CN115455746B - Nuclear power device operation monitoring data anomaly detection and correction integrated method - Google Patents
Nuclear power device operation monitoring data anomaly detection and correction integrated method Download PDFInfo
- Publication number
- CN115455746B CN115455746B CN202211402303.7A CN202211402303A CN115455746B CN 115455746 B CN115455746 B CN 115455746B CN 202211402303 A CN202211402303 A CN 202211402303A CN 115455746 B CN115455746 B CN 115455746B
- Authority
- CN
- China
- Prior art keywords
- nuclear power
- value
- operation monitoring
- monitoring data
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000012544 monitoring process Methods 0.000 title claims abstract description 96
- 238000000034 method Methods 0.000 title claims abstract description 67
- 238000001514 detection method Methods 0.000 title claims abstract description 45
- 238000012937 correction Methods 0.000 title claims abstract description 24
- 230000002159 abnormal effect Effects 0.000 claims abstract description 46
- 238000012795 verification Methods 0.000 claims abstract description 7
- 238000012549 training Methods 0.000 claims description 49
- 230000008569 process Effects 0.000 claims description 25
- 238000010276 construction Methods 0.000 claims description 7
- 239000012530 fluid Substances 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000013524 data verification Methods 0.000 claims 2
- 238000002474 experimental method Methods 0.000 description 10
- 238000010586 diagram Methods 0.000 description 9
- 230000000694 effects Effects 0.000 description 9
- 230000007246 mechanism Effects 0.000 description 8
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 230000005856 abnormality Effects 0.000 description 3
- 230000002547 anomalous effect Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 239000003381 stabilizer Substances 0.000 description 3
- 238000007405 data analysis Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000012938 design process Methods 0.000 description 1
- 238000013401 experimental design Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E30/00—Energy generation of nuclear origin
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Monitoring And Testing Of Nuclear Reactors (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The technical scheme adopted by the invention is as follows: a nuclear power plant operation monitoring data abnormity detection and correction integrated method comprises the following steps: reading a collection value of the operation monitoring data of the nuclear power device at the current moment; the GRU model generates a predicted value of the operation monitoring data of the nuclear power device at the current moment; judging whether the acquisition value of the nuclear power device operation monitoring data at the current moment is normal or not by the MLP model according to the predicted value and the acquisition value of the nuclear power device operation monitoring data at the current moment; if the nuclear power plant operation monitoring data is judged to be normal, the acquisition value of the nuclear power plant operation monitoring data at the current moment is used as the verification value of the nuclear power plant operation monitoring data at the moment for storage; and if the operation monitoring data is judged to be abnormal, storing the predicted value of the operation monitoring data of the nuclear power device at the current moment as the approved value of the operation monitoring data of the nuclear power device at the moment.
Description
Technical Field
The invention belongs to the technical field of nuclear power devices, and particularly relates to a nuclear power device operation monitoring data anomaly detection and correction integrated method.
Background
Processing and analysis capabilities are also increasing. In recent years, with the increasing maturity of technologies such as machine learning and deep learning, a method for diagnosing and predicting operation faults of a nuclear power plant based on data analysis becomes a research hotspot. The analysis of various methods proposed at present shows that the common application premise of the methods is that the operation data collected by the I & C system is accurate and complete. However, the working environment of data acquisition, transmission and storage equipment in the I & C system is severe, and equipment failure and external interference can cause data quality problems such as missing, drifting, jumping and the like of operation monitoring data. The analysis result obtained from the wrong operation monitoring data may cause the operator to make a misjudgment on the operation state of the device or cause the automatic controller to output a wrong control signal. Therefore, there is a need to develop a feasible and efficient method for detecting and correcting abnormal values occurring in the nuclear power plant operating parameter monitoring data.
At present, research on an abnormal detection method of operation monitoring data of a nuclear power plant is not common, and some schemes adopt Multivariate State Estimation (MSET) and Sequential Probability Ratio (SPRT) technologies to estimate, predict and judge the abnormal variables of a plant system. Some schemes establish a system reliability model by monitoring the internal reliability correlation of a plurality of related devices, thereby predicting the reliability state of the whole system and further discovering the abnormal state of system parameters. Henan et al establishes a system reliability model by monitoring the inherent reliability associations of a plurality of related devices, thereby predicting the reliability state of the entire system and further discovering the abnormal state of system parameters. Some schemes use a time series model predictive analysis method to predict noise signals generated by the leakage working condition of the steam generator, and compare the predicted signals with original signals to judge and find data abnormality, so as to further determine the specific time of the heat transfer pipeline leakage. In the prior art, a nuclear power plant abnormal operation state monitoring method based on a dynamic Hopfield artificial neural network also exists, and the method has good parameter abnormal change detection capability.
