CN111031064A - Method for detecting power grid false data injection attack - Google Patents
Method for detecting power grid false data injection attack Download PDFInfo
- Publication number
- CN111031064A CN111031064A CN201911357195.4A CN201911357195A CN111031064A CN 111031064 A CN111031064 A CN 111031064A CN 201911357195 A CN201911357195 A CN 201911357195A CN 111031064 A CN111031064 A CN 111031064A
- Authority
- CN
- China
- Prior art keywords
- time
- layer
- data
- input
- state estimation
- 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
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L63/00—Network architectures or network communication protocols for network security
- H04L63/14—Network architectures or network communication protocols for network security for detecting or protecting against malicious traffic
- H04L63/1408—Network architectures or network communication protocols for network security for detecting or protecting against malicious traffic by monitoring network traffic
- H04L63/1416—Event detection, e.g. attack signature detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L63/00—Network architectures or network communication protocols for network security
- H04L63/14—Network architectures or network communication protocols for network security for detecting or protecting against malicious traffic
- H04L63/1441—Countermeasures against malicious traffic
- H04L63/1466—Active attacks involving interception, injection, modification, spoofing of data unit addresses, e.g. hijacking, packet injection or TCP sequence number attacks
Abstract
The invention discloses a method for detecting false data injection attack of a power grid, which comprises the following steps: calculating to obtain a system running state estimated value according to data collected by a power grid terminal; extracting time-frequency characteristic data of each time interval from the state estimation value by using wavelet transformation; and inputting the time-frequency characteristic data into a deep neural network for classification of the existence/nonexistence of false data injection attack, and outputting a detection result. The method can be directly used for detecting the state estimation result of the power system, and the detection result can indicate whether the state estimation of the power system is influenced by the false data injection attack or not.
Description
Technical Field
The invention relates to a method for detecting whether a power grid has false data injection attack.
Background
The core link of the operation and control of the power system is state estimation, and the state estimation calculates the actual operation state of the power grid through power grid state information collected in a data collection and monitoring System (SCADA).
Malicious network attacks, such as False Data Injection Attack (FDIA), can tamper with the measurement data in the SCADA, thereby affecting the state estimation result of the smart grid, seriously threatening the stable operation of the grid, and even causing large-scale power outage loss, such as the grid attacks suffered by the recent ukraine and venezuela. How to detect whether the power grid has malicious attacks is a necessary premise for timely taking protective measures and ensuring the stable operation of the power grid.
FDIA is one of the conventional network attack modes with the greatest threat to power grid state estimation, and other attack modes include distributed denial of service attack, network congestion attack and the like, but only FDIA can bypass the traditional bad data identification mechanism based on residual error, and the method is repeatedly implemented and can avoid being detected by the existing detection program.
Disclosure of Invention
The invention provides a method for detecting false data injection attack of a power grid, which aims to: whether a false data injection attack exists in the power system state estimation is detected.
The technical scheme of the invention is as follows:
a method for detecting power grid false data injection attack comprises the following steps:
(1) calculating to obtain a system running state estimated value according to data collected by a power grid terminal;
(2) extracting time-frequency characteristic data from the state estimation value;
(3) and inputting the time-frequency characteristic data into a deep neural network for classification of the existence/nonexistence of false data injection attack, and outputting a detection result.
As a further improvement of the method: before the step (2) is implemented, residual error detection is carried out on the state estimation value, and if no bad data is found, the state estimation value is stored in a state historical database;
in step (2), the state estimation value is read from the state history database to extract the time-frequency characteristic data.
As a further improvement of the method: and (3) extracting time-frequency characteristic data by adopting wavelet transformation in the step (2).
