Method and device for constructing high-order tensor network of large-scale power grid, equipment and medium

Abstract

The present application relates to the technical field of smart grid, and discloses a large-scale power grid high-order tensor network construction method, device, equipment and medium, comprising: acquiring a plurality of category attribute sets of heterogeneous nodes in a large-scale power grid; performing feature extraction on the plurality of category attribute sets to obtain attribute features of the heterogeneous nodes; establishing multiple feature subspaces according to the attribute features of the heterogeneous nodes by using a deep hash mapping model; aligning the multiple feature subspaces to a unified feature dimension by using a breadth learning strategy; calculating distance measurement values between the heterogeneous nodes according to attribute feature vectors of the unified feature dimension by using a distance measurement method; determining a base tensor for representing an interaction relationship of the heterogeneous nodes according to the attribute feature vectors of the unified feature dimension and a weight value of the interaction relationship of the heterogeneous nodes; and constructing a large-scale power grid high-order tensor network according to the base tensor and the weight value of the interaction relationship of the heterogeneous nodes by using a tensor multiplication operation rule. The problem of lacking a model for efficiently representing a large-scale power grid is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart grid, and particularly relates to a large-scale power grid high-order tensor network construction method, device, equipment and medium. BACKGROUND

[0002] With the access of large-scale distributed resources, the power grid gradually develops into a high-order network with composite subjects, fast time-varying and heterogeneous interaction. The power grid data has characteristics such as complex and diverse types, different forms of expression, heterogeneous node attributes, and frequent node interaction. The large-scale power grid is a high-order, high-dimensional, heterogeneous and incomplete complex digital coupling space, covering power grids of various voltage levels, and is composed of power plants, substations, primary and secondary devices of various voltage levels and their connection networks. The calculation nodes cover various entities such as power equipment, power loads and power users. Due to the large number of entity nodes in the large-scale power grid, the different types of data, the difficulty of data fusion, and the large-scale calculation, the current model construction method for the large-scale power grid cannot efficiently represent the large-scale power grid, and there are problems such as slow analysis and calculation speed, large computing energy consumption, and the like, which hinder the development of the power grid. Therefore, there is a lack of an efficient large-scale power grid model in the prior art. SUMMARY

[0003] Therefore, the present application provides a large-scale power grid high-order tensor network construction method, device, equipment and medium to solve the problem of lacking an efficient large-scale power grid model in the prior art.

[0004] In a first aspect, the present application provides a large-scale power grid high-order tensor network construction method, comprising: obtaining a plurality of category attribute sets corresponding to heterogeneous nodes in a large-scale power grid; performing feature extraction on the plurality of category attribute sets to obtain heterogeneous node attribute features; using a deep hash mapping model to establish a plurality of feature subspaces according to the heterogeneous node attribute features, any one of the plurality of feature subspaces corresponding to a type of heterogeneous node interaction relationship; using a breadth learning strategy to align the plurality of feature subspaces to a unified feature dimension to obtain attribute feature vectors of the unified feature dimension; using a distance measurement method to calculate distance measurement values between the heterogeneous nodes according to the attribute feature vectors of the unified feature dimension, the distance measurement values being used to determine weights of the heterogeneous node interaction relationship; determining a base tensor used to represent the heterogeneous node interaction relationship according to the attribute feature vectors of the unified feature dimension and the weights of the heterogeneous node interaction relationship; and using a tensor multiplication operation rule to construct a large-scale power grid high-order tensor network according to the base tensor and the weights of the heterogeneous node interaction relationship.

[0005] In the embodiment of the present application, the tensor network is introduced to represent the large-scale power grid, the deep hash mapping model is used to establish multiple feature subspaces according to the extracted heterogeneous node attribute features, the distance measurement method is used to calculate the weight of the heterogeneous node interaction relationship according to the aligned multiple feature subspaces, then the base tensor representing the interaction relationship is determined, and the high-order tensor network representing the large-scale power grid is constructed using the base tensor. Since the tensor network has the distributed storage and parallel processing capability for high-dimensional heterogeneous features and complex correlation, the use of the tensor network for modeling the large-scale power grid realizes the effective fusion of high-dimensional and complex features, achieves the effect of improving the scalability of the large-scale power grid model and reducing the computational complexity, and solves the problem of lacking an efficient model for representing the large-scale power grid in the related art.

[0006] In an optional implementation, the method further comprises: obtaining heterogeneous power grid data; compressing the large-scale power grid high-order tensor network based on a high-order singular value analysis framework, a bit plane, a run-length encoding, and an arithmetic encoding; constructing a spatiotemporal mechanism nonlinear representation model corresponding to the heterogeneous power grid data; and generating a large-scale power grid high-order tensor network containing a dynamic evolution mechanism based on the spatiotemporal mechanism nonlinear representation model and the compressed large-scale power grid high-order tensor network.

[0007] In the embodiment of the present application, by constructing the spatiotemporal mechanism nonlinear representation model, the dynamic evolution mechanism of the large-scale power grid is accurately modeled, and the purpose of further improving the representation capability of the large-scale power grid high-order tensor network is achieved.

[0008] In an optional implementation, the feature extraction is performed on the multiple-category attribute sets to obtain the heterogeneous node attribute features, including: extracting features corresponding to the multiple-category attribute sets; storing the features corresponding to any one of the multiple-category attribute sets as a second-order tensor; determining multiple second-order tensors corresponding to the multiple-category attribute sets; and taking the multiple second-order tensors as the heterogeneous node attribute features.

[0009] In the embodiment of the present application, the second-order tensor is used to store the multiple-category attribute sets of the heterogeneous nodes, the purpose of merging the heterogeneous node attributes of the large-scale power grid is achieved, and the efficient representation of the heterogeneous node attribute features is achieved.

[0010] In an optional implementation, the deep hash mapping model includes a common hash encoding module and multiple hash encoding sub-modules, and the deep hash mapping model is used to establish multiple feature subspaces according to the heterogeneous node attribute features, any one of the multiple feature subspaces corresponding to one type of heterogeneous node interaction relationship, including: mapping the heterogeneous node attribute features to a common attribute feature space by using the common hash encoding module; and establishing one feature subspace according to the common attribute feature space by using each hash encoding sub-module.

[0011] In the embodiment of the present application, the multi-feature subspace is established using the deep hash mapping, the alignment and separation of the large-scale power grid heterogeneous node attribute space are realized, and the purpose of using the feature subspace to represent the interaction relationship of a class of heterogeneous nodes is achieved.

[0012] In an optional implementation, the attribute feature vector of the unified feature dimension includes the attribute feature vectors of the unified feature dimension corresponding to the heterogeneous nodes and the neighbor nodes thereof, the distance measurement value between the heterogeneous nodes is calculated according to the attribute feature vectors of the unified feature dimension by using a distance measurement method, and the distance measurement value is used to determine the weight value of the interaction relationship of the heterogeneous nodes, including: calculating the distance measurement value between the heterogeneous nodes according to the attribute feature vectors of the unified feature dimension by using a distance measurement method; merging the attribute feature vectors of the unified feature dimension corresponding to the heterogeneous nodes and the neighbor nodes thereof based on the distance measurement value by using graph convolution to obtain new attribute feature vectors of the heterogeneous nodes; and determining the weight value of the interaction relationship of the heterogeneous nodes according to the new attribute feature vectors of the heterogeneous nodes.

[0013] In the embodiment of the present application, by determining the weight value of the interaction relationship of the heterogeneous nodes, the purpose of unifying the heterogeneous nodes and the interaction relationship of the heterogeneous nodes is achieved.

[0014] In an optional implementation, the spatiotemporal mechanism nonlinear representation model corresponding to the heterogeneous power grid data is constructed, including: obtaining multi-factor spatiotemporal neighborhood features; performing joint modeling of spatiotemporal neighborhood information of the heterogeneous power grid data according to the multi-factor spatiotemporal neighborhood features by using a regularization method to obtain a power grid topology graph in a spatiotemporal dimension; analyzing the power grid topology graph in the spatiotemporal dimension by using a nonlinear Kalman filter state transition estimation method to determine a state transition mode of the power grid topology graph; and constructing the spatiotemporal mechanism nonlinear representation model corresponding to the heterogeneous power grid data based on the state transition mode of the power grid topology graph.

[0015] In the embodiment of the present application, the spatiotemporal mechanism nonlinear representation model corresponding to the heterogeneous power grid data is constructed according to the multi-factor spatiotemporal neighborhood features, the potential mode of the large-scale power grid time series is analyzed, and the purpose of integrating the spatiotemporal characteristics into the large-scale power grid high-order tensor network is achieved.

[0016] In an optional implementation, the spatiotemporal neighborhood information of the heterogeneous power grid data is jointly modeled according to the multi-factor spatiotemporal neighborhood features by using a regularization method to obtain a power grid topology graph in a spatiotemporal dimension, including: constructing a corresponding undirected weighted graph according to the heterogeneous power grid data; generating single spatiotemporal neighborhood information according to the undirected weighted graph by using a Laplacian operator and a time difference; and performing joint modeling of the spatiotemporal neighborhood information of the heterogeneous power grid data according to the single spatiotemporal neighborhood information by using a multi-index joint constraint regularization method to obtain the power grid topology graph in the spatiotemporal dimension.

[0017] In the embodiment of the present application, by generating single spatio-temporal neighborhood information and then using a multi-index joint constraint regularization method for joint modeling, the problem of strong and weak differences in single time or space characteristics is solved, and the representation capability of large-scale power grid high-order tensor network is further improved.

[0018] In a second aspect, the present application provides a large-scale power grid high-order tensor network construction device, comprising: an attribute set acquisition module for acquiring a plurality of category attribute sets corresponding to heterogeneous nodes in a large-scale power grid; a feature extraction module for extracting features from the plurality of category attribute sets to obtain heterogeneous node attribute features; a multiple feature subspace establishment module for establishing multiple feature subspaces according to the heterogeneous node attribute features using a deep hash mapping model, any one of the multiple feature subspaces corresponding to a type of heterogeneous node interaction relationship; a feature alignment module for aligning the multiple feature subspaces to a unified feature dimension using a breadth learning strategy to obtain attribute feature vectors of the unified feature dimension; a distance metric value calculation module for calculating distance metric values between the heterogeneous nodes according to the attribute feature vectors of the unified feature dimension using a distance metric method, the distance metric values being used to determine the weight values of the heterogeneous node interaction relationships; a base tensor determination module for determining a base tensor for representing the heterogeneous node interaction relationships according to the attribute feature vectors of the unified feature dimension and the weight values of the heterogeneous node interaction relationships; and a power grid high-order tensor network construction module for constructing a large-scale power grid high-order tensor network according to the base tensor and the weight values of the heterogeneous node interaction relationships using a tensor multiplication operation rule.

[0019] In an optional implementation, the device further comprises: a power grid data acquisition module for acquiring heterogeneous power grid data; a compression module for compressing the large-scale power grid high-order tensor network based on a high-order singular value analysis framework, a bit plane, a run-length encoding, and an arithmetic encoding; a mechanism model construction module for constructing a spatio-temporal mechanism nonlinear representation model corresponding to the heterogeneous power grid data; and a dynamic network generation module for generating a large-scale power grid high-order tensor network containing a dynamic evolution mechanism based on the spatio-temporal mechanism nonlinear representation model and the compressed large-scale power grid high-order tensor network.

[0020] In an optional implementation, the feature extraction module comprises: a feature extraction unit for extracting features corresponding to the plurality of category attribute sets; a feature storage unit for storing the extracted features corresponding to any one of the plurality of category attribute sets as a second-order tensor; a tensor determination unit for determining a plurality of second-order tensors corresponding to the plurality of category attribute sets; and a node attribute feature generation unit for taking the plurality of second-order tensors as the heterogeneous node attribute features.

[0021] In an optional implementation, the deep hash mapping model comprises a common hash coding module and a plurality of hash coding sub-modules, and the multiple feature subspace establishing module comprises: a feature mapping unit configured to map the heterogeneous node attribute features to a common attribute feature space by using the common hash coding module; and a subspace establishing unit configured to establish a one-dimensional feature subspace according to the common attribute feature space by using each hash coding sub-module.

[0022] In an optional implementation, the attribute feature vector of the unified feature dimension comprises attribute feature vectors of the unified feature dimension corresponding to the heterogeneous nodes and the neighbor nodes thereof, the distance metric value calculation module comprises: a calculation unit configured to calculate distance metric values between the heterogeneous nodes according to the attribute feature vectors of the unified feature dimension by using a distance metric method; a merging unit configured to merge the attribute feature vectors of the unified feature dimension corresponding to the heterogeneous nodes and the neighbor nodes thereof based on the distance metric values by using graph convolution, to obtain new attribute feature vectors of the heterogeneous nodes; and a determination unit configured to determine the weight values of the interaction relationships between the heterogeneous nodes according to the new attribute feature vectors of the heterogeneous nodes.

[0023] In an optional implementation, the mechanism model constructing module comprises: a spatio-temporal feature acquisition unit configured to acquire multi-factor spatio-temporal neighborhood features; a joint modeling unit configured to perform joint modeling of spatio-temporal neighborhood information of the heterogeneous power grid data according to the multi-factor spatio-temporal neighborhood features by using a regularization method, to obtain a power grid topology graph in a spatio-temporal dimension; an analysis unit configured to analyze the power grid topology graph in the spatio-temporal dimension by using a nonlinear Kalman filter state transition estimation method, to determine a state transition mode of the power grid topology graph; and a mechanism model constructing unit configured to construct a spatio-temporal mechanism nonlinear representation model corresponding to the heterogeneous power grid data based on the state transition mode of the power grid topology graph.

[0024] In an optional implementation, the joint modeling unit comprises: a graph constructing sub-unit configured to construct a corresponding undirected weighted graph according to the heterogeneous power grid data; a spatio-temporal neighborhood information generating sub-unit configured to generate single spatio-temporal neighborhood information according to the undirected weighted graph by using a Laplacian operator and a time difference; and a joint modeling sub-unit configured to perform joint modeling of spatio-temporal neighborhood information of the heterogeneous power grid data according to the single spatio-temporal neighborhood information by using a multi-index joint constraint regularization method, to obtain a power grid topology graph in a spatio-temporal dimension.

[0025] In a third aspect, the present application provides a computer device, comprising: a memory and a processor, which are communicatively connected with each other, and the memory stores computer instructions; the processor executes the computer instructions, thereby implementing the large-scale power grid high-order tensor network construction method of the first aspect or any of the corresponding embodiments thereof.

[0026] In a fourth aspect, the present application provides a computer readable storage medium, which stores computer instructions for causing a computer to execute the large-scale power grid high-order tensor network construction method of the first aspect or any of the corresponding embodiments thereof. BRIEF DESCRIPTION OF DRAWINGS

[0027] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings required to be used in the specific embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0028] Figure 1 is a flowchart of the large-scale power grid high-order tensor network construction method according to an embodiment of the present application;

[0029] Figure 2 is a flowchart of another large-scale power grid high-order tensor network construction method according to an embodiment of the present application;

[0030] Figure 3 is a schematic diagram of a heterogeneous power grid data corresponding undirected weighted graph according to an embodiment of the present application;

[0031] Figure 4 is a schematic diagram of the overall framework of the large-scale power grid high-order tensor network construction method according to an embodiment of the present application;

[0032] Figure 5 is a flowchart of the large-scale high-order tensor network feature learning method according to an embodiment of the present application;

[0033] Figure 6 is a heterogeneous space merging diagram according to an embodiment of the present application;

[0034] Figure 7 is a structural block diagram of the large-scale power grid high-order tensor network construction device according to an embodiment of the present application;

[0035] Figure 8 is a hardware structure schematic diagram of the computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0036] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0037] It should be noted that, in the description of the present application, the terms "first", "second" and the like are used to distinguish similar objects, and do not necessarily have to describe a particular order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices. The orientation or positional relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer" and the like is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. The terms "mount", "connect", "connect" should be understood broadly, for example, it can be fixedly connected, or detachably connected, or integrally connected; it can be mechanically connected, or electrically connected; it can be directly connected, or indirectly connected through an intermediate medium, or it can be the internal communication of two elements, it can be wireless connection, or wired connection. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0038] In addition, if "and / or" appears in the present application, it includes three parallel schemes, for example, "A and / or B" includes A scheme, or B scheme, or A and B scheme. In addition, the technical solutions of each embodiment can be combined with each other, but it must be based on the fact that the technical solutions can be realized by those skilled in the art, and when the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, and is not within the protection scope of the present application.

[0039] With the instant access of large-scale distributed resources, the power grid presents new characteristics such as complex topology, intensified operation state fluctuation, and bidirectional interaction between main grid and distribution network, gradually developing into a high-order network with complex main body, fast time-varying, and heterogeneous interaction. Large-scale power grid is a complex digital coupling space with high order, high dimension, heterogeneity, and incompleteness, with characteristics such as complex and diverse data types, different expression forms, heterogeneous node attributes, and frequent node interaction. The rapid development of power grid brings multiple technical challenges such as data fusion difficulty, large-scale computation, strong analysis timeliness, and high optimization demand. For example, large-scale power grid can cover multiple power grids at different voltage levels, including power plants, substations, and devices at different voltage levels and their connected networks. The calculation node types include power equipment, power load, and power users. The number of entity nodes in large-scale power grid calculation is huge and diverse, such as the number of transformer models in the production management system. The number of power grid calculation objects is also huge. If traditional graph calculation methods are used, it will still face great pressure when facing millions of topology nodes and hundreds of data fusion analysis, with a response delay of 5s / 10,000 nodes and poor analysis model scalability. The current technical route has obvious shortcomings in expressing and utilizing the high-order, complex, heterogeneous, and incomplete panoramic power grid coupling space, with high model complexity and slow analysis speed.

[0040] Accurate characterization of large-scale power grid is the basis for intelligent tasks such as load forecasting and fault warning. Existing characterization methods such as matrix decomposition and graph neural networks have the following shortcomings: (1) Continuous network space needs to be divided, logical mapping is complex, and time and topology information is easily lost, resulting in damage to the complete time and space power grid structure; (2) The incompleteness, computational complexity, and scalability of large-scale power grid data are not considered, and the analysis and calculation speed is slow; (3) The increase in network order leads to scattered data storage, and heterogeneous data increases modeling difficulty, resulting in low model representation efficiency; (4) Power grid calculation involves complex and diverse device description data, different devices have different data description forms, and the association between different device nodes lacks a unified description; (5) The large number of power grid topology node types makes it difficult to maintain the network topology. Therefore, how to use a more concise and efficient high-order network model to describe the complex power grid structure of multi-agent interaction, so as to meet the application requirements of better model, faster calculation, more accurate analysis, and less computing power consumption, is a key technical problem that needs to be solved in the future development of power grid. Considering that tensor network has good scalability and can completely save the time and space patterns of power grid, how to use high-order tensor network to completely represent the heterogeneous node attributes, heterogeneous node interaction, and complex mapping logic of large-scale power grid, to model the spatial structure information, heterogeneous information, and evolution mechanism of large-scale power grid as a whole, and to realize accurate panoramic modeling and representation of large-scale power grid, is a key difficulty that needs to be broken through in establishing large-scale power grid characterization methods.

[0041] According to the embodiment of the present application, a large-scale power grid high-order tensor network construction method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a terminal such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0042] A large-scale power grid high-order tensor network construction method is provided in this embodiment, which can be used in the terminal such as the central processing unit, server, etc. Figure 1 The flowchart of the large-scale power grid high-order tensor network construction method according to the embodiment of the present application is shown in FIG. 1, which includes the following steps: Figure 1

[0043] In step S101, a plurality of category attribute sets corresponding to heterogeneous nodes in a large-scale power grid are obtained. Optionally, the large-scale power grid includes a large number of heterogeneous nodes, i.e., nodes of different types and attributes. The heterogeneous nodes can be a plurality of categories of devices that constitute the power grid, such as generators, buses, loads, transformers, and lines. The heterogeneous nodes also include a plurality of different attributes, such as geographic location, voltage level, construction age, etc. The attribute set is a set of a plurality of attributes.

[0044] In step S102, feature extraction is performed on the plurality of category attribute sets to obtain heterogeneous node attribute features. Optionally, the plurality of category attribute sets obtained in step S101 are subjected to feature extraction using machine learning or other methods to obtain heterogeneous node attribute features. The extracted heterogeneous node attribute features can be stored in the form of a tensor or a matrix.

[0045] In step S103, a deep hash mapping model is used to establish a plurality of feature subspaces according to the heterogeneous node attribute features, and any feature subspace corresponds to an interaction relationship of a type of heterogeneous nodes. Optionally, the deep hash mapping model is used to map the heterogeneous node attribute features of step S102 to a low-dimensional common feature space. Feature extraction is then performed in the common feature space to obtain a plurality of feature subspaces. The attribute features on each feature subspace are fed back to the heterogeneous node relationship, which represents a specific interaction, thereby achieving the purpose of representing the interaction logic of the large-scale power grid heterogeneous nodes using the plurality of feature subspaces.

[0046] In step S104, a breadth learning strategy is used to align the plurality of feature subspaces to a unified feature dimension to obtain attribute feature vectors of the unified feature dimension. Optionally, the core problem considered by the breadth learning strategy is to mine more valuable information based on the fusion of different types of data. Illustratively, the plurality of subspaces can be aligned to a unified feature dimension based on the breadth learning strategy using an autoencoder, thereby achieving the unification of the attribute feature vector dimension. ​

[0047] In step S105, a distance metric method is used to calculate the distance metric value between the heterogeneous nodes according to the attribute feature vectors of the unified feature dimension, and the distance metric value is used to determine the weight value of the interaction relationship between the heterogeneous nodes. Optionally, the distance metric method includes information entropy, mutual information, correlation coefficient, etc. The distance metric method is used to calculate the distance metric value between the heterogeneous nodes according to the attribute feature vectors of the unified feature dimension in step S104, that is, to quantify the specific relationship between the heterogeneous nodes, so as to convert the attribute information of the heterogeneous nodes into the heterogeneous relationship between the feature vectors. The weight value of the interaction relationship between the heterogeneous nodes can be calculated according to the distance metric value, so as to achieve the purpose of determining the connection relationship between the heterogeneous nodes.

[0048] In step S106, the base tensor used to represent the interaction relationship between the heterogeneous nodes is determined according to the attribute feature vectors of the unified feature dimension and the weight value of the interaction relationship between the heterogeneous nodes. Optionally, the interaction relationship can be a single interaction relationship or a multiple interaction relationship, which is determined according to the analysis task requirement. Taking the single interaction relationship as an example, a group of heterogeneous node relationships with single relationship can be determined according to the attribute feature vectors obtained in step S104 and the weight value obtained in step S105, so as to obtain a group of base tensors representing the single relationship features.

[0049] In step S107, a high-order tensor network of the large-scale power grid is constructed by using a tensor multiplication operation rule according to the base tensor and the weight value of the interaction relationship between the heterogeneous nodes. Optionally, the high-order tensor network is constructed by using the tensor multiplication operation rule based on the base tensor determined in step S106 and the weight value calculated according to the distance metric value. It should be noted that different orders of base tensors can be used to construct different forms of high-order tensor networks according to different analysis task requirements. For example, a three-order base tensor can describe a ternary relationship, and a high-order tensor network can be constructed by using a tensor multiplication operation rule such as tensor module product according to the weight value of the interaction relationship between the nodes of the large-scale power grid corresponding to different orders of tensors.

[0050] In the embodiment of the present application, the tensor network is introduced to represent the large-scale power grid. The deep hash mapping model is used to establish multiple feature subspaces according to the extracted attribute features of the heterogeneous nodes, to represent the interaction relationship between the heterogeneous nodes. The distance metric method is used to calculate the weight value of the interaction relationship between the heterogeneous nodes according to the aligned multiple feature subspaces. Then, the base tensor representing the interaction relationship is determined, and the high-order tensor network representing the large-scale power grid is constructed by using the base tensor. Since the tensor network has the distributed storage and parallel processing capability for high-dimensional heterogeneous features and complex correlation relationships, the large-scale power grid is modeled by using the tensor network, which realizes the effective fusion of high-dimensional and complex features, improves the scalability of the large-scale power grid model, and reduces the computational complexity, thereby solving the problem of lacking an efficient model for representing the large-scale power grid in the related art.

[0051] A large-scale power grid high-order tensor network construction method is provided in the embodiment, which can be used in the terminal such as a central processing unit and a server, Figure 2 is a flowchart of another large-scale power grid high-order tensor network construction method according to the embodiment of the application, as shown in the figure, the flowchart comprises the following steps: Figure 2

[0052] In step S201, a plurality of category attribute sets corresponding to heterogeneous nodes in a large-scale power grid are obtained. For details, refer to step S101 of the embodiment shown in Figure 1 and will not be described here again.

[0053] In step S202, feature extraction is performed on the plurality of category attribute sets to obtain heterogeneous node attribute features. Optionally, step S202 comprises:

[0054] In step a1, features corresponding to the plurality of category attribute sets are extracted.

[0055] In step a2, the extracted features corresponding to any category attribute set are stored as a second-order tensor.

[0056] In step a3, a plurality of second-order tensors corresponding to the plurality of category attribute sets are determined.

[0057] In step a4, the plurality of second-order tensors are taken as the heterogeneous node attribute features.

[0058] Exemplarily, for the plurality of category attribute sets corresponding to the heterogeneous nodes of the large-scale power grid with different features, the differences in node attribute data are comprehensively considered, such as substation geographical location, voltage level, construction time limit, etc., features corresponding to different category attribute sets are extracted, and then classified and stored as a second-order tensor (matrix network). Each attribute set is stored as a separate second-order tensor, thereby realizing the classified representation of the heterogeneous node attributes.

[0059] In step S203, a plurality of feature subspaces are established according to the heterogeneous node attribute features by using a deep hash mapping model, and any feature subspace corresponds to an interaction relationship of a type of heterogeneous nodes. Optionally, the deep hash mapping model comprises a public hash coding module and a plurality of hash coding submodules, and the above step S203 comprises:

[0060] In step b1, the public hash coding module is used to map the heterogeneous node attribute features to a public attribute feature space.

[0061] In step b2, each hash coding submodule is used to establish a feature subspace according to the public attribute feature space.

[0062] ​Specifically, the different second-order tensors obtained in step 202 are taken as inputs of a deep hash mapping model, different heterogeneous attribute feature spaces corresponding to different categories of attributes are mapped (stored) to a common attribute feature space by a common hash coding module, and multiple (different) hash coding sub-modules are used to extract multiple feature subspaces of the heterogeneous node attributes, wherein the attribute features on each subspace represent a specific interaction relationship between the heterogeneous nodes, and thus the heterogeneous attribute space (multiple feature subspaces) in the entire large power grid corresponds to a set of high-order heterogeneous node interaction logics (interaction relationships).

[0063] In step S204, the multiple feature subspaces are aligned to a unified feature dimension by using a breadth learning strategy, and an attribute feature vector of the unified feature dimension is obtained. For details, please refer to Figure 1 In step S104 of the embodiment shown in the figure, details are not repeated here. It should be noted that by alignment, the relationship between features can be more accurately learned, thereby improving the generalization ability and performance.

[0064] In step S205, a distance measurement method is used to calculate the distance measurement value between the heterogeneous nodes according to the attribute feature vector of the unified feature dimension, and the distance measurement value is used to determine the weight value of the heterogeneous node interaction relationship. Optionally, the attribute feature vector of the unified feature dimension includes the attribute feature vector of the unified feature dimension corresponding to the heterogeneous node and its neighbor nodes, and step S205 includes:

[0065] In step c1, a distance measurement method is used to calculate the distance measurement value between the heterogeneous nodes according to the attribute feature vector of the unified feature dimension.

[0066] In step c2, the attribute feature vector of the unified feature dimension corresponding to the heterogeneous node and its neighbor nodes is merged based on the distance measurement value by using graph convolution to obtain a new attribute feature vector of the heterogeneous node.

[0067] In step c3, the weight value of the heterogeneous node interaction relationship is determined according to the new attribute feature vector of the heterogeneous node.

[0068] As an example, the distance measurement value (index) between nodes is calculated by using the information entropy, mutual information, correlation coefficient and other distance measurement methods, and the specific relationship between the heterogeneous nodes is quantified, so as to convert the attribute information of the heterogeneous nodes into the heterogeneous relationship of the vector edge, that is, the attribute information of the heterogeneous nodes is converted into the heterogeneous relationship of the attribute feature vector. The aligned attribute features of the heterogeneous nodes and their neighbor nodes are combined by using the graph convolution operation, and the weight value (relationship weight) of the interaction relationship between the heterogeneous nodes is determined based on the new attribute features of the heterogeneous nodes. Specifically, the distance measurement value between the heterogeneous nodes is calculated by using the aligned attribute feature vectors of the heterogeneous nodes in the large-scale power grid topology, that is, the attribute feature vectors with unified feature dimensions, as the weight value of the specific relationship between the heterogeneous node pairs, and then converted into the connection relationship between the heterogeneous nodes, so as to realize the unification of the heterogeneous nodes and the heterogeneous relationship.

[0069] In step S206, the base tensor for representing the interaction relationship between the heterogeneous nodes is determined according to the attribute feature vectors with unified feature dimensions and the weight value of the interaction relationship between the heterogeneous nodes. For details, please refer to Figure 1 The step S106 of the embodiment shown in the figure will not be repeated here.

[0070] In step S207, the high-order tensor network of the large-scale power grid is constructed by using the tensor multiplication rule according to the base tensor and the weight value of the interaction relationship between the heterogeneous nodes. For details, please refer to Figure 1 The step S107 of the embodiment shown in the figure will not be repeated here.

[0071] In an optional implementation, after step S207, it further includes:

[0072] In step S208, the heterogeneous power grid data is obtained. Optionally, the heterogeneous power grid data includes not only the heterogeneous node data such as the multi-category attribute set corresponding to the heterogeneous node pairs in the large-scale power grid, but also the heterogeneous power grid operation data, which can be collected by using the wireless sensor node (power grid information collection node).

[0073] Step S209, compressing the large-scale power grid high-order tensor network based on the high-order singular value analysis framework, bit plane, run-length encoding and arithmetic encoding. Specifically, although the high-order (heterogeneous) tensor network representing the large-scale power grid has been constructed, in some specific problems, instead of decomposing the tensor network first, the tensor network is used to describe the state of the entire system from the beginning, and the internal dimensions of the large-scale power grid high-order tensor network will rapidly increase with the action of the outside world. Therefore, it is necessary to compress the large-scale power grid high-order tensor network, so as to ensure that the construction of the tensor network can continue to run. The embodiment is based on the high-order singular value analysis framework (Singular Value Decomposition, HOSVD) and combined with bit plane, run-length and arithmetic encoding methods, to realize the compression and optimization of multi-time and space, multi-scale high-order tensor network. Specifically, let T be the input tensor, The result of compression and decompression. Each network data sampling pipeline accepts a main compression parameter, i.e. the error target , which can be expressed by RMSE (Root mean squared error) or MAE (Mean Absolute Error). Then, the specified error target is converted to SSE (Sum of Squares due to Error) by the following equivalent method, and the specific formula is as follows:

[0074] (1)

[0075] , where C is the total number of bit plane grid points I1…IN, and PSNR is the preset index. First, run the complete non-truncated HOSVD on the input data set T to obtain N orthogonal square factor matrices and the N-dimensional core of the matrix. The N-dimensional core of HOSVD is tiled into a one-dimensional vector of size C, then it is scaled and converted to a 64-bit integer, and C is used to sort it, i.e. traverse the dimensions in the core from right to left. In theory, the integer sequence is processed into a Cx64 binary matrix M. Second, the number of columns (bit planes) on the left of the matrix is compressed without loss, i.e. to the minimum value that makes the overall L2 error (mean square error) fall within the given target. This compression is achieved by run-length encoding (RLE) and arithmetic encoding (Arithmetic coding, AC). Finally, the orthogonal square factor matrices are compressed using the preset cost-benefit budget standard to obtain the compressed large-scale power grid high-order tensor network.

[0076] Step S210 involves constructing a spatiotemporal nonlinear representation model of the heterogeneous power grid data. Optionally, due to the different geographical locations of nodes in a large-scale power grid, and the potential patterns in the time series of heterogeneous power grid data, this embodiment models the dynamic evolution mechanism to achieve dynamic evolution deduction of the power grid topology. Specifically, step S210 includes:

[0077] Step d1: Obtain multi-factor spatiotemporal neighborhood features. Optionally, multi-factor spatiotemporal neighborhood features refer to the temporal and spatial features corresponding to multiple factors, such as information on the spatiotemporal changes of a certain attribute.

[0078] Step d2 involves using a regularization method to jointly model the spatiotemporal neighborhood information of heterogeneous power grid data based on multi-factor spatiotemporal neighborhood characteristics, thereby obtaining a power grid topology map in the spatiotemporal dimension. Optionally, step d2 includes: constructing a corresponding undirected weighted graph based on the heterogeneous power grid data; generating single spatiotemporal neighborhood information based on the undirected weighted graph using the Laplace operator and time difference; and using a multi-index joint constraint regularization method to jointly model the spatiotemporal neighborhood information of heterogeneous power grid data based on the single spatiotemporal neighborhood information, thereby obtaining a power grid topology map in the spatiotemporal dimension.

[0079] To address situations where there are strong or weak differences in single temporal or spatial characteristics, a modeling method based on the spatiotemporal mechanism of a single neighborhood is designed. Specifically, due to the different geographical locations of heterogeneous nodes in a large-scale power grid, data from real-world application scenarios generally exhibit spatial correlation. Furthermore, time series data from heterogeneous power grids can be generated from a finite number of patterns, and data adjacent to them at a given time slot show strong correlations. Therefore, to address spatial correlation, a modeling method is constructed such as... Figure 3 The undirected weighted graph G( shown) V , ε , W ),in, V This refers to the information collection nodes in the power grid, i.e. Figure 3 Nodes A, B…G in the dataset represent a set of wireless sensor nodes that perform data sampling. ε The edges represent the similarity relationships between wireless sensors. W This represents the weight changes between related sensors, i.e. Figure 3 The values ​​on the middle edge. Based on the graph structure representation, the Laplacian operator is defined as follows:

[0080] (2)

[0081] in, L For the Laplace operator, W This is the weight matrix. d N Weight matrix W No. N The harmony of actions, is a diagonal matrix constructor. For time correlation, since adjacent data shows strong correlation in time, the time correlation of data is represented by introducing a time difference operator, which is specifically represented as follows:

[0082] (3)

[0083] wherein, T is the total time sequence number of the grid information collection node data. By comprehensively considering the spatio-temporal neighborhood characteristics of various factors, a multi-index joint constraint regularization method is adopted to realize the joint modeling representation of spatio-temporal neighborhood information. Specifically, since the spatio-temporal neighborhood characteristic spaces of different factors are different in dimension, in order to realize subsequent unified modeling, it is necessary to perform order transformation on the spatio-temporal neighborhood characteristic spaces of different factors to realize the alignment of spatio-temporal neighborhood. Assuming that the spatio-temporal neighborhood characteristic space of factor a is A , the size of which is M x N , the spatial dimension corresponding to the characteristic matrix P is M×r , the time dimension corresponding to the characteristic matrix Q is N×r ; assuming that the spatio-temporal neighborhood characteristic space of factor b is B , the size of which is U x I , the spatial dimension corresponding to the characteristic matrix X is U x s , the time dimension corresponding to the characteristic matrix Y is I x s . The order transformation of the above characteristic matrices is performed by using a kernel function according to the following formula:

[0084] (4)

[0085] wherein, represents the mapping by using a kernel function, , , and , the sizes of M x f, N x f, U x f and I x f , f represent the unified characteristic space dimension. The embodiment comprehensively considers the spatio-temporal neighborhood characteristic spaces of various factors, and establishes a multi-index composite regularization constraint by joint constraint, so as to realize the joint modeling of spatio-temporal neighborhood information, and realizes the purpose of improving the generalization ability of the model in different application backgrounds. Specifically, the joint modeling of , , and is performed , and the mathematical expression of is as follows:

[0086] (5)

[0087] wherein, β 1and β 2represent control parameters of regularization terms constructed by different indexes, represents a regularization function, represents a loss function. It should be noted that the regularization can adopt L 1regularization, L 2regularization, L 1and L 2mixed regularization, graph regularization and the like. The embodiment is based on adaptive adjustment of the control parameters of different regularization terms by particle swarm optimization, so as to realize adaptive fusion of multi-factor space-time neighborhood information. Specifically, the adjustment is performed according to the following formula:

[0088] (6)

[0089] wherein, is a two-dimensional vector composed of parameters to be adaptively adjusted, v represents an update step, and represent individual and population optimums respectively, w represents an inertia factor, c 1and c 2represent acceleration coefficients, r 1and r 2represent random numbers between 0 and 1. At the same time, in order to improve the effect of adaptation, the fitness function can be designed in a targeted manner in combination with compound regularization constraints.

[0090] Step d3, the power grid topology graph in the space-time dimension is analyzed by using the nonlinear Kalman filtering state transition estimation method, and the state transition mode of the power grid topology graph is determined. Specifically, based on the nonlinear Kalman filtering state transition estimation means, the state transition mode of the power grid topology graph in the space-time dimension generated in step d2 is analyzed, so as to realize the deduction of the dynamic evolution of the power grid topology graph. Exemplarily, by using the nonlinear Kalman filtering dynamic estimation method in automatic control theory, a state transition space is constructed to determine the state transition mode of the power grid topology graph:

[0091] (7)

[0092] wherein, f ( ) and h ( ) represent different nonlinear functions, x k and x k-1They represent the first k The moment and the k -1 time period corresponding to the power grid topology map data w k Indicates state transition noise, v k Indicates measurement noise. z k This represents known sampled data within the spatiotemporal neighborhood information.

[0093] Step d4: Construct a nonlinear representation model of the spatiotemporal mechanism corresponding to heterogeneous power grid data based on the state transition pattern of the power grid topology diagram. Optionally, a nonlinear representation model of the spatiotemporal mechanism corresponding to heterogeneous power grid data is constructed from the state transition pattern of the power grid topology diagram.

[0094] Step S211: Based on the spatiotemporal mechanism nonlinear representation model and the compressed large-scale power grid high-order tensor network, a large-scale power grid high-order tensor network containing dynamic evolution mechanisms is generated. Optionally, the spatiotemporal mechanism nonlinear representation model is combined with the compressed large-scale power grid high-order tensor network to obtain a large-scale power grid high-order tensor network containing dynamic evolution mechanisms, thereby achieving accurate modeling of the dynamic evolution mechanism of large-scale power grids and solving the problem of efficient and accurate panoramic modeling and representation of large-scale power grids.

[0095] This embodiment innovatively employs a high-order tensor network model to represent a large-scale power grid. It merges, aligns, and separates the attribute spaces of heterogeneous nodes in the large-scale power grid to obtain different-order basis tensors representing the relationships between heterogeneous nodes. Based on a task-oriented approach, it constructs a highly scalable high-order tensor network for the data power grid, achieving the goal of comprehensively modeling and representing the topological and heterogeneous information of the large-scale power grid. By jointly modeling the spatiotemporal domain information of the digital network and constructing a nonlinear representation model of the spatiotemporal mechanism, it achieves the goal of accurately modeling the dynamic evolution mechanism of the large-scale power grid, thereby solving the problem of efficient and accurate panoramic modeling and representation of the large-scale power grid.

[0096] In one alternative implementation, Figure 4 This is a schematic diagram of the overall framework of a method for constructing high-order tensor networks for large-scale power grids according to an embodiment of the present invention, as shown below. Figure 4 As shown, firstly, heterogeneous power grid data is acquired. Secondly, high-order tensor network modeling of the digital power grid (large-scale power grid) is performed based on the heterogeneous power grid data. Specifically, this includes: merging of the attribute space of heterogeneous nodes in the large-scale power grid, the interaction transformation of heterogeneous node attributes to heterogeneous nodes, and the decoupling of the attribute space of heterogeneous nodes. For details, please refer to [link to relevant documentation]. Figure 2Steps S202 to S207 of the embodiment shown will not be repeated here. Again, the compressed large-scale power grid high-order tensor is obtained by the digital power grid high-order tensor network compression method, specifically including: feature compression by high-order singular value analysis, bit plane coding and the like, please refer to Figure 2 Steps S208 to S209 of the embodiment shown will not be repeated here. Then, the digital power grid dynamic evolution mechanism based on the high-order tensor network model is established, specifically including: single neighborhood space-time mechanism based modeling, joint modeling of space-time neighborhood information and construction of space-time mechanism nonlinear representation model, please refer to Figure 2 Steps S210 to S211 of the embodiment shown will not be repeated here. Finally, the high-order tensor network representing the large-scale power grid is obtained.

[0097] In an alternative embodiment, Figure 5 is a large-scale high-order tensor network feature learning method flowchart according to an embodiment of the application, corresponding to Figure 4 in which the high-order tensor network modeling of the digital power grid is performed according to the heterogeneous power grid data. As shown in Figure 5 the large-scale high-order tensor network feature learning method includes the following steps:

[0098] Step S501, extracting features corresponding to different categories of attribute sets. Please refer to Figure 2 Step S202 of the embodiment shown will not be repeated here.

[0099] Step S502, using a deep hash mapping model to input the extracted node features. Please refer to Figure 2 Step S203 of the embodiment shown will not be repeated here.

[0100] Step S503, heterogeneous attribute feature space transformation. Please refer to Figure 2 Step S203 of the embodiment shown will not be repeated here.

[0101] Step S504, obtaining a multi-feature subspace of heterogeneous node attributes. Please refer to Figure 2 Step S203 of the embodiment shown will not be repeated here.

[0102] Step S505, feature merging. Please refer to Figure 2 Step S204 of the embodiment shown will not be repeated here.

[0103] Step S506, calculating node and relationship weight. Please refer to Figure 2 Step S205 of the embodiment shown will not be repeated here.

[0104] Step S507, obtaining a base tensor representing single relationship features. Please refer toFigure 2 Step S206 of the embodiment shown will not be repeated here.

[0105] Step S508, constructing a high-order tensor network. For details, please refer to Figure 2 Step S207 of the embodiment shown will not be repeated here.

[0106] In an alternative embodiment, Figure 6 is a heterogeneous space merging schematic diagram according to an embodiment of the application, as Figure 6 As shown, it includes three stages of feature extraction, feature merging and subspace alignment. First, a set of hidden features is extracted from a plurality of category attribute sets (attribute A…attribute F) of the digital power grid heterogeneous nodes, and then input into a deep hash mapping model. Different hash coding modules are used to transform the heterogeneous attribute feature spaces corresponding to different category attributes to a common attribute feature space. According to different hash coding modules, a plurality of feature subspace of the heterogeneous node attributes is extracted, wherein the attribute features on each subspace are fed back to the heterogeneous node relationship, representing a specific interaction. Therefore, the entire digital power grid heterogeneous attribute space corresponds to a set of high-order heterogeneous node interaction logic. Finally, a breadth learning strategy is used to align the multiple subspaces to a unified feature dimension.

[0107] The embodiment innovatively uses a high-order tensor network model to represent a large-scale power grid. First, the heterogeneous node attribute space of the large-scale power grid is merged, aligned and separated to obtain different order basis tensors representing the heterogeneous node relationship. Then, based on the task guidance, a high-order tensor network with high scalability is constructed to realize the comprehensive modeling and representation of the large-scale power grid topology structure information and heterogeneous information. Then, based on the high-order singular value analysis framework, combined with bit plane, run and arithmetic coding methods, a power grid high-order tensor network compression method is used to reduce the internal dimension of the high-order tensor network. Finally, by jointly modeling the space-time field information of the digital network and constructing a space-time mechanism nonlinear representation model, the dynamic evolution mechanism of the large-scale power grid is accurately modeled, thereby solving the problem of efficient and accurate panoramic modeling and representation of the large-scale power grid.

[0108] In the embodiment, a large-scale power grid high-order tensor network construction device is also provided. The device is used to implement the above embodiments and preferred embodiments, and will not be repeated here. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware or a combination of software and hardware is also possible and is contemplated.

[0109] The embodiment provides a large-scale power grid high-order tensor network construction device, as Figure 7As shown, the method comprises: an attribute set obtaining module 701, configured to obtain a plurality of category attribute sets corresponding to heterogeneous nodes in a large-scale power grid; a feature extraction module 702, configured to perform feature extraction on the plurality of category attribute sets to obtain heterogeneous node attribute features; a multi-feature subspace establishing module 703, configured to establish a plurality of feature subspaces according to the heterogeneous node attribute features by using a deep hash mapping model, wherein any feature subspace corresponds to a type of heterogeneous node interaction relationship; a feature alignment module 704, configured to align the plurality of feature subspaces to a unified feature dimension by using a breadth learning strategy to obtain attribute feature vectors in the unified feature dimension; a distance metric value calculation module 705, configured to calculate distance metric values between the heterogeneous nodes according to the attribute feature vectors in the unified feature dimension by using a distance metric method, wherein the distance metric values are used to determine weights of the heterogeneous node interaction relationships; a base tensor determination module 706, configured to determine a base tensor used to represent the heterogeneous node interaction relationships according to the attribute feature vectors in the unified feature dimension and the weights of the heterogeneous node interaction relationships; and a power grid high-order tensor network construction module 707, configured to construct a large-scale power grid high-order tensor network according to the base tensor and the weights of the heterogeneous node interaction relationships by using a tensor multiplication operation rule.

[0110] In an optional implementation, the apparatus further comprises: a power grid data obtaining module, configured to obtain heterogeneous power grid data; a compression module, configured to compress the large-scale power grid high-order tensor network based on a high-order singular value analysis framework, a bit plane, a run-length encoding, and an arithmetic encoding; a mechanism model construction module, configured to construct a spatiotemporal mechanism nonlinear representation model corresponding to the heterogeneous power grid data; and a dynamic network generation module, configured to generate a large-scale power grid high-order tensor network containing a dynamic evolution mechanism based on the spatiotemporal mechanism nonlinear representation model and the compressed large-scale power grid high-order tensor network.

[0111] In an optional implementation, the feature extraction module comprises: a feature extraction unit, configured to extract features corresponding to the plurality of category attribute sets; a feature storage unit, configured to store the extracted features corresponding to any category attribute set as a second-order tensor; a tensor determination unit, configured to determine a plurality of second-order tensors corresponding to the plurality of category attribute sets; and a node attribute feature generation unit, configured to use the plurality of second-order tensors as the heterogeneous node attribute features.

[0112] In an optional implementation, the deep hash mapping model comprises a common hash encoding module and a plurality of hash encoding submodules, and the multi-feature subspace establishing module comprises: a feature mapping unit, configured to map the heterogeneous node attribute features to a common attribute feature space by using the common hash encoding module; and a subspace establishing unit, configured to establish a feature subspace according to the common attribute feature space by using each hash encoding submodule.

[0113] In an optional embodiment, the attribute feature vector of the unified feature dimension includes the attribute feature vector of the unified feature dimension corresponding to the heterogeneous node and the neighbor node thereof, the distance metric value calculation module includes: a calculation unit configured to calculate the distance metric value between the heterogeneous nodes according to the attribute feature vector of the unified feature dimension by using a distance metric method; a merging unit configured to merge the attribute feature vector of the unified feature dimension corresponding to the heterogeneous node and the neighbor node thereof based on the distance metric value by using graph convolution to obtain a new attribute feature vector of the heterogeneous node; and a determination unit configured to determine the weight value of the interaction relationship of the heterogeneous node according to the new attribute feature vector of the heterogeneous node.

[0114] In an optional embodiment, the mechanism model construction module includes: a spatio-temporal feature acquisition unit configured to acquire multi-factor spatio-temporal neighborhood features; a joint modeling unit configured to perform joint modeling of the spatio-temporal neighborhood information of the heterogeneous power grid data according to the multi-factor spatio-temporal neighborhood features by using a regularization method to obtain a power grid topology graph in a spatio-temporal dimension; an analysis unit configured to analyze the power grid topology graph in the spatio-temporal dimension by using a nonlinear Kalman filter state transition estimation method to determine a state transition mode of the power grid topology graph; and a mechanism model construction unit configured to construct a spatio-temporal mechanism nonlinear representation model corresponding to the heterogeneous power grid data based on the state transition mode of the power grid topology graph.

[0115] In an optional embodiment, the joint modeling unit includes: a graph construction subunit configured to construct a corresponding undirected weighted graph according to the heterogeneous power grid data; a spatio-temporal neighborhood information generation subunit configured to generate single spatio-temporal neighborhood information according to the undirected weighted graph by using a Laplacian operator and a time difference; and a joint modeling subunit configured to perform joint modeling of the spatio-temporal neighborhood information of the heterogeneous power grid data according to the single spatio-temporal neighborhood information by using a multi-index joint constraint regularization method to obtain a power grid topology graph in a spatio-temporal dimension.

[0116] Further function descriptions of the above-mentioned various modules and units are the same as those of the above-mentioned corresponding embodiments, which will not be described here again.

[0117] The large-scale power grid high-order tensor network construction device in the embodiment is presented in the form of functional units, where the units refer to ASIC (Application Specific Integrated Circuit, special-purpose integrated circuit) circuits, processors and memories executing one or more software or fixed programs, and / or other devices that can provide the above-mentioned functions.

[0118] The embodiment of the present application also provides a computer device with the large-scale power grid high-order tensor network construction device shown in the above Figure 7 .

[0119] Please refer to Figure 8 , Figure 8This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 8 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 8 Take a processor 10 as an example.

[0120] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0121] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0122] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0123] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0124] The computer device also comprises a communication interface 30 for the computer device to communicate with other devices or communication networks.

[0125] The embodiments of the present application also provide a computer readable storage medium, the method according to the embodiments of the present application can be implemented in hardware, firmware, or be implemented as computer code recorded in a storage medium, or be implemented through network downloading and originally stored in a remote storage medium or a non-transitory machine readable storage medium and to be stored in a local storage medium, so that the method described herein can be processed by such software on a storage medium using a general computer, a special processor or programmable or special hardware. Wherein, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk or a solid state disk, etc. Further, the storage medium can also include a combination of the above-mentioned types of memories. It can be understood that the computer, the processor, the microprocessor controller or the programmable hardware includes a storage component that can store or receive software or computer code, when the software or computer code is accessed and executed by the computer, the processor or the hardware, the method shown in the above embodiments is implemented.

[0126] Although the embodiments of the present application are described in conjunction with the accompanying drawings, various modifications and changes can be made by those skilled in the art without departing from the spirit and scope of the present application, and such modifications and changes fall within the scope defined by the appended claims.

Claims

1. A method for constructing a high-order tensor network of a large-scale power grid, characterized in that, The large-scale power grid comprises heterogeneous nodes, and the method comprises: Obtaining a plurality of category attribute sets corresponding to the heterogeneous nodes in the large-scale power grid; Performing feature extraction on the plurality of category attribute sets to obtain heterogeneous node attribute features; Using a deep hash mapping model to establish a plurality of feature subspaces according to the heterogeneous node attribute features, wherein any feature subspace corresponds to an interaction relationship of a type of heterogeneous nodes; Using a breadth learning strategy to align the plurality of feature subspaces to a unified feature dimension to obtain attribute feature vectors of the unified feature dimension; Using a distance measurement method to calculate distance measurement values between the heterogeneous nodes according to the attribute feature vectors of the unified feature dimension, wherein the distance measurement values are used to determine weights of the interaction relationship of the heterogeneous nodes; Determining a base tensor used to represent the interaction relationship of the heterogeneous nodes according to the attribute feature vectors of the unified feature dimension and the weights of the interaction relationship of the heterogeneous nodes; Using a tensor multiplication operation rule to construct a high-order tensor network of the large-scale power grid according to the base tensor and the weights of the interaction relationship of the heterogeneous nodes.

2. The method of claim 1, wherein, The method further comprises: Obtaining heterogeneous power grid data; Compressing the high-order tensor network of the large-scale power grid based on a high-order singular value analysis framework, a bit plane, a run-length encoding, and an arithmetic encoding; Constructing a spatiotemporal mechanism nonlinear representation model corresponding to the heterogeneous power grid data; Generating a high-order tensor network of the large-scale power grid containing a dynamic evolution mechanism based on the spatiotemporal mechanism nonlinear representation model and the compressed high-order tensor network of the large-scale power grid. 3.The method of claim 1, wherein, The feature extraction on the plurality of category attribute sets to obtain the heterogeneous node attribute features comprises: Extracting features corresponding to the plurality of category attribute sets; Storing the extracted features corresponding to any category attribute set as a second-order tensor; Determining a plurality of second-order tensors corresponding to the plurality of category attribute sets; Taking the plurality of second-order tensors as the heterogeneous node attribute features.

4. The method of claim 1, wherein, The deep hash mapping model comprises a common hash encoding module and a plurality of hash encoding submodules, and the establishment of the plurality of feature subspaces according to the heterogeneous node attribute features by using the deep hash mapping model, wherein any feature subspace corresponds to an interaction relationship of a type of heterogeneous nodes, comprises: Mapping the heterogeneous node attribute features to a common attribute feature space by using the common hash encoding module; Establishing a feature subspace by using each hash encoding submodule according to the common attribute feature space.

5. The method of claim 1, wherein, The attribute feature vectors of the unified feature dimension comprise attribute feature vectors of the unified feature dimension corresponding to the heterogeneous nodes and neighbor nodes of the heterogeneous nodes, and the calculation of distance measurement values between the heterogeneous nodes according to the attribute feature vectors of the unified feature dimension by using the distance measurement method, wherein the distance measurement values are used to determine the weights of the interaction relationship of the heterogeneous nodes, comprises: Calculating distance measurement values between the heterogeneous nodes according to the attribute feature vectors of the unified feature dimension by using the distance measurement method; Merging the attribute feature vectors of the unified feature dimension corresponding to the heterogeneous nodes and the neighbor nodes of the heterogeneous nodes based on the distance measurement values by using graph convolution to obtain new heterogeneous node attribute feature vectors; Determining the weights of the interaction relationship of the heterogeneous nodes according to the new heterogeneous node attribute feature vectors.

6. The method of claim 2, wherein, The construction of the spatiotemporal mechanism nonlinear representation model corresponding to the heterogeneous power grid data comprises: Obtaining multi-factor spatio-temporal neighborhood features; Performing joint modeling of spatio-temporal neighborhood information of heterogeneous power grid data according to the multi-factor spatio-temporal neighborhood features by using a regularization method to obtain a power grid topology graph in a spatio-temporal dimension; Analyzing the power grid topology graph in the spatio-temporal dimension by using a nonlinear Kalman filter state transition estimation method to determine a state transition mode of the power grid topology graph; Constructing a spatio-temporal mechanism nonlinear representation model corresponding to the heterogeneous power grid data based on the state transition mode of the power grid topology graph.

7. The method of claim 6, wherein, The joint modeling of spatio-temporal neighborhood information of the heterogeneous power grid data according to the multi-factor spatio-temporal neighborhood features by using the regularization method to obtain the power grid topology graph in the spatio-temporal dimension comprises: Constructing a corresponding undirected weighted graph according to the heterogeneous power grid data; Generating single spatio-temporal neighborhood information according to the undirected weighted graph by using a Laplacian operator and a time difference; Performing joint modeling of spatio-temporal neighborhood information of the heterogeneous power grid data according to the single spatio-temporal neighborhood information by using a multi-index joint constraint regularization method to obtain the power grid topology graph in the spatio-temporal dimension.

8. A large-scale power grid high-order tensor network construction apparatus, characterized in that, The large-scale power grid includes heterogeneous nodes, and the device comprises: An attribute set acquisition module configured to acquire a plurality of category attribute sets corresponding to the heterogeneous nodes in the large-scale power grid; A feature extraction module configured to perform feature extraction on the plurality of category attribute sets to obtain attribute features of the heterogeneous nodes; A multiple feature subspace establishment module configured to establish multiple feature subspaces according to the attribute features of the heterogeneous nodes by using a deep hash mapping model, wherein any one of the multiple feature subspaces corresponds to an interaction relationship of a category of the heterogeneous nodes; A feature alignment module configured to align the multiple feature subspaces to a unified feature dimension by using a breadth learning strategy to obtain attribute feature vectors in the unified feature dimension; A distance metric value calculation module configured to calculate distance metric values between the heterogeneous nodes according to the attribute feature vectors in the unified feature dimension by using a distance metric method, wherein the distance metric values are used to determine weights of the interaction relationship of the heterogeneous nodes; A basis tensor determination module configured to determine a basis tensor used to represent the interaction relationship of the heterogeneous nodes according to the attribute feature vectors in the unified feature dimension and the weights of the interaction relationship of the heterogeneous nodes; A power grid high-order tensor network construction module configured to construct a large-scale power grid high-order tensor network according to the basis tensor and the weights of the interaction relationship of the heterogeneous nodes by using a tensor multiplication operation rule. 9.The apparatus of claim 8, wherein, The device further comprises: A power grid data acquisition module configured to acquire heterogeneous power grid data; A compression module configured to compress the large-scale power grid high-order tensor network based on a high-order singular value analysis framework, a bit plane, a run-length encoding, and an arithmetic encoding; A mechanism model construction module configured to construct a spatio-temporal mechanism nonlinear representation model corresponding to the heterogeneous power grid data; A dynamic network generation module configured to generate a large-scale power grid high-order tensor network containing a dynamic evolution mechanism based on the spatio-temporal mechanism nonlinear representation model and the compressed large-scale power grid high-order tensor network. 10.The apparatus of claim 8, wherein, The feature extraction module comprises: A feature extraction unit configured to extract features corresponding to the plurality of category attribute sets; A feature storage unit configured to store the extracted features corresponding to any one of the category attribute sets as a second-order tensor. The tensor determination unit is configured to determine a plurality of second-order tensors corresponding to a plurality of category attribute sets; The node attribute feature generation unit is configured to take the plurality of second-order tensors as heterogeneous node attribute features.

11. The apparatus for constructing a high-order tensor network of a large-scale power grid according to claim 8, wherein, The deep hash mapping model comprises a common hash coding module and a plurality of hash coding sub-modules, and the multiple feature subspace establishment module comprises: The feature mapping unit is configured to map the heterogeneous node attribute features to a common attribute feature space by using the common hash coding module; The subspace establishment unit is configured to establish a one-dimensional feature subspace from the common attribute feature space by using each hash coding sub-module.

12. The apparatus for constructing a high-order tensor network of a large-scale power grid according to claim 8, wherein, The attribute feature vector of the unified feature dimension comprises attribute feature vectors of the unified feature dimension corresponding to the heterogeneous nodes and the neighbor nodes thereof, and the distance metric value calculation module comprises: The calculation unit is configured to calculate the distance metric values between the heterogeneous nodes according to the attribute feature vectors of the unified feature dimension by using a distance metric method; The merging unit is configured to merge the attribute feature vectors of the unified feature dimension corresponding to the heterogeneous nodes and the neighbor nodes thereof based on the distance metric values by using graph convolution, to obtain new heterogeneous node attribute feature vectors; The determination unit is configured to determine the weights of the heterogeneous node interaction relationships according to the new heterogeneous node attribute feature vectors.

13. The apparatus for constructing a high-order tensor network of a large-scale power grid according to claim 9, wherein, The mechanism model construction module comprises: The spatio-temporal feature acquisition unit is configured to acquire multi-factor spatio-temporal neighborhood features; The joint modeling unit is configured to perform joint modeling of the spatio-temporal neighborhood information of the heterogeneous power grid data according to the multi-factor spatio-temporal neighborhood features by using a regularization method, to obtain a power grid topology graph in a spatio-temporal dimension; The analysis unit is configured to analyze the power grid topology graph in the spatio-temporal dimension by using a nonlinear Kalman filter state transition estimation method, to determine a power grid topology graph state transition mode; The mechanism model construction unit is configured to construct a spatio-temporal mechanism nonlinear representation model corresponding to the heterogeneous power grid data based on the power grid topology graph state transition mode.

14. The apparatus according to claim 13, wherein, The joint modeling unit comprises: The graph construction sub-unit is configured to construct a corresponding undirected weighted graph from the heterogeneous power grid data; The spatio-temporal neighborhood information generation sub-unit is configured to generate single spatio-temporal neighborhood information from the undirected weighted graph by using a Laplacian operator and a time difference; The joint modeling sub-unit is configured to perform joint modeling of the spatio-temporal neighborhood information of the heterogeneous power grid data according to the single spatio-temporal neighborhood information by using a multi-index joint constraint regularization method, to obtain a power grid topology graph in a spatio-temporal dimension.

15. A computer device, comprising: The computer readable storage medium stores computer instructions, and the computer instructions are used to make the computer execute the large-scale power grid high-order tensor network construction method in any one of claims 1 to 7. The computer readable storage medium stores computer instructions, and the computer instructions are used to make the computer execute the large-scale power grid high-order tensor network construction method in any one of claims 1 to 7.

16. A computer-readable storage medium, characterized in that, ​

Images