CN114897138A - System fault diagnosis method based on attention mechanism and depth residual error network - Google Patents
System fault diagnosis method based on attention mechanism and depth residual error network Download PDFInfo
- Publication number
- CN114897138A CN114897138A CN202210484425.9A CN202210484425A CN114897138A CN 114897138 A CN114897138 A CN 114897138A CN 202210484425 A CN202210484425 A CN 202210484425A CN 114897138 A CN114897138 A CN 114897138A
- Authority
- CN
- China
- Prior art keywords
- network
- fault diagnosis
- attention mechanism
- data
- multiplied
- 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
- 230000007246 mechanism Effects 0.000 title claims abstract description 45
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000003745 diagnosis Methods 0.000 title claims abstract description 44
- 238000012549 training Methods 0.000 claims abstract description 20
- 238000003303 reheating Methods 0.000 claims abstract description 17
- 238000000926 separation method Methods 0.000 claims abstract description 17
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 17
- 238000013461 design Methods 0.000 claims abstract description 5
- 230000004913 activation Effects 0.000 claims description 22
- 238000011176 pooling Methods 0.000 claims description 17
- 230000006870 function Effects 0.000 claims description 14
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 11
- 238000005457 optimization Methods 0.000 claims description 8
- 230000006835 compression Effects 0.000 claims description 7
- 238000007906 compression Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 4
- 150000001875 compounds Chemical class 0.000 claims description 3
- 230000003213 activating effect Effects 0.000 claims description 2
- 238000000605 extraction Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 6
- 230000008878 coupling Effects 0.000 description 4
- 238000010168 coupling process Methods 0.000 description 4
- 238000005859 coupling reaction Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- 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
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M99/00—Subject matter not provided for in other groups of this subclass
- G01M99/005—Testing of complete machines, e.g. washing-machines or mobile phones
-
- 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
-
- 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
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E30/00—Energy generation of nuclear origin
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a system fault diagnosis method based on an attention mechanism and a deep residual error network, which is characterized in that an optimal network layer number and a convolution kernel number are obtained by using an artificial bee colony algorithm to design a network structure, so that the feature extraction capability of convolution operation on time sequence, multidimensional and noise interference data of a large-scale industrial system is fully exerted, and meanwhile, the attention mechanism based on channel attention and space attention is added among residual error blocks, so that the network can automatically learn the importance degrees of different channels and the importance degrees of different pixel points of input data, and the model training convergence speed and the fault diagnosis accuracy rate are accelerated. According to the method, the state variable time series data of the steam-water separation reheating system are input into the model, and high fault diagnosis accuracy is obtained under noise interference.
Description
Technical Field
The invention relates to the technical field of fault diagnosis of a nuclear power steam-water separation reheating system, in particular to a fault diagnosis algorithm based on an attention mechanism and a depth residual error network.
Background
A steam-water separation reheating system (MSR) is a main subsystem of a nuclear power two-loop system, and safe and efficient operation of the MSR has important significance on safety and economy of a nuclear power unit.
However, due to the increase of the number of nuclear power generating units under construction and the large amount of intermittent renewable energy sources in grid connection, the nuclear power generating units need to deeply participate in peak shaving frequency modulation, so that various typical faults are easy to occur in a steam-water separation reheating system under a time-varying working condition.
Therefore, the accurate diagnosis of the typical fault of the steam-water separation reheating system has important significance for safe and stable operation of the nuclear power unit under the time-varying working condition.
Current fault diagnosis techniques are largely classified into analytical model-based methods, empirical knowledge-based methods, and data-driven-based methods.
After the process system is analyzed by the method based on the analytical model, a function relation which can describe the system function and is formed by coupling mathematical formulas is abstracted, so that the analytical redundancy of the process system is obtained. Resolving the redundancy yields the difference between the measured value and the model estimate, i.e., the physical residual. The key to fault detection is to analyze the residual characteristics to identify, locate and isolate the fault. However, model-based methods rely on the accuracy of the model, and faults can only be identified efficiently when the model is accurate. However, the model established by analysis or experimental data cannot accurately reflect the steady-state and dynamic characteristics of the system, so that the method has a large limitation.
