CN113268729B - Smart grid attack positioning method based on convolutional neural network - Google Patents
Smart grid attack positioning method based on convolutional neural network Download PDFInfo
- Publication number
- CN113268729B CN113268729B CN202110486649.9A CN202110486649A CN113268729B CN 113268729 B CN113268729 B CN 113268729B CN 202110486649 A CN202110486649 A CN 202110486649A CN 113268729 B CN113268729 B CN 113268729B
- Authority
- CN
- China
- Prior art keywords
- data
- representing
- convolution
- layer
- measurement
- 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.)
- Active
Links
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 42
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000005259 measurement Methods 0.000 claims abstract description 52
- 238000001514 detection method Methods 0.000 claims abstract description 23
- 238000012549 training Methods 0.000 claims abstract description 10
- 238000007781 pre-processing Methods 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 10
- 238000013507 mapping Methods 0.000 claims description 9
- 230000004807 localization Effects 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 6
- 238000011478 gradient descent method Methods 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 238000004891 communication Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000003211 malignant effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/50—Monitoring users, programs or devices to maintain the integrity of platforms, e.g. of processors, firmware or operating systems
- G06F21/55—Detecting local intrusion or implementing counter-measures
- G06F21/554—Detecting local intrusion or implementing counter-measures involving event detection and direct action
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- 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
- G06N3/048—Activation functions
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S40/00—Systems for electrical power generation, transmission, distribution or end-user application management characterised by the use of communication or information technologies, or communication or information technology specific aspects supporting them
- Y04S40/20—Information technology specific aspects, e.g. CAD, simulation, modelling, system security
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Business, Economics & Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Economics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Computer Security & Cryptography (AREA)
- Human Resources & Organizations (AREA)
- Tourism & Hospitality (AREA)
- Water Supply & Treatment (AREA)
- Computer Hardware Design (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Public Health (AREA)
- General Business, Economics & Management (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Radar Systems Or Details Thereof (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a smart grid attack positioning method based on a convolutional neural network, which comprises the steps of firstly acquiring measurement data based on a nonlinear measurement equation of a smart grid under an alternating current model, transmitting the data to a bad data detection device for data preprocessing and removing bad data, and taking the preprocessed measurement data as a training sample; then, describing the attack positioning problem of the intelligent power grid as a multi-label classification problem, and extracting the intrinsic high-dimensional characteristics of the measured data under the communication model by adopting a convolutional neural network; finally, the trained convolutional neural network is used as a multi-label classifier, and the multi-label classifier is embedded into a bad data detection device to realize real-time online attack positioning. The invention can realize attack positioning of the intelligent power grid and has the following advantages which are not possessed by the prior art: the method and the device can acquire the nonlinear characteristics of the intelligent power grid more accurately, not only can detect the power grid attack, but also can acquire higher attack positioning accuracy and higher detection efficiency.
Description
Technical Field
The invention particularly relates to a smart grid attack positioning method based on a convolutional neural network, and belongs to the technical field of power system attack positioning.
Background
With the progress of network communication technology, the reliability of the operation of the smart grid is remarkably enhanced, but due to the dependence on data communication, the smart grid is easy to suffer from various malignant network attacks, and especially the hidden network attacks can avoid the traditional bad data detection device, and error data is injected into the power grid system so as to destroy the normal and stable operation of the power grid. However, most methods only consider the detection of a hidden attack and do not consider the localization of the attack. Therefore, in order to timely remove the attack and ensure the stable and safe operation of the system, how to quickly and accurately locate the attack of the smart grid is becoming more and more important in academia and industry.
Currently, the main stream technology of the smart grid attack positioning method mainly comprises the following steps: (1) Based on traditional machine learning and deep learning, the intelligent power grid attack positioning under the direct current model obtains good detection effect, but the direct current model-based method has larger deviation when the actual power grid is applied in consideration of the fact that the actual power grid has complex nonlinear characteristics, so that the detection precision is greatly reduced; (2) The attack positioning method based on the Kalman filter bank depends on a power grid model, and has the defects of use on the premise of detecting attack, insufficient anti-interference capability and the like.
