A scene-adaptive power distribution network fault handling method and device

Abstract

The application is suitable for the technical field of power distribution network fault disposal decision, and provides a scene adaptive power distribution network fault disposal method and device, the method comprises the following steps: nonlinear dimension reduction is performed on the collected power distribution network operation state data and external environment parameters by a stacked autoencoder to generate corresponding scene mode codes; a fault disposal process is matched and dynamically arranged from a preset process template library; in the process of executing the fault disposal process, the disposal state and risk probability of the next link are iteratively predicted by using a gating cycle unit; when the process is executed to the transfer supply analysis link, the power distribution network topology is divided into several feeder segment sub-areas according to key nodes, the maximum transferable capacity and transfer supply path skeleton of each sub-area are determined; a linear inequality system is constructed, the transfer supply path skeleton of each sub-area is substituted into the inequality system according to the fault current direction to solve, and a target cross-area fault transfer supply scheme is generated. The efficiency and accuracy of the power distribution network fault disposal are effectively improved.

Description

Technical Field

[0001] This application relates to the field of distribution network fault handling decision technology, and in particular to a scenario-adaptive distribution network fault handling method and apparatus. Background Technology

[0002] Existing power distribution network fault auxiliary decision-making systems mainly employ rule-based expert systems or offline contingency plan databases. The dispatch center pre-defines several fixed handling procedures and power transfer scheme templates based on typical network structures and common fault types. When a fault occurs, the system triggers an alarm based on fault indicators or protection action signals. Dispatchers, relying on experience, retrieve matching handling procedures from the contingency plan database and enumerate tie switches using a topology analysis module to select power transfer paths that meet voltage and capacity constraints. Finally, instructions are issued to field terminals in the form of operation tickets. Some systems also integrate fault waveform data and geographic information systems to assist in locating faulty sections.

[0003] Due to significant differences in fault characteristics, equipment status, communication conditions, and distributed resource output in power distribution networks under different seasons and operating modes, existing technologies employ uniform fixed procedures and contingency plan templates. These fail to adaptively adjust fault location priorities, contingency plan generation strategies, and the depth of power transfer analysis based on real-time weather, maintenance plans, communication quality, and other external environmental parameters. Consequently, during the rainy season when faults are concentrated, dispatchers still need to manually intervene in numerous potentially automated processes, resulting in severe delays in response. Furthermore, during peak maintenance periods, the lack of a conflict pre-detection mechanism between fixed operational sequences and planned power outages and restorations easily leads to the risk of operational errors. In mountainous areas with limited communication, processes relying on real-time remote interaction cannot operate normally. These factors severely restrict the speed of power restoration in power distribution networks, making existing fixed handling models insufficient to meet the demands for rapid and accurate fault handling. Summary of the Invention

[0004] This application provides a scenario-adaptive distribution network fault handling method and apparatus to address the problem that existing fixed handling modes cannot meet the needs for rapid and accurate fault handling.

[0005] The first aspect of this application provides a scenario-adaptive distribution network fault handling method, including: By using stacked autoencoders, nonlinear dimensionality reduction is performed on the collected power distribution network operation status data and external environmental parameters to generate corresponding scene mode codes; Based on the scenario pattern code, the fault handling process is matched and dynamically arranged from the preset process template library; During the execution of the fault handling process, the intermediate state information generated in the current handling stage is encoded into a handling state feature vector, and the gated loop unit is used to iteratively predict the handling state and risk probability of the next stage. When the process reaches the power transfer analysis stage, the distribution network topology is divided into several feeder segment sub-regions according to key nodes. The node and branch association matrix of each sub-region is truncated and singular value decomposition is performed to obtain a reduced-order approximate matrix in order to determine the maximum transferable capacity and power transfer path skeleton of each sub-region. With the constraints of voltage continuity, power balance, and continuous operation sequence between adjacent sub-regions, a set of linear inequalities is constructed. The power transfer path skeleton of each sub-region is substituted into the set of inequalities according to the fault current direction and solved sequentially. The target cross-region fault power transfer scheme is then generated by piecing them together.

[0006] Optionally, the step of performing nonlinear dimensionality reduction on the collected power distribution network operation status data and external environmental parameters through stacked autoencoders to generate corresponding scene mode codes includes: The collected power distribution network operation status data and external environmental parameters are normalized and then spliced ​​together to obtain the input feature vector; The input feature vector is nonlinearly reduced in dimensionality using the encoder in the pre-trained stacked autoencoder, and the vector output by the bottleneck layer of the encoder is extracted as a scene mode code; wherein, the stacked autoencoder is pre-trained with the goal of reconstructing the input feature vector, and the scene mode code is used to identify the type of the current fault handling scene.

[0007] Optionally, the step of matching and dynamically arranging fault handling procedures from a preset process template library based on the scenario pattern code includes: Calculate the similarity between the scenario pattern code and the scenario feature code corresponding to each process template in the preset process template library, and determine the process template with the highest similarity as the matching template; wherein, the scenario feature code is determined by the mean of the environmental parameter vector of historical typical fault handling cases under the corresponding scenario type; A basic handling process is generated based on the matching template, and the environmentally sensitive dimension component is extracted from the scene pattern code; When the environmentally sensitive dimension component exceeds the corresponding preset threshold, a sub-process is inserted or the original step is replaced at the corresponding position in the basic handling process to obtain a dynamically arranged fault handling process.

[0008] Optionally, encoding the intermediate state information generated in the current processing stage into a processing state feature vector includes: After the current handling phase is completed, the intermediate status information generated in this phase is collected. The intermediate status information includes the identified fault section, the generated candidate plan list, the evaluated transfer path set, the executed switching operation sequence, and the current handling time. The categorical variables in the intermediate state information are one-hot encoded, the numerical variables are normalized, and then concatenated with the scene pattern code to obtain the disposal state feature vector.

[0009] Optionally, the iterative prediction of the handling status and risk probability of the next stage using the gated loop unit includes: The disposal state feature vector is input into a pre-trained gated recurrent unit prediction module, which includes at least one gated recurrent unit layer and a top fully connected output layer. The prediction module of the gated loop unit outputs the predicted handling state vector for the next moment. The predicted handling state vector includes the predicted type of the next handling stage, the predicted remaining handling time, the predicted probability of operational conflict risk, and the predicted probability of transfer path failure risk.

[0010] Optionally, the step of performing truncated singular value decomposition on the node-branch association matrix of each sub-region to obtain a reduced-order approximate matrix includes: For each feeder segment sub-region, a node-branch association matrix is ​​constructed. The rows of the node-branch association matrix correspond to the nodes in the sub-region, and the columns correspond to the branches in the sub-region. The matrix elements represent the association relationship between nodes and branches and the branch impedance weights. Singular value decomposition is performed on the node-branch association matrix to obtain the singular value matrix and the corresponding left singular vector matrix and right singular vector matrix. Singular values ​​are truncated according to a preset energy retention ratio threshold, retaining the first... The largest singular value and its corresponding singular vector are used to reconstruct a reduced-order approximate matrix.

[0011] Optionally, determining the maximum transferable capacity and transfer path skeleton for each sub-region includes: The equivalent electrical parameters of the sub-region are reconstructed based on the reduced-order approximation matrix. The equivalent electrical parameters include equivalent node injected power, equivalent branch admittance, and equivalent boundary node voltage. Using the equivalent electrical parameters as input, a linear programming method is employed to solve for the maximum transferable capacity of the sub-region under the current fault conditions, satisfying node voltage constraints, branch thermal stability constraints, and distributed power output constraints.

[0012] Optionally, determining the maximum transferable capacity and transfer path skeleton for each sub-region further includes: From the left singular vector of the reduced-order approximation matrix, extract the nodes with the largest weights as key hubs of the transfer path; Based on the magnitude ranking of the equivalent branch admittances corresponding to the reduced-order approximation matrix, the connection relationships between the key pivot points are determined. Based on the key hub points and the connection relationships between them, the transfer path skeleton of the sub-region is generated.

[0013] Optionally, the process involves constructing a set of linear inequalities constrained by voltage continuity, power balance, and operational sequence continuity between adjacent sub-regions. The system then substitutes the power transfer path skeleton of each sub-region into this set of inequalities according to the fault current direction and solves them sequentially, piecing them together to generate the target cross-regional fault power transfer scheme. This includes: Starting from the sub-region where the fault point is located, determine the solution order of each sub-region sequentially along the direction of the fault current; At the splicing points of adjacent feeder segment sub-regions, voltage continuity constraints, power balance constraints, and operation sequence continuity constraints are constructed. The constraints at each splicing point are combined with the node voltage constraints, branch thermal stability constraints, and distributed power output constraints within each sub-region to form a set of linear inequalities. According to the solution order, the transfer path skeleton of each sub-region is substituted into the linear inequality set as the initial solution in turn, and the solution is performed with the goal of minimizing the number of switching operations and maximizing the recovery load, so as to obtain the transfer path of each sub-region. The supply transfer paths of each sub-region are spliced ​​together according to the solution order to generate the target cross-regional fault supply transfer scheme.

[0014] A second aspect of this application provides a scenario-adaptive power distribution network fault handling device, comprising: The scenario mode code generation module is used to perform nonlinear dimensionality reduction on the collected power distribution network operation status data and external environmental parameters through stacked autoencoders to generate corresponding scenario mode codes. The fault handling process matching module is used to match and dynamically arrange fault handling processes from a preset process template library based on the scenario pattern code. The handling status and risk probability prediction module is used to encode the intermediate status information generated in the current handling stage into a handling status feature vector during the execution of the fault handling process, and use the gated loop unit to iteratively predict the handling status and risk probability of the next stage. The reduced-order approximation matrix determination module is used when the process reaches the transfer analysis stage. It divides the distribution network topology into several feeder segment sub-regions according to key nodes, and performs truncated singular value decomposition on the node and branch association matrix of each sub-region to obtain a reduced-order approximation matrix, so as to determine the maximum transferable capacity and transfer path skeleton of each sub-region. The fault transfer scheme generation module is used to construct a set of linear inequalities under the constraints of voltage continuity, power balance and operation sequence continuity between adjacent sub-regions, and to solve the fault transfer path skeleton of each sub-region sequentially by substituting it into the set of inequalities according to the fault current direction, and splicing them together to generate the target cross-region fault transfer scheme.

[0015] As can be seen from the above technical solutions, this application has the following advantages: This application utilizes stacked autoencoders to perform nonlinear dimensionality reduction on distribution network operation status data and external environmental parameters to generate scenario pattern codes, achieving real-time perception and adaptive identification of different fault handling scenario types. Based on the scenario pattern codes, fault handling processes are matched and dynamically arranged from a process template library, enabling the handling strategy to be dynamically adjusted according to environmental factors such as thunderstorm level, maintenance intensity, and communication signal strength, overcoming the poor adaptability of fixed contingency plan templates. During process execution, gated loop units are used to encode intermediate handling states and iteratively predict the handling state and risk probability of the next stage to trigger risk warnings in advance. In the power transfer analysis stage, a reduced-order approximate matrix is ​​obtained by truncating singular value decomposition of the node and branch association matrices of each sub-region, reducing computational complexity while retaining key electrical modes, thereby quickly assessing the maximum transferable capacity of each sub-region and extracting the transfer path skeleton. A set of linear inequalities is constructed with voltage continuity, power balance, and operation sequence continuity between adjacent sub-regions as constraints, and is solved sequentially according to the fault current direction to generate cross-regional power transfer schemes. This ensures solution accuracy while achieving global coordination, effectively improving the efficiency and accuracy of distribution network fault handling. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a scenario-adaptive power distribution network fault handling method according to this application. Figure 2 This is a flowchart illustrating the matching and dynamic arrangement of fault handling procedures in this application; Figure 3 This is a flowchart illustrating the process of predicting the next stage of handling and the probability of risk in this application. Figure 4 This is a flowchart illustrating the process for determining the maximum transferable capacity and the backbone of the transfer path in this application. Figure 5 This is a flowchart illustrating the process of splicing together the target cross-regional fault transfer solution in this application. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] The implementation method in this embodiment can be implemented in a system, on a server, or on a terminal; no specific limitation is made. The method in this application will be described below from the perspective of system implementation. Please refer to... Figure 1 The following steps are included in the scenario-adaptive power distribution network fault handling method provided in this application embodiment: S1. The collected power distribution network operation status data and external environmental parameters are nonlinearly reduced by stacked autoencoders to generate corresponding scene mode codes; This system acquires real-time distribution network operation status data and external environmental parameters from the distribution network dispatch automation system, distribution automation system, fault indicators, distributed power generation monitoring system, meteorological early warning system, production management system, and communication network management system. The operation status data includes distribution network topology connections, switch status, feeder current and voltage, distributed power generation output, and fault indicator status. External environmental parameters include thunderstorm level, rainfall, maintenance schedule density, and communication signal strength. This embodiment achieves real-time perception and adaptive identification of the current fault handling scenario type by jointly performing nonlinear dimensionality reduction on multi-source heterogeneous environmental parameters and electrical status parameters, providing a basis for the dynamic orchestration of subsequent processes. The specific steps are as follows: 1. The collected power distribution network operation status data and external environmental parameters are normalized and then spliced ​​together to obtain the input feature vector; 2. The encoder in the pre-trained stacked autoencoder is used to perform nonlinear dimensionality reduction on the input feature vector, and the vector output by the bottleneck layer of the encoder is extracted as the scene mode code; wherein, the stacked autoencoder is pre-trained with the goal of reconstructing the input feature vector, and the scene mode code is used to identify the type of the current fault handling scenario.

[0019] Specifically, discrete quantities such as switch states are encoded using one-hot encoding; effective amplitude values ​​are extracted for feeder current and voltage; all continuous parameters are mapped to the [0,1] interval using minimum and maximum normalization methods; and seasonal identifiers are encoded using 1-of-4 encoding. All normalized parameters are concatenated in a preset order to form a one-dimensional input feature vector. The stacked autoencoder is an unsupervised learning neural network consisting of an encoder and a decoder. In this embodiment, the encoder consists of four fully connected layers stacked together, with 128, 64, 32, and 16 neurons in each layer, followed by a batch normalization layer and a ReLU activation function. The fourth layer is the bottleneck layer. The decoder structure is symmetrical to the encoder, with 32, 64, and 128 neurons in each layer. The last layer has the same dimension as the input feature vector, and the output value is constrained to the [0,1] interval using a Sigmoid activation function. The pre-training stage aims to minimize the reconstruction error, using mean squared error as the loss function and the Adam optimizer for training, enabling the encoder to learn a low-dimensional representation that retains the main changes in the input data. After pre-training, the 16-dimensional real-valued vector output from the encoder's bottleneck layer is extracted as a scene pattern code. This scene pattern code is a low-dimensional real-valued vector, where each component represents the feature response of the input data at different abstract dimensions. It is used to uniquely identify the type of the current fault handling scenario. Scene types include at least one of the following: high-incidence thunderstorm mode, peak maintenance mode, mountainous communication-restricted mode, and normal mode. In practical applications, the Euclidean distance between the scene pattern code obtained in real time and the typical center vectors of the above-mentioned modes is calculated, and the closest one is the current scene type.

[0020] S2. Based on the scenario pattern code, match and dynamically arrange the fault handling process from the preset process template library; The pre-built process template library is a collection of digital templates containing corresponding handling procedures for various scenario types. The construction process involves: for each scenario type, such as high-incidence thunderstorm mode, peak maintenance mode, mountainous communication-restricted mode, and normal mode, scheduling experts create corresponding fault handling flowcharts. These flowcharts are then converted into digital templates composed of handling stage nodes and directed edges using business process modeling and annotation standards. Each handling stage node includes attributes such as node type identifier, execution conditions, timeout time, and callable algorithm interfaces. Node types include fault location nodes, contingency plan generation nodes, transfer analysis nodes, instruction execution nodes, and feedback archiving nodes. Directed edges represent the execution order between nodes, supporting sequential, branching, and looping structures. Simultaneously, environmental parameter vectors from historical typical fault handling cases for each scenario type are collected, and their average value is used as the scenario feature code corresponding to the template. This process template library provides a standardized foundation for subsequent template matching and dynamic orchestration. Please refer to [link to relevant documentation]. Figure 2 The specific steps are as follows: S21. Calculate the similarity between the scenario pattern code and the scenario feature code corresponding to each process template in the preset process template library, and determine the process template with the highest similarity as the matching template; wherein, the scenario feature code is determined by the mean of the environmental parameter vector of historical typical fault handling cases under the corresponding scenario type; Specifically, the scene feature code is the mean vector of environmental parameter vectors from historical typical fault handling cases under each scene type, and its dimension is the same as the scene pattern code generated in step S1. In this embodiment, cosine similarity is used as the similarity metric. The cosine similarity values ​​of the calculated scene pattern code and each template scene feature code are sorted from high to low, and the template with the highest similarity is selected as the matching template. If the highest similarity is lower than the set threshold of 0.5, the default normal mode template is used to ensure that there is always an executable process template as a basis in extreme or unknown scenarios.

[0021] S22. Generate a basic handling process based on the matching template, and extract the environmentally sensitive dimension components from the scene pattern code; S23. When the environmentally sensitive dimension component exceeds the corresponding preset threshold, insert a sub-process or replace the original step at the corresponding position in the basic handling process to obtain a dynamically arranged fault handling process.

[0022] Specifically, the basic handling process refers to the standard handling process organized according to the predefined node sequence in the matching template, without any dynamic adjustments. Environmentally sensitive dimension components refer to specific dimension components strongly correlated with the external environment extracted from the scenario pattern code generated in step S1, including thunderstorm level components, maintenance plan intensity components, and mountainous communication signal strength components. The preset thresholds corresponding to the above-mentioned environmentally sensitive dimension components are determined based on historical distribution network operation data and dispatching expert experience: the thunderstorm level threshold is set to level 3, the maintenance plan intensity threshold is set to 0.7, and the mountainous communication signal strength threshold is set to -90dBm. When the thunderstorm level component exceeds the threshold of 3, a meteorological warning verification sub-process is inserted before the fault location node. This sub-process is used to obtain the latest lightning location information and assess the probability of line lightning strikes. When the maintenance plan density component exceeds the threshold of 0.7, an operation conflict detection sub-process is inserted before the instruction execution node. This sub-process is used to check whether the sequence of operations to be executed conflicts with the current planned power outage or restoration operations in terms of time or equipment. When the mountainous area communication signal strength component is below the threshold of -90dBm, the instruction execution node is replaced with a localized simplified instruction issuance sub-process. This sub-process generates operation instructions that can be directly executed on the local terminal without relying on real-time remote interaction. Through the above dynamic adjustments, a fault handling process example adapted to the current real-time scenario is obtained.

[0023] S3. During the execution of the fault handling process, the intermediate state information generated in the current handling stage is encoded into a handling state feature vector, and the gated loop unit is used to iteratively predict the handling state and risk probability of the next stage. After dynamically orchestrating and generating the fault handling process in step S2, the process engine sequentially activates and executes each handling stage node according to the pointing order of directed edges. The execution process includes fault location, contingency plan generation, supply chain analysis, instruction execution, and feedback archiving. This step introduces iterative prediction based on gated loop units during this execution process. By encoding and timing-modeling the intermediate states after each stage is completed, the handling status and potential risks of the next stage can be predicted in advance, thereby achieving proactive early warning. Please refer to [link / reference]. Figure 3 The specific steps are as follows: S31. After the current handling phase is completed, collect the intermediate status information generated in this phase. The intermediate status information includes the identified fault section, the generated candidate plan list, the evaluated transfer path set, the executed switch operation sequence, and the current handling time. Specifically, "completion of the current handling phase" refers to the situation where a handling phase node in the process engine has completed execution and returned the execution result. For example, the fault location node outputs the fault segment identifier after completing the location algorithm, or the transfer analysis node outputs the candidate transfer path set after completing the path search. After the execution of each phase node, the collection of intermediate status information is automatically triggered. Among them, the located fault segment identifier represents the unique number of the fault location, such as segment S2 of feeder F3; the generated candidate plan list is the set of identifiers of multiple handling plans output by the plan generation node; the evaluated transfer path set is a number of transfer paths that meet the preliminary electrical constraints selected by the transfer analysis node, and each path is represented by a switch sequence and a node sequence; the executed switch operation sequence is the switch operation record that has been successfully issued by the instruction execution node, sorted by operation time; and the current elapsed handling time is the time elapsed from the time the fault occurred to the current time.

[0024] S32. Perform one-hot encoding on categorical variables and normalization on numerical variables in the intermediate state information, and concatenate them with the scene pattern code to obtain the disposal state feature vector; Specifically, for categorical variables such as fault segment identifiers, candidate contingency plan lists, and evaluated transfer path sets, one-hot encoding is performed based on their corresponding totals to generate fixed-dimensional binary vectors. For executed switch operation sequences, they are first converted into operation type codes, and then the number of each type of operation is counted to form a fixed-length operation statistical vector. For the numerical variable of the current elapsed processing time, minimum and maximum normalization methods are used to map it to the [0,1] interval. After completing the above encoding and normalization processes, all sub-vectors are concatenated end-to-end in a fixed order, and then concatenated with the scene mode code vector generated in step S1 to obtain a fixed-dimensional processing status feature vector.

[0025] S33. Input the disposal state feature vector into the pre-trained gated recurrent unit prediction module, the gated recurrent unit prediction module includes at least one gated recurrent unit layer and a top fully connected output layer; S34. The prediction module of the gated loop unit outputs the predicted handling state vector for the next moment. The predicted handling state vector includes the predicted type of the next handling stage, the predicted remaining handling time, the predicted probability of operational conflict risk, and the predicted probability of failure of the transfer path.

[0026] This gated recurrent unit (ROU) is a variant of the recurrent neural network, capable of effectively processing time-series data and avoiding the gradient vanishing problem. In this embodiment, the gated ROU prediction module employs two stacked gated ROU layers, with each layer having a hidden state dimension of 64, and a fully connected output layer at the top. This prediction module requires pre-training using historical fault handling data. During training, the feature vector of the handling state at a historical moment is used as input, and the actual state vector at the next moment is used as the label. The mean squared error loss function and the Adam optimizer are used for parameter learning. Specifically, for each historical fault handling case, if a time overlap or equipment conflict actually occurs with the planned power outage / restoration operation during the subsequent instruction execution phase, the label value of that case is marked as 1; otherwise, it is marked as 0, thus forming a binary classification training label. For each historical fault handling case, if the selected power transfer path actually fails during execution due to distributed resource output fluctuations, load mutations, or switch failure, the label value of that case is marked as 1; otherwise, it is marked as 0. During model training, the aforementioned label values ​​and the labels of other predicted targets together constitute the actual state vector for the next moment. The gated recurrent unit prediction module learns the mapping relationship between the state feature vector and the state vector through backpropagation, so that the output of the corresponding node of the fully connected output layer is mapped to the [0,1] interval by the Sigmoid activation function, which are respectively used as the probability of operation conflict risk and the probability of transfer path failure.

[0027] In practical applications, the disposal status feature vector obtained in step S32 is input into the pre-trained gated loop unit prediction module, which outputs a predicted disposal status vector containing four prediction targets. The predicted next disposal stage type refers to predicting the next disposal stage node to be executed based on the current state, outputting the probability distribution of each stage, with the highest probability taken as the prediction result; the predicted remaining disposal time refers to the estimated time required from the current moment until the end of the entire fault disposal process; the predicted operational conflict risk probability refers to the possibility of operational sequence conflicts or equipment conflicts during subsequent instruction execution, with a value range of [0,1]; the predicted transfer path failure risk probability refers to the probability that the selected transfer path may fail during execution due to distributed resource fluctuations, load changes, or switch failures, with a value range of [0,1]. When the predicted risk probability exceeds a preset threshold of 0.7, the system pushes a warning message to the dispatcher interface and automatically adjusts the execution parameters of subsequent process nodes or suggests switching to a backup disposal strategy, thereby achieving early prediction and proactive prevention of fault disposal risks.