The method based on the empirical knowledge is suitable for complex industrial systems which are deeply known by people, and is fault recognition in the nuclear power field, and the classical method based on the empirical knowledge is an expert system. However, for such a complex system of a nuclear power plant, the currently established knowledge base is still insufficient to reflect the main accident characteristics, so that the development of an expert system is more dependent on the development of system knowledge.
With the rapid development of random computers and the internet, the operation data of the actual system can be more easily stored and mined, and thus a data-driven based method is widely used. However, when the method based on data driving is applied to the fault identification of a complex industrial system, there are some defects, such as large calculation amount, poor real-time performance, high requirement on the amount of data, and the like.
In recent years, a data-driven method represented by deep learning develops rapidly due to computer computing power improvement and big data and computing model progress, and has a wide prospect in fault diagnosis application.
However, the fault diagnosis method applied to complex industrial systems such as the steam-water separation reheating system is limited in practical application as follows: noise interference, data time sequence dimension and variable dimension coupling cause the fault diagnosis model aiming at the steam-water separation reheating system to be difficult to be applied to the ground.
Disclosure of Invention
The technical problems to be solved by the invention are as follows: the invention aims to solve the problem that a fault diagnosis method based on data driving is difficult to apply to fault diagnosis of an actual steam-water separation reheating system due to noise interference, data time sequence dimension and variable dimension coupling.
The invention provides the following technical scheme: the invention discloses a system fault diagnosis method based on an attention mechanism and a deep residual error network, which comprises the following steps,
s1, optimizing the number of network residual blocks and the number of convolution kernels of each layer in the network by using an artificial bee colony heuristic optimization algorithm so as to obtain a pareto optimal network structure meeting the maximum fault diagnosis accuracy rate and the minimum network calculation parameters;
s2, channel attention mechanisms and space attention mechanisms are added among the residual blocks in front and at back, so that the network training convergence speed and the fault diagnosis accuracy are improved;
s3, carrying out sliding window, edge zero filling and standardized processing operations to carry out preprocessing on the data;
and S4, inputting the two-dimensional image data into a network for training to finally obtain a trained network model, and performing state classification on the steam-water separation reheating system state data input into the network model so as to diagnose the system state.
Further, in the step S1, designing the network structure by using an artificial bee colony heuristic optimization algorithm, including the following steps,
taking the number of network residual blocks and the number of convolution kernels of each layer in the network as design variables;
taking the maximum fault diagnosis accuracy and the minimum network calculation parameter total amount after the network training for 20 periods as target variables;
the training period is set to be 100 times, then an artificial bee colony algorithm is used for automatically optimizing to obtain a pareto optimal solution set of fault diagnosis accuracy and network calculation parameter total amount, and the pareto optimal solution set corresponds to the number of optimal network residual block and the number of convolution kernels of each layer in the network, namely an optimal network structure.
Further, in the step S2, the channel attention mechanism includes the steps of,
sa1. compression, wherein the compression formula is as follows:
wherein u is c (v, w) represents a pixel (v, w) in the feature map of the channel c; z is a radical of c Representing the c global feature point; the input feature map size is H × W × C.
Sa2. activation, wherein the activation formula is as follows:
s c ＝σ(W 2 δ(W 1 z c ))
in the formula, sigma represents a sigmoid activation function; δ represents the Relu activation function; w 1 And W 2 Representing a weight of the fully connected layer; s c Represents the activated output value;
sa3. product, the product formula is:
in the formula (I), the compound is shown in the specification,
representing the weighted feature map.
