CN114743052A - Hydropower signal monitoring method, system and terminal based on graph structure pooling - Google Patents

Hydropower signal monitoring method, system and terminal based on graph structure pooling Download PDF

Info

Publication number
CN114743052A
CN114743052A CN202210384779.6A CN202210384779A CN114743052A CN 114743052 A CN114743052 A CN 114743052A CN 202210384779 A CN202210384779 A CN 202210384779A CN 114743052 A CN114743052 A CN 114743052A
Authority
CN
China
Prior art keywords
node
graph
representing
nodes
signal monitoring
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210384779.6A
Other languages
Chinese (zh)
Inventor
罗玮
马宇
张铮
郭仕锐
黄飞虎
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guoneng Daduhe Big Data Service Co ltd
Original Assignee
Guoneng Daduhe Big Data Service Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Guoneng Daduhe Big Data Service Co ltd filed Critical Guoneng Daduhe Big Data Service Co ltd
Priority to CN202210384779.6A priority Critical patent/CN114743052A/en
Publication of CN114743052A publication Critical patent/CN114743052A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/217—Validation; Performance evaluation; Active pattern learning techniques

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a hydropower signal monitoring method, a hydropower signal monitoring system and a hydropower signal monitoring terminal based on graph structure pooling, which relate to the technical field of anomaly monitoring and have the technical scheme key points that: performing feature extraction on monitoring data of each station of the hydroelectric system to obtain a node feature vector; aggregating adjacent nodes by improving a self-attention method, and updating the characteristic vectors of the nodes after acquiring an attention coefficient; adopting new graph convolution to obtain global graph information to retain nodes in the front of the fitness score sorting; repeating the operation, splicing the output of each layer and obtaining a final graph representation through MLP; and (4) carrying out graph classification on the graph representation by using a cross entropy loss function, and judging to obtain a monitoring result. The method comprehensively considers the local structural information and the global structural information of the diagram, so that the difference characteristics of the diagram information are more accurately extracted, and the method can be used for judging whether the hydroelectric system normally operates.

Description

Hydropower signal monitoring method, system and terminal based on graph structure pooling
Technical Field
The invention relates to the technical field of anomaly monitoring, in particular to a hydropower signal monitoring method, a hydropower signal monitoring system and a hydropower signal monitoring terminal based on graph structure pooling.
Background
The signal monitoring of the hydropower station monitoring system is the core work of operation on duty, and the relationship among all monitoring points is abstracted into a graph network, namely, the nodes represent the monitoring points, and the node states are monitoring signals. Therefore, the task of judging whether the whole system is normally operated is abstracted into a classification problem of a graph network, wherein the classification of the task into a correct class represents that the system is normally operated, and the classification of the task into an error class represents that the system is abnormally operated.
The method applied to the graph classification task involves predicting the labels of the input graph by using a given graph structure and an initial node-level representation. The existing graph classification method mainly includes two types, namely global graph pooling, and the architecture relies on node representation through GNN learning, and then node information is aggregated to generate a graph representation, namely, the expression of the whole graph is obtained through graph convolution and then classified by combining MLP, wherein SET2SET, global-attribute and sortPool are the most typical. The other is a graph pooling method based on hierarchy, namely, each step of pooling deletes unimportant nodes in the graph, and only the important nodes are reserved each time, so that fine structure information of the graph is obtained. Such as DiffPool, TOP-K Pool, and SAGPOOl.
However, in the existing graph classification method, the global graph pooling is flat in nature, only focuses on the global information of the graph and lacks the capture of the sub-graph structure, and the hierarchical graph pooling method focuses too much on capturing the sub-graph information of the graph and ignores the grasp of the whole graph structure. And the differential information between nodes is often ignored. For example, if there is a difference in one node in the same molecular structure, the functions of the whole molecule will be greatly different. Therefore, how to design a hydroelectric signal monitoring method, system and terminal based on graph structure pooling, which can overcome the above defects, is a problem that is urgently needed to be solved at present.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention aims to provide a hydropower signal monitoring method, a hydropower signal monitoring system and a hydropower signal monitoring terminal based on graph structure pooling, which comprehensively consider local structure information and global structure information of a graph to enable the extraction of the difference characteristics of the graph information to be more accurate and can be used for judging whether a hydropower system operates normally.
The technical purpose of the invention is realized by the following technical scheme:
in a first aspect, there is provided a hydroelectric signal monitoring method based on graph structure pooling, comprising the steps of:
s1: performing feature extraction on monitoring data of each station of the hydroelectric system to obtain a node feature vector;
s2: aggregating adjacent nodes by improving a self-attention method, and updating the characteristic vectors of the nodes after acquiring an attention coefficient;
s3: adopting new graph convolution to obtain global graph information to retain nodes in the front of the fitness score sorting;
s4: repeating the steps S2-S3, splicing the output of each layer and obtaining a final graph representation through MLP;
s5: and (4) carrying out graph classification on the graph representation by using a cross entropy loss function, and judging to obtain a monitoring result.
Further, the extraction process of the node feature vector specifically includes:
performing convolution on the initial characteristic information of the nodes for vectorization;
a fully connected layer is applied to reduce dimensionality.
Further, the extraction expression of the node feature vector specifically includes:
Figure 856607DEST_PATH_IMAGE002
wherein,
Figure 356858DEST_PATH_IMAGE004
a node feature vector representing an ith node; FC denotes a full connection layer; BN represents a regularized network and is responsible for regularization of parameters of each layer of the network;
Figure 856104DEST_PATH_IMAGE006
and (4) representing characteristic information of each monitoring site of the ith hydropower station.
Further, the expression of the improved self-attention method is specifically as follows:
Figure 390990DEST_PATH_IMAGE008
wherein,
Figure 707702DEST_PATH_IMAGE010
representing the attention coefficient between node i and node j;
Figure 210096DEST_PATH_IMAGE012
representing an activation function;
Figure 496721DEST_PATH_IMAGE014