However, the purpose of data anomaly detection is to find and correct data anomalies, so that detection and correction algorithms for anomalous data should be studied at the same time, and most of the currently studied methods separate anomaly detection and correction functions, occupy large computing resources, and are relatively inefficient.
Disclosure of Invention
The invention aims to solve the defects of the background technology and provides an integrated method for detecting and correcting the abnormal operation monitoring data of the nuclear power device.
The technical scheme adopted by the invention is as follows: a nuclear power plant operation monitoring data abnormity detection and correction integrated method comprises the following steps:
reading a collection value of the operation monitoring data of the nuclear power device at the current moment;
the GRU model generates a predicted value of the operation monitoring data of the nuclear power device at the current moment;
judging whether the acquisition value of the nuclear power device operation monitoring data at the current moment is normal or not by the MLP model according to the predicted value and the acquisition value of the nuclear power device operation monitoring data at the current moment;
if the nuclear power plant operation monitoring data is judged to be normal, the acquisition value of the nuclear power plant operation monitoring data at the current moment is used as the verification value of the nuclear power plant operation monitoring data at the moment for storage;
and if the operation monitoring data is judged to be abnormal, storing the predicted value of the operation monitoring data of the nuclear power device at the current moment as the approved value of the operation monitoring data of the nuclear power device at the moment.
In the above technical solution, the GRU model generates the predicted value of the operation monitoring data of the nuclear power plant at the current time based on the approval value of the operation monitoring data of the nuclear power plant at the historical time period corresponding to the current time.
In the technical scheme, the operation monitoring data of the nuclear power plant has time-varying characteristics and is easy to fluctuate, and the parameter values are generated in the operation process of a nuclear power plant system and equipment and reflect the operation state of the nuclear power plant system and the equipment, and comprise the temperature, the pressure and the flow of fluid in a pipeline, the rotating speed of a rotating machine and the operation current of electrical equipment.
In the technical scheme, a GRU model and an MLP model corresponding to each parameter in the operation monitoring data of the nuclear power plant are respectively constructed, and the acquired value of each parameter is input into the corresponding GRU model to obtain the predicted value of the parameter; and inputting the acquired value and the predicted value of each parameter into the corresponding MLP model to obtain a judgment result of whether the parameter is normal, and selecting the acquired value or the predicted value of the parameter as a check value according to the judgment result.
In the above technical solution, the process of constructing the single parameter GRU model includes: acquiring historical data a (t) of a certain parameter in operation monitoring data of a nuclear power plant, wherein the data sampling period is tau, a window with the length of q slides backwards along the time sequence, and the starting point of the window is set as t = t 0 Then the end point is t = t 0 + (q-1) τ; data a (t) in the window 0 )~a(t 0 + (q-1) τ) input GRU model, with a (t) 0 + q tau) as result mark, and carrying out iterative training; the time window slides backwards according to the time sequence, each time slides by a time unit tau, and training is carried out in sequence until the GRU model convergence reaches the endExpected, GRU model as the parameter; the historical data of the parameter is a value which is preset to be correct and has no deviation and is used for reflecting the objective and actual running state of the nuclear power device; and constructing GRU models corresponding to all parameters in the operation monitoring data of the nuclear power plant by adopting the process.
In the above technical solution, the construction process of the MLP model with a single parameter includes: acquiring historical data a (t) of a certain parameter in the operation monitoring data of the nuclear power plant, and taking the window data of a certain section as a true value and an acquisition value of the window data of the section; inputting the true value of the parameter into the GRU model which completes training to obtain the predicted value of the window data of the section; constructing a sample set based on the segment of window data; a single sample of the sample set comprises two parts, namely input data and a training label corresponding to the input data; the input data comprises a predicted value of a GRU model to a parameter at a certain moment and an acquired value of the parameter at the moment, and the training label is a mark for judging whether the parameter is abnormal at the moment; training an MLP model through a sample set to obtain the MLP model of the parameters; and constructing the MLP model corresponding to all the parameters in the operation monitoring data of the nuclear power plant by adopting the process.
In the technical scheme, the verification value of the operation monitoring data of the nuclear power device at the current moment is stored in a historical database; the GRU model acquires a check value of the nuclear power device operation monitoring data corresponding to a window time period with the terminal point being the current moment from a historical database, and the check value is used for generating a predicted value of the nuclear power device operation monitoring data at the next moment.
In the technical scheme, in the training process of the MLP model, firstly, the difference value between the predicted value and the acquired value in a single training sample is obtained, the difference value and the training label in the training sample are input into the MLP model for training, and the training of the MLP model is completed after the difference value and the training label in the training sample are converged;
in the using process of the MLP model, firstly, the difference value between the input predicted value and the input acquired value is obtained, and the difference value is input into the trained MLP model, so that the recognition conclusion can be obtained.
The invention has the beneficial effects that: the invention effectively improves the data quality problems of deletion, drift, jump and the like in the operation data acquired or stored by the instrument control system of the nuclear power device, so as to provide more reliable input for operation data analysis and an automatic controller. The invention provides a nuclear power device operation monitoring parameter abnormity detection and correction method based on a gated cyclic unit and a multilayer perceptron mixed model (GRU-MLP). Firstly, a short-time prediction algorithm of the operational monitoring data based on the GRU model is provided, a reference basis is provided for the abnormal detection and correction of the operational monitoring data, and a real-time correction mechanism is designed to improve the prediction accuracy of the GRU model to the operational data containing the abnormality. Then, by utilizing the nonlinear fitting capacity of the MLP model, a fixed threshold used under a prediction-anomaly detection mechanism is optimized to be a dynamic threshold, and the anomaly detection accuracy of the designed method is improved through the nonlinear fitting capacity of the MLP model. The invention provides a method for integrally predicting and correcting monitoring data of a nuclear power device, detecting abnormity and correcting, which effectively saves computing resources and improves the overall operation efficiency.
Drawings
FIG. 1 is a schematic flow diagram of the present invention;
FIG. 2 is a logic diagram of the present invention;
FIG. 3 is a schematic diagram of three pressure curves of the potentiostat at random jump;
FIG. 4 is a diagram illustrating a data anomaly detection accuracy curve under random hopping;
FIG. 5 is a schematic diagram of three curves of the water level of the constant-offset voltage stabilizer;
FIG. 6 is a graph illustrating data anomaly detection accuracy at a fixed offset;
FIG. 7 is a schematic diagram of three pressure curves of a potentiostat under linear growth;
FIG. 8 is a graph illustrating data anomaly detection accuracy under linear growth;
FIG. 9 is a schematic diagram of three polynomial curves for increasing the average temperature of the circuit;
FIG. 10 is a graph illustrating data anomaly detection accuracy under polynomial growth;
FIG. 11 is a graph of three curves of potentiostat pressure at logarithmic growth;
FIG. 12 is a graph illustrating data anomaly detection accuracy for logarithmic growth;
FIG. 13 is a graph illustrating the actual pressure curve of the pressure stabilizer;
FIG. 14 is a graph illustrating pressure collection values of the pressure stabilizer;
FIG. 15 is a schematic diagram of the GRU model prediction effect without real-time correction;
fig. 16 is a diagram illustrating the predicted effect of the real-time correction GRU model.
Detailed Description
The invention will be further described in detail with reference to the following drawings and specific examples, which are not intended to limit the invention, but are for clear understanding.
As shown in fig. 1, the invention provides an integrated method for detecting and correcting the abnormal operation monitoring data of a nuclear power plant, which comprises the following steps:
reading a collection value of the operation monitoring data of the nuclear power device at the current moment;
the GRU model generates a predicted value of the operation monitoring data of the nuclear power device at the current moment; the GRU model generates a predicted value of the nuclear power device operation monitoring data at the current moment based on the nuclear power device operation monitoring data at the historical time period corresponding to the current moment; storing a verification value of the operation monitoring data of the nuclear power device at the current moment into a historical database; the GRU model acquires a check value of the nuclear power device operation monitoring data corresponding to a window time period with the end point being the current moment from a historical database, and the check value is used for generating a predicted value of the nuclear power device operation monitoring data at the next moment;
judging whether the acquisition value of the nuclear power device operation monitoring data at the current moment is normal or not by the MLP model according to the predicted value and the acquisition value of the nuclear power device operation monitoring data at the current moment;
if the nuclear power plant operation monitoring data is judged to be normal, the acquisition value of the nuclear power plant operation monitoring data at the current moment is used as the verification value of the nuclear power plant operation monitoring data at the moment for storage;
and if the operation monitoring data is judged to be abnormal, storing the predicted value of the operation monitoring data of the nuclear power device at the current moment as the approved value of the operation monitoring data of the nuclear power device at the moment.
The nuclear power plant operating data is typically time series data. The nuclear power plant operation monitoring data are parameter values reflecting the operation states of the nuclear power plant system and the equipment generated in the operation process, such as the temperature, the pressure and the flow of fluid in a pipeline, the rotating speed of a rotating machine, the operation current of electrical equipment and other parameter values. The data has time-varying characteristics and is subject to fluctuations. The data has two expressions of real value and collected value. The true value refers to an objective actual data state of an operation parameter of a certain nuclear power device, and the acquired value refers to a data value obtained by observing the operation parameter through data acquisition equipment such as a sensor and a transmitter. There is a deviation between the collected value and the true value. This deviation is considered acceptable when the sensor, transmitter, etc. is operating properly, and therefore the collected value is usually used as the true value. When data acquisition equipment such as a detector and a sensor breaks down, the acquired value and the true value are greatly deviated, and the acquired value is regarded as abnormal.
The actual value used in this embodiment is a value that is preset to be correct and free of deviation, and the acquired value is a value acquired when the detection device fails. Model training uses the collected values and predicted values.
The GRU model extracts short-time memory information contained in the historical hidden information by using a reset gate r; updating the content of the new moment to the historical hidden information by using an updating gate z, wherein the opening degrees of z and r are calculated as shown in formulas (1) and (2):
wherein,
，W r 、W z the weight matrices for the update gate and reset gate respectively for the GRU unit,
as an input vector x t And the hidden state vector h output at the last moment t-1 The mosaic matrix of (1). z ∈ (0, 1), (1-z) denotes forgetting, (1-z) ∈ (0, 1), thereby enabling the merging of the "memory-forgetting" steps. The final output of the GRU model is the current moment conclusion y t And hidden information h transmitted to next time t As shown in formulas (3) and (4).
Wherein "+" indicates a corresponding multiplication of matrix elements and "+" indicates a matrix addition.
Representing hidden history information
The short-time memory content extracted after the processing of the reset gate r;
to express an integrated short-term memory
Externally inputting information with the current time
And forming current time information.
The GRU method focuses on processing predictions for time series data and is not suitable for direct detection of data anomalies. It is thus improved using a multi-layer perceptron (MLP) approach. The MLP is one of the simplest BP neural network models, is composed of a large number of simplest neuron structures, and realizes the approximation of nonlinear mapping from an input state space to an output state space by utilizing the function superposition of a large number of neurons. Compared with the traditional abnormity identification method, the method adopts the GRU model to improve the nonlinear judgment capability of the GRU model to judge whether the data is abnormal or not by using the fixed threshold value.
Respectively constructing a GRU model and an MLP model corresponding to each parameter in the operation monitoring data of the nuclear power plant, and inputting the true value of each parameter into the corresponding GRU model to obtain the predicted value of the parameter; and inputting the acquired value and the predicted value of each parameter into the corresponding MLP model to obtain a judgment result of whether the parameter is normal, and selecting the acquired value or the predicted value of the parameter as a check value according to the judgment result.
The purpose of predicting the nuclear power plant parameters through the GRU model is two: firstly, providing a reference basis for detecting the abnormity of the acquired value; and repairing the data points determined as abnormal values. The basic strategy of the designed prediction algorithm is as follows: and predicting a future value of a certain parameter by using the historical data of the parameter as a basis, and training a GRU model by adopting a sliding window method.
The GRU model construction process of the single parameter comprises the following steps: suppose that a certain parameter in the operation monitoring data of the nuclear power plant isa(t) a data sampling period ofτIn the length of time ofqIs slid backwards in time sequence, the starting point of the window is set to t = t 0 Then the end point time is t = t 0 +(q-1)τ. Data in windowa (t 0 )~ a (t 0 +(q-1)τ) Input into GRU model toa (t 0 +qτ) And performing iterative training for result marking. That is, a single sample of the training set of the GRU model includes the actual value of the parameter at a certain time as an input element and the actual value of the parameter at the next time as a training label. The time window is slid backwards in time sequence, one time unit at a timeτAnd training in sequence until the GRU model is converged to an expected value, thus obtaining the GRU model corresponding to the parameter. And constructing GRU models corresponding to all parameters in the operation monitoring data of the nuclear power plant by adopting the process.
Performing GRU model training to complete the prediction workWhen the current time is enabled, inputting a length taking the current time as an end pointlThe predicted value of the parameter at the next time can be obtained by the parameter verification value data.
The construction process of the single-parameter MLP model comprises the following steps: acquiring historical data a (t) of a certain parameter in the operation monitoring data of the nuclear power plant, and taking the window data of a certain section as the true value of the window data of the section; inputting the historical data of the parameter into the GRU model which is trained to obtain the predicted value of the window data of the section; constructing a sample set based on the segment of window data; a single sample of the sample set comprises two parts, namely input data and a training label corresponding to the input data; the input data comprises a predicted value of a GRU model to a parameter at a certain moment and an acquired value of the parameter at the moment, and the training label is a mark for judging whether the parameter is abnormal at the moment; training an MLP model through a sample set to obtain the MLP model of the parameters; and constructing an MLP model corresponding to all parameters in the operation monitoring data of the nuclear power plant by adopting the process. The training labels are set manually.
In the training process of the MLP model, firstly, the difference value between the predicted value and the acquired value in a single training sample is obtained, the difference value and the training label in the training sample are input into the MLP model for training, and the training of the MLP model is completed after the difference value and the training label in the training sample are converged;
in the use process of the MLP model, firstly, the difference value between the input predicted value and the input collection value is obtained, and the difference value is input into the trained MLP model, so that the identification conclusion can be obtained.
And on the basis of obtaining the detection conclusion of the abnormal acquisition value, replacing the identified abnormal value with the prediction result of the GRU model, thereby realizing the correction of the abnormal acquisition value of the operation parameter of the nuclear power plant. The overall algorithm flow is shown in fig. 2.
The nuclear power plant operation monitors data anomalies, typically manifested as random jumps in data value, fixed value drift, incremental drift, three of which are most common, manifested as linear incremental drift, polynomial incremental drift, and logarithmic incremental drift.
In the embodiment, the actual operation data of a certain type of nuclear power plant is used as experimental data to carry out experiments to verify the correctness of the invention, and the experimental design is shown in table 1. In order to facilitate observation and comparison, the real value, the acquisition index and the predicted value of the used operation monitoring data are all subjected to normalization processing. The normalization formula is shown in formula (6):
wherein X represents original data including a true value, an acquired value and a predicted value; y represents that the normalized data a is the maximum value and b is the minimum value.
The GRU model was trained using the true values. The acquisition value, the predicted value and a manually set label are used as a sample data set of the MLP model, the sample data set is divided, and 65% of the data are taken to form a training set to train the MLP model; the remaining 35% comprise the test set, which is tested for its ability to identify anomalies.
Finally, the effect of data anomaly detection is evaluated by using detection accuracy rate, and the value of the detection accuracy rate is the ratio of the number of correct points for anomaly state judgment to the total number of abnormal data points.
Table 1 abnormal data detection experiment set-up
The abnormal change form is not applied to the model construction stage, and is used for explaining that the algorithm can obtain high abnormal detection accuracy rate for various types of abnormal conditions, and the superiority is highlighted. As can be seen from analyzing the images in fig. 3, 5, 7, 9, and 11, the prediction algorithm of the present embodiment has high accuracy, and the maximum relative error occurring in all experimental processes is 0.001487645. As can be seen from the data anomaly detection experiment results shown in fig. 4, 6, 8, 10, and 12, the designed GRU-MLP algorithm has a high accuracy for detecting the random jump drift anomaly and the fixed value drift anomaly, and has a slightly inferior detection effect on the growth drift, but the final accuracy mostly exceeds 90% and is not lower than 85% at least, and as the data amount is accumulated, the data change patterns are increased, and the accuracy is further improved. The training curve is used for showing the training process of the construction of the MLP and GRU models. The test curve shows the using effect display process of the trained MLP and GRU models.
The correction effect and the important function of the real-time correction algorithm are verified by designing a comparison experiment. FIGS. 13 and 14 show the collected and actual values of the pressurizer pressure in 600 seconds, respectively. Experiments were performed according to this data.
FIGS. 15 and 16 are graphs comparing the results of two experiments. FIG. 15 is a graph of GRU model predicted effect without real-time correction. Fig. 16 shows the prediction effect of using the GRU model in the case of performing real-time correction using the MLP algorithm on abnormal data detected in the prediction process. Compared with two groups of experimental results, the real-time correction algorithm designed by the invention can effectively reduce the interference of outliers to the prediction effect and improve the prediction precision.
