A method and apparatus for time series plot modeling that integrates multi-parameter data analysis

Abstract

This invention discloses a time-series graph modeling method and apparatus that integrates multi-parameter data analysis. The method includes: acquiring multi-source information from a converter transformer, including operating condition data and multi-source sensor data; transforming the multi-source information into corresponding semantic feature vectors using a word vector model; performing linear transformations on each semantic feature vector and the concatenated feature vector obtained by splicing the semantic features to transform the multi-source information into a common feature space; processing the feature values ​​of each semantic feature vector after linear transformation using a knowledge graph technology based on the adaptive change of adjacency matrix weights with node features to obtain a knowledge graph corresponding to the multi-source information at different times; and processing the knowledge graph corresponding to the multi-source information at different times using a time-series graph attention neural network to analyze the changing patterns of the multi-source information.

Description

Technical Field

[0001] This invention relates to the field of data fusion processing technology, specifically to a time series graph modeling method and apparatus for fusing multi-parameter data analysis. Background Technology

[0002] With the large-scale development of the West-to-East Power Transmission Project, ultra-long-distance ultra-high voltage transmission lines have been continuously built, gradually forming a large ultra-high voltage power grid. As the hub of ultra-long-distance high-power DC transmission, the ultra-high voltage converter station rectifies the 500kV, 750kV or 1000kV AC power at the sending end through converter valves, converting it into ±800kV or ±1100kV DC power, realizing ultra-long-distance low-loss transmission of electrical energy. At the receiving end, the DC power is converted back to AC power through converter valves, and then transmitted to the next level AC substation.

[0003] Currently, the construction of smart grids and the enhancement of power supply reliability have risen to the level of national strategies. Power equipment condition monitoring and detection are gradually emerging. With the development of smart grids and the requirements for lean equipment management, converter stations are gradually forming centralized control or minimally staffed operation and maintenance models. However, UHV converter stations have a large number of devices, involving a significant amount of daily operation and maintenance work. If equipment malfunctions and causes a power outage, it will result in enormous economic losses, potentially causing severe power oscillations in the power grid, and even leading to regional power grid collapse. As a crucial piece of equipment in UHV converter stations, the converter transformer plays a vital role in the DC transmission system. Strengthening research on converter transformers is of great significance for the stable operation of converter stations.

[0004] In ultra-high voltage (UHV) converter stations, converter transformers experience approximately twice the failure rate of AC transformers due to DC bias and high-order harmonics. Meanwhile, issues such as instability in online digital systems and immature algorithms remain prominent. The biggest challenge for intelligent operation and maintenance (O&M) of converter stations is that data analysis technology lags behind the station's hardware configuration. From a data perspective, converter stations are equipped with numerous sensors for UHV converter transformers, collecting diverse observational data. The data collected within the converter transformers is varied, including operational condition data, various sensor parameters, historical data, and equipment maintenance case data. However, most online systems at the station analyze and process only certain types of data, neglecting the correlation between observational data to provide better results. Therefore, in converter station scenarios with such extreme fault conditions, new technologies need to be explored to support intelligent O&M of converter transformers.

[0005] Among related technologies, Chinese invention patent document CN113254663A describes a knowledge graph joint representation learning method that integrates graph convolution and translation models, including the following steps: 1) Constructing the corresponding direct adjacency matrix and indirect adjacency matrix based on the knowledge graph; 2) Designing a graph convolutional network, including an input layer, two hidden layers and an output layer, optimizing the attention coefficients of adjacent nodes to the center node, and obtaining the vector representation of the node by learning the structural information of the direct and indirect neighboring nodes; 3) Using a translation model to learn the semantic information of relations, obtaining the vector representation of entities and relations; 4) Integrating the graph convolutional network and the translation model, and obtaining the final vector representation of the knowledge graph through continuous iterative learning.