Further, in the step S2, the spatial attention mechanism includes the steps of,
performing pooling operation, namely performing channel-based global maximum pooling and global average pooling on the feature map with the input size of H multiplied by W multiplied by C to obtain two feature planes with the size of H multiplied by W multiplied by 1, and then performing channel splicing on the two feature planes to finally obtain the feature map with the size of H multiplied by W multiplied by 2;
dimension reduction operation, namely reducing the dimension of the feature map with the size of H multiplied by W multiplied by 2 into the feature map with the size of H multiplied by W multiplied by 1 through a convolution kernel of 7 multiplied by 7;
activating operation, namely using a sigmoid activation function to generate weights of different positions, so that the model obtains the capability of distinguishing importance degrees of different positions;
performing multiplication operation, namely assigning the weight coefficients of all the positions learned in the previous step to the initial feature map, so that the model has the distinguishing capability on different position features;
sb5. the spatial attention mechanism can be described as:
M s (F)＝σ(f 7×7 ([AvgPool(F)；MaxPool(F)]))
in the formula, sigma represents a sigmoid activation function; f. of 7×7 Represents a 7 × 7 convolution operation; avgpool (f) and maxpool (f) denote global average pooling and global maximum pooling, respectively.
Furthermore, two-dimensional image data with a fixed time length are acquired by using a sliding window method, and sliding window operation on a time axis is performed on the steam-water separation reheating system state data, so that time sequence data are converted into time sequence-variable two-dimensional image data, and the system state data can be subjected to convolution operation.
Further, the edge zero padding includes adding a pixel point of which the edge of the time sequence-variable two-dimensional image data is 0, so that time sequence-variable two-dimensional image data with a fixed variable length is obtained, and is finally stored as two-dimensional image data with a resolution of 600 × 600, so that the data can be stored as input data with a uniform format, and fault diagnosis operation is facilitated.
Further, the normalization processing includes normalizing the two-dimensional image data obtained by the edge zero padding operation, so as to remove the influence of dimensions on different variables;
the normalization process comprises expanding the input image data according to variable dimensions, calculating the mean value mu and standard deviation sigma of different variables at all time points, subtracting the mean value from the original input data X, and dividing by the standard deviation to obtain the data X after normalization process * ；
In the formula, μ — mean of all sample data; σ — standard deviation of all sample data.
Has the advantages that: compared with the prior art, the invention has the advantages that:
1. the method can effectively carry out fault diagnosis on time and variable coupling input data containing noise interference, ensures higher accuracy, and can be applied to fault diagnosis of a nuclear power steam-water separation reheating system; the network structure is optimized by using the artificial bee colony heuristic intelligent algorithm, the network structure which meets the pareto optimal solution set with higher network fault diagnosis accuracy and fewer network calculation parameters can be obtained, the network performance can be exerted as much as possible, and the fault diagnosis accuracy as high as possible can be obtained under the condition of limited calculation resources.
2. According to the method, a channel attention mechanism and a space attention mechanism are added between the residual blocks, wherein the channel attention mechanism can accelerate the model to extract effective channel characteristics, and the space attention mechanism can accelerate the model to extract effective space characteristics of input image data.
3. The invention adopts the data preprocessing operation of sliding window, edge zero padding and standardized processing, can obtain the two-dimensional image input data with uniform data format, and can facilitate the conversion of the actual steam-water separation reheating system data into the input data of the model.
Drawings
FIG. 1 is an overall flow chart of the present invention;
FIG. 2 is a schematic flow chart of an artificial bee colony heuristic optimization algorithm in the present invention;
FIG. 3 is a schematic diagram of a channel attention mechanism of the present invention;
FIG. 4 is a schematic diagram of the spatial attention mechanism of the present invention;
FIG. 5 is a schematic view of a sliding window according to the present invention;
FIG. 6 is a schematic diagram of edge zero padding according to the present invention;
FIG. 7 is a diagram of the results of model fault diagnosis in the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be more clearly and completely described below with reference to the accompanying drawings in the examples of the present invention.
As shown in fig. 1, a flow chart is constructed for a fault diagnosis method provided herein based on an attention mechanism and a deep residual error network, referred to as a-DRSN for short, and a specific flow of algorithm design mainly includes the following steps:
1. network structure designed using artificial bee colony heuristic optimization algorithm
Taking the number of network residual blocks and the number of convolution kernels of each layer in the network as design variables;
taking the maximum fault diagnosis accuracy and the minimum network calculation parameter total amount after the network training for 20 periods as target variables;
and setting the training period to be 100 times, and then automatically optimizing by using an artificial bee colony algorithm to obtain a pareto optimal solution set of the fault diagnosis accuracy and the network calculation parameter total amount, wherein the pareto optimal solution set corresponds to the number of the optimal network residual block and the number of convolution kernels of each layer in the network, namely the optimal network structure.
