CN116415197A - Underground pipe gallery abnormality detection network and method based on attention mechanism - Google Patents
Underground pipe gallery abnormality detection network and method based on attention mechanism Download PDFInfo
- Publication number
- CN116415197A CN116415197A CN202310236950.3A CN202310236950A CN116415197A CN 116415197 A CN116415197 A CN 116415197A CN 202310236950 A CN202310236950 A CN 202310236950A CN 116415197 A CN116415197 A CN 116415197A
- Authority
- CN
- China
- Prior art keywords
- layer
- attention mechanism
- conv
- lstm
- feature extraction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000007246 mechanism Effects 0.000 title claims abstract description 63
- 238000001514 detection method Methods 0.000 title claims abstract description 31
- 238000000034 method Methods 0.000 title claims abstract description 30
- 230000005856 abnormality Effects 0.000 title claims abstract description 26
- 238000000605 extraction Methods 0.000 claims abstract description 25
- 239000011159 matrix material Substances 0.000 claims description 34
- 238000004590 computer program Methods 0.000 claims description 5
- 230000015654 memory Effects 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 abstract description 8
- 230000000694 effects Effects 0.000 abstract description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000012544 monitoring process Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000012423 maintenance Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000002441 reversible effect Effects 0.000 description 2
- 101150008604 CAN1 gene Proteins 0.000 description 1
- 101150063504 CAN2 gene Proteins 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 239000002737 fuel gas Substances 0.000 description 1
- 239000007789 gas Substances 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- -1 heating power Substances 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000009440 infrastructure construction Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 229910052760 oxygen Inorganic materials 0.000 description 1
- 239000001301 oxygen Substances 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 239000000779 smoke Substances 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention provides an underground pipe gallery abnormality detection network and method based on an attention mechanism, which can ensure high accuracy of long-time abnormality detection and have good detection effect on abnormal conditions under multiple data. The network comprises: the device comprises an encoding module, a feature extraction module and a decoding module. The coding module comprises four convolution layers which are sequentially connected. The feature extraction module comprises a Conv-LSTM layer and an attention mechanism layer, wherein the Conv-LSTM layer comprises four Conv-LSTM layers with the same structure, and the attention mechanism layer comprises four attention mechanisms with the same structure. The decoding module comprises seven layers of deconvolution layers and connection layers, which are connected in sequence. The first convolution layer in the coding module is connected with the first Conv-LSTM of the Conv-LSTM layer of the feature extraction module; the first Conv-LSTM of the Conv-LSTM layer in the feature extraction module is connected to the first attention mechanism of the attention mechanism layer, and so on. The attention mechanism layer of the feature extraction module is reversely connected with the deconvolution layer of the decoding module.
Description
Technical Field
The invention belongs to the field of anomaly detection, and particularly relates to an underground pipe gallery anomaly detection network and method based on an attention mechanism. The method is particularly suitable for detecting the abnormality of the underground pipe gallery and preventing and controlling the risk under the conditions of long time and multiple data.
Background
Along with the urban construction process of China, the urban scale is gradually enlarged, means and methods for public infrastructure construction are greatly developed, and underground pipe gallery construction becomes an important one. The underground pipe gallery is an underground pipeline which concentrates pipelines such as electric power, communication, fuel gas, heating power, water supply, drainage, fire protection and the like, has the characteristics of concentration and concealment, can ensure that maintenance and maintenance of urban infrastructure are concentrated underground, does not occupy ground space and has no exposed line, and the convenience and safety of maintenance are improved. However, a large number of equipment and lines are concentrated in underground pipe galleries, which places high demands on the monitoring of the internal safety of the galleries.
At present, the comprehensive management of underground pipe galleries in China still depends on manual inspection, real-time monitoring and prevention of risks cannot be realized, inspection efficiency is low, and great potential safety hazards exist. In practice, environmental data such as voltage, methane concentration and temperature in the underground pipe gallery can be collected to monitor the internal condition of the pipe gallery, and whether the environment is abnormal or not can be judged by detecting abnormality of the time sequence of the data, so that the early warning of the safety state of the pipe gallery is realized.
