CN113591945A - Cross-power-level nuclear power device fault diagnosis method and system - Google Patents
Cross-power-level nuclear power device fault diagnosis method and system Download PDFInfo
- Publication number
- CN113591945A CN113591945A CN202110799159.4A CN202110799159A CN113591945A CN 113591945 A CN113591945 A CN 113591945A CN 202110799159 A CN202110799159 A CN 202110799159A CN 113591945 A CN113591945 A CN 113591945A
- Authority
- CN
- China
- Prior art keywords
- data
- fault diagnosis
- power level
- nuclear power
- parameter
- 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
- 238000003745 diagnosis Methods 0.000 title claims abstract description 96
- 238000000034 method Methods 0.000 title claims abstract description 63
- 238000013508 migration Methods 0.000 claims abstract description 51
- 230000005012 migration Effects 0.000 claims abstract description 51
- 238000009826 distribution Methods 0.000 claims abstract description 44
- 238000012544 monitoring process Methods 0.000 claims abstract description 39
- 238000013526 transfer learning Methods 0.000 claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 31
- 238000007781 pre-processing Methods 0.000 claims abstract description 18
- 238000012360 testing method Methods 0.000 claims abstract description 12
- 238000000605 extraction Methods 0.000 claims abstract description 9
- 238000010276 construction Methods 0.000 claims abstract description 7
- 238000005457 optimization Methods 0.000 claims description 28
- 238000004364 calculation method Methods 0.000 claims description 22
- 238000013527 convolutional neural network Methods 0.000 claims description 19
- 238000010606 normalization Methods 0.000 claims description 19
- 230000009467 reduction Effects 0.000 claims description 15
- 238000006243 chemical reaction Methods 0.000 claims description 14
- 239000002826 coolant Substances 0.000 claims description 14
- 239000011159 matrix material Substances 0.000 claims description 12
- 239000000126 substance Substances 0.000 claims description 10
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 238000010219 correlation analysis Methods 0.000 claims description 8
- 238000012546 transfer Methods 0.000 claims description 8
- 238000004422 calculation algorithm Methods 0.000 claims description 7
- 239000003086 colorant Substances 0.000 claims description 7
- 238000013507 mapping Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 4
- 238000005094 computer simulation Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 208000036829 Device dislocation Diseases 0.000 abstract description 2
- 238000005516 engineering process Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000012706 support-vector machine Methods 0.000 description 4
- 230000003044 adaptive effect Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 101100136092 Drosophila melanogaster peng gene Proteins 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000009257 reactivity Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 239000003381 stabilizer Substances 0.000 description 1
- 238000013024 troubleshooting Methods 0.000 description 1
- 238000012800 visualization 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
- 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
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Testing And Monitoring For Control Systems (AREA)
- Monitoring And Testing Of Nuclear Reactors (AREA)
Abstract
The invention provides a method and a system for fault diagnosis of a nuclear power device at a cross-power level, and particularly relates to a transfer learning method and a system for fault diagnosis of the nuclear power device under the difference of probability distribution of data of a training set and a test set. The invention comprises the following steps: the method comprises the steps of nuclear power device monitoring parameter acquisition, nuclear power device monitoring parameter preprocessing, nuclear power device migration fault diagnosis task data sorting, migratable feature extraction model construction in different fields, migration fault diagnosis model construction and target field fault identification. The method solves the problem that the generalization performance of the traditional data driving model is reduced under the condition of data probability distribution difference caused by different power levels, and can reliably apply the fault knowledge of the source power level to the fault identification of the target power level.
Description
Technical Field
The invention relates to a cross-power level nuclear power device fault diagnosis method and system, and belongs to the technical field of fault diagnosis.
Background
The nuclear power plant integrates a plurality of systems and components, such as a reactor coolant system, a nuclear auxiliary system, a steam turbine set and the like, and the safety of the nuclear power plant is very important in the whole process of the nuclear power plant. However, failure of critical components in the nuclear power plant may cause system abnormalities that may affect safe operation of the nuclear power plant, consuming a significant amount of maintenance costs. As an important component of nuclear power plants, the performance levels of instrumentation and control systems largely determine the safety and economics of the plant. The instrument control system can realize the functions of monitoring and diagnosing the running states of all the component systems and the process equipment of the nuclear power plant, and provides effective running support information for operators. Along with the conversion of the instrument control technology of the nuclear power plant from the analog technology to the digital technology, the realization of the monitoring and diagnosis of the running state of the nuclear power plant is facilitated. Therefore, more and more intelligent fault diagnosis methods are researched and applied to the nuclear power device, and the problems of excessive manual intervention, poor accuracy, low efficiency and the like in the traditional fault diagnosis are solved.
It is well known that nuclear power plant fault diagnosis techniques have experienced a long history of development. Generally, fault diagnosis methods are divided into three types, namely model-based, signal-based and data-driven, and are respectively and suitably applied to different scenes. Due to the increasing demands of instrument control systems on digitization and reliability and the rapid development of artificial intelligence in recent years, fault diagnosis technology based on data driving is developed vigorously and is widely researched. In order to overcome the problem of insufficient fault detection capability of a mathematical statistical reasoning method, Abiodun Ayodeji et al provides a nuclear power plant component-level fault diagnosis method of a data-driven Support Vector Machine (SVM), and verifies the effectiveness of the provided Multiclass SVM method based on Qinshan I stack type steam generator simulated fault data. However, shallow machine learning can only achieve good fault recognition effect in certain application scenarios. With the development of deep learning, the intelligent fault diagnosis model can deeply mine the corresponding rule of the multidimensional monitoring parameters and the faults, so that the fault diagnosis precision is improved. Peng et al combines correlation analysis and deep belief networks for system-level fault diagnosis of nuclear power plants, and verifies the superiority of the fault diagnosis performance relative to BP neural networks and SVM.