Compared with the traditional neural network, the convolutional neural network can extract the inherent nonlinear characteristics from the data, has been successfully applied to the classification problem of images and voices, but has few applications in the attack positioning problem of smart grid under an alternating current model. The smart grid attack localization problem can be essentially converted into a typical multi-tag classification problem. However, in the prior art, the problem of smart grid attack positioning under the alternating current model is rarely researched and developed from the perspective of multi-label classification.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a smart grid attack positioning method based on a convolutional neural network, which is simpler to implement without depending on model precision compared with positioning methods based on a grid model and the like.
The aim of the invention is realized by the following technical scheme: a smart grid attack positioning method based on a convolutional neural network comprises the following steps:
step 1: collecting network characteristic parameters of a smart grid, including a power grid topological structure, line parameters and power loads, and obtaining measurement data according to a nonlinear measurement equation of the smart grid, wherein the measurement data comprises a voltage phase angle and a voltage amplitude of a reference node, an injected active power and an injected reactive power of each node, and an active power and a reactive power of each branch, wherein the measurement data Z is normal o And measurement data Z subject to attack a The specific description is as follows:
Z o ＝h(x o )+ω (1)
Z a ＝Z+h(x a )-h(x o ) (2)
wherein h represents a nonlinear measurement equation, i.e., a functional relationship between system state and measurement truth, x o Representing the system state includes the true values of the voltage amplitude and voltage phase angle of all nodes of the system, x a Representing system state values under attack, x representing x o Or x a ，V k And theta k Respectively representing the voltage amplitude and the voltage phase angle of the node k, V m And theta m Respectively representing the voltage amplitude and the voltage phase angle of the node m, P k And Q k Respectively represent the injected active power and reactive power of the node k, P km And Q km Representing the active power and the reactive power of the line between the node k and the node M respectively, M representing the set of the node M, g km And b km Representing the conductance and susceptance, delta, between nodes k and m, respectively km ＝θ k -θ m Representing the phase angle difference between node k and node m, ω observe noise.
Step 2: data preprocessing, let d= [ Z ] o ,Z a ]The measurement data obtained in the step (1) are stored in D, then the data set D is transmitted to a smart grid bad data detection device, and the ith group of data D in the D is calculated according to the formula (4) i Residual R of (2) i I=1, 2, …, N represents the number of data groups, and the detection bad data threshold parameter λ is set and judged, if R i >Lambda, reject D i The method comprises the steps of carrying out a first treatment on the surface of the If R is i Less than or equal to lambda, D i Storing into a set T;
wherein,,representing the measured data D i And calculating the obtained system state estimated value.
Step 3: describing the smart grid attack positioning problem as a multi-label classification problem, and defining the characteristics (namely measurement data) of a training set asWhere j=1, 2, …, n represents the j-th measurement, n represents the feature number; define training set label (i.e. the category to which the measurement belongs) as +.>Wherein->A class of the j-th measurement representing the i-th group of data, if the j-th measurement is attacked, record +.>Otherwise, record->Defining the classification result of the multi-label classifier asWherein->Classification result of the j-th measurement representing the i-th group of data,/and/or>Or->If it isThen the j-th measurement representing the i-th set of data is attacked; if->Then the j-th measurement of the i-th set of data is not attacked;
step 4: the method comprises the following specific substeps of extracting intrinsic high-dimensional characteristics of measured data by adopting a Convolutional Neural Network (CNN), taking the CNN as a multi-label classifier:
(4.1) acquiring a layer 1 feature map based on the input data T according to formula (5):
c 1,u ＝ReLU(T*f 1,u +b 1,u ) (5)
wherein c 1,u A ith feature map representing layer 1 convolution, f 1,j A u-th convolution kernel representing a layer 1 convolution, b 1,u The u-th offset, representing the layer 1 convolution, represents the convolution operation, and ReLU represents the activation function, expressed specifically as:
(4.2) inputting implicit features generated by convolution in the (q-1) th convolution layer to the (q) th convolution layer according to a formula (7) to obtain a (q) th layer feature map;
c q,v ＝ReLU(c q-1,v *f q,v +b q,v ) (7)
wherein c q,v And c q-1,v Respectively represent the (th) and (th) -1 th convolution layers (v) th convolution kernels, f q,v A v-th convolution kernel representing a q-th layer convolution, b q,v A v-th offset representing a q-th layer convolution, q being an integer greater than 2;
(4.3) let q=q+1, repeat step (4.2) until q>q max Output the (q) max Feature mapping of convolutional layers, where q max Indicating the maximum number of convolution layers;
(4.4) inputting the q-th according to the formula (8) max Pooling is carried out on the feature mapping of the convolution layer;
wherein, maxpool β,γ Representing the use of beta x gamma region pairsAnd (5) taking the maximum value.
