CN110738274A - nuclear power device fault diagnosis method based on data driving - Google Patents
nuclear power device fault diagnosis method based on data driving Download PDFInfo
- Publication number
- CN110738274A CN110738274A CN201911026515.8A CN201911026515A CN110738274A CN 110738274 A CN110738274 A CN 110738274A CN 201911026515 A CN201911026515 A CN 201911026515A CN 110738274 A CN110738274 A CN 110738274A
- Authority
- CN
- China
- Prior art keywords
- data
- fault
- historical
- nuclear power
- power device
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000003745 diagnosis Methods 0.000 title claims abstract description 40
- 239000011159 matrix material Substances 0.000 claims abstract description 22
- 238000004364 calculation method Methods 0.000 claims abstract description 11
- 238000000513 principal component analysis Methods 0.000 claims abstract description 11
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 10
- 238000001514 detection method Methods 0.000 claims abstract description 10
- 238000004458 analytical method Methods 0.000 claims abstract description 8
- 238000012549 training Methods 0.000 claims abstract description 8
- 230000003213 activating effect Effects 0.000 claims abstract description 3
- 238000012216 screening Methods 0.000 claims abstract description 3
- 238000012706 support-vector machine Methods 0.000 claims description 21
- 238000005259 measurement Methods 0.000 claims description 9
- 230000009467 reduction Effects 0.000 claims description 8
- 238000011156 evaluation Methods 0.000 claims description 5
- 230000001186 cumulative effect Effects 0.000 claims description 4
- 230000010354 integration Effects 0.000 claims description 4
- 239000000126 substance Substances 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 claims description 2
- 230000001419 dependent effect Effects 0.000 claims description 2
- 230000002441 reversible effect Effects 0.000 claims description 2
- 230000008569 process Effects 0.000 abstract description 10
- 230000008901 benefit Effects 0.000 abstract description 3
- 238000000605 extraction Methods 0.000 abstract description 3
- 238000012544 monitoring process Methods 0.000 abstract description 3
- 230000000007 visual effect Effects 0.000 abstract description 2
- 239000002699 waste material Substances 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 10
- 230000005856 abnormality Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 238000007621 cluster analysis Methods 0.000 description 3
- 239000002826 coolant Substances 0.000 description 3
- 238000004451 qualitative analysis Methods 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000007405 data analysis Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 238000013079 data visualisation Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000012938 design process Methods 0.000 description 1
- 238000012854 evaluation process Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
- 239000003381 stabilizer Substances 0.000 description 1
- 230000036962 time dependent Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G21—NUCLEAR PHYSICS; NUCLEAR ENGINEERING
- G21C—NUCLEAR REACTORS
- G21C17/00—Monitoring; Testing ; Maintaining
- G21C17/02—Devices or arrangements for monitoring coolant or moderator
-
- G—PHYSICS
- G21—NUCLEAR PHYSICS; NUCLEAR ENGINEERING
- G21C—NUCLEAR REACTORS
- G21C17/00—Monitoring; Testing ; Maintaining
- G21C17/10—Structural combination of fuel element, control rod, reactor core, or moderator structure with sensitive instruments, e.g. for measuring radioactivity, strain
-
- 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
- Y02E30/30—Nuclear fission reactors
Abstract
The invention discloses a fault diagnosis method based on data drive nuclear power device, which comprises the following steps that nuclear power device operation data are collected and transmitted to a KPCA (kernel principal component analysis) abnormity detection model to obtain a covariance matrix of a higher-dimensional linear space, the covariance matrix is subjected to characteristic vector decomposition to extract high-dimensional information characteristics, and calculation statistics is carried out to obtain high-dimensional information characteristicsAnd comparing the real-time value of the SPE statistic with a historical normal operation threshold value to obtain fault information, activating a fault diagnosis module, and screening out fault reasons from multiple suspected faults. Compared with the prior art, the method has the advantages of visual monitoring information, good interpretability, easy extraction of historical data and avoidance of sensorsThe failure of the nuclear power device causes resource waste of subsequent diagnosis and analysis, the probability of misjudgment when the nuclear power device fails is reduced, the safety and the reliability of the nuclear power device are improved, the learning and training operation processes are easy, and the practicability is high.
Description
Technical Field
The invention relates to the technical field of nuclear power device fault diagnosis, in particular to fault diagnosis methods based on data driving nuclear power devices.
Background
The nuclear power device can be regarded as a typical Cyber-Physical System (CPS), advanced information technologies and automatic control technologies such as sensing, calculation, communication and control are integrated, a complex coupling System with the functions of mutual mapping, timely interaction and efficient cooperation of human, machine, object, environment, information and other elements in a Physical space and an information space is constructed, for the CPS, data is a soul, if an effective data-driven abnormity detection technology is lacked to guarantee the operation of the CPS, the CPS fails or fails, if maintenance and guarantee measures are not timely, accidents are easily caused, and serious damage is caused to a nuclear power device body and equipment related to the nuclear power device body, and if disastrous accidents are caused, the related technologies of large investment are continuously developed to cover shadows.
The fault diagnosis technology has been continuously developed since the 70 s of the 20 th century and has vigorous vitality. Existing nuclear power plant fault diagnosis methods may be broadly classified into analytical model-based, qualitative analysis, and data-driven methods.
