CN116743555A - Robust multi-mode network operation and maintenance fault detection method, system and product - Google Patents
Robust multi-mode network operation and maintenance fault detection method, system and product Download PDFInfo
- Publication number
- CN116743555A CN116743555A CN202310437633.8A CN202310437633A CN116743555A CN 116743555 A CN116743555 A CN 116743555A CN 202310437633 A CN202310437633 A CN 202310437633A CN 116743555 A CN116743555 A CN 116743555A
- Authority
- CN
- China
- Prior art keywords
- mode
- fusion
- result
- kpi
- representation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000012423 maintenance Methods 0.000 title claims abstract description 46
- 238000001514 detection method Methods 0.000 title claims abstract description 36
- 230000004927 fusion Effects 0.000 claims abstract description 80
- 239000013598 vector Substances 0.000 claims abstract description 53
- 238000000034 method Methods 0.000 claims abstract description 39
- 230000006870 function Effects 0.000 claims description 81
- 230000004913 activation Effects 0.000 claims description 21
- 230000028161 membrane depolarization Effects 0.000 claims description 21
- 238000000605 extraction Methods 0.000 claims description 19
- 238000005457 optimization Methods 0.000 claims description 18
- 238000012545 processing Methods 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 13
- 238000012549 training Methods 0.000 claims description 12
- 230000007246 mechanism Effects 0.000 claims description 8
- 230000014509 gene expression Effects 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 3
- 201000010099 disease Diseases 0.000 claims description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 13
- 238000009826 distribution Methods 0.000 description 8
- 238000003745 diagnosis Methods 0.000 description 7
- 238000004891 communication Methods 0.000 description 5
- 239000000284 extract Substances 0.000 description 4
- 230000002159 abnormal effect Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000013136 deep learning model Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000003491 array Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000001627 detrimental effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000011664 signaling Effects 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/06—Management of faults, events, alarms or notifications
- H04L41/069—Management of faults, events, alarms or notifications using logs of notifications; Post-processing of notifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/16—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks using machine learning or artificial intelligence
Abstract
The invention provides a robust multi-mode network operation and maintenance fault detection method, a system and a product, and relates to the technical field of network operation and maintenance, wherein the method comprises the following steps: step 1, inputting a log file into a corresponding single-mode model, and extracting characteristic representation of the log file; step 2, inputting the time sequence of the KPI into a corresponding single-mode model, and extracting the characteristic representation of the KPI time sequence; step 3, carrying out information fusion on the characteristic representations of different modes extracted in the step 1 and the step 2 to obtain a multi-mode characteristic vector; step 4, inputting the multi-mode feature vector in the step 3 into a first classifier to obtain corresponding output; step 5, inputting the features in the step 1 into a single-mode feature extractor to obtain corresponding outputs; and step 6, carrying out information fusion on the output of the step 4 and the step 5 to obtain a fusion information result, and carrying out operation on the fusion information result to obtain a final prediction result.
Description
Technical Field
The invention relates to the technical field of network operation and maintenance, in particular to a method, a system and a product for detecting operation and maintenance faults of a robust multi-mode network.
Background
In recent years, the importance of operation and maintenance technology in various industries is higher and higher, and as the digitalization degree is higher and higher, the system scale is larger and larger, the component monitoring granularity is finer and smaller, the monitoring data volume is larger and new technology and new components are continuously introduced, the network operation and maintenance difficulty is higher and higher, and operation and maintenance engineers are also submerged by a large amount of high-speed operation and maintenance monitoring data. The traditional network fault diagnosis method mainly depends on manual experience and professional knowledge to judge and solve the problems, and the method has certain subjectivity and limitation and can not meet the requirement of a large-scale network.
In recent years, with the development of machine learning technology, some data-driven fault diagnosis methods have been proposed. These methods rely primarily on machine learning algorithms to automatically diagnose network failures by analyzing data from network devices. However, the conventional data driving method generally can only use data of a single mode to perform diagnosis, and cannot use data of multiple modes in the operation and maintenance field to perform fault detection and diagnosis together. In addition, a single type of data may have problems such as insufficient data and data noise, which may affect the accuracy and reliability of fault diagnosis.
Therefore, in order to improve accuracy and reliability of network fault diagnosis, a method capable of simultaneously performing fault diagnosis using multi-modal data is required. However, in the network operation and maintenance field, due to signal-to-noise ratio of data, the multi-mode model often generates bias to one mode, so that the performance of the model is reduced. However, some existing depolarization methods often increase the performance of the non-independent co-distributed data at the expense of performance on the data that is independent of the training data. Both of these situations are detrimental to the robustness of the system.
Disclosure of Invention
The invention provides a robust multi-mode network operation and maintenance fault detection method, a system and a product, which are used for solving the problem that the robustness of a system is insufficient because the existing multi-mode learning model can only learn the distribution characteristics of log data and cannot learn the information characteristics of the data.
In a first aspect, an embodiment of the present invention provides a method for detecting an operation and maintenance fault of a robust multi-mode network, including the following steps:
step 1, inputting a log file into a corresponding single-mode model, and extracting characteristic representation of the log file;
step 2, inputting the time sequence of the KPI into a corresponding single-mode model, and extracting the characteristic representation of the KPI time sequence;
Step 3, carrying out information fusion on the characteristic representations of different modes extracted in the step 1 and the step 2 to obtain a multi-mode characteristic vector;
step 4, inputting the multi-mode feature vector in the step 3 into a first classifier to obtain corresponding output;
step 5, inputting the features in the step 1 into a single-mode feature extractor to obtain corresponding outputs;
and step 6, carrying out information fusion on the output of the step 4 and the step 5 to obtain a fusion information result, and carrying out operation on the fusion information result to obtain a final prediction result.
Based on the first aspect, in step 3, the information fusion is performed on the features of the different modes extracted in step 1 and step 2 to obtain a multi-mode feature vector, which includes:
the characteristic representation of the log file and the characteristic representation of the KPI time sequence are respectively subjected to linear layers, and added to be subjected to an activation function to obtain corresponding representations;
obtaining attention weights from the corresponding representations through a linear layer, and obtaining final attention weights by adopting softmax normalization;
weighting and expressing the attention weight to obtain a fusion expression of characteristic expressions of multiple groups of KPI time sequences;
and processing the fusion representation and the characteristic representation of the log file by using Hadamard products to obtain a multi-mode characteristic vector combined with a top-down attention mechanism.
Based on the first aspect, the multi-modal feature vector E m The definition is as follows:
E m ＝E mk ⊙e l (l)，
wherein E is mk For a fused representation of feature representations of multiple sets of KPI time sequences,the disease is Hadamard product, e l (l) A characteristic representation of the log file;
wherein E is mk The definition is as follows:
where j represents the j-th set of KPI performance index data,for the final attention weight corresponding to the j-th KPI performance index data, j E [1, n ] k ]，e k (k) For the representation of characteristics of KPI time series, k j For the j-th set of KPI performance index data (n in total k Group) of->Alpha is the attention weight obtained through the linear layer, < ->W a Is a linear layer, E Q The characteristic representation of the log file and the characteristic representation of the KPI time sequence are respectively subjected to linear layers, added and subjected to an activation function Relu (), and the obtained corresponding representation, < - >>T in (2) represents a linear layer W α Transpose of E Q,i ＝Relu(W l e l (l)+W k e k,i (k) T ),1≤i≤n k ，E Q,i Combining any one 1-n for characteristic representation of log files k A characteristic representation obtained from a characteristic representation of a KPI time series of the group, e l (l) E is a characteristic representation of the log file k,i (k) The feature represented as the ith KPI time series represents the corresponding input e k (k) Results of [ e ] k,i (k)] T T in (2) represents e k,i (k) Transpose of W l And W is k Is a linear layer.
Based on the first aspect, in step 4, the output is a classification result z that is not depolarized nd The definition is as follows:
z nd ＝softmax(FCN(E m ))，
wherein FCN (·) is a fully connected network, E m Is a multi-modal feature;
in step 5, the output is an n-dimensional feature vector E in a single mode n The definition is as follows:
E n ＝FCN(e l (l))，
wherein FCN (·) is a fully connected network, e l (l) Is a characteristic representation of the log file.
Based on the first aspect, in step 6, the information fusion is performed on the outputs of step 4 and step 5 to obtain a fused information result, and the fused information result is operated to obtain a final prediction result, which includes:
inputting the output of the step 5 into a sigmoid activation function, and then performing inner product with the output of the step 4) to obtain a fusion information result;
carrying out softmax operation on the fusion information result in an n-dimensional space to obtain a final prediction result;
wherein the final prediction result z pred The definition is as follows:
z pred ＝softmax(z)＝softmax(z nd ·σ(E n ))，
wherein z is nd E is the non-depolarization classification result in step 4 n For the n-dimensional feature vector in the single mode in step 5, σ (·) is a sigmoid activation function.
Based on the first aspect, the loss function of the robust multi-modal network is calculated by the following steps:
inputting the output of the step 5 into a second classifier, classifying n labels, and calculating the cross entropy loss of the label classification and the real labels to obtain a single-mode loss function;
Calculating the cross entropy loss of the final prediction result and the real result in the step 6 to obtain a multi-mode loss function;
introducing a loose control factor to optimize the multi-mode loss function to obtain a final loss function;
the final loss function LO is defined as follows:
LO＝γL 1 +L 2 ＝-[a]log(z pred ) γ -[a]log(softmax(c n (E n )))，
wherein the single mode loss function L 2 The definition is as follows:
L 2 ＝-[a]log(softmax(c n (E n )))，
wherein the multi-modal loss function L 1 The definition is as follows:
L 1 ＝-[a]log(z pred )，
wherein gamma E (0, 1) represents a loose control factor, a represents the true result, [ a ]]Representing the real label corresponding to the real result, z nd For the non-depolarized classification result in step 4, σ (·) is the sigmoid activation function, E n C is the n-dimensional feature vector in the single mode in the step 5 n Representing a second classifier.
Based on the detection method of the first aspect, the method further comprises:
in the training process, the calculated loss function is output by the first classifier for each round, loose optimization is performed, and loose control factors are adjusted as follows:
wherein L is 1 Representing the current first classifier output and calculated loss function value, L 1 Representing the loss function value output and calculated by the first classifier of the previous round.
Based on the same inventive concept as described above, in a second aspect, the present invention provides a robust multi-mode network operation and maintenance fault detection system, comprising:
The first extraction module is used for inputting the log file into the corresponding single-mode model and extracting the characteristic representation of the log file;
the second extraction module is used for inputting the time sequence of the KPI into the corresponding single-mode model and extracting the characteristic representation of the KPI time sequence;
the first fusion module is used for carrying out information fusion on the features of different modes in the step 1 and the step 2 to obtain a multi-mode feature vector;
the first classification module is used for inputting the multi-mode feature vector in the step 3 into a first classifier to obtain corresponding output;
the third extraction module is used for inputting the features in the step 1 into the single-mode feature extractor to obtain corresponding output;
and the second fusion module is used for carrying out information fusion on the output of the step 4 and the step 5 to obtain a fusion information result, and carrying out operation on the fusion information result to obtain a final prediction result.
In a third aspect, the present invention provides an electronic device comprising:
a memory for storing one or more programs;
a processor;
the method for robust multi-modal network operation and maintenance fault detection as claimed in any one of the first aspects is implemented when the one or more programs are executed by the processor.
In a fourth aspect, the present invention provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the robust multi-modal network operation and maintenance fault detection method according to any of the first aspects.