(4.5) mapping c the features extracted by the pooling layer according to equation (9) p Corresponding output characteristics are obtained through flattening layer operation;
c F ＝ReLU(w F *c p +b F ) (9)
wherein c F Feature map representing flattened layers, w F Representing weights of the flattened layers, b F Indicating the bias of the flattened layer.
(4.6) transmitting the output characteristics of the flattening layer as input characteristics to the fully connected layer according to the formula (10) to obtain a final classification result:
wherein w is D And b D Respectively representing the weight and bias of the fully connected layer, and sigmoid represents an activation function, which is specifically expressed as follows:
step 5: training CNN according to a small batch gradient descent method, taking a minimized cross entropy loss function as an optimization target, determining optimal learning parameters of CNN,
wherein s= { i 1 ,i 2 ,…,i E -representing a data set selected for each small lot, E representing the number of sets of data;
step 6: after the trained CNN is embedded into the intelligent power grid bad data detection device, real-time online attack positioning is realized, and extraction is performedAll the corresponding measurements are the attacked positions.
The invention has the following effective effects:
1. compared with attack positioning of the intelligent power grid based on a direct current model, the intelligent power grid attack positioning method based on the convolutional neural network can more accurately represent nonlinear characteristics of an actual intelligent power grid, and achieves higher attack positioning accuracy and higher detection efficiency.
2. Compared with a method based on a model, the smart grid attack positioning method based on the convolutional neural network is simpler and more efficient to implement, can detect whether the smart grid is attacked or not, can position the attack, and does not depend on the model.
Drawings
FIG. 1 is a block diagram of an IEEE-118 node power grid system in accordance with an embodiment of the present invention;
fig. 2 is a schematic diagram of functional modules of a smart grid attack positioning method;
fig. 3 is a CNN architecture diagram for smart grid attack localization under an ac model.
Detailed Description
The objects and effects of the present invention will become more apparent from the following description of the present invention when taken in conjunction with the accompanying drawings.
Example 1:
fig. 1 is a block diagram of an IEEE 118 node grid system. Taking the system as an example, the state estimation method provided by the invention is adopted for implementation;
FIG. 2 is a schematic diagram of a smart grid attack localization method based on convolutional neural networks;
fig. 3 is a CNN architecture diagram for smart grid attack localization under an ac model;
a smart grid attack positioning method based on a convolutional neural network comprises the following steps:
step 1: collecting network characteristic parameters of the intelligent power grid, including a power grid topological structure, line parameters and power loads, and obtaining measurement data according to a nonlinear measurement equation of the intelligent power grid, wherein the measurement data comprise voltage phase angles and voltage amplitude values of reference nodes, injected active power and reactive power of each node, and the parameters comprise the voltage phase angles and the voltage amplitude values of the reference nodesActive and reactive power of branch, wherein data Z is measured normally o And measurement data Z subject to attack a The specific description is as follows:
Z o ＝h(x o )+ω (1)
Z a ＝Z+h(x a )-h(x o ) (2)
wherein h represents a nonlinear measurement equation, i.e., a functional relationship between system state and measurement truth, x o Representing the system state includes the true values of the voltage amplitude and voltage phase angle of all nodes of the system, x a Representing system state values under attack, x representing x o Or x a ，V k And theta k Respectively representing the voltage amplitude and the voltage phase angle of the node k, V m And theta m Respectively representing the voltage amplitude and the voltage phase angle of the node m, P k And Q k Respectively represent the injected active power and reactive power of the node k, P km And Q km Representing the active power and the reactive power of the line between the node k and the node M respectively, M representing the set of the node M, g km And b km Representing the conductance and susceptance, delta, between nodes k and m, respectively km ＝θ k -θ m Representing the phase angle difference between node k and node m, ω observe noise.
