Power system safety self device action scheme generation method, device, equipment and storage medium

Abstract

The invention discloses a power system safety self device action scheme generation method and device, equipment and a storage medium. The method comprises the following steps: constructing an equipment attribute vector based on key attributes of various participating equipment of the power system, and converting the equipment attribute vector into an equipment representation vector through an equipment representation network; based on the equipment representation vector, a fault representation vector is constructed through aggregation, splicing and multi-layer perceptron processing; the method comprises the following steps: constructing an equipment topological graph according to a topological structure of an electric power system equipment network, adding isolated fault nodes, and using an equipment representation vector and a fault representation vector as node attributes; and through a multi-layer message propagation network and a link prediction network, predicting a link relationship between a fault node and an equipment node, and converting the link relationship into a security device action scheme after a fault occurs. According to the method, the hidden association between the fault and the equipment is learned and mined from the historical data, and the security and automation device action scheme is generated, so that the intelligence and rapidness of the security and automation device action scheme of the large-scale power system are realized.

Description

Technical Field

[0001] The present invention relates to the field of electric power systems, and in particular to a method, device, equipment and storage medium for generating an action plan for an anticipated fault safety device of a large-scale electric power system. Background Art

[0002] With the acceleration of the green and low-carbon transformation of electricity, the proportion of new energy installed capacity continues to increase, and the safety regulation strategy of large-scale power systems is also facing new challenges. The introduction of large-scale new energy into the power system, and the integration of more and more flexible loads into the power grid, has led to a sharp increase in the uncertainty factors affecting the power system, which has caused the operation scenarios that need to be evaluated in online or offline scheduling plans to grow exponentially. At the same time, the scale of the expected fault set of large-scale power networks is also very large. Based on detailed time domain simulation, it is difficult to calculate the safety and automatic device action plan for all expected faults in all operation scenarios in a short time, so that the safety risks of system operation cannot be grasped in time. Therefore, under the new situation of the growing scale of new energy installed capacity and the gradual expansion of power generation, how to make the generation of safety and automatic device action plans for large-scale power systems intelligent and rapid has become a problem that needs to be solved urgently.

[0003] The current method for generating action plans for safety and automatic devices used in the industry relies on a simulation system with complex calculations. It has a slow solution speed in large-scale power systems and is difficult to apply online in real time. As the system scale evolves and the number of possible faulty components increases, the number of device states increases exponentially, resulting in long calculation time or even inability to solve. The increasing penetration rate of new energy makes simulation system modeling more complex, further increasing the computational overhead. Although machine learning-based methods have also been explored in recent years, some key problems have not been effectively solved. First, the topological structure of the power system has not been effectively utilized, and the complex dependencies of implicit associations between devices cannot be mined; second, how to unify the heterogeneous attribute information of different types of equipment in the power system has not been considered, and there is also a lack of a bird's-eye view of the association between operating generator equipment, load equipment and other related equipment, resulting in a lack of flexibility and scalability in the model; finally, most of the methods proposed by the academic community have only been tested on small-scale power simulation models and have not been verified in the latest industrial environment, so their reliability needs to be further confirmed.

[0004] Therefore, there is a need for a method for generating action plans for anticipated fault safety devices in large-scale power systems to achieve intelligent and rapid generation of action plans for safety devices in power systems. Summary of the invention

[0005] Purpose of the invention: In order to overcome the shortcomings of the existing machine cutting plan generation algorithm, the present invention provides a method, device, equipment and storage medium for generating an action plan for an electric power system safety device, so as to solve the problem of high computing time overhead in the prior art in large-scale electric power systems.

[0006] Technical solution: To achieve the above-mentioned purpose, the method for generating an action plan of a power system safety device proposed in the present invention comprises the following steps:

[0007] Obtain the key attributes of each participating device in the power system and combine them into a device attribute vector;

[0008] Input the device attribute vector into the device representation network, and obtain the device representation vector with the same length and unified semantics through convolution operation;

[0009] Aggregate all device representation vectors of the same category into a category representation vector to represent the overall characteristics of all devices in the category, concatenate the category representation vectors of all categories, and input them into a multi-layer perceptron to obtain one or more fault representation vectors, which are used to represent the power system status information that needs to be paid attention to after the expected fault occurs;

[0010] According to the equipment association structure in the power system, the corresponding topological graph is constructed, and isolated fault nodes equal to the number of fault representation vectors are added. The equipment representation vector and the fault representation vector are used as node attributes to obtain the node attribute matrix X and the adjacency matrix A of the topological graph.

[0011] Using the node attribute matrix X and adjacency matrix A of the topology graph, we obtain the node embedding vector through a multi-layer message propagation network.

[0012] Based on the node embedding vector, the device nodes belonging to the safety and automation device and the fault nodes are spliced ​​in pairs to obtain a synthetic embedding vector, which is input into the node pair link prediction network to obtain the link relationship prediction result between the nodes;

[0013] According to the prediction results of the link relationship between nodes, the device nodes belonging to the safety and automatic device that have a link relationship with a certain fault node are collected as a node set. The devices included in the set are the devices that need to be operated after the corresponding fault occurs. The above operations are performed on each fault node to obtain the target operation device set for each fault, which is the safety and automatic device action plan required after these faults occur.

[0014] The device representation network, multi-layer perceptron, multi-layer message propagation network, and node-to-link prediction network constitute an end-to-end model of the system. The model is trained using a target equation using a pre-collected data set. The purpose of the training is to minimize the loss calculated by the target equation. Each sample of the data set includes the key attributes of each device involved in the power system, the equipment association structure of the power system, and the safety and automatic device action plan corresponding to each fault.

[0015] The objective equation used in model training is expressed as:

[0016]

[0017] Where l represents the loss function, N f is the number of possible fault types in the power system, N c is the number of equipment in the power system that belongs to the safety device. is the predicted probability that there is a link relationship between the faulty node i and the device j belonging to the safety device, y i,j Indicates whether fault type i in the data set requires operation of device j belonging to the safety device. If operation is required, it is 1, otherwise it is 0.

[0018] The calculation method of the multi-layer message propagation network is:

[0019]

[0020] Where σ(·) represents the activation function, || represents the concatenation operation, represents the aggregation function, N(i) represents the set of neighbor nodes of node i, represents the embedding vector of node i after k rounds of message propagation, W (k) represents the learnable parameters in the kth round of message passing.

[0021] The device representation network consists of a convolutional layer and a fully connected layer. The convolutional layer composed of multiple 1×1 convolution kernels is used to process the attribute vector into vectors of the same length.

[0022] The node attribute matrix X is a shape of (N e +N f )×D two-dimensional matrix, where N e is the number of devices in the power system, N f is the number of possible fault types in the power system, and D is the length of the device representation vector.

[0023] The adjacency matrix A is a (N e +N f )×(N e +N f), each row represents the relationship between a device or faulty node and other devices or faulty nodes, and the values ​​of the matrix elements are 0 or 1. When the i-th node and the j-th node have a link relationship, the adjacency matrix A in A ij is equal to 1, otherwise it is equal to 0.

[0024] The present invention provides a device for generating an action plan for a power system safety device, comprising:

[0025] The device attribute vector construction module obtains the key attributes of each device involved in the power system and combines them into an attribute vector;

[0026] The device representation vector construction module inputs the device attribute vector into the device representation network and obtains the device representation vector with the same length and unified semantics through convolution operation;

[0027] The fault representation vector construction module aggregates all the device representation vectors of the same category into a category representation vector, which represents the overall characteristics of all devices in the category. The category representation vectors of all categories are concatenated and input into the multi-layer perceptron to obtain one or more fault representation vectors, which are used to represent the power system status information that needs to be paid attention to after the expected fault occurs.

[0028] The equipment topology construction module constructs the corresponding topology according to the equipment association structure in the power system, and adds isolated fault nodes equal to the number of fault representation vectors. The equipment representation vector and the fault representation vector are used as node attributes to obtain the node attribute matrix X and the adjacency matrix A of the topology.

[0029] The node embedding vector construction module uses the node attribute matrix X and the adjacency matrix A of the topological graph to obtain the node embedding vector through a multi-layer message propagation network;

[0030] The link prediction module, based on the node embedding vector, concatenates the device nodes belonging to the safety and automation device and the fault nodes in pairs to obtain a synthetic embedding vector, which is input into the node pair link prediction network to obtain the link relationship prediction result between the nodes;

[0031] The safety and automatic device action plan generation module collects the device nodes belonging to the safety and automatic device that have a link relationship with a certain fault node into a node set based on the prediction results of the link relationship between nodes. The devices included in the set are the devices that need to be operated after the corresponding fault occurs. The above operations are performed on each fault node to obtain the target operation device set for each fault, which is the safety and automatic device action plan required after these faults occur.

[0032] A computer device provided by the present invention comprises: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the method for generating an action plan for an electric power system safety device are implemented as described above.

[0033] The present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method for generating an action plan for an electric power system safety device as described above are implemented.

[0034] Beneficial effects:

[0035] (1) This invention proposes for the first time a method for converting the problem of generating action plans for safety and self-driving devices in power systems into a problem of link prediction in graph neural networks, and designs an end-to-end method for generating action plans for safety and self-driving devices based on graph neural networks. The equipment association structure of the power system has an important impact on the operation and analysis of the power system. In the model, the equipment association structure of the power system is converted into a topological graph, which fully mines and captures the implicit complex dependencies between different devices, solving the problem that the previous learning-based algorithms ignored the link relationship between nodes.

[0036] (2) In the process of constructing the equipment representation vector, the convolution kernel with shared weights is used to semantically unify the equipment attribute vectors of different types and lengths, and convert the originally heterogeneous node information reflecting the current status, operating conditions and performance parameters of various types of equipment into equipment representation vectors with the same length and expression. This solves the difficulties in data information understanding and data alignment caused by the heterogeneity of power system equipment data, and unifies the prediction methods of equipment belonging to safety devices such as operating generator sets, load equipment, and parallel capacitors / reactors, thereby improving the flexibility of the model.

[0037] (3) The present invention connects a series of processes such as device representation vector construction, fault representation vector construction, and node embedding vector construction in series, directly predicts the link relationship between two nodes end-to-end, reduces the data, computing power, and time overhead of the entire network training process, and improves the efficiency of mining the precious information contained in the massive historical power system status data, making the generation of safety and self-driving device action plans faster and more accurate. After sufficient training, this method can be deployed on existing consumer-grade central processing units and graphics cards to have a millisecond-level safety and self-driving device action plan generation speed, and efficiently generate safety control measures for large-scale power systems, solving the problem of high computational overhead of traditional methods in large-scale power systems under the background of increasing installed capacity of high-energy and new energy. The method of the present invention was tested in real, specific scenarios in the industrial world, such as the data of the Northwest China Power System. The accuracy of the algorithm prediction meets the use requirements, and good results are achieved in the algorithm risk analysis and evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flow chart of a method for generating an action plan of a large-scale power system with high renewable energy penetration rate according to the present invention;

[0039] Figure 2 is a flow chart of a vector generation phase of the device of the present invention;

[0040] Figure 3 is a flow chart of the fault representation vector generation phase of the present invention;

[0041] Figure 4 It is a diagram of the message propagation calculation method of a single node in the multi-layer propagation network of the present invention. DETAILED DESCRIPTION

[0042] The present invention will be further described below in conjunction with the accompanying drawings.

[0043] The present invention proposes a method for generating an action plan for a large-scale power system anticipated fault safety device. Figure 1 As shown in the figure, this method generates an action plan for the safety and automatic device based on the key attributes of the equipment participating in the power system and the topological structure of the power system after a series of calculations and processing. In terms of data flow, this method first constructs the key attributes of the equipment participating in the power system into equipment attribute vectors, and then converts the equipment attribute vectors into equipment representation vectors with the same length and unified semantics through the equipment representation network. Then, the equipment representation vectors are aggregated and spliced ​​according to the equipment category, and input into the multi-layer perceptron to obtain the fault representation vector. Then, the equipment topology map is constructed based on the equipment representation vector, the fault representation vector and the topological structure of the power system. Finally, a multi-layer message propagation network and a link prediction network are used to predict the link relationship between the fault node and the equipment node belonging to the safety and automatic device, and the action plan for the safety and automatic device is generated.

[0044] According to the logic of data flow, this method can be broken down into three stages: device representation vector generation stage, fault representation generation stage and graph neural link prediction stage. The function of the device representation vector generation stage is to convert the key attributes of the equipment in the power system into device attribute vectors, and then input them into the device representation network to obtain the device representation vector. The function of the fault representation vector generation stage is to aggregate and splice the device representation vectors and input them into the multi-layer perceptron to obtain the fault representation vector. The function of the graph neural link prediction stage is to use the power system equipment association structure to build the equipment topology map, the node attributes are the device representation vector and the fault representation vector, and then generate the safety and self-driving device action plan based on the output results of the multi-layer message propagation network and the link prediction network. All three stages contain learnable parameters, which are initialized with random data, and the appropriate parameters are learned in the training step. After the training is completed, the parameters will be fixed, and the learned parameters will be used to predict the safety and self-driving device action plan in subsequent use.

[0045] Specifically, the process of the device representation vector generation phase of the method is as follows Figure 2 As shown, the following steps are included:

[0046] (11) Device attribute vector construction steps.

[0047] The purpose of this step is to obtain a simple vector description of the equipment in the power system. These vectors contain the operating status information of the equipment in the power system and are closely related to the operation of the entire system. Large-scale power systems containing new energy sources contain many types of equipment, including but not limited to generator sets, loads, buses, AC lines, transformers, DC systems, series compensators, and shunt capacitors / reactors. Each device has different attribute values. Although there are many types of attribute values ​​of equipment, only some of them are closely related to the operation of the power system, which are called key attributes. The device attribute vector constructed in this step is only composed of these key attributes, because redundant attributes will interfere with the performance of the deep learning model. To this end, it is necessary to select the corresponding key attributes from various devices and combine them into a device attribute vector. The way to combine into a vector is to arrange the key attributes in each device into a vector in a fixed order. The key attributes of different types of equipment are different. For example, the key attributes of generator sets include but are not limited to active power, reactive power, and voltage value, while the key attributes of series compensators include but are not limited to the first-end active value, the first-end reactive value, the terminal active value, and the terminal reactive value.

[0048] (12) Device representation vector construction step.

[0049] The purpose of this step is to convert the naive device attribute vectors with different lengths and different semantic representations into comprehensive device representation vectors with the same length and unified semantics. The device attribute vector simply arranges the key attributes and does not have comprehensive expression capabilities. At the same time, due to the variety of device categories, the number of their key attributes and the physical quantities they represent are also inconsistent, resulting in uneven lengths of device attribute vectors and different semantics of vector numerical representations at the same position. In this step, a device representation network is used to convert the device attribute vector into a device representation vector. The device representation network consists of a convolutional layer composed of multiple 1×1 convolutional kernels and multiple fully connected layers. In one embodiment of the present method, the convolutional layer of the device representation network contains 128 1×1 convolutional kernels, and the fully connected layer has 3 layers. The input and output dimensions of each fully connected layer are 128, and a rectified linear unit is connected to the tail as an activation function. The calculation formula of the convolutional layer is Where x is the input vector, y is the output vector, W and b are learnable parameters, and C r is the number of convolution kernels, C l is the length of the input vector, and ★ is the cross-correlation operator. The calculation formula of the fully connected layer is y = σ (xW T +b), where x is the input vector, y is the output vector, W and b are learnable parameters, and σ(·) is the activation function.

[0050] The process of the fault representation vector generation phase of the method is as follows Figure 3 As shown, the following steps are included:

[0051] (21) Device representation vector aggregation step.

[0052] The purpose of this step is to obtain the aggregate representation vector of each category of equipment. Although the device representation vectors are semantically unified, the device representation vectors of the same category of equipment are more closely related because their original physical quantities have the same meaning. In order to utilize the category relationship of the device representation vectors, fuse and extract their associated information, and reduce the dimension of the vectors, this step performs an aggregation operation on the device representation vectors. The specific aggregation operation can include but is not limited to additive aggregation. Average Aggregation Maximum aggregation max n x i , where n is the number of vectors, x i is the i-th vector, and the length of the vector before and after the aggregation operation remains the same. In one embodiment of the method, the aggregation operation uses additive aggregation.

[0053] (22) The device represents a vector concatenation operation.

[0054] The purpose of this step is to convert the representation vectors of multiple categories of devices into a single vector. The specific method of the splicing operation is to connect the multiple vectors to be spliced ​​in order, so that even if the spliced ​​vectors come from different device categories, the single vector after splicing will not lose information.

[0055] (23) Fault representation vector generation step.

[0056] The purpose of this step is to convert the spliced ​​vectors into several fault representation vectors, and the length of the fault representation vector is consistent with the length of the device representation vector. The fault representation vector is used to represent the power system status information that needs to be paid attention to after the expected fault occurs. Therefore, it can be calculated from the power grid status information. The state of the power system is determined by the state of the equipment. Therefore, after all the device representation vectors are aggregated and spliced, they contain all the information required for calculation, so the fault representation vector can be calculated from them. It should be noted that after the device representation vectors are aggregated and spliced, they can be converted into fault representation vectors after being input into the multi-layer perceptron "calculation". It is not directly aggregated and spliced ​​to become a fault representation vector. The number of fault representation vectors is the same as the number of faults in the expected fault set of the power system that needs to be focused. For example, if the expected fault set is a DC bipolar blocking fault of 20 lines, then 20 fault representation vectors are generated in this step, and each vector corresponds to these faults one by one. Multilayer perceptron is a basic structure in deep learning. It has a simple structure, provides nonlinear mapping and learnable parameters, maps a set of input vectors to a set of output vectors, and each node in it is a neuron with a nonlinear activation function. It is suitable here to nonlinearly map the "power system status" represented by the device representation vector to "power system status information that needs attention after this kind of fault occurs". The way to adjust the number of fault representation vectors generated is to adjust the output channel size of the multilayer perceptron, and the number of output channels is the number of fault representation vectors generated. In one embodiment of the present method, the multilayer perceptron has 3 layers, each with dimensions of 512, 256 and 128, and a rectified linear unit is used as the activation function of the multilayer perceptron.

[0057] The process of the graph neural link prediction phase of the method specifically includes the following steps:

[0058] (31) Steps for constructing device topology map.

[0059] The purpose of this step is to create a topological graph structure and organize the device representation vector and the fault representation vector in the format of the graph data. The link relationship of the device nodes in the graph data is mapped to the corresponding topological graph according to the device association structure in the power system, and then a number of isolated fault nodes equal to the number of fault representation vectors are added to the topological graph. The device association structure refers to the association relationship between different devices, and the device topological structure, device type relationship, device subnet link structure, etc. in the power system can be used, or a customized device complete graph structure, device group link structure, etc. can be used. In one embodiment of the present method, the device association structure uses the type relationship of the device, that is, according to the type of the device, the devices belonging to the same type are linked to each other, in order to improve the efficiency of information exchange within the type. In this way, a topological graph consisting of device nodes and fault nodes is obtained.

[0060] The specific mathematical form of the topological graph is composed of the node attribute matrix X and the adjacency matrix A. The node attribute matrix X is a shape of (N e +N f )×D two-dimensional matrix, where N e is the number of devices in the power system, N f is the number of possible fault types in the power system. Each column of the matrix is ​​the corresponding device representation vector or fault representation vector. Therefore, D is the length of the device representation vector or fault representation vector. According to the above steps, the length of the device representation vector and the fault representation vector are equal. The adjacency matrix A is a (N e +N f )×(N e +N f ), each row represents the relationship between a device or fault node and other devices or fault nodes. The currently constructed topology graph only has links between devices and device nodes, indicating the topological association of the corresponding devices in the power system. The values ​​of the matrix elements are 0 or 1. The construction method is that when the i-th node has a link relationship with the j-th node, let A in the adjacency matrix A ij is equal to 1, otherwise it is equal to 0.

[0061] (32) Node embedding vector construction step.

[0062] The purpose of this step is to fully explore the complex implicit associations between nodes based on the device topology map constructed in the previous step, so as to create a more expressive node embedding vector. This step uses a multi-layer message propagation network to utilize the structural characteristics of the device topology map. Its input is the node attribute matrix X and the adjacency matrix A. The multi-layer message propagation network needs to carry out multiple message propagation processes, and each message propagation requires the participation of all nodes. The message propagation process of a single node is as follows: Figure 4As shown in the figure, node 0 needs to do a k-th layer message propagation, and its neighbor nodes include node 1, node 2, node 3, and node 4. First, get the k-1 layer message propagation results of the neighbor nodes Then use an aggregation function to aggregate them. In one embodiment of the present method, an addition aggregation function is used. Then the aggregated result is combined with the k-1 layer message propagation result of node 0. Do the concatenation operation and finally input it into the fully connected layer to get the k-layer message propagation result of node 0 The fully connected layer contains learnable parameters. The starting point of the node iteration is the message propagation result of the 0th layer. It is the attribute vector of the node, that is, the column vector corresponding to the node in the node attribute matrix X.

[0063] The calculation formula of the multi-layer message propagation network is: Where σ(·) represents the activation function, || represents the concatenation operation, represents the aggregation function, N(i) represents the set of neighbor nodes of node i, represents the embedding vector of node i after k rounds of message propagation, W (k) represents the learnable parameters in the kth round of message passing, which will be adjusted during the model training step.

[0064] (33) Link prediction step.

[0065] The purpose of this step is to predict the link relationship between the device node and the faulty node. When constructing the topology graph, the faulty node is added to the graph as an isolated node. Therefore, in the initial state, the faulty node has no link relationship with all devices. Therefore, a link prediction network is used here to predict the link relationship between the faulty node and the device node belonging to the safety and self-service device. The link prediction network consists of a fully connected layer and an exponential normalization layer with a temperature hyperparameter. The input is obtained by concatenating the node embedding vectors of a pair of nodes. Here, a pair of nodes refers to the faulty node and the device node belonging to the safety and self-service device. The fully connected layer contains learnable parameters and outputs a vector of length 2, which is provided to the exponential normalization layer with a temperature hyperparameter. The calculation method of the exponential normalization layer with a temperature hyperparameter is as follows: where z i is each element of the input vector, T is the temperature hyperparameter, and p i is the probability that there is a link relationship between the node pairs. If the probability is greater than 0.5, then it is predicted that there is a link relationship between the node pairs, otherwise it is predicted that there is no link relationship.

[0066] (34) Steps for generating an automatic device action plan.

[0067] The purpose of this step is to generate an action plan for the safety and automation device based on the link relationship between the faulty node and the device nodes belonging to the safety and automation device. For any faulty node, the device nodes belonging to the safety and automation device that have a link relationship with it are collected as a node set, and then the corresponding device that needs to be operated after the fault occurs is the device set corresponding to the node set.

[0068] The three stages of the present invention all contain learnable parameters, which need to be trained before they can be used. After the training is completed, the corresponding learnable parameters are fixed, and the subsequent self-device action plan can be used to generate tasks. This method proposes an end-to-end model, which means that the device representation network, multi-layer perceptron, multi-layer message propagation network, and node-to-link prediction network involved in the three stages are trained simultaneously and do not need to be trained separately, which is beneficial to the robustness of the model and the economy of training. In the initial state, the learnable parameters of the three stages are initialized with random numbers and trained using the gradient descent method. A round of training process includes two stages: forward propagation and back propagation, where the loss function used in forward propagation is Where N f is the number of possible fault types in the power system, N c is the number of equipment in the power system that belongs to the safety device. is the predicted probability that there is a link relationship between the faulty node i and the device node j belonging to the safety device, y i,j Indicates whether fault type i in the data set requires operation of device j belonging to the safety device. If operation is required, it is 1, otherwise it is 0. Back propagation uses the gradient obtained in the forward propagation calculation process to adjust the learnable parameters. In one embodiment of the present invention, the number of training rounds is 1000, the learning rate is 0.0001, and the Adam optimizer is used.

[0069] The model training step uses the pre-collected data set to perform end-to-end training using the target equation. The purpose of the training is to minimize the loss of the target equation calculation. Each sample in the data set includes the key attributes of each device involved in the power system, the device association structure of the power system, and the safety and automatic device action plan corresponding to each fault. This step only needs to be executed once. When the safety and automatic device action plan for the target fault needs to be generated in the future, only steps (11) to (34) need to be executed. The safety and automatic device action plan obtained in step (34) is the safety and automatic device action plan finally generated by this method.

[0070] The present invention also provides a device for generating an action plan of an electric power system safety device, comprising:

[0071] The device attribute vector construction module obtains the key attributes of each device involved in the power system and combines them into an attribute vector;

[0072] The device representation vector construction module inputs the device attribute vector into the device representation network and obtains the device representation vector with the same length and unified semantics through convolution operation;

[0073] The fault representation vector construction module aggregates all device representation vectors of the same category into a category representation vector, which represents the overall characteristics of all devices in the category. The category representation vectors of all categories are concatenated and input into the multi-layer perceptron to obtain one or more fault representation vectors.

[0074] The equipment topology construction module constructs the corresponding topology according to the equipment association structure in the power system, and adds isolated fault nodes equal to the number of fault representation vectors. The equipment representation vector and the fault representation vector are used as node attributes to obtain the node attribute matrix X and the adjacency matrix A of the topology.

[0075] The node embedding vector construction module uses the node attribute matrix X and the adjacency matrix A of the topological graph to obtain the node embedding vector through a multi-layer message propagation network;

[0076] The link prediction module, based on the node embedding vector, concatenates the device nodes belonging to the safety and automation device and the fault nodes in pairs to obtain a synthetic embedding vector, which is input into the node pair link prediction network to obtain the link relationship prediction result between the nodes;

[0077] The safety and automatic device action plan generation module collects the device nodes belonging to the safety and automatic device that have a link relationship with a certain fault node into a node set based on the prediction results of the link relationship between nodes. The devices included in the set are the devices that need to be operated after the corresponding fault occurs. The above operations are performed on each fault node to obtain the target operation device set for each fault, which is the safety and automatic device action plan required after these faults occur.

[0078] It should be understood that the power system safety device action plan generation device in the embodiment of the present invention can implement all the technical solutions in the above-mentioned method embodiment, and the functions of its various functional modules can be specifically implemented according to the methods in the above-mentioned method embodiments. The specific implementation process can refer to the relevant description in the above-mentioned embodiments, which will not be repeated here.

[0079] The present invention also provides a computer device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the method for generating an action plan for an electric power system safety device are implemented as described above.

[0080] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method for generating an action plan for an electric power system safety device as described above are implemented.

[0081] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, devices (systems), computer equipment or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0082] The present invention is described with reference to a flowchart of a method according to an embodiment of the present invention. It should be understood that each process in the flowchart and a combination of processes in the flowchart can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart. Figure 1 A device that specifies functions in a process or multiple processes.

[0083] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A function specified in a process or multiple processes.

[0084] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 The steps of a specified function in a process or multiple processes.

[0085] The preferred embodiments of the present invention are described in detail above; however, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A method for generating an action plan for a power system safety device, characterized in that: The method comprises the following steps: Obtain the key attributes of each participating device in the power system and combine them into a device attribute vector; Input the device attribute vector into the device representation network, and obtain the device representation vector with the same length and unified semantics through convolution operation; Aggregate all device representation vectors of the same category into a category representation vector to represent the overall characteristics of all devices in the category, concatenate the category representation vectors of all categories, and input them into a multi-layer perceptron to obtain one or more fault representation vectors, which are used to represent the power system status information that needs to be paid attention to after the expected fault occurs; According to the equipment association structure in the power system, the corresponding topological graph is constructed, and isolated fault nodes equal to the number of fault representation vectors are added. The equipment representation vector and the fault representation vector are used as node attributes to obtain the node attribute matrix X and the adjacency matrix A of the topological graph. Using the node attribute matrix X and adjacency matrix A of the topology graph, we obtain the node embedding vector through a multi-layer message propagation network. Based on the node embedding vector, the device nodes belonging to the safety and automation device and the fault nodes are spliced ​​in pairs to obtain a synthetic embedding vector, which is input into the node pair link prediction network to obtain the link relationship prediction result between the nodes; According to the prediction results of the link relationship between nodes, the device nodes belonging to the safety and automatic device that have a link relationship with a certain fault node are collected as a node set. The devices included in the set are the devices that need to be operated after the corresponding fault occurs. The above operations are performed on each fault node to obtain the target operation device set for each fault, which is the safety and automatic device action plan required after these faults occur.

2. The method according to claim 1, characterized in that The device representation network, multi-layer perceptron, multi-layer message propagation network, and node-to-link prediction network constitute an end-to-end model of the system. The model is trained using a target equation using a pre-collected data set. The purpose of the training is to minimize the loss calculated by the target equation. Each sample of the data set includes the key attributes of each device involved in the power system, the equipment association structure of the power system, and the safety and automatic device action plan corresponding to each fault.

3. The method according to claim 2, characterized in that The objective equation used in model training is expressed as: Where l represents the loss function, N f is the number of possible fault types in the power system, N c is the number of equipment in the power system that belongs to the safety device. is the predicted probability that there is a link relationship between the faulty node i and the device j belonging to the safety device, y i,j Indicates whether fault type i in the data set requires operation of device j belonging to the safety device. If operation is required, it is 1, otherwise it is 0.

4. The method according to claim 1, characterized in that: The calculation method of the multi-layer message propagation network is: Where σ(·) represents the activation function, || represents the concatenation operation, represents the aggregation function, N(i) represents the set of neighbor nodes of node i, represents the embedding vector of node i after k rounds of message propagation, W (k) represents the learnable parameters in the kth round of message passing.

5. The method according to claim 1, characterized in that The device representation network consists of a convolutional layer and a fully connected layer. The convolutional layer composed of multiple 1×1 convolution kernels is used to process the attribute vector into vectors of the same length.

6. The method according to claim 1, characterized in that The node attribute matrix X is a shape of (N e +N f )×D two-dimensional matrix, where N e is the number of devices in the power system, N f is the number of possible fault types in the power system, and D is the length of the device representation vector.

7. The method according to claim 1, characterized in that The adjacency matrix A is a (N e +N f )×(N e +N f ), each row represents the relationship between a device or faulty node and other devices or faulty nodes, and the values ​​of the matrix elements are 0 or 1. When the i-th node and the j-th node have a link relationship, the adjacency matrix A in A ij is equal to 1, otherwise it is equal to 0, where N e is the number of devices in the power system, N f It is the number of possible fault types that may occur in the power system.

8. A device for generating an action plan for an electric power system safety device, characterized in that: include: The device attribute vector construction module obtains the key attributes of each device involved in the power system and combines them into an attribute vector; The device representation vector construction module inputs the device attribute vector into the device representation network and obtains the device representation vector with the same length and unified semantics through convolution operation; The fault representation vector construction module aggregates all the device representation vectors of the same category into a category representation vector, which represents the overall characteristics of all devices in the category. The category representation vectors of all categories are concatenated and input into the multi-layer perceptron to obtain one or more fault representation vectors, which are used to represent the power system status information that needs to be paid attention to after the expected fault occurs. The equipment topology construction module constructs the corresponding topology according to the equipment association structure in the power system, and adds isolated fault nodes equal to the number of fault representation vectors. The equipment representation vector and the fault representation vector are used as node attributes to obtain the node attribute matrix X and the adjacency matrix A of the topology. The node embedding vector construction module uses the node attribute matrix X and the adjacency matrix A of the topological graph to obtain the node embedding vector through a multi-layer message propagation network; The link prediction module, based on the node embedding vector, concatenates the device nodes belonging to the safety and automation device and the fault nodes in pairs to obtain a synthetic embedding vector, which is input into the node pair link prediction network to obtain the link relationship prediction result between the nodes; The safety and automatic device action plan generation module collects the device nodes belonging to the safety and automatic device that have a link relationship with a certain fault node into a node set based on the prediction results of the link relationship between nodes. The devices included in the set are the devices that need to be operated after the corresponding fault occurs. The above operations are performed on each fault node to obtain the target operation device set for each fault, which is the safety and automatic device action plan required after these faults occur.

9. A computer device, characterized in that: include: one or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the method for generating an action plan for an electric power system safety device as described in any one of claims 1-7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for generating an action plan for an electric power system safety device according to any one of claims 1 to 7 are implemented.

Images