The superiority of the calculated method in performance is verified through a comparison experiment. Two sets of comparative experiments were designed and the experimental content settings are shown in table 2.
TABLE 2 Experimental setup
The GRU-MLP method is compared with a window center (WinCen) value method and an ARIMA method, and the indexes are the state judgment accuracy and the abnormal data detection rate respectively. The results of the experiment are shown in tables 3 and 4. The state judgment accuracy rate is the ratio of the number of points judged to be correct (including normal state and abnormal state) to the total data number. The abnormal data detection rate represents a ratio of the number of detected abnormal data points to the total number of abnormal points.
TABLE 3 State judgment accuracy
TABLE 4 abnormal detection Rate
Comparing tables 3 and 4, it can be seen that the performance of the designed GRU-MLP method is much better than that of the other two methods. Particularly in terms of the detection rate of anomalous data. The main reasons are as follows:
(1) Compared with the common method, the designed GRU-MLP method utilizes the prediction value to correct the detected abnormal data in real time so as to weaken the interference of the abnormal data on the output conclusion of the prediction algorithm layer. However, winCen and ARIMA algorithms do not correct abnormal data in time, resulting in erroneous prediction results. Specifically, the WinCen algorithm and the ARIMA algorithm can be used for well predicting normal data in a fixed value deviation experiment, but the abnormal detection rate is low.
(2) The WinCen algorithm and the ARIMA algorithm adopt a fixed threshold strategy when abnormal data detection is carried out. Therefore, in its design process, the designer must relax the decision criteria to avoid identifying normal fluctuations in the data as anomalous. In the method designed by the invention, the dynamic adjustment of the detection threshold is realized by adopting an MLP model dynamic nonlinear fitting mode, and the accuracy of the anomaly detection is improved.
In conclusion, compared with the traditional methods such as WinCen and ARIMA, the nuclear power plant operation parameter prediction and abnormality detection method based on the GRU-MLP model has the remarkable advantages of high accuracy, real-time correction of data and the like. The method can be effectively applied to the engineering practice of abnormal detection and correction of the monitoring data of the nuclear power device, thereby improving the quality of the monitoring data and being used for fault-tolerant control of the measurement channel fault.
The innovation of the invention is the improvement of the operation mechanism. The supplement of the characteristics of the operating data of the nuclear power plant has already been elucidated in the aforesaid answer. The strong time-varying characteristics of the operating data of the nuclear power plant make the nuclear power plant have high requirements on the real-time performance of the algorithm. The method improves the traditional operation mechanism of separating the abnormal detection from the data correction, and corrects the data judged to be abnormal in real time while detecting the abnormal. Input data of a general GRU model is static data, and when data are abnormal, an error data acquisition value can be used for inputting the GRU model, so that an error prediction result is generated, and the accuracy of abnormal detection is influenced. The mechanism is applied to the nuclear power plant operation data with strong time variability, and unacceptable deviation can be generated, so that the nuclear safety is influenced. The method improves the operation mechanism of the GRU model aiming at the problem, and improves a 'prediction-abnormity detection' mechanism into a 'prediction-abnormity detection-real-time correction' mechanism, so that the prediction result of the GRU model can be used for inputting the abnormity detection of the MLP model, and can also be used as the basis for correcting the data which is identified as abnormity by the MLP model, thereby being suitable for the requirements of the data abnormity detection and correction of the nuclear power plant operation data.
Those not described in detail in this specification are within the skill of the art.
Claims (7)
1. A nuclear power plant operation monitoring data abnormity detection and correction integrated method is characterized in that: the method comprises the following steps:
reading a collection value of the operation monitoring data of the nuclear power device at the current moment;
the GRU model generates a predicted value of the operation monitoring data of the nuclear power device at the current moment;
judging whether the acquisition value of the nuclear power device operation monitoring data at the current moment is normal or not by the MLP model according to the prediction value and the acquisition value of the nuclear power device operation monitoring data at the current moment;
if the nuclear power plant operation monitoring data is judged to be normal, the acquisition value of the nuclear power plant operation monitoring data at the current moment is used as the verification value of the nuclear power plant operation monitoring data at the moment for storage;
if the operation monitoring data is judged to be abnormal, the predicted value of the operation monitoring data of the nuclear power device at the current moment is used as the approved value of the operation monitoring data of the nuclear power device at the moment for storage;
storing the approved value of the operation monitoring data of the nuclear power device at the current moment into a historical database; the GRU model obtains a nuclear power device operation monitoring data verification value corresponding to a window time period with the end point being the current moment from a historical database, and the nuclear power device operation monitoring data verification value is used for generating a predicted value of the nuclear power device operation monitoring data at the next moment.