Step 2: data preprocessing, let d= [ Z ] o ,Z a ]The measurement data obtained in the step (1) are stored in D, then the data set D is transmitted to a smart grid bad data detection device, and the ith group of data D in the D is calculated according to the formula (4) i Residual R of (2) i I=1, 2, …, N represents the number of data groups, and the detection bad data threshold parameter λ=10 is set and judged, if R i >Lambda, reject D i The method comprises the steps of carrying out a first treatment on the surface of the If R is i Less than or equal to lambda, D i Storing into a set T;
wherein,,representing the measured data D i And calculating the obtained state estimation value.
Step 3: describing the smart grid attack positioning problem as a multi-label classification problem, and defining the characteristics (namely measurement data) of a training set asWhere j=1, 2, …, n represents the j-th measurement, n represents the feature number; define training set label (i.e. the category to which the measurement belongs) as +.>Wherein->A class of the j-th measurement representing the i-th group of data, if the j-th measurement is attacked, record +.>Otherwise, record->Defining the classification result of the multi-label classifier asWherein->Classification result of the j-th measurement representing the i-th group of data,/and/or>Or->If it isThen the j-th measurement representing the i-th set of data is attacked; if->Then the j-th measurement of the i-th set of data is not attacked;
step 4: the method comprises the following specific substeps of extracting intrinsic high-dimensional characteristics of measured data by adopting a Convolutional Neural Network (CNN), taking the CNN as a multi-label classifier:
(4.1) acquiring a layer 1 feature map based on the input data T according to formula (5):
c 1,u ＝ReLU(T*f 1,u +b 1,u ) (5)
wherein c 1,u A ith feature map representing layer 1 convolution, f 1,j A u-th convolution kernel representing a layer 1 convolution, b 1,u The u-th offset, representing the layer 1 convolution, represents the convolution operation, and ReLU represents the activation function, expressed specifically as:
(4.2) inputting implicit features generated by convolution in the (q-1) th convolution layer to the (q) th convolution layer according to a formula (7) to obtain a (q) th layer feature map;
c q,v ＝ReLU(c q-1,v *f q,v +b q,v ) (7)
wherein c q,v And c q-1,v Respectively represent the (th) and (th) -1 th convolution layers (v) th convolution kernels, f q,v A v-th convolution kernel representing a q-th layer convolution, b q,v A v-th offset representing a q-th layer convolution, q being an integer greater than 2;
(4.3) let q=q+1, repeat step (4.2) until q>q max Output the (q) max Feature mapping of convolutional layers, where q max =5 represents the maximum number of convolution layers;
(4.4) inputting the characteristic mapping of the 5 th layer to carry out pooling operation according to the formula (8);
wherein, maxpool β,γ The use of the β×γ region is represented where β×γ=2×1, forAnd (5) taking the maximum value.
(4.5) mapping c the features extracted by the pooling layer according to equation (9) p Corresponding output characteristics are obtained through flattening layer operation;
c F ＝ReLU(w F *c p +b F ) (9)
wherein c F Feature map representing flattened layers, w F Representing weights of the flattened layers, b F Indicating the bias of the flattened layer.
(4.6) transmitting the output characteristics of the flattening layer as input characteristics to the fully connected layer according to the formula (10) to obtain a final classification result:
wherein w is D And b D Respectively representing the weight and bias of the fully connected layer, and sigmoid represents an activation function, which is specifically expressed as follows:
step 5: training CNN according to a small batch gradient descent method, taking a minimized cross entropy loss function as an optimization target, and determining optimal learning parameters of CNN;
wherein s= { i 1 ,i 2 ,…,i E Each time batch is selectedA data set is taken, and e=100 represents the number of groups of data;
step 6: after the trained CNN is embedded into the intelligent power grid bad data detection device, real-time online attack positioning is realized, and extraction is performedAll the corresponding measurements are the attacked positions.
