A power distribution equipment safe operation detection method and system

Abstract

The application relates to the field of power distribution, in particular to a power distribution equipment safe operation detection method and system, the method comprising the following steps: constructing and training a graph neural network about power distribution equipment; the nodes of the graph neural network represent each equipment, and operation data of each power distribution equipment is collected; edges represent electrical connection relationships between the equipment; the input is the operation data and the electrical connection relationship of the equipment, and the output is a fault probability distribution of each equipment for each fault type; predicted fault nodes are obtained by screening nodes according to the output fault probability distribution, the migration influence of the predicted fault nodes on other nodes in the case of each fault type is calculated, and the predicted fault probability is obtained by adjusting the fault probability of each node according to the migration influence of the node. The application has the effects of improving the accuracy and reliability of fault prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power distribution, and particularly relates to a power distribution equipment safe operation detection method and system. BACKGROUND

[0002] Power distribution equipment refers to various devices used for power distribution and control, which transmit electrical energy from power plants or substations to end users (such as homes, factories, commercial buildings, etc.). Power distribution equipment covers all aspects from power distribution, control, protection to monitoring and intelligent management. They play a key role in the power system, ensuring the safe, reliable and efficient distribution of electrical energy. Graph Neural Networks (GNNs) as a new machine learning method is particularly suitable for handling problems with complex relationships and structured data, so it can be applied to the safe operation detection of power distribution equipment, such as determining whether a device is in a fault state (normal / abnormal). Each node corresponds to a class probability distribution.

[0003] However, when a device fails, it will not only affect its own operating state, but also have a cascading effect on adjacent or associated devices through electrical connections. For example, if a node fails, it will affect the nodes related to it, thereby changing the failure probability of other related nodes. This effect can cause the failure probability to spread and spread among devices. Existing technologies cannot fully reflect the impact of device interaction on the overall system safety, which may underestimate or overestimate the actual failure risk of some devices. SUMMARY

[0004] In order to solve the above technical problem that the interaction between devices cannot fully reflect the impact on the overall system safety, the present application provides a power distribution equipment safe operation detection method and system.

[0005] In a first aspect, the application provides a power distribution equipment safe operation detection method, which adopts the following technical scheme: a power distribution equipment safe operation detection method, comprising the steps of: constructing and training a graph neural network about power distribution equipment; the nodes of the graph neural network represent each equipment, and the running data of each power distribution equipment is collected; the edges represent the electrical connection relationship between the equipments; the input is the running data of the equipment and the electrical connection relationship, and the output is the fault probability distribution of each equipment for each fault type; the predicted fault nodes are obtained by screening the nodes according to the output fault probability distribution, the migration influence of the predicted fault nodes on other nodes in the case of each fault type is calculated, and the predicted fault probability is obtained by adjusting the fault probability of each node according to the migration influence of the node; the calculation method of the migration influence is: the node influence order is adaptively set according to the size of the graph neural network; the similarity between each node in each fault type is calculated according to the historical data set in the training process of the graph neural network, and the similarity is normalized and taken as the edge weight; the migration influence is obtained according to the edge weight between the nodes, the fault probability of the fault node and the influence order, the edge weight and the migration influence are proportional, and the influence order and the migration influence are inversely proportional.

[0006] Optionally, the node influence order is adaptively set according to the size of the graph neural network, comprising: taking the product of the ratio of the node order of the node to the maximum node order of the graph neural network and the preset fixed order as the node influence order after rounding.

[0007] Optionally, the calculation method of the similarity is: constructing the fault label sequence of each node, counting the number of time points at which both fault label sequences exist, and taking the ratio of the number of time points to the total number of time points of the fault label as the similarity.

[0008] Optionally, the calculation formula of the migration influence is: ; wherein, represents the migration influence of node on node , represents the edge weight between node and node , represents the influence order of node , represents the fault probability of node existing fault ; represents the exponential function with as the base.

[0009] Optionally, the calculation formula of the migration influence is: ; wherein, represents the migration influence of node on node , representing nodes and edges between nodes edge weights of edges between nodes, representing nodes influence order of nodes, representing nodes failure probability of a failure .

[0010] Optionally, the predicted failure nodes are screened from the output failure probability distribution, including setting a failure probability threshold for each failure type, and determining that a node is a predicted failure node when the failure probability corresponding to any failure type in the failure probability distribution is greater than the failure probability threshold.

[0011] Optionally, the predicted failure nodes are screened from the output failure probability distribution, and an abnormal node is identified as a predicted failure node using an anomaly detection algorithm.

[0012] Optionally, the failure probability of each node is adjusted according to the migration influence of the node to obtain a predicted failure probability, and the product of the migration influence and the failure probability is taken as the predicted failure probability.

[0013] Optionally, the operating data of the device includes current, voltage, or power.

[0014] In a second aspect, the present application provides a power distribution equipment safe operation detection system, which adopts the following technical scheme:

[0015] A power distribution equipment safe operation detection system, comprising a processor and a memory, the memory storing computer program instructions, when the computer program instructions are executed by the processor, realizing the power distribution equipment safe operation detection method according to the above.

[0016] Beneficial effects are that the power distribution equipment safe operation detection method is generated into a computer program and stored in a memory to be loaded and executed by a processor, so that a system is made according to the memory and the processor, and use is facilitated.

[0017] The present application has the following technical effects:

[0018] 1. The present application not only focuses on the operating state and failure probability distribution of a single device, but also fully considers the electrical connection relationship between devices and its cascading effect. By combining migration influence with failure probability to adjust predicted failure probability, the dynamic interaction between devices is effectively captured, thereby significantly improving the accuracy and reliability of failure prediction and avoiding underestimation or overestimation of failure risk due to neglecting the interaction between devices.

[0019] 2、The calculation method of migration influence can reflect the interaction and fault propagation law between power distribution equipment by combining the size of the graph neural network, the similarity of nodes in the historical data set and the characteristics of the fault node. Specifically, the order of node influence is adaptively set according to the size of the graph neural network, so as to ensure that the nodes with high order have a larger influence range; the similarity between nodes in the historical data set is calculated and normalized as edge weight, so as to quantify the correlation strength between nodes; finally, the migration influence is calculated by comprehensively considering the edge weight, fault probability and influence order, wherein the edge weight and the fault probability are proportional to the migration influence, and the influence order is inversely proportional to the migration influence, so as to accurately capture the propagation effect of the fault between the devices and improve the accuracy of fault prediction and the ability of system safety evaluation. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is a method flowchart of a power distribution equipment safe operation detection method according to an embodiment of the present application.

[0021] Figure 2 is a method flowchart of migration influence calculation in a power distribution equipment safe operation detection method according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] The embodiments of the present application disclose a power distribution equipment safe operation detection method, referring to Figure 1 , comprising steps S1-S2, specifically as follows:

[0023] S1: Construct and train a graph neural network about power distribution equipment.

[0024] The node of the graph neural network (GNN) represents each device, and the running data of each power distribution device is collected; the edge represents the electrical connection relationship between the devices. Each node represents a power distribution device (such as a transformer, a circuit breaker, etc.). The feature vector of the node contains the running data of the device, such as current, voltage or power. Each edge represents the electrical connection relationship between two devices.

[0025] For example, the power distribution system can include: transformer A, circuit breaker B, switch C, switch D. For each device, collect running data such as current, voltage or power. The electrical connection between devices can be represented by an adjacency matrix. The device connection data is an adjacency matrix. For example, it is assumed that the power distribution system includes the following devices and their connection relationships: transformer A is connected to circuit breaker B and switch D, circuit breaker B is connected to switch C, switch C has no further connection, and switch D is only connected to transformer A, which is represented by an adjacency matrix as follows: The first row : Transformer A is connected to circuit breaker B and switch D. The second row : Breaker B is connected to transformer A and switch C. Third row : Switch C is connected to breaker B only. Fourth row : Switch D is connected to transformer A only.

[0026] The training of the graph neural network is performed on a historical data set, which is a two-dimensional sequence, for example, for current data, the historical data set is the current value corresponding to each time, each current value is provided with a label, which is 1 or 0, 1 represents a fault, and 0 represents normal. For example, transformer A: (3rd time fault). The construction and training of the graph neural network are prior art. The input of the graph neural network is the running data of the device and the electrical connection relationship, and the output is the fault probability distribution of each device for each fault type. For example, the output of transformer A can be: fault probability 0.9.

[0027] S2: According to the output fault probability distribution, the predicted fault node is screened, the migration influence of the predicted fault node on other nodes in the case of each fault type is calculated, and the fault probability of the node is adjusted according to the migration influence of each node to obtain the predicted fault probability.

[0028] In one embodiment, the method of screening nodes is: setting a fault probability threshold for each fault type, when the fault probability corresponding to any fault type in the fault probability distribution is greater than the fault probability threshold, the node is determined to be a predicted fault node.

[0029] For example, the fault threshold is 0.5, and the fault probability of transformer A is 0.9, so the node (transformer A) is a fault node. The setting of the fault threshold can be adjusted according to the actual application scene, which will not be described here.

[0030] In other embodiments, an abnormal node can also be identified as a predicted fault node using an anomaly detection algorithm. Common anomaly detection algorithms include Isolation Forest, Local Outlier Factor (LOF), etc., which are prior art and will not be described here. Through the anomaly detection algorithm, nodes that are significantly different from the normal operation mode can be detected. For example, using the Isolation Forest algorithm, the current value of transformer A The current value (25) at the 3rd time deviates significantly from the normal range and is marked as abnormal, so the node (transformer A) is a fault node.

[0031] Referring to Figure 2 , the calculation method of migration influence includes steps S20-S22:

[0032] S20: The order of node influence is adaptively set according to the size of the graph neural network.

[0033] In one embodiment, the product of the ratio of the node degree of the node to the maximum node degree of the graph neural network and the preset fixed degree is rounded off to obtain the node influence degree.

[0034] The node degree refers to the total number of nodes that can be indirectly influenced by the node through the connection relationship, including direct neighbor nodes, neighbors of the direct neighbor nodes, and the like. It is assumed that the node degree of a node a refers to the total number of all nodes (including the node a itself) that can be reached through edges starting from the node a. For example, if the node a is directly connected to the node b, and the node b is connected to the node c, the node degree of the node a is 3 (including the node a itself, the node b, and the node c). The maximum node degree refers to the maximum value of the degrees of all nodes in the graph neural network.

[0035] If the maximum node degree is large, it indicates that there are some highly connected nodes in the graph, and the size of the overall graph is large. If the maximum node degree is small, it indicates that the connectivity of the nodes in the graph is weak, and the size of the overall graph is small. By adaptively setting the node influence degree according to the size of the graph neural network, the subsequent step of quantifying the propagation range and degree of the fault or anomaly between nodes can be more accurate.

[0036] S21: According to the historical data set in the graph neural network training process, the similarity between each node in each fault type is calculated, and the similarity is normalized to be used as the edge weight.

[0037] In one embodiment, the calculation method of the similarity can be: constructing the fault label sequence of each node, counting the number of time points at which both fault label sequences exist, and taking the ratio of the number of time points to the total number of time points of the fault label as the similarity. For example, if the fault label of sequence 1 at time 3 is 1, and the fault label of sequence 2 at time 3 is 1, then time 3 is counted in the number of time points with faults. For example, if the fault label of sequence 1 at time 4 is 0, and the fault label of sequence 3 at time 4 is 1, then time 4 is not counted in the number of time points with faults. The normalization can use the existing standard normalization method, and the prior art will not be described again.

[0038] S22: According to the edge weight between the nodes, the fault probability of the fault node, and the influence degree, the migration influence is obtained, the edge weight and the migration influence are proportional, and the influence degree and the migration influence are inversely proportional.

[0039] In one embodiment, the calculation formula of the migration influence is: ; wherein, represents the migration influence of the node on the node , represents the edge weight between the node and the node , and the larger the edge weight, the more the migration influence of the node For Node The stronger the migration effect, the weaker the migration effect.

[0040] Representation node The influence order of the node If the node The influence range of is large, then it has a large impact on a single node. The migration impact will be reduced (because the total impact is spread over more nodes). Representation node There is a fault The greater the failure probability, the greater the impact on other nodes; Indicates This embodiment involves exponential operations, emphasizing the exponential decay effect, and is suitable for describing situations where the impact rapidly decreases with distance, and is applicable to scenarios where the impact rapidly decreases with distance (such as energy attenuation in signal propagation).

[0041] In other embodiments, the calculation formula for migration impact may also be: ;in, Representation node For Node The impact of migration, Representation node With node The edge weight between the edges, Representation node The order of influence, Representation node There is a fault This provides another mathematical expression for migration impact. This formula is simple to calculate, has low computational cost, and emphasizes linear distribution logic, making it suitable for describing situations where the impact is evenly distributed across multiple nodes. You can choose the migration impact calculation formula based on your application needs.

[0042] The method for adjusting the failure probability of each node according to the migration impact of the node to obtain the predicted failure probability is: taking the product of the migration impact and the failure probability as the predicted failure probability.

[0043] An embodiment of the present application also discloses a power distribution equipment safe operation detection system, including a processor and a memory, the memory storing computer program instructions, and when the computer program instructions are executed by the processor, the power distribution equipment safe operation detection method according to the present application is implemented.

[0044] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are known in the art and will not be described in detail here.

[0045] In this application, the memory can be, for example, without limitation, any suitable tangible medium that stores, contains, or maintains programming for use by or in connection with the instruction execution system, apparatus, or device. Program code can be stored within a memory. The memory can be implemented within the instruction execution system, apparatus, or device or external to it. By way of example, computer readable media can include volatile and non-volatile, removable and non-removable media implemented in a method or technology such as, but not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD), or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by the application, module, or both.

[0046] The above are only preferred embodiments of the present application, not to limit the protection scope of the present application, therefore: all equivalent changes made on the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A method for detecting safe operation of power distribution equipment, characterized in that: Including steps: Build and train a graph neural network for power distribution equipment. The nodes of the graph neural network represent each device and collect operating data from each distribution device. The edges represent the electrical connections between devices. The input is the operating data and electrical connection relationship of the device, and the output is the failure probability distribution of each device for each fault type. Filter nodes based on the output failure probability distribution to obtain predicted failure nodes, calculate the migration impact of predicted failure nodes on other nodes under each failure type, and adjust the failure probability of each node based on its migration impact to obtain the predicted failure probability; The migration impact is calculated by adaptively setting the node influence order based on the size of the graph neural network. Based on the historical dataset during the graph neural network training process, the similarity between nodes in each fault type is calculated and normalized as the edge weight. The migration impact is calculated based on the edge weight of the edge between nodes, the failure probability of the faulty node, and the influence order. The edge weight and failure probability are directly proportional to the migration impact, while the influence order is inversely proportional to the migration impact. Adaptively setting the node influence order according to the size of the graph neural network, including: taking the product of the ratio of the node order of the node to the maximum node order of the graph neural network and a preset fixed order as the node influence order; The node order refers to the total number of nodes that a node can indirectly affect through connection relationships, including direct neighbor nodes and neighbors of direct neighbor nodes; the maximum node order refers to the maximum value of the order of all nodes in the graph neural network.

2. The method for detecting safe operation of electric power distribution equipment according to claim 1, characterized in that: The similarity is calculated as follows: construct a fault label sequence for each node, count the number of time points where both fault label sequences have faults, and use the ratio of the number of time points to the total number of time points with fault labels as the similarity.

3. The method for detecting safe operation of electric power distribution equipment according to claim 1, characterized in that: The migration impact is calculated as: ;in, Representation node For Node The impact of migration, Representation node With node The edge weight between the edges, Representation node The order of influence, Representation node There is a fault probability of failure; Indicates An exponential function with base .

4. The method for detecting safe operation of electric power distribution equipment according to claim 1, characterized in that: The migration impact is calculated as: ;in, Representation node For Node The impact of migration, Representation node With node The edge weight between the edges, Representation node The order of influence, Representation node There is a fault probability of failure.

5. The method for detecting safe operation of electric power distribution equipment according to claim 1, characterized in that: Nodes are filtered according to the output fault probability distribution to obtain predicted fault nodes, including setting a fault probability threshold for each fault type. When the fault probability corresponding to any fault type in the fault probability distribution is greater than the fault probability threshold, the node is determined to be a predicted fault node.

6. The method for detecting safe operation of electric power distribution equipment according to claim 1, characterized in that: Nodes are filtered according to the output fault probability distribution to obtain predicted fault nodes, and abnormal nodes are identified as predicted fault nodes using an anomaly detection algorithm.

7. The method for detecting safe operation of electric power distribution equipment according to claim 1, characterized in that: The failure probability of each node is adjusted according to the migration impact of the node to obtain the predicted failure probability, and the product of the migration impact and the failure probability is used as the predicted failure probability.

8. The method for detecting safe operation of electric power distribution equipment according to claim 1, characterized in that: The operating data of the equipment includes current, voltage or power.

9. A power distribution equipment safe operation detection system, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the method for detecting safe operation of electric power distribution equipment according to any one of claims 1 to 8 is implemented.

Images