CN116958752B - Power grid infrastructure archiving method, device and equipment based on IPKCNN-SVM - Google Patents
Power grid infrastructure archiving method, device and equipment based on IPKCNN-SVM Download PDFInfo
- Publication number
- CN116958752B CN116958752B CN202311212335.5A CN202311212335A CN116958752B CN 116958752 B CN116958752 B CN 116958752B CN 202311212335 A CN202311212335 A CN 202311212335A CN 116958752 B CN116958752 B CN 116958752B
- Authority
- CN
- China
- Prior art keywords
- layer
- firefly
- svm
- training
- image
- 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
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000012549 training Methods 0.000 claims abstract description 73
- 238000012360 testing method Methods 0.000 claims abstract description 24
- 230000007246 mechanism Effects 0.000 claims abstract description 12
- 238000004458 analytical method Methods 0.000 claims abstract description 7
- 241000254158 Lampyridae Species 0.000 claims description 75
- 238000004590 computer program Methods 0.000 claims description 19
- 238000004422 calculation algorithm Methods 0.000 claims description 18
- 230000006870 function Effects 0.000 claims description 14
- 230000004913 activation Effects 0.000 claims description 13
- 238000003860 storage Methods 0.000 claims description 7
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000001902 propagating effect Effects 0.000 claims description 6
- 238000009826 distribution Methods 0.000 claims description 4
- 230000006872 improvement Effects 0.000 claims description 4
- 210000002569 neuron Anatomy 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- 230000004069 differentiation Effects 0.000 claims description 3
- 230000031700 light absorption Effects 0.000 claims description 3
- 230000000644 propagated effect Effects 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 2
- 238000013461 design Methods 0.000 abstract description 3
- 238000000605 extraction Methods 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 238000011176 pooling Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011056 performance test Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
- G06F16/355—Class or cluster creation or modification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/205—Parsing
- G06F40/211—Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/70—Labelling scene content, e.g. deriving syntactic or semantic representations
-
- 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
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Abstract
The utility model provides a method for power grid infrastructure architecture archiving based on IPKCNN-SVM, firstly, annotate the text data of each image in the pre-training set through the image, carry out dependency syntactic analysis to each text data and obtain sentence dependency, learn sentence dependency through self-attention mechanism and obtain attention output, load attention output as priori knowledge to CNN network and pretrain, input each image in the training set in CNN network and extract image feature through the input layer, forward propagate the image feature of extraction through a plurality of convolution layers, after forward propagate, begin to carry out reverse propagate from the classifier of last layer convolution layer, update the parameter of each convolution layer, finally input each image in the test set in the CNN network that trains, accomplish the classification archiving to power grid infrastructure architecture. The model provided by the design has short training time and high classification precision.
Description
Technical Field
The application belongs to the technical field of electronic archiving, and particularly relates to an IPKCNN-SVM-based power grid infrastructure archiving method, device and equipment.
Background
The management of the grid infrastructure files is widely focused as an important component for supporting the development of the grid infrastructure, however, the types of the grid infrastructure files are complex and various, the number of the grid infrastructure files is increased drastically, and the management difficulty of the grid infrastructure files is increased. Aiming at the impact of digital transformation of the power grid infrastructure files, a precise identification and classification method is explored to ensure the archiving quality of the infrastructure files, and the method has important significance in promoting the development of the power grid infrastructure.
In the prior art, an electronic archive automatic classifying system research based on text feature recognition provides an intelligent recognition system based on text features, and the recognition and classification of electronic archives are realized by utilizing the system, however, the method and the strategy only can process small-scale archives and low-dimensional feature archives, and have obvious limitation when the number of archives is increased and the feature complexity is increased sharply, the training time is longer, and the classification precision is low.
Disclosure of Invention
The application aims to solve the problems in the prior art and provides an IPKCNN-SVM-based power grid infrastructure archiving method, device and equipment capable of shortening training time and improving classification accuracy.
In order to achieve the above object, the technical scheme of the present application is as follows:
an IPKCNN-SVM based power grid infrastructure archiving method, the method comprising:
s1, dividing the obtained multiple power grid infrastructure building images into a pre-training set, a training set and a testing set, and marking building categories in the images;
s2, marking text data of each image in the pre-training set through the images, performing dependency syntactic analysis on each text data to obtain sentence dependency relationship, then learning the sentence dependency relationship through a self-attention mechanism to obtain attention output, and loading the attention output into a CNN network as priori knowledge to perform pre-training;
s3, inputting each image in a training set into a CNN network for joint training, wherein the joint training specifically comprises the steps of firstly extracting image features through an input layer, then forward propagating the extracted image features through a plurality of convolution layers, and after the forward propagation is completed, starting to reversely propagate from a classifier of a last convolution layer, and updating parameters of each convolution layer;
s4, inputting each image in the test set into a trained CNN network to finish classified archiving of the power grid infrastructure.
