A method for automatically checking operation mode of urban power distribution network

Abstract

The application relates to the technical field of data processing and computer simulation, and discloses a kind of urban distribution network operation mode automatic checking method, comprising: the mixed operation section is formed by combining the corrected current electrical quantity with target switch state;Based on the section output online branch embedding vector, the safety margin measure value is calculated using support vector machine to screen dynamic N-1 fault set;For the branch in fault set, simulate opening and closing target tie switch to form single-point supply fault scene topology, update node impedance matrix and extract static voltage-reactive control sensitivity operator;Based on the updated matrix, perform power flow calculation to calculate voltage out-of-limit deviation, combine the operator to limit the inverter reactive power output and update the active power output, and continue iteration until the accuracy or interception is met;Finally, according to the checking result, execute action sequence or lockout. The application excludes redundant fault branches to reduce calculation consumption, truly reflects the standby transfer operation state, and takes into account the equipment capacity limit to avoid overrunning.

Description

Technical Field

[0001] This invention relates to the field of data processing and computer simulation technology, specifically to an automatic verification method for the operation mode of urban power distribution networks. Background Technology

[0002] Before performing operations such as load transfer, fault recovery, and operation mode adjustment, urban power distribution networks need to conduct safety verification of the proposed operation mode due to factors such as the distribution network topology, bidirectional power flow distribution, distributed power source access, and equipment load limits. In the conventional verification process, the system temporarily suspends the issuance of switching action commands when generating a target topology request. It constructs a state to be verified based on real-time data to simulate fault scenarios. Only when it is confirmed that the voltage of each node is within the safe range can the verification pass signal be returned and the target action be executed; otherwise, the corresponding commands are blocked to avoid operational risks such as voltage exceeding limits.

[0003] In existing safety assessment data processing mechanisms, traditional methods typically require establishing simulation sections for each operable branch of the entire network and solving nonlinear power flow equations. As the scale of the distribution network expands, this indiscriminate full-scale simulation scanning generates a massive amount of computational data, leading to severe consumption of computing resources and making it difficult to meet the real-time requirements of online assessment data. Furthermore, some data-driven risk identification algorithms, when dealing with high-dimensional electrical features, suffer from poor generalization ability in their output assessment data due to the lack of equivalent physical constraints. Simultaneously, existing power flow solving algorithms are prone to iterative calculation divergence due to matrix rank reduction when simulating extreme conditions such as topology decoupling, resulting in unstable assessment results. Therefore, a data processing method is needed that can improve the efficiency of data dimensionality reduction and filtering, stably solve the simulation matrix, and generate high-precision assessment decision data. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an automatic verification method for urban power distribution network operation modes. It aims to solve the problems in existing technologies, such as the high computational power consumption and slow response caused by indiscriminate nonlinear power flow iterative scanning of all fault scenarios, the low data processing efficiency caused by relying on repeated matrix inversion when simulating backup power transfer topology changes, and the distortion of coordinated control iterative data caused by exceeding the physical hardware capacity limit of inverters in numerical simulations, which can easily lead to misjudgment and false blocking of system decisions.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an automatic verification method for the operation mode of an urban power distribution network, comprising:

[0006] The corrected current electrical quantities are combined with the target switch states in the target topology to form a hybrid operation profile; Based on the online branch embedding vector output from the hybrid operation section, the safety margin metric is calculated using the support vector machine safety boundary to filter the dynamic N-1 fault set. For the fault branches in the dynamic N-1 fault set, simulate the opening and closing of the target interconnection switch to form a single-point power transfer fault scenario topology, update the node impedance matrix and extract the static voltage-reactive power control sensitivity operator. Power flow calculation is performed based on the updated node impedance matrix. The voltage over-limit deviation is calculated and combined with the static voltage-reactive power control sensitivity operator to limit the reactive power output of the inverter and update the active power output. The process is iterated until active interception is triggered or the preset convergence accuracy threshold is met. Based on the safety verification results of all fault scenarios, execute the switch action sequence corresponding to the target topology or implement switch action sequence interlocking.

[0007] In one specific embodiment, the step of combining the corrected current electrical quantity with the target switch state in the target topology to form a hybrid operation profile includes: collecting the active power injection, reactive power injection, voltage amplitude, current switch state, and real-time output and apparent power upper limit of the inverter at the current moment; unifying the positive and negative directions of power and generating historical current carrying characteristics of branches; performing filtering judgment and missing measurement completion on real-time measurements to obtain the corrected current electrical quantity; and combining the corrected current electrical quantity with the target switch state in the target topology to form the hybrid operation profile.

[0008] Preferably, the step of performing filtering and missing measurement completion on real-time measurements includes: selecting an odd number of consecutive historical sample values ​​to calculate the median; determining data abnormality when the absolute deviation between the current sample value and the median exceeds a relative threshold; replacing the current sample value with the value of the previous valid time or correcting it by interpolation; generating pseudo-measurements by combining historical load curves and meteorological data, or directly using virtual measurements output by state estimation to complete the electrical parameters of nodes without real-time measurement devices.

[0009] In one specific embodiment, the step of outputting the online branch embedding vector based on the hybrid operation section includes: extracting branch node features and edge features based on the hybrid operation section; extracting normalized attention coefficients and performing feature aggregation on a graph attention network with fixed parameters to obtain a global branch node feature matrix; inputting the global branch node feature matrix into a multilayer perceptron with fixed weights to generate a soft assignment matrix; and performing spatial coarse-grained pooling based on the soft assignment matrix to output the dimensionality-reduced online branch embedding vector.

[0010] Preferably, the extraction of branch node features and edge features includes: mapping the original distribution network under the target topology to a dual graph, mapping the branches sharing the bus in the original distribution network to undirected edges between adjacent nodes in the dual graph; performing standardization processing on the active power, reactive power, voltage amplitude, current carrying capacity fluctuation variance, and historical overload frequency of the branches, and splicing them to form initial branch node features; the extraction of normalized attention coefficients and the performance of feature aggregation include: setting the neighborhood node set to include the target node itself, using independent attention heads to calculate the feature weights of the neighborhood branches in parallel, and using residual connections to complete the node feature update.

[0011] In one specific embodiment, the step of using a support vector machine to calculate a safety margin metric to filter a dynamic N-1 fault set includes: inputting the online branch embedding vector into a safety classification function to obtain the safety margin metric for the corresponding branch; adding the corresponding branch to the dynamic N-1 fault set when the safety margin metric is greater than zero, or when the safety margin metric is less than or equal to zero and greater than or equal to a negative fuzzy margin threshold; and removing the corresponding branch from the set to be precisely verified when the safety margin metric is less than the negative fuzzy margin threshold. The fuzzy margin threshold is calculated by reading the historical signal-to-noise ratio distribution sequence to calculate the maximum noise variance, combined with the maximum load fluctuation rate of the power grid.

[0012] Preferably, when establishing the support vector machine safety boundary, the network weights are trained using a joint loss function in the offline stage, and physical constraints are constructed by calculating the deviation between the output equivalent parameters and the Thevenin equivalent parameters of the real system; the support vector machine is used to perform soft-margin convex optimization to solve the classification hyperplane, and the solution results are solidified into the safety boundary hyperplane and normal vector and bias term parameters.