Although deep learning has achieved promising diagnostic results over the past few years, two challenges remain in the application of nuclear power plant fault diagnosis: (1) the nuclear power plant fault diagnosis tasks are usually performed based on the assumption that the training set and the test set obey the same distribution. However, in a complex monitoring environment of an actual nuclear power plant, fluctuation of working conditions, environmental noise interference and component material difference can cause inconsistent distribution of monitoring data. The data distribution difference can reduce the classification accuracy of the trained model on the test data. (2) The high safety requirements of nuclear power plants do not allow component level or system level failures to occur, making the monitoring data available for research nuclear power plants rare, and particularly the system level failure data with tags difficult to obtain. It is not practical to build fault diagnosis models for all distributed data based on traditional intelligent methods.
In order to solve the problem of low model generalization capability caused by data distribution difference and no label in the fault diagnosis of the nuclear power plant, a feasible intelligent technology is needed to reliably apply the training model to the fault identification of the test data with different probability distributions. As an emerging technology in the field of machine learning, transfer learning is a solution, and has been rapidly developed and widely studied in recent years in the fields of image recognition, text classification, and failure diagnosis. The key to the transfer learning technique is how to reduce the probability distribution difference of the source domain and target domain data. Long Wen et al propose a transfer learning method to the rotating machinery variable working condition fault diagnosis problem, measure the data distribution difference of source working condition and target working condition based on Maximum Mean redundancy (MMD), and combine the classification loss of the source working condition to form the neural network optimization goal, the test shows that the method based on transfer learning is obviously superior to the method without transfer learning in the variable working condition fault diagnosis task. Bin Yang et al propose a deep migration learning method based on a biochemical kernel induced MMD (PK-MMD), which can more accurately measure and reduce the feature distribution difference among different working conditions, and tests show that the method improves the calculation efficiency and accuracy of a migration diagnosis model. However, in the variable-operating-condition fault diagnosis of the nuclear power plant, when fault knowledge of a source operating condition is migrated to fault recognition of a target operating condition, the fault feature distributions under different operating conditions have large differences. Simply reducing the overall data distribution differences between domains tends to ignore the detailed feature space structure inside the domains.
To sum up, aiming at the problem that the generalization performance of the intelligent fault diagnosis of the nuclear power device is reduced when the working condition fluctuates, a set of reliable transfer learning fault diagnosis method or system is developed, the problem of distribution reduction of relevant local fields in different fields is considered, the data distribution difference is further reduced in detail, and the fault knowledge under the source power level is reliably applied to the fault identification of the target power level. The application of the transfer learning to the fault diagnosis of the nuclear power plant can widen the fault diagnosis application range and improve the generalization capability of the model, thereby having great practical significance to the stable operation of the nuclear power plant and the decision support of operators.
Disclosure of Invention
The invention aims to provide an intelligent fault diagnosis method and system for a transfer learning nuclear power device under different power levels, which aim to solve the problem that the generalization performance of a traditional data driving model is reduced under the condition of data probability distribution difference caused by different power levels and can reliably apply the fault knowledge of a source power level to the fault identification of a target power level.
The purpose of the invention is realized as follows: the method comprises the following steps:
step 1: collecting monitoring data of different operating conditions of the nuclear power device under different power levels;
step 2: carrying out data preprocessing operations such as parameter dimension reduction, parameter normalization and parameter-image conversion on the collected monitoring data;
and step 3: before fault diagnosis is carried out, data calibration and arrangement work of migration tasks is carried out, source power levels and target power levels are determined for different migration tasks, and labeled training sets and unlabeled test set data are divided;
and 4, step 4: developing migration learning fault diagnosis tasks under different power levels, and acquiring a well-trained cross-power level fault diagnosis model based on training set data;
and 5: and identifying the current operating condition by using the trained fault diagnosis model according to the collected operating data under the target power level.
The invention also includes such structural features:
1. the working conditions included in the collected operation monitoring parameters in the step 1 include normal working conditions, coolant loss accidents, steam pipeline rupture accidents outside containment, steam pipeline rupture accidents inside containment, steam generator heat transfer pipe rupture accidents and load shedding accidents.
2. The step 2 of obtaining the monitoring parameters of the nuclear power device comprises the following steps:
step 21: correlation analysis based on Pearson correlation coefficients is carried out on the operation parameters and the selected faults, and parameters related to low faults are eliminated;
step 22: the data normalization adopts a maximum difference method, and the calculation formula is as follows:
wherein x (n) is a data set of all operating states of the parameter under the same working condition, x*(n) is the normalized parameter value, xmin(n) and xmax(n) minimum and maximum values for the parameters;
step 23: and converting the normalized data into a two-dimensional matrix, and finally mapping the two-dimensional matrix into different colors in the two-dimensional image.