The step S2 includes:
s21, marking text data of each image in the pre-training set through the images, then giving a dependency relationship label, and coding sentence dependency relationship in each text data by using a Stanford NLP syntax analyzer to obtain dependency relationship codes:
；
in the above-mentioned method, the step of,encoding for dependencies>For the h-th dependency, h=1, 2, …, m, m represents the total number of dependencies;
s22, encoding and inputting the dependency relationship into a converter to obtain a characteristic representation:
；
；
in the above-mentioned method, the step of,encoding +.>Inputting the characteristic representation obtained after the converter, +.>、/>The characteristic representation output by the ith layer and the ith-1 layer full-connection layer respectively, < ->、/>Weights and biases of the i-th full link layers, i=1, 2, …, n, < >>Encoded by dependency +.>The first full-connection layer is input to obtain n which is the total layer number of the full-connection layer in the converter, the weight and bias of each full-connection layer are different, the number of neurons in each full-connection layer is the same as the number of dependency relations, and n is the total layer number of the full-connection layer in the converter>Is an activation function;
s23, learning sentence dependency relationship through a self-attention mechanism to obtain attention output:
；
；
in the above-mentioned method, the step of,for attention output, ++>Attention distribution weight represented for the ith feature, +.>Is->Transpose of->Representing a context vector;
s24, the attention output is used as a feature vector to be input into a CNN network for pretraining.
In the joint training of step S3, the extracted image features are propagated forward through a plurality of convolution layers according to the following formula:
；
in the above-mentioned method, the step of,for the j-th output feature of the k-th convolution layer,>the p-th input feature of the k-1 th convolutional layer; />，/>For the image feature set, +.>Weights for the k-th convolution layer, +.>For the k-th layer convolutional layer bias term, +.>Is an activation function;
after the forward propagation is completed, the backward propagation is started from the classifier of the last layer of convolution layer, the weight of each layer of convolution layer is updated, and the weight updating formula of the last layer of convolution layer is as follows:
；
wherein the saidUpdated weights for the last convolutional layer, +.>Error between output and input of classifier for last layer convolution layer, < >>For learning rate->Output of the last convolution layer;
the weight update formula of the rest convolution layers is as follows:
；
in the above-mentioned method, the step of,updated weights for convolutional layer b, +.>For the weight before update of convolutional layer b, +.>For the number of convolutional layers lying after convolutional layer b, +.>To be positioned behind convolution layer b +.>Individual rollsError between output and input of laminated classifier +.>Partial differentiation of ∈ ->To be positioned behind convolution layer b +.>The activation values generated by the activation functions of the individual convolutional layers.
The classifier of the CNN network is an SVM classifier, and parameters of the SVM classifier are optimized by utilizing a teaching and learning improved firefly algorithm after updating parameters of each convolution layer.
The optimization of the parameters of the SVM classifier by utilizing the teaching and learning improved firefly algorithm is as follows:
setting a search space of a firefly algorithm as a value range of parameters of an SVM classifier, initializing the position, brightness and attraction of the firefly, continuously optimizing the parameters of the SVM, taking the classification performance of the SVM classifier as the brightness of the firefly, updating the position of the firefly according to the following formula, and iteratively updating until the brightness converges:
；
；
；
；
；
in the above-mentioned method, the step of,for the final position of the firefly algorithm based on teaching and learning factors improvement, +.>For the position after the firefly update, +.>The individual moving direction of the fireflies at the time t+1 is obtained according to the difference between the fireflies u and the average position of the fireflies and the difference between the fireflies v and the average position of the fireflies; />、The difference between the firefly u and the firefly average position at time t+1, the difference between the firefly v and the firefly average position, and +.>For the mean position of firefly at time t, +.>、/>Taking 1 or 2 randomly as teaching factors; t is the current iteration number; />、/>The spatial positions of fireflies u and v at the moment t are respectively; />Is the attractive force between firefly u and firefly v; />Is constant and is generally taken as->；/>Is a random tree vector, ++>For maximum attractive force, typically 1 is taken; />For the light absorption coefficient, generally +.>，/>Is the Euclidean distance of firefly u and firefly v.