[0013] In one specific embodiment, the step of simulating the opening and closing of the target tie switch for the fault branch within the dynamic N-1 fault set to form a single-point power transfer fault scenario topology includes: for the undervoltage region formed by the simulated opening of the fault branch within the dynamic N-1 fault set, setting the end node of the fault branch closest to the undervoltage side as the root node of the breadth-first search and adding it to the breadth-first search queue; expanding outward layer by layer according to the distribution network connection topology, when the end of a search branch reaches the tie switch, adding the corresponding tie switch and the connected opposite node to the candidate tie switch set; comparing the power supply margin of the opposite feeder of each tie switch in the candidate tie switch set with the total load of the undervoltage region, and eliminating candidate tie switches whose power supply margin is not greater than the total load of the undervoltage region; arranging the remaining candidate tie switches that meet the load transfer requirements in descending order according to the power supply margin, determining the candidate tie switch ranked first as the target tie switch and simulating its closure to form the single-point power transfer fault scenario topology.

[0014] Preferably, the step of expanding outward layer by layer according to the distribution network connection topology includes: shielding the direction of the faulty branch that has been disconnected; extracting the difference between the apparent power safety limit of the feeder on the opposite side of the candidate tie switch and the current actual apparent power carried as the power supply margin; when the set of remaining candidate tie switches is empty after elimination, generating a transfer failure flag and terminating the reactive power control process of the current fault scenario.

[0015] In one specific embodiment, updating the node impedance matrix and extracting the static voltage-reactive power control sensitivity operator includes: generating the ground-state node impedance matrix corresponding to the target operating mode; using the Sherman-Morrison formula combined with admittance matrix perturbation, quickly calculating the updated node impedance matrix corresponding to the fault scenario based on the ground-state node impedance matrix; and extracting the imaginary part slice of the updated node impedance matrix to obtain the static voltage-reactive power control sensitivity operator.

[0016] Preferably, the calculation of the updated node impedance matrix corresponding to the fault scenario further includes: calculating the denominator scalar of the correction term generated by the perturbation decomposition variable of the admittance matrix; when the absolute value of the denominator scalar is less than the anti-overflow tolerance constant, it is determined that there is a risk of network dissociation; the matrix inversion algorithm is called to recalculate the node impedance matrix; and when the admittance matrix is ​​reduced in rank, a virtual grounding conductance parameter is added to the diagonal elements.

[0017] In one specific embodiment, the step of performing power flow calculation based on the updated node impedance matrix to calculate the voltage over-limit deviation includes: performing a Z-bus power flow iteration operation to obtain node voltage complex vectors based on the balanced node voltage vectors and the updated node impedance matrix; extracting the voltage amplitude of each node according to the node voltage complex vectors; when a node voltage amplitude is lower than the node voltage safety lower limit, using the node voltage safety lower limit as the target voltage of the corresponding over-limit node; when a node voltage amplitude is higher than the node voltage safety upper limit, using the node voltage safety upper limit as the target voltage of the corresponding over-limit node; and calculating the deviation between the target voltage and the voltage amplitude obtained by the current Z-bus iteration as the voltage over-limit deviation.

[0018] Preferably, before performing the Z-bus power flow iteration operation, the node power and voltage data involved in the calculation are converted to a per-unit system; during the iteration operation, the node injection current is calculated by complex voltage and conjugate complex number; if the magnitude of the complex voltage conjugate complex number is lower than the preset tolerance threshold, the current scenario calculation is terminated and an undervoltage alarm is output; the node voltage safety lower limit is 0.95 per-unit value, and the node voltage safety upper limit is 1.05 per-unit value.

[0019] In one specific embodiment, the step of limiting the reactive power output of the inverter and updating the active power output by combining the static voltage-reactive power control sensitivity operator includes: when the absolute value of the voltage over-limit deviation is greater than the preset convergence accuracy threshold, calculating the theoretical reactive power increment demand by combining the voltage over-limit deviation and the sensitivity coefficient of the current over-limit node relative to the reactive power output of the inverter extracted from the static voltage-reactive power control sensitivity operator; performing a limiting operation on the reactive power command according to the theoretical reactive power increment demand and the apparent power capacity constraint of the inverter to determine the new reactive power output; calculating the upper limit of the active power output allowed under the current reactive power output condition according to the apparent power capacity circle of the inverter; and updating the active power output of the inverter according to the upper limit of the allowed active power output.

[0020] Preferably, when extracting sensitivity coefficients for division or inverse mapping operations, if the absolute value of the coefficient is detected to be less than a preset tolerance threshold, a constant bias term is added to the denominator to replace the original value to prevent overflow during division by zero.

[0021] In one specific embodiment, the continuous iteration until active interception is triggered or a preset convergence accuracy threshold is met includes: using the converted node injection amount as the input for the next round of Z-bus power flow iteration; when the absolute value of the voltage over-limit deviation is detected to be no greater than the preset convergence accuracy threshold, directly locking the final inverter active power control command and reactive power control command; when the inverter capacity reaches the upper limit and the voltage over-limit deviation still cannot be eliminated, immediately triggering the active interception; if the iteration counter reaches the maximum allowable iteration limit and the convergence condition is still not met, synchronously triggering the active interception.

[0022] Preferably, a fixed maximum allowable number of iterations is set to prevent the continuous divergence caused by the ill-conditioned Jacobian matrix of the nonlinear power flow iterative equation.

[0023] In one specific embodiment, the step of executing the switching action sequence corresponding to the target topology or implementing the switching action sequence interlocking based on the safety verification results of all fault scenarios includes: marking the corresponding fault scenario as safe when the voltage amplitude of all nodes in a specific fault scenario simultaneously satisfies the node voltage safety judgment inequality; when it is detected that all scenarios in the set are marked as safe, returning the target operation mode verification pass signal and executing the switching action sequence corresponding to the target topology; and returning the target operation mode verification fail signal and implementing the switching action sequence interlocking as long as any scenario in the dynamic N-1 fault set triggers the set active interception mechanism.

[0024] Preferably, when executing the switching action sequence, a control message containing the power setting value is sent to the underlying device to perform terminal drive duty cycle adjustment; when implementing the interlock, the output path of the circuit breaker control command for the target topology is cut off, and an alarm message containing the fault scenario identifier and the maximum over-limit deviation value is pushed to the dispatch monitoring terminal.

[0025] This invention provides an automatic verification method for the operation mode of urban power distribution networks. It has the following beneficial effects: 1. This invention extracts high-dimensional features from hybrid operation sections and maps them to output online branch embedding vectors. It then uses support vector machine safety boundaries and fuzzy margin thresholds to calculate safety margin metrics, achieving low-dimensional dimensionality reduction of high-dimensional topological data and rapid screening of high-risk dynamic N-1 fault sets. This mechanism preemptively eliminates redundant fault branches in deep safety regions at the digital data processing level, avoiding indiscriminate nonlinear power flow iterative scanning of all open / closed scenarios from the source. This reduces CPU computing power consumption and dynamic memory usage during online verification of the computing system, and shortens data processing response time.