[0006] Chinese invention patent document CN113704500A discloses a knowledge graph community partitioning method based on graph neural networks, including the following steps: 1) constructing a knowledge graph and its adjacency matrix; 2) using graph neural networks to learn the representation of nodes in the knowledge graph to obtain the vector representation of the nodes; 3) calculating the similarity between nodes based on cosine similarity; 4) constructing an undirected weighted node relationship network based on the adjacency relationship between nodes of the same type and the node similarity; 5) setting modularity to evaluate the cohesion within the community; 6) partitioning the target nodes in the knowledge graph into communities using a community partitioning algorithm.

[0007] The knowledge graphs in the aforementioned solutions are limited to textual data, rather than being designed for the fusion of multi-source data. Summary of the Invention

[0008] The technical problem to be solved by this invention is how to achieve the fusion analysis of multimodal data of converter transformers and improve the intelligence level of the online monitoring system in the station.

[0009] The present invention solves the above-mentioned technical problems through the following technical means:

[0010] This paper proposes a time series graph modeling method that integrates multi-parameter data analysis, the method comprising:

[0011] Acquire multi-source information from the converter transformer, including operating condition data and multi-source sensor data;

[0012] The multi-source information is transformed using a word vector model to obtain the corresponding semantic feature vector;

[0013] Each of the semantic feature vectors and the concatenated feature vector obtained by concatenating the semantic features are subjected to linear transformation to transform the multi-source information into a common feature space.

[0014] Based on the knowledge graph technology that adapts the adjacency matrix weights to the node features, the feature values ​​of each semantic feature vector after linear transformation are processed to obtain the knowledge graph corresponding to the multi-source information at different times.

[0015] Based on a temporal graph attention neural network, the knowledge graph corresponding to the multi-source information at different times is processed to analyze the changing patterns of the multi-source information.

[0016] This invention addresses the heterogeneity and incomparability of features among different types of data by first mapping multimodal data to a common feature space. Then, it establishes correlations between different modalities in the multimodal data transformed in the previous step by constructing a knowledge graph. Finally, it uses graph neural networks combined with attention mechanisms to analyze the knowledge graph constructed from the multimodal data at each time point within a certain period, thereby analyzing the changing patterns of the multimodal data.

[0017] Further, the step of converting the multi-source information into corresponding semantic feature vectors via a word vector model includes:

[0018] The multi-source information is processed through the Word2Vec word vector model to obtain the corresponding semantic feature vectors {F1, F2, ..., F}. n}, where n represents the number of types of multi-source information.

[0019] Further, the step of performing linear transformations on each of the semantic feature vectors and the concatenated feature vector obtained by concatenating the semantic feature vectors to transform the multi-source information to a common feature space includes:

[0020] The semantic feature vectors are concatenated to obtain the concatenated feature vector F. s ;

[0021] The semantic feature vectors {F1,F2,…,F} are used to define the semantic feature vectors {F1,F2,…,F}. n} and the concatenated feature vector F s Perform linear transformations on each of them. The formula for the linear transformation is:

[0022]

[0023] In the formula, σ(·) represents the ReLU activation function; W m and b m F respectively m Transformation matrices and constants; F m This represents a semantic feature vector or a concatenated feature vector.

[0024] Furthermore, in the process of assigning each of the semantic feature vectors {F1,F2,…,F…} to… n} and the concatenated feature vector F sAfter performing linear transformations, it also includes:

[0025] The distance between each semantic feature vector and the concatenated feature vector is minimized using a loss function, which is:

[0026]

[0027] in, Represents the 2-norm; Let i represent the features after linear transformation of each semantic feature vector, i = 1, ..., n; This represents the features after the linear transformation of the concatenated feature vector;

[0028] The parameters of the linear transformation formula are iteratively optimized to minimize the loss value L, so as to map each semantic feature vector and the concatenated feature vector to the same space and transform them into the same dimension.

[0029] Furthermore, the knowledge graph technology based on the adaptive variation of adjacency matrix weights with node features processes the feature values ​​of each semantic feature vector after linear transformation to obtain the knowledge graph corresponding to the multi-source information at different times, including:

[0030] Each semantic feature vector, after linear transformation, corresponds to a node in the knowledge graph. Adjacency edges are constructed based on the connections between adjacent nodes, where the cosine similarity between the feature values ​​of adjacent nodes is used as the weight of the adjacency edges in the knowledge graph.