2. A method according to claim 1, characterized by: and the GRU model generates a predicted value of the operation monitoring data of the nuclear power device at the current moment based on the nuclear fixed value of the operation monitoring data of the nuclear power device in the historical time period corresponding to the current moment.
3. A method according to claim 1, characterized by: the nuclear power plant operation monitoring data has time-varying characteristics and is easy to fluctuate, and are parameter values which are generated in the operation process of a nuclear power plant system and equipment and reflect the operation state of the nuclear power plant system and equipment, wherein the parameter values comprise the temperature, the pressure and the flow of fluid in a pipeline, the rotating speed of a rotating machine and the operation current of electrical equipment.
4. A method according to claim 3, characterized by: respectively constructing a GRU model and an MLP model corresponding to each parameter in the operation monitoring data of the nuclear power plant, and inputting the acquired value of each parameter into the corresponding GRU model to obtain the predicted value of the parameter; and inputting the acquired value and the predicted value of each parameter into the corresponding MLP model to obtain a judgment result of whether the parameter is normal, and selecting the acquired value or the predicted value of the parameter as a check value according to the judgment result.
5. A method according to claim 4, characterized in that: the construction process of the single parameter GRU model comprises the following steps: acquiring historical data a (t) of a certain parameter in operation monitoring data of a nuclear power device, wherein the data sampling period is tau, a window with the length of q slides backwards along the time sequence, and the starting point of the window is set as t = t 0 Then the end point is t = t 0 + (q-1) τ; data a (t) in the window 0 )~a(t 0 + (q-1) τ) input GRU model, with a (t) 0 + q tau) as result mark, and carrying out iterative training; the time window is slid backwards in time sequence, one time unit tau at a time, and then in sequenceTraining until the convergence of the GRU model reaches the expectation finally, and taking the GRU model as the GRU model of the parameter; the historical data of the parameter is a value which is preset to be correct and has no deviation and is used for reflecting the objective and actual running state of the nuclear power device; and constructing GRU models corresponding to all parameters in the operation monitoring data of the nuclear power plant by adopting the process.
6. A method according to claim 5, characterized by: the construction process of the single-parameter MLP model comprises the following steps: acquiring historical data a (t) of a certain parameter in the operation monitoring data of the nuclear power plant, and taking the window data of a certain section as a true value and an acquisition value of the window data of the section; inputting the true value of the parameter into the GRU model which completes training to obtain the predicted value of the window data of the section; constructing a sample set based on the segment of window data; a single sample of the sample set comprises two parts, namely input data and a training label corresponding to the input data; the input data comprises a predicted value of a GRU model to a parameter at a certain moment and an acquired value of the parameter at the moment, and the training label is a mark for judging whether the parameter is abnormal at the moment; training an MLP model through a sample set to obtain the MLP model of the parameters; and constructing an MLP model corresponding to all parameters in the operation monitoring data of the nuclear power plant by adopting the process.