2, adding a channel attention mechanism and a space attention mechanism between the two residual blocks, and connecting the channel attention mechanism and the space attention mechanism in series, wherein the channel attention mechanism obtains the most important convolution kernel parameters in different convolution kernels through three steps of compression, activation and multiplication, and accelerates the training speed and the model accuracy; the spatial attention mechanism obtains the most important pixel point parameters of the characteristics corresponding to the same convolution kernel in the way through four steps of pooling, dimensionality reduction, activation and convolution, so that the extraction of important spatial characteristics by the network is accelerated, the network training is accelerated, and the model accuracy is improved.
2.1 mechanism for adding attention
The channel attention mechanism is shown in fig. 3, the channel attention mechanism is divided into three steps of compression, activation and multiplication, and the compression operation can be described as follows:
wherein u is c (v, w) represents a pixel (v, w) in the feature map of the channel c; z is a radical of c Representing the c global feature point; the input feature map size is H × W × C.
The activation operation may be described as:
s c ＝σ(W 2 δ(W 1 z c ))
in the formula, sigma represents a sigmoid activation function; δ represents the Relu activation function; w 1 And W 2 Representing a weight of the fully connected layer; s c Representing the activated output value.
The product operation can be described as:
in the formula (I), the compound is shown in the specification,
representing the weighted feature map.
2.2 the spatial attention mechanism is shown in FIG. 4, and is divided into four steps of pooling, dimensionality reduction, activation and multiplication;
firstly, pooling operation is carried out, namely, global maximum pooling and global average pooling based on channels are carried out on the feature map with the input size H multiplied by W multiplied by C, so as to obtain two feature planes with the size H multiplied by W multiplied by 1, and then the two feature planes are subjected to channel splicing, so as to finally obtain the feature map with the size H multiplied by W multiplied by 2.
The second step is a dimensionality reduction operation, namely, the characteristic diagram with the size of H multiplied by W multiplied by 2 is reduced to the characteristic diagram with the size of H multiplied by W multiplied by 1 through a convolution kernel of 7 multiplied by 7.
The third step is activation operation, namely, the sigmoid activation function is used for generating weights of different positions, so that the model obtains the capability of distinguishing importance degrees of different positions.
The fourth step is a multiplication operation, namely, the weight coefficients of all the positions learned in the previous step are assigned to the initial feature map, so that the model has the distinguishing capability for different position features.
Finally, the spatial attention mechanism can be described as:
M s (F)＝σ(f 7×7 ([AvgPool(F)；MaxPool(F)]))
in the formula, sigma represents a sigmoid activation function; f. of 7×7 Represents a 7 × 7 convolution operation; avgpool (f) and maxpool (f) denote global average pooling and global maximum pooling, respectively.
3. Carrying out sliding window, edge zero filling and standardized processing operations to carry out preprocessing on the data;
3.1 using sliding window method to collect two-dimensional image data with fixed time length, the sliding window is shown in FIG. 5; and performing sliding window operation on the steam-water separation reheating system state data on a time axis, so that the time sequence data is converted into time sequence-variable two-dimensional image data, and the system state data can be subjected to convolution operation.
3.2 use the zero padding operation of the edge to change the input data size to a unified data format of 600 × 600, as shown in fig. 6; performing edge zero padding, namely padding the edge of the time sequence-variable two-dimensional image data with pixels of which the value is 0 to obtain comfortable time sequence-variable two-dimensional image data with fixed variable length, and finally storing the data as two-dimensional image data with the resolution of 600 multiplied by 600, so that the data can be stored as input data with uniform format to facilitate fault diagnosis operation
The data is standardized to eliminate dimension influence and accelerate network convergence, and the standardized operation is shown as follows:
in the formula, μ — mean value of all sample data; σ — standard deviation of all sample data.
4. Taking the preprocessed data as input training data of a deep residual network added with an attention mechanism, building a network model by using a Pythrch, constructing the network model after artificial bee colony optimization and the attention mechanism addition, inputting the preprocessed training data into the network model, and training parameters of the network model; and simultaneously observing the change of the network fault diagnosis accuracy and the loss value, preventing the overfitting phenomenon caused by excessive training times, and stopping the network training when the accuracy of the training set and the accuracy of the test set of the network meet the requirements
In the training process, network parameters are updated by using an Adam optimization algorithm, and a softmax classifier is used for outputting a fault diagnosis result.
5. And (3) inputting the actual state data of the steam-water separation reheating system serving as input data into a trained network model, and checking the performance of the algorithm.
The depth network fault diagnosis model based on the attention mechanism has high accuracy rate on faults of a steam-water separation reheating system, the accuracy rate reaches over 99% in normal noise-free data, and meanwhile, the method has good identification accuracy on data containing noise signals.
As can be seen from fig. 7, under the condition of low noise, the deep residual error network DRSN without adding an attention mechanism, the conventional convolutional neural network VGG, and the multilayer perceptron MLP all have good accuracy, but as the noise signal in the data increases, the accuracy of other methods is significantly reduced, and the method provided by the present invention has an accuracy of 90% or more, so that it can be concluded that the method provided by the present invention has very good diagnostic accuracy for high-noise data.
The above embodiments are merely illustrative of the technical concept and structural features of the present invention, and are intended to be implemented by those skilled in the art, but the present invention is not limited thereto, and any equivalent changes or modifications made according to the spirit of the present invention should fall within the scope of the present invention.
Claims (7)
1. A system fault diagnosis method based on an attention mechanism and a depth residual error network is characterized in that: comprises the following steps of (a) carrying out,
s1, optimizing the number of network residual blocks and the number of convolution kernels of each layer in the network by using an artificial bee colony heuristic optimization algorithm so as to obtain a pareto optimal network structure meeting the maximum fault diagnosis accuracy rate and the minimum network calculation parameters;
s2, adding a channel attention mechanism and a space attention mechanism between the residual blocks and in front and back, and improving the convergence speed of network training and the accuracy rate of fault diagnosis;
s3, carrying out sliding window, edge zero filling and standardized processing operations to carry out preprocessing on the data;
and S4, inputting the two-dimensional image data into a network for training to finally obtain a trained network model, and performing state classification on the steam-water separation reheating system state data input into the network model so as to diagnose the system state.
2. The system fault diagnosis method based on attention mechanism and depth residual error network according to claim 1, characterized in that: in step S1, the method for designing a network structure using an artificial bee colony heuristic optimization algorithm includes the following steps,
taking the number of network residual blocks and the number of convolution kernels of each layer in the network as design variables;
taking the maximum fault diagnosis accuracy and the minimum network calculation parameter total amount after the network training for 20 periods as target variables;
and setting the training period to be 100 times, and then automatically optimizing by using an artificial bee colony algorithm to obtain a pareto optimal solution set of the fault diagnosis accuracy and the network calculation parameter total amount, wherein the pareto optimal solution set corresponds to the number of the optimal network residual block and the number of convolution kernels of each layer in the network, namely the optimal network structure.
3. The system fault diagnosis method based on attention mechanism and depth residual error network according to claim 1 or 2, characterized in that: in the step S2, the channel attention mechanism includes the steps of,
sa1. compression, wherein the compression formula is as follows:
wherein u is c (v, w) represents a pixel (v, w) in the feature map of the channel c; z is a radical of c Representing the c global feature point; the input feature map size is H × W × C.
Sa2. activation, wherein the activation formula is as follows:
s c ＝σ(W 2 δ(W 1 z c ))
in the formula, sigma represents a sigmoid activation function; δ represents the Relu activation function; w 1 And W 2 Representing a weight of the fully connected layer; s c Represents the activated output value;
sa3. product, the product formula is:
in the formula (I), the compound is shown in the specification,
representing the weighted feature map.