[0028] S4. When the process reaches the power transfer analysis stage, the distribution network topology is divided into several feeder segment sub-regions according to key nodes. The node and branch association matrix of each sub-region is truncated and singular value decomposition is performed to obtain a reduced-order approximate matrix in order to determine the maximum transferable capacity and power transfer path skeleton of each sub-region. In the fault handling process dynamically orchestrated in step S2, when the process engine sequentially activates each handling stage node according to the directed edge order and executes to the power transfer analysis stage—that is, when the faulty section has been isolated and a power transfer path for restoring power to the downstream power-loss loads needs to be found—this step is triggered. This step decomposes the power transfer analysis problem of large-scale distribution networks into two stages: independent evaluation of multiple sub-regions and cross-regional splicing coordination. By performing dimensionality reduction evaluation on each sub-region, the solution efficiency of power transfer analysis is significantly improved while ensuring engineering accuracy. Please refer to [link to relevant documentation]. Figure 4 The details are as follows: S41. For each feeder segment sub-region, construct a node-branch association matrix. The rows of the node-branch association matrix correspond to the nodes in the sub-region, and the columns correspond to the branches in the sub-region. The matrix elements represent the association relationship between nodes and branches and the branch impedance weights. S42. Perform singular value decomposition on the node-branch association matrix to obtain the singular value matrix and the corresponding left singular vector matrix and right singular vector matrix; S43. Truncate the singular values ​​according to the preset energy retention ratio threshold, retain the top k largest singular values ​​and their corresponding singular vectors, and reconstruct the reduced-order approximate matrix.

[0029] First, construct the Laplace matrix of the distribution network. ,in This is a similarity matrix between nodes, and its elements are... Defined as the reciprocal of electrical distance: In the formula: For nodes With nodes The equivalent impedance magnitude between It is a diagonal matrix, and its diagonal elements .

[0030] Then calculate The former The eigenvectors corresponding to the smallest non-zero eigenvalues The values ​​range from 8 to 15. These eigenvectors are arranged column-wise to form a matrix. ,in This represents the total number of nodes. (For...) K-means clustering is performed on the row vectors, and the number of cluster centers is automatically determined by the silhouette coefficient. The clustering result ensures that the average electrical distance between any two nodes within the same sub-region is less than the average electrical distance between nodes in different sub-regions, thereby achieving a partition with high cohesion and low coupling.

[0031] For each sub-region, construct a node-branch association matrix. Let there be m nodes and n branches within a sub-region; define the association matrix... Its elements are To reflect the importance of each branch in fault transfer, a branch importance weight is introduced. Construct a weighted correlation matrix Its elements are: Branch importance weight The calculation formula is: in: branch road Rated current carrying capacity, This represents the maximum rated current carrying capacity of all branches within the sub-region. branch road The historical failure rate, with a value range of [0,1], is obtained by normalization. branch road The topological betweenness of a branch is the number of shortest paths through that branch. It is the maximum topological betweenness number within the subregion.

[0032] This weighting ensures that branches with high carrying capacity, low historical failure rate, and high topological importance are more likely to be retained in subsequent low-rank approximations.

[0033] For weighted incidence matrix Perform standard singular value decomposition: in: Let be a left singular vector matrix, satisfying ; Let be a right singular vector matrix, satisfying ; For a singular value matrix, its diagonal elements for Singular values, all other elements are zero.

[0034] Singular values ​​are truncated according to a preset energy retention ratio threshold, which is set between 90% and 95%. Through statistical analysis of historical distribution network operation data, when the cumulative energy percentage of retained singular values ​​reaches over 90%, the reconstructed matrix can capture the main electrical transmission patterns related to power transfer analysis in the original matrix. However, values ​​below this threshold may lead to the loss of critical topology information. The specific method for truncating singular values ​​is to select the smallest positive integer... , so that the former is retained The cumulative energy percentage of each singular value is not less than the energy retention ratio threshold, that is: in: The energy retention threshold is set between 90% and 95%. (Before retention) The largest singular value and its corresponding left singular vector and right singular vector . Finally, reconstruct the reduced-order approximation matrix. for: in: ; ; ;here This is an adaptive low-rank approximation of the original weighted incidence matrix.

[0035] Due to rank Much smaller than the rank of the original matrix Subsequently, the number of variables in the linear programming solution based on this matrix was greatly reduced, and a significant improvement in computational speed was achieved while the loss of engineering accuracy was controllable.

[0036] Based on the reduced-order approximation matrix described above, the maximum transferable capacity assessment and transfer path skeleton extraction for each sub-region can be completed respectively. The specific process is as follows: 1. Reconstruct the equivalent electrical parameters of the sub-region based on the reduced-order approximation matrix. The equivalent electrical parameters include equivalent node injected power, equivalent branch admittance, and equivalent boundary node voltage. 2. Using equivalent electrical parameters as input, linear programming is used to solve for the maximum transferable capacity of the sub-region under the current fault conditions, satisfying node voltage constraints, branch thermal stability constraints, and distributed power output constraints.

[0037] Based on the rank of the reduced-order approximation matrix The main characteristic patterns of this sub-region are determined, and each characteristic pattern corresponds to an equivalent power transfer channel consisting of key nodes and key branches. The specific method for reconstructing the equivalent electrical parameters of this sub-region is as follows: Equivalent node injected power The calculation formula is: in: This represents the injected power vector for each node within the original sub-region. The left singular vector matrix reserved for step S43, This represents the injected power vector of the equivalent node after order reduction. This formula will transform the original... The injected power of the dimensional node is projected onto the transpose of the left singular vector matrix. The lower-order space preserves the power distribution pattern that contributes most to the power transfer analysis.

[0038] Equivalent branch admittance matrix The calculation formula is: in: This is the original branch admittance matrix (a diagonal matrix, where the diagonal elements are the admittance values ​​of each branch). The right singular vector matrix reserved in step S43 The singular value diagonal matrix reserved in step S43, This is the reduced-order equivalent branch admittance matrix. This formula uses left and right singular vectors to bidirectionally reduce the original branch admittance, preserving the equivalent admittance information corresponding to the main electrical modes.

[0039] The equivalent boundary node voltage is directly taken from the actual voltage measurement value of the sub-region boundary node. The boundary node is the node in the sub-region connected to the adjacent sub-region through a tie switch or a common node. Its voltage amplitude and phase angle are obtained from the real-time measurement data of the distribution automation system. Using these equivalent electrical parameters as input, a linear programming model is established, with the objective function being to maximize the transfer power. The constraints include: the node voltage amplitude is maintained within the range of 0.95 to 1.05 times the rated voltage, the transmission power of each branch does not exceed its rated current carrying capacity, and the output of distributed generation does not exceed its rated capacity. The simplex method is used to solve this linear programming model to obtain the maximum transferable capacity of the sub-region that can be transferred from adjacent feeders or distributed generation to the downstream load of the fault under the current fault conditions.