7. A method according to claim 6, characterized by: in the training process of the MLP model, firstly, the difference value between the predicted value and the acquired value in a single training sample is obtained, the difference value and the training label in the training sample are input into the MLP model for training, and the training of the MLP model is completed after the difference value and the training label in the training sample are converged;
in the use process of the MLP model, firstly, the difference value between the input predicted value and the input collection value is obtained, and the difference value is input into the trained MLP model, so that the identification conclusion can be obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211402303.7A CN115455746B (en) | 2022-11-10 | 2022-11-10 | Nuclear power device operation monitoring data anomaly detection and correction integrated method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211402303.7A CN115455746B (en) | 2022-11-10 | 2022-11-10 | Nuclear power device operation monitoring data anomaly detection and correction integrated method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115455746A CN115455746A (en) | 2022-12-09 |
| CN115455746B true CN115455746B (en) | 2023-04-07 |
Family
ID=84295685
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211402303.7A Active CN115455746B (en) | 2022-11-10 | 2022-11-10 | Nuclear power device operation monitoring data anomaly detection and correction integrated method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115455746B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116522088B (en) * | 2023-04-27 | 2024-01-26 | 哈尔滨工程大学 | Nuclear power plant operation data analysis method and system based on machine learning |
| CN116705364B (en) * | 2023-05-23 | 2024-01-02 | 上海发电设备成套设计研究院有限责任公司 | Reliability monitoring platform suitable for in-service nuclear power unit and nuclear turbine |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019216975A1 (en) * | 2018-05-07 | 2019-11-14 | Strong Force Iot Portfolio 2016, Llc | Methods and systems for data collection, learning, and streaming of machine signals for analytics and maintenance using the industrial internet of things |
| CN111459778B (en) * | 2020-03-12 | 2024-05-07 | 平安科技（深圳）有限公司 | Operation and maintenance system abnormal index detection model optimization method, device and storage medium |
| CN112668870B (en) * | 2020-12-25 | 2023-05-12 | 苏州热工研究院有限公司 | Nuclear power plant equipment state online intelligent monitoring method and computer terminal |
| CN113657454B (en) * | 2021-07-23 | 2024-02-23 | 杭州安脉盛智能技术有限公司 | Nuclear power rotating machinery state monitoring method based on autoregressive BiGRU |
| CN113569992B (en) * | 2021-08-26 | 2024-01-09 | 中国电子信息产业集团有限公司第六研究所 | Abnormal data identification method and device, electronic equipment and storage medium |
| CN114519382B (en) * | 2022-01-05 | 2024-10-18 | 哈尔滨工程大学 | Nuclear power plant key operation parameter extraction and abnormality monitoring method |
| CN115274009A (en) * | 2022-08-01 | 2022-11-01 | 东华大学 | Polyester melt quality online prediction method based on semi-supervised GRU regression model |
| CN115081629A (en) * | 2022-08-16 | 2022-09-20 | 杭州比智科技有限公司 | Deep learning method and system for sensitive data discovery and identification |
-
2022
- 2022-11-10 CN CN202211402303.7A patent/CN115455746B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN115455746A (en) | 2022-12-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111914873B (en) | Two-stage cloud server unsupervised anomaly prediction method | |
| CN115455746B (en) | Nuclear power device operation monitoring data anomaly detection and correction integrated method | |
| US11507049B2 (en) | Method for detecting abnormity in unsupervised industrial system based on deep transfer learning | |
| CN109492193B (en) | Abnormal network data generation and prediction method based on deep machine learning model | |
| CN109146246B (en) | Fault detection method based on automatic encoder and Bayesian network | |
| CN110705812A (en) | Industrial fault analysis system based on fuzzy neural network | |
| CN111723925B (en) | Fault diagnosis method, device, equipment and medium for on-road intelligent train air conditioning unit | |
| CN114297918A (en) | Aero-engine residual life prediction method based on full-attention depth network and dynamic ensemble learning | |
| CN115791174B (en) | Rolling bearing abnormality diagnosis method, system, electronic equipment and storage medium | |
| CN117556347A (en) | Power equipment fault prediction and health management method based on industrial big data | |
| CN111079348B (en) | Method and device for detecting slowly-varying signal | |
| CN114611409B (en) | Method and device for establishing abnormal detection model of power distribution terminal | |
| Xu et al. | Anomaly detection with gru based bi-autoencoder for industrial multimode process | |
| CN118188342A (en) | Fan-oriented fault early warning and life prediction method | |
| CN117792750A (en) | Information physical system anomaly detection method based on generation countermeasure network | |
| Dang et al. | seq2graph: Discovering dynamic non-linear dependencies from multivariate time series | |
| CN110874506A (en) | Low-temperature equipment fault prediction method | |
| CN113158448A (en) | Self-adaptive calculation method for equipment threshold of ship system | |
| CN114548701B (en) | Full-measurement-point-oriented coupling structure analysis and estimation process early warning method and system | |
| CN116304789A (en) | VP inclinometer fault diagnosis method and device | |
| KR102389317B1 (en) | Method of Determining Whether A Smart Farm Sensor has failed using a Recurrent Neural Network(RNN) | |
| Won et al. | Prediction of remaining useful lifetime of membrane using machine learning | |
| Zhuang et al. | Remaining useful life estimation under variable failure behaviors via transferable metric learning | |
| KR20220028726A (en) | Method and Apparatus for Fault Detection Using Pattern Learning According to Degradation | |
| KR20220028727A (en) | Method and Apparatus for Real Time Fault Detection Using Time series data According to Degradation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |