A power equipment management method and device based on a graph convolution network

Abstract

The application relates to a power equipment management method and device based on a graph convolution network, and belongs to the technical field of power equipment management.The method comprises the following steps: obtaining operation data of power equipment, obtaining a causal feature vector based on the operation data, calculating a causal feature vector similarity to determine the causal strength between each device and constructing a causal strength matrix.The causal feature vector and the matrix are input into a graph convolution network, an output node embedding matrix is output, a clustering center vector is obtained to determine the clustering labels of each device, and a device clustering set is formed.A coordination index of each set is obtained, and a resource allocation amount is obtained accordingly, and the resource allocation amount of each power equipment is calculated by combining a device resource allocation weight.A dispatching plan is generated according to the resource allocation amount of each power equipment, and the application ensures that key areas and important equipment can obtain reasonable management resources.

Description

TECHNICAL FIELD

[0001] The application relates to a power equipment management method and device based on a graph convolution network, and belongs to the technical field of power equipment management. BACKGROUND

[0002] With the deepening of the construction of new power systems, the scale of power grids continues to expand, the number of power equipment increases exponentially, the types of equipment become increasingly complex, and the operating conditions become increasingly diverse. Under this background, power equipment management is facing unprecedented challenges: on the one hand, the equipment operation data is growing explosively, and the traditional management mode based on manual experience and simple threshold judgment has been difficult to cope with the characteristics of high-dimensional, nonlinear, and strong coupling of complex systems; on the other hand, the complex causal correlation network is formed among the equipment through electrical connection, load transmission, and protection linkage, and local faults may quickly spread through coupling relationships to cause a chain reaction, posing a serious threat to the safe and stable operation of power grids.

[0003] The existing technology mainly relies on passive management strategies such as regular maintenance and post-disposal, lacks the ability to deeply mine and dynamically perceive the implicit correlation between equipment, and causes problems such as uneven resource allocation, unreasonable maintenance plan, and untimely fault warning, resulting in safety hazards. SUMMARY

[0004] In order to solve the problems existing in the prior art, the application provides a power equipment management method and device based on a graph convolution network.

[0005] The technical scheme of the application is as follows:

[0006] On the one hand, the application provides a power equipment management method based on a graph convolution network, comprising the following steps:

[0007] Obtaining the operation data of the power equipment, obtaining the causal feature vector of the power equipment based on the operation data;

[0008] Calculating the similarity between each causal feature vector, obtaining the causal strength between each power equipment through the similarity, and constructing a causal strength matrix based on the causal strength;

[0009] Inputting the causal feature vector and the causal strength matrix into a graph convolution network, outputting a node embedding matrix, obtaining a cluster center vector, determining the cluster label of each power equipment through the node embedding matrix and the cluster center vector, and constructing an equipment cluster set based on the power equipment with the same cluster label;

[0010] Obtaining the coordination index of each equipment cluster set;

[0011] Based on the aforementioned collaborative indicators, the resource allocation amount of each equipment cluster set is obtained, the resource allocation weight of each power equipment in the equipment cluster set is obtained, and the resource allocation amount of the power equipment is obtained based on the resource allocation weight and the resource allocation amount of the equipment cluster set.

[0012] A scheduling plan is generated based on the resource allocation of each power device in the device cluster set.

[0013] Preferably, the method further includes standardizing the running data, expressed by the formula:

[0014] ;

[0015] In the formula, Indicates the first Power equipment The operational data after standardization at all times. Indicates the first Power equipment Real-time runtime data, This represents the average value of the operating data of power equipment. The standard deviation of power equipment operating data This represents a preset constant.

[0016] Preferably, the causal feature vector of the power equipment is obtained based on the operational data, and the specific steps are as follows:

[0017] Based on the standardized runtime data, construct query vectors, key vectors, and value vectors;

[0018] The query vector is expressed by the formula:

[0019] ;

[0020] In the formula, Indicates the first Power equipment The query vector at time step, Represents the query weight matrix;

[0021] The key vector is expressed by the formula:

[0022] ;

[0023] In the formula, Indicates the first Power equipment The key vector at time step, Represents the key weight matrix. Indicates the first Power equipment Operational data after time-standardization processing;

[0024] The value vector is expressed by the formula:

[0025] ;

[0026] In the formula, Indicates the first Power equipment The value vector at time, Represents the value weight matrix;

[0027] The attention weights between each power device are obtained through the query vector and key vector, expressed by the following formula:

[0028] ;

[0029] In the formula, Indicates the first The power equipment and the first Power equipment Attention weight at any moment Represents the key vector transpose, Indicates the prior causal weight, Indicates the first The power equipment and the first Prior causal strength of an electrical device Indicates the index of the number of electrical equipment. Indicates the first Power equipment The key vector at time step, Represents the key vector transpose, Indicates the first The power equipment and the first Prior causal strength of an electrical device The dimension of the key vector. Indicates the quantity of electrical equipment;

[0030] The causal feature vector of the power equipment is obtained based on the attention weights and value vectors, expressed by the formula:

[0031] ;

[0032] In the formula, Indicates the first Power equipment The causal feature vector at time step, Represents the ReLU activation function. Representation layer normalization.

[0033] Preferably, the similarity between each causal feature vector is calculated, the causal strength between each power device is obtained through the similarity, and a causal strength matrix is ​​constructed based on the causal strength. The specific steps are as follows:

[0034] The similarity between each causal feature vector is calculated using the following formula:

[0035] ;

[0036] In the formula, Indicates the first The causal feature vector of the first power device and the first The similarity of causal feature vectors of individual electrical devices Indicates the first Power equipment The causal feature vector at time step, The L2 norm of a vector. Indicates the first The average causal feature vector of each power device This represents the Gaussian kernel width parameter. Indicates the first The average causal feature vector of each power device Indicates the management cycle;

[0037] The causal strength between the various power devices is obtained through the similarity, expressed by the following formula:

[0038] ;

[0039] In the formula, Indicates the first The power equipment and the first Causal strength between electrical devices This represents the similarity weighting factor. This represents the causal weighting factor.

[0040] Preferably, the causal feature vector and causal strength matrix are input into a graph convolutional network, which outputs a node embedding matrix to obtain cluster center vectors. The cluster labels for each power device are determined using the node embedding matrix and cluster center vectors. A device cluster set is then constructed based on power devices with the same cluster label. The specific steps are as follows:

[0041] Construct an average causal feature matrix based on the average causal feature vector;

[0042] The average causal feature matrix and causal strength matrix are input into the graph convolutional network, expressed by the formula:

[0043] ;

[0044] ;

[0045] ;

[0046] In the formula, This represents the hidden state of the first layer of the graph convolutional network. Degree matrix, This represents the initial input feature matrix of the graph convolutional network. This represents the weight matrix of the first layer of the graph convolutional network. This represents the self-connected causal strength matrix. Represents the average causal characteristic matrix. Represents the node embedding matrix. This represents the weight matrix of the second-layer graph convolutional network;

[0047] The self-connected causal strength matrix is ​​expressed by the formula:

[0048] ;

[0049] In the formula, Represents the identity matrix. Represents the causal strength matrix;

[0050] Based on the embedding vectors of each power device in the node embedding matrix, multiple cluster center vectors are generated using the k-means++ algorithm;

[0051] The soft allocation probability of power equipment is obtained based on the embedding vector and cluster center vector, expressed by the formula:

[0052] ;

[0053] In the formula, Indicates the first The electrical equipment belongs to the first Cluster center vectors The soft assignment probability, Represents the node embedding matrix. Embedding vectors of electrical devices This represents the number of cluster center vectors. Indicates the first Cluster center vectors, Indicates the index of the number of cluster center vectors;

[0054] The clustering labels for power equipment are determined based on the soft assignment probability, expressed by the following formula:

[0055] ;

[0056] In the formula, Indicates the first Clustering tags for electrical equipment Indices represent the index of the number of cluster center vectors with the highest probability of soft assignment. .

[0057] Preferably, the collaborative index of each device cluster set is obtained, and the specific steps are as follows:

[0058] The feature vectors of each device cluster set are constructed, expressed by the formula:

[0059] ;

[0060] In the formula, Indicates the first Feature vectors of each device cluster set Indicates the first A set of device clusters, Indicates the first The cardinality of a device cluster set, i.e., the number of electrical devices in the device cluster set. Represents the maximum value function. Describes the minimum value function;

[0061] The feature vectors of each device cluster set are used as input to the multilayer perceptron (MLP), and the collaborative index of each device cluster set is output, expressed by the formula:

[0062] ;

[0063] In the formula, Indicates the first Collaborative metrics for clusters of devices This represents a multilayer perceptron (MLP).

[0064] Preferably, the resource allocation amount of each equipment cluster is obtained based on the collaborative index, the resource allocation weight of each power equipment in the equipment cluster is obtained, and the resource allocation amount of the power equipment is obtained based on the resource allocation weight and the resource allocation amount of the equipment cluster. The specific steps are as follows:

[0065] The priority score of each device cluster set is obtained based on the aforementioned collaboration index, expressed by the formula:

[0066] ;

[0067] In the formula, Indicates the first The priority score of each device cluster set. Indicates the first Consistency index of a cluster of devices Indicates the first Key metrics for each device cluster set Indicates the first weighting factor. This represents the second weighting factor. Indicates the third weighting factor. This represents the fourth weighting factor;

[0068] The consistency index is expressed by the formula:

[0069] ;

[0070] In the formula, Represents the node embedding matrix. Embedding vectors of each power device;

[0071] The key indicators are expressed by the following formula:

[0072] ;

[0073] The resource allocation for each device cluster is determined by the priority score, expressed by the formula:

[0074] ;

[0075] In the formula, Indicates the first Resource allocation for each device cluster set Indicates the first The priority score of each device cluster set. Indicates the total amount of resources;

[0076] Construct device cluster set The resource allocation feature vector of each power device in the system is expressed by the formula:

[0077] ;

[0078] ;

[0079] In the formula, Represents a set of device clusters The Middle Resource allocation feature vector of each power device Indicates the first The power equipment and the first Causal strength between electrical devices;

[0080] The resource allocation feature vector is used as input to a multilayer perceptron (MLP), and the output resource allocation weights are expressed by the following formula:

[0081] ;

[0082] In the formula, Represents a set of device clusters The Middle Resource allocation weights for each power device;

[0083] The resource allocation weights are normalized, as expressed by the formula:

[0084] ;

[0085] In the formula, Represents a set of device clusters The Middle The normalized resource allocation weights for each power device Represents a set of device clusters The Middle Resource allocation weights for each power device;

[0086] The resource allocation amount of power equipment is obtained based on the normalized resource allocation weights and the resource allocation amount of the equipment cluster set, expressed by the formula:

[0087] ;

[0088] In the formula, Represents a set of device clusters The Middle Resource allocation for each power device.

[0089] Preferably, a scheduling plan is generated based on the resource allocation of each power device in the device cluster set, and the specific steps are as follows:

[0090] Based on the resource allocation amount and resource allocation weight of each power device in the aforementioned device cluster set, a scheduling feature vector is constructed, expressed by the formula:

[0091] ;

[0092] In the formula, Indicates the first Dispatch feature vectors of individual power equipment;

[0093] Using the scheduling feature vector as input to a multilayer perceptron (MLP), the output scheduling plan is expressed by the formula:

[0094] ;

[0095] In the formula, Indicates the first Dispatch plan for each power equipment.

[0096] Preferably, the mean of the operating data is expressed by the formula:

[0097] ;

[0098] The standard deviation of the operational data is expressed by the formula:

[0099] .

[0100] In another aspect, the present invention also provides an electronic device having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the power equipment management method based on graph convolutional networks as described in any embodiment of the present invention.

[0101] The present invention has the following beneficial effects:

[0102] 1. This invention employs an attention mechanism to construct causal feature vectors, enabling the capture of mutual influence and causal dependencies among power equipment. This process considers not only the operating state parameters of individual devices but also incorporates causal association information between devices, resulting in more comprehensive and accurate feature representations. Compared to traditional static feature extraction methods, this approach better reflects the complex coupling characteristics and nonlinear relationships of power systems, highlights the influence of key equipment on surrounding equipment, and effectively uncovers the causal relationships behind the data.

[0103] 2. This invention quantifies the implicit relationships between devices into a computable structured data network by calculating the causal strength between devices and constructing a relationship matrix. This matrix representation clearly shows the potential fault propagation paths and the scope of impact, helping to identify key node devices in the system.

[0104] 3. The graph convolutional network clustering method of this invention fully utilizes the topological connections and causal relationships between power equipment, making the clustering results more consistent with the physical operation of the power system. Compared with traditional clustering algorithms that rely solely on distance metrics, this invention can effectively discover functionally highly related and electrically connected equipment structures, grouping equipment with significant mutual influence into a single group for management.

[0105] 4. This invention utilizes a multilayer perceptron to evaluate the collaborative indicators of a device cluster, achieving a comprehensive evaluation shift from focusing on individual device performance to focusing on the collaborative capabilities of the group. This indicator not only reflects the synchronicity and coordination of the devices within the cluster, but more importantly, it embodies the collaborative work efficiency of the device group as a whole.

[0106] 5. This invention allocates resources based on collaborative indicators, establishing a direct mapping relationship from operational data to management decisions, thus achieving intelligent allocation. The allocation mechanism comprehensively considers the overall priority of the equipment cluster and the causal weights between devices within the cluster, achieving reasonable resource allocation across different levels. This adjustment mechanism avoids subjective biases and resource waste caused by traditional experience-based judgments, ensuring that key areas and important equipment receive appropriate management resources, improving resource utilization efficiency and management fairness. Attached Figure Description

[0107] Figure 1 This is a flowchart illustrating the implementation of the method in an embodiment of the present invention. Detailed Implementation

[0108] 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, and 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.

[0109] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0110] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0111] The terms “comprising” and “including” indicate the presence of the described feature, whole, step, operation, element and / or component, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0112] The term “and / or” refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes these combinations.

[0113] Example 1:

[0114] See Figure 1 This embodiment provides a power equipment management method based on graph convolutional networks, including the following steps:

[0115] S1. Obtain the operating data of the power equipment and perform standardized processing, expressed as a formula:

[0116] ;

[0117] In the formula, Indicates the first Power equipment The operational data after standardization at all times. Indicates the first Power equipment Real-time runtime data, This represents the average value of the operating data of power equipment. The standard deviation of power equipment operating data This represents a preset constant used to prevent the denominator from being zero;

[0118] The mean of the operational data is expressed by the formula:

[0119] ;

[0120] In the formula, Indicates the number of electrical equipment. Indicates the management cycle;

[0121] The standard deviation of the operational data is expressed by the formula:

[0122] .

[0123] The operating data includes current, voltage, power, equipment temperature, and ambient temperature.

[0124] Standardization eliminates the differences in dimensions and numerical ranges of operating data from different power equipment, enabling subsequent calculations and analyses of various monitoring data such as voltage, current, and temperature to be performed on a unified scale.

[0125] S2. Construct query vectors, key vectors, and value vectors based on the standardized operational data. Obtain attention weights between power devices using the query vectors and key vectors. Obtain causal feature vectors of power devices based on the attention weights and value vectors. The specific steps are as follows:

[0126] S21. Construct query vectors, key vectors, and value vectors based on the standardized running data;

[0127] The query vector is expressed by the formula:

[0128] ;

[0129] In the formula, Indicates the first Power equipment The query vector at time step, Represents the query weight matrix;

[0130] The key vector is expressed by the formula:

[0131] ;

[0132] In the formula, Indicates the first Power equipment The key vector at time step, Represents the key weight matrix. Indicates the first Power equipment Operational data after time-standardization processing;

[0133] The value vector is expressed by the formula:

[0134] ;

[0135] In the formula, Indicates the first Power equipment The value vector at time, Represents the value weight matrix;

[0136] Information exchange between power equipment is simulated by query vectors, key vectors, and value vectors.

[0137] S22. Obtain the attention weights between each power device using the query vector and key vector, expressed as a formula:

[0138] ;

[0139] In the formula, Indicates the first The power equipment and the first Power equipment Attention weight at any moment Represents the key vector transpose, Indicates the prior causal weight, Indicates the first The power equipment and the first Prior causal strength of an electrical device Indicates the index of the number of electrical equipment. Indicates the first Power equipment The key vector at time step, Represents the key vector transpose, Indicates the first The power equipment and the first Prior causal strength of an electrical device Indicates the dimension of the key vector;

[0140] Among them, if the first The electrical equipment is directly connected to the first If an electrical device has an electrical connection, then the a priori causal strength is... =1, if the first A power device connects to the first power device via a power device. If an electrical device has an electrical connection, then the a priori causal strength is... =0.5, otherwise prior causality strength =0;

[0141] For example: If electrical equipment A is directly electrically connected to electrical equipment B, electrical equipment B is directly electrically connected to electrical equipment C, and electrical equipment C is directly electrically connected to electrical equipment D, then the prior causal strength between electrical equipment A and electrical equipment B is 1, the prior causal strength between electrical equipment A and electrical equipment C is 0.5, and the prior causal strength between electrical equipment A and electrical equipment D is 0.

[0142] When calculating attention weights, both feature similarity and prior causal strength among power devices are considered. Feature similarity is calculated by the dot product of the query vector and the key vector, reflecting the potential correlation strength between power devices. Prior causal strength is derived from the topology of the power devices and introduced as a bias term in the calculation process, conforming to the physical laws of causal propagation in power systems. Finally, through layer normalization and nonlinear activation processing, a causal feature vector with stable distribution and strong expressive power is obtained.

[0143] S23. Obtain the causal feature vector of the power equipment based on the attention weights and value vectors, expressed by the formula:

[0144] ;

[0145] In the formula, Indicates the first Power equipment The causal feature vector at time step, Represents the ReLU activation function. Representation layer normalization.

[0146] S3. Calculate the similarity between each causal feature vector, obtain the causal strength between each power device through the similarity, and construct a causal strength matrix based on the causal strength. The specific steps are as follows:

[0147] S31. Calculate the similarity between each causal feature vector, expressed by the formula:

[0148] ;

[0149] In the formula, Indicates the first The causal feature vector of the first power device and the first The similarity of causal feature vectors of individual electrical devices Indicates the first Power equipment The causal feature vector at time step, The L2 norm of a vector. Indicates the first The average causal feature vector of each power device This represents the Gaussian kernel width parameter, used to control the degree to which feature differences affect similarity. Indicates the first The average causal feature vector of each power device;

[0150] Similarity is calculated using cosine similarity, which involves summing the dot product of the causal feature vectors of two electrical devices over the entire time span and dividing by the square root of the product of the magnitudes of their respective feature vectors. This method effectively assesses the synchronicity and consistency of equipment operating modes over time, capturing similarities in dynamic behavior between devices. Feature space similarity is calculated using a Gaussian kernel function, which calculates the squared Euclidean distance between the average causal feature vectors of the devices and applies an exponential decay function. This method primarily assesses the similarity of the static distribution of electrical devices in the feature space, reflecting the similarity of their inherent characteristics. Multiplying these two similarities yields the final similarity score, ensuring that only electrical devices similar in both time and feature space achieve high similarity.

[0151] The average causal feature vector is expressed by the formula:

[0152] ;

[0153] ;

[0154] S32. Obtain the causal strength between each power device through the similarity, expressed by the formula:

[0155] ;

[0156] In the formula, Indicates the first The power equipment and the first Causal strength between electrical devices This represents the similarity weighting factor. Indicates causal weighting factor;

[0157] S33. Construct a causal intensity matrix based on the causal intensity.

[0158] S4. Input the causal feature vector and causal strength matrix into a graph convolutional network, outputting a node embedding matrix. Based on the embedding vectors of each power device in the node embedding matrix, use the k-means++ algorithm to generate multiple cluster center vectors. Determine the cluster label of each power device through the embedding vectors and cluster center vectors. Construct a device cluster set based on power devices with the same cluster label. The specific steps are as follows:

[0159] S41. Construct an average causal feature matrix based on the average causal feature vector, expressed by the formula:

[0160] ;

[0161] In the formula, Represents the average causal characteristic matrix;

[0162] S42. Input the average causal feature matrix and causal strength matrix into the graph convolutional network. The specific steps are as follows:

[0163] S421. The average causal feature matrix is ​​used as the initial input feature matrix of the graph convolutional network. Together with the causal strength matrix, it is input into the first layer of the graph convolutional network. The hidden state of the first layer of the graph convolutional network is output, expressed by the formula:

[0164] ;

[0165] ;

[0166] In the formula, This represents the hidden state of the first layer of the graph convolutional network. Degree matrix, This represents the initial input feature matrix of the graph convolutional network. This represents the weight matrix of the first layer of the graph convolutional network. Represents the self-connected causal strength matrix;

[0167] The self-connected causal strength matrix is ​​expressed by the formula:

[0168] ;

[0169] In the formula, Represents the identity matrix. Represents the causal strength matrix;

[0170] The degree matrix is ​​expressed by the formula:

[0171] ;

[0172] In the formula, The degree matrix represents the first degree. One diagonal element, Represents the self-connected causal strength matrix No. Line 1 Column elements;

[0173] S422. Input the hidden states of the first-layer graph convolutional network into the second-layer graph convolutional network, and output the node embedding matrix, which is expressed by the formula:

[0174] ;

[0175] In the formula, Represents the node embedding matrix. This represents the weight matrix of the second-layer graph convolutional network;

[0176] S43. Based on the embedding vectors of each power device in the node embedding matrix, the k-means++ algorithm is used to generate multiple cluster center vectors;

[0177] S44. Determine the cluster label of each power device using the embedding vector and cluster center vector. The specific steps are as follows:

[0178] S441. Based on the embedding vector and cluster center vector, obtain the soft allocation probability of the power equipment, expressed by the formula:

[0179] ;

[0180] In the formula, Indicates the first The electrical equipment belongs to the first Cluster center vectors The soft assignment probability, Represents the node embedding matrix. Embedding vectors of electrical devices This represents the number of cluster center vectors. Indicates the first Cluster center vectors, Indicates the index of the number of cluster center vectors;

[0181] S442. Determine the clustering label of the power equipment based on the soft assignment probability, expressed by the formula:

[0182] ;

[0183] In the formula, Indicates the first Cluster labels for each power device are used to identify which cluster center vector each power device belongs to. Indices represent the index of the number of cluster center vectors with the highest probability of soft assignment. .

[0184] S45. Construct a set of equipment clusters based on the same cluster label for power equipment.

[0185] S5. Obtain the collaboration index of each device cluster set. The specific steps are as follows:

[0186] S51. Construct the feature vector of each device cluster set, expressed by the formula:

[0187] ;

[0188] In the formula, Indicates the first Feature vectors of each device cluster set Indicates the first A set of device clusters, Indicates the first The cardinality of a device cluster set, i.e., the number of electrical devices in the device cluster set. Represents the maximum value function. Describes the minimum value function;

[0189] S52. Using the feature vectors of each device cluster set as input to the multilayer perceptron (MLP), the collaborative index of each device cluster set is output, expressed by the formula:

[0190] ;

[0191] In the formula, Indicates the first Collaborative metrics for clusters of devices This represents a multilayer perceptron (MLP).

[0192] S6. Based on the collaborative index, obtain the priority score of each equipment cluster set, determine the resource allocation amount of each equipment cluster set through the priority score, obtain the resource allocation weight of each power equipment in the equipment cluster set, and obtain the resource allocation amount of the power equipment based on the resource allocation weight and the resource allocation amount of the equipment cluster set. The specific steps are as follows:

[0193] S61. Obtain the priority score of each device cluster set based on the aforementioned collaboration index, expressed by the formula:

[0194] ;

[0195] In the formula, Indicates the first The priority score of each device cluster set. Indicates the first Consistency index of a cluster of devices Indicates the first Key metrics for each device cluster set Indicates the first weighting factor. This represents the second weighting factor. Indicates the third weighting factor. This represents the fourth weighting factor;

[0196] The consistency index is expressed by the formula:

[0197] ;

[0198] In the formula, Represents the node embedding matrix. Embedding vectors of each power device;

[0199] The key indicators are expressed by the following formula:

[0200] ;

[0201] S62. The resource allocation for each device cluster is determined based on the priority score, expressed by the formula:

[0202] ;

[0203] In the formula, Indicates the first Resource allocation for each device cluster set Indicates the first The priority score of each device cluster set. Indicates the total amount of resources;

[0204] S63. Obtain the resource allocation weights of each power device in the device cluster set. The specific steps are as follows:

[0205] S631, Constructing Equipment Cluster Sets The resource allocation feature vector of each power device in the system is expressed by the formula:

[0206] ;

[0207] ;

[0208] In the formula, Represents a set of device clusters The Middle Resource allocation feature vector of each power device Indicates the first The power equipment and the first Causal strength between electrical devices;

[0209] S632. The resource allocation feature vector is used as the input of a multilayer perceptron (MLP), and the resource allocation weights are output as follows:

[0210] ;

[0211] In the formula, Represents a set of device clusters The Middle Resource allocation weights for each power device;

[0212] S633. The resource allocation weights are normalized, expressed by the following formula:

[0213] ;

[0214] In the formula, Represents a set of device clusters The Middle The normalized resource allocation weights for each power device Represents a set of device clusters The Middle Resource allocation weights for each power device;

[0215] S64. Based on the normalized resource allocation weights and the resource allocation amounts of the equipment cluster sets, the resource allocation amounts of the power equipment are obtained, expressed by the formula:

[0216] ;

[0217] In the formula, Represents a set of device clusters The Middle The resource allocation of each power device is determined by the amount of resources allocated to it. The higher the resource allocation of a power device, the higher the priority of the corresponding scheduling plan, and the more detailed the scheduling plan.

[0218] Power equipment with low resource allocation executes a dispatch plan every 2 hours, while power equipment with high resource allocation may execute a dispatch plan every 1-1.5 hours. For example, in voltage regulation, because power equipment with high resource allocation has a shorter time interval, the range of each adjustment will be more precise and detailed than that of power equipment with low resource allocation (power equipment with high resource allocation adjusts by 0.2 each time, while power equipment with low resource allocation adjusts by 0.3 each time).

[0219] S7. Generate a scheduling plan based on the resource allocation of each power device in the device cluster set. The specific steps are as follows:

[0220] S71. Construct a scheduling feature vector based on the resource allocation amount and resource allocation weight of each power device in the device cluster set, expressed by the formula:

[0221] ;

[0222] In the formula, Indicates the first Dispatch feature vectors of individual power equipment;

[0223] S72. Using the scheduling feature vector as input to a multilayer perceptron (MLP), the output scheduling plan is expressed by the formula:

[0224] ;

[0225] In the formula, Indicates the first The scheduling plan for each power equipment includes voltage regulation, charging or discharging, power regulation, etc.

[0226] Example 2:

[0227] This embodiment provides an electronic device that stores a computer program. When the computer program is executed by a processor, it implements the power equipment management method based on graph convolutional networks as described in any embodiment of the present invention.

[0228] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0229] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. 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 implementation should not be considered beyond the scope of this application.

[0230] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0231] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0232] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A power equipment management method based on graph convolutional networks, characterized in that, Includes the following steps: Obtain operational data of power equipment, and obtain causal feature vectors of power equipment based on the operational data; The similarity between each causal feature vector is calculated using the following formula: ； In the formula, Indicates the first The causal feature vector of the first power device and the first The similarity of causal feature vectors of individual electrical devices Indicates the first Power equipment The causal feature vector at time step, Indicates the first Power equipment The causal feature vector at time step, The L2 norm of a vector. Indicates the first The average causal feature vector of each power device This represents the Gaussian kernel width parameter. Indicates the first The average causal feature vector of each power device Indicates the management cycle; The causal strength between the various power devices is obtained through the similarity, expressed by the following formula: ； In the formula, Indicates the first The power equipment and the first Causal strength between electrical devices This represents the similarity weighting factor. Indicates causal weighting factor. Indicates the first The power equipment and the first The a priori causal strength of an electrical device; Construct a causality strength matrix based on the causality strength; The causal feature vector and causal strength matrix are input into a graph convolutional network, which outputs a node embedding matrix to obtain cluster center vectors. The cluster labels of each power device are determined by the node embedding matrix and cluster center vectors, and a set of device clusters is constructed based on the power devices with the same cluster label. Obtain the collaborative metrics of each device cluster set; Based on the aforementioned collaborative indicators, the resource allocation amount of each equipment cluster set is obtained, the resource allocation weight of each power equipment in the equipment cluster set is obtained, and the resource allocation amount of the power equipment is obtained based on the resource allocation weight and the resource allocation amount of the equipment cluster set. A scheduling plan is generated based on the resource allocation of each power device in the device cluster set.

2. The power equipment management method based on graph convolutional networks according to claim 1, characterized in that, The method further includes standardizing the runtime data, expressed as a formula: ； In the formula, Indicates the first Power equipment The operational data after standardization at all times. Indicates the first Power equipment Real-time runtime data, This represents the average value of the operating data of power equipment. The standard deviation of power equipment operating data This represents a preset constant.

3. The power equipment management method based on graph convolutional networks according to claim 2, characterized in that, The specific steps for obtaining the causal feature vector of the power equipment based on the aforementioned operational data are as follows: Based on the standardized runtime data, construct query vectors, key vectors, and value vectors; The query vector is expressed by the formula: ； In the formula, Indicates the first Power equipment The query vector at time step, Represents the query weight matrix; The key vector is expressed by the formula: ； In the formula, Indicates the first Power equipment The key vector at time step, Represents the key weight matrix. Indicates the first Power equipment Operational data after time-standardization processing; The value vector is expressed by the formula: ； In the formula, Indicates the first Power equipment The value vector at time, Represents the value weight matrix; The attention weights between each power device are obtained through the query vector and key vector, expressed by the following formula: ； In the formula, Indicates the first The power equipment and the first Power equipment Attention weight at any moment Represents the key vector transpose, Indicates the prior causal weight, Indicates the index of the number of electrical equipment. Indicates the first Power equipment The key vector at time step, Represents the key vector transpose, Indicates the first The power equipment and the first Prior causal strength of an electrical device The dimension of the key vector. Indicates the quantity of electrical equipment; The causal feature vector of the power equipment is obtained based on the attention weights and value vectors, expressed by the formula: ； In the formula, Represents the ReLU activation function. Representation layer normalization.

4. The power equipment management method based on graph convolutional networks according to claim 1, characterized in that, The specific steps for constructing a cluster set of power equipment based on the same cluster label are as follows: Construct an average causal feature matrix based on the average causal feature vector; The average causal feature matrix and causal strength matrix are input into the graph convolutional network, expressed by the formula: ； ； ； In the formula, This represents the hidden state of the first layer of the graph convolutional network. Degree matrix, This represents the initial input feature matrix of the graph convolutional network. This represents the weight matrix of the first layer of the graph convolutional network. This represents the self-connected causal strength matrix. Represents the average causal characteristic matrix. Represents the node embedding matrix. This represents the weight matrix of the second-layer graph convolutional network; The self-connected causal strength matrix is ​​expressed by the formula: ； In the formula, Represents the identity matrix. Represents the causal strength matrix; Based on the embedding vectors of each power device in the node embedding matrix, multiple cluster center vectors are generated using the k-means++ algorithm; The soft allocation probability of power equipment is obtained based on the embedding vector and cluster center vector, expressed by the formula: ； In the formula, Indicates the first The electrical equipment belongs to the first Cluster center vectors The soft assignment probability, Represents the node embedding matrix. Embedding vectors of electrical devices This represents the number of cluster center vectors. Indicates the first Cluster center vectors, Indicates the index of the number of cluster center vectors; The clustering labels for power equipment are determined based on the soft assignment probability, expressed by the following formula: ； In the formula, Indicates the first Clustering tags for electrical equipment Indices represent the index of the number of cluster center vectors with the highest probability of soft assignment. .

5. The power equipment management method based on graph convolutional networks according to claim 4, characterized in that, The specific steps to obtain the collaboration index of each device cluster set are as follows: The feature vectors of each device cluster set are constructed, expressed by the formula: ； In the formula, Indicates the first Feature vectors of each device cluster set Indicates the first A set of device clusters, Indicates the first The cardinality of a device cluster set, i.e., the number of electrical devices in the device cluster set. Represents the maximum value function. Describes the minimum value function; The feature vectors of each device cluster set are used as input to the multilayer perceptron (MLP), and the collaborative index of each device cluster set is output, expressed by the formula: ； In the formula, Indicates the first Collaborative metrics for clusters of devices This represents a multilayer perceptron (MLP).

6. The power equipment management method based on graph convolutional networks according to claim 5, characterized in that, The resource allocation amount of power equipment is obtained based on the resource allocation weight and the resource allocation amount of the equipment cluster set. The specific steps are as follows: The priority score of each device cluster set is obtained based on the aforementioned collaboration index, expressed by the formula: ； In the formula, Indicates the first The priority score of each device cluster set. Indicates the first Consistency index of a cluster of devices Indicates the first Key metrics for each device cluster set Indicates the first weighting factor. This represents the second weighting factor. Indicates the third weighting factor. This represents the fourth weighting factor; The consistency index is expressed by the formula: ； In the formula, Represents the node embedding matrix. Embedding vectors of each power device; The key indicators are expressed by the following formula: ； The resource allocation for each device cluster is determined by the priority score, expressed by the formula: ； In the formula, Indicates the first Resource allocation for each device cluster set Indicates the first The priority score of each device cluster set. Indicates the total amount of resources; Construct device cluster set The resource allocation feature vector of each power device in the system is expressed by the formula: ； ； In the formula, Represents a set of device clusters The Middle Resource allocation feature vector of each power device Indicates the first The power equipment and the first Causal strength between electrical devices; The resource allocation feature vector is used as input to a multilayer perceptron (MLP), and the output resource allocation weights are expressed by the following formula: ； In the formula, Represents a set of device clusters The Middle Resource allocation weights for each power device; The resource allocation weights are normalized, as expressed by the formula: ； In the formula, Represents a set of device clusters The Middle The normalized resource allocation weights for each power device Represents a set of device clusters The Middle Resource allocation weights for each power device; The resource allocation amount of power equipment is obtained based on the normalized resource allocation weights and the resource allocation amount of the equipment cluster set, expressed by the formula: ； In the formula, Represents a set of device clusters The Middle Resource allocation for each power device.

7. The power equipment management method based on graph convolutional networks according to claim 6, characterized in that, The scheduling plan is generated based on the resource allocation of each power device in the device cluster set. The specific steps are as follows: Based on the resource allocation amount and resource allocation weight of each power device in the aforementioned device cluster set, a scheduling feature vector is constructed, expressed by the formula: ； In the formula, Indicates the first Dispatch feature vectors of individual power equipment; Using the scheduling feature vector as input to a multilayer perceptron (MLP), the output scheduling plan is expressed by the formula: ； In the formula, Indicates the first Dispatch plan for each power equipment.

8. The power equipment management method based on graph convolutional networks according to claim 2, characterized in that, The mean of the operational data is expressed by the formula: ； The standard deviation of the operational data is expressed by the formula: 。 9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the power equipment management method based on graph convolutional networks as described in any one of claims 1 to 8.

Images