[0040] 1. Extract the nodes with the largest weights from the left singular vectors of the reduced-order approximation matrix as key hubs in the transfer path; 2. Based on the magnitude ranking of the equivalent branch admittances corresponding to the reduced-order approximation matrix, determine the connection relationship between key pivot points; 3. Generate the transfer path skeleton for sub-regions based on the key hub points and the connection relationships between them.

[0041] Specifically, the left singular vector matrix retained in step S43 For each column, calculate the absolute value of each element. This absolute value reflects the weight of the corresponding node in the feature pattern. Take the node with the largest absolute value as the key pivot point of the pattern; then combine all... The key pivot points of each pattern are merged and deduplicated to obtain the set of key pivot points for the sub-region. Next, the amplitudes of the equivalent branch admittances calculated in step 2 are sorted in descending order, and the top two amplitudes are selected. Equivalent branches are selected as candidate edges connecting key hubs. Finally, the set of key hubs is used as the vertices of the graph, and the selected equivalent branches are used as the edges to construct a simplified topology graph, which serves as the backbone of the transfer path for this sub-region. The transfer path backbone retains only the hub nodes and main branches that play a decisive role in the transfer, excluding minor intermediate nodes, thus providing a concise and effective initial solution for the cross-regional path splicing in the subsequent step S5.

[0042] S5. With the constraints of voltage continuity, power balance and operation sequence continuity between adjacent sub-regions, construct a set of linear inequalities, and substitute the power transfer path skeleton of each sub-region into the set of inequalities according to the fault current direction to solve them in sequence, and splice them to generate the target cross-region fault power transfer scheme.

[0043] After obtaining the maximum transferable capacity assessment results and transfer path skeleton for each sub-region in step S4, these relatively independent sub-region schemes need to be spliced ​​together into a complete cross-regional transfer scheme encompassing the entire downstream region of the fault. This step transforms the electrical constraints of the cross-regional splicing into a standard system of linear inequalities and employs a decomposition and coordination strategy of solving zone by zone along the fault current direction and iteratively verifying the boundaries to generate a globally feasible transfer scheme while ensuring electrical continuity at the splicing points. Please refer to [link to relevant documentation]. Figure 5 The specific steps are as follows: S51. Starting from the sub-region where the fault point is located, determine the solution order of each sub-region in sequence along the direction of the fault current; The sub-region where the fault point is located refers to the feeder segment sub-region where the fault occurs, which has already been determined in the fault location step S3. Starting from this sub-region, the solution order for the remaining sub-regions is determined sequentially along the fault current direction: if the fault current flows from upstream to the fault point, the solution order for each sub-region is determined sequentially upstream from the sub-region where the fault point is located. The upstream sub-region sequence is denoted as... If there are sub-regions downstream of the fault that require power restoration, the solution order is determined sequentially from the sub-region where the fault point is located downstream. The sequence of downstream sub-regions is denoted as... This embodiment employs a bidirectional expansion strategy, first solving upstream to the power supply side, and then downstream to the load end. This ensures that when solving the latter sub-region, the voltage and power boundary conditions of the former sub-region at the splicing point have already been substituted as known quantities, thereby achieving effective transfer of boundary conditions.

[0044] S52. At the splicing point of adjacent feeder segment sub-regions, construct voltage continuity constraints, power balance constraints, and operation sequence continuity constraints, and combine the constraints at each splicing point with the node voltage constraints, branch thermal stability constraints, and distributed power output constraints within each sub-region to form a set of linear inequalities; Specifically, for two adjacent feeder segment sub-regions and Let sub-regions be defined The export boundary node is sub-region The entry boundary node is ,and and The connection point, either via a handover switch or a public node, is designated as the splicing point. At the splicing point The following three types of electrical constraints are constructed respectively: Voltage continuity constraint: assuming sub-region At the splicing point The voltage phasor at that point is sub-region At the splicing point The voltage phasor at that point is ,in , Voltage amplitude, , This refers to the voltage phase angle. Voltage continuity constraints require that the voltage amplitude difference and phase angle difference on both sides of the splicing point be within allowable ranges, i.e. and ,in The voltage amplitude deviation threshold is set between 0.01 and 0.05 times the rated voltage, with a preferred value of 0.02 times the rated voltage. The phase angle deviation threshold is defined as 0.01 to 0.1 radians, with a preferred value of 0.05 radians. The above inequality can be transformed into a system of linear inequalities: Power balance constraint: Assume the flow passes through the splice point From sub-region Injection sub-region The active power is reactive power is Subregion The rated capacity of the side tie switch is sub-region The load requirement of the side boundary node is and Power balance constraints require that the injected power must meet downstream load demand without exceeding the transmission capacity of upstream equipment, while also satisfying power factor limitations. To maintain linearity, a linearization approximation can be used. This involves introducing a power factor angle. The power factor is defined as the factor used in power distribution networks. =0.9 (lag). Therefore, the relationship between reactive power and active power can be approximated as: Substituting into the capacity constraint, we get: Therefore, the linear constraint for active power is: Similarly, the linear constraint for reactive power is: in .

[0045] Operation sequence continuity constraint: Let the sequence of switching operations involved in the transfer path of sub-region A be as follows. The sequence of switching operations involved in the transfer path of sub-region B is as follows: ,in: and These are the total number of switch operations in subregions A and B, respectively, and are positive integers; each This indicates a switch operation command, such as closing the tie switch K1 or opening the section switch K2.

[0046] splicing point The position of the switch (connecting switch) in the operating sequence must satisfy the following condition: the switch is in The middle is the last operation object, in Let `m` be the first object being operated on. Indicates operation The execution time constraint is as follows: Furthermore, the time difference between the two must not be less than the minimum time interval between the switching actions. : in Depending on the switch type: for circuit breakers, the typical value is 0.2 seconds; for load switches, the typical value is 1 second; for disconnect switches, the typical value is 2 seconds. For safety reasons, a conservative value is used in this embodiment. 1 second. This constraint can be linearized as: .

[0047] Combining all the above linear constraints, we form a system of linear inequalities concerning the decision variable vector x: in: Let be a vector of decision variables, with dimension . ( (A positive integer), its components include: the switch status in each sub-region's power transfer path (value is 0 or 1, 0 indicates open, 1 indicates closed), power allocation coefficient (continuous variable, value range 0~1, representing the proportion of power allocated to a certain path), and boundary node voltage amplitude (continuous variable, value range 0.95~1.05 times the rated voltage). It is a coefficient matrix, and the matrix elements are composed of the coefficients of the decision variables in each constraint inequality; It is a constant vector with dimension N; its elements are the constant terms on the right-hand side of each constraint inequality.

[0048] S53. Following the solution order, substitute the transfer path skeleton of each sub-region as the initial solution into the system of linear inequalities, and solve the problem with the goal of minimizing the number of switching operations and maximizing the recovery load to obtain the transfer path of each sub-region. Specifically, following the solution order determined in step S51, the transfer path is solved sequentially for each sub-region. For the current sub-region, the transfer path skeleton extracted in step S4 is substituted into the system of linear inequalities as the initial solution to narrow the search range of the feasible region and accelerate solution convergence. The objective function is constructed as follows: in: The number of switching operations that need to be performed in the current sub-region transfer path is a non-negative integer; The maximum number of operations allowed is usually set to 10. Operations exceeding this value are considered too complex and should not be used. The load power that can be restored through this transfer path; The total load power in the downstream area of ​​the fault is obtained in real time from the distribution automation system. This is a weighting coefficient, with a value ranging from 0 to 1, preferably... =0.4 indicates a greater emphasis on load recovery.

[0049] The objective function is a linear combination; the first term penalizes excessive switching operations, and the second term rewards restoring more load. This is achieved by satisfying a system of linear inequalities that satisfy the steps. Under constraints, a mixed-integer linear programming solver is used to solve for the optimal transfer path within the sub-region, yielding the decision variables. The optimal values ​​are obtained to determine the specific switching operation sequence and power allocation scheme.

[0050] S54. Connect the transfer paths of each sub-region in the order of solution to generate the target cross-region fault transfer scheme.

[0051] Specifically, the power transfer paths obtained from solving each sub-region are sequentially spliced ​​together according to the solution order determined in step S51 to form a complete power transfer path covering the entire downstream area of ​​the fault. After splicing, the electrical quantities at the splicing points of adjacent sub-regions are checked for consistency: assuming the splicing point voltage calculated by the previous sub-region is... The splicing point voltage calculated for the latter sub-region is Rated voltage is If the relative error is: Then introduce a boundary relaxation factor. The value ranges from 0.8 to 1.2. After adjustment, the power constraint of a sub-region is: in The initial value is 1.0. If the constraint needs to be relaxed, it is gradually increased to 1.2; if it needs to be tightened, it is decreased to 0.8. Then the solution is repeated until the constraint conditions of all splicing points are satisfied. The upper limit of iteration is set to 5 times. If it is still not satisfied, a conservative strategy is adopted: the average of the two results is taken as the final boundary condition. The final output is a complete fault transfer path scheme, including the switching operation sequence of each sub-region and the timing sequence of cross-regional interconnection switching operations. The scheme is stored in the database in structured data form for dispatcher review or automatic execution.

[0052] Furthermore, an embodiment of the scenario-adaptive distribution network fault handling device of this application includes the following: The scenario mode code generation module is used to perform nonlinear dimensionality reduction on the collected power distribution network operation status data and external environmental parameters through stacked autoencoders to generate corresponding scenario mode codes. The fault handling process matching module is used to match and dynamically arrange fault handling processes from a preset process template library based on scenario pattern codes. The handling status and risk probability prediction module is used to encode the intermediate status information generated in the current handling stage into a handling status feature vector during the execution of the fault handling process, and use the gated loop unit to iteratively predict the handling status and risk probability of the next stage. The reduced-order approximation matrix determination module is used when the process reaches the transfer analysis stage. It divides the distribution network topology into several feeder segment sub-regions according to key nodes, and performs truncated singular value decomposition on the node and branch association matrix of each sub-region to obtain a reduced-order approximation matrix, so as to determine the maximum transferable capacity and transfer path skeleton of each sub-region. The fault transfer scheme generation module is used to construct a set of linear inequalities under the constraints of voltage continuity, power balance and operation sequence continuity between adjacent sub-regions, and to solve the fault transfer path skeleton of each sub-region sequentially by substituting it into the set of inequalities according to the fault current direction, and splicing them together to generate the target cross-region fault transfer scheme.

[0053] For specific limitations regarding the device, please refer to the limitations of the method above, which will not be repeated here. Each module in the above device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0054] It is understood that those skilled in the art can combine various implementation methods in the above embodiments under the guidance of the above examples to obtain technical solutions with multiple implementation methods.

[0055] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A scenario-adaptive distribution network fault handling method, characterized in that, include: By using stacked autoencoders, nonlinear dimensionality reduction is performed on the collected power distribution network operation status data and external environmental parameters to generate corresponding scene mode codes; Based on the scenario pattern code, the fault handling process is matched and dynamically arranged from the preset process template library; During the execution of the fault handling process, the intermediate state information generated in the current handling stage is encoded into a handling state feature vector, and the gated loop unit is used to iteratively predict the handling state and risk probability of the next stage. When the process reaches the power transfer analysis stage, the distribution network topology is divided into several feeder segment sub-regions according to key nodes. The node and branch association matrix of each sub-region is truncated and singular value decomposition is performed to obtain a reduced-order approximate matrix in order to determine the maximum transferable capacity and power transfer path skeleton of each sub-region. With the constraints of voltage continuity, power balance, and continuous operation sequence between adjacent sub-regions, a set of linear inequalities is constructed. The power transfer path skeleton of each sub-region is substituted into the set of inequalities according to the fault current direction and solved sequentially. The target cross-region fault power transfer scheme is then generated by piecing them together.

2. The scenario-adaptive distribution network fault handling method according to claim 1, characterized in that, The process of performing nonlinear dimensionality reduction on the collected power distribution network operation status data and external environmental parameters using a stacked autoencoder to generate corresponding scene mode codes includes: The collected power distribution network operation status data and external environmental parameters are normalized and then spliced ​​together to obtain the input feature vector; The input feature vector is nonlinearly reduced in dimensionality using the encoder in the pre-trained stacked autoencoder, and the vector output by the bottleneck layer of the encoder is extracted as a scene mode code; wherein, the stacked autoencoder is pre-trained with the goal of reconstructing the input feature vector, and the scene mode code is used to identify the type of the current fault handling scene.

3. The scenario-adaptive distribution network fault handling method according to claim 1, characterized in that, The step of matching and dynamically arranging fault handling procedures from a preset process template library based on the scenario pattern code includes: Calculate the similarity between the scenario pattern code and the scenario feature code corresponding to each process template in the preset process template library, and determine the process template with the highest similarity as the matching template; wherein, the scenario feature code is determined by the mean of the environmental parameter vector of historical typical fault handling cases under the corresponding scenario type; A basic handling process is generated based on the matching template, and the environmentally sensitive dimension component is extracted from the scene pattern code; When the environmentally sensitive dimension component exceeds the corresponding preset threshold, a sub-process is inserted or the original step is replaced at the corresponding position in the basic handling process to obtain a dynamically arranged fault handling process.

4. The scenario-adaptive distribution network fault handling method according to claim 1, characterized in that, Encoding the intermediate state information generated in the current handling stage into a handling state feature vector includes: After the current handling phase is completed, the intermediate status information generated in this phase is collected. The intermediate status information includes the identified fault section, the generated candidate plan list, the evaluated transfer path set, the executed switching operation sequence, and the current handling time. The categorical variables in the intermediate state information are one-hot encoded, the numerical variables are normalized, and then concatenated with the scene pattern code to obtain the disposal state feature vector.

5. The scenario-adaptive distribution network fault handling method according to claim 4, characterized in that, The method of iteratively predicting the handling status and risk probability of the next stage using a gated loop unit includes: The disposal state feature vector is input into a pre-trained gated recurrent unit prediction module, which includes at least one gated recurrent unit layer and a top fully connected output layer. The prediction module of the gated loop unit outputs the predicted handling state vector for the next moment. The predicted handling state vector includes the predicted type of the next handling stage, the predicted remaining handling time, the predicted probability of operational conflict risk, and the predicted probability of transfer path failure risk.

6. The scenario-adaptive distribution network fault handling method according to claim 1, characterized in that, The process of performing truncated singular value decomposition on the node-branch association matrix of each sub-region to obtain a reduced-order approximate matrix includes: For each feeder segment sub-region, a node-branch association matrix is ​​constructed. The rows of the node-branch association matrix correspond to the nodes in the sub-region, and the columns correspond to the branches in the sub-region. The matrix elements represent the association relationship between nodes and branches and the branch impedance weights. Singular value decomposition is performed on the node-branch association matrix to obtain the singular value matrix and the corresponding left singular vector matrix and right singular vector matrix. Singular values ​​are truncated according to a preset energy retention ratio threshold, retaining the first... The largest singular value and its corresponding singular vector are used to reconstruct a reduced-order approximate matrix.

7. The scenario-adaptive distribution network fault handling method according to claim 6, characterized in that, The process of determining the maximum transferable capacity and transfer path skeleton for each sub-region includes: The equivalent electrical parameters of the sub-region are reconstructed based on the reduced-order approximation matrix. The equivalent electrical parameters include equivalent node injected power, equivalent branch admittance, and equivalent boundary node voltage. Using the equivalent electrical parameters as input, a linear programming method is employed to solve for the maximum transferable capacity of the sub-region under the current fault conditions, satisfying node voltage constraints, branch thermal stability constraints, and distributed power output constraints.

8. The scenario-adaptive distribution network fault handling method according to claim 7, characterized in that, The process of determining the maximum transferable capacity and transfer path skeleton for each sub-region also includes: From the left singular vector of the reduced-order approximation matrix, extract the nodes with the largest weights as key hubs of the transfer path; Based on the magnitude ranking of the equivalent branch admittances corresponding to the reduced-order approximation matrix, the connection relationships between the key pivot points are determined. Based on the key hub points and the connections between them, the transfer path skeleton for the sub-region is generated.

9. The scenario-adaptive distribution network fault handling method according to claim 1, characterized in that, The process involves constructing a set of linear inequalities constrained by voltage continuity, power balance, and operational sequence continuity between adjacent sub-regions. The power transfer path skeletons of each sub-region are then substituted into this set of inequalities according to the fault current direction and solved sequentially. These inequalities are then pieced together to generate the target cross-regional fault power transfer scheme, including: Starting from the sub-region where the fault point is located, determine the solution order of each sub-region sequentially along the direction of the fault current; At the splicing points of adjacent feeder segment sub-regions, voltage continuity constraints, power balance constraints, and operation sequence continuity constraints are constructed. The constraints at each splicing point are combined with the node voltage constraints, branch thermal stability constraints, and distributed power output constraints within each sub-region to form a set of linear inequalities. According to the solution order, the transfer path skeleton of each sub-region is substituted into the linear inequality set as the initial solution in turn, and the solution is performed with the goal of minimizing the number of switching operations and maximizing the recovery load, so as to obtain the transfer path of each sub-region. The supply transfer paths of each sub-region are spliced ​​together according to the solution order to generate the target cross-regional fault supply transfer scheme.

10. A scenario-adaptive power distribution network fault handling device, characterized in that, The scenario-adaptive distribution network fault handling method according to any one of claims 1 to 9 includes: The scenario pattern code generation module is used to perform nonlinear dimensionality reduction on the collected power distribution network operation status data and external environmental parameters through stacked autoencoders to generate corresponding scenario pattern codes. The fault handling process matching module is used to match and dynamically arrange fault handling processes from a preset process template library based on the scenario pattern code. The handling status and risk probability prediction module is used to encode the intermediate status information generated in the current handling stage into a handling status feature vector during the execution of the fault handling process, and use the gated loop unit to iteratively predict the handling status and risk probability of the next stage. The reduced-order approximation matrix determination module is used when the process reaches the transfer analysis stage. It divides the distribution network topology into several feeder segment sub-regions according to key nodes, and performs truncated singular value decomposition on the node and branch association matrix of each sub-region to obtain a reduced-order approximation matrix, so as to determine the maximum transferable capacity and transfer path skeleton of each sub-region. The fault transfer scheme generation module is used to construct a set of linear inequalities under the constraints of voltage continuity, power balance and operation sequence continuity between adjacent sub-regions, and to solve the fault transfer path skeleton of each sub-region sequentially by substituting it into the set of inequalities according to the fault current direction, and to splice them together to generate the target cross-region fault transfer scheme.

Images