An intelligent AC / DC fault positioning method and system based on multi-scale feature adaptive fusion and a medium
By combining time-series data convolution, gated cyclic units, and graph convolution operations, a multi-scale feature adaptive fusion method was developed to solve the problem of identifying and locating multi-region and multi-type faults in new energy access flexible DC systems, thereby improving the accuracy of fault location and the stability of the power grid.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- STATE GRID HUBEI ELECTRIC POWER RES INST
- Filing Date
- 2026-02-09
- Publication Date
- 2026-06-02
Smart Images
Figure CN122133060A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of smart grid technology, specifically a method, system and medium for intelligent AC / DC fault location based on multi-scale feature adaptive fusion. Background Technology
[0002] New energy sources (such as wind farms and photovoltaic power stations) are connected to the grid via power electronic converters in flexible direct current transmission systems (VSC-HVDC), which is fundamentally different from traditional synchronous generators. This difference becomes particularly apparent when system faults occur, mainly manifested in limited fault current amplitude, phase control, waveform distortion, reduced system inertia, and more complex transient processes. These altered fault characteristics mean that protection principles based on traditional synchronous power source assumptions face the risk of decreased adaptability when applied to flexible direct current systems with new energy connections, potentially leading to undetermined fault locations, protection failure to operate (failing to operate when it should), or erroneous operation (operating when it shouldn't).
[0003] Considering transient stability during faults, to ensure that VSC-HVDC and renewable energy sources do not disconnect from the grid during faults and other transient processes, the grid dispatching and operation departments must limit the access capacity and output limit of renewable energy sources to reserve sufficient stability margin for the system. This directly reduces the system's theoretical maximum absorption capacity. Secondly, in complex fault scenarios, if the protection device misjudges or delays action due to the characteristics of renewable energy faults, the fault may not be quickly isolated. Once this happens, it not only causes a huge instantaneous loss of electricity, but also requires a long time for system recovery, seriously affecting the daily absorption capacity.
[0004] In summary, the large-scale integration of new energy sources (photovoltaics and wind power) has altered the fault characteristics of near-field VSC-HVDC systems. Furthermore, the structure and control / protection strategies of VSC-HVDC systems are more complex than traditional AC systems. These changes present challenges to traditional VSC-HVDC fault identification, location, and protection applications, including decreased sensitivity and increased risks of false tripping or failure to trip. This necessitates the application of more intelligent, faster, and more adaptable fault identification or location technologies to assist protection systems in fault isolation.
[0005] Current fault location technologies mostly focus on faults in a single area (such as DC transmission lines), and cannot locate multiple types of faults in multiple areas, including AC systems, AC connecting lines, bridge arm reactors, DC lines, and converter valves. Fault location technology that covers multiple areas and multiple fault types in both AC and DC systems remains a major challenge. Summary of the Invention
[0006] To address the aforementioned issues, this invention presents an intelligent AC / DC fault location method, system, and medium based on multi-scale feature adaptive fusion, which enables intelligent fault location in multi-region AC / DC transmission systems based on a limited number of fault samples.
[0007] The technical solution adopted in this invention is as follows:
[0008] A smart AC / DC fault location method based on multi-scale feature adaptive fusion is proposed. The method employs a fault location model, which includes a time-series data convolution module, a gated recurrent unit, a multi-scale feature adaptive fusion module, and a graph convolution operation module. The method comprises the following steps:
[0009] S1. Collect fault datasets X and test set X from various regions of the DC and AC transmission systems under renewable energy access. c ;
[0010] S2. Input the fault dataset X into the time series data convolution module CNN to obtain convolutional features. ;
[0011] S3. Input the fault dataset X into the gated recurrent unit (GRU) to obtain long-short-term features. ;
[0012] S4. Based on convolutional features and long-term and short-term characteristics The gating weight α of the multi-scale feature adaptive fusion module is obtained, and the multi-scale feature adaptive fusion module is based on... , And the fused features are obtained from the gating weight α. ;
[0013] S5, the graph convolution operation module will After graph convolution operation, the fault location prediction value is output using an activation function;
[0014] S6. Calculate the label smooth cross-entropy loss function and update the parameters of the fault location model through backpropagation;
[0015] S7, using test set X c As input to the fault location model, repeat steps S2-S5 and calculate the accuracy of the fault location prediction. The accuracy is the ratio of the total number of correctly predicted samples to the total number of samples in the test set. If the requirements are not met, repeat steps S2-S6. If the requirements are met, the training ends and a well-trained fault location model is obtained.
[0016] S8. Select fault data and input the fault location model trained in S7 to locate the fault.
[0017] Furthermore, step S1 specifically includes:
[0018] Referring to the topology of AC / DC transmission systems, lines or equipment areas with different potentials and currents are treated as nodes, and the connection relationships of each node Gn are determined according to the electrical structure of the system to form a graph structure. Each node in the graph structure is used as a fault area and electrical measurement point, and various types of faults are set in each area of the entire AC / DC transmission system.
[0019] Electrical data is extracted from fault waveforms at each electrical measurement point and divided into datasets. and test set , where t, n, These represent the indices for time, node Gn, and electrical quantity type, respectively. Each fault sample data is labeled to form a true label y according to the actual fault location.
[0020] Furthermore, step S2 specifically includes:
[0021] F1 is obtained by passing X through the first one-dimensional convolution layer:
[0022] ;
[0023] in" "It consists of convolutional layers, BN normalization layers, and mappings learned by activation functions;"
[0024] f1 is processed by the residual convolution block ResidualConvBlock to obtain f2:
[0025] ;
[0026] Where GELU is the activation function;
[0027] f2 is processed by a max pooling layer to obtain the time series data convolution module output f. cnn :
[0028] .
[0029] Furthermore, step S3 specifically includes:
[0030] Data from X at each time t The input gated loop unit is processed by the following algorithm:
[0031] ;
[0032] ;
[0033] ;
[0034] ;
[0035] in" "" indicates the Hadamard product, a matrix calculation method. This represents the output of the GRU at time t-1. This represents the data in X at time t; It is the sigmoid activation function; , , , , , These are the weight matrix parameters for the update gate, reset gate, and candidate state, respectively. , , Here are the corresponding bias parameters, and tanh is the activation function;
[0036] The above calculations yield the GRU output feature "f". gru = ".
[0037] Furthermore, step S4 specifically includes:
[0038] It is obtained by directly concatenating the outputs of CNN and GRU in S2 and S3. ;
[0039] The gating weight α is obtained based on the overall output features of CNN and GRU, and the gating weight α is expressed as:
[0040] ;
[0041] By dynamically adjusting the contribution ratio of CNN and GRU features using gating weight α, fused features are obtained. :
[0042] .
[0043] Furthermore, step S5 specifically includes:
[0044] Based on the connection relationships between nodes in the graph structure, the adjacency matrix A is obtained, and its normalization formula is... for:
[0045]
[0046] Where D is the degree matrix of A, and I is the identity matrix;
[0047] The fault location prediction value Y is output by the graph convolution operation module:
[0048]
[0049] in and Here are the weight matrix and bias parameters, and GELU is the activation function;
[0050] The graph structure is a data structure composed of nodes and edges, representing the relationships between objects. Lines or equipment areas with different potentials and currents are taken as several nodes, and the connection relationship of each node Gn is determined according to the electrical structure of the system to form a graph structure.
[0051] Furthermore, step S6 specifically includes:
[0052] Based on the prediction results of the S5 fault location model, the label smooth cross-entropy loss function L is obtained:
[0053] ;
[0054] Where C is the total number of fault area categories, k represents a specific category, q(k) is the true distribution of the constructed fault location areas, and p(k) is the probability distribution of each type of fault area output by the fault location model; the cross-entropy loss function optimizes the fault location model by minimizing the difference between the predicted distribution and the true distribution.
[0055] An intelligent AC / DC fault location system based on multi-scale feature adaptive fusion includes: a computer-readable storage medium and a processor;
[0056] The computer-readable storage medium is used to store executable instructions;
[0057] The processor is used to read executable instructions stored in the computer-readable storage medium and execute the AC / DC fault intelligent location method based on multi-scale feature adaptive fusion.
[0058] A non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the aforementioned intelligent AC / DC fault location method based on multi-scale feature adaptive fusion.
[0059] The fault location model designed in this invention integrates basic convolutional features, temporal features, and spatial features of each node in the data. Through feature weight learning and gating mechanisms, it organically combines multiple algorithms to achieve fault location in multiple regions of AC / DC systems under various fault types. With a limited number of fault samples, it can significantly improve the accuracy of fault location in multi-region AC / DC transmission systems. Attached Figure Description
[0060] Figure 1 This is a flowchart illustrating an intelligent AC / DC fault location method based on multi-scale feature adaptive fusion provided by an embodiment of the present invention.
[0061] Figure 2 This is a topology diagram of a flexible DC transmission system for accessing new energy sources according to an embodiment of the present invention;
[0062] Figure 3 This is a structural diagram of the gated loop unit according to an embodiment of the present invention;
[0063] Figure 4 This is a structural diagram of the feature adaptive fusion module according to an embodiment of the present invention;
[0064] Figure 5 This is a schematic diagram of the graph structure based on the power system topology and fault characteristics according to an embodiment of the present invention;
[0065] Figure 6 This describes the changes in the loss function and accuracy of the fault location model during the training process in this embodiment of the invention. Detailed Implementation
[0066] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0067] Please see Figure 1 The first aspect of this invention provides an intelligent AC / DC fault location method based on multi-scale feature adaptive fusion, comprising the following steps:
[0068] S1. Taking an end-to-end flexible DC transmission system connected to new energy (photovoltaic power plants) as an example, collect fault data under various operating conditions, perform preliminary data processing, that is, extract data from simulation waveforms, and then combine the electrical quantity vector data of each measurement point into a dataset. Preferably, step S1 specifically includes the following steps:
[0069] exist Figure 2 Points G1-G14 are configured to simulate single-phase ground faults on the converter transformer side, single-phase ground faults on the AC connection bus, DC line short-circuit and ground faults, converter valve short-circuit faults, and converter valve submodule short-circuit faults, respectively. Based on these fault settings, the fault areas are divided into 14 zones (G1-G14). Fault parameters, such as grounding impedance and fault duration, are adjusted to perform multiple fault simulations. Each fault sample data point is labeled to form a real-world label y, based on the actual fault occurrence area.
[0070] Will Figure 2 Fourteen nodes, G1-G14, are used as electrical measurement points, based on... Figure 2The AC / DC transmission system topology is constructed as a graph structure consisting of nodes G1-G14, such as... Figure 5 As shown. The measurement points of the fourteen nodes G1-G14 are as follows: three-phase AC system bus (nodes G1, G2, G3), three-phase AC connection bus (nodes G4, G5, G6), upper arm of the ABC three-phase bridge (nodes G7, G8, G9), lower arm of the ABC three-phase bridge (nodes G10, G11, G12), positive DC bus (node G13), and negative DC bus (node G14). Figure 5 The lines connecting nodes in a graph represent the edges of the graph structure.
[0071] Electrical data is extracted from fault simulation waveforms and divided into datasets. and test set Where t, n, These represent the indexes for time, node (G1-G14), and electrical quantity type (voltage or current), respectively.
[0072] S2, Input X into the time series data convolution module to obtain... Preferably, step S2 specifically includes the following steps:
[0073] F1 is obtained by passing X through a one-dimensional convolutional layer:
[0074]
[0075] in" "It consists of mappings learned from one-dimensional convolution, Batch Normalization (BN) layers, activation functions, etc. The one-dimensional convolution operation involves sliding the convolution kernel across the input sequence and calculating the dot product between the kernel and the corresponding part of the input sequence at each position. That is, a linear combination of sliding windows is applied to the input data, with each window corresponding to a position of the convolution kernel. An element in the output sequence is obtained by calculating the sum of the products of the data within the window and the elements of the convolution kernel."
[0076] f1 is processed by a residual convolution block to obtain f2:
[0077]
[0078] GELU (Gaussian Error Linear Unit) is the activation function.
[0079] f2 is processed by a max pooling layer to obtain the time series data convolution module output f. cnn :
[0080]
[0081] S3, obtain the X input gated loop unit Preferably, step S3 specifically includes the following steps:
[0082] The gated recurrent unit consists of the recurrent neural unit from the initial time, t-1 to time t. , , , The operation process consists of "".
[0083] Specifically, update the gate output. Reset door output Candidate hidden state The output of the gated recurrent unit (GRU) at time t The expressions are as follows:
[0084]
[0085]
[0086]
[0087]
[0088] in" "" indicates the Hadamard product, a matrix calculation method. This represents the output of the GRU at time t-1. This represents the data in X at time t. It is the sigmoid activation function; , , , , , These are the weight matrix parameters for the update gate, reset gate, and candidate state, respectively. , , These are the corresponding bias parameters. The expression for the tanh activation function is as follows:
[0089]
[0090] The above calculations yield the GRU output feature "f". gru = ".
[0091] S4, according to and The gating weights α of the feature adaptive fusion module are obtained, and the fused features are obtained from the feature adaptive fusion module. Preferably, step S4 specifically includes the following steps:
[0092] The gating weights α are obtained by passing the outputs of CNN and GRU through activation functions:
[0093]
[0094] in This is the weight matrix of the gate layer. This is the bias vector. After... The element-wise weight matrix is obtained, where the value of each element is between 0 and 1, representing the weight ratio of the corresponding feature in the fused output.
[0095] By dynamically adjusting the contribution ratio of CNN and GRU features using gating weight α, fused features are obtained. :
[0096]
[0097] S5, will After graph convolution operation, and using an activation function to output the fault location prediction value; preferably, step S5 specifically includes the following steps:
[0098] According to the diagram structure ( Figure 3 The adjacency matrix A is obtained by analyzing the connections between nodes in the matrix, and its normalization formula is... for:
[0099]
[0100] Where D is the degree matrix of A, and I is the identity matrix.
[0101] The following fault location prediction value Y is output after graph convolution operation:
[0102]
[0103] in and These are the weight matrix and bias parameters.
[0104] S6. Calculate the label smooth cross-entropy loss function and update the parameters of the fault location model through backpropagation; preferably, step S6 specifically includes the following steps:
[0105] The true distribution q(k) is defined as:
[0106]
[0107] The original output of the fault location model is normalized to obtain the probability distribution p(k) of each type of fault region:
[0108]
[0109] The expression for the label smoothing cross-entropy loss function L is obtained as follows:
[0110]
[0111] Where C is the total number of fault area categories, k represents a specific category, ϵ is the smoothing parameter (usually set to 0.1), and y is the true label (the index of the integer category). This refers to the output of various models in S5. This represents the output of each of the S5 models for the first to the Cth type of fault.
[0112] The parameters of the fault location model are updated through backpropagation.
[0113] S7. Using the test set Xc as input to the fault location model, repeat S2-S5 and calculate the fault location accuracy (the ratio of the total number of correctly predicted samples to the total number of samples in the test set). If the requirement is not met, repeat S2-S6. If the requirement is met, end the training and obtain the fault location model. Figure 6 The figure shows the changes in loss function and accuracy during training. Using this invention, the model accuracy of the proposed fault adaptive fusion method is above 95% with 300 small fault dataset samples.
[0114] S8. Select the fault data to be diagnosed (e.g., select the A-phase upper arm ground fault), input the fault location model trained in S7, and the model output is 7, corresponding to... Figure 2 The G7 node indicates that the fault occurred on the upper arm of phase A. The fault location is correct, and the fault location is complete.
[0115] Another aspect of the present invention provides an AC / DC fault intelligent location system based on multi-scale feature adaptive fusion, comprising: a computer-readable storage medium and a processor;
[0116] The computer-readable storage medium is used to store executable instructions;
[0117] The processor is used to read executable instructions stored in the computer-readable storage medium and execute the AC / DC fault intelligent location method based on multi-scale feature adaptive fusion as described in the first aspect.
[0118] In another aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent AC / DC fault location method based on multi-scale feature adaptive fusion as described in the first aspect.
[0119] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0120] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0121] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0122] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A smart AC / DC fault location method based on multi-scale feature adaptive fusion, characterized in that, The fault localization model is employed, which includes a time-series data convolution module, a gated recurrent unit, a multi-scale feature adaptive fusion module, and a graph convolution operation module. The method includes the following steps: S1. Collect fault datasets X and test set X from various regions of the DC and AC transmission systems under renewable energy access. c ; S2. Input the fault dataset X into the time series data convolution module CNN to obtain convolutional features. ; S3. Input the fault dataset X into the gated recurrent unit (GRU) to obtain long-short-term features. ; S4. Based on convolutional features and long-term and short-term characteristics The gating weight α of the multi-scale feature adaptive fusion module is obtained, and the multi-scale feature adaptive fusion module is based on... , And the fused features are obtained from the gating weight α. ; S5, the graph convolution operation module will After graph convolution operation, the fault location prediction value is output using an activation function; S6. Calculate the label smooth cross-entropy loss function and update the parameters of the fault location model through backpropagation; S7, using test set X c As input to the fault location model, repeat steps S2-S5 and calculate the accuracy of the fault location prediction. The accuracy is the ratio of the total number of correctly predicted samples to the total number of samples in the test set. If the requirements are not met, repeat steps S2-S6. If the requirements are met, the training ends and a well-trained fault location model is obtained. S8. Select fault data and input the fault location model trained in S7 to locate the fault.
2. The AC / DC fault intelligent location method based on multi-scale feature adaptive fusion as described in claim 1, characterized in that: Step S1 specifically includes: Referring to the topology of AC / DC transmission systems, lines or equipment areas with different potentials and currents are treated as nodes, and the connection relationships of each node Gn are determined according to the electrical structure of the system to form a graph structure. Each node in the graph structure is used as a fault area and electrical measurement point, and various types of faults are set in each area of the entire AC / DC transmission system. Electrical data is extracted from fault waveforms at each electrical measurement point and divided into datasets. and test set , where t, n, These represent the indices for time, node Gn, and electrical quantity type, respectively. Each fault sample data is labeled to form a true label y according to the actual fault location.
3. The AC / DC fault intelligent location method based on multi-scale feature adaptive fusion as described in claim 1, characterized in that: Step S2 specifically includes: F1 is obtained by passing X through the first one-dimensional convolution layer: ； in" "It consists of convolutional layers, BN normalization layers, and mappings learned by activation functions;" f1 is processed by the residual convolution block ResidualConvBlock to obtain f2: ； Where GELU is the activation function; f2 is processed by a max pooling layer to obtain the time series data convolution module output f. cnn : 。 4. The AC / DC fault intelligent location method based on multi-scale feature adaptive fusion as described in claim 1, characterized in that: Step S3 specifically includes: Data from X at each time t The input gated loop unit is processed by the following algorithm: ； ； ； ； in" "" indicates the Hadamard product, a matrix calculation method. This represents the output of the GRU at time t-1. This represents the data in X at time t; It is the sigmoid activation function; , , , , , These are the weight matrix parameters for the update gate, reset gate, and candidate state, respectively. , , Here are the corresponding bias parameters, and tanh is the activation function; The GRU output feature "f" is obtained after the above calculations. gru = ".
5. The AC / DC fault intelligent location method based on multi-scale feature adaptive fusion as described in claim 1, characterized in that: Step S4 specifically includes: It is obtained by directly concatenating the outputs of CNN and GRU in S2 and S3. ; The gating weight α is obtained based on the overall output features of CNN and GRU, and the gating weight α is expressed as: ； By dynamically adjusting the contribution ratio of CNN and GRU features using gating weight α, fused features are obtained. : 。 6. The AC / DC fault intelligent location method based on multi-scale feature adaptive fusion as described in claim 1, characterized in that: Step S5 specifically includes: Based on the connection relationships between nodes in the graph structure, the adjacency matrix A is obtained, and its normalization formula is... for: ； Where D is the degree matrix of A, and I is the identity matrix; The fault location prediction value Y is output by the graph convolution operation module: ； in and Here are the weight matrix and bias parameters, and GELU is the activation function; The graph structure is a data structure composed of nodes and edges, representing the relationships between objects. Lines or equipment areas with different potentials and currents are taken as several nodes, and the connection relationship of each node Gn is determined according to the electrical structure of the system to form a graph structure.
7. The AC / DC fault intelligent location method based on multi-scale feature adaptive fusion as described in claim 1, characterized in that: Step S6 specifically includes: Based on the prediction results of the S5 fault location model, the label smooth cross-entropy loss function L is obtained: ； Where C is the total number of fault area categories, k represents a specific category, q(k) is the true distribution of the constructed fault location areas, and p(k) is the probability distribution of each type of fault area output by the fault location model; the cross-entropy loss function optimizes the fault location model by minimizing the difference between the predicted distribution and the true distribution.
8. An intelligent AC / DC fault location system based on multi-scale feature adaptive fusion, comprising: Computer-readable storage media and processors; The computer-readable storage medium is used to store executable instructions; The processor is used to read executable instructions stored in the computer-readable storage medium and execute the AC / DC fault intelligent location method based on multi-scale feature adaptive fusion as described in any one of claims 1-7.
9. A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the AC / DC fault intelligent location method based on multi-scale feature adaptive fusion as described in any one of claims 1-7.