3. The establishing of the transfer learning fault diagnosis model based on the sub-field difference reduction in the step 4 comprises the following steps:
step 41: parameter set for nuclear power plant source power level
Obey to a probability distribution Ps(x) Parameter set at target power level
Obey to a probability distribution Pt(x) Inputting the data into a deep convolution neural network to extract features;
step 42: in order to reduce the classification loss while reducing the feature distribution difference, the optimization goals of the computational model are:
wherein the content of the first and second substances,
a source domain classification label is represented that,
representing prediction probability of source domain classifier output, cross entropy loss function
The weight coefficient lambda is larger than 0 and is used for coordinating the proportion of the classification loss and the sub-field self-adaptive loss in the total loss; dH(Xs,Xt) Calculating the sub-domain feature distribution difference of 1 network layer;
step 43: for the obtained model optimization target, carrying out convolution neural network parameter theta based on random gradient descent optimization algorithmfAnd fault classifier parameter θcIs iteratively updated.
4. A cross-power level nuclear power plant fault diagnostic system comprising:
the system comprises a data acquisition module, a data processing module and a data processing module, wherein the data acquisition module is used for acquiring monitoring data of different operating conditions of the nuclear power plant under different power levels, and the operating conditions contained in the acquired operating monitoring parameters comprise normal operating conditions, coolant loss accidents (cold pipe sections), coolant loss accidents (heat pipe sections), steam pipeline rupture accidents outside containment vessels, steam pipeline rupture accidents inside containment vessels, steam generator heat transfer pipe rupture accidents and load shedding accidents;
the data preprocessing module is used for performing data preprocessing operations such as parameter dimension reduction, parameter normalization, parameter-image conversion and the like on the acquired monitoring data;
the migration task data sorting module is used for sorting and dividing the preprocessed data aiming at the migration tasks among different power levels so as to input the preprocessed data to the migration learning model building module;
the transfer learning fault diagnosis model building module is used for building a fault diagnosis model based on a deep convolutional neural network and sub-field reduction transfer learning;
and the fault type output module is used for carrying out fault diagnosis on the nuclear power device under a target power level according to the transfer learning fault diagnosis model and outputting the type of the operation working condition.
5. The data preprocessing module specifically comprises:
the data correlation analysis unit is used for calculating the operation parameters and Pearson correlation coefficients of the selected faults and further rejecting parameters low in correlation with the faults;
the data normalization unit is used for carrying out normalization calculation on the monitoring parameters so as to eliminate dimension influence; the data normalization adopts a maximum difference method, and the calculation formula is as follows:
wherein x (n) is a data set of all operating states of the parameter under the same working condition, x*(n) is the normalized parameter value, xmin(n) and xmax(n) minimum and maximum values for the parameters;
and the parameter-image conversion unit is used for converting the normalized data into a two-dimensional matrix, and finally mapping the two-dimensional matrix into different colors in a two-dimensional image so as to adapt to the feature extraction unit.
6. The migration task data sorting module comprises:
the migration task confirming unit is used for determining a source power level and a target power level to be migrated based on migration tasks with different set power levels;
and the training data sorting unit is used for forming a training data pair by the data with the working condition type label under the source power level and the non-label data under the target power level after the task is definitely migrated.
7. The transfer learning fault diagnosis model building module comprises:
the migratable feature extraction unit is used for extracting data features of the source power level and the target power level by utilizing the deep convolutional neural network; the deep convolutional neural network comprises a plurality of convolutional layers and a pooling layer;
a model optimization objective calculation unit for reducing the classification loss while reducing the feature distribution difference, calculating an optimization objective of the model:
wherein the content of the first and second substances,
a source domain classification label is represented that,
representing prediction probability of source domain classifier output, cross entropy loss function
The weight coefficient lambda is larger than 0 and is used for coordinating the proportion of the classification loss and the sub-field self-adaptive loss in the total loss; dH(Xs,Xt) Calculating the sub-domain feature distribution difference of 1 network layer;
training modelA type iteration unit for performing the parameter theta of the convolutional neural network by adopting an optimization algorithm of random gradient descent based on the obtained model optimization targetfAnd fault classifier parameter θcThe back propagation iterative update of;
and the transfer learning fault diagnosis model construction unit is used for obtaining a transfer learning model under the fault diagnosis task after training is finished.
Compared with the prior art, the invention has the beneficial effects that: the invention provides a deep migration learning method and a deep migration learning system for cross-power level fault diagnosis of a nuclear power device, wherein a deep convolutional neural network is adopted to extract the operating condition characteristics of multi-dimensional system-level monitoring parameters, and the deep migration learning method and the deep convolutional neural network can be used for mining migratable characteristics of different power levels; the method comprises the steps of adopting a transfer learning method with characteristic reduction in the sub-field to ensure that the characteristic difference between a source power level and a target power level is reduced, determining an optimization target of a model by setting the difference between the characteristic of a working condition identification loss and the power level under the source power level, aligning the characteristic distribution of different working conditions in a finer granularity manner, and finally realizing the transfer of fault knowledge under the source power level to the target power level. Compared with the data-driven fault diagnosis method of the traditional nuclear power device, the migration learning fault diagnosis method has wider application range, breaks through the assumption that a training set and a test set need to obey the same distribution, and further improves the generalization performance of the fault diagnosis model.
Drawings
FIG. 1 is a flow chart of a transfer learning fault diagnosis method for a nuclear power plant at different power levels provided by the present invention;
FIG. 2 is a schematic diagram of a monitoring data preprocessing operation provided by the present invention;
FIG. 3 is a basic flow chart of fault diagnosis based on the deep migration learning method in the actual operation process of the present invention;
FIG. 4 is a schematic diagram illustrating the monitoring parameter-image conversion in the data preprocessing operation provided by the present invention;
FIG. 5 is a flow chart illustrating the calculation of an optimization objective in a migration fault diagnosis model provided by the present invention;
fig. 6 is a structural diagram of a nuclear power plant migration learning fault diagnosis system provided by the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.
The technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide an intelligent fault diagnosis method and system for a transfer learning nuclear power device under different power levels, which aim to solve the problem that the generalization performance of a traditional data driving model is reduced under the condition of data probability distribution difference caused by different power levels and can reliably apply the fault knowledge of a source power level to the fault identification of a target power level.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Fig. 1 is a flowchart of a migration learning fault diagnosis method for a nuclear power plant at different power levels, as shown in fig. 1, the migration learning fault diagnosis method for a nuclear power plant at different power levels includes:
step S1: monitoring data of different operating conditions of the nuclear power plant at different power levels are collected.
The step S1 specifically includes: the collected operation monitoring parameters include normal working conditions, coolant loss accidents (cold pipe sections), coolant loss accidents (hot pipe sections), out-of-containment steam pipe rupture accidents, in-containment steam pipe rupture accidents, steam generator heat transfer pipe rupture accidents and load shedding accidents.
Step S2: and carrying out data preprocessing operations such as parameter dimension reduction, parameter normalization, parameter-image conversion and the like on the acquired monitoring data.
As shown in fig. 2, when performing the data preprocessing, the step S2 specifically includes
Step S21: correlation analysis based on Pearson correlation coefficients is carried out on the operation parameters and the selected faults, and parameters related to low faults are eliminated;
step S22: the data normalization related to the fault adopts a maximum difference method, and the calculation formula is as follows:
wherein x (n) is a data set of all operating states of the parameter under the same working condition, x*(n) is the normalized parameter value, xmin(n) and xmax(n) minimum and maximum values for the parameters;
step S23: and converting the normalized data into a two-dimensional matrix, and finally mapping the two-dimensional matrix into different colors in a two-dimensional image.
Step S3: before fault diagnosis is carried out, data arrangement work of migration tasks is carried out, source power levels and target power levels are determined for different migration tasks, and labeled training sets and unlabeled test set data are divided.
Step S4: developing migration learning fault diagnosis tasks under different power levels, and acquiring a well-trained cross-power level fault diagnosis model based on training set data;
specifically, the migration feature extraction, the optimization target calculation, and the parameter iterative update in step S4 are respectively as follows:
step S41: parameter set for nuclear power plant source power level
Obey to a probability distribution Ps(x) Parameter set at target power level
Obey to a probability distribution Pt(x) Firstly, inputting the data into a deep convolutional neural network to extract features;
step S42: in order to reduce the classification loss while reducing the feature distribution difference, the optimization goals of the computational model are:
wherein the content of the first and second substances,
a source domain classification label is represented that,
representing prediction probability of source domain classifier output, cross entropy loss function
A weight coefficient lambda > 0 for coordinating the ratio of the classification loss to the sub-domain adaptive loss in the total loss, DH(Xs,Xt) The method is used for calculating the sub-domain feature distribution difference of 1 network layer and has the following formula:
wherein, Ps (c)And Pt (c)Probability distribution, weight coefficient of the sub-domain c
And
corresponding representative sample
And
probability of belonging to the sub-field c, having
The key point of the sub-field difference lies in the calculation of weight coefficients, and relates to a sub-field division method, if the sub-field is divided according to the operation working condition, a sample xkThe weighting factors for the sub-realm c are:
wherein the content of the first and second substances,
represents a sample xkFor the one-hot category label of category c, ∑ ycA one-hot class label sum representing the domain sample for class c;
step S43: for the obtained model optimization target, carrying out convolution neural network parameter theta based on random gradient descent optimization algorithmfAnd fault classifier parameter θcBack-propagation iterative update of (2):
and eta is the learning rate of network training, and after the training is finished, a transfer learning model under the fault diagnosis task is obtained.
Step S5: and identifying the current operating condition by using the trained fault diagnosis model according to the collected operating data under the target power level.
The method for diagnosing the migration learning fault of the nuclear power device under different power levels is applied to actual operation, and the specific flow is shown in fig. 3.
Step 1: the method comprises the steps that system-level monitoring parameters are collected or relevant parameters of a simulation simulator are obtained based on a nuclear power device intelligent instrument and a control system, wherein the collected data can comprise 100%, 90%, 80% and 70% of equal power levels, meanwhile, the working conditions under all the power levels can comprise normal working conditions, coolant loss accidents (cold pipe sections), coolant loss accidents (heat pipe sections), out-of-containment steam pipeline breakage accidents, in-containment steam pipeline breakage accidents, steam generator heat transfer pipe breakage accidents, load shedding accidents and other system-level transients, and all the system-level parameters can comprise data such as voltage stabilizer water level, steam generator pressure, coolant average temperature, down-flow, up-flow, total containment leakage, control rod reactivity, steam pipeline radiation quantity, steam turbine load power and the like.
Step 2: before the migration learning troubleshooting task is carried out, the data are sorted and calibrated. The data of the working conditions at different power levels are calibrated into fields A, B, C, D and the like, and different classification labels are calibrated for different working condition types at each power level.
And step 3: firstly, correlation analysis based on Pearson coefficients is carried out on all data in the step 2, and then monitoring parameters which are low in correlation with the selected working condition are removed.
And 4, step 4: and for the obtained fault related parameters, carrying out data normalization based on a maximum difference method, wherein the calculation formula is as follows:
wherein x (n) is a data set of all operating states of the parameter under the same working condition, x*(n) is the normalized parameter value, xmin(n) and xmaxAnd (n) is the minimum and maximum values of the parameter.
And 5: the resulting pre-processing parameters are subjected to parameter-image conversion to better adapt to the subsequent deep convolutional neural network, as shown in fig. 4. In sequence-image conversion, it is necessary to ensure the integrity and undistortion of information, so that the input form is suitable for the deep convolutional neural network, and simultaneously, the characteristic information in the conversion is ensured not to be distorted. Therefore, the acquired high-dimensional monitoring parameters at the same time are sequentially converted into n-x-n parameter matrixes, and the parameters are converted into different colors at corresponding positions in the image according to the normalized numerical value of each position in the matrixes. Therefore, the RGB value of the image extracted from the deep convolutional neural network is the running state information contained in the parameter.
Step 6: taking the migration fault diagnosis task a- > B as an example, the purpose is to apply the fault knowledge at the source power level a to the operating condition identification at the target power level B by using the correlation between the power levels a and B. The calculation flow of the optimized target of the migration fault diagnosis task A- > B is shown in FIG. 5. Therefore, at this time, the condition label in a is known, the condition label in B is unknown, the training set is composed of part data in a and B, and the test set is composed of the rest part of B data. Training set data is input into a pre-trained deep convolutional neural network to extract shared migratable features.
And 7: parameter set for nuclear power plant source power level
Obey to a probability distribution Ps(x) Parameter set at target power level
Obey to a probability distribution Pt(x) Based on the features extracted by the deep convolutional neural network under the source and target power levels, the optimization objective of the calculation model is as follows:
wherein the content of the first and second substances,
a source domain classification label is represented that,
to representPrediction probability of source domain classifier output, cross entropy loss function
A weight coefficient lambda > 0 for coordinating the ratio of the classification loss to the sub-domain adaptive loss in the total loss, DH(Xs,Xt) The method is used for calculating the sub-domain feature distribution difference of 1 network layer and has the following formula:
wherein, Ps (c)And Pt (c)Probability distribution, weight coefficient of the sub-domain c
And
corresponding representative sample
And
probability of belonging to the sub-field c, having
The key point of the sub-field difference lies in the calculation of weight coefficients, and relates to a sub-field division method, if the sub-field is divided according to the operation working condition, a sample xkThe weighting factors for the sub-realm c are:
wherein the content of the first and second substances,
represents a sample xkOn for class ce-hot sort tag, ∑ ycA one-hot class label sum representing the domain sample for class c;
and 8: for the obtained model optimization target, as shown in the migration model training in the flowchart 3, the convolutional neural network parameter θ is performed based on the optimization algorithm of random gradient descentfAnd fault classifier parameter θcBack-propagation iterative update of (2):
and eta is the learning rate of network training, and when the training is completed by the maximum iteration number, a transfer learning model under the fault diagnosis task A- > B is obtained.
And step 9: and identifying the type of the current operating condition by using the trained fault diagnosis model according to the collected operating data under the target power level. And (3) visually displaying the characteristic difference reduction effect among the power levels by utilizing a characteristic visualization technology T-SNE, and inputting the characteristic difference reduction effect into an operation support system by combining the current working condition identification result so as to assist an operator to make a decision. The results obtained contribute to the safety and economy of the nuclear power plant.
The fault diagnosis method and the fault diagnosis system adopt a deep migration learning strategy to realize fault diagnosis tasks of the nuclear power device under different power levels, are beneficial to solving the problem that a traditional data driving model is low in generalization capability under the condition of data distribution difference, can fully utilize fault knowledge under a source power level to develop the fault diagnosis task under a target power level based on correlation among fields, further widen the application of an intelligent fault diagnosis model in the nuclear power device, and obtain a better fault classification effect compared with the traditional machine learning.
Fig. 6 is a structural diagram of a migration learning fault diagnosis system of a nuclear power plant under different power levels, which includes:
the data acquisition module M1 is used for acquiring monitoring data of different operating conditions of the nuclear power plant under different power levels, wherein the operating conditions contained in the acquired operating monitoring parameters include normal operating conditions, coolant loss accidents (cold pipe sections), coolant loss accidents (heat pipe sections), steam pipeline rupture accidents outside containment, steam pipeline rupture accidents inside containment, steam generator heat transfer pipe rupture accidents and load shedding accidents;
the data preprocessing module M2 is used for performing data preprocessing operations such as parameter dimension reduction, parameter normalization and parameter-image conversion on the acquired monitoring data;
the data preprocessing module M2 specifically includes:
the data correlation analysis unit is used for calculating the operation parameters and Pearson correlation coefficients of the selected faults and further rejecting parameters low in correlation with the faults;
and the data normalization unit is used for performing normalization calculation on the monitoring parameters so as to eliminate dimensional influence. The data normalization adopts a maximum difference method, and the calculation formula is as follows:
wherein x (n) is a data set of all operating states of the parameter under the same working condition, x*(n) is the normalized parameter value, xmin(n) and xmax(n) minimum and maximum values for the parameters;
and the parameter-image conversion unit is used for converting the normalized data into a two-dimensional matrix, and finally mapping the two-dimensional matrix into different colors in a two-dimensional image so as to adapt to the feature extraction unit.
The migration task data sorting module M3 is used for sorting and dividing the preprocessed data for the migration tasks among different power levels to input the preprocessed data to the migration learning model building module;
the migration task data sorting module M3 specifically includes:
the migration task confirming unit is used for determining a source power level and a target power level to be migrated based on migration tasks with different set power levels;
and the training data sorting unit is used for forming a training data pair by the data with the working condition type label under the source power level and the non-label data under the target power level after the task is definitely migrated.
The transfer learning fault diagnosis model building module M4 is used for building a fault diagnosis model based on deep convolutional neural network and sub-field reduction transfer learning;
the migration learning fault diagnosis model building module M4 specifically includes:
the migratable feature extraction unit is used for extracting data features of the source power level and the target power level by utilizing the deep convolutional neural network; the deep convolutional neural network comprises a plurality of convolutional layers and a pooling layer;
a model optimization objective calculation unit for reducing the classification loss while reducing the feature distribution difference, calculating an optimization objective of the model:
wherein the content of the first and second substances,
a source domain classification label is represented that,
representing prediction probability of source domain classifier output, cross entropy loss function
And the weight coefficient lambda is more than 0 and is used for coordinating the proportion of the classification loss and the sub-field adaptive loss in the total loss. DH(Xs,Xt) The method is used for calculating the sub-domain feature distribution difference of 1 network layer and has the following formula:
wherein, Ps (c)And Pt (c)Respectively, the probability distribution of the sub-domain c. Weight coefficient
And
corresponding representative sample
And
probability of belonging to the sub-field c, having
The key point of the sub-field difference lies in the calculation of weight coefficients, and relates to a sub-field division method, if the sub-field is divided according to the operation working condition, a sample xkThe weighting factors for the sub-realm c are:
wherein the content of the first and second substances,
represents a sample xkFor the one-hot category label of category c, ∑ ycA one-hot class label sum representing the domain sample for class c;
a training model iteration unit for performing a convolution neural network parameter theta by using an optimization algorithm of random gradient descent based on the obtained model optimization targetfAnd fault classifier parameter θcBack-propagation iterative update of (2):
wherein eta is the learning rate of network training;
and the transfer learning fault diagnosis model construction unit is used for obtaining a transfer learning model under the fault diagnosis task after training is finished.
And the fault type output module M5 is used for carrying out fault diagnosis on the nuclear power device at a target power level according to the transfer learning fault diagnosis model and outputting the type of the operating condition.
In summary, the invention discloses a deep migration learning method and system for nuclear power plant fault diagnosis, and particularly relates to a migration learning method and system for performing nuclear power plant fault diagnosis under the difference of probability distribution of training set and test set data. The invention comprises the following steps: the method comprises the steps of nuclear power device monitoring parameter acquisition, nuclear power device monitoring parameter preprocessing, nuclear power device migration fault diagnosis task data sorting, migratable feature extraction model construction in different fields, migration fault diagnosis model construction and target field fault identification. The principles and implementation procedures of the present invention have been described based on specific embodiments, and it is easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention, and does not limit the technical scope of the present invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention still belong to the technical scope of the present invention.
Claims (8)
1. A cross-power level nuclear power plant fault diagnosis method is characterized by comprising the following steps: the method comprises the following steps:
step 1: collecting monitoring data of different operating conditions of the nuclear power device under different power levels;
step 2: carrying out data preprocessing operations such as parameter dimension reduction, parameter normalization and parameter-image conversion on the collected monitoring data;
and step 3: before fault diagnosis is carried out, data calibration and arrangement work of migration tasks is carried out, source power levels and target power levels are determined for different migration tasks, and labeled training sets and unlabeled test set data are divided;
and 4, step 4: developing migration learning fault diagnosis tasks under different power levels, and acquiring a well-trained cross-power level fault diagnosis model based on training set data;
and 5: and identifying the current operating condition by using the trained fault diagnosis model according to the collected operating data under the target power level.
2. The method of claim 1 for diagnosing a nuclear power plant fault across power levels, wherein: the working conditions included in the collected operation monitoring parameters in the step 1 include normal working conditions, coolant loss accidents, steam pipeline rupture accidents outside containment, steam pipeline rupture accidents inside containment, steam generator heat transfer pipe rupture accidents and load shedding accidents.
3. A cross-power level nuclear power plant fault diagnosis method according to claim 1 or 2, characterized in that: the step 2 of obtaining the monitoring parameters of the nuclear power device comprises the following steps:
step 21: correlation analysis based on Pearson correlation coefficients is carried out on the operation parameters and the selected faults, and parameters related to low faults are eliminated;
step 22: the data normalization adopts a maximum difference method, and the calculation formula is as follows:
wherein x (n) is a data set of all operating states of the parameter under the same working condition, x*(n) is the normalized parameter value, xmin(n) and xmax(n) minimum and maximum values for the parameters;
step 23: and converting the normalized data into a two-dimensional matrix, and finally mapping the two-dimensional matrix into different colors in the two-dimensional image.
4. The method of claim 3 for diagnosing a nuclear power plant fault across power levels, wherein: the establishing of the transfer learning fault diagnosis model based on the sub-field difference reduction in the step 4 comprises the following steps:
step 41: parameter set for nuclear power plant source power level
Obey to a probability distribution Ps(x) Parameter set at target power level
Obey to a probability distribution Pt(x) Inputting the data into a deep convolution neural network to extract features;
step 42: in order to reduce the classification loss while reducing the feature distribution difference, the optimization goals of the computational model are:
wherein the content of the first and second substances,
a source domain classification label is represented that,
representing prediction probability of source domain classifier output, cross entropy loss function
The weight coefficient lambda is larger than 0 and is used for coordinating the proportion of the classification loss and the sub-field self-adaptive loss in the total loss; dH(Xs,Xt) Represents calculation 1Sub-domain feature distribution differences of individual network layers;
step 43: for the obtained model optimization target, carrying out convolution neural network parameter theta based on random gradient descent optimization algorithmfAnd fault classifier parameter θcIs iteratively updated.
5. A cross-power level nuclear power plant fault diagnostic system, characterized by: the method comprises the following steps:
the system comprises a data acquisition module, a data processing module and a data processing module, wherein the data acquisition module is used for acquiring monitoring data of different operating conditions of the nuclear power plant under different power levels, and the operating conditions contained in the acquired operating monitoring parameters comprise normal operating conditions, coolant loss accidents (cold pipe sections), coolant loss accidents (heat pipe sections), steam pipeline rupture accidents outside containment vessels, steam pipeline rupture accidents inside containment vessels, steam generator heat transfer pipe rupture accidents and load shedding accidents;
the data preprocessing module is used for performing data preprocessing operations such as parameter dimension reduction, parameter normalization, parameter-image conversion and the like on the acquired monitoring data;
the migration task data sorting module is used for sorting and dividing the preprocessed data aiming at the migration tasks among different power levels so as to input the preprocessed data to the migration learning model building module;
the transfer learning fault diagnosis model building module is used for building a fault diagnosis model based on a deep convolutional neural network and sub-field reduction transfer learning;
and the fault type output module is used for carrying out fault diagnosis on the nuclear power device under a target power level according to the transfer learning fault diagnosis model and outputting the type of the operation working condition.
6. The system of claim 5, wherein the system further comprises a power level-crossing nuclear power plant fault diagnostic system, wherein the system further comprises: the data preprocessing module specifically comprises:
the data correlation analysis unit is used for calculating the operation parameters and Pearson correlation coefficients of the selected faults and further rejecting parameters low in correlation with the faults;
the data normalization unit is used for carrying out normalization calculation on the monitoring parameters so as to eliminate dimension influence; the data normalization adopts a maximum difference method, and the calculation formula is as follows:
wherein x (n) is a data set of all operating states of the parameter under the same working condition, x*(n) is the normalized parameter value, xmin(n) and xmax(n) minimum and maximum values for the parameters;
and the parameter-image conversion unit is used for converting the normalized data into a two-dimensional matrix, and finally mapping the two-dimensional matrix into different colors in a two-dimensional image so as to adapt to the feature extraction unit.
7. The system of claim 5, wherein the system further comprises a power level-crossing nuclear power plant fault diagnostic system, wherein the system further comprises: the migration task data sorting module comprises:
the migration task confirming unit is used for determining a source power level and a target power level to be migrated based on migration tasks with different set power levels;
and the training data sorting unit is used for forming a training data pair by the data with the working condition type label under the source power level and the non-label data under the target power level after the task is definitely migrated.