[0026] 2. This invention performs a breadth-first search of the undervoltage region and reconstructs the single-point power transfer topology for the selected dynamic N-1 fault set. It uses the Sherman-Morrison formula combined with admittance matrix perturbation to directly update the ground-state node impedance matrix quickly. This algorithm simplifies the repeated high-order complex matrix inversion operations required in traditional topology changes to low-dimensional vector multiplication operations, significantly reducing computational complexity. Combined with the extracted imaginary part slice of the static voltage-reactive power control sensitivity operator, it improves the convergence speed of subsequent Z-bus nonlinear power flow iteration calculations and the execution efficiency of the system program.

[0027] 3. In the Z-bus power flow iteration process, this invention combines the control sensitivity operator with the inverter's apparent power capacity circle constraint to achieve coordinated control data calculation for updating the inverter's reactive power limit and active power output upper limit. This overcomes the problem of calculation result distortion caused by traditional digital simulation ignoring the capacity limits of the underlying hardware. Based on the safety verification results of all fault scenarios, the system automatically issues control messages or generates blocking commands, improving the data processing reliability and safety control capabilities of the distribution network automatic verification system. Attached Figure Description

[0028] Figure 1 This is a system architecture diagram according to an embodiment of the present invention.

[0029] Figure 2 This is a general flowchart of an automatic verification method for the operation mode of an urban power distribution network according to an embodiment of the present invention.

[0030] Figure 3This is a schematic diagram of the transformation from the original distribution network to the dual diagram according to an embodiment of the present invention. Detailed Implementation

[0031] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] See attached document Figure 1 The present invention provides an automatic verification system for the operation mode of urban power distribution network, which may include a cross-section construction module, a fault screening module, a topology reconstruction module, a coordinated regulation module, and a decision control module.

[0033] The cross-section construction module is used to collect real-time electrical quantities and inverter operating parameters of distribution network nodes, and combine them with the historical current carrying characteristics of branches and the topology of the target to be executed to form a hybrid operation cross-section.

[0034] The fault screening module is used to convert the target topology into a dual graph, extract node features and edge features, and input them into a jointly trained graph attention network and pooling layer to generate branch embedding vectors. Combined with the safety boundary fitted by support vector machine, a dynamic N-1 fault set is screened.

[0035] The topology reconfiguration module is used to simulate branch interruption within a dynamic N-1 fault set, perform undervoltage region search and determine the target tie switch, form a backup power transfer topology, and update the node impedance matrix corresponding to the fault scenario.

[0036] The coordinated control module is used to perform Z-bus nonlinear power flow iterative calculations based on the updated node impedance matrix, and generate coordinated control decision data for the active and reactive power output of the inverter under the apparent power capacity constraint of the inverter based on the voltage over-limit deviation.

[0037] The decision control module is used to output a pass or fail signal for the target operating mode based on the voltage safety verification results of each fault scenario, and generate safety assessment data to indicate the execution or blocking of the corresponding switch action sequence.

[0038] Before the actual adjustment of the urban power distribution network operation mode triggers online verification, the method of this invention first performs offline construction and training of the graph neural network model and safety boundary by the fault screening module in the aforementioned system, providing a physical constraint basis for online fault screening. The specific offline construction and training process includes: Extract historical operational section data and perform the conversion from the original distribution network to its dual diagram. Obtain multiple sets of historical operational section data of the distribution network as historical samples, and extract the branch electrical parameters and historical power flow status from them. (See attached diagram) Figure 3 The original distribution network branches, such as... Figure 3 Branches e1, e2, e3, and e4 in the graph are mapped to nodes in the dual graph, i.e., appended nodes. Figure 3 The nodes e1, e2, e3, and e4 in the graph map the branches that share a bus in the original distribution network to the edges between adjacent nodes in the dual graph. For example, if branch e1 and branch e2 share node 2, then an adjacent edge is established between node e1 and node e2 in the dual graph; while branch e2 and branch e4 do not share a bus node, then they are represented as non-adjacent edges in the graph. This completes the topology transformation of the original distribution network to the dual graph, and the extracted branch electrical parameters and historical power flow status are used as the node features of the dual graph.

[0039] A joint loss function is used to train the graph attention network and pooling layers. The graph attention network and pooling layers are constructed by inputting the features of the dual graph nodes into the graph attention network for feature aggregation, followed by spatial dimensionality reduction through the pooling layer to output branch embedding vectors. During training, the deviation between the network's output equivalent parameters and the Thevenin equivalent parameters of the real system is calculated to construct a Thevenin equivalent physical penalty term; that is, the mean square error between the predicted equivalent impedance values ​​of the dual graph boundary nodes and the real impedance values ​​derived using Thevenin's theorem is used as a physical constraint for network training. This physical penalty term is weighted and summed with the InfoNCE contrastive loss, which enhances the distinction between positive and negative sample features, to construct the joint loss function. The network weights of the graph attention network and pooling layers are updated using the backpropagation algorithm until the joint loss function converges, resulting in the trained model.

[0040] A support vector machine (SVM) is used to fit the safety boundary hyperplane through soft-spacing convex optimization. The embedding vectors of the actual branches corresponding to historical operating sections are used as input, combined with safety state labels indicating whether node voltages exceed limits under actual operating conditions. SVM is then used to perform soft-spacing convex optimization to solve for the classification hyperplane. The solved classification hyperplane is used as the branch safety boundary to delineate the safety domain and the fault domain. After completing the offline training of the model and the construction of the safety boundary hyperplane, the model can be deployed in the main process of the system for online real-time verification.

[0041] See attached document Figure 2 This invention provides an automatic verification method for the operation mode of an urban power distribution network, executed through the aforementioned system, comprising the following steps: S1, construct a hybrid operation profile; collect the active power injection, reactive power injection, voltage amplitude, current switch status, and real-time output and apparent power limit of the inverter at the current moment, unify the positive and negative directions of power and generate the historical current carrying characteristics of the branch, perform filtering judgment and missing measurement completion on the real-time measurements, and combine the corrected current electrical quantities with the target switch status in the target topology to form a hybrid operation profile.

[0042] S2, Filter the dynamic N-1 fault set; Convert the original distribution network under the target topology into a line graph to generate a dual graph, extract branch node features and edge features based on the hybrid operation section, extract normalized attention coefficients and perform feature aggregation by inputting a graph attention network with fixed parameters, perform spatial coarse-grained mapping through a multilayer perceptron to output online branch embedding vectors, calculate the safety margin metric value using a support vector machine safety boundary, and filter the dynamic N-1 fault set based on the interval where the safety margin metric value is located and the fuzzy margin threshold.

[0043] S3, reconstruct the fault scenario topology; generate the ground state node impedance matrix corresponding to the target operating mode; simulate the interruption of fault branches in the dynamic N-1 fault set; perform a breadth-first search to extract the candidate tie switch set with the end node on the undervoltage side as the root node; compare the power supply margin of each tie switch to the feeder on the opposite side with the total load of the undervoltage area to determine the target tie switch; simulate closure to form a single-point transfer fault scenario topology; use the Sherman-Morrison formula combined with admittance matrix perturbation to update the node impedance matrix; and extract the static voltage-reactive power control sensitivity operator.

[0044] S4 executes Z-bus power flow iteration and inverter output coordinated control data calculation; based on the updated node impedance matrix, performs Z-bus nonlinear power flow calculation to obtain node voltage; when there is a node voltage exceeding the limit, calculates the voltage exceeding the limit deviation; combines the static voltage-reactive power control sensitivity operator to calculate the theoretical reactive power increment demand; under the inverter's apparent power capacity constraint, limits the reactive power output and simultaneously updates the allowable active power output upper limit and active power output, thereby generating inverter reactive power output limiting and updated active power output control decision data; converts the updated active and reactive power output into node injection quantities and continuously performs Z-bus power flow iteration; when the voltage exceeding the limit deviation reaches the preset convergence accuracy, locks the inverter control decision data; when the inverter capacity reaches the upper limit and the voltage exceeding the limit is still not eliminated, or when the maximum allowable iteration number is reached, triggers active interception.

[0045] S5, output safety assessment result data; perform traversal verification on the dynamic N-1 fault set. When the node voltages of all fault scenarios are within the preset safety range, output the safety assessment result data corresponding to the target topology. The safety assessment result data is used to indicate the execution of the switching action sequence and simultaneously send the locked inverter control decision data; when any fault scenario triggers the active interception mechanism, output the safety assessment result data corresponding to the target topology. The safety assessment result data is used to indicate the implementation of the switching action sequence blocking and push alarm information containing over-limit deviation.

[0046] To make the technical solution of the present invention clearer, the internal operating logic and mathematical principles of each of the above steps are explained in detail below in conjunction with the specific implementation process.

[0047] The network model and security boundaries constructed in the aforementioned offline phase establish a judgment benchmark for online fault screening. When implementing online verification, the automatic verification system needs to obtain fused data of the current actual electrical state and the proposed topology as a unified calculation input. Based on the working logic of the cross-section construction module, the process of constructing the hybrid operation cross-section in step S1 specifically includes the following sub-steps: S11 collects real-time electrical quantities and inverter operating parameters. The data acquisition and monitoring control system collects the active power injection, reactive power injection, and voltage amplitude of each node in the urban distribution network at the current moment, collects the current switching status of each branch, and obtains the real-time active power output, reactive power output, and apparent power upper limit of each distributed inverter. For the output of distributed photovoltaic and wind power, it can be included in the node active power injection as a negative value of the load, or it can be measured as an independent power source and included in the node injection power.

[0048] S12: Unify the positive and negative directions of power and generate historical current-carrying characteristics. Before performing subsequent power flow calculations, the load power, distributed generation output, and inverter output are uniformly converted according to the node injection power sign rules adopted by the distribution management system, so that the node active power injection, node reactive power injection, and node injection current have a consistent positive and negative direction. Simultaneously, the historical database of the distribution management system is accessed to obtain the time-series data of current-carrying capacity and historical heavy load frequency for each branch of the entire network over the past 30 days. Historical current-carrying characteristics are generated based on the historical time-series data of current-carrying capacity and historical heavy load frequency of the branches.

[0049] S13: Perform filtering and missing measurement completion on real-time electrical quantities. For each measurement point, select an odd number of consecutive historical sample values ​​and calculate the corresponding median. When the absolute deviation between the current sample value and the median exceeds a preset relative threshold, the current sample value is determined to be abnormal. Considering the short-term fluctuation characteristics of normal load in the distribution network, the aforementioned preset relative threshold is set to 3% to 5%. When an anomaly is determined, the value of the previous valid time point is used to replace the current sample value, or interpolation is used to correct the current sample value. For nodes without real-time measurement devices, including low-voltage load nodes and distributed power grid-connected nodes without real-time measurement devices, pseudo-measurements are generated based on historical load curves and meteorological data, or virtual measurements output by the distribution network state estimation module are directly used as data supplementation.

[0050] S14. Generate a hybrid operation profile by combining the target topology switch states. Read the target topology corresponding to this operation mode verification request and obtain the target states of each branch switch under the target topology. Combine the current node active power injection, node reactive power injection, and node voltage amplitude (corrected in the previous steps) with the target switch states corresponding to the target topology to form a hybrid operation profile. Considering that the operation mode adjustment command has not yet been actually executed, the node source load power will not undergo synchronous abrupt changes with the switch states of the target topology. The node active power injection, node reactive power injection, and node voltage amplitude in the hybrid operation profile all use the corrected current real-time values, while the branch switch states use the target states in the target topology.

[0051] The hybrid operation profile generated through the above steps simultaneously reflects the current real source-load operating conditions and the target topology to be executed, realizing an accurate simulation of the target operation mode before the actual execution of the target switching action, and providing a common data foundation for subsequent dynamic N-1 fault screening, fault scenario topology reconstruction and power flow calculation.

[0052] After obtaining the hybrid operation profile constructed in step S1, to avoid performing nonlinear power flow iterative scanning on all branch disconnection scenarios within the distribution network, the automatic verification system performs spatial dimensionality reduction by extracting the mapping relationship of the trained network model and combines boundary constraints to complete the preliminary screening of high-risk fault scenarios. The specific implementation of step S2 includes the following sub-steps: S21, extract the topological correlation matrix based on the hybrid operation section and construct the initial features of the dual graph. (See appendix) Figure 3 The original distribution network under the target topology is transformed into a line diagram. The urban distribution network corresponding to the target topology is represented as the original distribution network diagram: ; in: : Original distribution network diagram under the target topology; : The original set of busbar nodes in the distribution network, such as Figure 3 Nodes 1 to 5 are shown in the diagram; : The original distribution network branch collection, such as Figure 3 Branches e1 to e4 are shown.

[0053] The original distribution network is converted into a line diagram, and the dual diagram is obtained through the dual diagram conversion formula: ; in: : The dual diagram corresponding to the original distribution network; The set of nodes in the dual graph, where each branch in the original distribution network corresponds to a node in the dual graph, such as... Figure 3 The dual graph formed after transformation consists of nodes e1 to e4. Dual graph edge set: When two branches in the original distribution network are connected to the same busbar, dual graph edges are established between the corresponding two dual graph nodes; for example... Figure 3 As shown, in the original graph, branch e2 and branch e3 are connected by an adjacent dual edge because they share node 3; in the original graph, branch e1 and branch e4 are connected by an adjacent dual edge because they share node 2.

[0054] Establish a node-branch association matrix based on the target operating mode and target topology: ; in: Node-branch association matrix; : The number of busbar nodes in the original distribution network; The number of operable branches in the original distribution network; : OK The space of columnar real matrix.

[0055] Using the node-branch incidence matrix, the adjacency matrix of the dual graph can be obtained through the formula for calculating the adjacency matrix of the dual graph: ; in: : Adjacency matrix of dual graph; Node-branch association matrix transpose; Branch correlation matrix obtained based on node-branch association; :Depend on A diagonal matrix formed by the diagonal elements.

[0056] Based on the mixed operation section, the active power, reactive power, and voltage amplitude per unit value of the first and last busbars of each branch are extracted as the static electrical characteristics of the corresponding branch's corresponding diametrical diagram nodes. The historical current carrying capacity time-series data of each branch over the past 30 days are retrieved to calculate the variance of the corresponding branch's current carrying capacity fluctuation and to obtain the historical heavy load frequency of the branch.

[0057] To eliminate scale differences in data with different dimensions during the forward propagation of the neural network, the active and reactive power, per-unit voltage amplitude, current carrying capacity fluctuation variance, and historical overload frequency of the branch's head and tail buses are Z-score standardized according to the branch dimension. The standardized multi-dimensional values ​​are then concatenated into vectors to form the initial branch node features of each branch node in the dual graph.

[0058] By combining real-time cross-sectional estimation results, branch conductance, branch susceptance, and inter-node voltage phase angle difference are extracted, and the edge features of the dual graph are formed through the edge feature vector formula: ; in: Dual graph nodes With dual graph nodes Edge features between; Branch circuit conductivity; Branch circuit susceptance; :node ,node Phase angle difference between voltages; , : Dual graph node index.

[0059] S22, calculate the normalized attention coefficients and perform feature aggregation using the trained network weights. Input the aforementioned branch node features and dual graph edge features into the graph attention network whose parameters have been fixed in the offline stage. In the dual graph space, let the dual graph node corresponding to the target branch be a node. ,node The set of neighboring nodes is represented as .

[0060] To prevent isolated nodes from occurring due to power grid outages, a neighborhood node set is defined. Internal forced inclusion node This ensures that the denominator of subsequent attention calculations is always greater than zero.

[0061] use Each independent attention head computes the influence weights of neighboring branches on the target branch in parallel. In the first... In each attention head, neighboring nodes Relative to the target node The attention coefficient is obtained using the normalized attention coefficient calculation formula: ; in: For the first Neighboring nodes in each attention head For the target node The attention coefficient is used to characterize the sensitivity of the target branch to power flow transfer caused by changes in the state of the neighboring branch; For the first The feedforward neural network weight vector is already fixed in the attention head; For the first The fixed node feature transformation weight matrix in each attention head; For the first The fixed edge feature mapping matrix in each attention head; For the target node Branch node characteristics; Representing neighboring nodes Branch node characteristics; For normalized calculation of neighboring nodes Branch node characteristics; For nodes and nodes Edge features between; For nodes and nodes Edge features between; This represents the concatenation operation of feature vectors in the hidden layer dimension; The activation function is a modified linear unit with leakage coefficient, and the slope of the negative half-axis is fixed at 0.2. For nodes The set of neighboring nodes; Indicates the attention head index; For the neighboring node index.

[0062] The features of neighboring branch nodes are weighted and aggregated based on the normalized attention coefficients, and the features are updated using the residual connection through the branch node feature update formula: ; in: The nodes output by the attention network layer in the previous diagram. Branch node characteristics; : The node after the current network layer update Branch node characteristics; Exponential linear unit activation function; : Pay attention to the number of heads; :right The feature aggregation results corresponding to each attention head are averaged. : For nodes Perform a weighted summation based on the features of the neighboring nodes; : No. Nodes in each attention head For the target node The resulting weighted feature contribution.

[0063] After forward propagation iteration processing using a multi-layer graph attention network, the feature vectors generated by each branch node in the entire network are arranged and concatenated according to the node index to obtain the global branch node feature matrix. ; S23 performs spatial coarse-grained mapping based on a multilayer perceptron and outputs online branch embedding vectors.

[0064] The global node feature matrix is ​​further input into the multilayer perceptron with fixed weights, and a soft assignment matrix is ​​generated using the soft assignment matrix calculation formula: ; in: : Soft assignment matrix, representing the probability distribution of a single physical branch mapped to each fault primitive in the reduced-dimensional space; The number of branches in the original distribution network; The number of equivalent fault primitives, and satisfying the following conditions: ; : Multilayer perceptron nonlinear mapping; : Softmax normalization mapping function; : The global node feature matrix output by the graph attention network.

[0065] Spatial coarse-grained pooling is performed based on the soft allocation matrix, and the dimensionality-reduced online branch embedding vector is output through an equivalent feature mapping operation: ; in: : An online branch embedding vector matrix containing the dimensionality reduction features of all branches in the entire network; Soft allocation matrix The transpose of .

[0066] From the matrix By extracting the corresponding vector by row index, the node can be obtained. Online embedding vectors under the current hybrid operation section .

[0067] S24, use support vector machine to calculate the safety margin metric and filter the fault scenario set.

[0068] The parameters of the safe boundary hyperplane, obtained by soft-margin convex optimization using support vector machine in the offline phase, are obtained. The safe classification function is expressed as follows: ; in: : The security classification function corresponding to the branch embedding vector; : Branch low-dimensional embedding vector; : Fixed safety boundary hyperplane normal vector; : Fixed safety boundary hyperplane bias term.

[0069] The extracted online branch embedding vectors are input into the safety classification function, and the safety margin metric value of the corresponding branch is obtained using the safety margin metric calculation formula: ; in: : No. The safety margin measure of a branch relative to the safety boundary, its positive or negative state characterizes the theoretical tendency of the system voltage to exceed the limit after the branch is disconnected, and the magnitude of the value characterizes the high-dimensional distance from the safety critical surface. : No. The online embedding vector corresponding to the current mixed operation section of each branch.

[0070] Determine the numerical range of the safety margin metric and perform tiered filtering: When satisfied When, it indicates the first If the online characteristics of a branch are located on the over-limit side of the safety boundary, the corresponding branch will be added to the dynamic N-1 fault set.

[0071] When satisfied When, it indicates the first The branch is located on the safety side, but within the fuzzy margin region near the safety boundary. To prevent misjudgment of feature offset caused by measurement noise in the data acquisition and monitoring control system, the corresponding branch is also added to the dynamic N-1 fault set.

[0072] in: : Fuzzy margin threshold. The maximum noise variance is calculated by reading the historical measurement signal-to-noise ratio distribution sequence, and then converted to the embedding vector space by combining the maximum load fluctuation rate of the power grid to form the fuzzy margin threshold.

[0073] When satisfied When the distance exceeds the absolute safety depth threshold, it indicates that the branch is located on the safe side and the distance from the safety boundary exceeds the absolute safety depth threshold. At this time, the branch characteristics are in the deep safety region, and the probability of misjudging the state under the current measurement noise level is close to zero. Therefore, the corresponding branch is removed from the set to be accurately verified.

[0074] Based on the above-mentioned numerical range judgment conditions, all branch combinations that satisfy the retention rules are extracted to form a dynamic N-1 fault set. This provides a high-risk scenario target after dimensionality reduction and convergence for subsequent topology reconstruction and accurate verification steps, while avoiding full nonlinear power flow iteration.

[0075] After obtaining the dynamic N-1 fault set output in step S2, perform dynamic search of the underpressure area and backup power supply topology reconfiguration for each fault scenario in sequence.

[0076] The specific implementation of step S3 includes the following sub-steps: S31, perform a breadth-first search of the underpressure area and extract a set of candidate contact switches.

[0077] The dual graph nodes corresponding to each branch in the dynamic N-1 fault set are inversely mapped back to the actual fault branches in the original distribution network, simulating the corresponding fault branches in an open state in the computational model. For the undervoltage region formed by the open fault branch, the terminal node of the fault branch closest to the undervoltage side is set as the root node of the breadth-first search, and the root node is added to the breadth-first search queue. The search process expands outward layer by layer according to the distribution network connection topology. To avoid backtracking or the formation of deadlocks in the search path, the direction of the already open fault branch is forcibly blocked during the search process.

[0078] When the end of a search branch reaches a tie switch, the corresponding tie switch and its connected opposite node are added to the candidate tie switch set, and the current search branch stops expanding outward. The process continues scanning the remaining search branches that have not yet reached the topology boundary until the search of all topology ends in the underpressure region is completed, obtaining the candidate tie switch set around the underpressure region.

[0079] S32, calculate the power supply margin and execute the target tie switch decision. Analyze the composition of the candidate tie switch set. When the candidate tie switch set contains only a single available tie switch, directly determine the selected tie switch as the target tie switch. When the candidate tie switch set includes multiple available tie switches, extract the apparent power safety limit and the current actual apparent power carried by the feeder opposite to each candidate tie switch, and calculate the target tie switch using the power supply margin calculation formula. Power supply margin of each candidate interconnecting switch to the side feeder: ; in: : No. The power supply margin of the feeder on the opposite side of each candidate tie switch is in MVA. : No. The apparent power safety limit for each candidate interconnecting switch opposite the feeder output circuit breaker or main transformer is in MVA. : No. The apparent power currently carried by the feeder of each candidate tie switch, in MVA; Candidate contact switch index; The corresponding parameter belongs to the feeder side.

[0080] The real-time load demand of all nodes within the undervoltage area is obtained and accumulated to obtain the total load of the undervoltage area. The power supply margin is compared with the total load of the undervoltage area, and candidate tie switches with a power supply margin not greater than the total load of the undervoltage area are eliminated. The remaining candidate tie switches that meet the load transfer requirements are sorted in descending order of power supply margin, and the candidate tie switch ranked first is determined as the target tie switch. If the set of remaining candidate tie switches is empty after elimination, it indicates that there is no single-point transfer channel in the system that meets the full transfer condition. At this time, a transfer failure flag is output to the topology reconfiguration module, and subsequent reactive power verification operations for the current fault scenario are terminated.

[0081] The target tie switch is closed in the computational model to form a single-point power transfer fault scenario topology that includes the opening of the fault branch and the closing state of the backup tie switch.

[0082] S33, combined with admittance perturbation, updates the node impedance matrix and extracts the control sensitivity operator.

[0083] Using the target topology and the static parameters of the entire network contained in the generated hybrid operation section, the ground state complex admittance matrix corresponding to the target operation mode is generated. It has a dimension of N×N (N is the total number of distribution network nodes) and a dimension of Siemens. It is used to reflect the set of network node admittance parameters of the distribution network under normal operating conditions.

[0084] While ensuring the ground-state complex admittance matrix is ​​full rank, inverting the ground-state complex admittance matrix yields the ground-state nodal impedance matrix: ; in: The ground-state node impedance matrix corresponding to the target topology has an N×N dimension and a dimension of Ω. Ground state complex admittance matrix The inverse matrix.

[0085] The opening of a faulty branch and the closing of the target communication switch cause changes in the global admittance matrix. The disturbance in the admittance matrix caused by the state change can be expressed as a disturbance decomposition formula: ; in: The admittance matrix disturbance caused by the opening of the fault branch and the closing of the target communication switch has a dimension of N×N. , The column vectors corresponding to the admittance matrix perturbation have a dimension of N×1 and represent the branch deployment and withdrawal relationship between specific nodes in the network topology. The vector elements correspond to Siemens units. :vector The transpose of .

[0086] The Sherman-Morrison formula is used to quickly calculate the updated node impedance matrix corresponding to the fault scenario based on the ground-state node impedance matrix: ; in: The updated node impedance matrix corresponding to the current fault scenario topology has N×N dimensions and Ω as its unit. Ground-state node impedance matrix; Impedance correction matrix terms derived from Sherman-Morrison theory; : The scalar of the denominator of the correction term.

[0087] When the denominator scalar When the absolute value is less than the overflow tolerance constant, the overflow tolerance constant is set to 10. -8 If the topology change causes the network to be at risk of being disjointed, the Sherman-Morrison formula correction path fails, and the matrix inversion algorithm is called to recalculate the node impedance matrix. When the admittance matrix is ​​reduced in rank, a virtual grounding large conductance is added to the diagonal elements to ensure the non-singularity of the matrix.

[0088] Extract the updated node impedance matrix The imaginary part slice is used to obtain the static voltage-reactive power control sensitivity operator. It has N×N dimensions and kV / Mvar units, and is used to characterize the correspondence between changes in reactive power injection at nodes and changes in node voltage, and serves as the basis for calculating subsequent theoretical reactive power increment demand.

[0089] Combining the updated node impedance matrix and static voltage-reactive power control sensitivity operator output in step S3, the system control module takes into account the inverter's apparent power capacity constraint and performs the calculation process of nonlinear power flow verification and active and reactive power coordinated control data.

[0090] The specific implementation of step S4 includes the following sub-steps: S41, Extract the updated node impedance matrix corresponding to the current fault scenario. and static voltage-reactive power control sensitivity operator The system first initializes the power flow iteration counter and sets the initial loop parameters. Read the network nodes in the current fault scenario. Initial active power injection and initial reactive power injection To ensure the consistency of numerical calculations of voltage, current, and power during power flow iteration and to avoid orders-of-magnitude deviations caused by actual value calculations, all electrical quantity parameters input to the model must be pre-converted to a per-unit value system. In the next iteration, the node The injection current is: ; in: : No. Node at the next iteration The injected current; : No. Node at the next iteration The active power injected; : No. Node at the next iteration The reactive power injected; : No. Node at the next iteration Complex voltage; :node The conjugate of a complex voltage; Imaginary unit; Node index; : Number of Z-bus power flow iterations.

[0091] In this division operation, if the conjugate of the complex voltage is... The module length is lower than the preset tolerance constant (the value can be set to 10). -6 If the system determines that the corresponding node has experienced a deep short circuit and loss of pressure, it will directly terminate the current scenario verification and output a warning that it cannot self-heal.

[0092] Based on the slack node voltage vector and the updated node impedance matrix, perform Z-bus power flow iteration operation: ; in: : No. The complex vector of node voltages obtained in the next iteration; : Balance node voltage vector; : The updated node impedance matrix corresponding to the current fault scenario topology; : No. The node injection current vector is formed in the next iteration.

[0093] The above equations use the node impedance matrix to reflect the global topology of the network and approximate the true voltage distribution of the system through iterative updates.

[0094] S42, extract the voltage amplitude of each node based on the complex vector of the node voltage. This is done when all node voltages are within a preset safe range. Internally, the current fault scenario is determined to have passed the safety check. A safety lower limit is typically set in the operating procedures. 0.95 PU, safe limit It is 1.05 pu.

[0095] When a node voltage falls below a preset safety lower limit, the node voltage safety lower limit is used as the target voltage for the corresponding node that exceeds the limit. ; When a node voltage exceeds a preset safety limit, the node voltage safety limit will be used as the target voltage for the corresponding node that exceeds the limit. ; in: The target safety boundary voltage that the over-limit node needs to achieve; Node voltage safety lower limit; Node voltage safety limit.

[0096] The deviation between the target voltage of the over-limit node and the voltage amplitude obtained from the current Z-bus iteration is calculated as the voltage over-limit deviation: ; in: : No. The node that exceeds the limit during the next iteration The corresponding voltage over-limit deviation; The target node that exceeds the limit obtained by the current Z-bus iteration. The voltage amplitude.

[0097] Determine if the absolute value of the voltage over-limit deviation is not greater than the preset convergence accuracy threshold. The preset convergence accuracy threshold can be set to 10. -4 When the convergence condition is met, the simulated values ​​of the current active and reactive power output of the inverter are locked, and the locked values ​​are used as the final output active power control decision data and reactive power control decision data.

[0098] When the absolute value of the voltage over-limit deviation exceeds the preset convergence accuracy threshold, the sensitivity coefficient of the current over-limit node relative to the reactive power output of the inverter participating in the regulation is extracted from the static voltage-reactive power control sensitivity operator. Based on the voltage over-limit deviation and the static voltage-reactive power control sensitivity operator, the theoretical reactive power increment demand is calculated: ; in: : No. In the next iteration, the out-of-limit nodes are addressed. The theoretical incremental reactive power demand obtained through calculation; : No. The node that exceeds the limit during the next iteration The corresponding voltage over-limit deviation; Over-limit nodes extracted from the static voltage-reactive power control sensitivity operator The voltage sensitivity coefficient for regulating the reactive power output of the inverter.

[0099] Because some remote nodes may have extremely low sensitivity, if the aforementioned calculations are performed, The absolute value is less than the tolerance threshold of 10. -6 The system forcibly adds a constant bias term to replace the original value to prevent data corruption caused by division by zero.

[0100] S43, Obtain the upper limit of the apparent power hardware capacity of the distributed inverters participating in voltage coordination support. Based on the theoretical incremental reactive power demand and the apparent power capacity constraint of the inverter, the reactive power regulation value is limited: ; in: The inverter's reactive power output prior to this round of adjustments; The reactive power output of the inverter after being limited by the apparent power capacity constraint of the inverter; This round of calculations is based on the theoretical reactive power increment demand obtained from voltage over-limit deviation. :node The apparent power hardware capacity limit of the distributed inverter.

[0101] After determining the new reactive power output, calculate the upper limit of the active power output allowed under the current reactive power output conditions based on the inverter's apparent power capacity circle: ; in: : No. After the adjustment, under the current reactive power output conditions, the node... The maximum allowable active power output of the inverter; Since the apparent power of hardware devices is limited by the junction temperature and current carrying capacity of power electronic devices, when the demand for reactive power support increases, the safe available space for active power output will inevitably be compressed.

[0102] Update the inverter's active power output decision value based on the allowed active power output limit: ; in: The active power output of the inverter prior to this round of adjustments; The updated inverter has active power output; : The upper limit of active power output allowed by the inverter under the current reactive power output state.

[0103] S44 recalculates the updated inverter active and reactive power outputs into node active and reactive power injections, and uses the conversion results as the node injection amounts for the next Z-bus power flow iteration. The system drives the iteration counter to continue, returning to perform the recalculation of node injection current and node voltage.

[0104] The process of continuously repeating Z-bus power flow calculation, voltage over-limit judgment, theoretical reactive power increment calculation, inverter reactive power limiting, active power capacity update, and re-performing Z-bus power flow calculation is carried out according to the established logic.

[0105] When the absolute value of the voltage over-limit deviation is not greater than the preset convergence accuracy threshold, the final active power control decision data and reactive power control decision data of the inverter are directly locked.

[0106] When the inverter capacity reaches its upper limit and the voltage deviation still cannot be eliminated, it is determined that the current fault scenario cannot be restored to a safe state through inverter control, and active interception is immediately triggered. Simultaneously, to prevent non-convergence of the nonlinear power flow equations due to ill-conditioned Jacobian matrices, the system has a maximum permissible number of iterations of 50. If the iteration counter reaches the maximum permissible number of iterations and the convergence condition is still not met, the system is treated as a grid instability, and active interception is simultaneously triggered, reporting a serious abnormality signal to the main control center.

[0107] The system receives the verification status and inverter control decision data of each scenario in the dynamic N-1 fault set output in step S4. The safety assessment data processing system summarizes the scenario status and outputs the target operation mode verification conclusion based on the judgment conditions.

[0108] The specific implementation of step S5 includes the following sub-steps: S51, the decision control module (or safety assessment data processing system) traverses the dynamic N-1 fault set. For each independent fault scenario, it extracts the total network node voltage amplitude locked after Z-bus nonlinear power flow iterative convergence and inverter collaborative control data calculation. Compliance verification is performed using the following node voltage safety judgment inequality: ; in: :node Complex voltage under final power flow conditions; :node The voltage amplitude is measured using a per-unit system consistent with power flow calculations; : Lower limit of node voltage safety, corresponding to a value of 0.95 per unit; : The upper limit of node voltage safety, which corresponds to a value of 1.05 per unit.

[0109] When the voltage amplitudes of all nodes within the topology in a single-point power transfer failure scenario simultaneously satisfy the above-mentioned inequality, the corresponding failure scenario is marked as safe. This type of verification ensures that a single topology reconfiguration scheme will not trigger local overvoltage or undervoltage risks when encountering potential equipment failures.

[0110] S52 summarizes the safety judgment indicators for all fault scenarios. When it is detected that all single-point power transfer fault scenario topologies in the dynamic N-1 fault set are marked as safe, a target operating mode verification pass signal is returned to the system control module. Based on the verification pass signal, the system generates safety assessment result data corresponding to the target topology. The safety assessment result data is used to instruct the dispatch system to remove the action restrictions on the target operating mode and guide the implementation of the corresponding switching action sequence. After outputting the assessment data, according to the inverter active and reactive power control decision data locked in the simulation verification process, the system generates an inverter control message containing the target node address and power setting value, and outputs the message to the distribution network data bus for the dispatch monitoring terminal to refer to or forward to the underlying execution terminal for parameter updates.

[0111] S53, as long as any scenario in the dynamic N-1 fault cluster triggers the active interception mechanism set in step S4, that is, after the inverter reaches the upper limit of the apparent power hardware capacity, the voltage over-limit deviation is still not eliminated, the system determines that the planned target operation mode cannot resist the risk of over-limit transfer voltage.

[0112] At this point, the system returns a signal indicating that the target operation mode verification failed. At the data transmission level, it intercepts the circuit breaker position change command output path for the target topology and generates a control message to instruct the switch action sequence to implement blocking. Simultaneously, the system pushes simulation evaluation alarm data to the dispatch center monitoring terminal. The alarm data includes the fault scenario identifier that triggered the interception, the location of the over-limit node, and the maximum over-limit deviation value. This data interaction triggers the replanning process for the distribution network operation mode.

[0113] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. An automatic verification method for the operation mode of an urban power distribution network, characterized in that, Includes the following steps: The corrected current electrical quantities are combined with the target switch states in the target topology to form a hybrid operation profile; Based on the online branch embedding vector output from the hybrid operation section, the safety margin metric is calculated using the support vector machine safety boundary to filter the dynamic N-1 fault set. For the fault branches in the dynamic N-1 fault set, simulate the opening and closing of the target interconnection switch to form a single-point power transfer fault scenario topology, update the node impedance matrix and extract the static voltage-reactive power control sensitivity operator. Based on the updated node impedance matrix, Z-bus nonlinear power flow iterative calculation is performed to calculate the voltage over-limit deviation and combine it with the static voltage-reactive power control sensitivity operator to generate control decision data for inverter reactive power output limiting and updating active power output. The process is iterated until active interception is triggered or the preset convergence accuracy threshold is met. Based on the safety verification results of all fault scenarios, output safety assessment result data to indicate the execution of the corresponding switch action sequence of the target topology or to implement the switch action sequence interlocking.

2. The automatic verification method for urban power distribution network operation mode according to claim 1, characterized in that, The steps for forming the hybrid operating section include: Collect the active power injection, reactive power injection, voltage amplitude, current switching status, and real-time output and apparent power limit of the inverter at the current moment; Unify the positive and negative directions of power and generate historical current carrying characteristics of branches. Perform filtering and judgment on real-time measurements and fill in missing measurements to obtain the corrected current electrical quantity. The corrected current electrical quantity is combined with the target switch state in the target topology to form the hybrid operation section.

3. The automatic verification method for urban power distribution network operation mode according to claim 1, characterized in that, The step of outputting the online branch embedding vector includes: Based on the hybrid operation section, branch node features and edge features are extracted. The graph attention network with fixed parameters is used to extract normalized attention coefficients and perform feature aggregation to obtain a global branch node feature matrix. The global branch node feature matrix is ​​input into a multilayer perceptron with fixed weights to generate a soft assignment matrix; Spatial coarse-grained pooling is performed based on the soft allocation matrix to output the dimensionality-reduced online branch embedding vector.

4. The automatic verification method for urban power distribution network operation mode according to claim 1, characterized in that, The steps for filtering the dynamic N-1 fault set include: The online branch embedding vector is input into the security classification function to obtain the security margin metric value of the corresponding branch. When the safety margin metric is greater than zero, or when the safety margin metric is less than or equal to zero and greater than or equal to a negative fuzzy margin threshold, the corresponding branch is added to the dynamic N-1 fault set. When the safety margin metric is less than the negative fuzzy margin threshold, the corresponding branch is removed from the set to be precisely verified.

5. The automatic verification method for urban power distribution network operation mode according to claim 1, characterized in that, The steps in forming the topology for a single-point supply failure scenario include: For the undervoltage area caused by the failure of the faulty branch, the end node is added to the breadth-first search queue and the candidate tie switch set is extracted by expanding it outward layer by layer. By comparing the power supply margin with the total load of the undervoltage area, candidate tie switches with a power supply margin not greater than the total load of the undervoltage area are eliminated. The remaining candidate tie switches that meet the load transfer requirements are sorted in descending order according to the power supply margin. The candidate tie switch ranked first is determined as the target tie switch and simulated to close, thus forming the topology of the single-point transfer fault scenario.

6. The automatic verification method for urban power distribution network operation mode according to claim 1, characterized in that, The steps of updating the nodal impedance matrix and extracting the static voltage-reactive power control sensitivity operator include: Generate the ground-state node impedance matrix corresponding to the target operating mode; The Sherman-Morrison formula is combined with admittance matrix perturbation to quickly calculate the updated node impedance matrix corresponding to the fault scenario based on the ground state node impedance matrix. Extract the imaginary part slice of the updated node impedance matrix to obtain the static voltage-reactive power control sensitivity operator.

7. The automatic verification method for urban power distribution network operation mode according to claim 1, characterized in that, The steps for calculating voltage over-limit deviation include: Based on the balanced node voltage vector and the updated node impedance matrix, a Z-bus power flow iteration operation is performed to obtain the node voltage complex vector; Extract the voltage amplitude of each node based on the complex vector of the node voltage; When a node voltage amplitude is lower than the node voltage safety lower limit, the node voltage safety lower limit is used as the target voltage for the corresponding over-limit node; When a node voltage amplitude exceeds the node voltage safety limit, the node voltage safety limit is used as the target voltage for the corresponding over-limit node; The deviation between the target voltage and the voltage amplitude obtained by the current Z-bus iteration is calculated and used as the voltage over-limit deviation.

8. The automatic verification method for urban power distribution network operation mode according to claim 1, characterized in that, The steps involved in generating regulatory decision data include: When the absolute value of the voltage over-limit deviation is greater than the preset convergence accuracy threshold, the theoretical reactive power increment demand is calculated based on the voltage over-limit deviation and the static voltage-reactive power control sensitivity operator. Based on the theoretical reactive power increment demand and the inverter apparent power capacity constraint, the reactive power control value is limited to determine the new reactive power output. Based on the inverter's apparent power capacity circle, calculate the upper limit of the active power output allowed under the current reactive power output conditions and update the active power output decision value.

9. The automatic verification method for urban power distribution network operation mode according to claim 8, characterized in that, The steps of continuous iteration until active interception is triggered or a preset convergence accuracy threshold is met include: The updated active power output decision value and the new reactive power output are used as the node injection amount for the next round of Z-bus power flow iteration. When the absolute value of the voltage over-limit deviation is not greater than the preset convergence accuracy threshold, the inverter control decision data is directly locked. When the inverter capacity reaches its upper limit and the voltage over-limit deviation still cannot be eliminated, the active interception is immediately triggered. If the convergence condition is not met even after the iteration counter reaches the maximum allowable number of iterations, the active interception is triggered synchronously.

10. The automatic verification method for urban power distribution network operation mode according to claim 1, characterized in that, The step of outputting safety assessment result data for indicating the execution of the target topology corresponding to the switch action sequence or the implementation of the switch action sequence interlocking includes: When the voltage amplitude of all nodes in the single-point power transfer fault scenario topology simultaneously satisfies the node voltage safety judgment inequality, the corresponding fault scenario is marked as safe. When it is detected that all single-point power transfer fault scenario topologies in the dynamic N-1 fault set are marked as safe, the safety assessment result data is output to indicate the execution of the corresponding switching action sequence of the target topology, and the locked inverter control decision data is sent out simultaneously. If any scenario in the dynamic N-1 fault set triggers the set active interception mechanism, the safety assessment result data for indicating the implementation of the switch action sequence interlocking will be output, and an alarm message containing the over-limit deviation will be pushed.

Images