[0031]

[0032] In the formula: A ij This represents the element in the i-th row and j-th column of the adjacency matrix. Represents the feature vector of node i; This represents the feature vector of node j.

[0033] Furthermore, an attention mechanism is added to each graph attention neural network included in the temporal graph attention neural network. The attention mechanism is used to process the feature vectors h of the knowledge graph nodes i and j. i h j After splicing, with one dimensional vector Calculate the inner product, and then calculate the attention score of node j for node i:

[0034]

[0035] In the formula: || represents the concatenation operation; LeakyReLU represents the activation function; W represents the linear transformation matrix; Indicated Transpose; N i This represents the set of nodes associated with the i-th node;

[0036] Based on the attention score, the output feature quantity of the knowledge graph corresponding to the multi-source information at a certain moment is obtained, and the calculation formula of the output feature quantity is as follows:

[0037]

[0038] In the formula: σ represents the output feature; σ(·) represents the activation function ReLU.

[0039] Furthermore, the temporal graph attention neural network includes graph attention neural networks that share network parameters.

[0040] Furthermore, this invention also proposes a time series graph modeling device that integrates multi-parameter data analysis, the device comprising:

[0041] The multi-source information acquisition module is used to acquire multi-source information of the converter transformer, including operating condition data and multi-source sensor data.

[0042] The conversion module is used to convert the multi-source information into corresponding semantic feature vectors through a word vector model;

[0043] The linear transformation module is used to perform linear transformations on each of the semantic feature vectors and the concatenated feature vector obtained by concatenating the semantic features, so as to transform the multi-source information to a common feature space.

[0044] The knowledge graph module is used to process the feature values ​​of each semantic feature vector after linear transformation, based on the knowledge graph technology that adapts the adjacency matrix weights to the node features, to obtain the knowledge graph corresponding to the multi-source information at different times.

[0045] The analysis module is used to process the knowledge graph corresponding to the multi-source information at different times based on the time-series graph attention neural network, and to analyze the changing patterns of the multi-source information.

[0046] Furthermore, the present invention also proposes an apparatus comprising a memory and a processor; wherein the processor runs a program corresponding to the executable program code stored in the memory to implement the method described above.

[0047] Furthermore, the present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0048] The advantages of this invention are:

[0049] (1) This invention addresses the characteristics of existing converter transformer data, which often have multi-source heterogeneity and uncertain correlations among sources. It maps multi-source data to a common feature space and transforms it to the same dimension. Then, based on the correlation between different source data, it uses knowledge graph design to construct adaptive graph data. This graph construction is for the perception data of equipment such as power transformers, which is different from the knowledge graph which is limited to text data. This solution obtains the knowledge graph corresponding to multi-source data at each time step. Finally, it uses a graph neural network with a joint temporal mechanism and attention mechanism to perform data mining and fusion analysis on the knowledge graph containing multi-source information to obtain the changing patterns of multi-source data.

[0050] (2) The loss function proposed in this invention is used to minimize the distance between individual semantic feature vectors and concatenated feature vectors. By iteratively optimizing the parameters of the linear transformation, data with different and incomparable structures are mapped to the same space and transformed into the same dimension.

[0051] (3) This invention utilizes knowledge graphs to connect different modal data and calculates the adjacency matrix of adjacent nodes as the weight value of the association between adjacent modal data. It proposes an adaptive weight learning algorithm that incorporates an attention mechanism and establishes an adaptive weight knowledge graph. This method of assigning different weights to adjacent edges according to the degree of association is beneficial for the transformation data to learn its knowledge information according to the degree of association with other source data during the graph propagation process, thereby enhancing its own data features and improving diagnostic and early warning capabilities.

[0052] (4) The graph construction proposes a time-series graph with shared weights, which can model the correlation of heterogeneous sensing data from various sources of power equipment. The time-series graph neural network can analyze the data change trend based on the graph to achieve diagnosis and early warning.

[0053] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating a time series graph modeling method that integrates multi-parameter data analysis in one embodiment of the present invention.