4. The system fault diagnosis method based on attention mechanism and depth residual error network according to claim 3, characterized in that: in the step S2, the spatial attention mechanism includes the steps of,
performing pooling operation, namely performing channel-based global maximum pooling and global average pooling on the feature map with the input size of H multiplied by W multiplied by C to obtain two feature planes with the size of H multiplied by W multiplied by 1, and then performing channel splicing on the two feature planes to finally obtain the feature map with the size of H multiplied by W multiplied by 2;
dimension reduction operation, namely reducing the dimension of the feature map with the size of H multiplied by W multiplied by 2 into the feature map with the size of H multiplied by W multiplied by 1 through a convolution kernel of 7 multiplied by 7;
activating operation, namely using a sigmoid activation function to generate weights of different positions, so that the model obtains the capability of distinguishing importance degrees of different positions;
performing multiplication operation, namely assigning the weight coefficients of all the positions learned in the previous step to the initial feature map, so that the model has the distinguishing capability on different position features;
sb5. the spatial attention mechanism can be described as:
M s (F)＝σ(f 7×7 ([AvgPool(F)；MaxPool(F)]))
in the formula, sigma represents a sigmoid activation function; f. of 7×7 Represents a 7 × 7 convolution operation; avgpool (f) and maxpool (f) denote global average pooling and global maximum pooling, respectively.
5. The system fault diagnosis method based on attention mechanism and depth residual error network according to claim 4, characterized in that: the method comprises the steps of collecting two-dimensional image data with a fixed time length by using a sliding window method, and carrying out sliding window operation on state data of the steam-water separation reheating system on a time axis, so that time sequence data are converted into time sequence-variable two-dimensional image data, and the system state data can be subjected to convolution operation.
6. The system fault diagnosis method based on attention mechanism and depth residual error network according to claim 5, characterized in that: and the edge zero padding comprises the step of padding the edge of the time sequence-variable two-dimensional image data with a pixel point of 0 so as to obtain the time sequence-variable two-dimensional image data with a fixed variable length, and the time sequence-variable two-dimensional image data is finally stored as the two-dimensional image data with the resolution of 600 multiplied by 600, so that the data can be stored as input data with a uniform format, and fault diagnosis operation is facilitated.
7. The system fault diagnosis method based on attention mechanism and depth residual error network according to claim 6, characterized in that: the standardization processing comprises the step of carrying out standardization processing on two-dimensional image data obtained by the edge zero filling operation so as to remove the influence of dimensions on different variables;
the normalization process comprises expanding the input image data according to variable dimensions, calculating the mean value mu and standard deviation sigma of different variables at all time points, subtracting the mean value from the original input data X, and dividing by the standard deviation to obtain the data X after normalization process * ；
In the formula, μ — mean value of all sample data; σ — standard deviation of all sample data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210484425.9A CN114897138A (en) | 2022-05-06 | 2022-05-06 | System fault diagnosis method based on attention mechanism and depth residual error network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210484425.9A CN114897138A (en) | 2022-05-06 | 2022-05-06 | System fault diagnosis method based on attention mechanism and depth residual error network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114897138A true CN114897138A (en) | 2022-08-12 |
Family
ID=82719377
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210484425.9A Pending CN114897138A (en) | 2022-05-06 | 2022-05-06 | System fault diagnosis method based on attention mechanism and depth residual error network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114897138A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116417013A (en) * | 2023-06-09 | 2023-07-11 | 中国海洋大学 | Underwater propeller fault diagnosis method and system |
| CN117388716A (en) * | 2023-12-11 | 2024-01-12 | 四川长园工程勘察设计有限公司 | Battery pack fault diagnosis method, system and storage medium based on time sequence data |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112577747A (en) * | 2020-12-07 | 2021-03-30 | 东南大学 | Rolling bearing fault diagnosis method based on space pooling network |
| CN113255882A (en) * | 2021-04-30 | 2021-08-13 | 南通大学 | Bearing fault diagnosis method based on improved convolution capsule network |
| US20210319546A1 (en) * | 2020-04-10 | 2021-10-14 | Samsung Display Co., Ltd. | Image-based defects identification and semi-supervised localization |
| CN114239641A (en) * | 2021-11-16 | 2022-03-25 | 电子科技大学 | Fault diagnosis method for selecting kernel convolution residual error network by combined attention machine mechanism |
-
2022
- 2022-05-06 CN CN202210484425.9A patent/CN114897138A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210319546A1 (en) * | 2020-04-10 | 2021-10-14 | Samsung Display Co., Ltd. | Image-based defects identification and semi-supervised localization |
| CN112577747A (en) * | 2020-12-07 | 2021-03-30 | 东南大学 | Rolling bearing fault diagnosis method based on space pooling network |
| CN113255882A (en) * | 2021-04-30 | 2021-08-13 | 南通大学 | Bearing fault diagnosis method based on improved convolution capsule network |
| CN114239641A (en) * | 2021-11-16 | 2022-03-25 | 电子科技大学 | Fault diagnosis method for selecting kernel convolution residual error network by combined attention machine mechanism |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116417013A (en) * | 2023-06-09 | 2023-07-11 | 中国海洋大学 | Underwater propeller fault diagnosis method and system |
| CN116417013B (en) * | 2023-06-09 | 2023-08-25 | 中国海洋大学 | Underwater propeller fault diagnosis method and system |
| CN117388716A (en) * | 2023-12-11 | 2024-01-12 | 四川长园工程勘察设计有限公司 | Battery pack fault diagnosis method, system and storage medium based on time sequence data |
| CN117388716B (en) * | 2023-12-11 | 2024-02-13 | 四川长园工程勘察设计有限公司 | Battery pack fault diagnosis method, system and storage medium based on time sequence data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112149316B (en) | Aero-engine residual life prediction method based on improved CNN model | |
| CN112200244B (en) | Intelligent detection method for anomaly of aerospace engine based on hierarchical countermeasure training | |
| CN102707256B (en) | Fault diagnosis method based on BP-Ada Boost nerve network for electric energy meter | |
| CN113255848B (en) | Water turbine cavitation sound signal identification method based on big data learning | |
| CN114897138A (en) | System fault diagnosis method based on attention mechanism and depth residual error network | |
| CN112001306A (en) | Electroencephalogram signal decoding method for generating neural network based on deep convolution countermeasure | |
| CN111523081B (en) | Aeroengine fault diagnosis method based on enhanced gate control circulating neural network | |
| CN110046562A (en) | A kind of wind power system health monitor method and device | |
| CN111815806B (en) | Method for preprocessing flight parameter data based on wild value elimination and feature extraction | |
| CN110929765A (en) | Convolution self-coding fault monitoring method based on batch imaging | |
| CN110829417A (en) | Electric power system transient stability prediction method based on LSTM double-structure model | |
| CN105678343A (en) | Adaptive-weighted-group-sparse-representation-based diagnosis method for noise abnormity of hydroelectric generating set | |
| CN111275108A (en) | Method for performing sample expansion on partial discharge data based on generation countermeasure network | |
| CN117290800B (en) | Timing sequence anomaly detection method and system based on hypergraph attention network | |
| CN112952830B (en) | Method for predicting transient stability of power system after fault | |
| CN114048688A (en) | Method for predicting service life of bearing of wind power generator | |
| Li et al. | Intelligent fault diagnosis of aeroengine sensors using improved pattern gradient spectrum entropy | |
| CN114581699A (en) | Transformer state evaluation method based on deep learning model in consideration of multi-source information | |
| CN113935413A (en) | Distribution network wave recording file waveform identification method based on convolutional neural network | |
| CN113758709A (en) | Rolling bearing fault diagnosis method and system combining edge calculation and deep learning | |
| CN114295967A (en) | Analog circuit fault diagnosis method based on migration neural network | |
| CN113159088A (en) | Fault monitoring and diagnosis method based on multi-feature fusion and width learning | |
| CN116720057A (en) | River water quality prediction method and system based on feature screening and weight distribution | |
| CN113821888B (en) | Vibration data fault diagnosis method based on periodic impact feature extraction and echo state network | |
| CN113151842B (en) | Method and device for determining conversion efficiency of wind-solar complementary water electrolysis hydrogen production |
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 |