The fault diagnosis based on the analytical model comprises the steps of firstly establishing a refined model of a system and equipment to obtain prior information of the system, then comparing observable information of an object with the prior information to obtain a residual error, and finally realizing the purpose of fault diagnosis through analysis processing of the residual error. The ideal state of the method is to establish an accurate model, and nuclear power plant systems and equipment are numerous, the operation mechanism is complex, and an accurate simulation model is difficult to establish for all equipment. Therefore, the main disadvantages of the method are that the modeling process is complex, the factors to be considered are many and complicated, and the established model is not enough to reflect the association relationship between each device and system of the nuclear power plant.
The qualitative analysis-based method requires a large amount of expert experience knowledge as a diagnosis rule, and utilizes the accumulated experience of field experts in long-term practice to establish a knowledge base for reasoning and analysis so as to obtain the fault reasons of related systems or equipment. The method is suitable for a system with difficult establishment of a mechanism model and less sensor data. The disadvantage of this type of method is the difficulty of knowledge acquisition; when the rules are more, the problems of matching conflict, combination explosion and the like exist in the reasoning process.
Therefore, analytical model-based methods and qualitative analysis-based methods are more suitable for systems with fewer input, output or state parameters, and are too costly to use for complex systems with large amounts of monitoring data.
The data-driven-based method does not need a process-accurate analytical model, and a data analysis model is constructed through massive learning and training of historical data, so that the modeling process is relatively simple. Diagnosing typical faults of a pressurized water reactor steam generator by utilizing residual space analysis in Upadhyaya, B, R of university of Tennessee, USA abroad; KamalHadad of Iran university adopts a fault diagnosis method combining a BP artificial neural network and wavelet transformation. In China, Zhao Yun Fei et al use BP artificial neural network in diagnosis of partial accidents in AP1000 nuclear power station. However, the method has obvious defects that historical data is difficult to acquire, and learning and training cannot be completed; the interpretability is poor.
In conclusion, the conventional nuclear power fault diagnosis method is difficult to obtain historical data, weak in fault identification capability and poor in fault diagnosis accuracy and interpretability.
Disclosure of Invention
The invention aims to provide fault diagnosis methods based on kernel principal component analysis, support vector machine and cluster analysis for a nuclear power plant, and combines the feature extraction and abnormality detection capability of the kernel principal component analysis, the pattern recognition capability of the support vector machine and the data visualization capability of the cluster analysis, thereby improving the accuracy and interpretability of fault diagnosis.
A fault diagnosis method for a data-driven nuclear power plant, comprising the steps of:
step 1: acquiring historical normal operation data, namely unidirectionally acquiring historical normal operation measurement data from a historical operation database of the nuclear power device to form historical normal operation data;
and step 3: transmitting the operation data to a KPCA (kernel principal component analysis) anomaly detection model to obtain a covariance matrix of a higher-dimensional linear space; the operation data is the historical normal operation data or the real-time operation data;
and 4, step 4: performing eigenvector decomposition on the covariance matrix to obtain an eigenvalue and an eigenvector of a high-dimensional principal component space; meanwhile, the number of the principal elements is calculated according to the cumulative variance percentage (CPV)
；
And 5: determining a pivot space
And residual space
Calculating statistics based on the operating data based on the deterministic element space and the residual element spaceAnd SPE statistics;
wherein the content of the first and second substances,
the contribution ratio calculation formula is as follows:
for the SPE statistic, the calculation is as follows:
step 6: based onCalculating the historical normal operation data according to the steps 3, 4 and 5 to obtain statistics
And a historical normal operation threshold for the SPE statistics;
based on the real-time operation data, calculating according to the steps 3, 4 and 5And a real-time value of the SPE statistic;
comparing the real-time value with the historical normal operation threshold value, if the real-time value has
Or the SPE statistic exceeds the historical normal operation threshold, judging that a fault exists in the nuclear power device;
and 7: when the nuclear power device is judged to have faults, inputting the real-time fault data collected in the step 6 into a support vector machine for mode classification, and screening out fault reasons from multiple suspected faults;
the support vector machine is a mathematical model obtained by transferring historical fault sample data to the support vector machine with the supervision classification mode identification method and performing off-line training on the historical fault sample data when the nuclear power device is off-line.
, calculating the percentage of contribution rate of each measurement parameter to the T2 statistic in step 6, if only measurement parameters have contribution rate to the T2 statistic higher than the historical normal operation threshold, determining that the sensor has a fault, prompting an operator to repair the corresponding fault sensor and the corresponding line, and not starting step 7, and if the contribution rate of a plurality of measurement parameters to the T2 statistic is higher than the historical normal operation threshold, determining that the nuclear power plant has fault, and immediately activating step 7.
, step 8, if the preliminary diagnosis result of the support vector machine is a general fault, confirming and verifying the diagnosis result of the support vector machine, firstly, calling sample data after the fault occurs from the historical fault sample data and taking the time length of the sample data as ms, then, reading the running data after the true fault with the same time length ms to obtain groups of data matrixes of [2m, n ], wherein n is the dimension of the measured data, and if the diagnosis result is a special fault, confirming the diagnosis result of the support vector machine and evaluating the severity of the fault, then, calling k fault degrees from the historical fault sample data, wherein the time length of the sample number under each fault degree is m, and then, still taking the running data after the true fault with the same time length ms to obtain groups of data matrixes of [ (k +1) m, n ];
and step 9: after the [ (k +1) m, n ] dimensional data matrix is obtained, the KPCA in the step 3 is started again according to actual conditions to carry out kernel principal component decomposition, and a covariance matrix of a higher-dimensional linear space is obtained;
step 11: respectively performing data fitting on the historical fault sample data and the real-time fault data by utilizing an Euclidean distance function and combining a model-based clustering algorithm on the dimensionality reduction data matrix to obtain respective polynomial fitting curves, wherein time in the curves is used as an independent variable, and a distance value Dit is used as a dependent variable; performing data integration sigma Dit on all the fitting curves obtained by calculation according to time; after the integral values of the real-time fault data and the historical fault sample data are obtained, synchronously comparing the obtained parameters, and selecting the historical fault sample data integral value sigma Dit with the smallest difference value with the integral value sigma dt of the real-time fault data, wherein the corresponding fault information is the final evaluation result;
if the difference between Σ dt and all Σ Dit is large, two Σ Dit having the smallest difference from Σ dt are selected, and steps 8 to 10 are repeated within the failure degrees corresponding to the two Σ Dit until a relatively accurate failure result is achieved.
And , the fault analysis result can be directly displayed on the human-computer interface.
Compared with the prior art, the method is not limited by the coupling influence between system abnormity and sensor abnormity in the fault monitoring process, can distinguish and position sensor-level, system-level and equipment-level abnormity simultaneously through the change of characteristic parameters, avoids resource waste of subsequent diagnosis and analysis caused by the fault of the sensor, has higher accuracy, reduces the probability of misjudgment when the nuclear power device fails, improves the safety and reliability of the nuclear power device, is easy to learn and train the operation process, and has strong practicability.
Drawings
FIG. 1 is a table of operating parameters collected in a nuclear power system;
FIG. 2 is an overall functional flow of fault diagnosis of types of base data driven nuclear power systems;
FIG. 3 is a flow chart of an anomaly detection process;
FIG. 4 is a flow chart of a fault validation and evaluation process;
FIG. 5 is a graph of the rate of contribution of measured parameters to the T2 statistic for a single sensor failure;
FIG. 6 is a graph of the contribution rate of measured parameters to the T2 statistic at system level failure;
FIG. 7 historical fault samples and real-time fault data for a voltage regulator water level;
FIG. 8 historical fault samples and real-time fault data for fluctuating tube temperatures;
FIG. 9 historical fault samples and real-time fault data for charging line pressure;
FIG. 10 support vector machine preliminary diagnostic results;
FIG. 11 cluster analysis results based on Euclidean distance;
fig. 12 shows the clustering result based on euclidean distance and data integration.
Detailed Description
In order to better understand the present invention, the basic concepts involved in the present invention will first be briefly described:
the KPCA has the basic idea that nonlinear separable original spaces are changed into linear separable high-dimensional feature spaces through nonlinear mapping, and principal component analysis is completed in the new space.
The Euclidean distance function has the advantages of easiness in understanding, simplicity in use and the like, has outstanding shortcomings, and particularly solves the problem that when multi-element data analysis is carried out, the distance value of the Euclidean distance has a large relation with dimension selection of each parameter, results are different when different dimensions are selected, and the correlation among the parameters is not considered in the Euclidean distance.
A support vector machine: the VC dimension and structure risk minimization principle based on the statistical learning theory is provided, the generalization capability of a learning machine can be improved by seeking for structure risk minimization according to limited data sample information, the minimization of experience risk and a confidence range is realized, and tasks such as mode classification, regression analysis and the like can be performed.
Typical failure: design benchmark accidents mainly include main coolant system pipeline breakage, control rod out-of-control lifting, control rod drop accidents and the like.
General failure: the phenomenon that the states of equipment such as a pump, a valve and the like deviate from the set values of corresponding working conditions due to faults or misoperation mainly comprises misoperation of the valve, non-opening according to specified conditions, non-closing according to specified conditions and the like.
The technical solution of the present invention is further described in step by embodiments in conjunction with the accompanying drawings.
As shown in FIG. 2, the invention is a fault diagnosis method for nuclear power devices based on data driving, as shown in FIG. 3, feature vector decomposition is carried out on a covariance matrix to obtain a feature value and a feature vector of a high-dimensional pivot space, and meanwhile, the number of pivot is calculated according to cumulative variance percentage (CPV)
(ii) a As shown in fig. 4, when it is determined that the nuclear power plant has a fault, the collected real-time fault data is input into a support vector machine for pattern classification, and the most probable fault cause is screened out from a plurality of suspected faults.
The method adopts a 300MW Qinshan -stage nuclear power station full-range simulator as an actual object to be diagnosed, and sets that after the nuclear power station normally operates for 200s, a cold pipe section 1cm of the main coolant system occurs2And (4) a minor breach fault, wherein the fault belongs to a special fault.
Firstly, historical operating data such as measured parameters of pressure, temperature, flow, water level and the like are unidirectionally collected from an operating database of the nuclear power plant to form historical normal measuring data, and part of the measuring parameters are shown in fig. 1.
The principle of distinguishing the faults of the sensors and the abnormalities in the system and equipment based on the KPCA (kernel principal component analysis) abnormality detection method is that the design and manufacturing process in the nuclear power device is high, and the abnormality of a plurality of sensors cannot happen simultaneously, so that the sensor faults are considered to happen when only measuring points in all measuring point parameters are obviously abnormal, and if all the parameters are obviously deviated from normal values, the sensor faults are considered to happenFaults are considered to occur in the nuclear power system and devices. Due to the fact that the nuclear power system is provided with a plurality of parameters, the KPCA can be used for analyzing high-order statistic variation of related parameters, and abnormal reasons can be analyzed quickly and accurately. Transferring the measured data of the figure 1 to a KPCA (kernel principal component analysis) anomaly detection model, and obtaining a covariance matrix of a higher-dimensional linear space through a radial basis kernel function; performing eigenvector decomposition on the covariance matrix to obtain an eigenvalue and an eigenvector of a high-dimensional principal component space; meanwhile, the number of the principal elements is calculated according to the cumulative variance percentage (CPV)
。
To verify the KPCA detection in the event of a single sensor failure, the regulator surge tube temperature was first randomly selected in real-time operating data and injected with a reading offset failure, while the other parameters remained in the original state. After all parameters are subjected to feature extraction through KPCA, all measuring points of the KPCA are subjected to T2The trend of the contribution rate of the statistics with time is shown in FIG. 5, and T corresponding to the fluctuation tube temperature of the voltage stabilizer can be observed2The contribution rate obviously exceeds other measuring points, and only exceeds the set contribution rate threshold, so that the abnormality of the sensor can be quickly detected, and the operator can be assisted to directly deal with related problems. The subsequent unnecessary analysis and calculation caused by the abnormality of the sensor is avoided, and the possibility of misdiagnosis is reduced.
Then, randomly insert 1cm cold pipe section on the actual object2To verify the results of KPCA detection of anomalies in systems and devices. FIG. 6 shows all pairs of test points T after the occurrence of the fault2Unlike the case where a single sensor abnormality occurs, the trend graph of the time-dependent change in the contribution rate of the statistics is such that the contribution rates of parameters including (pressurizer pressure, coolant flow rate of 1# loop, core outlet temperature) at a plurality of measurement points are relatively high and exceed the set T2The anomaly results may enable the follow-up module to proceed to fault location, identification, and severity assessment.
A large amount of training sample data are obtained by randomly selecting a small cold pipe section break, a small heat pipe section break, a steam generator heat transfer pipe break, a main pump break shaft and a single group of control rods from historical fault sample data, and meanwhile, relevant measurement parameters of a reactor, an loop system and a chemical and volume control system are extracted, as shown in the figures 7, 8 and 9, part of actual operation data in a fault sample database are displayed, in order to show the effectiveness of the training sample data, the actual fault data and the historical sample data under three most similar faults are compared and displayed, and the sample data under other faults with larger parameter change difference are not displayed at .
After sample data is obtained, a support vector machine is used for carrying out initial diagnosis on faults, a kernel function of the support vector machine is selected as a Radial Basis Function (RBF), data is classified to (0, 1) by default, the noise intensity is set to be 30dBW, meanwhile, is selected for constructing a multi-classification nonlinear SVM by a method, and after a classical grid search algorithm is combined with the sample data for training and optimizing, the optimal values of a penalty factor c in the SVM and the optimal value of the width g in the RBF kernel function are 1.121 and 578.18 respectively.
Fig. 10 shows the preliminary diagnosis result of the support vector machine. In the figure, 0-3 represent fault type labels respectively, and other fault phenomena are too far away from the present example and are not shown in detail, in the figure, the horizontal axis represents time, and the vertical axis represents fault labels, wherein 0 represents normal operation, 1 represents 1# cold pipe section micro-break, 2 represents 1# heat pipe section micro-break, and 3 represents pressure regulator steam space rupture. From the test results, at the start of the failure, most of the data was classified as normal operation because the failure characteristics were not obvious, but after 15s, the diagnosis results thereof substantially coincided with the actual conditions. And besides the classification result at the starting moment, the final diagnosis accuracy is 94.89%, and the related requirements of the primary fault diagnosis are met.
After the initial diagnosis is carried out by using the support vector machine, because the model has the characteristic of black box and cannot be understood and convinced by operating personnel, the result of the model needs to be verified, and meanwhile, for the small break of the cold pipe section in the test example, steps are carried out to evaluate the fault severity so as to take more targeted decision and relief measures in the actual operation process.
In the process of dimension reduction by using KPCA, all historical sample data and real-time fault data system are integrated into an integral data matrix, so that kernel principal component decomposition in a high-dimensional space can be carried out in a system scale in a physical sense, an approximate value of which the dimension reduced low-dimensional vector can approximately represent original high-dimensional data is obtained, then, after control variables are analyzed through comparison, the data are determined to be reduced to be between and (0, 1), the noise intensity is set to be 30dBW to simulate the noise of an actual process, a kernel function is selected as a Radial Basis Function (RBF), the default value of the data sample size is 300 groups of data under each fault, and in order to observe the visualization effect, all measured point data are directly reduced to 3 dimensions.
Subsequently, on the basis of kernel principal component analysis, the clustering calculation process is continuously tested, and through control variables and comparative analysis, the clustering effect of the data on the reduced-dimension data is observed by using a euclidean distance function, as shown in fig. 11, a real-time clustering curve is obtained by using the euclidean distance.
As can be seen from FIG. 11, another benefits of using kernel principal component analysis and Euclidean distance are that the distance value is easy to understand and faster in calculation speed, and from FIG. 11, it can be seen that the distance value of the simulation data with the minimum fault degree is always at the bottom, the actual data is next to the simulation data corresponding to the fault degree 2, and other data are also strictly ordered in sequence according to the severity of the fault degree, so that the operator can visually observe the change of the fault degree.