Table 1 shows the comparison of the results of the attack detection rate and the attack positioning accuracy rate of the smart grid using the feedback neural network method and the method according to the embodiment of the invention. As can be seen from Table 1, the method of the present invention achieves higher attack detection rate and attack localization accuracy than the feedback neural network method. By analyzing the operation experimental result of the intelligent power grid by adopting the technology, the invention can find that: the method can obtain higher attack detection rate and attack positioning accuracy than the feedback neural network method, verifies that the method has higher accuracy for intelligent power grid attack positioning, and can better meet the requirements of power grid safety control and stable operation.
Table 1: attack detection rate and attack positioning accuracy rate comparison
| Method | Attack detection rate | Attack location accuracy |
| Feedback neural network method | 89％ | 73.3％ |
| The positioning method of the invention | 96％ | 91.4％ |
In summary, the effective effects of the invention are as follows: the invention can realize the positioning of the attack of the intelligent power grid and has the following advantages which are not possessed by the prior art: the method can detect the power grid attack, can perform attack positioning on the power grid on line in real time, has higher positioning accuracy and higher positioning efficiency, does not depend on a model, and ensures the safety control and stable operation of the power grid.
Claims (3)
1. The smart grid attack positioning method based on the convolutional neural network is characterized by comprising the following steps of:
step 1: collecting network characteristic parameters of the intelligent power grid;
step 2: preprocessing data;
step 3: describing the smart grid attack positioning problem as a multi-label classification problem;
step 4: extracting intrinsic high-dimensional characteristics of the measured data by adopting a Convolutional Neural Network (CNN), and taking the CNN as a multi-label classifier;
step 5: training CNN according to a small batch gradient descent method, taking a minimized cross entropy loss function as an optimization target, and determining optimal learning parameters of CNN;
step 6: after the trained CNN is embedded into the intelligent power grid bad data detection device, real-time online attack positioning is realized, and extraction is performedAll the corresponding measurements are the attacked positions;
step 1: network characteristic parameters of the intelligent power grid are collected, and the network characteristic parameters are specifically as follows:
the method comprises the steps of obtaining measurement data according to a nonlinear measurement equation of a smart grid, wherein the measurement data comprise voltage phase angles and voltage amplitudes of reference nodes, injected active power and reactive power of each node, and each branchActive and reactive power of (1), wherein the data Z is measured normally o And measurement data Z subject to attack a The specific description is as follows:
Z o ＝h(x o )+ω (1)
Z a ＝Z+h(x a )-h(x o ) (2)
wherein h represents a nonlinear measurement equation, i.e., a functional relationship between system state and measurement truth, x o Representing the system state includes the true values of the voltage amplitude and voltage phase angle of all nodes of the system, x a Representing system state values under attack, x representing x o Or x a ，V k And theta k Respectively representing the voltage amplitude and the voltage phase angle of the node k, V m And theta m Respectively representing the voltage amplitude and the voltage phase angle of the node m, P k And Q k Respectively represent the injected active power and reactive power of the node k, P km And Q km Representing the active power and the reactive power of the line between the node k and the node M respectively, M representing the set of the node M, g km And b km Representing the conductance and susceptance, delta, between nodes k and m, respectively km ＝θ k -θ m Representing the phase angle difference between node k and node m, ω observing noise;