a parameter matrix for representing the self-adaptive retention of the characteristic information of the current node;
Figure 636847DEST_PATH_IMAGE016
a parameter matrix representing characteristics of a source node in the learning directed graph;
Figure 249094DEST_PATH_IMAGE017
a node feature vector representing an ith node;
Figure 864883DEST_PATH_IMAGE019
a node feature vector representing a jth node;
Figure 472975DEST_PATH_IMAGE021
indicating a splicing operation.
Further, the update formula of the node feature vector is specifically as follows:
Figure 982454DEST_PATH_IMAGE023
wherein,
Figure 985176DEST_PATH_IMAGE025
representing the updated node feature vector;
Figure 212895DEST_PATH_IMAGE027
representing the number of first-order neighbor nodes of node i.
Further, the expression of the new graph convolution is specifically as follows:
Figure 825011DEST_PATH_IMAGE029
wherein,
Figure 454575DEST_PATH_IMAGE031
representing the fitness score of node i;
Figure 18412DEST_PATH_IMAGE032
representing an activation function;
Figure 218580DEST_PATH_IMAGE034
、
Figure 916278DEST_PATH_IMAGE036
respectively representing the feature vectors of the nodes i and j; a denotes a graph adjacency matrix which is,
Figure 398685DEST_PATH_IMAGE038
indicating that two sites are directly adjacent; n (i) represents the number of first-order neighboring nodes of node i;
Figure 523636DEST_PATH_IMAGE039
a parameter matrix for representing the self-adaptive retention of the characteristic information of the current node;
Figure 335734DEST_PATH_IMAGE040
a parameter matrix representing characteristics of a source node in the learning directed graph;
Figure 322276DEST_PATH_IMAGE042
a parameter matrix representing characteristics of learning target nodes;
Figure 395274DEST_PATH_IMAGE044
representing a hamiltonian operation.
Further, the process of retaining the nodes in the front rank of the fitness score specifically includes:
multiplying the fitness vector by the feature matrix to obtain a global feature vector of the graph:
sorting the fitness scores of the global feature vectors of the graph, and giving out node indexes in front of the score sorting in the pooled graph;
and reserving nodes in the front row according to the node indexes and sorting the fitness scores.
Further, the expression shown in the final graph is specifically as follows:
Figure 940394DEST_PATH_IMAGE046
wherein Y represents the final diagram representation;
Figure 364422DEST_PATH_IMAGE048
、
Figure 154655DEST_PATH_IMAGE050
、
Figure 82159DEST_PATH_IMAGE052
respectively representing feature matrixes of nodes of layers 1, 2 and 3;
Figure 100002_DEST_PATH_IMAGE053
indicating a splicing operation.
In a second aspect, there is provided a hydroelectric signal monitoring system based on pooling of graph structures, comprising:
the characteristic extraction module is used for extracting the characteristics of the monitoring data of each station of the hydropower system to obtain a node characteristic vector;
the node updating module is used for aggregating adjacent nodes by improving a self-attention method, acquiring an attention coefficient and then updating a node feature vector;
the node retaining module is used for retaining the nodes in the front of the fitness score sorting by adopting the global graph information obtained by convolution of the new graph;
the output splicing module is used for repeating node updating and node retention, splicing the output of each layer and obtaining a final graph representation through MLP (Multi-level Linear programming);
and the classification judgment module is used for classifying the graph representation by using the cross entropy loss function and judging to obtain a monitoring result.
In a third aspect, a computer terminal is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the program, the adaptive structure-aware pooling method for graph-oriented classification according to any one of the first aspect is implemented.
Compared with the prior art, the invention has the following beneficial effects:
1. the self-adaptive structure perception pooling method for diagram classification can be used for judging whether a hydroelectric system operates normally or not, and comprehensively considers local structure information and global structure information of a diagram to enable the extraction of the difference characteristics of the diagram information to be more accurate for constructing a node unstructured characteristic relationship network;
2. in the aspect of the graph substructure, the expression of the node is updated in the first-order neighbors of the node by using an attention mechanism, so that the model can better capture the local characteristics of the graph;
3. in the aspect of capturing the global characteristics of the graph, the invention adopts a new graph convolution operation to score the nodes, retains the most important nodes in the graph, avoids the calculation of node clustering and soft distribution matrix, keeps the sparsity of graph calculation, improves the robustness of the model and further reduces the parameters.
Drawings
The accompanying drawings, which are included to provide a further understanding of the embodiments of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the principles of the invention. In the drawings:
FIG. 1 is an overall flow chart in an embodiment of the present invention;
fig. 2 is a block diagram of a system in an embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to examples and accompanying drawings, and the exemplary embodiments and descriptions thereof are only used for explaining the present invention and are not meant to limit the present invention.
Example 1: a hydroelectric signal monitoring method based on graph structure pooling, comprising the steps of:
s1: performing feature extraction on monitoring data of each station of the hydroelectric system to obtain a node feature vector;
s2: aggregating adjacent nodes by improving a self-attention method, and updating the characteristic vectors of the nodes after acquiring an attention coefficient;
s3: adopting new graph convolution to obtain global graph information to retain nodes in the front of the fitness score sorting;
s4: repeating the steps S2-S3, splicing the output of each layer and obtaining a final graph representation through MLP;
s5: and (4) carrying out graph classification on the graph representation by using a cross entropy loss function, and judging to obtain a monitoring result.
In the embodiment, the network inputs monitoring signals of each monitoring point with a characteristic format of 43 × 256, wherein 43 is the number of the monitoring points, and 256 is the signal characteristic of a single monitoring point; the model output is a graph label with a size of 43 x 1. The classification into the correct type indicates that the system operates normally, and the classification into the error type indicates that the system operates abnormally.
In step S1, a feature extraction operation is first performed. Recording a data set as
Figure 4372DEST_PATH_IMAGE054
，
Figure DEST_PATH_IMAGE055
Characteristic information of each monitoring station of the hydropower station,
Figure 728746DEST_PATH_IMAGE056
to representLabel information (label) of the site. Specifically, we perform convolution on the node initial feature information for vectorization and apply fully connected layers to reduce dimensionality.
The extraction expression of the node feature vector is specifically as follows:
Figure 40779DEST_PATH_IMAGE002
wherein,
Figure DEST_PATH_IMAGE057
a node feature vector representing an ith node; FC denotes a full connection layer; BN represents a regularized network and is responsible for regularization of parameters of each layer of the network;
Figure 806478DEST_PATH_IMAGE058
and (4) representing characteristic information of each monitoring site of the ith hydropower station. Note that feature extraction is performed on the entire sequence, and the weights are shared across all the series.
In step S2, according to the existing graph attention mechanism, such as Token2Token, Source2Token, that too much focuses on nodes with similar features but ignores the deficiency of nodes with more different features, the present invention proposes a new self-attention mechanism, and the expression of the improved self-attention method is specifically:
Figure DEST_PATH_IMAGE059
wherein,
Figure 257182DEST_PATH_IMAGE060
representing the attention coefficient between node i and node j;
Figure DEST_PATH_IMAGE061
represents an activation function, such as RELU;
Figure 235896DEST_PATH_IMAGE062
parameter for representing adaptive retention of characteristic information of current nodeA number matrix;
Figure DEST_PATH_IMAGE063
a parameter matrix representing characteristics of a source node in the learning directed graph;
Figure 554882DEST_PATH_IMAGE064
a node feature vector representing an ith node;
Figure DEST_PATH_IMAGE065
a node feature vector representing a jth node;
Figure 801187DEST_PATH_IMAGE066
indicating a splicing operation.
The updating formula of the node feature vector is specifically as follows:
Figure DEST_PATH_IMAGE067
wherein,
Figure 62272DEST_PATH_IMAGE068
representing the updated node feature vector;
Figure DEST_PATH_IMAGE069
representing the number of first-order neighbor nodes of node i.
In step S3, a new graph convolution operation is used to hold global graph information and to retain the most important nodes by repeating message passing on the graph in terms of global scoring of the nodes. The expression of the new graph convolution is specifically:
Figure 230080DEST_PATH_IMAGE070
wherein,
Figure DEST_PATH_IMAGE071
representing the fitness score of node i;
Figure 729588DEST_PATH_IMAGE072
representing an activation function;
Figure DEST_PATH_IMAGE073
、
Figure 564820DEST_PATH_IMAGE074
respectively representing the feature vectors of the nodes i and j; a denotes a graph adjacency matrix which is,
Figure DEST_PATH_IMAGE075
indicating that two sites are directly adjacent; n (i) represents the number of first-order neighboring nodes of node i;
Figure 747539DEST_PATH_IMAGE076
a parameter matrix for representing the self-adaptive retention of the characteristic information of the current node;
Figure DEST_PATH_IMAGE077
a parameter matrix representing characteristics of a source node in the learning directed graph;