In order to make the clustering result clearer and more visual, the data are fittedThe curves and the integration are shown in fig. 12 to obtain the cluster evaluation curve of the data. It can be seen from the figure that the actual data substantially completely coincides with the simulated data corresponding to the degree of failure 3 in the first 100s after the occurrence of the failure. And in the later stage of the fault, the actual clustering evaluation curve is positioned between the evaluation curves corresponding to the fault degree 2 and the fault degree 3. Because the low-dimensional parameters subjected to kernel principal component analysis dimensionality reduction can approximately represent the estimated value of the original data, the high interpretability of the low-dimensional parameters enables operating personnel to finally determine that the actually occurring fault degree is 2 (2 cm) of the fault degree2Breach) and degree of failure 3 (4 cm)2Lacerations).
The above description is only a preferred embodiment of the present invention, and is not intended to limit the technical scope of the present invention, so that any minor modifications, equivalent changes and modifications made to the above embodiment according to the technical spirit of the present invention are within the technical scope of the present invention.
Claims (4)
1, A fault diagnosis method for a nuclear power device based on data driving, characterized in that, the fault diagnosis method comprises the following steps:
step 1: acquiring historical normal operation data, namely unidirectionally acquiring historical normal operation measurement data from a historical operation database of the nuclear power device to form historical normal operation data;
step 2, unidirectionally acquiring real-time operation data from a system database of the nuclear power device, wherein the type and the dimension of the real-time operation data correspond to the historical normal operation data ;
and step 3: transmitting the operation data to a KPCA (kernel principal component analysis) anomaly detection model to obtain a covariance matrix of a higher-dimensional linear space; the operation data refers to the historical normal operation data or the real-time operation data;
and 4, step 4: performing eigenvector decomposition on the covariance matrix to obtain an eigenvalue and an eigenvector of a high-dimensional principal component space; meanwhile, the number of the principal elements is calculated according to the cumulative variance percentage (CPV)
；
And 5: determining a pivot space
And residual space
Calculating statistics based on the operating data based on the deterministic element space and the residual element space
And SPE statistics;
wherein the content of the first and second substances,
the contribution ratio calculation formula is as follows:
for the SPE statistic, the calculation is as follows:
step 6: based on the historical normal operation data, calculating according to the steps 3, 4 and 5 to obtain statistics
And a historical normal operation threshold for the SPE statistics;
based on the real-time operation data, calculating according to the steps 3, 4 and 5And a real-time value of the SPE statistic;
comparing the real-time value with the historical normal operation threshold value, if the real-time value is in the real-time valueIs provided withOr the SPE statistic exceeds the historical normal operation threshold, judging that a fault exists in the nuclear power device;
and 7: when the nuclear power device is judged to have faults, inputting the real-time fault data collected in the step 6 into a support vector machine for mode classification, and screening out fault reasons from multiple suspected faults;
and the support vector machine is a model obtained by transferring historical fault sample data to the support vector machine with the supervision classification mode identification method and performing off-line training on the historical fault sample data when the nuclear power device is off-line.
2. The fault diagnosis method according to claim 1, characterized in that: in step 6, each measurement parameter pair T is calculated2Percentage contribution of statistic if only measurement parameters are given to T2Judging that the sensor has a fault if the contribution rate of the statistic is higher than the historical normal operation threshold value, prompting an operator to overhaul the corresponding faulty sensor and the corresponding line thereof, and not starting the step 7; if multiple measured parameters are paired with T2And (4) judging that the fault occurs in the nuclear power device if the contribution rate of the statistic is higher than the historical normal operation threshold value, and immediately activating the step 7.
3. The fault diagnosis method according to claim 1, characterized in that:
step 8, confirming and verifying the diagnosis result of the support vector machine if the preliminary diagnosis result of the support vector machine is a general fault, firstly calling sample data after the fault occurs from the historical fault sample data and taking the time length of the sample data as ms, then reading the operation data after the true fault with the same time length ms to obtain groups of [2m, n ] dimensional data matrixes, wherein n is the dimension of the measured data, and confirming the diagnosis result of the support vector machine and evaluating the severity of the fault if the diagnosis result is a special fault, at the moment, calling k fault degrees from the historical fault sample data and taking the time length of the sample number under each fault degree as m, and then still taking the operation data after the true fault with the same time length ms to obtain groups of [ (k +1) m, n ] dimensional data matrixes;
and step 9: after the [ (k +1) m, n ] dimensional data matrix is obtained, the KPCA in the step 3 is started again according to actual conditions to carry out kernel principal component decomposition, and a covariance matrix of a higher-dimensional linear space is obtained;
step 10, repeating the covariance matrix obtained in the step 9 to the step 4 to obtain the eigenvalue and the eigenvector of the high-dimensional principal component space and obtain the number of the kernel principal componentsl(ii) a Obtaining a nonlinear kernel principal component eigenvector of the comprehensive data matrix, and calculating to obtain a projection of the nonlinear principal component eigenvector on a feature space, namely a final dimension-reduced data estimation value; carrying out reverse decomposition on the obtained dimension reduction matrix according to the step 8 to obtain independent dimension reduction data matrixes;
step 11: respectively performing data fitting on the historical fault sample data and the real-time fault data by utilizing an Euclidean distance function and combining a model-based clustering algorithm on the dimensionality reduction data matrix to obtain respective polynomial fitting curves, wherein time in the curves is used as an independent variable, and a distance value Dit is used as a dependent variable; performing data integration sigma Dit on all the fitting curves obtained by calculation according to time; after the integral values of the real-time fault data and the historical fault sample data are obtained, synchronously comparing the obtained parameters, and selecting the historical fault sample data integral value sigma Dit with the smallest difference value with the integral value sigma dt of the real-time fault data, wherein the corresponding fault information is the final evaluation result;
if the difference between Σ dt and all Σ Dit is large, two Σ Dit having the smallest difference from Σ dt are selected, and steps 8 to 10 are repeated within the failure degrees corresponding to the two Σ Dit until a relatively accurate failure result is achieved.
4. The method of any one of claims 1-3 through wherein the results of the fault analysis are displayed directly on a human-machine interface.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911026515.8A CN110738274A (en) | 2019-10-26 | 2019-10-26 | nuclear power device fault diagnosis method based on data driving |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911026515.8A CN110738274A (en) | 2019-10-26 | 2019-10-26 | nuclear power device fault diagnosis method based on data driving |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110738274A true CN110738274A (en) | 2020-01-31 |
Family
ID=69271533
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911026515.8A Pending CN110738274A (en) | 2019-10-26 | 2019-10-26 | nuclear power device fault diagnosis method based on data driving |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110738274A (en) |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111680725A (en) * | 2020-05-28 | 2020-09-18 | 哈尔滨工业大学 | Gas sensor array multi-fault isolation algorithm based on reconstruction contribution |
| CN111767657A (en) * | 2020-07-09 | 2020-10-13 | 哈尔滨工程大学 | Nuclear power system fault diagnosis method and system |
| CN111797533A (en) * | 2020-07-09 | 2020-10-20 | 哈尔滨工程大学 | Nuclear power device operation parameter abnormity detection method and system |
| CN111816338A (en) * | 2020-06-08 | 2020-10-23 | 核动力运行研究所 | Health monitoring and fault positioning system and method for nuclear power plant information system |
| CN111881594A (en) * | 2020-08-05 | 2020-11-03 | 哈尔滨工程大学 | Non-stationary signal state monitoring method and system for nuclear power equipment |
| CN111881176A (en) * | 2020-07-07 | 2020-11-03 | 中国人民解放军海军工程大学 | Marine nuclear power anomaly detection method based on logical distance characterization safety operation domain |
| CN112036087A (en) * | 2020-09-02 | 2020-12-04 | 哈尔滨工程大学 | Multi-strategy fused nuclear power key equipment fault diagnosis method and system |
| CN112036568A (en) * | 2020-07-09 | 2020-12-04 | 中国人民解放军海军工程大学 | Intelligent diagnosis method for damage fault of primary loop coolant system of nuclear power plant |
| CN112033463A (en) * | 2020-09-02 | 2020-12-04 | 哈尔滨工程大学 | Nuclear power equipment state evaluation and prediction integrated method and system |
| CN112199890A (en) * | 2020-10-11 | 2021-01-08 | 哈尔滨工程大学 | System-level fault diagnosis method for integrated nuclear power device |
| CN112992392A (en) * | 2021-02-19 | 2021-06-18 | 哈尔滨工程大学 | Leakage test section before pressure-bearing pipeline breaks |
| CN113254925A (en) * | 2021-02-01 | 2021-08-13 | 中国人民解放军海军工程大学 | Network intrusion detection system based on PCA and SVM |
| CN113406537A (en) * | 2020-03-16 | 2021-09-17 | 上海长庚信息技术股份有限公司 | Quantitative evaluation method for fault degree of power equipment |
| CN113591406A (en) * | 2021-07-26 | 2021-11-02 | 西安交通大学 | Method and system for optimizing arrangement and fault diagnosis of heat pipe cooling reactor measuring points |
| CN113719446A (en) * | 2021-08-31 | 2021-11-30 | 华能国际电力股份有限公司上海石洞口第一电厂 | Steam feed pump state monitoring system based on data mining |
| CN114112390A (en) * | 2021-11-23 | 2022-03-01 | 哈尔滨工程大学 | Early fault diagnosis method for nonlinear complex system |
| CN114118289A (en) * | 2021-12-02 | 2022-03-01 | 中国石油大学(北京) | Data-driven identification method and system for operating conditions of finished oil pipeline |
| CN114666117A (en) * | 2022-03-17 | 2022-06-24 | 国网浙江省电力有限公司信息通信分公司 | Network security situation measuring and predicting method for power internet |
| CN115095534A (en) * | 2022-04-11 | 2022-09-23 | 中核核电运行管理有限公司 | KPCA-based CANDU6 reactor main pump fault diagnosis method |
| CN115169709A (en) * | 2022-07-18 | 2022-10-11 | 华能汕头海门发电有限责任公司 | Power station auxiliary machine fault diagnosis method and system based on data driving |
| CN117076915A (en) * | 2023-10-17 | 2023-11-17 | 中海油能源发展股份有限公司采油服务分公司 | Intelligent fault attribution analysis method and system for FPSO crude oil process system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130035910A1 (en) * | 2010-09-29 | 2013-02-07 | Yingwei Zhang | Continuous annealing process fault detection method based on recursive kernel principal component analysis |
| CN103699117A (en) * | 2013-12-18 | 2014-04-02 | 中广核核电运营有限公司 | Method and system for diagnosing failure based on actual working conditions of nuclear power plant and simulation system |
| CN107301884A (en) * | 2017-07-24 | 2017-10-27 | 哈尔滨工程大学 | A kind of hybrid nuclear power station method for diagnosing faults |
| CN107316057A (en) * | 2017-06-07 | 2017-11-03 | 哈尔滨工程大学 | Based on the nuclear power unit method for diagnosing faults being locally linear embedding into K nearest neighbor classifiers |