the step 2 data preprocessing is specifically as follows, let d= [ Z o ,Z a ]The measurement data obtained in the step 1 are stored in D, the data set D is transmitted to a bad data detection device of the smart grid, and the ith group of data D in the D is calculated according to the formula (4) i Residual R of (2) i I=1, 2, …, N represents the number of data groups, and the detection bad data threshold parameter λ is set and judged, if R i >Lambda, reject D i The method comprises the steps of carrying out a first treatment on the surface of the If R is i Less than or equal to lambda, D i Storing into a set T;
wherein,,representing the measured data D i Calculating the obtained system state estimated value;
step 3 describes the smart grid attack positioning problem as a multi-label classification problem, and specifically includes defining training set characteristics, namely measurement data as T i ＝(T i 1 ,...,T i j ,...,T i n ) Where j=1, 2, …, n represents the j-th measurement, n represents the feature number; defining training set labels, i.e. measuring belonging categories asWherein->A class of the j-th measurement representing the i-th group of data, if the j-th measurement is attacked, record +.>Otherwise, record->Defining the result of the multi-label classifier classification as +.>Wherein->Classification result of the j-th measurement representing the i-th group of data,/and/or>Or alternativelyIf->Then the j-th measurement representing the i-th set of data is attacked; if->Then the j-th measurement of the i-th set of data is not attacked;
the step 4 adopts a Convolutional Neural Network (CNN) to extract the intrinsic high-dimensional characteristics of the measured data, and takes the CNN as a multi-label classifier, and the specific substeps are as follows:
(4.1) acquiring a layer 1 feature map based on the input data T according to formula (5):
c 1,u ＝ReLU(T*f 1,u +b 1,u ) (5)
wherein c 1,u A ith feature map representing layer 1 convolution, f 1,j A u-th convolution kernel representing a layer 1 convolution, b 1,u The u-th offset, representing the layer 1 convolution, represents the convolution operation, and ReLU represents the activation function, expressed specifically as:
(4.2) inputting implicit features generated by convolution in the (q-1) th convolution layer to the (q) th convolution layer according to a formula (7) to obtain a (q) th layer feature map;
c q,v ＝ReLU(c q-1,v *f q,v +b q,v ) (7)
wherein c q,v And c q-1,v Respectively represent the (th) and (th) -1 th convolution layers (v) th convolution kernels, f q,v A v-th convolution kernel representing a q-th layer convolution, b q,v A v-th offset representing a q-th layer convolution, q being an integer greater than 2;
(4.3) let q=q+1, repeat step (4.2) until q>q max Output the (q) max Feature mapping of convolutional layers, where q max Indicating the maximum number of convolution layers;
(4.4) inputting the q-th according to the formula (8) max Pooling is carried out on the feature mapping of the convolution layer;
wherein, maxpool β,γ Representing the use of beta x gamma region pairsTaking the maximum value;
(4.5) mapping c the features extracted by the pooling layer according to equation (9) p By flattening the layer operation, corresponding output features are obtained,
c F ＝ReLU(w F *c p +b F ) (9)
wherein c F Feature map representing flattened layers, w F Representing weights of the flattened layers, b F Indicating the offset of the flattened layer,
(4.6) transmitting the output characteristics of the flattening layer as input characteristics to the fully connected layer according to the formula (10) to obtain a final classification result:
wherein w is D And b D Respectively representing the weight and bias of the fully connected layer, and sigmoid represents an activation function, which is specifically expressed as follows:
2. the smart grid attack localization method based on convolutional neural network according to claim 1, wherein step 5 trains CNN according to a small batch gradient descent method, takes minimizing cross entropy loss function as an optimization target, determines an optimal learning parameter of CNN,
，
wherein s= { i 1 ,i 2 ,…,i E The data set selected for each small lot is denoted, and E denotes the number of sets of data.