The fluctuation mode of the modeling sequence of the long-short-term memory network (Long Short Term Memory, abbreviated as LSTM) can predict the attribute value at a certain moment, and then the residual error between the predicted value and the actual value is utilized to detect the abnormality, but the abnormality aims at one-dimensional data, and the time change condition of one-dimensional features is captured. For multi-dimensional data, a convolution long-short-term memory network (Convolutional LSTM, conv-LSTM for short) is generally adopted, namely an LSTM method added with convolution, so as to capture multi-dimensional characteristics of the data. However, the spatial signature generated by the convolutional encoder depends in time on the previous time step, so the performance of Conv-LSTM may deteriorate with increasing sequence length. In order to solve the problem, the invention provides an underground pipe gallery abnormality detection network and method based on an attention mechanism, and the attention mechanism is added on the basis of a Conv-LSTM method, so that the method can adaptively select relevant hidden states on different time steps, further ensure that long-time abnormality detection has higher precision, and the detection effect is not deteriorated with time.
Disclosure of Invention
The invention aims to provide an underground pipe gallery abnormality detection network and method based on an attention mechanism, which can ensure high accuracy of long-time abnormality detection and have good detection effect on abnormal conditions under multiple data.
In a first aspect, the present invention relates to an underground pipe gallery abnormality detection network based on an attention mechanism, and the structure is shown in fig. 1, and the network comprises three modules: the device comprises an encoding module, a feature extraction module and a decoding module.
The coding module comprises four convolution layers, namely a first convolution layer, a second convolution layer, a third convolution layer and a fourth convolution layer, which are sequentially connected. The coding module is used for coding the input feature matrix.
The feature extraction module includes a Conv-LSTM layer and an attention mechanism layer. The Conv-LSTM layer comprises four Conv-LSTM layers of identical structure, namely a first Conv-LSTM layer, a second Conv-LSTM layer, a third Conv-LSTM layer and a fourth Conv-LSTM layer. The attention mechanism layer comprises four attention mechanisms with the same structure, namely a first attention mechanism, a second attention mechanism, a third attention mechanism and a fourth attention mechanism. The feature extraction module is used for extracting effective data features.
The decoding module comprises seven layers, namely a first deconvolution layer, a first connection layer, a second deconvolution layer, a second connection layer, a third deconvolution layer, a third connection layer and a fourth deconvolution layer, which are sequentially connected. The decoding module is used for decoding the extracted effective data features.
The first convolution layer in the encoding module is connected to the first Conv-LSTM of the Conv-LSTM layer of the feature extraction module, and so on.
The first Conv-LSTM of the Conv-LSTM layer in the feature extraction module is connected to the first attention mechanism of the attention mechanism layer, and so on.
The fourth attention mechanism of the attention mechanism layer of the feature extraction module is connected with the first deconvolution layer of the decoding module; the third attention mechanism of the attention mechanism layer of the feature extraction module is connected with the first connection layer of the decoding module; the second attention mechanism of the attention mechanism layer of the feature extraction module is connected with the second connection layer of the decoding module; the first attention mechanism of the attention mechanism layer of the feature extraction module is connected to the third connection layer of the decoding module.
In a second aspect of the present invention, an attention mechanism-based underground pipe gallery abnormality detection method is provided, where a flowchart is shown in fig. 2, and the method includes the following steps:
step one: calculating a correlation feature matrix
Normally, certain sensors in the piping lane interact to present a specific form of equilibrium. However, when a mutation occurs at a certain time, it is inevitable that some sensors are mutated, and then such an abnormality may be expressed in relation to different sensors. The correlation between the plurality of sensors (the plurality of data) is considered, and a feature function is constructed by the correlation to perform abnormality detection. Then the degree of correlation between the data variables at a certain time instant is calculated taking into account the degree of correlation, i.e. the degree of correlation of the time-series data, during a certain time instant before the time instant t.
Given a time period from the instant t-w to the instant t (positive integer), w is the time step (0<w<t), assuming that the current two sequences are
And->
The correlation between the two time data sequences is:
where denominator k is the scaling factor (k=w). Then will
As a matrix S t Is used for constructing an n x n feature matrix S by the elements of the ith row and the jth column t . Feature matrix S t Not only can shape similarity and value scale correlation between two time series be captured, but also robustness to input noise is achieved because some time series instability has little effect on the feature matrix. By selecting different w values, features at different scales can be obtained.
Step two: encoding feature matrix using full-reel encoder
Encoding feature matrix S using a full-reel encoder t Spatial mode of (c). Specifically, S will be at different scales t Connected into tensor χ t,0 It is then fed to four convolutional layers, which are applied to the feature matrix S using a full-reel encoder t Further encoding is performed for the spatial mode of (c). Through the process ofFour-layer convolution kernel, which is used for integrating the original characteristic matrix S t Coded into feature maps of different sizes, four layers in total. Let l represent a positive integer not less than 2, χ t,l-1 Representing the feature map of layer l-1, then the output value of layer l is:
χ t,l ＝f(W l *χ t,l-1 +b l )
where x represents convolution, f () represents activation function, W l B of the representation l Convolution kernel, b l Representing the deflectable term, χ t,l Representing the output feature map of the first layer.
Step three: conv-LSTM feature extraction based on attention mechanism
LSTM is a recurrent neural network for processing time series data, and can capture the correlation of historical time data, but the time change condition of one-dimensional characteristics is captured, so that the convolution can be added on the basis of LSTM to capture the time change condition of three-dimensional characteristics. Conv-LSTM combines convolutional neural networks (Convolutional Neural Networks, abbreviated CNN) and LSTM, so Conv-LSTM networks are able to extract both temporal and spatial information. The core idea of Conv-LSTM is to translate all inputs, hidden states and different gating values into a three-dimensional tensor. In addition, since Conv-LSTM is performed on each layer of feature map, if the time step is longer, the LSTM performance will be reduced, so that the attention mechanism is adopted to assign weights to the hidden features of each previous time step and provide information from each coding hidden state to the decoder.
And (3) respectively extracting the characteristic features on the four layers of characteristic diagrams obtained in the step one.
Specifically, given the output feature map χ of the first layer t,l And the previous hidden state H t-1,l Current hidden state H t,l ＝ConvLSTM(χ t,l ,H t-1,l ). Furthermore, it is contemplated that not all previous steps are associated with the current hidden state H t,l Equally relevant, the hidden states associated with the current step are adaptively selected using a time-awareness mechanism, and representations of these feature maps are aggregated to form a refined output of the feature mapsAnd (3) out:
wherein H is i,l Attention weight a for hidden state at any moment i For each time H i And H t The calculated attention mechanism coefficients (last moment):
where Vec represents a vector, χ represents a rescaling factor (χ=5).
Step four: reconstruction of feature maps using a deconvolution decoder
And finally, performing reverse reconstruction by using a deconvolution decoder, wherein the feature mapping extracted in the step three is required to be added during reconstruction, and finally, a reconstructed feature matrix is obtained
Step five: determining anomalies using a reconstruction error matrix
By reconstructing feature matrices
Subtracting the original feature matrix S t And (3) obtaining a reconstruction error matrix, comparing elements in the reconstruction error matrix with elements in a threshold matrix, and judging that an abnormal phenomenon exists if the number of the elements in the reconstruction error matrix is larger than that of the elements in the threshold matrix and exceeds a set value, namely, realizing abnormal detection.
Compared with the prior art, the invention not only improves the abnormality detection efficiency, but also has the following advantages:
(1) The long-time abnormality detection can be ensured to have higher precision, and the method is suitable for detecting the whole process of data evolution in a pipe gallery;
(2) The correlation characteristic matrix is calculated by utilizing the multiple data, and the method has good detection effect on abnormal conditions under the multiple data.
Drawings
FIG. 1 is a diagram of an underground pipe gallery anomaly detection network based on an attention mechanism;
FIG. 2 is a flow chart of a method for detecting anomalies in an underground pipe gallery based on an attention mechanism;
FIG. 3 is a schematic diagram of a utility tunnel sensor deployment;
fig. 4 is a graph of abnormality detection results.
Detailed Description
The invention provides an underground pipe gallery abnormality detection network and method based on an attention mechanism. The following describes specific embodiments of the present invention with reference to the drawings:
the deployment schematic diagram of the sensors in the underground pipe gallery refers to fig. 3, every 400 meters in the pipe gallery is a monitoring bin section, every 20 meters is deployed with one sensor, 20 sensors are needed in total, a core controller is connected with each sensor through CAN1 and CAN2, and data collected by different sensors are collected by the core controller, so that 20 time series data are generated.
Executing the first step: calculating a correlation feature matrix
In a specific embodiment three time scales are chosen: 10, 30, 50. I.e. each time step, there are three feature matrices. The size of each feature matrix is consistent with the dimension of the time series data. The correlation of the sensors with each other is calculated according to three time steps 10, 30, 50, respectively, and a 20 x 3 raw feature matrix is generated. Three-dimensional features such as image processing are formed, and one-dimensional data is stereoscopically formed. If there are 1400 time series, 20 sensors, w= {10, 30, 50}, there is 20×20×3 dimensional data per time point, and the input data dimension is 1400×30×30×3.
Executing the second step: encoding feature matrix using full-reel encoder
The feature size of the encoder input is 20×20×3, and the four feature sizes finally extracted are 20×20×32, 10×10× 64,6 ×6× 128,3 ×3×256, respectively. Multi-layer convolution is performed on the data at each time instant: the convolution step length of the first layer is 1, and the convolution kernel is 3; the convolution step length of the second layer is 2, and the convolution kernel is 3; the convolution step length of the third layer is 2, and the convolution kernel is 2; the fourth layer convolution step is 2 and the convolution kernel is 2. At the time of the third layer convolution, the feature size was already 6×6 small enough, and the one-more layer convolution was used to more fully capture the detailed information in the convolution layer.
Executing the third step: conv-LSTM feature extraction based on attention mechanism
And (3) respectively extracting the characteristic features on the four layers of characteristic diagrams obtained in the step one. Let l=1, 2,3,4.
Given the output feature map χ of the first layer t,l And the previous hidden state H t-1,l Current hidden state H t,l ＝ConvLSTM(χ t,l ,H t-1,l ). Furthermore, it is contemplated that not all previous steps are associated with the current hidden state H t,l Equally relevant, a time-attention mechanism is employed to adaptively select hidden states associated with the current step and aggregate representations of these feature maps to form refined outputs of feature maps:
wherein H is i,l Attention weight a for hidden state at any moment i For each time H i And H t The calculated attention mechanism coefficients (last moment):
where Vec represents a vector, χ represents a rescaling factor (χ=5).
Executing the fourth step: reconstruction of feature maps using a deconvolution decoder
Deconvolution to recover the initial matrix, following the reverse order, the first Conv-LSTM layer
To deconvolution neural networksA layer. The output feature map is concatenated with the output of the previous Conv-LSTM layer, allowing the decoder process to be stacked and further fed into the next deconvolution layer. Four deconvolution layers are used: the first deconvolution layer, the second deconvolution layer, the third deconvolution layer, and the fourth deconvolution layer are provided with 128 cores of size 2×2×256, 64 cores of size 2×02×1128, 32 cores of size 3×23×364, and 3 cores of size 3×43×564 in order, and steps of 2×62, 2×72, 2×82, and 1×91, respectively. The first deconvolution test, the second deconvolution layer, the third deconvolution layer, and the fourth deconvolution layer result in feature maps of sizes 6×6×128, 10×10×64, 20×20×32, and 20×20×3, respectively.
Executing the fifth step: determining anomalies using a reconstruction error matrix
By reconstructing feature matrices
Subtracting the original feature matrix S t And (3) obtaining a reconstruction error matrix, comparing elements in the reconstruction error matrix with elements in a threshold matrix, and judging that an abnormal phenomenon exists if the number of the elements in the reconstruction error matrix is larger than that of the elements in the threshold matrix and exceeds a set value, namely, realizing abnormal detection. Fig. 4 shows the abnormality detection result in which the broken line is a threshold value and the gray line is an abnormality judgment mark, and it can be seen that four abnormalities occur in the piping lane.
After the occurrence of the abnormality is judged in the practical application, early warning can be carried out, and the monitoring is informed to optionally execute subsequent operations, such as controlling and opening a fan of an executing device, reducing the temperature, humidity, smoke concentration and combustible gas concentration in a pipe gallery and improving the oxygen concentration, and adjusting the internal environment of the pipe gallery to keep safe.
It should be appreciated that embodiments of the invention may also be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in non-transitory computer-readable memory. The methods may be implemented in a computer program using standard programming techniques, including a non-transitory computer readable storage medium configured with a computer program, where the storage medium so configured causes a computer to operate in a specific and predefined manner, in accordance with the methods and drawings described in the specific embodiments. Each program may be implemented in a high level procedural or object oriented programming language to communicate with a computer system. However, the program(s) can be implemented in assembly or machine language, if desired. In any case, the language may be a compiled or interpreted language. Furthermore, the program can be run on a programmed application specific integrated circuit for this purpose.
Further, the method may be implemented in any type of computing platform operatively connected to a suitable computing platform, including, but not limited to, a personal computer, mini-computer, mainframe, workstation, network or distributed computing environment, separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, and so forth. Aspects of the invention may be implemented in machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, optical read and/or write storage medium, RAM, ROM, etc., such that it is readable by a programmable computer, which when read by a computer, is operable to configure and operate the computer to perform the processes described herein. Further, the machine readable code, or portions thereof, may be transmitted over a wired or wireless network. When such media includes instructions or programs that, in conjunction with a microprocessor or other data processor, implement the steps described above, the invention described herein includes these and other different types of non-transitory computer-readable storage media. The invention also includes the computer itself when programmed according to the methods and techniques of the present invention.
The present invention is not limited to the above embodiments, but is merely preferred embodiments of the present invention, and the present invention should be construed as being limited to the above embodiments as long as the technical effects of the present invention are achieved by the same means. Various modifications and variations are possible in the technical solution and/or in the embodiments within the scope of the invention.
Claims (8)
1. Underground pipe gallery anomaly detection network based on attention mechanism, characterized by comprising:
the device comprises an encoding module, a feature extraction module and a decoding module.
2. The attention-based utility tunnel anomaly detection network of claim 1 wherein the encoding module comprises four convolutional layers:
a first convolution layer, a second convolution layer, a third convolution layer, and a fourth convolution layer;
the four convolution layers are connected in sequence.
3. The attention mechanism based underground pipe gallery anomaly detection network of claim 2, wherein the feature extraction module comprises:
Conv-LSTM layer and attention mechanism layer;
the Conv-LSTM layer comprises four Conv-LSTM layers with the same structure, namely a first Conv-LSTM layer, a second Conv-LSTM layer, a third Conv-LSTM layer and a fourth Conv-LSTM layer;
the attention mechanism layer comprises four attention mechanisms with the same structure, namely a first attention mechanism, a second attention mechanism, a third attention mechanism and a fourth attention mechanism.
4. The attention-based utility tunnel anomaly detection network of claim 3 wherein the decoding module comprises seven layers:
a first deconvolution layer, a first connection layer, a second deconvolution layer, a second connection layer, a third deconvolution layer, a third connection layer, and a fourth deconvolution layer;
the seven layers are connected in sequence.
5. The attention-based underground pipe gallery anomaly detection network of claim 4,
the first convolution layer in the coding module is connected with the first Conv-LSTM of the Conv-LSTM layer of the feature extraction module, and so on;
the first Conv-LSTM of the Conv-LSTM layer in the feature extraction module is connected with the first attention mechanism of the attention mechanism layer, and so on;
the fourth attention mechanism of the attention mechanism layer of the feature extraction module is connected with the first deconvolution layer of the decoding module; the third attention mechanism of the attention mechanism layer of the feature extraction module is connected with the first connection layer of the decoding module; the second attention mechanism of the attention mechanism layer of the feature extraction module is connected with the second connection layer of the decoding module; the first attention mechanism of the attention mechanism layer of the feature extraction module is connected to the third connection layer of the decoding module.
6. The underground pipe gallery abnormality detection network method based on the attention mechanism is characterized by comprising the following steps of:
step one: calculating a correlation feature matrix;
step two: encoding the feature matrix by using a full-coiler encoder;
step three: conv-LSTM feature extraction based on an attention mechanism;
step four: reconstructing the feature map using a deconvolution decoder;
step five: and judging abnormality by using the reconstruction error matrix.
7. A computer device, comprising: a memory for storing a computer program; a processor for implementing the method of claim 6 when executing the computer program.
8. A readable storage medium, characterized in that it has stored thereon a computer program which, when executed by a processor, implements the method according to claim 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310236950.3A CN116415197A (en) | 2023-03-13 | 2023-03-13 | Underground pipe gallery abnormality detection network and method based on attention mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310236950.3A CN116415197A (en) | 2023-03-13 | 2023-03-13 | Underground pipe gallery abnormality detection network and method based on attention mechanism |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116415197A true CN116415197A (en) | 2023-07-11 |
Family
ID=87052426
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310236950.3A Pending CN116415197A (en) | 2023-03-13 | 2023-03-13 | Underground pipe gallery abnormality detection network and method based on attention mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116415197A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117190078A (en) * | 2023-11-03 | 2023-12-08 | 山东省计算中心（国家超级计算济南中心） | Abnormality detection method and system for monitoring data of hydrogen transportation pipe network |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111880998A (en) * | 2020-07-30 | 2020-11-03 | 平安科技（深圳）有限公司 | Service system anomaly detection method and device, computer equipment and storage medium |
| CN115169430A (en) * | 2022-04-27 | 2022-10-11 | 北京理工大学 | Cloud network end resource multidimensional time sequence anomaly detection method based on multi-scale decoding |
-
2023
- 2023-03-13 CN CN202310236950.3A patent/CN116415197A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111880998A (en) * | 2020-07-30 | 2020-11-03 | 平安科技（深圳）有限公司 | Service system anomaly detection method and device, computer equipment and storage medium |
| CN115169430A (en) * | 2022-04-27 | 2022-10-11 | 北京理工大学 | Cloud network end resource multidimensional time sequence anomaly detection method based on multi-scale decoding |
Non-Patent Citations (2)
| Title |
|---|
| CHUXU ZHANG ET AL: "A Deep Neural Network for Unsupervised Anomaly Detection and Diagnosis in Multivariate Time Series Data", 《THE THIRTY-THIRD AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE》, vol. 33, no. 1, pages 1409 - 1416, XP055787183, DOI: 10.1609/aaai.v33i01.33011409 * |
| 陈敏 等: "基于注意力机制和稀疏图卷积的行人轨迹预测", 《激光与光电子学进展》, vol. 60, no. 10, pages 1 - 15 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117190078A (en) * | 2023-11-03 | 2023-12-08 | 山东省计算中心（国家超级计算济南中心） | Abnormality detection method and system for monitoring data of hydrogen transportation pipe network |
| CN117190078B (en) * | 2023-11-03 | 2024-02-09 | 山东省计算中心（国家超级计算济南中心） | Abnormality detection method and system for monitoring data of hydrogen transportation pipe network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112460495B (en) | Monitoring point layout method and system for leakage monitoring positioning | |
| CN111024158B (en) | Kitchen electrical hazard intelligent monitoring method combined with edge calculation | |
| CN112861071B (en) | High-speed rail traction system anomaly detection method based on depth self-coding | |
| CN110334948B (en) | Power equipment partial discharge severity evaluation method and system based on characteristic quantity prediction | |
| CN116415197A (en) | Underground pipe gallery abnormality detection network and method based on attention mechanism | |
| CN109681391B (en) | Blade root bolt fracture fault detection method and medium | |
| CN111784061B (en) | Training method, device and equipment for power grid engineering cost prediction model | |
| CN109299208B (en) | Intelligent visual risk assessment method for transmission tower in typhoon disaster | |
| CN112884008A (en) | Prediction evaluation method and device for running state of power information acquisition system | |
| CN110866682B (en) | Underground cable early warning method and device based on historical data | |
| CN116777088B (en) | Power supply emergency repair environment monitoring method and system for guaranteeing life safety | |
| CN113435103A (en) | Power distribution room abnormity detection method, system, server, edge gateway and medium | |
| CN111766513B (en) | Capsule network-based variable-working-condition multi-fault diagnosis method for three-phase induction motor | |
| CN116993537A (en) | Power load abnormality detection method and system based on serial GRU (generic routing unit) self-encoder | |
| CN117113202A (en) | Power loop energy consumption detection method and equipment based on joint error stacking model | |
| CN110188939B (en) | Wind power prediction method, system, equipment and storage medium of wind power plant | |
| CN111354497B (en) | Nuclear power station breach accident information diagnosis method | |
| CN117092581A (en) | Segment consistency-based method and device for detecting abnormity of electric energy meter of self-encoder | |
| CN116973703A (en) | Acoustic diagnosis method for discharge fault and abnormal operation state of dry type air-core reactor | |
| CN115423810A (en) | Blade icing form analysis method for wind generating set | |
| Li et al. | Multi-time-step and multi-parameter prediction for real-world proton exchange membrane fuel cell vehicles (PEMFCVs) toward fault prognosis and energy consumption prediction | |
| CN114565581A (en) | Detection method, recording medium and system for low-value insulator of distribution line | |
| CN116498435B (en) | Method and device for monitoring use state based on diesel generator set | |
| CN112699608B (en) | Time sequence repairing method suitable for data loss caused by sensor power failure | |
| CN114092322A (en) | Method and device for generating anomaly detection model and method and device for detecting anomaly |
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 |