[0055] Figure 2 This is a schematic diagram of the overall process of a time series graph modeling method that integrates multi-parameter data analysis in one embodiment of the present invention;

[0056] Figure 3 This is a diagram of a multi-source data mapping framework based on self-supervised learning in one embodiment of the present invention;

[0057] Figure 4 This is a schematic diagram of knowledge graph comparison based on adaptive weights in one embodiment of the present invention;

[0058] Figure 5 This is a schematic diagram of a timing graph attention neural network in one embodiment of the present invention;

[0059] Figure 6 This is a schematic diagram of the Attention mechanism in one embodiment of the present invention;

[0060] Figure 7 This is a schematic diagram of the structure of a time series graph modeling device that integrates multi-parameter data analysis in one embodiment of the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] like Figures 1 to 2 As shown, the first embodiment of the present invention proposes a time series graph modeling method that integrates multi-parameter data analysis. The method includes the following steps:

[0063] S10. Obtain multi-source information of the converter transformer, wherein the multi-source information includes operating condition data and multi-source sensor data;

[0064] It should be noted that the multi-source information of the converter transformer obtained in this embodiment includes acoustic, optical, electrical, gaseous, and chemical data such as voltage data, current data, gas data, and temperature data. These multi-source data can be collected through corresponding sensors.

[0065] S20. The multi-source information is transformed using a word vector model to obtain the corresponding semantic feature vector;

[0066] It should be noted that the word vector model used in this embodiment includes, but is not limited to, the Word2Vec word vector model, and this embodiment does not impose any specific limitations.

[0067] In this embodiment, the multi-source information is processed by the Word2Vec word vector model to obtain the corresponding semantic feature vector {F1,F2,…,F…}. n}, where n represents the number of types of multi-source information.

[0068] S30. Perform linear transformations on each of the semantic feature vectors and the concatenated feature vectors obtained by concatenating the semantic features to transform the multi-source information into a common feature space.

[0069] It should be noted that each modality of data in multi-source information corresponds to a semantic feature vector. By concatenating the semantic feature vectors, a concatenated feature vector is obtained. The feature obtained by linearly transforming the concatenated feature vector contains information on all data types in the multi-source information.

[0070] To address the multi-source heterogeneity of multimodal data, semantic feature vectors and concatenated feature vectors corresponding to multiple modalities are mapped to a common feature space, laying the foundation for subsequent data analysis and processing.

[0071] S40. Based on the knowledge graph technology that adapts the weight of the adjacency matrix to the node features, the feature values ​​of each semantic feature vector after linear transformation are processed to obtain the knowledge graph corresponding to the multi-source information at different times.

[0072] It should be noted that the knowledge graph described in this embodiment is different from the traditional knowledge graph. It uses the feature information of different source data to construct a knowledge graph whose adjacency matrix weights adapt to the node features. It can not only describe whether there is a relationship between nodes, but also reflect the degree of relationship between nodes.

[0073] S50. Based on the temporal graph attention neural network, the knowledge graph corresponding to the multi-source information at different times is processed to analyze the changing pattern of the multi-source information.

[0074] This embodiment addresses the characteristics of existing converter transformer data, which often exhibits multi-source heterogeneity and uncertain correlations among sources. It maps multi-source data to a common feature space, transforming it to the same dimension. Then, based on the relationships between different source data, it utilizes an adaptive graph data construction technique designed with knowledge graphs to obtain the knowledge graph corresponding to the multi-source data at each time step. Finally, it employs a graph neural network combining temporal and attention mechanisms to perform data mining and fusion analysis on the knowledge graph containing multi-source information, revealing the changing patterns of the multi-source data.

[0075] In one embodiment, such as Figure 3 As shown, step S30, which involves performing linear transformations on each semantic feature vector and the concatenated feature vector obtained by concatenating the semantic features, to transform the multi-source information to a common feature space, specifically includes the following steps:

[0076] S31. Concatenate the semantic feature vectors to obtain the concatenated feature vector F. s ;

[0077] S32. The semantic feature vectors {F1, F2, ..., F...} are... n} and the concatenated feature vector F s Perform linear transformations on each of them. The formula for the linear transformation is:

[0078]

[0079] In the formula, σ(·) represents the ReLU activation function; W m and b m F respectively m Transformation matrices and constants; F m This represents a semantic feature vector or a concatenated feature vector.

[0080] It should be noted that the feature vector F s The linear transformation process is similar to the feature fusion process of multi-source data features, therefore the transformed features Containing {F1,F2,…,F n Information from}

[0081] In one embodiment, after step S32, the method further includes the following steps:

[0082] The distance between each semantic feature vector and the concatenated feature vector is minimized using a loss function, which is:

[0083]

[0084] in, Represents the 2-norm; Let i represent the features after linear transformation of each semantic feature vector, i = 1, ..., n; This represents the features after the linear transformation of the concatenated feature vector;

[0085] The parameters of the linear transformation formula are iteratively optimized to minimize the loss value L, so as to map each semantic feature vector and the concatenated feature vector to the same space and transform them into the same dimension.

[0086] This embodiment innovatively proposes a loss function to minimize the distance between individual and concatenated semantic features. By iteratively optimizing the parameters of the linear transformation, it maps multimodal data with different and incomparable structures to the same space and transforms them into the same dimension. After a certain number of iterations of training, the feature vector is obtained. As a mapping result of multi-source data.

[0087] In one embodiment, step S40: Based on the knowledge graph technology where the adjacency matrix weights adapt to the node features, the feature values ​​of each semantic feature vector after linear transformation are processed to obtain the knowledge graph corresponding to the multi-source information at different times, including the following steps:

[0088] Each semantic feature vector, after linear transformation, corresponds to a node in the knowledge graph. Adjacency edges are constructed based on the connections between adjacent nodes, where the cosine similarity between the feature values ​​of adjacent nodes is used as the weight of the adjacency edges in the knowledge graph.

[0089]

[0090] In the formula: A ij This represents the element in the i-th row and j-th column of the adjacency matrix. Represents the feature vector of node i; This represents the feature vector of node j.

[0091] It should be noted that there are certain similarities and relationships between data from different sources. Existing knowledge graphs cannot reflect the relationships between data from different sources, which affects the stability of the data during propagation. Figure 4 As shown in (a), the adjacency matrix of existing knowledge graphs is a square matrix used to represent a finite graph. Each element represents whether there is an edge connecting the nodes. If there is an edge between nodes, the corresponding position in the adjacency matrix is ​​1; if there is no edge between nodes, the corresponding position is 0. This method of constructing the adjacency matrix can only describe whether there is a connection between nodes, but it cannot reflect the degree of connection between nodes, thus limiting the stability of data features during the propagation of the graph.

[0092] This embodiment addresses the problems existing in current knowledge graphs by utilizing feature information from different source data to construct a knowledge graph where the adjacency matrix weights adaptively change with node features. This adjacency matrix of the knowledge graph based on adaptive weights can intuitively represent the degree of association between nodes, such as... Figure 4 As shown in (b), by assigning different weights to adjacent edges according to their degree of association, it is beneficial for the flow-transformed data to learn its knowledge information according to the degree of association with other source data during the graph propagation process, thereby enhancing its own data characteristics and improving its diagnostic and early warning capabilities.

[0093] Furthermore, the cosine similarity ranges from [-1, 1], where a negative value indicates that the two connected rheostat data have a negative correlation, and a zero value indicates that the two data have almost no correlation.

[0094] In one embodiment, to achieve fault diagnosis and early warning, it is necessary to analyze multimodal data from different time periods. This embodiment proposes a temporal graph attention neural network to obtain data features from different time periods, such as... Figure 5 As shown, from time T1 to time T k The data is used to obtain the corresponding knowledge graph through adaptive weight-based knowledge graph technology, and then the data features I1, I2, ..., I at the corresponding time point are obtained through graph attention neural network. n .

[0095] Assuming time T i The data has N data types, and the feature vector of each node is h. i If the vector dimension is F, then at time T i A knowledge graph can be represented as h = {h1, h2, ..., h} N A linear transformation of the node feature vector h yields a new feature vector. Dimension The formula is as follows:

[0096]

[0097]

[0098]

[0099] In the formula: W is the matrix of the linear transformation; Let F represent the dimension.

[0100] Furthermore, an attention mechanism is added to each graph attention neural network within the temporal graph attention neural network. If node j is a neighbor of node i, then the attention mechanism can be used to calculate the importance of node j to node i, and the attention score is:

[0101] e ij =Attention(Wh i ,Wh j )

[0102]

[0103] The attention mechanism works as follows: it processes the feature vectors of nodes i and j... Piecing them together, and then with dimensional vector The formula for calculating the inner product is as follows:

[0104]

[0105] In the formula: || represents the concatenation operation; LeakyReLU represents the activation function; Indicated Transpose; N i This represents the set of nodes associated with the i-th node.

[0106] More intuitively, such as Figure 6 As shown, the feature vector of node i after the attention mechanism is as follows:

[0107]

[0108] It should be noted that during training, the temporal graph attention neural network model in this embodiment shares network parameters among the graph attention neural networks.

[0109] This embodiment first maps multimodal data to a common feature space, solving the problem of heterogeneity and incomparability of features among different types of data. Then, it constructs a knowledge graph to establish the correlation between different modalities of the multimodal data transformed in the previous step. Finally, it uses a graph neural network combined with an attention mechanism to analyze the knowledge graph constructed by the multimodal data at each time point within a certain period, thereby analyzing the data change patterns.

[0110] In addition, such as Figure 7 As shown, the second embodiment of the present invention proposes a time series graph modeling device that integrates multi-parameter data analysis, the device comprising:

[0111] The multi-source information acquisition module 10 is used to acquire multi-source information of the converter transformer, including operating condition data and multi-source sensor data.

[0112] The conversion module 20 is used to convert the multi-source information into corresponding semantic feature vectors through a word vector model;

[0113] The linear transformation module 30 is used to perform linear transformations on each of the semantic feature vectors and the concatenated feature vectors obtained by concatenating the semantic features, so as to transform the multi-source information to a common feature space.

[0114] Knowledge graph module 40 is used to process the feature values ​​of each semantic feature vector after linear transformation based on knowledge graph technology that adapts the adjacency matrix weights to the node features, and obtain the knowledge graph corresponding to the multi-source information at different times.

[0115] Analysis module 50 is used to process the knowledge graph corresponding to the multi-source information at different times based on the time-series graph attention neural network, and analyze the changing patterns of the multi-source information.

[0116] This embodiment addresses the characteristics of existing converter transformer data, which often exhibits multi-source heterogeneity and uncertain correlations among sources. It maps multi-source data to a common feature space, transforming it to the same dimension. Then, based on the relationships between different source data, it utilizes an adaptive graph data construction technique designed with knowledge graphs to obtain the knowledge graph corresponding to the multi-source data at each time step. Finally, it employs a graph neural network combining temporal and attention mechanisms to perform data mining and fusion analysis on the knowledge graph containing multi-source information, revealing the changing patterns of the multi-source data.

[0117] In one embodiment, the conversion module 20 uses the Word2Vec word vector model to process the multimodal data to obtain the corresponding semantic feature vectors {F1, F2, ..., F...}. n}, where n represents the number of types of multi-source information.

[0118] In one embodiment, the linear transformation module 30 includes:

[0119] The feature concatenation unit is used to concatenate the semantic feature vectors to obtain the concatenated feature vector F. s ;

[0120] Linear transformation unit, used to transform each of the semantic feature vectors {F1,F2,…,F…} n} and the concatenated feature vector F s Perform linear transformations on each of them. The formula for the linear transformation is:

[0121]

[0122] In the formula, σ(·) represents the ReLU activation function; W m and b m F respectively m Transformation matrices and constants; F m This represents a semantic feature vector or a concatenated feature vector.

[0123] It should be noted that the feature vector F s The linear transformation process is similar to the feature fusion process of multi-source data features, therefore the transformed features Containing {F1,F2,…,F n Information from}

[0124] In one embodiment, the apparatus further includes an iterative training module for minimizing the distance between each of the semantic feature vectors and the concatenated feature vector using a loss function, wherein the loss function is:

[0125]

[0126] in, Represents the 2-norm; Let i represent the features after linear transformation of each semantic feature vector, i = 1, ..., n; This represents the features after the linear transformation of the concatenated feature vector;

[0127] The parameters of the linear transformation formula are iteratively optimized to minimize the loss value L, so as to map each semantic feature vector and the concatenated feature vector to the same space and transform them into the same dimension.

[0128] In one embodiment, the knowledge graph module 40 is specifically used for:

[0129] Each semantic feature vector, after linear transformation, corresponds to a node in the knowledge graph. Adjacency edges are constructed based on the connections between adjacent nodes, where the cosine similarity between the feature values ​​of adjacent nodes is used as the weight of the adjacency edges in the knowledge graph.

[0130]

[0131] In the formula: A ij This represents the element in the i-th row and j-th column of the adjacency matrix. Represents the feature vector of node i; This represents the feature vector of node j.

[0132] In one embodiment, an attention mechanism is added to each graph attention neural network included in the temporal graph attention neural network. If node j is a neighbor of node i, the attention mechanism can be used to calculate the importance of node j to node i, and the attention score is:

[0133] e ij =Attention(Wh i ,Wh j )

[0134]

[0135] The attention mechanism works as follows: it processes the feature vectors of nodes i and j... Piecing them together, and then with dimensional vector The formula for calculating the inner product is as follows:

[0136]

[0137] In the formula: || represents the concatenation operation; LeakyReLU represents the activation function; Indicated Transpose; N i This represents the set of nodes associated with the i-th node.

[0138] More intuitively, such as Figure 6 As shown, the feature vector of node i after the attention mechanism is as follows:

[0139]

[0140] It should be noted that during training, the temporal graph attention neural network model in this embodiment shares network parameters among the graph attention neural networks.

[0141] It should be noted that other embodiments or implementation methods of the time series graph modeling device that integrates multi-parameter data analysis described in this invention can refer to the above-described method embodiments, and will not be repeated here.

[0142] Furthermore, a third embodiment of the present invention proposes an apparatus comprising a memory and a processor; wherein the processor reads executable program code stored in the memory to run a program corresponding to the executable program code, for implementing the time series graph modeling method for fusion multi-parameter data analysis as described above.

[0143] Furthermore, the present invention also proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the time series graph modeling method for fusion multi-parameter data analysis as described above.

[0144] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0145] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0146] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0147] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0148] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A time series graph modeling method integrating multi-parameter data analysis, characterized in that, The method includes: Acquire multi-source information from the converter transformer, including operating condition data and multi-source sensor data; The multi-source information is transformed using a word vector model to obtain the corresponding semantic feature vector; Each of the semantic feature vectors and the concatenated feature vector obtained by concatenating the semantic feature vectors are subjected to linear transformation to transform the multi-source information into a common feature space. Based on a knowledge graph technology that adapts the adjacency matrix weights to node features, the feature values ​​of each semantic feature vector after linear transformation are processed to obtain a knowledge graph corresponding to the multi-source information at different times. This includes mapping the feature values ​​of each semantic feature vector after linear transformation to a node in the knowledge graph, using the lines between adjacent nodes as adjacency edges, and using the cosine similarity between the feature values ​​of adjacent nodes as the adjacency edge weights of the knowledge graph. The adjacency matrix is ​​constructed as follows: In the formula: This represents the element in the i-th row and j-th column of the adjacency matrix. Represents the feature vector of node i; Represents the feature vector of node j; Based on a temporal graph attention neural network, the knowledge graph corresponding to the multi-source information at different times is processed to analyze the changing patterns of the multi-source information; After performing linear transformations on each of the semantic feature vectors and the concatenated feature vector obtained by concatenating the semantic feature vectors, the method further includes: The distance between each semantic feature vector and the concatenated feature vector is minimized using a loss function, which is: in, Represents the 2-norm; This represents the feature after linear transformation of each of the aforementioned semantic feature vectors. =1,......,n; This represents the features after the linear transformation of the concatenated feature vector; The parameters of the linear transformation formula are iteratively optimized to minimize the loss value L, so as to map each semantic feature vector and the concatenated feature vector to the same space and transform them into the same dimension.

2. The time series graph modeling method integrating multi-parameter data analysis as described in claim 1, characterized in that, The step of converting the multi-source information into corresponding semantic feature vectors via a word vector model includes: The multi-source information is processed using the Word2Vec word vector model to obtain the corresponding semantic feature vectors. ,in, This indicates the number of types of multi-source information.

3. The time series graph modeling method integrating multi-parameter data analysis as described in claim 1, characterized in that, The step of performing linear transformations on each of the semantic feature vectors and the concatenated feature vector obtained by concatenating the semantic feature vectors to transform the multi-source information to a common feature space includes: The semantic feature vectors are concatenated to obtain the concatenated feature vector. ; Each of the semantic feature vectors and the concatenated feature vector Perform linear transformations on each of them. The formula for the linear transformation is: In the formula, Represents the activation function ReLU; and They are respectively The transformation matrix and constants; This represents a semantic feature vector or a concatenated feature vector.

4. The time series graph modeling method integrating multi-parameter data analysis as described in claim 1, characterized in that, The temporal graph attention neural network incorporates an attention mechanism into each graph attention neural network, the attention mechanism being used to process the feature vectors of knowledge graph nodes i and j. , After splicing, it is combined with a 2 dimensional vector Calculate the inner product, and then calculate the attention score of node j for node i: In the formula: || represents the concatenation operation; LeakyReLU represents the activation function; Represents a linear transformation matrix; Indicated Transpose; This represents the set of nodes associated with the i-th node; Based on the attention score, the output feature quantity of the knowledge graph corresponding to the multi-source information at a certain moment is obtained, and the calculation formula of the output feature quantity is as follows: In the formula: Indicates the output feature quantity; This represents the activation function ReLU.

5. The time series graph modeling method integrating multi-parameter data analysis as described in claim 1, characterized in that, The temporal graph attention neural network includes graph attention neural networks that share network parameters.

6. A time series graph modeling device integrating multi-parameter data analysis, characterized in that, The device includes: The multi-source information acquisition module is used to acquire multi-source information of the converter transformer, including operating condition data and multi-source sensor data. The conversion module is used to convert the multi-source information into corresponding semantic feature vectors through a word vector model; The linear transformation module is used to perform linear transformations on each of the semantic feature vectors and the concatenated feature vectors obtained by concatenating the semantic feature vectors, so as to transform the multi-source information to a common feature space. The knowledge graph module is used to process the feature values ​​of the semantic feature vectors after linear transformation, based on the knowledge graph technology of adaptively changing adjacency matrix weights with node features, to obtain the knowledge graph corresponding to the multi-source information at different times. This includes mapping the feature values ​​of each semantic feature vector after linear transformation to a node in the knowledge graph, using the connections between adjacent nodes as adjacency edges, and using the cosine similarity between the feature values ​​of adjacent nodes as the adjacency edge weights of the knowledge graph. The adjacency matrix is ​​constructed as follows: In the formula: This represents the element in the i-th row and j-th column of the adjacency matrix. Represents the feature vector of node i; Represents the feature vector of node j; The analysis module is used to process the knowledge graph corresponding to the multi-source information at different times based on the time-series graph attention neural network, and analyze the changing patterns of the multi-source information. The device further includes an iterative training module, used to minimize the distance between each semantic feature vector and the concatenated feature vector using a loss function, wherein the loss function is: in, Represents the 2-norm; This represents the feature after linear transformation of each of the aforementioned semantic feature vectors. =1,......,n; This represents the features after the linear transformation of the concatenated feature vector; The parameters of the linear transformation formula are iteratively optimized to minimize the loss value L, so as to map each semantic feature vector and the concatenated feature vector to the same space and transform them into the same dimension.

7. A device, characterized in that, The device includes a memory and a processor; wherein the processor runs a program corresponding to the executable program code by reading executable program code stored in the memory, so as to implement the method as described in any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-5.

Images