect all the minimum points to get the lower envelope of s(t). According to the local maximum and local minimum of the signal s(t), the average values ​​of the upper envelope v1(t) and the lower envelope v2(t) are calculated:

[0134]

[0135] Then find s(t) and m 11 The difference h 11 :

[0136] h 11 =s(t)-m 11 (Formula 3)

[0137] In an ideal situation, 11 It should meet the IMF definition, because h 11 The construction of should satisfy the symmetry. After the first round of extraction, this value has a probability of becoming a local maximum. The new local extreme point generated in equation (3) indicates that the correct mode that meets the requirements is lost in the initial step. After repeated decomposition and extraction, the extraction process can restore the original signal and represent it as a low-amplitude driving wave.

[0138] The decimation process has two goals: to remove the driving wave and to make the waveform more symmetrical. To achieve these two goals, the decimation process must be repeated as many times as required to make an IMF after subtracting the decimated signal. In the subsequent decimation process, h 11 It can only be considered as an IMF prototype if h 11 It is not IMF, so it is considered as the new s(t) and equation (2) is repeated 2k times.

[0139] h 1k =h 1(k-1) -m 1k (Formula 4)

[0140] In formula (4), h 1k and h 1(k-1) are the data obtained from the kth and k-1th screening, respectively, m 1k h 1(k-1) The median (average) of the upper and lower envelopes, S calculated by formula (5) D The value can be used to see whether the result of each "screening" is an IMF component that meets the requirements:

[0141]

[0142] A critical stopping criterion is needed here: if S D If the value is between 0.2 and 0.3, let c1 = h 1k Consider it as an IMF, and make r = s(t)-c1. Then consider r as the new s(t). Repeat the above process to get the second IMF. c2 , the third IMF c3 …until r(t) becomes a monotonic function or the absolute value of r(t) is less than a certain threshold and can be regarded as a measurement error. Then the signal composed of the superposition of the IMF components and the residual of each layer is This shows the completeness of EMD decomposition. In the above formula, r is the residual, representing the average trend in the signal, and each IMF component c1 represents the content waveform of the signal in each frequency band. The smaller the i value, the higher the frequency. The partial time distribution of different frequency elements changes with the change of the signal itself.

[0143] After EMD processing, the component representing the overall trend in IMF is extracted as preprocessed data and input into the subsequent predictor.

[0144] After obtaining the preprocessed data, i The target variable trend curve at the moment can be expressed by the following linear model:

[0145] z i =φ+θt i =f(t i , φ, θ) (Formula 6)

[0146] In the formula, φ and θ are random variables, z i is the estimated value of the target variable at time i. By introducing a time domain window, the objective function of the linear model can be described as an online optimization problem:

[0147]

[0148] Among them, y i is the true value of the target variable at time i, N is the window length, f(t i , φ, θ) are the estimated values ​​of the target variable. The optimal solution at time k is recorded as (φ * ,θ * ) k , substitute into the linear model, and get the estimated value z of the target variable at time k k .

[0149] In the process of online prediction, the above linear model should be continuously updated adaptively using data from the past period of time. Therefore, the algorithm flow of the key operating parameter trend prediction based on the adaptive exponential model is as follows (the algorithm used in step S300):

[0150] 1. Collect t k At time s, the data z with window length s k , perform empirical mode decomposition on it and extract the IMF component containing the overall trend.

[0151] 2. Assign initial values ​​to the linear model parameters.

[0152] 3. At time k, based on t ks to k Data at the moment Train a linear model f(t i, φ, θ).

[0153] 4. This linear model f(t i , φ, θ) can estimate k s The predicted value at time k Right now:

[0154]

[0155] 5. Wait for the next moment data to arrive and jump to step 1 (the data collection part above).