As a further improvement of the method: the specific steps of the step (2) are as follows:
(2-1) taking state estimation value data of a plurality of time intervals, and then respectively taking state estimation values of a plurality of sampling moments for each time interval; the node voltage amplitude value and the phase angle estimation value obtained by the state estimation value at each sampling moment respectively form a signal vector x (t) corresponding to each sampling moment t, wherein the length of x (t) is 2M；
(2-2) for the signal vector x (t) at each sampling instant, at a scaling factor a-2jAnd the position parameter is b-2jPerforming discrete wavelet transform under the condition of x k, j, k being integers, thereby decomposing the signal vector x (t) into wavelet functions of at most M layers:
wherein phijk(t)＝2-j2φ(2-jt-k), phi (t) is a scale function, i.e. a parent wavelet function, phijk(t)＝2-j/2ψ(2- jt-k), ψ (t) is the mother wavelet function, ajkAnd djkAre corresponding approximation coefficients; obtaining a corresponding to each time tjkAnd djk；
(2-3) for each time interval, respectively calculating d of all sampling time moments in the time intervaljkAs the time-frequency characteristic data of the time period.
As a further improvement of the method: the deep neural network in the step (3) is a cyclic neural network, the neural network comprises a data input layer, a first GRU structural layer, a second GRU structural layer, a first full-connection layer and a second full-connection layer which are sequentially arranged, and the output of the upper layer is the input of the next layer;
d obtained for each time intervaljkMean value and djkVariance, sequentially input to the data in time orderAn input layer;
the GRU structural layer adopts a gate control cycle unit to extract the time sequence characteristics of input data; let { f }T-ω-1,...,fTRepresents the input of GRU structural layer, { gT-ω-1,...,gTRepresents the output of the GRU structure layer, the relationship between input and output is:
zT＝sigm(ωfzfT+ωgzgT-1+bz)，
rT＝sigm(ωfrfT+ωgrgT-1+br)；
wherein f isTRepresents d corresponding to T periodjkMean value and djkA two-dimensional vector formed by the variances; gTDenotes fTAn output obtained after the input; omega is a constant set artificially, and the numerical value is larger when the time interval quantity is larger; sigm stands for sigmoid function used as activation function; omegafg、ωgg、bg、ωfz、ωgz、bz、ωfr、ωgrAnd brAll the parameters are undetermined parameters obtained through training and learning;
the relationship between the input and output of the fully connected layer is:
y＝actv(ωd×x+bd)；
x and y represent the input and output of the full link layer, respectively; actv is the activation function; omegadAnd bdAll the parameters are undetermined parameters obtained through training and learning;
the output y of the second fully connected layer is the output of the neural network, and a value of 0 represents no attack, and a value of 1 represents an attack.
Compared with the prior art, the invention has the following beneficial effects: (1) the method can be directly used for detecting the state estimation result of the power system, the detection result can indicate whether the state estimation of the power system is influenced by the false data injection attack or not, and measures can be taken in timeA foundation is laid for ensuring the stable operation of the power grid; (2) the method uses all the time djkCompared with the method that the approximate coefficient is directly adopted to represent the time-frequency characteristic of the state estimation value, the method has the advantages that the mean value and the variance of the state estimation value are used as the time-frequency characteristic of the input signal vector, so that the problem of overlarge data volume is avoided, the calculated amount is obviously reduced, and the detection efficiency is improved; (3) the neural network uses a multilayer gating circulation unit, the GRU has the advantages of simple structure, less training time and quick convergence, and the double-layer GRU can extract more abstract features from input data; (4) although the time-space characteristics are more abstract after two layers of GRUs, excessive parameter training can cause the overfitting problem, the method further adopts a full connection layer, a random elimination mechanism is introduced, and partial characteristics are randomly eliminated to prevent overfitting.
Drawings
FIG. 1 is a flow chart of an embodiment of the method.
FIG. 2 is an architecture diagram of a neural network used in the present method.
Detailed Description
The technical scheme of the invention is explained in detail in the following with the accompanying drawings:
the false data injection attack detection model provided by the invention is divided into two links of training and detection. The training aims to determine relevant parameters in the deep neural network so as to realize online detection of false data injection attacks. Therefore, the two links are similar in process, and the difference lies in that the unknown parameters in the deep neural network are determined by utilizing the training link, and the trained deep neural network is utilized to carry out online detection in the detection link, namely the two classification problems of the existence/nonexistence of false data injection attack are realized.
Fig. 1 is a flowchart of a method for detecting a false data injection attack according to the present invention. And the power grid terminal uploads the collected running state information to a state estimation module in the SCADA, a system running state estimation value is obtained through calculation, and if no bad data is found in a residual error detection link, the estimation value is stored in a state historical database. The dotted line frame is the core of the invention, FDIA detection step: taking a plurality of recent time periods from a state historical database, taking state estimation values of a plurality of sampling moments in each time period, extracting time-frequency characteristics through wavelet transformation, storing the characteristic data of each time period into a characteristic historical database, extracting the characteristic data of the recent time periods from the characteristic historical database, inputting the characteristic data into an attack detection link based on a deep neural network, classifying whether false data is injected into the attack or not, and outputting a detection result.
The method comprises the following specific steps:
(1) and calculating to obtain a system running state estimated value according to data collected by the power grid terminal.
Residual error detection is carried out on the state estimation value, and if no bad data is found, the state estimation value is stored in a state history database;
(2) reading a state estimation value from a state historical database, and extracting time-frequency characteristic data by adopting wavelet transformation:
(2-1) taking state estimation value data of a plurality of time intervals, and then respectively taking state estimation values of a plurality of sampling moments for each time interval; the node voltage amplitude value and the phase angle estimation value obtained by the state estimation value at each sampling moment respectively form a signal vector x (t) corresponding to each sampling moment t, wherein the length of x (t) is 2M。
(2-2) for the signal vector x (t) at each sampling instant, at a scaling factor a-2jAnd the position parameter is b-2jDiscrete Wavelet Transform (DWT) is performed at x k, j, k being integers, so that the signal vector x (t) is decomposed into wavelet functions of at most M layers:
wherein phijk(t)＝2-j/2φ(2-jt-k), phi (t) is a scale function, i.e. a parent wavelet function, phijk(t)＝2-j/2ψ(2-jt-k), ψ (t) is the mother wavelet function, ajkAnd djkAre corresponding approximation coefficients; obtaining a corresponding to each time tjkAnd djk。
(2-3) if the time-frequency characteristic of the state estimation value is directly represented by the approximate coefficient, the data quantity is too hugeTo a problem of (a). The invention adopts the statistical characteristics of the approximate coefficient in the wavelet transformation as the input of the following neural network so as to reduce the data amount and the calculated amount and keep the characteristics in the data. The method comprises the following steps: for each time interval, d of all sampling moments in the time interval is respectively obtainedjkAs the time-frequency characteristic data of the time period.
(3) Inputting the time-frequency characteristic data into a deep neural network for classification of false data injection attack or not, and outputting a detection result:
as shown in fig. 2, the deep neural network is a recurrent neural network, and the neural network includes a data input layer, a first GRU structural layer, a second GRU structural layer, a first full-link layer, and a second full-link layer, which are sequentially arranged, where an output of a previous layer is an input of a next layer.
1) Data input layer
D obtained for each time intervaljkMean value and djkAnd the variances are sequentially input to the data input layer according to the time sequence.
2) First GRU structural layer
The GRU structure layer adopts a Gated Recurrent Unit (GRU) to extract the time sequence characteristics of input data, the GRU is a variant of a long-time and short-time memory network, the structure is simpler, and the GRU structure layer has the advantages of less time consumption and quick convergence when a large amount of data are trained.
In the present invention, the GRU is used to extract timing characteristics of the data input layer.
Let { f }T-ω-1,...,fTRepresents the input of GRU structural layer, { gT-ω-1,...,gTRepresents the output of the GRU structure layer, the relationship between input and output is:
zT＝sigm(ωfzfT+ωgzgT-1+bz)，
rT＝sigm(ωfrfT+ωgrgT-1+br)；
wherein f isTRepresents d corresponding to T periodjkMean value and djkA two-dimensional vector formed by the variances; gTDenotes fTAn output obtained after the input; omega is an artificially set constant, and the larger the constant is, the more sampling time periods are represented; sigm stands for sigmoid function used as activation function; the rest parameters are as follows: omegafg、ωgg、bg、ωfz、ωgz、bz、ωfr、ωgrAnd brAll the parameters are undetermined parameters obtained by training and learning.
3) Second GRU structural layer
The layer has the same structure and calculation mode as the previous layer. The layer takes the output of the GRU of the previous layer as the input of the GRU of the layer. More abstract features can be extracted from input data by adopting multiple layers of GRUs.
4) First full connection layer
The time-space characteristics of the input of the data input layer are more abstract after the input passes through two layers of GRUs, but excessive parameter training can cause an overfitting problem. A random elimination mechanism is introduced into the layer, and partial features are eliminated randomly to prevent over-fitting.
The fully-connected layer can realize the mapping from the characteristics to the judgment result, and the input-output relation of the fully-connected layer is represented by the following formula:
y＝actv(ωd×x+bd)；
x and y represent the input and output of the full link layer, respectively; actv is an activation function, preferably softmax; omegadAnd bdAll are undetermined parameters obtained through training and learning.
5) Second full connection layer
On the basis of the previous layer, partial features are continuously and randomly removed to prevent overfitting, and a detection result is output, wherein the structure of the detection result is the same as that of the previous layer. The output y of the second fully connected layer is the output of the neural network, and a value of 0 represents no attack, and a value of 1 represents an attack.
The following illustrates the implementation of the above method in this example:
the present embodiment employs an IEEE118 node system to verify the validity of the FDIA detection method proposed herein. 200000 times of operation conditions of the system are randomly generated, FDIA with known global topology is injected under 100 conditions, FDIA with known partial topology is injected under 400 conditions, and the training set and the test set are as follows: 1 for allocation. The calculation condition is a dual-core computer with Intel Pentium G32603.3GHz CPU and 4G memory.
As shown in table 1, the analysis results confirmed the effectiveness of the FDIA detection method proposed herein. The accuracy of the FDIA detection method herein reaches over 90% regardless of the test set or the training set, indicating that the DNN herein avoids the overfitting problem. The positive error rate in table 1 represents that the original case without attack is mistaken as having an attack, and the negative error rate is opposite to the positive error rate. From the aspect of training time, the training time required for the DNN to be 7633s is the most time-consuming part in the whole process, so the training part of the DNN can only be performed off-line, and the on-line detection is extremely short and only needs 12 ms.
The detection performance of the FDIA detection method mentioned in Table 1 is as follows:
Claims (5)
1. a method for detecting power grid false data injection attack is characterized by comprising the following steps:
(1) calculating to obtain a system running state estimated value according to data collected by a power grid terminal;
(2) extracting time-frequency characteristic data from the state estimation value;
(3) and inputting the time-frequency characteristic data into a deep neural network for classification of the existence/nonexistence of false data injection attack, and outputting a detection result.
2. The method of detecting a grid false data injection attack as claimed in claim 1, wherein: before the step (2) is implemented, residual error detection is carried out on the state estimation value, and if no bad data is found, the state estimation value is stored in a state historical database;
in step (2), the state estimation value is read from the state history database to extract the time-frequency characteristic data.
3. The method of detecting a grid false data injection attack as claimed in claim 1, wherein: and (3) extracting time-frequency characteristic data by adopting wavelet transformation in the step (2).
4. The method for detecting the power grid false data injection attack as claimed in claim 3, wherein the specific steps of the step (2) are as follows:
(2-1) taking state estimation value data of a plurality of time intervals, and then respectively taking state estimation values of a plurality of sampling moments for each time interval; the node voltage amplitude value and the phase angle estimation value obtained by the state estimation value at each sampling moment respectively form a signal vector x (t) corresponding to each sampling moment t, wherein the length of x (t) is 2M；
(2-2) for the signal vector x (t) at each sampling instant, at a scaling factor a-2jAnd the position parameter is b-2jPerforming discrete wavelet transform under the condition of x k, j, k being integers, thereby decomposing the signal vector x (t) into wavelet functions of at most M layers:
wherein phijk(t)＝2-j/2φ(2-jt-k), phi (t) is a scale function, i.e. a parent wavelet function, phijk(t)＝2-j/2ψ(2-jt-k), ψ (t) is the mother wavelet function, ajkAnd djkAre corresponding approximation coefficients; obtaining a corresponding to each time tjkAnd djk；
(2-3) for each time interval, respectively calculating d of all sampling time moments in the time intervaljkAs the time-frequency characteristic data of the time period.
5. The method for detecting the power grid false data injection attack as claimed in claim 4, wherein the deep neural network in the step (3) is a recurrent neural network, the neural network comprises a data input layer, a first GRU structural layer, a second GRU structural layer, a first full connection layer and a second full connection layer which are arranged in sequence, and the output of the previous layer is the input of the next layer;
d obtained for each time intervaljkMean value and djkThe variance is sequentially input to the data input layer according to the time sequence;
the GRU structural layer adopts a gate control cycle unit to extract the time sequence characteristics of input data; let { f }T-ω-1,...,fTRepresents the input of GRU structural layer, { gT-ω-1,...,gTRepresents the output of the GRU structure layer, the relationship between input and output is:
zT＝sigm(ωfzfT+ωgzgT-1+bz)，
rT＝sigm(ωfrfT+ωgrgT-1+br)；
wherein f isTRepresents d corresponding to T periodjkMean value and djkA two-dimensional vector formed by the variances; gTDenotes fTAn output obtained after the input; omega is a constant set artificially, and the numerical value is larger when the time interval quantity is larger; sigm stands for sigmoid function used as activation function; omegafg、ωgg、bg、ωfz、ωgz、bz、ωfr、ωgrAnd brAll the parameters are undetermined parameters obtained through training and learning;
the relationship between the input and output of the fully connected layer is:
y＝actv(ωd×x+bd)；
x and y represent the input and output of the full link layer, respectively; actv is the activation function; omegadAnd bdAll the parameters are undetermined parameters obtained through training and learning;
the output y of the second fully connected layer is the output of the neural network, and a value of 0 represents no attack, and a value of 1 represents an attack.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911357195.4A CN111031064A (en) | 2019-12-25 | 2019-12-25 | Method for detecting power grid false data injection attack |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911357195.4A CN111031064A (en) | 2019-12-25 | 2019-12-25 | Method for detecting power grid false data injection attack |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111031064A true CN111031064A (en) | 2020-04-17 |
Family
ID=70213199
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911357195.4A Pending CN111031064A (en) | 2019-12-25 | 2019-12-25 | Method for detecting power grid false data injection attack |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111031064A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113765880A (en) * | 2021-07-01 | 2021-12-07 | 电子科技大学 | Power system network attack detection method based on space-time correlation |
| CN116232742A (en) * | 2023-03-08 | 2023-06-06 | 中国信息通信研究院 | False data attack detection method and system based on state estimation |
| CN116886355A (en) * | 2023-07-03 | 2023-10-13 | 华北电力大学 | DDOS and false data injection collaborative attack optimization method of power system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107016236A (en) * | 2017-03-23 | 2017-08-04 | 新疆电力建设调试所 | Power network false data detection method for injection attack based on non-linear measurement equation |
| US20180176249A1 (en) * | 2016-12-21 | 2018-06-21 | Abb Inc. | System and method for detecting false data injection in electrical substations |
| CN110365647A (en) * | 2019-06-13 | 2019-10-22 | 广东工业大学 | A kind of false data detection method for injection attack based on PCA and BP neural network |
-
2019
- 2019-12-25 CN CN201911357195.4A patent/CN111031064A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180176249A1 (en) * | 2016-12-21 | 2018-06-21 | Abb Inc. | System and method for detecting false data injection in electrical substations |
| CN107016236A (en) * | 2017-03-23 | 2017-08-04 | 新疆电力建设调试所 | Power network false data detection method for injection attack based on non-linear measurement equation |
| CN110365647A (en) * | 2019-06-13 | 2019-10-22 | 广东工业大学 | A kind of false data detection method for injection attack based on PCA and BP neural network |
Non-Patent Citations (6)
| Title |
|---|
| DEBI PRASAD MISHRA,ET.AL: "《A Combined Wavelet and Data-Mining Based Intelligent Protection Scheme for Microgrid》", 《IEEE TRANSACTIONS ON SMART GRID》 * |
| 刘明宇等: "《基于深度学习的交通流量预测》", 《系统仿真学报》 * |
| 刘松等: "《基于门控递归单元神经网络的高速公路行程时间预测》", 《应用数学和力学》 * |
| 李元诚等: "《检测电网假数据注入攻击的方法》", 《电力系统自动化》 * |
| 雨霖LING: "《解决RNN的梯度消失问题》", 《机器学习算法与自然语言处理-公众号》 * |
| 高扬: "《循环神经网络》", 《人工智能与机器人先进技术丛书 智能摘要与深度学习》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113765880A (en) * | 2021-07-01 | 2021-12-07 | 电子科技大学 | Power system network attack detection method based on space-time correlation |
| CN116232742A (en) * | 2023-03-08 | 2023-06-06 | 中国信息通信研究院 | False data attack detection method and system based on state estimation |
| CN116232742B (en) * | 2023-03-08 | 2023-10-24 | 中国信息通信研究院 | False data attack detection method, system, electronic equipment and medium based on state estimation |
| CN116886355A (en) * | 2023-07-03 | 2023-10-13 | 华北电力大学 | DDOS and false data injection collaborative attack optimization method of power system |
| CN116886355B (en) * | 2023-07-03 | 2024-01-23 | 华北电力大学 | DDOS and false data injection collaborative attack optimization method of power system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109308522B (en) | GIS fault prediction method based on recurrent neural network | |
| CN111612651B (en) | Abnormal electric quantity data detection method based on long-term and short-term memory network | |
| CN111031064A (en) | Method for detecting power grid false data injection attack | |
| CN108830127A (en) | A kind of rotating machinery fault feature intelligent diagnostic method based on depth convolutional neural networks structure | |
| CN109814523A (en) | Method for diagnosing faults based on CNN-LSTM deep learning method and more attribute time series datas | |
| CN111751133B (en) | Intelligent fault diagnosis method of deep convolutional neural network model based on non-local mean embedding | |
| CN113343633A (en) | Thermal runaway fault classification and risk prediction method and system for power lithium battery | |
| CN111352977B (en) | Time sequence data monitoring method based on self-attention bidirectional long-short term memory network | |
| CN112101480A (en) | Multivariate clustering and fused time sequence combined prediction method | |
| CN103103570B (en) | Based on the aluminium cell condition diagnostic method of pivot similarity measure | |
| CN114239725B (en) | Electric larceny detection method for data poisoning attack | |
| CN112308124B (en) | Intelligent electricity larceny prevention method for electricity consumption information acquisition system | |
| CN113866455A (en) | Bridge acceleration monitoring data anomaly detection method, system and device based on deep learning | |
| Li et al. | Intelligent fault diagnosis of aeroengine sensors using improved pattern gradient spectrum entropy | |
| CN110650124A (en) | Network flow abnormity detection method based on multilayer echo state network | |
| CN110610203A (en) | Electric energy quality disturbance classification method based on DWT and extreme learning machine | |
| CN116400168A (en) | Power grid fault diagnosis method and system based on depth feature clustering | |
| CN115618202A (en) | Mechanical fault diagnosis method based on manifold embedding and key feature selection | |
| Hu et al. | Research on the fault identification method of oil pumping unit based on residual network | |
| CN115146842A (en) | Multivariate time series trend prediction method and system based on deep learning | |
| CN114841063A (en) | Aero-engine residual life prediction method based on deep learning | |
| CN115062686A (en) | Multi-KPI (Key performance indicator) time sequence abnormity detection method and system based on multi-angle features | |
| CN111653087A (en) | Construction method of urban traffic anomaly detection model | |
| Wang et al. | Construction and analysis of cross-layer aggregation neural network for AMI intrusion detection | |
| CN117290756B (en) | Power transmission line fault identification method and device based on federal learning and electronic equipment |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200417 |