| CN108062565A (en) * | 2017-12-12 | 2018-05-22 | 重庆科技学院 | Double pivots-dynamic kernel principal component analysis method for diagnosing faults based on chemical industry TE processes |
-
2019
- 2019-10-26 CN CN201911026515.8A patent/CN110738274A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130035910A1 (en) * | 2010-09-29 | 2013-02-07 | Yingwei Zhang | Continuous annealing process fault detection method based on recursive kernel principal component analysis |
| CN103699117A (en) * | 2013-12-18 | 2014-04-02 | 中广核核电运营有限公司 | Method and system for diagnosing failure based on actual working conditions of nuclear power plant and simulation system |
| CN107316057A (en) * | 2017-06-07 | 2017-11-03 | 哈尔滨工程大学 | Based on the nuclear power unit method for diagnosing faults being locally linear embedding into K nearest neighbor classifiers |
| CN107301884A (en) * | 2017-07-24 | 2017-10-27 | 哈尔滨工程大学 | A kind of hybrid nuclear power station method for diagnosing faults |
| CN108062565A (en) * | 2017-12-12 | 2018-05-22 | 重庆科技学院 | Double pivots-dynamic kernel principal component analysis method for diagnosing faults based on chemical industry TE processes |
Non-Patent Citations (4)
| Title |
|---|
| MA J 等: "Fault detection and identification in NPP instruments using kernel principal component analysis", 《INTERNATIONAL CONFERENCE ON NUCLEAR ENGINEERING》 * |
| 杜兴富: "基于支持向量机的核动力装置故障诊断", 《中国优秀博硕士学位论文全文数据库(硕士) 工程科技Ⅱ辑》 * |
| 王航: "模型驱动的核电站混合式故障诊断策略研究", 《中国优秀博士学位论文全文数据库(博士) 工程科技Ⅱ辑》 * |
| 陈超: "基于数据驱动的核电厂故障诊断技术研究", 《中国优秀博硕士学位论文全文数据库(硕士) 工程科技Ⅱ辑》 * |
Cited By (34)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113406537A (en) * | 2020-03-16 | 2021-09-17 | 上海长庚信息技术股份有限公司 | Quantitative evaluation method for fault degree of power equipment |
| CN113406537B (en) * | 2020-03-16 | 2024-05-07 | 上海长庚信息技术股份有限公司 | Quantitative evaluation method for fault degree of power equipment |
| CN111680725A (en) * | 2020-05-28 | 2020-09-18 | 哈尔滨工业大学 | Gas sensor array multi-fault isolation algorithm based on reconstruction contribution |
| CN111680725B (en) * | 2020-05-28 | 2023-05-05 | 哈尔滨工业大学 | Gas sensor array multi-fault isolation algorithm based on reconstruction contribution |
| CN111816338A (en) * | 2020-06-08 | 2020-10-23 | 核动力运行研究所 | Health monitoring and fault positioning system and method for nuclear power plant information system |
| CN111881176A (en) * | 2020-07-07 | 2020-11-03 | 中国人民解放军海军工程大学 | Marine nuclear power anomaly detection method based on logical distance characterization safety operation domain |
| CN111767657A (en) * | 2020-07-09 | 2020-10-13 | 哈尔滨工程大学 | Nuclear power system fault diagnosis method and system |
| CN111797533A (en) * | 2020-07-09 | 2020-10-20 | 哈尔滨工程大学 | Nuclear power device operation parameter abnormity detection method and system |
| CN111797533B (en) * | 2020-07-09 | 2022-05-13 | 哈尔滨工程大学 | Nuclear power device operation parameter abnormity detection method and system |
| CN111767657B (en) * | 2020-07-09 | 2022-04-22 | 哈尔滨工程大学 | Nuclear power system fault diagnosis method and system |
| CN112036568A (en) * | 2020-07-09 | 2020-12-04 | 中国人民解放军海军工程大学 | Intelligent diagnosis method for damage fault of primary loop coolant system of nuclear power plant |
| CN112036568B (en) * | 2020-07-09 | 2023-10-17 | 中国人民解放军海军工程大学 | Intelligent diagnosis method for damage faults of primary loop coolant system of nuclear power device |
| CN111881594A (en) * | 2020-08-05 | 2020-11-03 | 哈尔滨工程大学 | Non-stationary signal state monitoring method and system for nuclear power equipment |
| CN111881594B (en) * | 2020-08-05 | 2022-07-26 | 哈尔滨工程大学 | Non-stationary signal state monitoring method and system for nuclear power equipment |
| CN112033463A (en) * | 2020-09-02 | 2020-12-04 | 哈尔滨工程大学 | Nuclear power equipment state evaluation and prediction integrated method and system |
| CN112033463B (en) * | 2020-09-02 | 2022-09-06 | 哈尔滨工程大学 | Nuclear power equipment state evaluation and prediction integrated method and system |
| CN112036087A (en) * | 2020-09-02 | 2020-12-04 | 哈尔滨工程大学 | Multi-strategy fused nuclear power key equipment fault diagnosis method and system |
| CN112199890A (en) * | 2020-10-11 | 2021-01-08 | 哈尔滨工程大学 | System-level fault diagnosis method for integrated nuclear power device |
| CN113254925B (en) * | 2021-02-01 | 2022-11-15 | 中国人民解放军海军工程大学 | Network intrusion detection system based on PCA and SVM |
| CN113254925A (en) * | 2021-02-01 | 2021-08-13 | 中国人民解放军海军工程大学 | Network intrusion detection system based on PCA and SVM |
| CN112992392A (en) * | 2021-02-19 | 2021-06-18 | 哈尔滨工程大学 | Leakage test section before pressure-bearing pipeline breaks |
| CN112992392B (en) * | 2021-02-19 | 2022-12-09 | 哈尔滨工程大学 | Leakage test section before pressure-bearing pipeline breaks |
| CN113591406A (en) * | 2021-07-26 | 2021-11-02 | 西安交通大学 | Method and system for optimizing arrangement and fault diagnosis of heat pipe cooling reactor measuring points |
| CN113591406B (en) * | 2021-07-26 | 2024-03-29 | 西安交通大学 | Heat pipe cooling reactor measuring point optimal arrangement and fault diagnosis method and system |
| CN113719446A (en) * | 2021-08-31 | 2021-11-30 | 华能国际电力股份有限公司上海石洞口第一电厂 | Steam feed pump state monitoring system based on data mining |
| CN114112390B (en) * | 2021-11-23 | 2024-03-22 | 哈尔滨工程大学 | Nonlinear complex system early fault diagnosis method |
| CN114112390A (en) * | 2021-11-23 | 2022-03-01 | 哈尔滨工程大学 | Early fault diagnosis method for nonlinear complex system |
| CN114118289A (en) * | 2021-12-02 | 2022-03-01 | 中国石油大学(北京) | Data-driven identification method and system for operating conditions of finished oil pipeline |
| CN114118289B (en) * | 2021-12-02 | 2024-04-23 | 中国石油大学(北京) | Method and system for identifying operation condition of finished oil pipeline based on data driving |
| CN114666117A (en) * | 2022-03-17 | 2022-06-24 | 国网浙江省电力有限公司信息通信分公司 | Network security situation measuring and predicting method for power internet |
| CN115095534A (en) * | 2022-04-11 | 2022-09-23 | 中核核电运行管理有限公司 | KPCA-based CANDU6 reactor main pump fault diagnosis method |
| CN115169709A (en) * | 2022-07-18 | 2022-10-11 | 华能汕头海门发电有限责任公司 | Power station auxiliary machine fault diagnosis method and system based on data driving |
| CN117076915B (en) * | 2023-10-17 | 2024-01-09 | 中海油能源发展股份有限公司采油服务分公司 | Intelligent fault attribution analysis method and system for FPSO crude oil process system |
| CN117076915A (en) * | 2023-10-17 | 2023-11-17 | 中海油能源发展股份有限公司采油服务分公司 | Intelligent fault attribution analysis method and system for FPSO crude oil process system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110738274A (en) | nuclear power device fault diagnosis method based on data driving | |
| JP5421913B2 (en) | Fuzzy classification method for fault pattern matching cross-reference for related applications | |
| CN107844799B (en) | Water chilling unit fault diagnosis method of integrated SVM (support vector machine) mechanism | |
| CN104390657B (en) | A kind of Generator Unit Operating Parameters measurement sensor fault diagnosis method and system | |
| CN101446831B (en) | Decentralized process monitoring method | |
| CN105700518A (en) | Fault diagnosis method during industrial process | |
| CN110441065A (en) | Gas turbine online test method and device based on LSTM | |
| WO2011034805A1 (en) | Supervised fault learning using rule-generated samples for machine condition monitoring | |
| CN109472097B (en) | Fault diagnosis method for online monitoring equipment of power transmission line | |
| WO2021114320A1 (en) | Wastewater treatment process fault monitoring method using oica-rnn fusion model | |
| CN112199890A (en) | System-level fault diagnosis method for integrated nuclear power device | |
| CN116383636A (en) | Coal mill fault early warning method based on PCA and LSTM fusion algorithm | |
| Lindner et al. | Data-driven fault detection with process topology for fault identification | |
| CN112036087A (en) | Multi-strategy fused nuclear power key equipment fault diagnosis method and system | |
| CN115392782A (en) | Method and system for monitoring and diagnosing health state of process system of nuclear power plant | |
| CN116292367A (en) | Power plant fan system abnormal condition detection method based on one-dimensional convolution | |
| CN114519382A (en) | Nuclear power plant key operation parameter extraction and abnormity monitoring method | |
| CN112926656A (en) | Method, system and equipment for predicting state of circulating water pump of nuclear power plant | |
| CN110244690B (en) | Multivariable industrial process fault identification method and system | |
| US11339763B2 (en) | Method for windmill farm monitoring | |
| CN106874589A (en) | A kind of alarm root finding method based on data-driven | |
| CN113673600A (en) | Industrial signal abnormity early warning method, system, storage medium and computing equipment | |
| Baraldi et al. | A modified Auto Associative Kernel Regression method for robust signal reconstruction in nuclear power plant components | |
| Lee et al. | Event diagnosis method for a nuclear power plant using meta-learning | |
| CN113076211B (en) | Quality-related fault diagnosis and false alarm feedback method based on fault reconstruction |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200131 |