CN113778811A - Method and system for fault monitoring of software system based on deep convolution transfer learning - Google Patents

Method and system for fault monitoring of software system based on deep convolution transfer learning Download PDF

Info

Publication number
CN113778811A
CN113778811A CN202111157772.2A CN202111157772A CN113778811A CN 113778811 A CN113778811 A CN 113778811A CN 202111157772 A CN202111157772 A CN 202111157772A CN 113778811 A CN113778811 A CN 113778811A
Authority
CN
China
Prior art keywords
data set
network
software system
target domain
fault
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
Application number
CN202111157772.2A
Other languages
Chinese (zh)
Inventor
吴勇
廖明霞
董一英
沈谷峰
杨婷婷
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chongqing University of Post and Telecommunications
Original Assignee
Chongqing University of Post and Telecommunications
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Chongqing University of Post and Telecommunications filed Critical Chongqing University of Post and Telecommunications
Priority to CN202111157772.2A priority Critical patent/CN113778811A/en
Publication of CN113778811A publication Critical patent/CN113778811A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/3003—Monitoring arrangements specially adapted to the computing system or computing system component being monitored
    • G06F11/302—Monitoring arrangements specially adapted to the computing system or computing system component being monitored where the computing system component is a software system
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
    • G06F11/3452—Performance evaluation by statistical analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
    • G06F11/3466—Performance evaluation by tracing or monitoring
    • G06F11/3476—Data logging
    • 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)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Quality & Reliability (AREA)
  • Probability & Statistics with Applications (AREA)
  • Computer Hardware Design (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Test And Diagnosis Of Digital Computers (AREA)

Abstract

本发明涉及一种基于深度卷积迁移学习软件系统故障监测方法及系统，属于计算机软件测试领域，包括以下步骤：收集已有负载S下的软件系统负载数据集，构建源域数据集；对每一组原始响应时间都进行点数分割，构建源域样本数据集；构建目标域数据集，并对目标域数据集中的每组原始响应时间进行点数分割，构建目标域样本数据集；将源域样本数据集和目标域样本数据集利用深度卷积迁移学习，实现对软件系统进行故障监测。本发明可以在面对多负载下故障样本较少或某故障样本缺失等情况发生时，仍旧可以获得较为理想的故障监测效果，且在新负载下的数据集，不需要重新训练网络模型，可以节约大量时间。

Figure 202111157772

The invention relates to a software system fault monitoring method and system based on deep convolution transfer learning, belonging to the field of computer software testing, comprising the following steps: collecting a software system load data set under an existing load S, constructing a source domain data set; A set of original response times are divided by points to construct the source domain sample data set; the target domain data set is constructed, and each group of original response times in the target domain data set is divided by points to construct the target domain sample data set; the source domain sample data set is constructed; The dataset and target domain sample dataset utilize deep convolution transfer learning to implement fault monitoring for software systems. The present invention can still obtain a relatively ideal fault monitoring effect when there are few fault samples under multiple loads or a certain fault sample is missing, and the data set under the new load does not need to retrain the network model, and can Save a lot of time.

Figure 202111157772

Description

Fault monitoring method and system based on deep convolution migration learning software system
Technical Field
The invention belongs to the field of computer software testing, and relates to a method and a system for monitoring system faults based on deep convolution transfer learning software.
Background
With the increasing size and complexity of computer software, the quality of the computer software is difficult to be effectively controlled and guaranteed. In the software system, when the load is applied to the adjacent boundary in the running process of a large number of users, the software system can have faults of different degrees. How to effectively extract and utilize the existing response time information to quickly and accurately identify and predict software faults is a key problem in the field of software fault monitoring at present.
The software system cannot respond or stop running, which causes poor user experience for users, and may cause great surface influence on company image, even may cause great damage. Therefore, by monitoring, monitoring and downtime prediction of the software system, the equipment is expanded or distributed for maintenance when the equipment fails or is about to fail, and the method has important significance for improving the reliability and the economy of the software system.
In the process of monitoring the actual load, the software system is usually operated under the condition of different loads, the response time is short, and the fault state of the software system is less. Therefore, the fault state data monitored and collected by the software system in the actual load has the characteristics of multiple loads, fewer fault state samples and even the defect of a certain fault state sample. When the traditional diagnosis method is faced with fault data samples under different loads, a network model needs to be re-established when the loads change, and the process of the traditional diagnosis method takes a lot of time. Moreover, most of the traditional diagnosis methods rely on a large amount of fault label data, and when the conditions that the fault state samples are insufficient or the fault state samples are missing occur, the generalization capability of the traditional network model is poor, and the fault monitoring effect is not ideal.
Disclosure of Invention
In view of the above, an object of the present invention is to provide a method and a system for monitoring a fault of a software system based on deep convolution migration learning, which can still obtain an ideal fault monitoring effect when the conditions of fewer fault samples under multiple loads or missing a certain fault sample occur, and can save a lot of time without retraining a network model for a data set under a new load. The transfer learning is a learning method for solving problems in different but related fields by using existing knowledge, and the method realizes field knowledge sharing by transferring the knowledge obtained by learning in a source field into a target field, thereby solving the problem of poor performance of a training model caused by few learning samples and unbalanced sample distribution in the target field. Compared with methods such as incremental learning, multi-task learning and self-learning, the migration learning emphasizes the correlation between learning tasks and utilizes the correlation to complete the migration between knowledge. The concept of Deep learning originates from the field of artificial intelligence machine learning, and a Deep Neural Network (DNN) model composed of multiple hidden layers is a remarkable characteristic of the Deep learning model. Compared with a shallow neural network model, the DNN can combine bottom layer features to form more abstract high-level feature representation, so that implicit feature expression of data is found, and features of information are effectively extracted and represented through layer-by-layer conversion of data features. Transfer Learning (Transfer Learning) is a machine Learning method, which transfers knowledge in one field (i.e., a source field) to another field (i.e., a target field) to enable the target field to obtain a better Learning effect.
In order to achieve the purpose, the invention provides the following technical scheme:
on one hand, the invention provides a software system fault monitoring method based on deep convolution transfer learning, which comprises the following steps:
s1: collecting a software system load data set under the existing load S, and constructing a source domain sample data set;
s2: point division is carried out on each group of original response time, and a source domain data set is constructed;
s3: constructing a target domain sample data set, and performing point segmentation on each group of original response time in the target domain data set to construct a target domain data set;
s4: and carrying out fault monitoring on the software system by using the source domain data set and the target domain data set through deep convolution transfer learning.
Further, in step S1, the software system load sample data set under the existing load S is classified into w states according to the fault type, and the original response time under each fault type
Figure BDA0003284895330000021
Where w represents the data class, w is 1, 2, 3 … n, x0～xnRepresented as the 1 st to n +1 th group fault signals in the w fault state.
Further, in the step S2, the source domain data set construction method includes the following steps:
s21: setting a window sliding step length s and a window length l according to the number N of data points, and generating a sample number t; sample di＝{X0，X1，X2，...XL1, 2, 3, ·, t; obtaining a source domain data set M from a samples＝{d1，，，d2，，d3，…dt，}；
S22: setting a source domain test set in a source domain data set
Figure BDA0003284895330000022
And source domain training set
Figure BDA0003284895330000023
R, the source domain training set
Figure BDA0003284895330000024
Sample number a ═ t · r, source domain test set
Figure BDA0003284895330000025
The sample number b is t (1-r).
Further, in step S3, the machine response time of the software system under different loads in the four states of normal operation state, data abnormality, local user abnormality, and downtime is collected
Figure BDA0003284895330000026
Constructing a target domain sample data set according to the machine response time
Figure BDA0003284895330000027
Wherein, w' is 1, 2, 3, 4, which respectively represents four states of normal operation state, program data abnormity, local error and downtime.
Further, setting window sliding step length and window length, and constructing a target domain data set MT(ii) a Setting the proportion of the test set and the training set, and constructing a target domain data training set
Figure BDA0003284895330000028
And target domain data test set
Figure BDA0003284895330000029
Further, in step S4, the fault monitoring includes the following steps:
s41: training source domain data
Figure BDA00032848953300000210
Inputting a set of one-dimensional depth convolution neural network I to pre-train and initialize network parameters, and testing the set through a source domain
Figure BDA0003284895330000031
Testing the network effect, if the testing effect is ideal, pre-training to finish determining parameters and finishing training the network, otherwise, continuously adjusting the network to perform back propagation and continuously updating the parameters until the network achieves the ideal effect on the test set to finish training;
s42: targeting domain dataset M using convolutional neural network hierarchyTPerforming transfer learning, freezing the global mean pooling layer L in the feature extraction module and the feature classification module of the one-dimensional deep convolutional neural network IGAnd L in the full connection layerFAdding a new Softmax layer for the network model I to adapt to the target domain data set
Figure BDA0003284895330000032
Completing network level adjustment and constructing new network I2；
S43: to network I2Fine tuning is performed by locking feature classification modules D, and L1，L2，L3Weight parameter of layer, unfreezing L4Layer parameters, obtaining network I after fine adjustment3；
S44: acquiring original fault signals of the software system in real time and transmitting the signals to a network I3And obtaining a fault monitoring result of the current software system.
Further, the one-dimensional depth convolution neural network I model construction method in step S41 includes the following steps:
(1) construction of a convolution pooling layer Lj：
Lj＝{Cj，Pj，Bj}
In the formula, Cj、Pj、BjThe convolution layer, the pooling layer and the normalization layer are respectively used for feature extraction; j is the number of the convolution pooling module;
(2) stacking 4 convolution pooling layers to construct a feature extraction module S', S ═ L1，L2，L3，L4}；
(3) Adding a characteristic classification module D, D ═ LG，LF，LsoftmaxThe feature classification module comprises a global mean pooling layer LGAll-connected layer LFSoftmax layer
Figure BDA0003284895330000033
And completing the network construction.
Further, in the step S43, the target domain training data set is used
Figure BDA0003284895330000034
To network I3Training is carried out to enable the network to extract deep abstract features from the target domain data set
Figure BDA0003284895330000035
Via the full connection layer LFSoftmax layer
Figure BDA0003284895330000036
And outputting the fault probability distribution of each fault type of the target domain, wherein the maximum probability of the fault probability distribution corresponds to the fault type and serves as a diagnosis result.
On the other hand, the invention provides a load diagnosis system based on a deep convolution migration learning software system, which comprises a source domain sample data set construction module, a source domain data set construction module, a target domain data set construction module and a fault monitoring module;
the source domain sample data set construction module collects a software system load data set under the existing load S and constructs a source domain sample data set;
the source domain data set construction module performs point number segmentation on each group of original response time to construct a source domain data set;
the target domain data set construction module constructs a target domain sample data set, and performs point segmentation on each group of original response time in the target domain data set to construct a target domain sample data set;
and the fault monitoring module carries out fault monitoring on the software system by using the deep convolution transfer learning on the source domain data set and the target domain data set.
Further, when the fault monitoring module detects a fault, the method includes: training set of source domain data
Figure BDA0003284895330000041
Inputting one-dimensional deep convolution neural network I to pre-train and initialize network parameters, and passing through a source domain test set
Figure BDA0003284895330000042
Testing the network effect of the network, if the test effect is ideal, pre-training to complete the determination of parameters and complete the training of the network, otherwise, continuously adjusting the network to perform back propagation and continuously updating the parameters until the network achieves the ideal effect on the test set to complete the training; targeting domain dataset M using convolutional neural network hierarchyTPerforming transfer learning, freezing the global mean pooling layer L in the feature extraction module S' and the feature classification module of the one-dimensional deep convolutional neural network IGAnd weighting parameters of LF in the full connection layer, and adding a new Softmax layer for adapting the network model I to the target domain data set
Figure BDA0003284895330000043
New network I is constructed by adjusting transmission completion network level2(ii) a Acquiring original fault signals of the software system in real time and transmitting the signals to a network I3And obtaining a fault monitoring result of the current software system.
The invention has the beneficial effects that: compared with the traditional fault monitoring method, the deep winding machine migration learning method provided by the invention still has higher fault monitoring precision when a few sample data sets or missing sample data sets are faced. 2. Compared with the traditional fault monitoring method, the deep convolution transfer learning method provided by the invention utilizes the convolutional neural network hierarchical structure transfer learning, and can save a large amount of time when new load training is faced.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objectives and other advantages of the invention may be realized and attained by the means of the instrumentalities and combinations particularly pointed out hereinafter.
Drawings
For the purposes of promoting a better understanding of the objects, aspects and advantages of the invention, reference will now be made to the following detailed description taken in conjunction with the accompanying drawings in which:
FIG. 1 is a schematic overall flow chart of a fault monitoring method based on a deep convolution migration learning software system according to the present invention;
FIG. 2 is a schematic diagram of a one-dimensional convolutional network structure according to the present invention;
FIG. 3 is a schematic diagram of the network migration learning of the present invention;
fig. 4 is a schematic diagram of network migration learning fine tuning according to the present invention.
Detailed Description
The embodiments of the present invention are described below with reference to specific embodiments, and other advantages and effects of the present invention will be easily understood by those skilled in the art from the disclosure of the present specification. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention. It should be noted that the drawings provided in the following embodiments are only for illustrating the basic idea of the present invention in a schematic way, and the features in the following embodiments and examples may be combined with each other without conflict.
Wherein the showings are for the purpose of illustrating the invention only and not for the purpose of limiting the same, and in which there is shown by way of illustration only and not in the drawings in which there is no intention to limit the invention thereto; to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product; it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The same or similar reference numerals in the drawings of the embodiments of the present invention correspond to the same or similar components; in the description of the present invention, it should be understood that if there is an orientation or positional relationship indicated by terms such as "upper", "lower", "left", "right", "front", "rear", etc., based on the orientation or positional relationship shown in the drawings, it is only for convenience of description and simplification of description, but it is not an indication or suggestion that the referred device or element must have a specific orientation, be constructed in a specific orientation, and be operated, and therefore, the terms describing the positional relationship in the drawings are only used for illustrative purposes, and are not to be construed as limiting the present invention, and the specific meaning of the terms may be understood by those skilled in the art according to specific situations.
As shown in fig. 1, the present invention provides a method for monitoring system faults based on deep convolution transfer learning software, which includes the following steps:
s1, collecting the software system load data set under the existing load S, and constructing the source domain sample data set
Figure BDA0003284895330000051
Figure BDA0003284895330000052
The software system load sample data set under the existing load S is classified into w states according to fault types, and the original response time under each fault type
Figure BDA0003284895330000053
Where w represents the data class, w is 1, 2, 3 … n, xNRepresented as the nth set of fault signals in the w fault state.
S2, performing point number segmentation on each group of original response time to construct a source domain data set;
to be provided with
Figure BDA0003284895330000054
One set of signals x in (1)1For example, for x1Point segmentation is carried out to construct a source domain sample data set, and the specific steps are as follows:
and S21, setting a window sliding step length S and a window length l according to the number N of the data points, and generating a sample with the number t. Sample di＝{X0，X1，X2，...XL1, 2, 3, ·, t; obtaining a source domain data set M from a samples＝{d1，，，d2，，d3，…dt，}；
S22, setting a source domain test set in a source domain data set
Figure BDA0003284895330000055
And source domain training set
Figure BDA0003284895330000056
R, the source domain training set
Figure BDA0003284895330000057
Sample number a ═ t · r, source domain test set
Figure BDA0003284895330000058
Sample number b ═ t · (1-r); in the present embodiment, the ratio r is preferably 0.3.
S3, constructing a target domain sample data set
Figure BDA0003284895330000059
Dividing the number of points of each group of original response time in the target domain data set to construct a target domain data set;
collecting original response time of software system under different loads in four states of normal operation state, abnormal program data, local error and downtime
Figure BDA00032848953300000510
Constructing a target domain sample data set according to the response time
Figure BDA00032848953300000511
Figure BDA00032848953300000512
W is 1, 2, 3 and 4, which respectively represent a normal running state, tooth surface abrasion, planet gear tooth breakage and rolling element bearing loss;
setting window sliding step size and window length according to step S21, and constructing target domain data set MTSetting the ratio of the test set to the training set according to the step S22, and constructing a training set of target domain data
Figure BDA0003284895330000061
And target domain data test set
Figure BDA0003284895330000062
S4, carrying out fault monitoring on the software system by using the deep convolution transfer learning of the source domain data set and the target domain data set, which comprises the following specific steps:
s41, training set of source domain data
Figure BDA0003284895330000063
Inputting one-dimensional deep convolution neural network I to pre-train and initialize network parameters, and passing through a source domain test set
Figure BDA0003284895330000064
And testing the network effect of the network, if the test effect is ideal, pre-training to finish determining parameters and finishing training the network, otherwise, continuously adjusting the network to perform back propagation and continuously updating the parameters until the network achieves the ideal effect on the test set to finish training. The initialization of the internal parameters of the network comprises the steps of setting learning rate, activating function, weighting parameters, extracting characteristics and the like.
As shown in fig. 2, the method for constructing the one-dimensional depth convolution neural network I model includes:
(1) construction of a convolution pooling layer Lj：
Lj＝{Cj，Pj，Bj}
In the formula, Cj、Pj、BjThe convolution layer, the pooling layer and the normalization layer are respectively used for feature extraction; j is the convolution pooling module number.
(2) Superposing 4 convolution pooling layers to construct a feature extraction module S, S ═ L1，L2，L3，L4}。
(3) Adding a characteristic classification module D, D ═ LG，LF，LsoftmaxThe feature classification module comprises a global mean pooling layer LGAll-connected layer LFSoftmax layer
Figure BDA0003284895330000065
And completing the network construction.
The source domain data training set passes through C of each convolution pooling layer in the feature extraction modulej、Pj、BjConvolution kernel operation, pooling operation, normalization operation output characteristics of
Figure BDA0003284895330000066
Superposition of 4 convolutional pooling layers S ═ L1，L2，L3，L4Get the final characteristics
Figure BDA0003284895330000067
Final characteristics
Figure BDA0003284895330000068
Outputting the characteristic value y after passing through the global mean pooling layerfg(ii) a Full connection layer pair yfgPerforming characteristic combination and Dropodt operation to output characteristic value ytAnd is combined with ytAnd (4) inputting the probability distribution of each fault type of the source domain into a Softmax classifier, and taking the maximum probability of the probability distribution corresponding to the fault type as a diagnosis result.
S42, utilizing the convolutional neural network hierarchy structure to carry out the data set M of the target domainTPerforming transfer learning, freezing a feature extraction module S of the network model I and a global mean pooling layer L in the feature classification moduleGAnd L in the full connection layerFAdding a new Softmax layer for the network model I to adapt to the target domain data set
Figure BDA0003284895330000069
Completing network level adjustment and constructing new network I2As shown in fig. 3.
Training set using target domain data
Figure BDA00032848953300000610
For new network 12Training and updating Softmax layer
Figure BDA00032848953300000611
Pass the target domain test set
Figure BDA00032848953300000612
And testing the network, finishing the transfer learning if the testing effect is ideal, and otherwise, continuing to perform network iteration and performing back propagation until the network achieves the ideal effect on the testing set.
S43, network I2Fine tuning is performed by locking feature classification modules D, and L1，L2，L3Weight parameter of layer, unfreezing L4Layer parameters, obtaining network I after fine adjustment3As shown in fig. 4.
Training a data set using a target domain
Figure BDA0003284895330000071
To network I3Training is carried out to enable the network to extract deep abstract features from the target domain data set
Figure BDA0003284895330000072
Via the full connection layer LFSoftmax layer
Figure BDA0003284895330000073
And outputting the fault probability distribution of each fault type of the target domain, wherein the maximum probability of the fault probability distribution corresponds to the fault type and serves as a diagnosis result.
S44, real-time obtaining step S3SoftwareThe original fault signal of the system is transmitted to the network I in step S433And obtaining a fault monitoring result of the current software system.
The invention also provides a system load diagnosis system based on the deep convolution transfer learning software, which comprises: the system comprises a source domain sample data set construction module, a source domain data set construction module, a target domain data set construction module and a fault monitoring module;
a source domain data set construction module collects a software system load data set under the existing load S and constructs a source domain sample data set;
the source domain data set construction module performs point number segmentation on each group of original response time to construct a source domain data set;
the target domain data set construction module constructs a target domain sample data set, and performs point segmentation on each group of original response time in the target domain sample data set to construct a target domain data set;
and the fault monitoring module carries out fault monitoring on the software system by using the deep convolution transfer learning on the source domain data set and the target domain data set.
In the above embodiment, in the fault monitoring module, the fault monitoring includes the following steps:
training set of source domain data
Figure BDA0003284895330000074
Inputting one-dimensional deep convolution neural network I to pre-train and initialize network parameters, and passing through a source domain test set
Figure BDA0003284895330000075
Testing the network effect of the network, if the test effect is ideal, pre-training to complete the determination of parameters and complete the training of the network, otherwise, continuously adjusting the network to perform back propagation and continuously updating the parameters until the network achieves the ideal effect on the test set to complete the training;
targeting domain dataset M using convolutional neural network hierarchyTPerforming transfer learning, freezing a feature extraction module S of the network model I and a global mean pooling layer L in the feature classification moduleGAnd L in the full connection layerFAdding a new Softmax layer for the network model I to adapt to the target domain data set
Figure BDA0003284895330000076
Completing network level adjustment and constructing new network I2；
To network I2Fine tuning is performed by locking feature classification modules D, and L1，L2，L3Weight parameter of layer, unfreezing L4Layer parameters, obtaining network I after fine adjustment3；
Acquiring original fault signals of the software system in real time and transmitting the signals to a network I3And obtaining a fault monitoring result of the current software system.
In conclusion, the invention constructs the one-dimensional deep convolutional neural network, performs the migration learning by utilizing the hierarchical structure of the one-dimensional convolutional neural network, and provides the software system fault monitoring method based on the deep convolutional migration learning. The invention uses the existing source domain data set to pre-train the one-dimensional convolutional neural network, and uses the hierarchical structure of the one-dimensional convolutional neural network to complete the transfer learning of the target domain data set.
Finally, the above embodiments are only intended to illustrate the technical solutions of the present invention and not to limit the present invention, and although the present invention has been described in detail with reference to the preferred embodiments, it will be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions, and all of them should be covered by the claims of the present invention.

Claims (10)

1.一种基于深度卷积迁移学习软件系统故障监测方法，其特征在于：包括以下步骤：1. a software system fault monitoring method based on deep convolution migration learning, is characterized in that: comprise the following steps: S1：收集已有负载S下的软件系统负载数据集，构建源域样本数据集；S1: Collect the software system load data set under the existing load S, and construct the source domain sample data set; S2：对每一组原始响应时间都进行点数分割，构建源域数据集；S2: Divide each group of original response times by points to construct a source domain data set; S3：构建目标域样本数据集，并对目标域数据集中的每组原始响应时间进行点数分割，构建目标域数据集；S3: Construct the target domain sample data set, and divide each group of original response times in the target domain data set by points to construct the target domain data set; S4：将源域数据集和目标域数据集利用深度卷积迁移学习，实现对软件系统进行故障监测。S4: The source domain dataset and the target domain dataset are used for deep convolution transfer learning to realize fault monitoring of the software system. 2.根据权利要求1所述的基于深度卷积迁移学习软件系统故障监测方法，其特征在于：在步骤S1中，将已有负载S下的软件系统负载样本数据集，按照故障类型分类为w个状态，每种故障类型下的原始响应时间
Figure FDA0003284895320000011
其中w代表数据类别，w＝1，2，3…n，x0～xn表示为在w故障状态下的第1～n+1组故障信号。2. The software system fault monitoring method based on deep convolution transfer learning according to claim 1, wherein in step S1, the software system load sample data set under the existing load S is classified as w according to the fault type states, raw response time under each fault type
Figure FDA0003284895320000011
Where w represents the data type, w=1, 2, 3...n, x 0 ～x n are represented as the 1st ～n+1 groups of fault signals in the w fault state. 3.根据权利要求1所述的基于深度卷积迁移学习软件系统故障监测方法，其特征在于：在所述步骤S2中，所述源域数据集构建方法包括以下步骤：3. The software system fault monitoring method based on deep convolution transfer learning according to claim 1, wherein in the step S2, the source domain data set construction method comprises the following steps: S21：根据数据点数N设定窗口滑动步长s、窗口长度l，生成样本个数为t；样本di＝{X0，X1，X2，...XL}，i＝1，2，3，...，t；根据样本得到源域数据集Ms＝{d1，，，d2，，d3，...dt，}；S21: Set the window sliding step size s and the window length l according to the number of data points N , and the number of generated samples is t ; 2, 3,..., t; obtain the source domain data set M s ={d 1 ,,,d 2 ,,d 3 ,...d t ,} according to the sample; S22：设定源域数据集中源域测试集
Figure FDA0003284895320000012
与源域训练集
Figure FDA0003284895320000013
的比例为r，则源域训练集
Figure FDA0003284895320000014
样本数a＝t·r，源域测试集
Figure FDA0003284895320000015
样本数b＝t·(1-r)。S22: Set the source domain test set in the source domain dataset
Figure FDA0003284895320000012
with the source domain training set
Figure FDA0003284895320000013
The ratio of r is r, then the source domain training set
Figure FDA0003284895320000014
Number of samples a = t r, source domain test set
Figure FDA0003284895320000015
Number of samples b=t·(1-r). 4.根据权利要求1所述的基于深度卷积迁移学习软件系统故障监测方法，其特征在于：所述步骤S3中，采集不同负载下的软件系统在正常运行状态、数据异常、局部用户异常及宕机四种状态下的机器响应时间
Figure FDA0003284895320000016
根据所述机器响应时间构建目标域样本数据集
Figure FDA0003284895320000017
其中，w′＝1，2，3，4，分别代表正常运行状态、程序数据异常、局部错误及宕机四种状态。4. The method for monitoring software system faults based on deep convolution transfer learning according to claim 1, characterized in that: in the step S3, the software systems under different loads are collected in normal operation state, data abnormality, local user abnormality and Machine response time in four states of downtime
Figure FDA0003284895320000016
Build a target domain sample dataset based on the machine response time
Figure FDA0003284895320000017
Among them, w'=1, 2, 3, and 4, respectively representing four states of normal running state, abnormal program data, partial error and downtime. 5.根据权利要求4所述的基于深度卷积迁移学习软件系统故障监测方法，其特征在于：设定窗口滑动步长和窗口长度，构建目标域数据集MT；设定测试集与训练集比例，构建目标域数据训练集
Figure FDA0003284895320000018
和目标域数据测试集
Figure FDA0003284895320000019
5. the software system fault monitoring method based on deep convolution transfer learning according to claim 4, is characterized in that: setting window sliding step size and window length, build target domain data set MT ; Set test set and training set Proportion, construct the target domain data training set
Figure FDA0003284895320000018
and the target domain data test set
Figure FDA0003284895320000019
 6.根据权利要求1所述的基于深度卷积迁移学习软件系统故障监测方法，其特征在于：所述步骤S4中，故障监测包括以下步骤：6. The fault monitoring method based on deep convolution transfer learning software system according to claim 1, is characterized in that: in described step S4, fault monitoring comprises the following steps: S41：将源域数据训练
Figure FDA00032848953200000110
集输入一维深度卷积神经网络I进行预训练对网络参数初始化，通过源域测试集
Figure FDA00032848953200000111
测试网络效果，若测试效果理想则预训练完成确定参数，并完成训练网络，反之继续调整网络进行反向传播不断更新参数直到网络在测试集上达到理想效果完成训练；S41: Train the source domain data
Figure FDA00032848953200000110
Set input one-dimensional deep convolutional neural network I for pre-training to initialize network parameters, and pass the source domain test set
Figure FDA00032848953200000111
Test the network effect. If the test effect is ideal, the pre-training is completed to determine the parameters, and the training network is completed. Otherwise, continue to adjust the network for back-propagation and continuously update the parameters until the network achieves the ideal effect on the test set to complete the training; S42：利用卷积神经网络层级结构对目标域数据集MT进行迁移学习，冻结所述一维深度卷积神经网络I的特征提取模块、特征分类模块中的全球均值池化层LG和全连接层中LF的权重参数，为网络模型I适应目标域数据集添加新的Softmax层
Figure FDA0003284895320000021
完成网络层级调整构建新网络I2；S42: Use the convolutional neural network hierarchical structure to perform migration learning on the target domain data set MT , and freeze the global mean pooling layer LG and the global mean pooling layer LG in the feature extraction module and the feature classification module of the one-dimensional deep convolutional neural network I. The weight parameter of LF in the connection layer, adding a new Softmax layer for the adaptation of the network model I to the target domain dataset
Figure FDA0003284895320000021
Complete the network level adjustment to construct a new network I 2 ; S43：对网络I2进行微调，通过锁定特征分类模块D，以及L1，L2，L3层的权重参数，解冻L4层的参数，得到微调后的网络I3；S43: Fine-tune the network I 2 , and obtain the fine-tuned network I 3 by locking the feature classification module D, and the weight parameters of the L 1 , L 2 , and L 3 layers, and unfreezing the parameters of the L 4 layer; S44：实时获取软件系统原始故障信号，传输至网络I3中，得到当前软件系统的故障监测结果。S44: Acquire the original fault signal of the software system in real time, transmit it to the network I3 , and obtain the fault monitoring result of the current software system. 7.根据权利要求6所述的基于深度卷积迁移学习软件系统故障监测方法，其特征在于：步骤S41中所述一维深度卷积神经网络I模型构建方法包括以下步骤：7. the software system fault monitoring method based on deep convolution transfer learning according to claim 6, is characterized in that: the one-dimensional deep convolutional neural network I model construction method described in step S41 comprises the following steps: (1)构建卷积池化层Lj：(1) Construct the convolution pooling layer L j : Lj＝{Cj，Pj，Bj}L j ={C j , P j , B j } 式中，Cj、Pj、Bj分别为卷积层、池化层、归一化层，用于特征提取；j为卷积池化模块编号；In the formula, C j , P j , and B j are the convolution layer, pooling layer, and normalization layer, which are used for feature extraction; j is the number of the convolution pooling module; (2)叠加4个卷积池化层，构建特征提取模块S′，S′＝{L1，L2，L3，L4}；(2) Stacking four convolution pooling layers to construct a feature extraction module S', S'={L 1 , L 2 , L 3 , L 4 }; (3)添加特征分类模块D，D＝{LG，LF，Lsoftmax}，特征分类模块包括全球均值池化层LG、全连接层LF、Softmax层
Figure FDA0003284895320000022
完成网络构建。(3) Add a feature classification module D, D={L G , L F , L softmax }, the feature classification module includes a global mean pooling layer LG , a fully connected layer LF , and a Softmax layer
Figure FDA0003284895320000022
Complete the network construction. 8.根据权利要求6所述的基于深度卷积迁移学习软件系统故障监测方法，其特征在于：所述步骤S43中，使用目标域训练数据集
Figure FDA0003284895320000023
对网络I3进行训练，使得网络对目标域数据集提取深层抽象特征
Figure FDA0003284895320000024
经由全连接层LF、Softmax层
Figure FDA0003284895320000025
输出目标域各个故障类型的故障概率分布，其最大概率对应故障类别作为诊断结果。8. The fault monitoring method for a software system based on deep convolution transfer learning according to claim 6, wherein in the step S43, a target domain training data set is used
Figure FDA0003284895320000023
Train the network I3 so that the network extracts deep abstract features from the target domain dataset
Figure FDA0003284895320000024
Via the fully connected layer LF and the Softmax layer
Figure FDA0003284895320000025
The fault probability distribution of each fault type in the target domain is output, and the maximum probability corresponds to the fault type as the diagnosis result. 9.一种基于深度卷积迁移学习软件系统负载诊断系统，其特征在于：包括源域样本数据集构建模块、源域数据集构建模块、目标域数据集构建模块和故障监测模块；9. A software system load diagnosis system based on deep convolution transfer learning, characterized in that: it comprises a source domain sample data set building module, a source domain data set building module, a target domain data set building module and a fault monitoring module; 所述源域样本数据集构建模块收集已有负载S下的软件系统负载数据集，构建源域样本数据集；The source domain sample data set construction module collects the software system load data set under the existing load S, and constructs the source domain sample data set; 所述源域数据集构建模块对每一组原始响应时间都进行点数分割，构建源域数据集；The source domain data set building module divides each group of original response times by points to construct a source domain data set; 所述目标域数据集构建模块构建目标域样本数据集，并对目标域数据集中的每组原始响应时间进行点数分割，构建目标域样本数据集；The target domain data set building module constructs a target domain sample data set, and divides each group of original response times in the target domain data set by points to construct a target domain sample data set; 所述故障监测模块将源域数据集和目标域数据集利用深度卷积迁移学习，实现对软件系统进行故障监测。The fault monitoring module utilizes the deep convolution transfer learning of the source domain data set and the target domain data set to realize fault monitoring of the software system. 10.根据权利要求9所述的基于深度卷积迁移学习软件系统负载诊断系统，其特征在于：所述故障监测模块进行故障检测时，包括：将源域数据训练集
Figure FDA0003284895320000031
输入一维深度卷积神经网络I进行预训练对网络参数初始化，通过源域测试集
Figure FDA0003284895320000032
对网络进行测试其网络效果，若测试效果理想则预训练完成确定参数，并完成训练网络，反之继续调整网络进行反向传播不断更新参数直到网络在测试集上达到理想效果完成训练；利用卷积神经网络层级结构对目标域数据集MT进行迁移学习，冻结所述一维深度卷积神经网络I的特征提取模块S′、特征分类模块中的全球均值池化层LG和全连接层中LF的权重参数，为网络模型I适应目标域数据集添加新的Softmax层
Figure FDA0003284895320000033
输完成网络层级调整构建新网络I2；实时获取软件系统原始故障信号，传输至网络I3中，得到当前软件系统的故障监测结果。10. The system load diagnosis system based on deep convolution transfer learning software system according to claim 9, characterized in that: when the fault monitoring module performs fault detection, it comprises: the source domain data training set
Figure FDA0003284895320000031
Input the one-dimensional deep convolutional neural network I for pre-training to initialize the network parameters, and pass the source domain test set
Figure FDA0003284895320000032
Test the network effect of the network. If the test effect is ideal, the pre-training is completed to determine the parameters, and the training network is completed. Otherwise, continue to adjust the network for back-propagation and continuously update the parameters until the network achieves the ideal effect on the test set to complete the training; use convolution The neural network hierarchical structure performs migration learning on the target domain data set MT, and freezes the feature extraction module S′ of the one-dimensional deep convolutional neural network I, the global mean pooling layer LG in the feature classification module, and the fully connected layer L. The weight parameter of F , adding a new Softmax layer to adapt the network model I to the target domain dataset
Figure FDA0003284895320000033
After completing the network level adjustment and constructing a new network I 2 ; obtain the original fault signal of the software system in real time, transmit it to the network I 3 , and obtain the fault monitoring result of the current software system.
CN202111157772.2A 2021-09-28 2021-09-28 Method and system for fault monitoring of software system based on deep convolution transfer learning Pending CN113778811A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111157772.2A CN113778811A (en) 2021-09-28 2021-09-28 Method and system for fault monitoring of software system based on deep convolution transfer learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111157772.2A CN113778811A (en) 2021-09-28 2021-09-28 Method and system for fault monitoring of software system based on deep convolution transfer learning

Publications (1)

Publication Number Publication Date
CN113778811A true CN113778811A (en) 2021-12-10

Family

ID=78854917

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111157772.2A Pending CN113778811A (en) 2021-09-28 2021-09-28 Method and system for fault monitoring of software system based on deep convolution transfer learning

Country Status (1)

Country Link
CN (1) CN113778811A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114239733A (en) * 2021-12-21 2022-03-25 华中科技大学 Machine tool response modeling method and system based on transfer learning and response prediction method
CN114782362A (en) * 2022-04-20 2022-07-22 广州东焊智能装备有限公司 Intelligent welding detection system based on deep migration learning
CN115017970A (en) * 2022-04-24 2022-09-06 西安建筑科技大学 A method and system for abnormal gas use behavior detection based on transfer learning

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2185908A1 (en) * 1994-04-12 1995-10-19 Brent Gregory Architecture and methods for a hardware description language source level analysis and debugging system
US20040117392A1 (en) * 2002-12-16 2004-06-17 Hermann Burgmeier Value mapping
CN110751186A (en) * 2019-09-26 2020-02-04 北京航空航天大学 Cross-project software defect prediction method based on supervised expression learning
US20200150622A1 (en) * 2018-11-13 2020-05-14 Guangdong University Of Technology Method for detecting abnormity in unsupervised industrial system based on deep transfer learning
CN112163640A (en) * 2020-10-30 2021-01-01 北京信息科技大学 Planetary gearbox fault diagnosis method and system based on deep convolution migration learning

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2185908A1 (en) * 1994-04-12 1995-10-19 Brent Gregory Architecture and methods for a hardware description language source level analysis and debugging system
US20040117392A1 (en) * 2002-12-16 2004-06-17 Hermann Burgmeier Value mapping
US20200150622A1 (en) * 2018-11-13 2020-05-14 Guangdong University Of Technology Method for detecting abnormity in unsupervised industrial system based on deep transfer learning
CN110751186A (en) * 2019-09-26 2020-02-04 北京航空航天大学 Cross-project software defect prediction method based on supervised expression learning
CN112163640A (en) * 2020-10-30 2021-01-01 北京信息科技大学 Planetary gearbox fault diagnosis method and system based on deep convolution migration learning

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
宫丽娜等: "软件缺陷预测技术研究进展", 《软件学报》 *
电视中心工艺技术系统建设丛书编委会: "《电视中心工艺技术系统建设测量与测试》", 31 December 2017 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114239733A (en) * 2021-12-21 2022-03-25 华中科技大学 Machine tool response modeling method and system based on transfer learning and response prediction method
CN114782362A (en) * 2022-04-20 2022-07-22 广州东焊智能装备有限公司 Intelligent welding detection system based on deep migration learning
CN114782362B (en) * 2022-04-20 2023-07-18 广州东焊智能装备有限公司 Welding intelligent detection system based on deep migration learning
CN115017970A (en) * 2022-04-24 2022-09-06 西安建筑科技大学 A method and system for abnormal gas use behavior detection based on transfer learning

Similar Documents

Publication Publication Date Title
WO2024021246A1 (en) Cross-device incremental bearing fault diagnosis method based on continuous learning
CN112163640A (en) Planetary gearbox fault diagnosis method and system based on deep convolution migration learning
Tang et al. A novel adaptive convolutional neural network for fault diagnosis of hydraulic piston pump with acoustic images
He et al. Improved deep transfer auto-encoder for fault diagnosis of gearbox under variable working conditions with small training samples
CN111709448B (en) A Mechanical Fault Diagnosis Method Based on Transition Relational Network
CN108197648B (en) Hydroelectric generating set fault diagnosis method and system based on LSTM deep learning model
CN113778811A (en) Method and system for fault monitoring of software system based on deep convolution transfer learning
CN110849626B (en) Self-adaptive sparse compression self-coding rolling bearing fault diagnosis system
CN109635677B (en) Composite fault diagnosis method and device based on multi-label classification convolutional neural network
Qin et al. A new supervised multi-head self-attention autoencoder for health indicator construction and similarity-based machinery RUL prediction
CN113392881B (en) Rotary machine fault diagnosis method based on transfer learning
CN110096827B (en) A method for parameter optimization of shield machine based on deep neural network
CN110222371A (en) Engine residual life on-line prediction method based on Bayes and neural network
CN113837000A (en) A Few-Sample Fault Diagnosis Method Based on Task Ranking Meta-Learning
CN110244689A (en) An AUV Adaptive Fault Diagnosis Method Based on Discriminant Feature Learning Method
Tang et al. Deep transfer learning strategy in intelligent fault diagnosis of rotating machinery
Zhu et al. A transfer learning method using high-quality pseudo labels for bearing fault diagnosis
CN113204921B (en) Remaining service life prediction method and system of aircraft turbofan engine
CN112949402A (en) Fault diagnosis method for planetary gear box under minimum fault sample size
CN111444871A (en) A Multi-scale Deep Convolutional Neural Network Fault Diagnosis Method for Planetary Gearboxes
CN114152442A (en) Rolling bearing cross-working condition fault detection method based on migration convolutional neural network
Chang et al. Temporal convolution-based sorting feature repeat-explore network combining with multi-band information for remaining useful life estimation of equipment
CN117009916A (en) Actuator fault diagnosis method based on multi-sensor information fusion and transfer learning
CN115051929A (en) Network fault prediction method and device based on self-supervision target perception neural network
WO2019178930A1 (en) Fault diagnosis method for mechanical device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20211210

RJ01 Rejection of invention patent application after publication