The invention has the following advantages:
(1) The embodiment of the invention fully utilizes the multi-mode data in the network operation and maintenance, and greatly enhances the reliability and accuracy of fault prediction;
(2) The embodiment of the invention adopts depolarization optimization to solve the problem that the model learns the bias of the data due to low signal-to-noise ratio of the data in the network operation and maintenance;
(3) According to the embodiment of the invention, after the depolarization optimization is introduced, a loose optimization strategy is added, and the effectiveness of the bias optimization is further ensured by introducing a loose control factor, so that the system can keep higher accuracy when carrying out fault prediction on data with different distributions, and the robustness of the system is greatly increased.
Drawings
Fig. 1 is a flowchart of a robust multi-mode network operation and maintenance fault detection method according to an embodiment of the present invention;
fig. 2 is a schematic frame flow diagram of a multi-mode information fusion module according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a frame flow for robust multi-mode operation and maintenance fault detection according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a robust multi-mode network operation and maintenance fault detection system according to an embodiment of the present invention;
fig. 5 is a schematic block diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The invention provides a robust multi-mode network operation and maintenance fault detection method, a system and a product, which can utilize the multi-mode information of a machine KPI performance index and a log file to carry out fault detection on a network operation and maintenance system, and can maintain the stability and reliability of a prediction result when processing data with different distributions; firstly, respectively extracting characteristics of KPI performance indexes and log files, then processing the extracted multi-modal characteristics by utilizing a multi-modal information fusion module to obtain corresponding multi-modal characteristics, carrying out preliminary classification by a classifier, carrying out depolarization treatment to obtain single-modal characteristic vectors required by depolarization, and combining the single-modal characteristic vectors with the previous classification results to obtain a final fault detection classification result; in addition, in the training stage of the model, a loose optimization strategy is introduced, so that the dependence of the model on data distribution is reduced, and the robustness of system fault detection is enhanced.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
Referring to fig. 1, fig. 1 is a flowchart of a robust multi-mode network operation and maintenance fault detection method according to an embodiment of the present invention. The robust multi-mode network operation and maintenance fault detection method comprises the following steps:
step 1, inputting a log file into a corresponding single-mode model, and extracting characteristic representation of the log file;
step 1 is directed to obtaining a characteristic representation of a log file, inputting the log file in text form to an encoderObtaining the dimension d z Is a feature vector of (1); specifically, word segmentation and serialization are carried out on a log file in a text form within the time t according to a word list, and then the serialized data I is input into a transducer model to obtain a corresponding characteristic representation
Step 2, inputting the time sequence of the KPI into a corresponding single-mode model, and extracting the characteristic representation of the KPI time sequence;
step 2 is aimed at obtaining KPI performance index feature representation, and n is defined as k KPI performance index k input encoder in group time series formObtaining n k The group dimension is d z Is a feature vector of (1); specifically, KPI performance index data k in a time series form within t time is input into a transducer model to obtain a corresponding characteristic representation e of the KPI time series in the time series form within t time k (k) The method comprises the steps of carrying out a first treatment on the surface of the The KPI performance indexes are KPI performance indexes, such as network delay, bandwidth occupancy rate, packet loss rate, CPU usage rate, and the like, and generally exist in the form of time sequence series.
Step 3, carrying out information fusion on the characteristic representations of different modes extracted in the step 1 and the step 2 to obtain a multi-mode characteristic vector;
in step 3, the information fusion is performed on the features of the different modes extracted in step 1 and step 2 to obtain a multi-mode feature vector, which includes:
the characteristic representation of the log file and the characteristic representation of the KPI time sequence are respectively subjected to linear layers, and added to be subjected to an activation function to obtain corresponding representations;
obtaining attention weights from the corresponding representations through a linear layer, and obtaining final attention weights by adopting softmax normalization;
weighting and expressing the attention weight to obtain a fusion expression of characteristic expressions of multiple groups of KPI time sequences;
and processing the fusion representation and the characteristic representation of the log file by using Hadamard products to obtain a multi-mode characteristic vector combined with a top-down attention mechanism.
Wherein, in the embodiment of the invention, the activation function is Relu ().
Wherein the multi-modal feature vector E m The definition is as follows:
E m ＝E mk ⊙e l (l)，
wherein E is mk For a fused representation of the feature representations of the m sets of KPI time sequences,the disease is Hadamard product, e l (l) A characteristic representation of the log file;
wherein E is mk The definition is as follows:
where j represents the j-th set of KPI performance index data,for the final attention weight corresponding to the j-th KPI performance index data, j E [1, n ] k ]，e k (k) For the representation of characteristics of KPI time series, k j For the j th group KPI Performance index data (total n k Group) of->Alpha is the attention weight obtained through the linear layer, < ->W a Is a linear layer, E Q The characteristic representation of the log file and the characteristic representation of the KPI time sequence are respectively subjected to linear layers, added and subjected to an activation function Relu (), and the obtained corresponding representation, < - >>T in (2) represents a linear layer W α Transpose of E Q,i ＝Relu(W l e l (l)+W k e k,i (k) T ),1≤i≤n k ，E Q,i Combining any one 1-n for characteristic representation of log files k A characteristic representation obtained from a characteristic representation of a KPI time series of the group, e l (l) E is a characteristic representation of the log file k,i (k) The feature represented as the ith KPI time series represents the corresponding input e k (k) Results of [ e ] k,i (k)] T T in (2) represents e k,i (k) Transpose of W l And W is k Is a linear layer.
Step 3 is aimed at obtaining multi-modal feature vectors, and representing the features of the log file extracted in step 1 as e l (l) And step 2, extracting a characteristic representation e of the KPI time series k (k) Input multi-mode encoderThe single-mode characteristics obtained from the bottom to the top are fused through a top-to-bottom attention mechanism, so that the model is facilitated to extract information E with finer granularity of data m ＝e m (e l (l),e k (k) A) is provided; specifically, the information fusion is carried out on the characteristics of different modes obtained in the step 1 and the step 2 in a bottom-up mode, so as to obtain multi-mode information characteristics; will firstRespectively via linear layers->And adding the activated functions Relu (& gt) to obtain the corresponding representation +/>Wherein E is Q,i ＝Relu(W l e l (l)+W k e k,i (k) T ),1≤i≤n k ，E Q,i Combining any one 1-n for characteristic representation of log files k A feature representation derived from the feature representation of the KPI time series of the group; then E is again carried out Q Through a linear layer->Obtaining the corresponding attention weight alpha, and obtaining the final attention weight after normalization by adopting softmax>Wherein (1)> T in (2) represents a linear layer W α Is a transpose of (2); thus n can be reduced k Weighted representation of group KPI performance indexes to obtain fusion representation E of feature representation of corresponding group KPI time series mk Fusion representation E of feature representations of multiple sets of KPI time series mk And characteristic representation e of log file l (l) By doing Hadamard product, the multi-modal feature vector combined with the top-down attention mechanism can be obtained >Wherein E is m ＝E mk ⊙e l (l)。
Exemplary referring to fig. 2, fig. 2 is a diagram illustrating a multi-modal information fusion according to an embodiment of the present inventionThe frame flow diagram of the integrated module is specifically that the features of the bottom-up single-mode feature log file are represented as e l (l) And a time series of characteristic representations e of KPI performance indicators k (k) Respectively through the linear layers W l And W is k The outputs are then summed up through an activation function Relu (), resulting in the corresponding representation E Q Then E is carried out Q Through a linear layer W α Obtaining a corresponding attention weight alpha, and obtaining a final attention weight after normalization by adopting softmaxWill n k Weighted representation of group KPI performance indexes to obtain fusion representation E of feature representation of corresponding group KPI time series mk Then the fusion representation E of the characteristic representations of the multiple sets of KPI time series mk And characteristic representation e of log file l (l) By Hadamard product, the multi-modal feature vector E combined with the top-down attention mechanism can be obtained m The attention weight acquisition process of the log file is to acquire the attention weight by adopting a top-down attention mechanism.
Step 4, inputting the multi-mode feature vector in the step 3 into a first classifier to obtain corresponding output;
Wherein the first classifier is a multi-modal classifier;
step 5, inputting the features in the step 1 into a single-mode feature extractor to obtain corresponding outputs;
in step 4), the output is a non-depolarized classification result z nd The definition is as follows:
z nd ＝softmax(FCN(E m ))，
wherein FCN (·) is a fully connected network, E m Is a multi-modal feature;
in step S510, the output is an n-dimensional feature vector E in a single mode n The definition is as follows:
E n ＝FCN(e l (l))，
wherein FCN (& gt) is a fully connected network,e l (l) Is a characteristic representation of the log file.
Step 4, aiming at obtaining a non-depolarized classification result, and passing the multi-modal characteristics through a multi-modal classifierObtaining a non-depolarized classification result z nd The method comprises the steps of carrying out a first treatment on the surface of the Step 5 is aimed at obtaining the final classification result, inputting the log file l into a new unimodal feature extractor +.>Obtaining the corresponding n-dimensional characteristic E under single mode n Finally, the classified fault result z is obtained pred ＝softmax(z nd ·σ(E n ) A) is provided; specifically, in step 4, the multimodal feature vector E obtained in step 3 is used m Inputting into a multi-modal classifier to obtain corresponding output, specifically E m As an input to a fully connected network, it is taken from d z Mapping the dimensions to n dimensions of the types (including no faults) of the preset fault types to obtain a non-depolarization classification result z nd ，z nd ＝softmax(FCN(E m ) Wherein FCN (·) is a fully connected network, E) m Is a multi-modal feature; specifically, in step 5, the feature representation in step 1 is input into a single-mode feature extractor to obtain a corresponding output; d, recording log file in step 1 z The dimension characteristic representation is mapped to n dimensions through a fully connected network to obtain n dimension characteristic vectors E under a corresponding single mode n ，E n ＝FCN(e l (l) Wherein FCN (·) is a fully connected network, e) l (l) Is a characteristic representation of the log file.
And step 6, carrying out information fusion on the output of the step 4 and the step 5 to obtain a fusion information result, and carrying out operation on the fusion information result to obtain a final prediction result.
The invention provides a robust multi-mode network operation and maintenance fault detection system which can utilize multi-mode information of machine KPI performance indexes and log files to carry out fault detection on the network operation and maintenance system and can maintain stability and reliability of a prediction result when processing data with different distributions. According to the invention, firstly, the KPI and the log file are respectively subjected to feature extraction, then the extracted multi-modal features are processed by utilizing a multi-modal information fusion module to obtain corresponding multi-modal features, the corresponding multi-modal features are subjected to preliminary classification by a classifier, then the single-modal feature vectors required by depolarization are obtained by depolarization processing, and the single-modal feature vectors are combined with the previous classification results to obtain the final fault detection classification results. In addition, in the training stage of the model, a loose optimization strategy is introduced, so that the dependence of the model on data distribution is reduced, and the robustness of system fault detection is enhanced.
In step 6, the information fusion is performed on the outputs of step 4 and step 5 to obtain a fusion information result, and the fusion information result is operated to obtain a final prediction result, which includes:
inputting the output of the step 5 to a sigmoid activation function, and then performing inner product with the output of the step 4 to obtain a fusion information result;
carrying out softmax operation on the fusion information result in an n-dimensional space to obtain a final prediction result;
wherein the final prediction result z pred The definition is as follows:
z pred ＝softmax(z)＝softmax(z nd ·σ(E n ))，
wherein z is nd E is the non-depolarization classification result in step 4 n For the n-dimensional feature vector in the single mode in step 5, σ (·) is a sigmoid activation function.
In step 6, step 6 and step 5 aim at eliminating bias influence, because the data signal-to-noise ratio in the operation and maintenance field is very low, a large amount of data are normal data, and the abnormal data only occupy a small part, the deep learning model possibly generates data bias in the training and learning process, and the model can strongly correlate the system state with the log file in the learning process, and neglect the contribution of the KPI performance index of another machine. Thus, step 6 is introduced to eliminate this bias. After step 6 is completed, the loss function LO of the whole system can be obtained, and back propagation is performed to update the model parameters. Care should be taken that: the corresponding procedure in step 6 is not counter-propagated to prevent the encoder from learning the deviation of the data directly, i.e. the parameters between the unimodal feature extractor in step 5 and step 1 do not calculate the mutual gradient, for example, the specific corresponding procedure can be seen in the x-section of fig. 3.
Specifically, first, the n-dimensional unimodal log file feature vector E in step 5 is set n Inputting a sigmoid activation function, and then matching with the classification result z in the step 4 nd And (3) performing inner product to obtain a corresponding fusion information result z:
z＝z nd ·σ(E n )，
then, the fused information result z is subjected to softmax operation on an n-dimensional space to obtain a final prediction result z pred ，
z pred ＝softmax(z)＝softmax(z nd ·σ(E n ))，
Wherein z is the result of fusion information, z nd E is the non-depolarization classification result in step 4 n For the n-dimensional feature vector in the single mode in step 5, σ (·) is a sigmoid activation function.
In addition, the loss function of the robust multi-modal network is calculated by the following steps:
inputting the output of the step 5 into a second classifier, classifying n labels, and calculating the cross entropy loss of the label classification and the real labels to obtain a single-mode loss function;
calculating the cross entropy loss of the final prediction result and the real result in the step 6 to obtain a multi-mode loss function;
introducing a loose control factor to optimize the multi-mode loss function to obtain a final loss function;
the final loss function LO is defined as follows:
LO＝γL 1 +L 2 ＝-[a]log(z pred ) γ -[a]log(softmax(c n (E n )))，
wherein the single mode loss function L 2 The definition is as follows:
L 2 ＝-[a]log(softmax(c n (E n )))，
wherein the multi-modal loss function L 1 The definition is as follows:
L 1 ＝-[a]log(z pred )，
wherein gamma E (0, 1) represents a loose control factor, a represents the true result, [ a ]]Representing the real label corresponding to the real result, z nd For the non-depolarized classification result in step 4, σ (·) is the sigmoid activation function, E n C is the n-dimensional feature vector in the single mode in the step 5 d Representing a second classifier; wherein the second classifier is a single-mode classifier.
Inputting the characteristics of the step 5 into a second classifier, and calculating the losses of the final prediction results and the real results of the step (step 5) and the step 6; performing loose optimization processing on the loss of the final predicted result and the real result to obtain a final loss, and continuously iterating the training model; specifically, the cross entropy loss function L of the classification result in the step 6 and the real label is calculated 1 The n-dimensional single mode characteristic E of the step 5 n Input into a classifierIn the process, n label classification is carried out, and a cross entropy loss function L of the output of the step 5 and the real label is calculated 2 。
In the embodiment of the invention, the loss function comprises a cross entropy loss function L 1 And a cross entropy loss function L 2 The two parts are as follows:
L 1 ＝-[a]log(z pred )，
L 2 ＝-[a]log(softmax(c n (E n )))，
based on the cross entropy loss function L 1 And a cross entropy loss function L 2 Adding to obtain the original loss function ce=l 1 +L 2 ；
Introducing a loose control factor gamma to the cross entropy loss function L 1 And part of the loose optimization processing is performed, and the optimized loss function LO is defined as follows:
LO＝γL 1 +L 2 ＝-[a]log(z pred ) γ -[a]log(softmax(c n (E n )))，
wherein a represents the true result, [ a ]]Representing the real label corresponding to the real result, gamma E (0, 1) representing the loose control factor, z nd For the non-depolarized classification result in step 4, σ (·) is the sigmoid activation function, E n C is the n-dimensional feature vector in the single mode in the step 5 n Representing a second classifier.
The principle of the above-described calculation of the loss function is that for the cross entropy loss function, its derivative is:
in the method, in the process of the invention,a epsilon { + -1 } respectively corresponds to whether the classification result is correct or not, and p is the prediction probability; it can be seen that the derivative of the loosely optimized loss function is compared to the original derivative:
wherein p is d Is the final prediction result obtained in step 6, and p b The prediction result obtained in the step 4; from 0 to<γ<1 and 0.ltoreq.p d Less than or equal to 1, can obtainTherefore, the method can update the training model parameters by using smaller gradient change values to reduce model bias learning capacity caused by the characteristics of log functions, further ensure the effectiveness of bias optimization and improve the robustness of the system.
Further, based on the robust multi-mode network operation and maintenance fault detection method, the method further comprises the following steps:
In the training process, the calculated loss function is output by the first classifier for each round, loose optimization is performed, and loose control factors are adjusted as follows:
wherein L is 1 Representing the current first classifier output and calculated loss function value, L 1 Representing the loss function value output and calculated by the first classifier of the previous round; wherein the first classifier is a multi-modal classifier.
Referring to fig. 3, in an exemplary embodiment, fig. 3 is a schematic flow diagram of a framework for robust multi-mode operation and maintenance fault detection provided by the embodiment of the present invention, specifically, firstly, extracting a feature representation of a log file and a feature representation of a KPI performance index through a single-mode model, performing multi-mode information fusion on the extracted feature representation of the log file and the feature representation of the KPI performance index, and inputting the fused multi-mode information into a fault classifier to obtain a non-depolarized prediction result, where the fault classifier is a multi-mode classifier; then, a depolarization module is adopted to enable the model to reduce bias by training log file data, the depolarization module comprises a step 5 and a step 6, and in the specific implementation process, the depolarization module realizes depolarization processing through the step 5 and the step 6, specifically comprises the following steps: inputting the extracted characteristic representation of the log file into a single-mode characteristic extractor to obtain an n-dimensional characteristic vector under a single mode, inputting the characteristic vector into a sigmoid activation function, carrying out information fusion on a Hadamard product and a non-depolarization prediction result to obtain a fusion information result, and carrying out softmax operation to obtain a final prediction result [0.8,0.1,0.1], wherein the final prediction result [0.8,0.1,0.1] is a classification result obtained by a multi-mode classifier on the fusion information result; the characteristic representation of the log file is input into a fault classifier to obtain a corresponding prediction result [0.7,0.2,0.1], the fault classifier is a single-mode classifier, cross entropy loss functions are calculated on the final prediction result [0.8,0.1,0.1] and the prediction result [0.7,0.2,0.1] respectively, loose processing is carried out on the cross entropy loss functions calculated on the final prediction result [0.8,0.1,0.1] to obtain an optimized loss function, and the optimized prediction result [1, 0] is obtained.
In the implementation process, through the step 1, the log file is input into a corresponding single-mode model, and the characteristic representation of the log file is extracted; step 2, inputting the time sequence of the KPI into a corresponding single-mode model, and extracting the characteristic representation of the KPI time sequence; step 3, carrying out information fusion on the characteristic representations of different modes extracted in the step 1 and the step 2 to obtain a multi-mode characteristic vector; step 4, inputting the multi-mode feature vector in the step 3 into a first classifier to obtain corresponding output; step 5, inputting the features in the step 1 into a single-mode feature extractor to obtain corresponding outputs; and step 6, carrying out information fusion on the output of the step 4 and the step 5 to obtain a fusion information result, and carrying out operation on the fusion information result to obtain a final prediction result. The robust multi-mode network operation and maintenance fault detection method provided by the embodiment of the invention fully utilizes multi-mode data in network operation and maintenance, and greatly enhances the reliability and accuracy of fault prediction; the problem that the model learns biased data due to low data signal-to-noise ratio in network operation and maintenance is solved by adopting depolarization optimization; after the depolarization optimization is introduced, a loose optimization strategy is added, and the effectiveness of the bias optimization is further ensured by introducing a loose control factor, so that the system can keep higher accuracy when carrying out fault prediction on data with different distributions, and the robustness of the system is greatly improved.
Based on the same inventive concept, the embodiment of the invention also provides a robust multi-mode network operation and maintenance fault detection system. Referring to fig. 4, fig. 4 is a schematic diagram of a robust multi-mode network operation and maintenance fault detection system according to an embodiment of the present invention, including:
the first extraction module 210 is configured to input the log file into a corresponding single-mode model, and extract a feature representation of the log file in step 1;
a second extraction module 220, configured to input the time sequence of KPI indicators into a corresponding single-mode model, and extract a feature representation of the KPI time sequence in step 2;
the first fusion module 230 is configured to perform information fusion on the features of different modes extracted in the step 1 and the step 2 to obtain a multi-mode feature vector;
the first classification module 240 is configured to input the multi-modal feature vector in step 3 into a first classifier to obtain a corresponding output;
the third extraction module 250 is configured to input the features in step 1 to the single-mode feature extractor to obtain corresponding outputs in step 5;
and a second fusion module 260, configured to perform information fusion on the outputs of step 4 and step 5 to obtain a fused information result, and perform operation on the fused information result to obtain a final prediction result.
In the above implementation process, the present invention provides a robust multi-mode network operation and maintenance fault detection system, where a log file is input into a corresponding single-mode model through a first extraction module 210, and a feature representation of the log file is extracted; the second extraction module 220 inputs the time sequence of the KPI indicators into the corresponding single-mode model, and extracts the characteristic representation of the KPI time sequence; the first fusion module 230 carries out information fusion on the features of different modes extracted in the first extraction module and the second extraction module to obtain a multi-mode feature vector; the first classification module 240 inputs the multi-modal feature vectors in the first fusion module into a first classifier to obtain corresponding outputs; the third extraction module 250 inputs the features in the first extraction module into a single-mode feature extractor to obtain corresponding outputs; the second fusion module 260 performs information fusion on the outputs of the first classification module and the third extraction module to obtain a fusion information result, and performs operation on the fusion information result to obtain a final prediction result.
Specifically, first, in order to make full use of mass data in the network operation and maintenance field, a first fusion module is introduced, and the first fusion module can fuse the characteristic representation of the log file extracted by the first extraction module and the characteristic representation of the KPI time sequence extracted by the second extraction module through the top-down attention weight, extract data characteristics from a finer granularity angle and acquire multi-mode characteristic vectors. Next, because the signal-to-noise ratio of the data in the operation and maintenance field is very low, a large amount of data is normal data, and the abnormal data only occupies a small part, the deep learning model may be affected by data bias in the training learning process, for example: for fault types with abnormal KPI and normal log files, as most of log files are normal in normal corresponding results, the model can strongly correlate the normal system state with the normal log files in the learning process, and the contribution of the KPI performance index of the other machine is ignored.
Furthermore, during the training phase of the model, cross entropy loss functions are typically employed to optimize the model parameters. When the model finally predicts a label with a probability p that is smaller than the 1, a larger loss occurs due to the nature of the log function. In practice, however, the set of faults is usually large, i.e. the number of labels n is large (e.g. n=100), and when p is relatively small (e.g. p=0.4), p is relatively small enough 1/n=0.01 is a relatively large number, in which case a loss of-log 0.4≡0.40 still occurs. At this time, if the gradient is updated greatly according to the loss function, the model is forced to learn a certain mapping relation between the anomaly and the single-mode data, so that the model has a stronger capability of remedying a certain bias. Therefore, in order to ensure the robustness of the system, the invention introduces a loosely optimized strategy. By introducing a loose control factor gamma, the loss function value can be reduced, so that the model can obtain better performance on various distributed data.
The robust multi-mode network operation and maintenance fault detection method provided by the invention fully utilizes multi-mode data in network operation and maintenance, and greatly enhances the reliability and accuracy of fault prediction; the depolarization processing is adopted, so that the problem that the model learns biased data due to low data signal-to-noise ratio in network operation and maintenance is solved; after the depolarization treatment is introduced, a loose optimization strategy is added, and the effectiveness of bias optimization is further ensured by introducing a loose control factor, so that the system can keep higher accuracy when carrying out fault prediction on data with different distributions, and the robustness of the system is greatly improved.
Referring to fig. 5, fig. 5 is a schematic block diagram of an electronic device according to an embodiment of the present invention. The electronic device comprises a memory 101, a processor 102 and a communication interface 103, wherein the memory 101, the processor 102 and the communication interface 103 are electrically connected with each other directly or indirectly to realize data transmission or interaction. For example, the components may be electrically connected to each other via one or more communication buses or signal lines. The memory 101 may be used to store software programs and modules, such as program instructions/modules corresponding to the class incremental learning system based on the dynamic class prototype generation mechanism provided in the embodiments of the present invention, and the processor 102 executes the software programs and modules stored in the memory 101, thereby executing various functional applications and data processing. The communication interface 103 may be used for communication of signaling or data with other node devices.
The Memory 101 may be, but is not limited to, a random access Memory (Random Access Memory, RAM), a Read Only Memory (ROM), a programmable Read Only Memory (Programmable Read-Only Memory, PROM), an erasable Read Only Memory (Erasable Programmable Read-Only Memory, EPROM), an electrically erasable Read Only Memory (Electric Erasable Programmable Read-Only Memory, EEPROM), etc.
The processor 102 may be an integrated circuit chip with signal processing capabilities. The processor 102 may be a general purpose processor including a central processing unit (Central Processing Unit, CPU), a network processor (Network Processor, NP), etc.; but also digital signal processors (Digital Signal Processing, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), field programmable gate arrays (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.
It will be appreciated that the configuration shown in fig. 5 is merely illustrative, and that the electronic device may also include more or fewer components than shown in fig. 5, or have a different configuration than shown in fig. 5. The components shown in fig. 5 may be implemented in hardware, software, or a combination thereof.
In the embodiments provided in the present invention, it should be understood that the disclosed system and method may be implemented in other manners as well. The apparatus embodiments described above are merely illustrative, for example, of the flowcharts and block diagrams in the figures that illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
In addition, functional modules in the embodiments of the present invention may be integrated together to form a single part, or each module may exist alone, or two or more modules may be integrated to form a single part.
The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Claims (10)
1. The robust multi-mode network operation and maintenance fault detection method is characterized by comprising the following steps of:
step 1, inputting a log file into a corresponding single-mode model, and extracting characteristic representation of the log file;
step 2, inputting the time sequence of the KPI into a corresponding single-mode model, and extracting the characteristic representation of the KPI time sequence;
step 3, carrying out information fusion on the characteristic representations of different modes extracted in the step 1 and the step 2 to obtain a multi-mode characteristic vector;
step 4, inputting the multi-mode feature vector in the step 3 into a first classifier to obtain corresponding output;
Step 5, inputting the features in the step 1 into a single-mode feature extractor to obtain corresponding outputs;
and step 6, carrying out information fusion on the output of the step 4 and the step 5 to obtain a fusion information result, and carrying out operation on the fusion information result to obtain a final prediction result.
2. The detection method according to claim 1, wherein in step 3, the information fusion is performed on the features of different modes extracted in step 1 and step 2 to obtain a multi-mode feature vector, which includes:
the characteristic representation of the log file and the characteristic representation of the KPI time sequence are respectively subjected to linear layers, and added to be subjected to an activation function to obtain corresponding representations;
obtaining attention weights from the corresponding representations through a linear layer, and obtaining final attention weights by adopting softmax normalization;
weighting and expressing the attention weight to obtain a fusion expression of characteristic expressions of multiple groups of KPI time sequences;
and processing the fusion representation and the characteristic representation of the log file by using Hadamard products to obtain a multi-mode characteristic vector combined with a top-down attention mechanism.
3. The detection method according to claim 2, wherein the multi-modal feature vector E m The definition is as follows:
E m ＝E mk ⊙e l (l)，
wherein E is mk For a fused representation of the feature representations of the m sets of KPI time sequences,the disease is Hadamard product, e l (l) A characteristic representation of the log file;
wherein E is mk The definition is as follows:
where j represents the j-th set of KPI performance index data,for the final attention weight corresponding to the j-th KPI performance index data, j E [1, n ] k ]，e k (k) For the representation of characteristics of KPI time series, k j For the j-th set of KPI performance index data (n in total k Group),alpha is the attention weight obtained through the linear layer, < ->W α Is a linear layer, E Q The characteristic representation of the log file and the characteristic representation of the KPI time sequence are respectively subjected to linear layers, added and subjected to an activation function Relu (), and the obtained corresponding representation, < - >>T in (2) represents a linear layer W α Transpose of E Q,i ＝Relu(W l e l (l)+W k [e k,i (k)] T ),1≤i≤n k ，E Q,i Combining any one 1-n for characteristic representation of log files k A characteristic representation obtained from a characteristic representation of a KPI time series of the group, e l (l) E is a characteristic representation of the log file k,i (k) The feature represented as the ith KPI time series represents the corresponding input e k (k) Results of [ e ] k,i (k)] T T in (2) represents e k,i (k) Transpose of W l And W is k Is a linear layer.
4. The method according to claim 1, wherein in step 4, the output is a classification result z which is not depolarized nd The definition is as follows:
z nd ＝softmax(FCN(E m ))，
wherein FCN (·) is a fully connected network, E m Is a multi-modal feature;
in step 5, the output is an n-dimensional feature vector E in a single mode n The definition is as follows:
E n ＝FCN(e l (l))，
wherein FCN (·) is a fully connected network, e l (l) Is a characteristic representation of the log file.
5. The detection method according to claim 1, wherein in step 6, the information fusion is performed on the outputs of step 4 and step 5 to obtain a fused information result, and the fused information result is operated to obtain a final prediction result, which includes:
inputting the output of the step 5 into a sigmoid activation function, and then performing inner product with the output of the step 4) to obtain a fusion information result;
carrying out softmax operation on the fusion information result in an n-dimensional space to obtain a final prediction result;
wherein the final prediction result z pred The definition is as follows:
z pred ＝softmax(z)＝softmax(z nd ·σ(E n ))，
wherein z is nd E is the non-depolarization classification result in step 4 n For the n-dimensional feature vector in the single mode in step 5, σ (·) is a sigmoid activation function.
6. The method of claim 1, wherein the loss function of the robust multi-modal network is calculated by:
inputting the output of the step 5 into a second classifier, classifying n labels, and calculating the cross entropy loss of the label classification and the real labels to obtain a single-mode loss function;
Calculating the cross entropy loss of the final prediction result and the real result in the step 6 to obtain a multi-mode loss function;
introducing a loose control factor to optimize the multi-mode loss function to obtain a final loss function;
the final loss function LO is defined as follows:
LO＝γL 1 +L 2 ＝-[a]log(z pred ) γ -[a]log(softmax(c n (E n )))，
wherein the single mode loss function L 2 The definition is as follows:
L 2 ＝-[a]log(softmax(c n (E n )))，
wherein the multi-modal loss function L 1 The definition is as follows:
L 1 ＝-[a]log(z pred )，
wherein gamma E (0, 1) represents a loose control factor, a represents the true result, [ a ]]Representing the real label corresponding to the real result, z nd For the non-depolarized classification result in step 4, σ (·) is the sigmoid activation function, E n C is the n-dimensional feature vector in the single mode in the step 5 n Representing a second classifier.
7. The method of detecting according to claim 6, further comprising:
in the training process, the calculated loss function is output by the first classifier for each round, loose optimization is performed, and loose control factors are adjusted as follows:
wherein L is 1 Representing the current first classifier output and calculated loss function value,representing the loss function value output and calculated by the first classifier of the previous round.
8. A robust multi-modal network operation and maintenance fault detection system, comprising:
The first extraction module is used for inputting the log file into the corresponding single-mode model and extracting the characteristic representation of the log file;
the second extraction module is used for inputting the time sequence of the KPI into the corresponding single-mode model and extracting the characteristic representation of the KPI time sequence;
the first fusion module is used for carrying out information fusion on the features of different modes in the step 1 and the step 2 to obtain a multi-mode feature vector;
the first classification module is used for inputting the multi-mode feature vector in the step 3 into a first classifier to obtain corresponding output;
the third extraction module is used for inputting the features in the step 1 into the single-mode feature extractor to obtain corresponding output;
and the second fusion module is used for carrying out information fusion on the output of the step 4 and the step 5 to obtain a fusion information result, and carrying out operation on the fusion information result to obtain a final prediction result.
9. An electronic device, comprising:
a memory for storing one or more programs;
a processor;
the method for robust multi-modal network operation and maintenance fault detection according to any of claims 1-7 is implemented when the one or more programs are executed by the processor.
10. A computer readable storage medium having stored thereon a computer program, which when executed by a processor implements the robust multi-modal network operation and maintenance failure detection method according to any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310437633.8A CN116743555A (en) | 2023-04-21 | 2023-04-21 | Robust multi-mode network operation and maintenance fault detection method, system and product |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310437633.8A CN116743555A (en) | 2023-04-21 | 2023-04-21 | Robust multi-mode network operation and maintenance fault detection method, system and product |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116743555A true CN116743555A (en) | 2023-09-12 |
Family
ID=87906815
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310437633.8A Pending CN116743555A (en) | 2023-04-21 | 2023-04-21 | Robust multi-mode network operation and maintenance fault detection method, system and product |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116743555A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117475240A (en) * | 2023-12-26 | 2024-01-30 | 创思(广州)电子科技有限公司 | Vegetable checking method and system based on image recognition |
-
2023
- 2023-04-21 CN CN202310437633.8A patent/CN116743555A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117475240A (en) * | 2023-12-26 | 2024-01-30 | 创思(广州)电子科技有限公司 | Vegetable checking method and system based on image recognition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111124840B (en) | Method and device for predicting alarm in business operation and maintenance and electronic equipment | |
| CN108737406B (en) | Method and system for detecting abnormal flow data | |
| CN111914873A (en) | Two-stage cloud server unsupervised anomaly prediction method | |
| US20210209477A1 (en) | Anomaly detection using a non-mirrored dimensional-reduction model | |
| CN111314331A (en) | Unknown network attack detection method based on conditional variation self-encoder | |
| CN110110318B (en) | Text steganography detection method and system based on cyclic neural network | |
| CN104217216A (en) | Method and device for generating detection model, method and device for detecting target | |
| CN111652290B (en) | Method and device for detecting countermeasure sample | |
| CN112039903B (en) | Network security situation assessment method based on deep self-coding neural network model | |
| CN111711608B (en) | Method and system for detecting abnormal flow of power data network and electronic equipment | |
| Wang et al. | Learning efficient binarized object detectors with information compression | |
| CN114978613B (en) | Network intrusion detection method based on data enhancement and self-supervision feature enhancement | |
| CN110851654A (en) | Industrial equipment fault detection and classification method based on tensor data dimension reduction | |
| CN109766259B (en) | Classifier testing method and system based on composite metamorphic relation | |
| CN116743555A (en) | Robust multi-mode network operation and maintenance fault detection method, system and product | |
| CN114048468A (en) | Intrusion detection method, intrusion detection model training method, device and medium | |
| KR20230107558A (en) | Model training, data augmentation methods, devices, electronic devices and storage media | |
| CN116451139A (en) | Live broadcast data rapid analysis method based on artificial intelligence | |
| Wan et al. | Confnet: predict with confidence | |
| Deng et al. | Semi-supervised ensemble fault diagnosis method based on adversarial decoupled auto-encoder with extremely limited labels | |
| CN117407665A (en) | Retired battery time sequence data missing value filling method based on generation countermeasure network | |
| CN111107082A (en) | Immune intrusion detection method based on deep belief network | |
| CN116361788A (en) | Binary software vulnerability prediction method based on machine learning | |
| CN115661539A (en) | Less-sample image identification method embedded with uncertainty information | |
| CN115310837A (en) | Complex electromechanical system fault detection method based on causal graph attention neural network |
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 |