The parameters of the SVM classifier are penalty factors of the SVM classifierWidth parameter->。
An IPKCNN-SVM-based power grid infrastructure archiving device comprises an image preprocessing module, a CNN network parameter training module and a classification module;
the image preprocessing module is used for dividing the obtained multiple power grid infrastructure building images into a pre-training set, a training set and a testing set, and marking building categories in the images;
the CNN network parameter training module is used for marking text data of each image in a pre-training set through images, performing dependency syntactic analysis on each text data to obtain sentence dependency relationship, learning the sentence dependency relationship through a self-attention mechanism to obtain attention output, loading the attention output into a CNN network to perform pre-training, inputting each image in the training set into the CNN network, extracting image features through an input layer, forward transmitting the extracted image features through a plurality of convolution layers, and after the forward transmission is completed, starting backward transmitting from a classifier of a last layer of convolution layer, and updating parameters of each convolution layer;
and the classification module is used for inputting each image in the test set into the trained CNN network to finish classification filing of the power grid infrastructure.
The archiving device further comprises an SVM parameter optimizing module;
the SVM parameter optimizing module is used for optimizing parameters of the SVM classifier by utilizing a teaching and learning improved firefly algorithm after updating parameters of each convolution layer.
An IPKCNN-SVM based grid-based capital construction archiving device, the archiving device comprising a memory and a processor;
the memory is used for storing computer program codes and transmitting the computer program codes to the processor;
the processor is configured to perform the archiving method described above in accordance with instructions in the computer program code.
A computer readable storage medium having stored thereon a computer program which when executed by a processor implements the archiving method described above.
Compared with the prior art, the application has the beneficial effects that:
1. the application relates to an IPKCNN-SVM-based power grid infrastructure archiving method, which comprises the steps of firstly dividing an obtained plurality of power grid infrastructure building images into a pre-training set, a training set and a testing set, marking out building types in the images, marking out text data of each image in the pre-training set through the images, performing dependency syntax analysis on each text data to obtain sentence dependency relationship, learning the sentence dependency relationship through a self-attention mechanism to obtain attention output, loading the attention output as priori knowledge into a CNN network for pre-training, inputting each image in the training set into the CNN network for joint training, specifically, extracting image features through an input layer, carrying out forward propagation on the extracted image features through a plurality of convolution layers, carrying out backward propagation from a classifier of a last layer of convolution layer, updating parameters of each convolution layer, and finally inputting each image in the testing set into the trained CNN network to finish classifying and archiving of the power grid infrastructure building; according to the design, on one hand, the dependency relationship between words is learned through a self-attention mechanism, the obtained attention output is fused into a convolutional neural network model as priori knowledge to participate in pre-training, after the priori knowledge is added, the model can be helped to better understand and capture potential modes and associated information in text data, relevant features are more accurately identified and extracted, time-consuming fitting irrelevant features are avoided, convergence speed and classification precision of the model in the process of processing a large number of texts and data are improved, and on the other hand, a training mode combining forward propagation and reverse propagation is introduced into the convolutional neural network, so that training speed is finally effectively improved. Therefore, the method provided by the application has the advantages of short model training time and high classification precision.
2. The application relates to an IPKCNN-SVM-based power grid infrastructure archiving method, which adopts an SVM classifier as a classifier of a CNN network and optimizes parameters of the SVM classifier by utilizing teaching and learning improved firefly algorithm; the firefly algorithm is a global optimization algorithm, and the design can enhance global searching capability in the optimization process after introducing teaching factors, so that the firefly algorithm can better search the optimal solution in the SVM classifier parameter space, better optimize SVM classifier parameters, improve the distinguishing performance of the model on different categories, and further improve the classification precision of the model. Therefore, the method and the device can further improve the model classification precision.
Drawings
FIG. 1 is a flow chart of an archiving method according to the present application.
Fig. 2 is a classification accuracy test result of three models for grid-based building images with different signal-to-noise ratios.
Fig. 3 is a block diagram of the archival device according to the present application.
Fig. 4 is a block diagram of the configuration of the filing apparatus according to the present application.
Detailed Description
The present application will be described in further detail with reference to the following detailed description and the accompanying drawings.
Example 1:
referring to fig. 1, an IPKCNN-SVM-based power grid infrastructure archiving method is sequentially performed according to the following steps:
1. dividing the obtained multiple power grid infrastructure building images into a pre-training set, a training set and a testing set, and marking building categories in the images;
2. firstly, marking text data of each image in a pre-training set through the images, then giving a dependency relationship label, and coding sentence dependency relationship in the text data by using a Stanford NLP syntactic analyzer to obtain dependency relationship coding:
；
in the above-mentioned method, the step of,encoding for dependencies>For the h-th dependency, h=1, 2, …, m, m represents the total number of dependencies;
then, the dependency is encoded into a transducer to obtain a characteristic representation:
；
；
in the above-mentioned method, the step of,encoding +.>Inputting the characteristic representation obtained after the converter, +.>、/>The characteristic representation output by the ith layer and the ith-1 layer full-connection layer respectively, < ->、/>Weights and biases of the i-th full link layers, i=1, 2, …, n, < >>Encoded by dependency +.>The first full-connection layer is input to obtain n which is the total layer number of the full-connection layer in the converter, the weight and bias of each full-connection layer are different, the number of neurons in each full-connection layer is the same as the number of dependency relations, and n is the total layer number of the full-connection layer in the converter>Is an activation function;
learning sentence dependency relationship through a self-attention mechanism to obtain attention output:
；
；
in the above-mentioned method, the step of,for attention output, ++>Attention distribution weight represented for the ith feature, +.>Is->Transpose of->Representing a context vector;
finally, the attention output of each image is used as a feature vector to be input into a CNN network for pre-training;
3. inputting each image in a training set into a CNN (computer network) for joint training, wherein the CNN has the structure that:
input layer- & gt convolution layer 1- & gt pooling layer 1- & gt classifier 1- & gt convolution layer 2- & gt pooling layer 2- & gt classifier 2- & gt … - & gt convolution layer a-1- & gt pooling layer a-1- & gt classifier a-1- & gt convolution layer a- & gt pooling layer- & gt full connection layer- & gt classifier a;
after the feature vector of the text data of each image is input into the input layer of the CNN network, the feature vector is propagated forward through a plurality of convolution layers according to the following formula:
；
in the above-mentioned method, the step of,for the j-th output feature of the k-th convolution layer,>the p-th input feature of the k-1 th convolutional layer; />，/>For the image feature set, +.>Weights for the k-th convolution layer, +.>For the k-th layer convolutional layer bias term, +.>Is an activation function;
4. after the forward propagation is completed, the backward propagation is started from the classifier of the last layer of convolution layer, the weight of each layer of convolution layer is updated, and the weight updating formula of the last layer of convolution layer is as follows:
；
wherein the saidUpdated weights for the last convolutional layer, +.>Error between output and input of classifier for last layer convolution layer, < >>For learning rate->Output of the last convolution layer;
the weight update formula of the rest convolution layers is as follows:
；
in the above-mentioned method, the step of,updated weights for convolutional layer b, +.>For the weight before update of convolutional layer b, +.>For the number of convolutional layers lying after convolutional layer b, +.>To be positioned behind convolution layer b +.>Error between classifier output and input of each convolution layer +.>Partial differentiation of ∈ ->To be positioned behind convolution layer b +.>An activation value generated by an activation function of the plurality of convolutional layers;
5. the classifier adopted by the CNN network is replaced by an SVM classifier, the SVM classifier is trained by using the output data of the full-connection layer, and the punishment factors of the SVM classifier are improved by using the teaching and learning improved firefly algorithmWidth parameter->Optimizing to obtain a trained IPKCNN-SVM model, wherein the optimizing process comprises the following steps:
setting the search space of firefly algorithm as punishment factorOr width parameter->Initializing the position, brightness and attraction of fireflies, continuously optimizing parameters, taking the classification performance of the SVM classifier as the brightness of the fireflies, updating the position of the fireflies according to the following formula, and iteratively updating until the brightness converges:
；
；
；
；
；
in the above-mentioned method, the step of,for the final position of the firefly algorithm based on teaching and learning factors improvement, +.>For the position after the firefly update, +.>The individual moving direction of the fireflies at the time t+1 is obtained according to the difference between the fireflies u and the average position of the fireflies and the difference between the fireflies v and the average position of the fireflies; />、The difference between the firefly u and the firefly average position at time t+1, the difference between the firefly v and the firefly average position, and +.>For the mean position of firefly at time t, +.>、/>Taking 1 or 2 randomly as teaching factors; t is the current iteration number; />、/>The spatial positions of fireflies u and v at the moment t are respectively; />Is the attractive force between firefly u and firefly v; />Is constant and is generally taken as->；/>Is a random tree vector, ++>For maximum attractive force, typically 1 is taken; />For the light absorption coefficient, generally +.>，/>The Euclidean distance is firefly u and firefly v;
6. and inputting each image in the test set into a trained IPKCNN-SVM model to finish classified archiving of the power grid infrastructure.
Performance test:
1. inputting a plurality of power grid infrastructure building images obtained in a power grid infrastructure region into the IPKCNN-SVM model, the CNN model and the CNN-SVM model for training and classifying regression, testing the training duration and classifying regression result accuracy of the three models, and testing the results shown in Table 1:
table 1 training duration test results for three models
As can be seen from table 1, the training time of the CNN model is 92.8703s, the test time is 0.477s, and the accuracy is 91.11%; the CNN-SVM model is respectively faster than 0.2563s and 0.401s in training time and testing time, which shows that the combination of the SVM and the CNN can improve the classification and identification efficiency of the model, and the accuracy is improved by 4.45%; compared with the CNN-SVM model, the IPKCNN-SVM model provided by the application has the advantages that the training time and the testing time are respectively reduced by 29.3% and 23.68%, and the accuracy is improved by 1.66%, so that the training time and the classifying time are obviously shortened, and the accuracy is effectively improved.
2. The method comprises the steps of inputting power grid infrastructure building images with different signal to noise ratios into the IPKCNN-SVM model, the CNN model and the CNN-SVM model for training and classifying regression, and testing the accuracy of classifying regression results of the three models, wherein the test results are shown in figure 2;
as can be seen from fig. 2, under the strong interference of 1dB gaussian white noise, the accuracy of the IPKCNN-SVM model of the present application can reach 82.22%, whereas the accuracy of the CNN-SVM model and the CNN model is only 80% and 75%; the accuracy of the three models is steadily improved when noise interference is gradually weakened, and under the small interference of 5dB noise, the accuracy of the IPKCNN-SVM model reaches 97.18 percent, and compared with the accuracy under the noise-free condition, the accuracy of the CNN-SVM model and the accuracy of the CNN model under the small interference of 5dB noise are 2.22 percent and 6.67 percent lower than those of the IPKCNN-SVM model.
Example 2:
referring to fig. 3, an IPKCNN-SVM-based power grid infrastructure archives device includes an image preprocessing module, a CNN network parameter training module, an SVM parameter optimizing module, and a classification module; the image preprocessing module is used for dividing the obtained multiple power grid infrastructure building images into a pre-training set, a training set and a testing set, marking building categories in the images, and specifically executing the steps 1 in the embodiment 1; the CNN network parameter training module is used for marking text data of each image in a pre-training set through images, performing dependency syntactic analysis on each text data to obtain sentence dependency relationship, learning the sentence dependency relationship through a self-attention mechanism to obtain attention output, loading the attention output into a CNN network to perform pre-training, inputting each image in the training set into the CNN network, extracting image features through an input layer, forward propagating the extracted image features through a plurality of convolution layers, and after the forward propagating is completed, starting backward propagating from a classifier of a last layer of convolution layer, and updating parameters of each convolution layer; the SVM parameter optimizing module is used for optimizing parameters of the SVM classifier by utilizing data of a CNN network full convolution layer after updating parameters of each convolution layer through teaching and learning improvement firefly algorithm, and the module specifically executes the step 5 as in the embodiment 1; the classification module is used for inputting each image in the test set into a trained CNN network to finish classification filing of the power grid infrastructure building; this module performs in particular step 6 as in example 1.
Example 3:
referring to fig. 4, an IPKCNN-SVM based power grid infrastructure archival device includes a memory and a processor;
the memory is used for storing computer program codes and transmitting the computer program codes to the processor; the processor is used for executing the archiving method according to the instructions in the computer program code;
a computer readable storage medium having stored thereon a computer program which when executed by a processor implements the archiving method described above.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein. The scheme in the embodiment of the application can be realized by adopting various computer languages, such as object-oriented programming language Java, an transliteration script language JavaScript and the like.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present application have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiments and all such alterations and modifications as fall within the scope of the application.
Claims (6)
1. An IPKCNN-SVM-based power grid infrastructure archiving method is characterized in that:
the method comprises the following steps:
s1, dividing the obtained multiple power grid infrastructure building images into a pre-training set, a training set and a testing set, and marking building categories in the images;
s2, marking text data of each image in a pre-training set through the images, performing dependency syntactic analysis on each text data to obtain sentence dependency relationship, then learning the sentence dependency relationship through a self-attention mechanism to obtain attention output, and loading the attention output into a CNN network as priori knowledge to perform pre-training, wherein the method specifically comprises the following steps:
s21, marking text data of each image in the pre-training set through the images, then giving a dependency relationship label, and coding sentence dependency relationship in each text data by using a Stanford NLP syntax analyzer to obtain dependency relationship codes:
in the above equation, r is the dependency code,for the h-th dependency, h=1, 2, …, m, m represents the total number of dependencies;
s22, encoding and inputting the dependency relationship into a converter to obtain a characteristic representation:
s i ＝tanh(w i s i-1 +b i )；
S＝{s 1 ，s 2 ，...，s i ...，s n }；
in the above formula, S is a characteristic representation obtained by inputting the dependency relation code r into the converter, and S i 、s i-1 Respectively represents full connection by the ith layer and the ith-1 layerCharacterization of layer output, w i 、b i Weights and biases of the i-th full link layers, i=1, 2, …, n, s, respectively 1 The method comprises the steps that dependence relation codes r are input into a first full-connection layer to obtain, n is the total number of layers comprising the full-connection layer in the converter, the weight and bias of each full-connection layer are different, the number of neurons comprising the full-connection layer is the same as the number of dependence relation, and tanh is an activation function;
s23, learning sentence dependency relationship through a self-attention mechanism to obtain attention output:
in the above formula, D is the attention output, a i The attention distribution weight represented for the ith feature,is s i Transpose of s w Representing a context vector;
s24, inputting the attention output as a feature vector into a CNN network for pre-training;
s3, inputting each image in a training set into a CNN network for joint training, wherein the joint training specifically comprises the steps of firstly extracting image features through an input layer, then forward propagating the extracted image features through a plurality of convolution layers, and after the forward propagation is completed, starting to reversely propagate from a classifier of a last convolution layer, and updating parameters of each convolution layer;
s4, replacing a classifier adopted by the CNN network with an SVM classifier, training the SVM classifier by utilizing output data of a CNN network full-connection layer, optimizing punishment factors rho and width parameters sigma of the SVM classifier by utilizing a teaching and learning improved firefly algorithm to obtain a trained IPKCNN-SVM model, inputting each image in a test set into the trained IPKCNN-SVM model, and finishing classification archiving of the power grid infrastructure.
2. The IPKCNN-SVM based power grid infrastructure archiving method of claim 1, wherein:
in the joint training of step S3, the extracted image features are propagated forward through a plurality of convolution layers according to the following formula:
in the above-mentioned method, the step of,for the j-th output feature of the k-th convolution layer,>the p-th input feature of the k-1 th convolutional layer; p epsilon M p ，M p For the image feature set, +.>Weights for the k-th convolution layer, +.>A k-th convolution layer bias term, f is an activation function;
after the forward propagation is completed, the backward propagation is started from the classifier of the last layer of convolution layer, the weight of each layer of convolution layer is updated, and the weight updating formula of the last layer of convolution layer is as follows:
wherein the saidE is the updated weight of the last convolution layer FC Error between output and input of classifier of the last layer of convolution layer, eta is learning rate and x a Output of the last convolution layer;
the weight update formula of the rest convolution layers is as follows:
in the above-mentioned method, the step of,for the updated weights of convolution layer b, W b For the weight before updating of convolution layer b, L is the number of convolution layers after convolution layer b,/>Error E between output and input of classifier of the first convolution layer after convolution layer b l Partial differentiation of alpha l The activation value generated for the activation function of the first convolutional layer after convolutional layer b.
3. The IPKCNN-SVM based power grid infrastructure archiving method of claim 1, wherein:
in the step S4, the optimization of the parameters of the SVM classifier by using the teaching and learning improved firefly algorithm is as follows:
setting a search space of a firefly algorithm as a value range of parameters of an SVM classifier, initializing the position, brightness and attraction of the firefly, continuously optimizing the parameters of the SVM, taking the classification performance of the SVM classifier as the brightness of the firefly, updating the position of the firefly according to the following formula, and iteratively updating until the brightness converges:
X(t+1)＝X fa (t+1)+ΔX tlbo (t+1)；
in the above formula, X (t+1) is the final position of firefly algorithm based on teaching and learning factor improvement, X fa (t+1) is the position after the update of firefly, ΔX tlbo (t+1) is the individual moving direction of the firefly at time t+1 obtained from the difference between the firefly u and the average position of the firefly and the difference between the firefly v and the average position of the firefly; ΔX 1 (t+1)、ΔX 2 (t+1) is the difference between the average positions of firefly u and firefly at time t+1, the difference between the average positions of firefly v and firefly,for the mean position of firefly at time T, T F1 、T F2 Taking 1 or 2 randomly as teaching factors; t is the current iteration number; x is X u (t)、X v (t) are the spatial positions of firefly u and firefly v at time t, respectively; beta uv (r uv ) Is the attractive force between firefly u and firefly v; alpha t Is constant and is generally taken as [0,1 ]]；/>As a random tree vector, beta 0 For maximum attractive force, typically 1 is taken; gamma is the light absorption coefficient, generally taken as [0.01, 100 ]]，r uv Is the Euclidean distance of firefly u and firefly v.
4. An IPKCNN-SVM-based power grid-based foundation construction archiving device is characterized in that:
the archiving device comprises an image preprocessing module, a CNN network parameter training module and a classification module;
the image preprocessing module is used for dividing the obtained multiple power grid infrastructure building images into a pre-training set, a training set and a testing set, and marking building categories in the images;
the CNN network parameter training module is used for marking text data of each image in a pre-training set through the images, performing dependency syntactic analysis on each text data to obtain sentence dependency relationship, learning the sentence dependency relationship through a self-attention mechanism to obtain attention output, and loading the attention output into a CNN network to perform pre-training, and specifically comprises the following steps:
s21, marking text data of each image in the pre-training set through the images, then giving a dependency relationship label, and coding sentence dependency relationship in each text data by using a Stanford NLP syntax analyzer to obtain dependency relationship codes:
in the above equation, r is the dependency code,for the h-th dependency, h=1, 2, …, m, m represents the total number of dependencies;
s22, encoding and inputting the dependency relationship into a converter to obtain a characteristic representation:
s i ＝tanh(w i s i-1 +b i )；
S＝{s 1 ，s 2 ，...，s i ...，s n }；
in the above formula, S is a characteristic representation obtained by inputting the dependency relation code r into the converter, and S i 、s i-1 Separate tableThe characteristic representation output by the ith layer and the ith-1 layer full connection layer is shown as w i 、b i Weights and biases of the i-th full link layers, i=1, 2, …, n, s, respectively 1 The method comprises the steps that dependence relation codes r are input into a first full-connection layer to obtain, n is the total number of layers comprising the full-connection layer in the converter, the weight and bias of each full-connection layer are different, the number of neurons comprising the full-connection layer is the same as the number of dependence relation, and tanh is an activation function;
s23, learning sentence dependency relationship through a self-attention mechanism to obtain attention output:
in the above formula, D is the attention output, a i The attention distribution weight represented for the ith feature,is s i Transpose of s w Representing a context vector;
s24, inputting the attention output as a feature vector into a CNN network for pre-training;
then inputting each image in the training set into a CNN network, extracting image features through an input layer, forward propagating the extracted image features through a plurality of convolution layers, and after the forward propagation is finished, starting to reversely propagate from a classifier of the last convolution layer, and updating parameters of each convolution layer;
the classification module is used for replacing a classifier adopted by the CNN network with an SVM classifier, training the SVM classifier by utilizing output data of a full-connection layer of the CNN network, optimizing penalty factors rho and width parameters sigma of the SVM classifier by utilizing a teaching and learning improved firefly algorithm to obtain a trained IPKCNN-SVM model, and inputting each image in a test set into the trained IPKCNN-SVM model to finish classifying and archiving of a power grid infrastructure building.
5. An IPKCNN-SVM-based power grid infrastructure archiving device is characterized in that:
the archiving device includes a memory and a processor;
the memory is used for storing computer program codes and transmitting the computer program codes to the processor;
the processor is configured to perform the method of any one of claims 1 to 3 according to instructions in the computer program code.
6. A computer-readable storage medium, characterized by: a computer program stored on a computer readable storage medium, which when executed by a processor, implements the method of any one of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311212335.5A CN116958752B (en) | 2023-09-20 | 2023-09-20 | Power grid infrastructure archiving method, device and equipment based on IPKCNN-SVM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311212335.5A CN116958752B (en) | 2023-09-20 | 2023-09-20 | Power grid infrastructure archiving method, device and equipment based on IPKCNN-SVM |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116958752A CN116958752A (en) | 2023-10-27 |
| CN116958752B true CN116958752B (en) | 2023-12-15 |
Family
ID=88449568
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311212335.5A Active CN116958752B (en) | 2023-09-20 | 2023-09-20 | Power grid infrastructure archiving method, device and equipment based on IPKCNN-SVM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116958752B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107748895A (en) * | 2017-10-29 | 2018-03-02 | 北京工业大学 | UAV Landing landforms image classification method based on DCT CNN models |
| WO2019076227A1 (en) * | 2017-10-20 | 2019-04-25 | 北京达佳互联信息技术有限公司 | Human face image classification method and apparatus, and server |
| CN112001187A (en) * | 2020-08-26 | 2020-11-27 | 重庆理工大学 | Emotion classification system based on Chinese syntax and graph convolution neural network |