[0156] At time k, solve equation (7) and get the optimal solution (φ * ,θ * ) k The estimated value of the target variable at time k can be obtained:

[0157] z k =φ * +θ * t k ;

[0158] The alarm threshold ξ of the known target variable satisfies:

[0159] ξ=φ * +θ * (t k +T k );

[0160] The remaining shutdown time can be obtained:

[0161]

[0162] Where, T k represents the time when the abnormal event occurs, ξ represents the alarm threshold of the target variable, φ * ,θ * is the optimal solution of the random variable at time k.

[0163] The length of the time window changes dynamically during the prediction process, and its calculation logic is to obtain the latest trend characteristics of the data as much as possible. Therefore, the algorithm requires that the absolute value of the difference between the first and last values ​​of the time window is less than a threshold, which can ensure that the data trend can be captured as soon as possible after the change.

[0164] The present invention establishes an early warning and diagnosis model by monitoring the operating status of the power plant. With the help of data-driven process monitoring, high-dimensional data is transformed into low-dimensional data, and information is obtained from it to capture the abnormalities of the correlation between process variables, so as to inform the operator to respond in time. The abnormalities in the operating process are monitored, analyzed, and warned, and the problems including detecting whether the current operating status is normal and predicting the time of occurrence of the abnormal event are solved respectively, providing auxiliary support for ensuring the smooth and safe operation of the power plant.

[0165] In addition, based on the above-mentioned event warning method based on nuclear power control logic, the present invention also provides an event warning device based on nuclear power control logic, such as Figure 6 As shown, including:

[0166] A data acquisition module 11 is used to acquire real-time data;

[0167] The abnormality judgment module 12 is used to calculate the real-time data using an abnormality monitoring algorithm based on multi-modal DPCA to determine whether an abnormal event will occur, wherein the abnormal event is at least one of a reactor shutdown event, a shutdown event, and a load shedding event;

[0168] The abnormality prediction module 13 is used to predict the change trend of real-time data by using a key operating parameter trend prediction algorithm based on an adaptive linear model when it is determined that an abnormal event will occur, so as to predict the time when the abnormal event will occur;

[0169] The output module 14 is used to output the time when the abnormal event occurs and to generate an alarm.

[0170] In one embodiment, the abnormality determination module 12 is specifically used for:

[0171] Group the real-time data to obtain several groups of data;

[0172] The covariance matrix of several groups of data is calculated using an anomaly monitoring algorithm based on multimodal DPCA;

[0173] Based on the covariance matrix, the Hotelling statistic and residual statistic of each group of data are calculated using the anomaly monitoring algorithm based on multimodal DPCA;

[0174] Based on the Hotelling statistics and residual statistics of the several groups of data and the preset Hotelling statistics thresholds and residual statistics thresholds corresponding to the several groups of data, it is determined whether an abnormal event will occur.

[0175] In one embodiment, a method for calculating Hotelling statistic thresholds and residual statistic thresholds for a plurality of sets of data includes:

[0176] S241, obtaining a historical sample data set, and dividing the historical sample data set into several groups of initial data sets;

[0177] S242, calculating the covariance matrix of several groups of initial data sets;

[0178] S243, calculating the eigenvalues ​​and eigenvectors of several groups of covariance matrices;

[0179] S244, sorting the plurality of groups of eigenvalues, screening the plurality of groups of eigenvalues ​​by using the cumulative variance contribution rate, and merging the plurality of groups of initial data sets to generate a plurality of groups of intermediate data sets;

[0180] S245, modeling the multiple groups of intermediate data sets to obtain multiple groups of PCA models, and then calculating the residuals between the groups of PCA models;

[0181] S246, merging the multiple sets of intermediate data sets based on a preset threshold and the residuals between the PCA models to generate multiple sets of new data sets;

[0182] S247, repeat S242 to S246 until no new multiple sets of data sets are generated, and then execute step S248;

[0183] S248. Using the PCA method, the Hotelling statistic threshold and the residual statistic threshold of each group of PCA models are finally calculated.

[0184] In one embodiment, the abnormality prediction module 13 is specifically used for:

[0185] When it is judged that an abnormal event will occur, the empirical mode decomposition method is used to denoise the real-time data and extract trends to obtain pre-processed data;

[0186] The preprocessed data are input into the pre-built linear model to obtain the estimated value of the target variable;

[0187] Based on the alarm threshold of the target variable and the estimated value of the target variable, the time when the abnormal event occurs is calculated.

[0188] In one embodiment, the linear model is:

[0189] z i =φ+θt i =f(t i , φ, θ)

[0190] In the formula, φ and θ are random variables, z i is the estimated value of the target variable at time i.

[0191] In one embodiment, the objective function of the linear model is:

[0192]

[0193] In the formula, y i is the true value of the target variable at time i, N is the window length, f(t i , φ, θ) are the estimated values ​​of the target variable.

[0194] In one embodiment, the calculation formula for the time when an abnormal event occurs is:

[0195]

[0196] Where, T k represents the time when the abnormal event occurs, ξ represents the alarm threshold of the target variable, φ * ,θ * is the optimal solution of the random variable at time k.

[0197] In summary, the event warning method and device based on nuclear power control logic provided by the present invention first obtains real-time data; then uses an abnormal monitoring algorithm based on multimodal DPCA to calculate the real-time data to determine whether an abnormal event will occur; then when it is determined that an abnormal event will occur, a key operating condition parameter trend prediction algorithm based on an adaptive linear model is used to predict the change trend of the real-time data to predict the time when the abnormal event will occur; finally, the time when the abnormal event occurs is output and an alarm is issued. By performing abnormal detection on event-related parameters, the input parameter range is divided according to different events, and data processing, feature extraction, multivariate statistical analysis, dynamic principal component analysis and other algorithms are used to carry out research on shutdown events and four scene abnormal detection, and then trend prediction is introduced to determine how long the abnormal event will occur. It performs regression modeling based on historical power plant operation data and the correlation between parameters, predicts the operation trend of the relevant parameters of a specific event in the future, and finally merges the information and outputs it as warning information, thereby providing auxiliary support for ensuring the stable and safe operation of the power plant.

[0198] In addition, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the above-mentioned event warning method based on nuclear power control logic are implemented.

[0199] In addition, the present invention also provides a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed, the steps of the above-mentioned event warning method based on nuclear power control logic are implemented.

[0200] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0201] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. An event early warning method based on nuclear power control logic, characterized in that: include: S100, obtaining real-time data; S200, calculating the real-time data to determine whether an abnormal event will occur; S300, when it is determined that an abnormal event will occur, a key operating condition parameter trend prediction algorithm based on an adaptive linear model is used to predict the change trend of real-time data to predict the time when the abnormal event will occur; S400: Output the time when the abnormal event occurs and generate an alarm.

2. The event early warning method based on nuclear power control logic according to claim 1 is characterized in that: In S200, an abnormal monitoring algorithm based on multi-modal DPCA is used to monitor abnormal events during the operation of the unit, and the abnormal events include at least one of a shutdown event, a shutdown event, and a load rejection event.

3. The event warning method based on nuclear power control logic according to claim 1 or 2 is characterized in that: S200 includes: S210, grouping the real-time data to obtain several groups of data; S220, using an abnormal monitoring algorithm based on multi-modal DPCA to calculate the covariance matrix of several groups of data; S230, based on the covariance matrix, using the anomaly monitoring algorithm based on multimodal DPCA to calculate the Hotelling statistic and the residual statistic of each group of data; S240: judging whether an abnormal event will occur based on the Hotelling statistics and residual statistics of the plurality of data groups and the preset Hotelling statistics thresholds and residual statistics thresholds corresponding to the plurality of data groups.

4. The event early warning method based on nuclear power control logic according to claim 3 is characterized in that: S240 includes: S241, obtaining a historical sample data set, and dividing the historical sample data set into several groups of initial data sets; S242, calculating the covariance matrix of several groups of initial data sets; S243, calculating the eigenvalues ​​and eigenvectors of several groups of covariance matrices; S244, sorting several groups of eigenvalues, screening several groups of eigenvalues ​​using cumulative variance contribution rates, merging several groups of initial data sets, and generating multiple groups of intermediate data sets; S245, modeling the multiple groups of intermediate data sets to obtain multiple groups of PCA models, and then calculating the residuals between the groups of PCA models; S246, merging the multiple sets of intermediate data sets based on a preset threshold and the residuals between the PCA models to generate multiple sets of new data sets; S247, repeat S242 to S246 until no new multiple sets of data sets are generated, and then execute S248; S248. Using the PCA method, the Hotelling statistic threshold and the residual statistic threshold of each group of PCA models are finally calculated.

5. The event early warning method based on nuclear power control logic according to claim 1 is characterized in that: The process of establishing a fault detection model includes: (1) Denoted as the initial data set, the PCA model is constructed through the following iterative process; (2) For each group PCA model The number of principal components retained was determined by cross-validation and the test data set was used as part of; (3) The similarity between the two PCA models is compared, and the residual is used as the similarity measure: (4) If the residual is less than the threshold f, the corresponding data groups are merged and the newly generated data groups are expressed as Go to step (2), otherwise stop. (5) Determine the detection threshold: The value of the threshold is the maximum residual of the corresponding modality training set itself.

6. The event early warning method based on nuclear power control logic according to claim 1 is characterized in that: S300 includes: S310, when it is determined that an abnormal event will occur, the empirical mode decomposition method is used to denoise the real-time data and extract trends to obtain preprocessed data; S320, inputting the preprocessed data into a pre-built linear model to obtain an estimated value of the target variable; S330: Calculate the time when the abnormal event occurs based on the alarm threshold of the target variable and the estimated value of the target variable.

7. The event early warning method based on nuclear power control logic according to claim 1 is characterized in that: In S300, the key operating parameter trend prediction algorithm based on the adaptive exponential model includes: (1) Collect t k At time s, the data z with window length s k , perform empirical mode decomposition on it and extract the IMF component containing the overall trend; (2) Assign initial values ​​to the linear model parameters; (3) At time k, based on t ks to k Data at the moment Train a linear model f(t i , φ, θ); (4) Through the linear model f(t i , φ, θ) estimate k s The predicted value at time k as follows: (5) Wait for the next moment data to arrive and jump to step (1); At time k, the pair Solve and get the optimal solution (φ * ,θ * ) k , get the estimated value of the target variable at time k: z k =φ * +θ * t k ; The alarm threshold ξ of the known target variable satisfies: ξ=φ * +θ * (t k +T k ); Calculate the remaining shutdown time: Where, T k represents the time when the abnormal event occurs, ξ represents the alarm threshold of the target variable, φ * ,θ * represents the optimal solution of the random variable at time k.

8. An event warning device based on nuclear power control logic, characterized in that: include: Data acquisition module, used to acquire real-time data; The abnormality judgment module is used to calculate the real-time data using the abnormality monitoring algorithm based on multi-modal DPCA to determine whether an abnormal event will occur; The abnormality prediction module is used to predict the change trend of real-time data and the time when the abnormal event will occur by using the key operating parameter trend prediction algorithm based on the adaptive linear model when it is judged that an abnormal event will occur; The output module is used to output the time when the abnormal event occurs and to generate an alarm.

9. A computer device comprising a memory and a processor, wherein the memory stores computer-readable instructions, characterized in that: When the processor executes the computer-readable instructions, the steps of the event warning method based on nuclear power control logic described in any one of claims 1-7 are implemented.

10. A computer-readable storage medium having computer-readable instructions stored thereon, characterized in that: When the computer-readable instructions are executed, the steps of the event warning method based on nuclear power control logic described in any one of claims 1-7 are implemented.

Images