8. The system of claim 5, wherein the system further comprises a power level-crossing nuclear power plant fault diagnostic system, wherein the system further comprises: the transfer learning fault diagnosis model building module comprises:
the migratable feature extraction unit is used for extracting data features of the source power level and the target power level by utilizing the deep convolutional neural network; the deep convolutional neural network comprises a plurality of convolutional layers and a pooling layer;
a model optimization objective calculation unit for reducing the classification loss while reducing the feature distribution difference, calculating an optimization objective of the model:
wherein the content of the first and second substances,
a source domain classification label is represented that,
representing prediction probability of source domain classifier output, cross entropy loss function
The weight coefficient lambda is larger than 0 and is used for coordinating the proportion of the classification loss and the sub-field self-adaptive loss in the total loss; dH(Xs,Xt) Calculating the sub-domain feature distribution difference of 1 network layer;
a training model iteration unit for performing a convolution neural network parameter theta by using an optimization algorithm of random gradient descent based on the obtained model optimization targetfAnd fault classifier parameter θcThe back propagation iterative update of;
and the transfer learning fault diagnosis model construction unit is used for obtaining a transfer learning model under the fault diagnosis task after training is finished.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110799159.4A CN113591945A (en) | 2021-07-15 | 2021-07-15 | Cross-power-level nuclear power device fault diagnosis method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110799159.4A CN113591945A (en) | 2021-07-15 | 2021-07-15 | Cross-power-level nuclear power device fault diagnosis method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113591945A true CN113591945A (en) | 2021-11-02 |
Family
ID=78247502
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110799159.4A Pending CN113591945A (en) | 2021-07-15 | 2021-07-15 | Cross-power-level nuclear power device fault diagnosis method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113591945A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110866365A (en) * | 2019-11-22 | 2020-03-06 | 北京航空航天大学 | Mechanical equipment intelligent fault diagnosis method based on partial migration convolutional network |
| CN111651937A (en) * | 2020-06-03 | 2020-09-11 | 苏州大学 | Method for diagnosing similar self-adaptive bearing fault under variable working conditions |
| CN111709448A (en) * | 2020-05-20 | 2020-09-25 | 西安交通大学 | Mechanical fault diagnosis method based on migration relation network |
| CN112183581A (en) * | 2020-09-07 | 2021-01-05 | 华南理工大学 | Semi-supervised mechanical fault diagnosis method based on self-adaptive migration neural network |
| US20210133559A1 (en) * | 2019-11-04 | 2021-05-06 | Cisco Technology, Inc. | Iot-based network architecture for detecting faults using vibration measurement data |
| CN112906644A (en) * | 2021-03-22 | 2021-06-04 | 重庆大学 | Mechanical fault intelligent diagnosis method based on deep migration learning |
-
2021
- 2021-07-15 CN CN202110799159.4A patent/CN113591945A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210133559A1 (en) * | 2019-11-04 | 2021-05-06 | Cisco Technology, Inc. | Iot-based network architecture for detecting faults using vibration measurement data |
| CN110866365A (en) * | 2019-11-22 | 2020-03-06 | 北京航空航天大学 | Mechanical equipment intelligent fault diagnosis method based on partial migration convolutional network |
| CN111709448A (en) * | 2020-05-20 | 2020-09-25 | 西安交通大学 | Mechanical fault diagnosis method based on migration relation network |
| CN111651937A (en) * | 2020-06-03 | 2020-09-11 | 苏州大学 | Method for diagnosing similar self-adaptive bearing fault under variable working conditions |
| CN112183581A (en) * | 2020-09-07 | 2021-01-05 | 华南理工大学 | Semi-supervised mechanical fault diagnosis method based on self-adaptive migration neural network |
| CN112906644A (en) * | 2021-03-22 | 2021-06-04 | 重庆大学 | Mechanical fault intelligent diagnosis method based on deep migration learning |
Non-Patent Citations (1)
| Title |
|---|
| 毛冠通: "基于迁移学习的滚动轴承在线故障诊断", 《航空科学技术》, vol. 31, no. 01, pages 61 - 67 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110361176B (en) | Intelligent fault diagnosis method based on multitask feature sharing neural network | |
| CN106951695A (en) | Plant equipment remaining life computational methods and system under multi-state | |
| CN113283027B (en) | Mechanical fault diagnosis method based on knowledge graph and graph neural network | |
| CN102707256B (en) | Fault diagnosis method based on BP-Ada Boost nerve network for electric energy meter | |
| CN110348624B (en) | Sand storm grade prediction method based on Stacking integration strategy | |
| CN109672406B (en) | Photovoltaic power generation array fault diagnosis and classification method based on sparse representation and SVM | |
| CN113486578A (en) | Method for predicting residual life of equipment in industrial process | |
| CN109472097B (en) | Fault diagnosis method for online monitoring equipment of power transmission line | |
| CN114676742A (en) | Power grid abnormal electricity utilization detection method based on attention mechanism and residual error network | |
| CN111784061B (en) | Training method, device and equipment for power grid engineering cost prediction model | |
| CN114462718A (en) | CNN-GRU wind power prediction method based on time sliding window | |
| CN112507479B (en) | Oil drilling machine health state assessment method based on manifold learning and softmax | |
| Wang et al. | A deep transfer learning method for system-level fault diagnosis of nuclear power plants under different power levels | |
| CN111122811A (en) | Sewage treatment process fault monitoring method of OICA and RNN fusion model | |
| CN116186633A (en) | Power consumption abnormality diagnosis method and system based on small sample learning | |
| CN115409369A (en) | Comprehensive energy system reliability evaluation method based on mechanism and data hybrid driving | |
| CN201716727U (en) | Geographical simulation system based on remote sensing and GIS | |
| CN114462459A (en) | Hydraulic machine fault diagnosis method based on 1DCNN-LSTM network model | |
| CN115791174B (en) | Rolling bearing abnormality diagnosis method, system, electronic equipment and storage medium | |
| CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
| CN117113086A (en) | Energy storage unit load prediction method, system, electronic equipment and medium | |
| CN113591945A (en) | Cross-power-level nuclear power device fault diagnosis method and system | |
| Chen | Knowledge distillation with error-correcting transfer learning for wind power prediction | |
| CN116415485A (en) | Multi-source domain migration learning residual service life prediction method based on dynamic distribution self-adaption | |
| CN115217534A (en) | Method and system for monitoring service quality state of steam turbine |
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 |