| WO2020244066A1 (en) * | 2019-06-04 | 2020-12-10 | 平安科技（深圳）有限公司 | Text classification method, apparatus, device, and storage medium |
| CN113158667A (en) * | 2021-04-09 | 2021-07-23 | 杭州电子科技大学 | Event detection method based on entity relationship level attention mechanism |
| CN114898151A (en) * | 2022-05-13 | 2022-08-12 | 重庆理工大学 | Image classification method based on deep learning and support vector machine fusion |
| CN115641473A (en) * | 2022-10-21 | 2023-01-24 | 长沙理工大学 | Remote sensing image classification method based on CNN-self-attention mechanism hybrid architecture |
| CN116363536A (en) * | 2023-05-31 | 2023-06-30 | 国网湖北省电力有限公司经济技术研究院 | Unmanned aerial vehicle inspection data-based power grid infrastructure equipment defect archiving method |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11205103B2 (en) * | 2016-12-09 | 2021-12-21 | The Research Foundation for the State University | Semisupervised autoencoder for sentiment analysis |
| US20190095764A1 (en) * | 2017-09-26 | 2019-03-28 | Panton, Inc. | Method and system for determining objects depicted in images |
-
2023
- 2023-09-20 CN CN202311212335.5A patent/CN116958752B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019076227A1 (en) * | 2017-10-20 | 2019-04-25 | 北京达佳互联信息技术有限公司 | Human face image classification method and apparatus, and server |
| CN107748895A (en) * | 2017-10-29 | 2018-03-02 | 北京工业大学 | UAV Landing landforms image classification method based on DCT CNN models |
| WO2020244066A1 (en) * | 2019-06-04 | 2020-12-10 | 平安科技（深圳）有限公司 | Text classification method, apparatus, device, and storage medium |
| CN112001187A (en) * | 2020-08-26 | 2020-11-27 | 重庆理工大学 | Emotion classification system based on Chinese syntax and graph convolution neural network |
| CN113158667A (en) * | 2021-04-09 | 2021-07-23 | 杭州电子科技大学 | Event detection method based on entity relationship level attention mechanism |
| CN114898151A (en) * | 2022-05-13 | 2022-08-12 | 重庆理工大学 | Image classification method based on deep learning and support vector machine fusion |
| CN115641473A (en) * | 2022-10-21 | 2023-01-24 | 长沙理工大学 | Remote sensing image classification method based on CNN-self-attention mechanism hybrid architecture |
| CN116363536A (en) * | 2023-05-31 | 2023-06-30 | 国网湖北省电力有限公司经济技术研究院 | Unmanned aerial vehicle inspection data-based power grid infrastructure equipment defect archiving method |
Non-Patent Citations (3)
| Title |
|---|
| Xiakun Sun ; Lan Liu,et.al.Classification for Remote Sensing Data With Improved CNN-SVM Method.IEEE.2019,全文. * |
| 多层局部感知卷积神经网络的高光谱图像分类;池涛;王洋;陈明;;四川大学学报(自然科学版)(01);全文 * |
| 池涛 ; 王洋 ; 陈明 ; .多层局部感知卷积神经网络的高光谱图像分类.四川大学学报(自然科学版).2020,(01),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116958752A (en) | 2023-10-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109992782B (en) | Legal document named entity identification method and device and computer equipment | |
| CN103049792B (en) | Deep-neural-network distinguish pre-training | |
| CN110750665A (en) | Open set domain adaptation method and system based on entropy minimization | |
| CN112000772B (en) | Sentence-to-semantic matching method based on semantic feature cube and oriented to intelligent question and answer | |
| CN112699247A (en) | Knowledge representation learning framework based on multi-class cross entropy contrast completion coding | |
| CN111477247A (en) | GAN-based voice countermeasure sample generation method | |
| CN114090780B (en) | Prompt learning-based rapid picture classification method | |
| CN112000770A (en) | Intelligent question and answer oriented sentence-to-sentence matching method based on semantic feature map | |
| CN111882042B (en) | Neural network architecture automatic search method, system and medium for liquid state machine | |
| CN114398976A (en) | Machine reading understanding method based on BERT and gate control type attention enhancement network | |
| CN113741886B (en) | Sentence-level program repairing method and system based on graph | |
| CN113641819A (en) | Multi-task sparse sharing learning-based argument mining system and method | |
| CN112463935B (en) | Open domain dialogue generation method and system with generalized knowledge selection | |
| CN116958752B (en) | Power grid infrastructure archiving method, device and equipment based on IPKCNN-SVM | |
| CN111414466A (en) | Multi-round dialogue modeling method based on depth model fusion | |
| CN113836903B (en) | Enterprise portrait tag extraction method and device based on situation embedding and knowledge distillation | |
| CN115840884A (en) | Sample selection method, device, equipment and medium | |
| CN113313250B (en) | Neural network training method and system adopting mixed precision quantization and knowledge distillation | |
| CN115422945A (en) | Rumor detection method and system integrating emotion mining | |
| CN115270795A (en) | Small sample learning-based named entity recognition technology in environmental assessment field | |
| CN114519104A (en) | Action label labeling method and device | |
| CN109815323B (en) | Human-computer interaction training question-answer generation algorithm | |
| CN109271494B (en) | System for automatically extracting focus of Chinese question and answer sentences | |
| Zhou et al. | Short-spoken language intent classification with conditional sequence generative adversarial network | |
| CN116991391B (en) | Code generation and deficiency supplementing method based on large language 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 |