3. The smart grid attack location method based on convolutional neural network according to claim 1, wherein the step 6 is implemented by embedding the trained CNN into a smart grid bad data detection device, and extractingAll the corresponding measurements are the attacked positions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110486649.9A CN113268729B (en) | 2021-05-01 | 2021-05-01 | Smart grid attack positioning method based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110486649.9A CN113268729B (en) | 2021-05-01 | 2021-05-01 | Smart grid attack positioning method based on convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113268729A CN113268729A (en) | 2021-08-17 |
| CN113268729B true CN113268729B (en) | 2023-07-28 |
Family
ID=77229968
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110486649.9A Active CN113268729B (en) | 2021-05-01 | 2021-05-01 | Smart grid attack positioning method based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113268729B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110826888A (en) * | 2019-10-29 | 2020-02-21 | 西安交通大学 | Data integrity attack detection method in power system dynamic state estimation |
| CN111092897A (en) * | 2019-12-23 | 2020-05-01 | 深圳供电局有限公司 | Active immune attack recognition method and charging device |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9904893B2 (en) * | 2013-04-02 | 2018-02-27 | Patternex, Inc. | Method and system for training a big data machine to defend |
| CN107016236B (en) * | 2017-03-23 | 2021-04-20 | 新疆电力建设调试所 | Power grid false data injection attack detection method based on nonlinear measurement equation |
| CN109165504B (en) * | 2018-08-27 | 2021-05-07 | 广西大学 | Power system false data attack identification method based on anti-generation network |
| CN110719275B (en) * | 2019-09-30 | 2021-04-02 | 南方电网科学研究院有限责任公司 | Method for detecting power terminal vulnerability attack based on message characteristics |
| CN111353153B (en) * | 2020-03-04 | 2022-11-01 | 南京邮电大学 | GEP-CNN-based power grid malicious data injection detection method |
| CN112565187B (en) * | 2020-11-03 | 2023-05-09 | 特变电工新疆新能源股份有限公司 | Power grid attack detection method, system, equipment and medium based on logistic regression |
-
2021
- 2021-05-01 CN CN202110486649.9A patent/CN113268729B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110826888A (en) * | 2019-10-29 | 2020-02-21 | 西安交通大学 | Data integrity attack detection method in power system dynamic state estimation |
| CN111092897A (en) * | 2019-12-23 | 2020-05-01 | 深圳供电局有限公司 | Active immune attack recognition method and charging device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113268729A (en) | 2021-08-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Graph convolutional network-based method for fault diagnosis using a hybrid of measurement and prior knowledge | |
| CN109165504B (en) | Power system false data attack identification method based on anti-generation network | |
| CN110659693B (en) | K-nearest neighbor classification-based power distribution network rapid topology identification method, system and medium | |
| CN108108877A (en) | A kind of transmission line of electricity damage to crops caused by thunder methods of risk assessment based on BP neural network | |
| CN109145516B (en) | Analog circuit fault identification method based on improved extreme learning machine | |
| CN108435819B (en) | Energy consumption abnormity detection method for aluminum profile extruder | |
| CN108090628A (en) | A kind of grain feelings security detection and analysis method based on PSO-LSSVM algorithms | |
| CN112327104B (en) | Fault detection and positioning method for power distribution network with distributed power supply | |
| CN114638256B (en) | Transformer fault detection method and system based on acoustic wave signals and attention network | |
| CN113496440B (en) | User abnormal electricity consumption detection method and system | |
| Xu et al. | Oversmoothing relief graph convolutional network-based fault diagnosis method with application to the rectifier of high-speed trains | |
| CN111582406A (en) | Power equipment state monitoring data clustering method and system | |
| CN117151657A (en) | Plastic packaging material management method and system | |
| CN113514743B (en) | Method for constructing GIS partial discharge pattern recognition system based on multidimensional features | |
| CN113780432B (en) | Intelligent detection method for operation and maintenance abnormity of network information system based on reinforcement learning | |
| CN114330486A (en) | Power system bad data identification method based on improved Wasserstein GAN | |
| CN113268729B (en) | Smart grid attack positioning method based on convolutional neural network | |
| CN116258972A (en) | Deep learning-based rock high-steep slope structural surface extraction method | |
| CN116316629A (en) | Data-driven optimal power flow calculation method considering topology feature learning | |
| CN114692729A (en) | New energy station bad data identification and correction method based on deep learning | |
| CN114821187A (en) | Image anomaly detection and positioning method and system based on discriminant learning | |
| CN109506936B (en) | Bearing fault degree identification method based on flow chart and non-naive Bayes inference | |
| CN116340833B (en) | Fault diagnosis method based on countermeasure migration network in improved field | |
| CN114021433B (en) | Construction method and application of dominant instability mode identification model of power system | |
| CN118035952B (en) | Table box identification method and system based on kirchhoff law regression model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |