A method for determining relationships of power distribution equipment and related components

Abstract

The application discloses a power distribution equipment relationship determination method and related components, and relates to the field of power distribution networks, which comprises the following steps: obtaining the attributes of each power distribution equipment, the attributes including business attributes and interconnection attributes, the business attributes being related to the functions of the power distribution equipment, and the interconnection attributes being related to the communication modes of the power distribution equipment; performing feature extraction on each attribute of each power distribution equipment; determining the attention scores between the attributes, the attention scores being positively correlated with the relevance between the attributes; and visually displaying the attributes and the relevance between the attributes. The feature extraction on the attributes of each power distribution equipment can more intuitively determine the attributes of each equipment and improve the accuracy of the portrait of a single power distribution equipment. The determination of the attention scores and the visual display based on the attention scores can determine the attributes of a power distribution equipment and the relationship between the attributes of multiple power distribution equipments, without manual query and determination by technical personnel.

Description

Technical Field

[0001] This invention relates to the field of power distribution networks, and in particular to a method for determining the relationships of power distribution equipment and related components. Background Technology

[0002] The distribution network Internet of Things (DIoT) contains a vast amount of data from diverse and heterogeneous devices, including information most closely related to the stable and reliable operation of the distribution network. To construct device profiles for the DIoT, it is necessary to mine the attribute and operational status data of these devices, followed by efficient mining using machine learning or decision tree algorithms. Accurately extracting the relationships between the attributes of massive amounts of DIoT devices from this diverse and heterogeneous data—relationships between devices and between devices and business operations—and automatically constructing device profiles for management is crucial. Related technologies often combine traditional expert-based intelligent decision-making systems, fault location systems, and transmission network planning decisions. However, these knowledge bases largely rely on traditional knowledge management methods where experts extract, organize, and store data in databases in chart form. The knowledge structure they can store is relatively simple; the attributes of each distribution device are relatively fixed, making personalized adjustments impossible based on the actual distribution equipment. Furthermore, it requires specialized technicians to search the database, wasting manpower. Summary of the Invention

[0003] The purpose of this invention is to provide a method and related components for determining the relationship between power distribution equipment, which can more intuitively determine the attributes of each equipment, improve the accuracy of the profile of a single power distribution equipment, and determine the relationship between the attributes of multiple power distribution equipment by determining the attributes of one power distribution equipment, without the need for technicians to manually query and determine.

[0004] To solve the above-mentioned technical problems, the present invention provides a method for determining the relationship of power distribution equipment, comprising:

[0005] Obtain the attributes of each power distribution device, including service attributes and interconnection attributes. The service attributes are related to the function of the power distribution device, and the interconnection attributes are related to the communication method of the power distribution device.

[0006] Feature extraction is performed on each of the attributes of each of the aforementioned power distribution equipment;

[0007] Determine the attention scores among the attributes after feature extraction, wherein the attention scores represent a positive correlation between the attributes;

[0008] The attention scores are used to visualize each attribute and the relationships between them.

[0009] On the other hand, the attributes of each power distribution device are obtained, including:

[0010] The service attributes and interconnection attributes of each power distribution device are obtained. The service attributes include the type, rated power, voltage level, operating status and power grid level of the power distribution device. The power grid level includes the transmission layer, distribution layer and terminal layer. The interconnection attributes include the physical connection between two power distribution devices, communication protocol, communication interface and network topology of the power distribution devices.

[0011] On the other hand, feature extraction is performed on each of the attributes of each of the aforementioned power distribution equipment, including:

[0012] The attribute is input into the BERT model to obtain the fragments output by the BERT model, and each fragment corresponds to one attribute;

[0013] Each of the segments is marked to distinguish it from the other segments.

[0014] On the other hand, after obtaining the fragment output by the BERT model, the process also includes:

[0015] The fragment is input into multiple stacked encoders and decoders to obtain the output vector;

[0016] The vector is obtained after processing by the multi-head self-attention mechanism of the encoder and the decoder, the feedforward neural network, the residual connection, and the layer normalization. The multi-head self-attention mechanism is used to calculate the attention score for each segment. The attention score is related to the context information between words corresponding to the attribute. The feedforward neural network is used to further process the output vector of the multi-head self-attention mechanism. The residual connection is used to solve the gradient vanishing problem. The layer normalization is used to stabilize the distribution of intermediate features.

[0017] The output vector is used as the attribute feature of each attribute.

[0018] On the other hand, determining the attention scores among the various attributes includes:

[0019] Determine the attribute characteristics of N attributes

[0020] in, Let i be the attribute feature of the i-th attribute. Let F be the attribute feature of the Nth attribute, where 1 ≤ i ≤ N, and i and N are both integers. The feature dimension of each attribute is F, R. F For feature space;

[0021] Determine the attention scores among the various attributes

[0022] Among them, e ijLet be the attention score of the i-th attribute relative to the j-th attribute, 1≤j≤N, where j is an integer. And W is a learnable parameter.

[0023] On the other hand, after determining the attention scores among the various attributes, the following also includes:

[0024] According to the normalization formula The attention scores are normalized;

[0025] Where, α ij Let N be the normalized attention score of the i-th attribute relative to the j-th attribute. i For the neighborhood of i, softmax j (e ij ) is the normalization function for the attention score of the i-th attribute relative to the j-th attribute, exp(e ij Let be an exponential function with base e of the attention score of the i-th attribute relative to the j-th attribute, and LeakyReLU be the activation function;

[0026] The normalized attention score is used as the correlation between the two attributes, and the process proceeds to the step of visualizing each attribute and the correlation between the attributes.

[0027] On the other hand, after normalizing the attention score according to the normalization formula, the method further includes:

[0028] According to the update relation The attributes are updated so that the attention score is determined in the next round based on the updated attributes;

[0029] in, Let σ be the updated i-th attribute, and σ be the activation function. The attributes of the i-th node obtained after feature aggregation and activation function are...

[0030] To address the aforementioned technical problems, the present invention also provides a system for determining the relationship of power distribution equipment, comprising:

[0031] An attribute acquisition unit is used to acquire the attributes of each power distribution device. The attributes include service attributes and interconnection attributes. The service attributes are related to the function of the power distribution device, and the interconnection attributes are related to the communication method of the power distribution device.

[0032] The feature extraction unit is used to extract features from each of the attributes of each of the power distribution devices;

[0033] An attention score determination unit is used to determine the attention score among the various attributes after feature extraction, wherein the attention score indicates a positive correlation between the attributes;

[0034] The display unit is used to visualize the individual attributes and the relationships between the attributes based on the attention scores.

[0035] To address the aforementioned technical problems, the present invention also provides a device for determining the relationship of power distribution equipment, comprising:

[0036] Memory, used to store computer programs;

[0037] A processor is used to implement the steps of the above-described method for determining the relationship of power distribution equipment when executing the computer program.

[0038] To address the aforementioned technical problems, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for determining the relationship of the aforementioned power distribution equipment.

[0039] This invention discloses a method and related components for determining the relationships between power distribution equipment, relating to the field of power distribution networks. The method includes acquiring the attributes of each power distribution device, including service attributes and interconnection attributes. Service attributes are related to the function of the power distribution device, and interconnection attributes are related to the communication method of the power distribution device. Features are extracted from each attribute of each power distribution device. Attention scores are determined among the attributes, with the attention score indicating a positive correlation between the attributes. The attributes and their relationships are then visualized. Feature extraction of the attributes of each power distribution device allows for a more intuitive determination of the attributes, improving the accuracy of the profile of a single power distribution device. Determining attention scores and visualizing them allows for the determination of the attributes of a single power distribution device as well as the relationships between the attributes of multiple power distribution devices, eliminating the need for manual querying by technical personnel. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the prior art and embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 A flowchart of a method for determining the relationship of power distribution equipment provided by the present invention;

[0042] Figure 2A schematic diagram of the structure provided by the present invention for visualization;

[0043] Figure 3 A schematic diagram of the structure of a BERT model provided by the present invention;

[0044] Figure 4 This invention provides a schematic diagram of the structure of a system for determining the relationship of power distribution equipment.

[0045] Figure 5 This is a schematic diagram of the structure of a device for determining the relationship of power distribution equipment provided by the present invention. Detailed Implementation

[0046] The core of this invention is to provide a method for determining the relationship between power distribution equipment and related components, which can more intuitively determine the attributes of each equipment, improve the accuracy of the profile of a single power distribution equipment, and determine the relationship between the attributes of multiple power distribution equipment by determining the attributes of one power distribution equipment, without the need for technicians to manually query and determine.

[0047] 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 with reference to the accompanying drawings. 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.

[0048] Figure 1 This invention provides a flowchart of a method for determining the relationship of power distribution equipment, which includes:

[0049] S11: Obtain the attributes of each power distribution device. The attributes include service attributes and interconnection attributes. Service attributes are related to the function of the power distribution device, and interconnection attributes are related to the communication method of the power distribution device.

[0050] In a specific distribution IoT scenario, distribution IoT technology is applied to the monitoring and management of microgrids to achieve efficient utilization of renewable energy and reliable power supply. In this scenario, the microgrid includes multiple power distribution devices, such as solar photovoltaic panels, wind turbines, energy storage systems, and microgrid controllers. These devices work together to achieve autonomous power supply and energy interconnection and sharing within the microgrid.

[0051] Specifically, taking solar photovoltaic panels, wind turbines, energy storage systems, and microgrid controllers as examples, the business attributes of solar photovoltaic panels include their power, area, and voltage, while their interconnection attributes include connection methods (grid-connected or off-grid), communication protocols (e.g., Modbus or CAN), and communication interfaces (e.g., RS485 or Ethernet). The business attributes of wind turbines include their rated power, speed, and blade length, while their interconnection attributes include connection methods, communication protocols, and communication interfaces. The business attributes of energy storage systems include their capacity, charging and discharging efficiency, and battery type, while their interconnection attributes include connection methods, communication protocols, and communication interfaces. The business attributes of microgrid controllers include their processing capabilities, control algorithms, and operating methods, while their interconnection attributes include connection methods, communication protocols, and communication interfaces.

[0052] S12: Extract features from each attribute of each power distribution device;

[0053] Considering that the storage methods of attributes obtained from various power distribution devices may be inconsistent, the obtained attributes cannot be directly used to determine the attributes of the power distribution devices. Therefore, feature extraction is required for each attribute. Specifically, each attribute of each power distribution device is extracted. For example, the business attributes of the energy storage system, including the energy storage system capacity, charging and discharging efficiency, and battery type, are extracted and differentiated separately.

[0054] S13: Determine the attention scores among the attributes after feature extraction. The attention scores indicate a positive correlation between the attributes.

[0055] S14: Visualize the individual attributes and the relationships between them based on the attention scores.

[0056] A joint device attribute relationship extraction model is constructed using GAT (Graph Attention Network). For a specific power distribution device, the extracted attributes of that device are used as nodes, and the relationships between those attributes are used as edges to construct a knowledge graph. This process is repeated for each existing power distribution device. Finally, the extracted device attributes and the relationships between them are integrated into a single overall knowledge graph. Each entity in the knowledge graph represents a business attribute, and edges represent the relationships and interconnections between these attributes. The relationships represented by the edges are reflected based on attention scores.

[0057] Figure 2 A schematic diagram of the structure provided by the present invention for visualization;

[0058] from Figure 2 As can be seen, device A has multiple attribute nodes: Z A1 ZA2 Z A3 and Z A4 Relationships between nodes: R1, R2, and R3. For the profile of device B, we can see that it has a similar topology and node representation to the profile of device A. Attribute node: Z B1 Z B2 Z B3 Z B4 and Z B5 Relationships between nodes: R1, R2, and R3.

[0059] For device B, attribute Z needs to be added to the knowledge graph. B5 This approach utilizes aligned knowledge graphs to provide more meta-paths (such as R1+R2 or R3+R2) to offer richer cross-domain prior knowledge for new devices. Based on multi-source knowledge graphs, it constructs a mechanism for direct logical reasoning and indirect logical reasoning through meta-path definitions to obtain the context of the new device. Finally, it employs an autoencoder combined with existing knowledge graph structures to learn context combinations and corresponding weights, thereby completing the profile representation Z of the new device's attributes. B5 Therefore, for new devices, their direct or indirect context representations combined with weight parameters can be used to derive a profile representation of the new device, ultimately achieving incremental construction of device profiles.

[0060] This invention discloses a method for determining the relationships between power distribution equipment, relating to the field of power distribution networks. The method includes acquiring the attributes of each power distribution device, including service attributes and interconnection attributes. Service attributes are related to the function of the power distribution device, and interconnection attributes are related to the communication method of the power distribution device. Features are extracted from each attribute of each power distribution device. Attention scores are determined among the attributes, with the attention score indicating a positive correlation between the attributes. The attributes and their relationships are then visualized. Feature extraction of the attributes of each power distribution device allows for a more intuitive determination of the attributes, improving the accuracy of the profile of a single power distribution device. Determining attention scores and visualizing them allows for the identification of individual attributes of a single power distribution device, as well as the determination of relationships between the attributes of multiple power distribution devices, eliminating the need for manual querying by technical personnel.

[0061] Based on the above embodiments:

[0062] In some embodiments, the attributes of each power distribution device are obtained, including:

[0063] Obtain the business attributes and interconnection attributes of each power distribution device. The business attributes include the type, rated power, voltage level, operating status and power grid level of the power distribution device. The power grid level includes the transmission layer, distribution layer and terminal layer. The interconnection attributes include the physical connection between two power distribution devices, communication protocol, communication interface and the network topology of the power distribution devices.

[0064] Solar photovoltaic panels, wind turbines, energy storage systems, and microgrid controllers are classified as types. All specific devices involved within the scope are defined as entities. These devices also have several attributes. For example, the entity "Wind Turbine A" is of type "Wind Turbine" and has attributes such as "Rated Power: 1.5MW", "Speed ​​Rotation: 750 rpm", and "Blade Length: 20 meters". Another example is the entity "Solar Photovoltaic Panel B", which is of type "Solar Photovoltaic Panel" and has attributes such as "Rated Power: 20000W", "Area: 2 square meters", "Rated Voltage: 12V-48V", and "Communication Interface: RS485".

[0065] In some embodiments, feature extraction is performed on various attributes of each power distribution device, including:

[0066] The attributes are input into the BERT model to obtain fragments output by the BERT model, with each fragment corresponding to one attribute;

[0067] Each segment contains a marker to distinguish it from other segments.

[0068] Figure 3 A schematic diagram of the structure of a BERT model provided by the present invention;

[0069] BERT is used as a word vector representation and feature vector extraction model for smart microgrid equipment data. For the business attributes and interconnection attributes of power distribution equipment, the BERT model can perform preprocessing, word vector representation, and attribute extraction through the following steps:

[0070] First, the description of the power distribution equipment is preprocessed. The BERT model uses a word segmentation algorithm to divide the input sequence into words or subwords, and generates an initial word vector representation for each word or subword. These word vector representations contain semantic and contextual information of the words. For the power distribution equipment in the smart microgrid scenario described above, for business attributes, attributes such as equipment type, rated power, and rated voltage can be split into multiple segments. For example, equipment type can be one segment, rated power another, and voltage a third. Each segment can be marked with a special tag "[CLS]" at the beginning, and together with other segments, it constitutes the input of BERT. For interconnection attributes, attributes such as connection method, communication protocol, and communication interface between devices can be split into multiple segments. For example, connection method can be one segment, communication protocol another, and communication interface a third. Similarly, each segment can be marked with a special tag at the beginning, and together with other segments, it serves as the input of BERT. To enable the BERT model to learn word position information and distinguish the relationships between different sentences, position embedding and segment embedding are added to the word vectors. Position embedding is used to represent the position of each word in the input sequence, while segment embedding is used to distinguish different sentences or text segments.

[0071] This invention proposes to extract features from attribute relationship matrices based on autoencoders. An autoencoder is an unsupervised learning model consisting of an encoder and a decoder, which extracts features from the input data by reconstructing the input data. The first and second layers of the autoencoder are the encoding process, obtaining a low-dimensional representation of the initial attribute relationship vector. The second and third layers are the decoding process, outputting a reconstructed vector with the same dimension as the input attribute relationship matrix. The network is trained by minimizing the reconstruction loss. Since autoencoders can learn features from large-scale datasets and have the advantage of minimizing information loss, they are suitable for solving the problem of extracting relationships from complex device profiles. In practical applications, attribute information descriptions of new devices exist in the power distribution IoT. Therefore, an attribute relationship matrix can be constructed by comparing the attribute information descriptions of new devices with those of existing devices, and then fed into the autoencoder for feature extraction. Furthermore, based on the corresponding weights of the existing device profile context, a weighted representation of the new device profile can be derived using the existing device profile weighted representation.

[0072] In some embodiments, after obtaining the fragment output by the BERT model, the method further includes:

[0073] The fragment is input into multiple stacked encoders and decoders to obtain the output vector;

[0074] The vector is obtained after processing by the multi-head self-attention mechanism of the encoder and decoder, the feedforward neural network, the residual connection and the layer normalization. The multi-head self-attention mechanism is used to calculate the attention score for each segment. The attention score is related to the context information between the words corresponding to the attribute. The feedforward neural network is used to further process the output vector of the multi-head self-attention mechanism. The residual connection is used to solve the gradient vanishing and gradient exploding problems. The layer normalization is used to stabilize the distribution of intermediate features.

[0075] The output vector is used as the attribute feature of each attribute.

[0076] The embedded representation of the input is fed into BERT's feature extraction module, which consists of multiple stacked encoders and decoders. Each encoder comprises multiple sub-layers, including a multi-head self-attention mechanism, a feedforward neural network, residual connections, and normalization layers. The multi-head self-attention mechanism calculates an attention score for each word and uses these scores as weights to combine word representations, capturing contextual information between words. The feedforward neural network further processes the output of the attention mechanism through two linear transformations and activation functions. Residual connections address the vanishing and exploding gradient problems and help the model converge faster. Layer normalization makes the distribution of intermediate features more stable, improving the stability and generalization ability of network training. Similarly, the decoder also consists of multiple sub-layers, including multi-head attention, a feedforward neural network, residual connections, and normalization layers. In this way, the encoder and decoder can encode and decode the context of the input sequence, capturing contextual information and generating context-related word vector representations for each word. The low-dimensional vector obtained after passing through multiple encoders and decoders is used as the result of attribute extraction. The final output of the BERT model is a word vector encoded from each word in the input sequence of power distribution equipment, containing semantic information and contextual relationships of various attributes. For example, if wind turbine A and wind turbine B have similar rated power, their BERT-encoded word vectors will have high similarity. Similarly, if solar photovoltaic panel A and solar photovoltaic panel B have similar communication interfaces, their BERT-encoded word vectors will have high similarity. Through BERT's contextual representation, more accurate attribute analysis and applications in the power distribution IoT can be performed.

[0077] In some embodiments, determining the attention scores among the attributes includes:

[0078] Determine the attribute characteristics of N attributes

[0079] in, Let i be the attribute feature of the i-th attribute. Let F be the attribute feature of the Nth attribute, where 1 ≤ i ≤ N, and i and N are both integers. The feature dimension of each attribute is F, R. F For feature space;

[0080] Determine the attention scores among the various attributes

[0081] Among them, e ij Let be the attention score of the i-th attribute relative to the j-th attribute, 1≤j≤N, where j is an integer. And W is a learnable parameter.

[0082] Graph Attention Networks (GAT) are used to extract device attribute relationships. In power distribution IoT devices, GAT allows for updating node embeddings to capture the complex relationships and importance between entity nodes. GAT employs a self-attention mechanism to accurately extract relationships between device business attributes and device interconnection attributes. First, the attention score of the current node and each neighboring node is calculated. Then, this attention score is multiplied by the feature of each node and accumulated. Finally, after a non-linear mapping, the result is used as the feature of the current node.

[0083] For each node in the power distribution IoT device, the word vectors output by BERT encoding are used as the initial node representation. These representations can be the node's business attributes, interconnection attributes, or other feature vectors. These node representations will be used as input to the GAT model. The attribute features of N attributes are then determined. The GAT model uses a self-attention mechanism to calculate the importance and relevance between nodes. For each node, the relationship weight between the node and its neighbors is determined by calculating the attention score between the node and its neighbors. To calculate the attention weight between a node and its neighbors, the GAT model introduces an attention mechanism. The relative importance of a node is determined by calculating the attention score between two nodes. The calculation of the attention score involves two steps: First, a linear transformation is performed between the node's representation and the representations of its neighbors to obtain the feature propagation of each node to its neighbors. Then, for each node and its neighbors, the attention score is calculated. Attention scores are obtained by performing dot product and scaling operations through feature propagation between nodes.

[0084] In some embodiments, after determining the attention scores among the attributes, the method further includes:

[0085] According to the normalization formula Normalize the attention score;

[0086] Where, α ij Let N be the normalized attention score of the i-th attribute relative to the j-th attribute. iFor the neighborhood of i, softmax j (e ij Let exp(e) be the normalization function for the attention score of the i-th attribute relative to the j-th attribute. ij Let be an exponential function with base e for the attention score of the i-th attribute relative to the j-th attribute, and LeakyReLU be the activation function;

[0087] The normalized attention score is used as the correlation between two attributes, and then the process proceeds to visualize the individual attributes and the correlation between them.

[0088] After calculating the attention score, it is softmax normalized to limit the attention weights to between 0 and 1, and to ensure that the sum of the attention weights of all neighbors is 1. This allows the attention weight of each node to represent its importance to its neighboring nodes.

[0089] In some embodiments, after normalizing the attention score according to the normalization formula, the method further includes:

[0090] According to the update relation The attributes are updated so that the attention score can be determined in the next round based on the updated attributes;

[0091] in, Let σ be the updated i-th attribute, and σ be the activation function. The attributes of the i-th node obtained after feature aggregation and activation function are...

[0092] The attention weights of each node are weighted and summed with the feature propagation of its neighbors to obtain an aggregated representation of the node. This aggregated representation includes the node's own feature representation and the feature representations of its neighbors. The aggregated representation is then input into an activation function to generate a new node representation. This new node representation is used in the next round of node embedding updates. To better capture the relationships between nodes, the GAT model typically undergoes multiple iterations. In each iteration, the above steps are repeated to update the node representation. Each iteration considers the node's neighbors and incorporates their information into the node's representation. After multiple layers of GAT blocks, the graph node representation vector integrates its own business attribute information as well as the interconnection attribute information between related devices.

[0093] By utilizing the node embedding update process of the graph attention network described above, the relationships and importance between nodes can be leveraged to update node representations, thereby better capturing the complex relationships and semantic information between nodes. This method can help understand the interactions and influences between nodes in power distribution IoT devices and support more accurate predictions and decisions. GAT incorporates the importance of different node attributes into the feature extraction process, enabling more accurate and effective extraction of relationships between attributes; its feature extraction process is not based on a fixed graph structure and can dynamically learn node features. Thus, joint device attribute relationship extraction is achieved, yielding entity relationships in triples (head entity, entity relationship, tail entity).

[0094] Figure 4 This is a schematic diagram of a relationship determination system for power distribution equipment provided by the present invention. The relationship determination system for power distribution equipment includes:

[0095] The attribute acquisition unit 41 is used to acquire the attributes of each power distribution device. The attributes include service attributes and interconnection attributes. The service attributes are related to the function of the power distribution device, and the interconnection attributes are related to the communication method of the power distribution device.

[0096] Feature extraction unit 42 is used to extract features from various attributes of each power distribution device;

[0097] Attention score determination unit 43 is used to determine the attention score between each attribute after feature extraction. The attention score indicates that the correlation between the attributes is positive.

[0098] Display unit 44 is used to visualize the various attributes and the relationships between them based on the attention score.

[0099] Based on the above embodiments:

[0100] The attribute acquisition unit 41 is specifically used to acquire the business attributes and interconnection attributes of each power distribution device. The business attributes include the type, rated power, voltage level, operating status and power grid level of the power distribution device. The power grid level includes the transmission layer, distribution layer and terminal layer. The interconnection attributes include the physical connection between two power distribution devices, communication protocol, communication interface and network topology structure formed by the power distribution devices.

[0101] The feature extraction unit 42 is specifically used to input attributes into the BERT model and obtain fragments output by the BERT model, with each fragment corresponding to an attribute;

[0102] Each segment contains a marker to distinguish it from other segments.

[0103] Also includes:

[0104] The encoding / decoding unit is used to input segments into multiple stacked encoders and decoders to obtain the output vector;

[0105] The vector is obtained after processing by the multi-head self-attention mechanism of the encoder and decoder, the feedforward neural network, the residual connection and the layer normalization. The multi-head self-attention mechanism is used to calculate the attention score for each segment. The attention score is related to the context information between the words corresponding to the attribute. The feedforward neural network is used to further process the output vector of the multi-head self-attention mechanism. The residual connection is used to solve the gradient vanishing and gradient exploding problems. The layer normalization is used to stabilize the distribution of intermediate features.

[0106] The output vector is used as the attribute feature of each attribute.

[0107] Attribute feature determination unit, used to determine the attribute features of N attributes.

[0108] in, Let i be the attribute feature of the i-th attribute. Let F be the attribute feature of the Nth attribute, where 1 ≤ i ≤ N, and i and N are both integers. The feature dimension of each attribute is F, R. F For feature space;

[0109] Attention score determination unit 43 is specifically used to determine the attention scores between various attributes.

[0110]

[0111] Among them, e ij Let be the attention score of the i-th attribute relative to the j-th attribute, 1≤j≤N, where j is an integer. And W is a learnable parameter.

[0112] Normalization unit, used according to the normalization formula Normalize the attention score;

[0113] Where, α ij Let N be the normalized attention score of the i-th attribute relative to the j-th attribute. i For the neighborhood of i, softmax j (e ij Let exp(e) be the normalization function for the attention score of the i-th attribute relative to the j-th attribute. ij Let be an exponential function with base e for the attention score of the i-th attribute relative to the j-th attribute, and LeakyReLU be the activation function;

[0114] The normalized attention score is used as the correlation between two attributes, and then the process proceeds to visualize the individual attributes and the correlation between them.

[0115] Update unit, used to update the relation The attributes are updated so that the attention score can be determined in the next round based on the updated attributes;

[0116] in, Let σ be the updated i-th attribute, and σ be the activation function. The attributes of the i-th node obtained after feature aggregation and activation function are...

[0117] Figure 5 This is a schematic diagram of a relationship determination device for power distribution equipment provided by the present invention. The relationship determination device for power distribution equipment includes:

[0118] Memory 51 is used to store computer programs;

[0119] The processor 52 is used to implement the steps of the above-described method for determining the relationship of power distribution equipment when executing a computer program.

[0120] Please refer to the above embodiments for a description of the device for determining the relationship of power distribution equipment provided in this application, and it will not be repeated here.

[0121] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for determining the relationship of power distribution equipment.

[0122] The description of the computer-readable storage medium provided in this application is given in the above embodiments and will not be repeated here.

[0123] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0124] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0125] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for determining the relationship of power distribution equipment, characterized in that, include: Obtain the attributes of each power distribution device, including service attributes and interconnection attributes. The service attributes are related to the function of the power distribution device, and the interconnection attributes are related to the communication method of the power distribution device. Feature extraction is performed on each of the attributes of each of the aforementioned power distribution equipment; Determine the attention scores among the attributes after feature extraction, wherein the attention scores represent a positive correlation between the attributes; The attention scores are used to visualize each attribute and the relationships between them.

2. The method for determining the relationship of power distribution equipment as described in claim 1, characterized in that, Obtain the attributes of each power distribution device, including: The service attributes and interconnection attributes of each power distribution device are obtained. The service attributes include the type, rated power, voltage level, operating status and power grid level of the power distribution device. The power grid level includes the transmission layer, distribution layer and terminal layer. The interconnection attributes include the physical connection between two power distribution devices, communication protocol, communication interface and network topology of the power distribution devices.

3. The method for determining the relationship of power distribution equipment as described in claim 1, characterized in that, Feature extraction is performed on each of the attributes of each of the aforementioned power distribution equipment, including: The attribute is input into the BERT model to obtain the fragments output by the BERT model, and each fragment corresponds to one attribute; Each of the segments is marked to distinguish it from the other segments.

4. The method for determining the relationship of power distribution equipment as described in claim 3, characterized in that, After obtaining the fragment output by the BERT model, the following is also included: The fragment is input into multiple stacked encoders and decoders to obtain the output vector; The vector is obtained after processing by the multi-head self-attention mechanism of the encoder and the decoder, the feedforward neural network, the residual connection, and the layer normalization. The multi-head self-attention mechanism is used to calculate the attention score for each segment. The attention score is related to the context information between words corresponding to the attribute. The feedforward neural network is used to further process the output vector of the multi-head self-attention mechanism. The residual connection is used to solve the gradient vanishing problem. The layer normalization is used to stabilize the distribution of intermediate features. The output vector is used as the attribute feature of each attribute.

5. The method for determining the relationship of power distribution equipment as described in any one of claims 1 to 4, characterized in that, Determining the attention scores among the various attributes includes: Determine the attribute characteristics of N attributes in, Let i be the attribute feature of the i-th attribute. Let F be the attribute feature of the Nth attribute, where 1 ≤ i ≤ N, and i and N are both integers. The feature dimension of each attribute is F, R. F For feature space; Determine the attention scores among the various attributes Among them, e ij Let be the attention score of the i-th attribute relative to the j-th attribute, 1≤j≤N, where j is an integer. And W is a learnable parameter.

6. The method for determining the relationship of power distribution equipment as described in claim 5, characterized in that, After determining the attention scores among the various attributes, the following is also included: According to the normalization formula The attention scores are normalized; Where, α ij Let N be the normalized attention score of the i-th attribute relative to the j-th attribute. i For the neighborhood of i, softmax j (e ij ) is the normalization function for the attention score of the i-th attribute relative to the j-th attribute, exp(e ij Let be an exponential function with base e of the attention score of the i-th attribute relative to the j-th attribute, and LeakyReLU be the activation function; The normalized attention score is used as the correlation between the two attributes, and the process proceeds to the step of visualizing each attribute and the correlation between the attributes.

7. The method for determining the relationship of power distribution equipment as described in claim 6, characterized in that, After normalizing the attention score according to the normalization formula, the method further includes: According to the update relation The attributes are updated so that the attention score is determined in the next round based on the updated attributes; in, Let σ be the updated i-th attribute, and σ be the activation function. The attributes of the i-th node obtained after feature aggregation and activation function are...

8. A system for determining the relationship of power distribution equipment, characterized in that, include: An attribute acquisition unit is used to acquire the attributes of each power distribution device. The attributes include service attributes and interconnection attributes. The service attributes are related to the function of the power distribution device, and the interconnection attributes are related to the communication method of the power distribution device. The feature extraction unit is used to extract features from each of the attributes of each of the power distribution devices; An attention score determination unit is used to determine the attention score among the various attributes after feature extraction, wherein the attention score indicates a positive correlation between the attributes; The display unit is used to visualize the individual attributes and the relationships between the attributes based on the attention scores.

9. A device for determining the relationship of power distribution equipment, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the method for determining the relationship of power distribution equipment as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method for determining the relationship of power distribution equipment as described in any one of claims 1 to 7.

Images