CN113378967A - Structural health monitoring multivariate data anomaly diagnosis method based on convolutional neural network and transfer learning - Google Patents
Structural health monitoring multivariate data anomaly diagnosis method based on convolutional neural network and transfer learning Download PDFInfo
- Publication number
- CN113378967A CN113378967A CN202110720189.1A CN202110720189A CN113378967A CN 113378967 A CN113378967 A CN 113378967A CN 202110720189 A CN202110720189 A CN 202110720189A CN 113378967 A CN113378967 A CN 113378967A
- Authority
- CN
- China
- Prior art keywords
- data
- neural network
- convolutional neural
- training
- model
- 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.)
- Granted
Links
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
-
- 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
- G06N3/02—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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic 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)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Testing And Monitoring For Control Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a structural health monitoring multivariate data anomaly diagnosis method based on a convolutional neural network and transfer learning, wherein multivariate monitoring data of a large-scale structure A is processed in a visualized manner by time series segmental data and is converted into a time domain response image, manual marking is carried out according to time domain response image data corresponding to a data segment, and various samples with various anomaly types and with manual marking are selected to form a data set A; inputting the data set A into a convolutional neural network model A for anomaly detection, and training the model A; visualizing the multivariate monitoring data of a certain large structure B, and manually marking to form a data set B; adding a data set B on the basis of the model A, performing transfer learning training, improving the generalization performance of the classification model, enabling the convolutional neural network model to adapt to data with different distributions, and taking the transfer learning trained model as a multi-element data anomaly detector; the invention can solve the problems that the prior structure health monitoring multivariate data has no detection method and the like.
Description
Technical Field
The invention belongs to the technical field of migration learning, convolutional neural networks and civil engineering structure health monitoring, and particularly relates to a structure health monitoring multivariate data abnormity diagnosis method based on convolutional neural networks and migration learning.
Background
At present, large-scale infrastructures such as large-span bridges, super high-rise buildings, large-span space structures and ocean platforms are increasing, and the infrastructures can be inevitably affected by disasters in a long-term use period, so that the reliability of the infrastructures is reduced. Even if the structure is not affected by disasters, the structure can inevitably cause material aging, structural deformation and self defects amplification due to long-time environmental erosion, load effect, fatigue effect and other comprehensive influences, so that certain damage is generated, the bearing capacity of the structure is reduced, even catastrophic accidents such as structural collapse are caused, and huge casualties and economic losses are caused. Structural Health Monitoring (SHM), as an effective means to ensure Structural safety, has been widely used in engineering practice. At the present time, a large-scale structural health monitoring system containing multiple types of sensors is installed in many major engineering structures, the number of single large structural sensors can reach thousands, and the total amount of generated data can reach hundreds of TB per year. The mass monitoring data records performance evolution information and response rules of the structure in the service process and response information of the structure under rare conditions such as earthquake, typhoon, fire, ship collision, traffic accidents and the like.
The structural health monitoring system works outdoors for a long time, due to factors such as sensor faults and data transmission faults, data obtained by the structural health monitoring system inevitably comprises various types of abnormal data which cannot completely represent the real condition of the structure and can seriously interfere with the automatic data analysis and early warning functions of the monitoring system, and in order to ensure the accuracy of a data analysis result, the abnormal data must be detected and cleaned. Due to the fact that the structural health monitoring data volume is huge (efficiency, precision and scale are considered, an artificial expert detection method is not applicable any more), abnormal data types are multiple (a traditional single-target and two-classification method is not applicable), abnormal data features have uncertainty (a method based on a modeling threshold value is difficult to apply), even if professionals participate in the abnormal data features, the situation of over-processing or under-processing is easy to occur, time and labor are wasted, and the requirement of real-time data analysis of a monitoring system cannot be met.
At present, detection work of abnormal data at home and abroad is mostly focused on detection of dynamic data such as acceleration and the like, and more various and complex data of other types of sensors (such as strain, temperature, humidity, inclination angle, displacement, GPS and the like) lack of effective abnormal detection methods. The difference between the monitoring data of the sensors of the various types and the vibration acceleration data is as follows, the sampling frequency of the vibration acceleration is generally 20HZ, the data quality is relatively good, the interval of the data sampling frequency of the sensors of the other types is 1HZ-20HZ, the data sampling frequency of a Global Positioning System (GPS) is generally 1HZ, the data sampling frequency of temperature, humidity, inclination angle, displacement and the like is different from 5HZ-10HZ, and the sampling frequency of the strain data is generally 20 HZ; the vibration acceleration data graph has obvious symmetry and has the characteristic that the average value is 0 under the general condition; other types of sensor data have random fluctuation trend, and data images are generally asymmetric and have no characteristic that the average value is 0; the related research of the vibration acceleration data is more, the abnormal detection work research is more, the related research of other types of sensor data is relatively lacked, and various types of data have no corresponding abnormal data detection research.
Disclosure of Invention
In order to solve the problems, the invention provides a structural health monitoring multivariate data abnormity diagnosis method based on a convolutional neural network and transfer learning, which can realize the automatic processing of the whole process of abnormity pattern learning, model training, transfer learning and structural health monitoring multivariate data abnormity diagnosis.
The invention is realized by the following scheme:
a structural health monitoring multivariate data anomaly diagnosis method based on a convolutional neural network and transfer learning comprises the following steps:
the method comprises the following steps:
the method comprises the following steps: converting the multivariate monitoring data of the structure A into a time domain response image through time series data visualization processing, performing artificial labeling according to time domain response image data corresponding to the same data segment, selecting various abnormal type samples with artificial labeling, and forming a training set D { A }T,L}；
Step two: will train set D { ATL is input into A convolutional neural network model CNN-A for anomaly detection;
step three: a small amount of multivariate monitoring data of the structure B are visualized, and the step I is repeated to form a small training set D { B }T,L}；
Step four: adding A training set D { B) on the basis of A convolutional neural network model CNN-ATAnd L, performing migration learning training, improving the generalization performance of the classification model, enabling the convolutional neural network model to adapt to data with different distributions, and performing structural health monitoring and data anomaly detection by using the trained model as an anomaly data detector.
Further, in the first step,
step one, the multivariate data to be detected is segmented into N data segments d according to the time interval TiThe length of the data segment depends on the sampling frequency fs of the data and the length of the data segment diLength diDrawing a data image p corresponding to each segment of data in time domain to form a data set D { A };
step two, randomly extracting N from the data set D { A }TEach picture sample forms a training set D { AT}；
Step one and three, in training set D { ATIn the method, according to the characteristics of each image p, a label L is markedP；
Step one, step four, repeat step one and step three, until training set D { ATIn (C) } NTEach sample is labeled to generate a training set D { A }T,L}。
Further, in the second step, the first step,
the convolutional neural network sequentially comprises an input layer, a convolutional layer, a pooling layer, a batch standardization layer, a freezing layer, a full connection layer, a Softmax classification layer and an output layer;
the training ratio of the convolutional neural network model CNN-A is 0.85, the initial learning rate alphA is 0.0005, the learning rates of freezing layers are all 0, batch processing is adopted during training, the batch processing size BatchSize is 128, A cross entropy function E (W) is selected as an objective function, and the formulA is as follows:
wherein, P is the total number of samples;
i (-) is the total number of samples, and the function of I (-) is defined as I (true) 1, I (false) 0;
Lpa label for sample p;
The global maximum iteration number is 8, random gradient descent with momentum is selected, the momentum parameter beta is 0.8, and shuffle is started when training is started.
Further, in the fourth step,
the global maximum iteration times are 4, random gradient descent with momentum is selected, and the momentum factor beta is 0.8;
the transfer learning training mode is as follows: freezing the L1-L5 layer of the convolutional neural network model CNN-A, retraining the L6 full link layer, the L7softmax classification layer and the L8 output layer only by using the datA of the structure B so as to improve the generalization performance of the network, and carrying out anomaly detection on the multivariate datA of A, B two structures by using the trained classification model.
Further, in step two, the training set D { A) obtained in step oneTAnd L is rich, the convolutional neural network model CNN-A directly carries out structural health monitoring and multivariate datA anomaly detection without carrying out the third step and the fourth step.
Further, A pre-training model AlexNet model is used for replacing the convolutional neural network model CNN-A, and transfer learning is carried out on the AlexNet model;
step A: visualizing a small amount of multivariate data of the structure A, and forming a small data set D { A } according to a step one modeT,L}；
And B: adding a training set to perform transfer learning training on the basis of the AlexNet model, so that the generalization performance of the classification model is improved, the convolutional neural network model can adapt to data with different distributions, and the trained model is used as an abnormal data detector to perform data abnormality detection;
the training rate is 0.85, the initial learning rate is 0.0005, the learning rates of the freezing layers are all 0, batch processing is adopted during training, the batch processing size is 128, a cross entropy function is selected as an objective function, the global maximum iteration number is 4, and random gradient descent with momentum is selected;
and retraining the last three layers of the AlexNet model, namely a full link layer, a classification layer and an output layer only by using the data of the structure C so as to improve the generalization performance of the network, wherein the trained classification model can be used for carrying out multi-element data anomaly detection.
Further, migration learning is carried out by adopting a mixed data set,
and (3) forming a training set D { S) containing A, B various multivariate data of two structures according to the method of the step one by carrying out data visualization and manual labeling on the multivariate monitoring data of the structure A and the structure B from time sequence dataTL, and will D { S }TL is divided into D { S } with a large data size1L and D { S ] with a small data amount2,L}；
Step S1, set D { S1L, inputting the parameters into a convolutional neural network model A for anomaly detection, training the convolutional neural network model, and setting the parameters to be the same as the parameters in the step two;
step S2, set D { S2And L, inputting the input signal into A convolutional neural network model CNN-A for anomaly detection, and performing migration training, wherein the parameter setting is the same as that in the step four.
The invention has the beneficial effects
(1) According to the invention, the abnormal detection problem of the structural health monitoring multivariate data is converted into the image classification problem, so that the automatic processing of the whole process of abnormal pattern learning, model training, transfer learning and structural health monitoring multivariate data abnormal diagnosis is realized, and the whole process is intelligent and accurate;
(2) the method can diagnose the structural health monitoring multivariate data with various abnormal modes at the same time, and solves the problem of multi-classification of the structural health monitoring multivariate data;
(3) the invention has high automation degree, except in the data marking process, the whole abnormal data diagnosis process is automated;
(4) the invention has high processing speed and can meet the real-time data preprocessing requirement of the online early warning of the structural health monitoring;
(5) the overall accuracy of the diagnosis result of one embodiment of the invention can reach 96%;
(6) the transfer learning model of the invention can be trained and completed only by applying a small number of data sets.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a schematic diagram of structural health monitoring vibration acceleration data differentiated from other multivariate data;
FIG. 3(a) is an architecture diagram of a convolutional neural network of the convolutional neural network-based multivariate data anomaly detection method of the present invention; (b) the image is input when automatic detection is carried out;
FIG. 4(a) is a flowchart of a multivariate data anomaly detection method based on transfer learning according to the present invention; (b) the invention is A transfer learning architecture diagram of the multivariate datA anomaly detection method based on the CNN-A model transfer learning; (c) the framework diagram of the transfer learning of the multivariate data anomaly detection method based on the AlexNet model transfer learning;
FIG. 5 is A diagram illustrating the automatic detection of multivariate datA anomaly based on CNN-A transfer learning according to an embodiment of the present invention; wherein (a) is a test result, (b) a manual detection result, and (c) is an automatic detection result of transfer learning.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments; 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.
With reference to fig. 1 to 5; data for structure A, B was obtained, and data for structure C was obtained in the same manner as A, B.
A structural health monitoring multivariate data anomaly diagnosis method based on a convolutional neural network and transfer learning comprises the following steps:
the method comprises the following steps: in combination with the figures 3 and 4 of the drawings,
the first embodiment is as follows:
the method comprises the following steps: converting the multivariate monitoring data of a large structure A into a time domain response image through time series data visualization processing, performing artificial labeling according to the time domain response image data corresponding to the same data segment, selecting various abnormal type samples with artificial labeling, and forming a training set D { A }T,L}；
Step two: will train set D { ATL is input into A convolutional neural network model CNN-A for anomaly detection;
step three: visualizing a small amount of multivariate monitoring data of a large structure B, repeating the step one to form a small training set D { B }T,L}；
Step four: adding A training set D { B) on the basis of A convolutional neural network model CNN-ATAnd L, performing migration learning training, improving the generalization performance of the classification model, enabling the convolutional neural network model to adapt to data with different distributions, and performing structural health monitoring and data anomaly detection by using the trained model as an anomaly data detector.
In the first step, the first step is carried out,
step one by one, segmenting multivariate data to be detected into N data segments d according to time interval T (in the example, T is one hour and 3600s)iThe length of the data segment depends on the sampling frequency fs of the data and the length of the data segment diLength diFs × T20 × 3600 ═ 72000, data images p corresponding to the time domain of each segment of data are respectively rendered, and a data set D { a } is formed;
step two, randomly extracting N from the data set D { A }TEach picture sample forms a training set D { AT}；
Step one and three, in training set D { ATIn the method, according to the characteristics of each image p, (the image size in the example is 100 multiplied by 1, only red R channel) is marked with a label LP(ii) a In this example, the normal mode is labeled "1", the deletion mode is labeled "2", the mutation mode is labeled "4", and so on;
step one, step four, repeat step one and step three, until training set D { ATIn (C) } NTEach sample is labeled to generate a training set D { A }T,L}。
In the second step, the first step is carried out,
the convolutional neural network sequentially comprises an input layer, a convolutional layer, a pooling layer, a batch standardization layer, a freezing layer, a full connection layer, a Softmax classification layer and an output layer; the specific network parameters are shown in table 1;
TABLE 1 convolutional neural network architecture Table
The training ratio of the convolutional neural network model CNN-A is 0.85, the initial learning rate alphA is 0.0005, the learning rates of freezing layers are all 0, batch processing is adopted during training, the batch processing size BatchSize is 128, A cross entropy function E (W) is selected as an objective function, and the formulA is as follows:
wherein, P is the total number of samples;
i (-) is the total number of samples, and the function of I (-) is defined as I (true) 1, I (false) 0;
Lpa label for sample p;
The global maximum iteration number is 8, random gradient descent with momentum is selected, the momentum parameter beta is 0.8, and shuffle is started when training is started.
In the fourth step of the method, the first step of the method,
the global maximum iteration times are 4, random gradient descent with momentum is selected, and the momentum factor beta is 0.8;
the transfer learning training mode is as follows: freezing the L1-L5 layer of the convolutional neural network model CNN-A, retraining the L6 full link layer, the L7softmax classification layer and the L8 output layer only by using the datA of the structure B so as to improve the generalization performance of the network, and carrying out anomaly detection on the multivariate datA of A, B two structures by using the trained classification model.
In step two, the training set D { A) obtained in step oneTAnd L is rich, the convolutional neural network model CNN-A directly carries out structural health monitoring and multivariate datA anomaly detection without carrying out the third step and the fourth step.
The second embodiment is as follows:
using A pre-training model AlexNet (or other) model to replace the convolutional neural network model CNN-A, and performing transfer learning on the AlexNet model;
step A: visualizing a small amount of multivariate data of the structure A, and forming a small data set D { A } according to a step one modeT,L}；
And B: adding a training set to perform transfer learning training on the basis of the AlexNet model, so that the generalization performance of the classification model is improved, the convolutional neural network model can adapt to data with different distributions, and the trained model is used as an abnormal data detector to perform data abnormality detection;
the training rate is 0.85, the initial learning rate is 0.0005, the learning rates of the freezing layers are all 0, batch processing is adopted during training, the batch processing size is 128, a cross entropy function is selected as an objective function, the global maximum iteration number is 4, and random gradient descent with momentum is selected;
and retraining the last three layers of the AlexNet model, namely a full link layer, a classification layer and an output layer only by using the data of the structure C so as to improve the generalization performance of the network, wherein the trained classification model can be used for carrying out multi-element data anomaly detection.
The third concrete implementation mode:
performing migration learning by adopting a mixed data set, performing data visualization and manual labeling on multivariate monitoring data of a structure A and a structure B from time series data, and forming a training set D { S } containing A, B two-structure various multivariate data according to the method of the step oneTL, and will D { S }TL is divided into D { S } with a large data size1L and D { S ] with a small data amount2,L}；
Step S1, set D { S1L, inputting the parameters into a convolutional neural network model A for anomaly detection, training the convolutional neural network model, and setting the parameters to be the same as the parameters in the step two;
step S2, set D { S2And L, inputting the input signal into A convolutional neural network model CNN-A for anomaly detection, and performing migration training, wherein the parameter setting is the same as that in the step four.
An embodiment of the specific implementation method comprises the following steps:
the embodiment is to diagnose abnormal data of multivariate monitoring data such as strain, displacement, temperature, humidity, GPS and the like of a certain real large-span bridge on the basis of the first specific embodiment. This application demonstrates the feasibility and utility of the present invention. The following specific examples illustrate the effects of the present invention.
Fig. 5 shows the multi-datA abnormal datA detection results of strain, displacement, temperature, humidity and GPS datA (22 channels in total) in 2018 of A certain large-span bridge health monitoring system, where each color represents A different abnormal pattern, "1-normal" represents A normal pattern, "2-missing" represents A missing pattern, "3-outler" represents an outlier pattern, "4-mutation" represents A mutation pattern, "5-minor" represents A sub-minimum pattern, "6-tend" represents A trend pattern, "7-square" represents an overrange pattern, "8-constant" represents A constant pattern, fig. 5(A) represents A datA set classification situation, fig. 5(b) is A result of an artificial mark, and fig. 5(c) is A migration learning model prediction result based on A model CNN-A.
The structural health monitoring multivariate data anomaly diagnosis method based on the convolutional neural network and the transfer learning, which is provided by the invention, is introduced in detail, the principle and the implementation mode of the invention are explained, and the description of the embodiment is only used for helping to understand the method and the core idea of the method; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.
Claims (7)
1. A structural health monitoring multivariate data anomaly diagnosis method based on a convolutional neural network and transfer learning is characterized by comprising the following steps:
the method comprises the following steps:
the method comprises the following steps: converting the multivariate monitoring data of the structure A into a time domain response image through time series data visualization processing, performing artificial labeling according to time domain response image data corresponding to the same data segment, selecting various abnormal type samples with artificial labeling, and forming a training set D { A }T,L}；
Step two: will train set D { ATL is input into A convolutional neural network model CNN-A for anomaly detection;
step three: a small amount of multivariate monitoring data of the structure B are visualized, and the step I is repeated to form a small training set D { B }T,L}；
Step four: adding A training set D { B) on the basis of A convolutional neural network model CNN-ATAnd L, performing migration learning training to enable the convolutional neural network model to adapt to data with different distributions, and using the trained model as an abnormal data detector to perform structural health monitoring and data abnormality detection.
2. The method of claim 1, further comprising: in the first step, the first step is carried out,
step one, the multivariate data to be detected is segmented into N data segments d according to the time interval TiThe length of the data segment depends on the sampling frequency fs of the data and the length of the data segment diLength diDrawing a data image p corresponding to each segment of data in time domain to form a data set D { A };
step two, randomly extracting N from the data set D { A }TEach picture sample forms a training set D { AT}；
Step one and three, in training set D { ATIn the method, according to the characteristics of each image p, a label L is markedP；
Step one, step four, repeat step one and step three, until training set D { ATIn (C) } NTEach sample is labeled to generate a training set D { A }T,L}。
3. The method of claim 2, further comprising: in the second step, the first step is carried out,
the convolutional neural network sequentially comprises an input layer, a convolutional layer, a pooling layer, a batch standardization layer, a freezing layer, a full connection layer, a Softmax classification layer and an output layer;
the training ratio of the convolutional neural network model CNN-A is 0.85, the initial learning rate alphA is 0.0005, the learning rates of freezing layers are all 0, batch processing is adopted during training, the batch processing size BatchSize is 128, A cross entropy function E (W) is selected as an objective function, and the formulA is as follows:
wherein, P is the total number of samples;
i (-) is the total number of samples, and the function of I (-) is defined as I (true) 1, I (false) 0;
Lpa label for sample p;
the global maximum iteration number is 8, random gradient descent with momentum is selected, the momentum parameter beta is 0.8, and shuffle is started when training is started.
4. The method of claim 3, further comprising: in the fourth step of the method, the first step of the method,
the global maximum iteration times are 4, random gradient descent with momentum is selected, and the momentum factor beta is 0.8;
the transfer learning training mode is as follows: freezing the L1-L5 layer of the convolutional neural network model CNN-A, retraining the L6 full link layer, the L7softmax classification layer and the L8 output layer only by using the datA of the structure B so as to improve the generalization performance of the network, and carrying out anomaly detection on the multivariate datA of A, B two structures by using the trained classification model.
5. The method of claim 3, further comprising: in step two, the training set D { A) obtained in step oneTAnd L is rich, the convolutional neural network model CNN-A directly carries out structural health monitoring and multivariate datA anomaly detection without carrying out the third step and the fourth step.
6. The method of claim 1, further comprising: replacing the convolutional neural network model CNN-A with A pre-training model AlexNet model, and performing transfer learning on the AlexNet model;
step A: visualizing a small amount of multivariate data of the structure A, and forming a small data set D { A } according to a step one modeT,L}；
And B: adding a training set to perform transfer learning training on the basis of the AlexNet model, so that the convolutional neural network model can adapt to data with different distributions, and using the trained model as an abnormal data detector to perform data abnormality detection;
the training rate is 0.85, the initial learning rate is 0.0005, the learning rates of the freezing layers are all 0, batch processing is adopted during training, the batch processing size is 128, a cross entropy function is selected as an objective function, the global maximum iteration number is 4, and random gradient descent with momentum is selected;
and retraining the last three layers of the AlexNet model, namely a full link layer, a classification layer and an output layer only by using the data of the structure C so as to improve the generalization performance of the network, wherein the trained classification model can be used for carrying out multi-element data anomaly detection.
7. The method of claim 1, further comprising: the mixed data set is adopted for the migration learning,
and (3) forming a training set D { S) containing A, B various multivariate data of two structures according to the method of the step one by carrying out data visualization and manual labeling on the multivariate monitoring data of the structure A and the structure B from time sequence dataTL, and will D { S }TL is divided into D { S } with a large data size1L and D { S ] with a small data amount2,L}；
Step S1, set D { S1L, inputting the parameters into a convolutional neural network model A for anomaly detection, training the convolutional neural network model, and setting the parameters to be the same as the parameters in the step two;
step S2, set D { S2And L, inputting the input signal into A convolutional neural network model CNN-A for anomaly detection, and performing migration training, wherein the parameter setting is the same as that in the step four.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110720189.1A CN113378967B (en) | 2021-06-28 | 2021-06-28 | Structural health monitoring multivariate data anomaly diagnosis method based on convolutional neural network and transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110720189.1A CN113378967B (en) | 2021-06-28 | 2021-06-28 | Structural health monitoring multivariate data anomaly diagnosis method based on convolutional neural network and transfer learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113378967A true CN113378967A (en) | 2021-09-10 |
| CN113378967B CN113378967B (en) | 2022-11-08 |
Family
ID=77579436
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110720189.1A Active CN113378967B (en) | 2021-06-28 | 2021-06-28 | Structural health monitoring multivariate data anomaly diagnosis method based on convolutional neural network and transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113378967B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113866455A (en) * | 2021-09-30 | 2021-12-31 | 中铁桥隧技术有限公司 | Bridge acceleration monitoring data anomaly detection method, system and device based on deep learning |
| CN113988202A (en) * | 2021-11-04 | 2022-01-28 | 季华实验室 | Mechanical arm abnormal vibration detection method based on deep learning |
| CN117218602A (en) * | 2023-10-20 | 2023-12-12 | 青岛理工大学 | Structural health monitoring data abnormality diagnosis method and system |
| CN117723782A (en) * | 2024-02-07 | 2024-03-19 | 山东大学 | Sensor fault identification positioning method and system for bridge structure health monitoring |
Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180357542A1 (en) * | 2018-06-08 | 2018-12-13 | University Of Electronic Science And Technology Of China | 1D-CNN-Based Distributed Optical Fiber Sensing Signal Feature Learning and Classification Method |
| CN109190567A (en) * | 2018-09-10 | 2019-01-11 | 哈尔滨理工大学 | Abnormal cervical cells automatic testing method based on depth convolutional neural networks |
| CN109710636A (en) * | 2018-11-13 | 2019-05-03 | 广东工业大学 | A kind of unsupervised industrial system method for detecting abnormality based on depth migration study |
| CN109919135A (en) * | 2019-03-27 | 2019-06-21 | 华瑞新智科技（北京）有限公司 | Behavioral value method, apparatus based on deep learning |
| CN110220709A (en) * | 2019-06-06 | 2019-09-10 | 北京科技大学 | Fault Diagnosis of Roller Bearings based on CNN model and transfer learning |
| CN110458210A (en) * | 2019-07-25 | 2019-11-15 | 武汉理工大学 | A kind of driver's state monitoring method and system based on transfer learning |
| US10565471B1 (en) * | 2019-03-07 | 2020-02-18 | Capital One Services, Llc | Systems and methods for transfer learning of neural networks |
| CN111291096A (en) * | 2020-03-03 | 2020-06-16 | 腾讯科技（深圳）有限公司 | Data set construction method and device, storage medium and abnormal index detection method |
| CN111582298A (en) * | 2020-03-18 | 2020-08-25 | 宁波送变电建设有限公司永耀科技分公司 | Sensing abnormal data real-time detection method based on artificial intelligence |
| WO2021011931A1 (en) * | 2019-07-17 | 2021-01-21 | The Regents Of The University Of California | Biologically inspired sleep algorithm for artificial neural networks |
| AU2020103613A4 (en) * | 2020-11-23 | 2021-02-04 | Agricultural Information and Rural Economic Research Institute of Sichuan Academy of Agricultural Sciences | Cnn and transfer learning based disease intelligent identification method and system |
| CN112353402A (en) * | 2020-10-22 | 2021-02-12 | 平安科技（深圳）有限公司 | Training method of electrocardiosignal classification model, electrocardiosignal classification method and device |
| CN112381060A (en) * | 2020-12-04 | 2021-02-19 | 哈尔滨工业大学 | Building earthquake damage level classification method based on deep learning |
| CN112734692A (en) * | 2020-12-17 | 2021-04-30 | 安徽继远软件有限公司 | Transformer equipment defect identification method and device |
| CN112884060A (en) * | 2021-03-09 | 2021-06-01 | 联仁健康医疗大数据科技股份有限公司 | Image annotation method and device, electronic equipment and storage medium |
| CN112906644A (en) * | 2021-03-22 | 2021-06-04 | 重庆大学 | Mechanical fault intelligent diagnosis method based on deep migration learning |
-
2021
- 2021-06-28 CN CN202110720189.1A patent/CN113378967B/en active Active
Patent Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180357542A1 (en) * | 2018-06-08 | 2018-12-13 | University Of Electronic Science And Technology Of China | 1D-CNN-Based Distributed Optical Fiber Sensing Signal Feature Learning and Classification Method |
| CN109190567A (en) * | 2018-09-10 | 2019-01-11 | 哈尔滨理工大学 | Abnormal cervical cells automatic testing method based on depth convolutional neural networks |
| CN109710636A (en) * | 2018-11-13 | 2019-05-03 | 广东工业大学 | A kind of unsupervised industrial system method for detecting abnormality based on depth migration study |
| US10565471B1 (en) * | 2019-03-07 | 2020-02-18 | Capital One Services, Llc | Systems and methods for transfer learning of neural networks |
| CN109919135A (en) * | 2019-03-27 | 2019-06-21 | 华瑞新智科技（北京）有限公司 | Behavioral value method, apparatus based on deep learning |
| CN110220709A (en) * | 2019-06-06 | 2019-09-10 | 北京科技大学 | Fault Diagnosis of Roller Bearings based on CNN model and transfer learning |
| WO2021011931A1 (en) * | 2019-07-17 | 2021-01-21 | The Regents Of The University Of California | Biologically inspired sleep algorithm for artificial neural networks |
| CN110458210A (en) * | 2019-07-25 | 2019-11-15 | 武汉理工大学 | A kind of driver's state monitoring method and system based on transfer learning |
| CN111291096A (en) * | 2020-03-03 | 2020-06-16 | 腾讯科技（深圳）有限公司 | Data set construction method and device, storage medium and abnormal index detection method |
| CN111582298A (en) * | 2020-03-18 | 2020-08-25 | 宁波送变电建设有限公司永耀科技分公司 | Sensing abnormal data real-time detection method based on artificial intelligence |
| CN112353402A (en) * | 2020-10-22 | 2021-02-12 | 平安科技（深圳）有限公司 | Training method of electrocardiosignal classification model, electrocardiosignal classification method and device |
| AU2020103613A4 (en) * | 2020-11-23 | 2021-02-04 | Agricultural Information and Rural Economic Research Institute of Sichuan Academy of Agricultural Sciences | Cnn and transfer learning based disease intelligent identification method and system |
| CN112381060A (en) * | 2020-12-04 | 2021-02-19 | 哈尔滨工业大学 | Building earthquake damage level classification method based on deep learning |
| CN112734692A (en) * | 2020-12-17 | 2021-04-30 | 安徽继远软件有限公司 | Transformer equipment defect identification method and device |
| CN112884060A (en) * | 2021-03-09 | 2021-06-01 | 联仁健康医疗大数据科技股份有限公司 | Image annotation method and device, electronic equipment and storage medium |
| 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 (3)
| Title |
|---|
| HOO-CHANG SHIN等: "Deep Convolutional Neural Networks for Computer-Aided Detection: CNN Architectures, Dataset Characteristics and Transfer Learning", 《IEEE TRANSACTIONS ON MEDICAL IMAGING》 * |
| ZHONGLING HUANG等: "Transfer Learning with Deep Convolutional Neural Network for SAR Target Classification with Limited Labeled Data", 《REMOTE SENS》 * |
| 钟昌源等: "基于分组注意力模块的实时农作物病害叶片语义分割模型", 《农业工程学报》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113866455A (en) * | 2021-09-30 | 2021-12-31 | 中铁桥隧技术有限公司 | Bridge acceleration monitoring data anomaly detection method, system and device based on deep learning |
| CN113988202A (en) * | 2021-11-04 | 2022-01-28 | 季华实验室 | Mechanical arm abnormal vibration detection method based on deep learning |
| CN117218602A (en) * | 2023-10-20 | 2023-12-12 | 青岛理工大学 | Structural health monitoring data abnormality diagnosis method and system |
| CN117723782A (en) * | 2024-02-07 | 2024-03-19 | 山东大学 | Sensor fault identification positioning method and system for bridge structure health monitoring |
| CN117723782B (en) * | 2024-02-07 | 2024-05-03 | 山东大学 | Sensor fault identification positioning method and system for bridge structure health monitoring |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113378967B (en) | 2022-11-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113378967B (en) | Structural health monitoring multivariate data anomaly diagnosis method based on convolutional neural network and transfer learning | |
| CN108764601B (en) | Structural health monitoring abnormal data diagnosis method based on computer vision and deep learning technology | |
| CN110108456A (en) | A kind of rotating machinery health evaluating method of depth convolutional neural networks | |
| CN109272123B (en) | Sucker-rod pump working condition early warning method based on convolution-circulation neural network | |
| CN108225906B (en) | Inhaul cable corrosion monitoring and identifying and fatigue life evaluating method based on computer vision | |
| CN108596507B (en) | Disaster damage early warning method for mariculture net cage | |
| CN111625988A (en) | Bridge health management analysis and prediction system and method based on deep learning | |
| CN110276385B (en) | Similarity-based mechanical part residual service life prediction method | |
| CN115951014A (en) | CNN-LSTM-BP multi-mode air pollutant prediction method combining meteorological features | |
| CN111595541A (en) | Multi-dimensional structure damage identification method based on mass vibration transmissibility data convolutional neural network processing | |
| WO2021114320A1 (en) | Wastewater treatment process fault monitoring method using oica-rnn fusion model | |
| CN115950609B (en) | Bridge deflection anomaly detection method combining correlation analysis and neural network | |
| CN115267945A (en) | Thunder and lightning early warning method and system based on graph neural network | |
| CN114139590A (en) | Method for estimating ocean temperature | |
| CN115290057B (en) | Channel software row state monitoring system and method | |
| CN116960962A (en) | Mid-long term area load prediction method for cross-area data fusion | |
| CN116663126A (en) | Bridge temperature effect prediction method based on channel attention BiLSTM model | |
| CN110727986A (en) | BIM and weighted gray cloud clustering model-based ancient building maintenance model acquisition method | |
| CN116484219A (en) | Water supply network water quality abnormal pollution source identification method based on gate control graph neural network | |
| CN110570022A (en) | mGBDT-based building energy consumption prediction method and system | |
| CN114547868A (en) | Cable-stayed bridge performance online evaluation prediction method based on BIM | |
| CN112069621A (en) | Method for predicting residual service life of rolling bearing based on linear reliability index | |
| CN112036403A (en) | Intelligent detection method for missing of bolt pin of power transmission tower based on attention mechanism | |
| CN115688555B (en) | Bridge life prediction method | |
| CN117094236B (en) | High-precision calibration method for deep water drilling gas invasion data analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |