Short-circuit fault self-recovery power protection system with high response speed

Abstract

The invention relates to a short-circuit fault self-recovery power protection system with high response speed, and the system comprises a data collection module which is responsible for capturing original data and reporting the original data to a pre-grouping module; the pre-grouping module is used for dividing the branches into a class A with a multi-recovery mode and a class B with a single-recovery mode, and outputting a feeder group division result at the same time; the optimal recovery scheme solving module is used for carrying out load flow calculation on the A-type branches by adopting a multi-objective optimization model and screening to obtain an optimal recovery scheme; the similarity calculation and grouping module is used for calculating cosine similarity and clustering the optimal recovery mode; the strategy library is used for storing the optimal recovery scheme corresponding to each group; and the online fault processing module is used for taking out an optimal recovery scheme in the group and guiding fault isolation and power supply recovery. The problem that the response speed is difficult to reduce due to the fact that a power grid recovery operation face adopts a grouping mode for a plurality of micro-grids is solved, field operation and manual intervention are reduced, and the overall duration of fault isolation and power supply reconstruction is shortened.

Description

Technical Field

[0001] The present invention relates to a power grid self-recovery technology, and in particular to a short-circuit fault self-recovery power protection system with high response speed. Background Art

[0002] When a branch of a power system fails, it often causes power outages downstream and affects the normal operation of the system. To achieve rapid recovery of the power system under branch failures and minimize the scope of power outages under system failures, the power grid's power restoration strategy is generally divided into three steps: first, locating the fault point, then isolating the faulty branch, and finally performing recovery operations.

[0003] The restoration operation of the power grid mainly involves opening and closing switch pairs to bypass the faulty branch. For power grids that rely on the main grid for power supply, the opening and closing operations of switch pairs can be performed in steps by switching units within the power plant. In areas with a large number of distributed power sources such as wind power and photovoltaic power, there may be multiple microgrids, and the branches may intersect, which requires centralized scheduling of multiple microgrids.

[0004] Compared with a single main grid, multiple microgrids have more switch pairs and more complex relationships. Existing related technologies divide complex lines and switches into several groups based on the similarity of network topology and post-fault reconstruction requirements to reduce the amount of calculation. However, the recovery plan for some branches is not unique. If multiple feasible recovery plans for the same branch are calculated one by one, it means an increase in the amount of calculation and the number of operations, which also reduces the response speed of the system. Summary of the Invention

[0005] Aiming at the problem that the power grid restoration operation faces the problem of multiple microgrids being grouped but the restoration schemes for some branches are not unique and it is still difficult to reduce the response speed, a short-circuit fault self-recovery power protection system with high response speed is proposed.

[0006] The technical solution of the present invention is:

[0007] A high-response short-circuit fault self-recovery power protection system includes a data acquisition module, a pre-grouping module, an optimal recovery method solution module, a similarity calculation and grouping module, a strategy library, and an online fault processing module. The details are as follows:

[0008] The data acquisition module is responsible for capturing line branch parameters, switch and tie line information, load data, distributed power supply information and historical fault records, and reporting these raw data to the pre-grouping module;

[0009] The pre-grouping module divides branches into Class A with multiple restoration methods and Class B with single restoration method based on the number of available tie lines and islanding risk, and outputs the feeder group division results. For Class A branches, the pre-grouping module pushes the corresponding branch list, feeder group structure, and related parameters to the optimal restoration solution solution module.

[0010] The optimal restoration solution module performs multi-objective power flow optimization on Class A branches, generates the optimal restoration solution for each branch, feeds these optimal restoration solutions into the similarity calculation and grouping module, and simultaneously writes them into the strategy library for persistent storage.

[0011] Similarity calculation and grouping module: This module constructs feature vectors based on switch action sequences, voltage deviations, and power loss changes, calculates cosine similarities between groups, completes clustering, and updates the optimal recovery solution corresponding to each cluster group into the strategy library.

[0012] The policy library is used to store the optimal recovery plan corresponding to each group;

[0013] When a new fault branch is detected, the online fault processing module obtains the real-time parameters and topology information of the branch from the data acquisition module, then queries the policy library to quickly locate the cluster group to which it belongs, and finally retrieves the optimal recovery plan within the group to guide fault isolation and power restoration.

[0014] Furthermore, the similarity calculation and grouping module is provided with a dynamically adjustable similarity threshold θ, so as to flexibly change the grouping accuracy under different fault scenarios.

[0015] Furthermore, the similarity threshold θ is updated according to the number of manual interventions, satisfying the following formula:

[0016] θ next =clamp(θ current -ω(h0-h))

[0017] in,

[0018]

[0019] Among them, θ current is the current threshold, θ next is the updated threshold, h0 is the expected number of manual interventions, h is the actual number of manual interventions during this fault, ω is the adjustment coefficient used to control the step size of each update; θ min is the set minimum threshold, θ max The maximum threshold value set.

[0020] Furthermore, the optimal recovery method solution module adopts the following multi-objective optimization model:

[0021]

[0022] Where i is the i-th branch, j is the j-th switch operation, n is the total number of branches, and m is the total number of switch operations; f loss,i Indicates network loss or load unpowered, Δt j represents the switching operation delay, α and β are weight factors;

[0023] Constraints:

[0024] Line current limit: I ij ≤I max,ij ;

[0025] Node voltage limit: 0.9≤V i ≤1.1;

[0026] Network radial structure: Use the spanning tree algorithm to ensure that no closed loops are formed;

[0027] Perform power flow calculations for each of the optional tie lines or distributed generation grid-connected combinations for Class A branches;

[0028] First, perform the power flow calculation and record the network loss, or how much load has not been restored, and record these data as f loss,1 ,f loss, 2,…,f loss,n , and finally add up to get the total network loss

[0029] At the same time, record the number of all closing / opening steps or cumulative delays required for the solution and record it as the total operation delay

[0030] Substitute the total network loss and the sum of the operation delay into the multi-objective optimization model, and the solution with the lowest multi-objective optimization score is selected as the optimal restoration solution;

[0031] Then perform similarity analysis on the restoration methods and group the lines according to the similarity;

[0032] Substitute the total network loss and the sum of operation delay into the objective function

[0033] Repeat the above calculation for all feasible solutions and finally compare their objective function values;

[0034] The solution with the smallest value is determined as the optimal recovery solution.

[0035] Furthermore, in the optimal restoration solution solving module, when the similarity of the optimal restoration solutions of two or more branches is higher than the threshold θ, they are grouped into the same group, and the optimal restoration solutions of the group are shared in the strategy library to reduce repeated power flow calculations and multi-objective optimization operations.

[0036] Furthermore, in the pre-grouping module, the "number of available interconnection lines" and the "distributed power supply island potential" are added together as the "number of recovery methods". If the value is greater than or equal to 2, the corresponding branch is judged as a Class A branch, otherwise it is judged as a Class B branch; among them, the Class A branch has at least two power supply reconstruction methods after a fault, and the Class B branch only relies on a single switching scheme after a fault.

[0037] Furthermore, the data collection module is used to collect and integrate the following information:

[0038] Line branch parameters, including impedance Z ij , capacity S max,ij , initial state x m ;

[0039] Switch and tie line information, including installation location L k , action delay and maximum breaking capacity;

[0040] Load data, including active power P of each node D,i and reactive power Q D,i ;

[0041] Distributed power information, including node location DG n , output upper and lower limits P G,n ,Q G,n and grid-connected / island mode;

[0042] Historical fault records, including fault location, type, and recovery methods used.

[0043] Furthermore, the pre-grouping module includes:

[0044] The preliminary branch classification unit determines whether the branch is Class A or Class B based on the number of available tie lines and the possibility of distributed generation islanding;

[0045] The feeder group division unit takes the substation or main power outlet as the starting point and uses breadth-first search to perform topological search. Once a tie line or distributed power grid connection point is encountered, it is divided to form a new feeder group.

[0046] Furthermore, the similarity calculation and grouping module operates on the A and B branches respectively:

[0047] The following eigenvectors are extracted for the optimal restoration solution of Class A branches:

[0048] Switch action sequence s, where +1, -1, and 0 represent closing, opening, or no operation;

[0049] Voltage deviation ΔV;

[0050] Power loss change ΔP loss ;

[0051] And use the cosine similarity formula:

[0052]

[0053] Determine whether the optimal recovery methods are similar. If the similarity is higher than θ, the corresponding branches are divided into the same group, where x is the feature vector corresponding to one branch and y is the feature vector corresponding to the other branch.

[0054] For Class B branches, perform similarity comparison with existing grouping features:

[0055]

[0056] Where GroupSet is the set of existing grouping features, b is the feature vector of the class B branch to be grouped, and g is the group representative vector in the existing grouping features. The purpose of similarity comparison of the existing grouping features is to find the group that is most similar to b among all existing groups.

[0057] If it exceeds θ, it is merged into the group; otherwise it remains independent.

[0058] Furthermore, the policy library stores a unified optimal recovery plan for each group. When the online fault processing module detects a short circuit fault in a branch, it immediately locates the group to which the branch belongs and searches for the switch action sequence in the corresponding plan to achieve the goals of rapid self-recovery and minimal power outage range.

[0059] The beneficial effects of the present invention are:

[0060] 1. By preliminarily grouping complex power grids containing distributed generation according to the various possible reconstruction methods after a fault and screening the optimal solution, the online stage can directly look up the table and execute the pre-calculated power restoration strategy, thereby reducing on-site calculations and manual intervention, significantly shortening the overall time for fault isolation and power supply reconstruction, and improving the speed of fault self-recovery.

[0061] 2. In the offline stage, multi-scheme branches and single-scheme branches are first distinguished, and lines with similar switching operation sequences, similar voltage deviations and network loss changes are grouped into the same group according to their similarity. This avoids repeated power flow and multi-objective optimization calculations for each line, reducing the system's calculation workload and the difficulty of offline management.

[0062] 3. In the similarity calculation and grouping process, a similarity threshold is introduced that can be dynamically updated with the number of manual interventions or changes in the operating environment. When the system automation level is high and the number of manual interventions is relatively small, the merged groups can be reduced; on the contrary, when the number of manual interventions increases, the similarity threshold is increased to strengthen the consistency within the group, ensure the adaptive balance between grouping accuracy and group number management, and further improve the efficiency and accuracy of online fault recovery. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a flow chart of the grouping algorithm of the present invention;

[0064] Figure 2 This is a flow chart of the adaptive similarity value algorithm of the present invention. DETAILED DESCRIPTION

[0065] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0066] The high-response short-circuit fault self-recovery power protection system proposed in this paper aims to improve the self-healing capability and fault recovery speed of distribution networks containing distributed generation (DG) in the event of line faults. By independently identifying branches with multiple recovery modes and performing power flow simulation and optimal solution screening, similarity analysis and grouping based on the optimal solution are finally formed into a unified offline strategy library, which can be quickly called up when online, reducing the amount of calculation and the scope of power outages.

[0067] like Figure 1 As shown, first, it is necessary to collect detailed parameters of all line branches in the distribution network, including the impedance Z of each line ij , capacity S max,ij , and the initial on-off state x m If x m =1, the line is put into operation; if x m =0, it is disconnected due to maintenance or other reasons;

[0068] On the one hand, impedance and capacity can be used to determine the load capacity and operating safety margin of the line in the power flow model. On the other hand, the initial state helps to clarify the distribution of currently operating and outage lines when simulating fault scenarios, and combines the fault removal action to evaluate the feasibility of the recovery steps.

[0069] When a fault occurs, the most important operation is often achieved by switch action, including isolating the fault section and switching the backup interconnection line. Therefore, the installation position of all switches needs to be recorded. k ;

[0070] Installation position L kIncluding the nodes where the tie lines are located, the geographical coordinates of the branch switches and the switch action characteristics.

[0071] During the multi-objective optimization and power flow calculation process, the installation location L k It can be used to evaluate the impact of different switching operation combinations on system voltage and current distribution, and determine whether additional overload or transient shock will be caused.

[0072] Load data mainly includes the active power P of each node in the dispatching system D,i and reactive power Q D,i ;

[0073] Through the active power P D,i and reactive power Q D,i , the power flow calculation can accurately evaluate the voltage and current distribution of each line and each node.

[0074] Preferably, in the multi-objective optimization link, load priority or load type is used to refine the recovery strategy so that important loads are prioritized when a failure occurs.

[0075] Preferably, for loads with time-varying characteristics, typical load scenarios can be selected for simulation offline, and the strategy can be slightly adjusted based on load forecasts in a real-time environment.

[0076] The access point and output characteristics of the distributed power source are extremely critical for fault self-healing. Therefore, the present invention needs to obtain the location of the distributed power source DG n Its output upper and lower limits, including active power P G,n and reactive Q G,n , while recording the distributed power grid connection or island mode information.

[0077] If a branch can supply power to downstream loads by means of distributed generation islanding after a fault, it belongs to the category of "multiple recovery methods".

[0078] The power flow initialization also incorporates the distributed power grid connection data, so as to accurately determine whether the distributed power can share the load and improve the voltage support capability in the subsequent fault simulation, and evaluate its operational safety when switching between different topologies.

[0079] In addition, historical fault records need to be organized for use as a reference or to assist in strategy selection during the offline phase.

[0080] Historical fault records include the fault location, type, time, and recovery method used.

[0081] If a branch has been connected to different distributed power sources for restoration multiple times in history, it means that the branch is actually a "multi-restoration mode" branch.

[0082] If historical fault records show that certain solutions cannot meet current limits or voltage constraints in specific scenarios, such low-feasibility solutions are eliminated.

[0083] After obtaining the above data, a Newton-Raphson (AC) or approximate (DC) power flow calculation is performed to obtain the voltage amplitude V of each node. i , phase angle θ i (If it is AC power flow), and line power flow distribution:

[0084]

[0085] in:

[0086] It represents the node active power injection value calculated by the power flow equation;

[0087] The sum of active power injection specified for the node;

[0088] The node reactive power injection value calculated by the power flow equation;

[0089] The sum of reactive injections specified for the node.

[0090] Through this initialization, the current grid operating status can be judged and used as a comparison benchmark in fault scenarios, providing starting operating condition data for subsequent fault simulation and optimal recovery plan evaluation.

[0091] Create an N×N adjacency matrix A:

[0092] If there is a branch between node i and node j, let A ij =1; otherwise, let A ij =0.

[0093] The lines are then pre-grouped. The goal of the pre-grouping step is to first identify the two types of branches: "multiple recovery methods" and "single recovery method", and preliminarily divide each feeder according to the grid topology, so as to conduct more accurate optimal recovery analysis and similarity grouping later.

[0094] This step mainly includes the following:

[0095] The first step is to calculate the number of recovery methods.

[0096] To determine whether a branch has multiple restoration methods, statistics are required from the following dimensions:

[0097] Number of available tie lines: When a branch fails, are there more than one backup tie line for load transfer? If the upstream and downstream of the branch can be connected to different power supply paths, each path can be considered a potential recovery solution.

[0098] Possibility of DG islanding: If there is a distributed power source on the downstream node or the node of the branch, it can form an island operation in the event of a fault. In this case, this mechanism can also be regarded as a single recovery method.

[0099] For the convenience of quantification, the number of recovery methods can be defined as:

[0100] "Number of recovery methods" = ("Number of available contact lines") + ("DG island potential"),

[0101] The values ​​of the distributed power supply island potential are as follows:

[0102] If the islanding condition is met, it can be set to 1; if the islanding condition is not met, it can be set to 0;

[0103] If the above sum is ≥ 2, it means that the branch has at least two power supply reconstruction paths after the fault, and can be judged as a Class A branch; otherwise, it is classified as Class B.

[0104] A Class A branch specifically refers to a branch that, after a fault, can, based on the flow and switch status, transfer load with the help of more than one backup interconnection line, or can provide island operation by DG to continue to supply power to downstream users. In this case, it is considered to have "multiple recovery modes".

[0105] Class B branches specifically refer to branches that have only a single backup line or no backup line. Once a fault occurs, the only feasible solution is usually to isolate the fault and rely on the original main feeder or a simple trip-reclosing operation. If neither the tie line nor the distributed power island is available, that is, the branch has only a single recovery method, it is recorded as a Class B branch.

[0106] In the second step, after the preliminary classification of Class A and Class B branches, in order to maintain the topological hierarchy in subsequent fault analysis, the distribution network needs to be divided into feeder groups. That is, starting from the substation or main power outlet, a breadth-first search is performed based on the electrical connection relationship of the nodes.

[0107] When a tie line or a distributed generation grid connection point is encountered during the search process, a new subgroup is formed at that point to avoid excessive mixing across regions.

[0108] Start the breadth-first search from the substation bus and visit the connected branches in sequence;

[0109] If a DG grid connection point or tie line node is encountered, the branch extension is stopped and the group boundary is divided here;

[0110] Repeat until all Class A or Class B branches are included in a feeder group.

[0111] Once completed, the entire distribution network is divided into several independent primary "feeder groups", each of which contains a group of Class A branches and Class B branches.

[0112] Through the above-mentioned pre-grouping steps of preliminary branch classification and feeder group division, Class A critical branches with multiple recovery methods can be quickly identified in the offline stage, and the network can be split and managed according to feeder group blocks.

[0113] Afterwards, the optimal restoration solution is calculated for the Class A branch:

[0114] Establish a multi-objective optimization model. The purpose of the multi-objective optimization model is to integrate the two goals of reducing load loss or network loss and reducing operating costs into one framework.

[0115] The details are as follows:

[0116]

[0117] Where: i is the i-th branch, j is the j-th switch operation, n is the total number of branches, and m is the total number of switch operations; f loss,i Indicates network loss or load unpowered quantity, etc., Δt j is the switching operation delay; α and β are weights used to balance the operating loss and the number of operations.

[0118] Constraints:

[0119] Line current limit: I ij ≤I max,ij .

[0120] Node voltage limit: 0.9≤V i ≤1.1.

[0121] Network radial structure: Use the spanning tree algorithm to ensure that no closed loop is formed.

[0122] Perform power flow calculations for each of the optional tie lines or distributed generation grid-connected combinations for Class A branches;

[0123] First, perform the power flow calculation and record the network loss, or how much load has not been restored, and record these data as f loss,1 ,f loss,2 ,…,f loss,n , and finally add up to get the total network loss

[0124] At the same time, record the number of all closing / opening steps or cumulative delays required for the solution and record it as the total operation delay

[0125] The total network loss and the sum of operation delays are substituted into the multi-objective optimization model, and the solution with the lowest multi-objective optimization score is selected as the optimal recovery solution.

[0126] Then perform similarity analysis on the restoration methods and group the lines according to the similarity;

[0127] Substitute the total network loss and the sum of operation delay into the objective function

[0128] Repeat the above calculation for all feasible solutions and finally compare their objective function values;

[0129] The solution with the smallest value is determined as the optimal recovery solution.

[0130] The following features are extracted from the optimal restoration solution for each Class A branch:

[0131] Switch action sequence ([L1,L2,...,L M ]); voltage deviation ΔV; power loss change ΔP loss ;

[0132] Combine the above features into a unified vector;

[0133] Specifically, for the switch action sequence, firstly, all switches or tie lines in the power grid that may participate in fault recovery are numbered as {L1, L2, ..., L M}, when a branch needs to close some switches and open other switches in its optimal restoration plan, these operations can be mapped to a vector of length M, denoted as s∈R M ;

[0134] s[k]=+1 means closing the kth switch; s[k]=-1 means opening the switch; if no operation is performed, 0 is assigned.

[0135] Consider voltage deviation and power loss variation as one-dimensional real numbers;

[0136] That is, we get the eigenvector [s1,s2,…,s M ,ΔV,ΔP loss ],

[0137] The cosine similarity is used to calculate the features:

[0138]

[0139] Where x is the eigenvector corresponding to one branch, and y is the eigenvector corresponding to the other branch;

[0140] If the similarity is high, the optimal restoration solutions of the corresponding branches are similar.

[0141] Cluster the optimal restoration plans; those with a similarity higher than θ are grouped together. After a branch failure occurs within the same group, the restoration strategy template of the group can be shared to reduce repeated calculations.

[0142] For Class B branches, perform similarity comparison with existing grouping features:

[0143]

[0144] GroupSet is a set of existing grouping features, b is the feature vector of the class B branch to be grouped, and g is the group representative vector in the existing grouping features. The purpose of similarity comparison of the existing grouping features is to find the group most similar to b among all existing groups.

[0145] If it exceeds θ, it is merged into the group; otherwise it remains independent.

[0146] For branches that cannot be classified into existing groups, perform separate power flow simulations on these branches and save the strategies.

[0147] Preferably, a dynamically updated similarity threshold θ is used, which is dynamically updated according to the number of manual interventions;

[0148] The dynamic update formula of the similarity threshold θ is as follows:

[0149] θ next =clamp(θ current -ω(h0-h))

[0150] in,

[0151]

[0152] Among them, θ current is the current threshold, θ next is the updated threshold, h0 is the expected number of manual interventions, h is the actual number of manual interventions during this fault, and ω is the adjustment coefficient, which is used to control the step size of each update. If ω = 0.1, only 10% of the difference is adjusted each time; θ min is the set minimum threshold, θ max The maximum threshold value set.

[0153] In the above formula, the similarity threshold θ is allowed to vary dynamically between a maximum and a minimum value. The number of manual interventions is used as an online feedback signal to dynamically adjust θ. This ensures that the self-healing grouping maintains sufficient accuracy while not being infinitely refined, achieving the optimal balance under different fault scenarios.

[0154] Finally, a self-recovery strategy library is constructed, and a group ID is assigned to each group. Each group includes the optimal switching sequence, load switching priority, power distribution information, and a list of applicable branches.

[0155] When the system detects a branch fault, the online fault processing module obtains the group ID of the faulty branch, finds the group to which the branch belongs and the optimal solution strategy, and performs closing / opening operations to quickly restore the non-faulty section.

[0156] The above-described embodiment merely represents one embodiment of the present invention. While the description is relatively specific and detailed, it should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make various modifications and improvements without departing from the spirit of the present invention, and these modifications and improvements fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A short-circuit fault self-recovery power protection system with high response speed, characterized in that: It includes data acquisition module, pre-grouping module, optimal recovery method solution module, similarity calculation and grouping module, strategy library, and online fault handling module, as follows: The data acquisition module is responsible for capturing line branch parameters, switch and tie line information, load data, distributed power supply information and historical fault records, and reporting these raw data to the pre-grouping module; The pre-grouping module divides branches into Class A with multiple restoration methods and Class B with single restoration method based on the number of available tie lines and islanding risk, and outputs the feeder group division results. For Class A branches, the pre-grouping module pushes the corresponding branch list, feeder group structure, and related parameters to the optimal restoration solution solution module. The optimal restoration solution module performs multi-objective power flow optimization on Class A branches, generates the optimal restoration solution for each branch, feeds these optimal restoration solutions into the similarity calculation and grouping module, and simultaneously writes them into the strategy library for persistent storage. Similarity calculation and grouping module: This module constructs feature vectors based on switch action sequences, voltage deviations, and power loss changes, calculates cosine similarities between groups, completes clustering, and updates the optimal recovery solution corresponding to each cluster group into the strategy library. The policy library is used to store the optimal recovery plan corresponding to each group; When a new fault branch is detected, the online fault processing module obtains the real-time parameters and topology information of the branch from the data acquisition module, then queries the policy library to quickly locate the cluster group to which it belongs, and finally retrieves the optimal recovery plan within the group to guide fault isolation and power restoration.

2. The high-response short-circuit fault self-restoring power protection system according to claim 1 is characterized in that: The similarity calculation and grouping module is provided with a dynamically adjustable similarity threshold θ, so as to flexibly change the grouping accuracy under different fault scenarios.

3. The high-response short-circuit fault self-restoring power protection system according to claim 2, characterized in that: The similarity threshold θ is updated according to the number of manual interventions and satisfies the following formula: i next =clamp(θ current -ω(h0-h)) in, Among them, θ current is the current threshold, θ next is the updated threshold, h0 is the expected number of manual interventions, h is the actual number of manual interventions during this fault, ω is the adjustment coefficient used to control the step size of each update; θ min is the set minimum threshold, θ max The maximum threshold value set.

4. The high-response short-circuit fault self-restoring power protection system according to claim 1, characterized in that: The optimal recovery method solution module adopts the following multi-objective optimization model: Where i is the i-th branch, j is the j-th switch operation, n is the total number of branches, and m is the total number of switch operations; f loss,i Indicates network loss or load unpowered, Δt j represents the switching operation delay, α and β are weight factors; Constraints: Line current limit: I ij ≤I max,ij ; Node voltage limit: 0.9≤V i ≤1.1; Network radial structure: Use the spanning tree algorithm to ensure that no closed loops are formed; Perform power flow calculations for each of the optional tie lines or distributed generation grid-connected combinations for Class A branches; First, perform the power flow calculation and record the network loss, or how much load has not been restored, and record these data as f loss,1 ,f loss,2 ,…,f loss,n , and finally add up to get the total network loss At the same time, record the number of all closing / opening steps or cumulative delays required for the solution and record it as the total operation delay Substitute the total network loss and the sum of the operation delay into the multi-objective optimization model, and the solution with the lowest multi-objective optimization score is selected as the optimal restoration solution; Then perform similarity analysis on the restoration methods and group the lines according to the similarity; Substitute the total network loss and the sum of operation delay into the objective function Repeat the above calculation for all feasible solutions and finally compare their objective function values; The solution with the smallest value is determined as the optimal recovery solution.

5. The high-response short-circuit fault self-restoring power protection system according to claim 2, characterized in that: In the optimal restoration solution solving module, when the similarity of the optimal restoration solutions of two or more branches is higher than the threshold θ, they are grouped into the same group, and the optimal restoration solutions of the group are shared in the strategy library to reduce repeated power flow calculations and multi-objective optimization operations.

6. The high-response short-circuit fault self-restoring power protection system according to claim 1, characterized in that: In the pre-grouping module, the "number of available tie lines" and the "distributed generation island potential" are added together to form the "number of recovery methods." If this value is greater than or equal to 2, the corresponding branch is classified as a Class A branch; otherwise, it is classified as a Class B branch. Class A branches have at least two power supply reconstruction methods after a fault, while Class B branches rely on only a single switching solution after a fault.

7. The high-response short-circuit fault self-restoring power protection system according to claim 1, characterized in that: The data acquisition module is used to collect and integrate the following information: Line branch parameters, including impedance Z ij , capacity S max,ij , initial state x m ; Switch and tie line information, including installation location L k , action delay and maximum breaking capacity; Load data, including active power P of each node D,i and reactive power Q D,i ; Distributed power information, including node location DG n , output upper and lower limits P G,n ,Q G,n and grid-connected / island mode; Historical fault records, including fault location, type, and recovery methods used.

8. The high-response short-circuit fault self-restoring power protection system according to claim 1, characterized in that: The pre-grouping module includes: The preliminary branch classification unit determines whether the branch is Class A or Class B based on the number of available tie lines and the possibility of distributed generation islanding; The feeder group division unit takes the substation or main power outlet as the starting point and uses breadth-first search to perform topological search. Once a tie line or distributed power grid connection point is encountered, it is divided to form a new feeder group.

9. The high-response short-circuit fault self-restoring power protection system according to claim 2, characterized in that: The similarity calculation and grouping module operates on the A and B branches respectively: The following eigenvectors are extracted for the optimal restoration solution of Class A branches: Switch action sequence s, where +1, -1, and 0 represent closing, opening, or no operation; Voltage deviation ΔV; Power loss change ΔP loss ; And use the cosine similarity formula: Determine whether the optimal recovery methods are similar. If the similarity is higher than θ, the corresponding branches are divided into the same group, where x is the feature vector corresponding to one branch and y is the feature vector corresponding to the other branch. For Class B branches, perform similarity comparison with existing grouping features: Where GroupSet is the set of existing grouping features, b is the feature vector of the class B branch to be grouped, and g is the group representative vector in the existing grouping features. The purpose of similarity comparison of the existing grouping features is to find the group that is most similar to b among all existing groups. If it exceeds θ, it is merged into the group; otherwise it remains independent.

10. The high-response short-circuit fault self-restoring power protection system according to claim 1, characterized in that: The policy library stores a unified optimal recovery plan for each group. When the online fault processing module detects a short circuit fault in a branch, it immediately locates the group to which the branch belongs and searches for the switch action sequence in the corresponding plan to achieve the goals of rapid self-recovery and minimal power outage scope.

Images