Figure 432336DEST_PATH_IMAGE042
a parameter matrix representing characteristics of the learning target node;
Figure 906174DEST_PATH_IMAGE044
representing a hamiltonian operation.
And multiplying the fitness vector and the feature matrix to obtain a global feature vector of the graph. The specific expression is as follows:
Figure DEST_PATH_IMAGE079
wherein,
Figure DEST_PATH_IMAGE081
representing graph global feature vectors;
Figure DEST_PATH_IMAGE083
representing a fitness vector; x represents a feature matrix;
Figure DEST_PATH_IMAGE085
representing the broadcast hadamard product.
Use function
Figure DEST_PATH_IMAGE087
The fitness scores are sorted, and the score in the pooling graph G is given as the top
Figure DEST_PATH_IMAGE089
Node index of
Figure DEST_PATH_IMAGE091
As follows:
Figure DEST_PATH_IMAGE093
pool map
Figure DEST_PATH_IMAGE095
By selecting these tops
Figure 679134DEST_PATH_IMAGE096
And each node is formed to reserve part of the high-branch nodes. It avoids the computation of node clustering and soft allocation matrices to maintain the sparsity of graph computation. Pruned node feature matrix
Figure 95072DEST_PATH_IMAGE098
Given by:
Figure DEST_PATH_IMAGE100
wherein,
Figure 4515DEST_PATH_IMAGE091
for indexing the slices. The obtained node feature matrix
Figure DEST_PATH_IMAGE102
And putting the next layer of network structure to continue training.
In step S4, the final graph shows an expression specifically:
Figure DEST_PATH_IMAGE103
wherein Y represents the final diagram representation;
Figure DEST_PATH_IMAGE104
、
Figure DEST_PATH_IMAGE105
、
Figure DEST_PATH_IMAGE106
respectively representing feature matrixes of nodes of layers 1, 2 and 3;
Figure 153DEST_PATH_IMAGE066
indicating a splicing operation.
In step S5, a model output is calculated
Figure DEST_PATH_IMAGE108
Error from target value y
Figure DEST_PATH_IMAGE110
Cross entropy (Cross entropy) calculation was used:
Figure DEST_PATH_IMAGE112
wherein,
Figure DEST_PATH_IMAGE114
a target true value representing the node i,
Figure DEST_PATH_IMAGE116
and (5) outputting the model of the node i.
Updating network parameters according to the total loss:
Figure DEST_PATH_IMAGE118
in which
Figure DEST_PATH_IMAGE120
The parameters for minimizing the loss value in the gradient descent are selected and the model is updated.
Steps S2-S3 are repeated until the parameters converge or the maximum number of iterations is reached. The maximum number of iterations is generally set to 100 times, and early-stop technology can be adopted to finish training in advance (the waiting time is 20 times).
Example 2: a hydropower signal monitoring system based on graph structure pooling is used for realizing the method described in embodiment 1 and comprises a feature extraction module, a node updating module, a node reserving module, an output splicing module and a classification judgment module as shown in figure 2.
The characteristic extraction module is used for extracting characteristics of monitoring data of each station of the hydropower system to obtain a node characteristic vector; the node updating module is used for aggregating adjacent nodes by improving a self-attention method, acquiring an attention coefficient and then updating a node feature vector; the node retaining module is used for retaining the nodes in the front of the fitness score sorting by adopting the global graph information obtained by convolution of the new graph; the output splicing module is used for repeating node updating and node retaining, splicing the output of each layer and obtaining a final graph representation through MLP; and the classification judgment module is used for classifying the graph representation by using the cross entropy loss function and judging to obtain a monitoring result.
The working principle is as follows: in the invention, in order to construct a node unstructured feature relation network, the local structure information and the global structure information of the graph are comprehensively considered, so that the extraction of the graph information difference features is more accurate; in the aspect of the graph substructure, the expression of the node is updated in the first-order neighbors of the node by using an attention mechanism, so that the model can better capture the local features of the graph; in the aspect of capturing global features of the graph, a new graph convolution operation is adopted to score nodes, the most important nodes in the graph are reserved, node clustering and soft distribution matrix calculation are avoided, the sparsity of graph calculation is kept, the robustness of a model is improved, and parameters are further reduced.
As will be appreciated by one skilled in the art, 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 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.
The above embodiments are provided to further explain the objects, technical solutions and advantages of the present invention in detail, it should be understood that the above embodiments are merely exemplary embodiments of the present invention and are not intended to limit the scope of the present invention, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (10)

1. A hydroelectric signal monitoring method based on graph structure pooling is characterized by comprising the following steps:
s1: performing feature extraction on monitoring data of each station of the hydroelectric system to obtain a node feature vector;
s2: aggregating adjacent nodes by improving a self-attention method, and updating the characteristic vectors of the nodes after acquiring an attention coefficient;
s3: adopting new graph convolution to obtain global graph information to retain nodes in the front of the fitness score sorting;
s4: repeating the steps S2-S3, splicing the output of each layer and obtaining a final graph representation through MLP;
s5: and (4) classifying the graph representation by using a cross entropy loss function, and judging to obtain a monitoring result.
2. A hydroelectric signal monitoring method based on graph structure pooling according to claim 1, wherein the extraction process of the node feature vector comprises:
performing convolution on the initial characteristic information of the nodes for vectorization;
a fully connected layer is applied to reduce dimensionality.
3. A hydroelectric signal monitoring method based on graph structure pooling according to claim 1, wherein the extraction expression of the node feature vector is specifically:
Figure 655712DEST_PATH_IMAGE002
wherein,
Figure 899086DEST_PATH_IMAGE004
a node feature vector representing an ith node; FC denotes a full connection layer; BN represents a regularized network and is responsible for regularization of parameters of each layer of the network;
Figure 6719DEST_PATH_IMAGE006
and (4) representing characteristic information of each monitoring site of the ith hydropower station.
4. A graph structure pooling-based hydroelectric signal monitoring method according to claim 1 wherein the expression of said modified self-attention method is embodied as:
Figure 418240DEST_PATH_IMAGE008
wherein,
Figure 91667DEST_PATH_IMAGE010
representing the attention coefficient between node i and node j;
Figure 428976DEST_PATH_IMAGE012
representing an activation function;
Figure 758326DEST_PATH_IMAGE014
a parameter matrix for representing the self-adaptive retention of the characteristic information of the current node;
Figure 35855DEST_PATH_IMAGE016
a parameter matrix representing characteristics of a source node in the learning directed graph;
Figure 767050DEST_PATH_IMAGE017
a node feature vector representing the ith node;
Figure 25993DEST_PATH_IMAGE019
a node feature vector representing a jth node;
Figure 94837DEST_PATH_IMAGE021
indicating a splicing operation.
5. A hydroelectric signal monitoring method based on graph structure pooling according to claim 4, wherein the update formula of the node feature vector is specifically:
Figure 894166DEST_PATH_IMAGE023
wherein,
Figure 496180DEST_PATH_IMAGE025
representing the updated node feature vector;
Figure 988341DEST_PATH_IMAGE027
representing the number of first-order neighbor nodes of node i.
6. A graph structure pooling-based hydroelectric signal monitoring method according to claim 1 wherein the new graph convolution expression is specifically:
Figure 744813DEST_PATH_IMAGE029
wherein,
Figure DEST_PATH_IMAGE031
representing the fitness score of node i;
Figure 160882DEST_PATH_IMAGE032
representing an activation function;
Figure 69932DEST_PATH_IMAGE034
、
Figure 732995DEST_PATH_IMAGE036
respectively representing the feature vectors of the nodes i and j; a denotes a graph adjacency matrix which is,
Figure 979693DEST_PATH_IMAGE038
indicating that two sites are directly adjacent; n (i) represents the number of first-order neighboring nodes of node i;
Figure DEST_PATH_IMAGE039
a parameter matrix for representing the self-adaptive retention of the characteristic information of the current node;
Figure 933873DEST_PATH_IMAGE040
a parameter matrix representing characteristics of a source node in the learning directed graph;
Figure 759747DEST_PATH_IMAGE042
a parameter matrix representing characteristics of the learning target node;
Figure 842978DEST_PATH_IMAGE044
representing a hamiltonian operation.
7. A hydroelectric signal monitoring method based on graph structure pooling according to claim 1, wherein the process of retaining the nodes in the front of the fitness score ranking is specifically as follows:
multiplying the fitness vector by the feature matrix to obtain a global feature vector of the graph:
sorting the fitness scores of the global feature vectors of the graph, and giving out node indexes in front of the score sorting in the pooled graph;
and reserving nodes in the front row according to the node indexes and sorting the fitness scores.
8. A graph structure pooling-based hydroelectric signal monitoring method according to claim 1 wherein said final graph representation is expressed by the specific expression:
Figure 59196DEST_PATH_IMAGE046
wherein Y represents the final graph representation;
Figure 269598DEST_PATH_IMAGE048
、
Figure 435131DEST_PATH_IMAGE050
、
Figure DEST_PATH_IMAGE052
respectively representing feature matrixes of nodes of layers 1, 2 and 3;
Figure DEST_PATH_IMAGE053
indicating a splicing operation.
9. A hydroelectric signal monitoring system based on graph structure pooling, comprising:
the characteristic extraction module is used for extracting the characteristics of the monitoring data of each station of the hydropower system to obtain a node characteristic vector;
the node updating module is used for aggregating adjacent nodes by improving a self-attention method, acquiring an attention coefficient and then updating a node feature vector;
the node retention module is used for retaining nodes in the front of the fitness score ranking by adopting the new graph convolution to obtain global graph information;
the output splicing module is used for repeating node updating and node retaining, splicing the output of each layer and obtaining a final graph representation through MLP;
and the classification judgment module is used for classifying the graph representation by using the cross entropy loss function and judging to obtain a monitoring result.
10. A computer terminal comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor when executing the program implements the adaptive structure-aware pooling method of graph classification according to any of the claims 1-8.
CN202210384779.6A 2022-04-13 2022-04-13 Hydropower signal monitoring method, system and terminal based on graph structure pooling Pending CN114743052A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210384779.6A CN114743052A (en) 2022-04-13 2022-04-13 Hydropower signal monitoring method, system and terminal based on graph structure pooling

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210384779.6A CN114743052A (en) 2022-04-13 2022-04-13 Hydropower signal monitoring method, system and terminal based on graph structure pooling

Publications (1)

Publication Number Publication Date
CN114743052A true CN114743052A (en) 2022-07-12

Family

ID=82281171

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210384779.6A Pending CN114743052A (en) 2022-04-13 2022-04-13 Hydropower signal monitoring method, system and terminal based on graph structure pooling

Country Status (1)

Country Link
CN (1) CN114743052A (en)

Similar Documents

Publication Publication Date Title
CN109816032B (en) Unbiased mapping zero sample classification method and device based on generative countermeasure network
CN112685504B (en) Production process-oriented distributed migration chart learning method
JP7172612B2 (en) Data expansion program, data expansion method and data expansion device
CN111401149B (en) Lightweight video behavior identification method based on long-short-term time domain modeling algorithm
CN107783998A (en) The method and device of a kind of data processing
CN114064627A (en) Knowledge graph link completion method and system for multiple relations
CN111414863B (en) Enhanced integrated remote sensing image classification method
CN115293919A (en) Graph neural network prediction method and system oriented to social network distribution generalization
CN113159115A (en) Vehicle fine-grained identification method, system and device based on neural architecture search
CN104008177A (en) Method and system for rule base structure optimization and generation facing image semantic annotation
CN114743052A (en) Hydropower signal monitoring method, system and terminal based on graph structure pooling
CN115861306A (en) Industrial product abnormity detection method based on self-supervision jigsaw module
AU2021103314A4 (en) Remote sensing image scene classification method based on deep neural network automatic learning
CN115631057A (en) Social user classification method and system based on graph neural network
CN116225752A (en) Fault root cause analysis method and system for micro-service system based on fault mode library
CN113159976B (en) Identification method for important users of microblog network
AU2020103440A4 (en) A method for optimizing the convergence performance of data learning with minimal computational steps
CN114494284A (en) Scene analysis model and method based on explicit supervision area relation
Lim et al. Analyzing deep neural networks with noisy labels
CN112738724A (en) Method, device, equipment and medium for accurately identifying regional target crowd
CN113868460A (en) Image retrieval method, device and system
CN117688975B (en) Meteorological event prediction method and system based on evolution rule mining
Chang et al. Combining SOM and GA-CBR for flow time prediction in semiconductor manufacturing factory
CN118471327B (en) Genome prediction method and device based on genotype and environment interaction heterograms
CN118296387B (en) Model bi-directional iteration-based training sample optimization method and model bi-directional iteration-based training sample optimization system

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