Distribution network feeder fault automatic isolation and self-healing method, device and equipment

Abstract

The application discloses a distribution network feeder fault automatic isolation and self-healing method, device and equipment, relates to the technical field of distribution network fault isolation, and comprises the following steps: obtaining an adjacent section set based on obtained power grid operation data; generating a fusion contract basis according to the obtained fault severity and adjacent section attributes; screening an influence section from the adjacent section set based on the fusion contract basis; and obtaining an isolation strategy through sparse classification according to the influence section and the power grid topology, and performing self-healing. The application is used to solve the problem that the traditional method is difficult to accurately identify the fault influence range and efficiently generate the optimal isolation and self-healing scheme.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of distribution network fault isolation, and more particularly to a distribution network feeder fault automatic isolation and self-healing method, device and equipment. BACKGROUND

[0002] As the last mile of the power system extending from the transmission link to the user end, the distribution network has the significant characteristics of multiple nodes, miscellaneous branches, and wide coverage, and directly faces various types of user loads in a complex and variable operating environment. With the large-scale access of distributed power sources and the rapid growth of new loads such as electric vehicle charging piles, the operating state of the distribution network is increasingly dynamic, and the causes of faults are more diverse, with short-circuit and grounding faults occurring from time to time.

[0003] After a fault occurs, it is a core requirement to quickly isolate the fault and restore power supply to the non-fault area to ensure user power supply reliability and reduce power loss. Traditional distribution network fault handling relies on manual inspection and experience-based operation, and it often takes several hours or even longer from fault reporting, on-site investigation to isolation and recovery, which has been difficult to meet the requirements of high reliability and less interruption of power supply in modern society. Therefore, relying on automation technology, data analysis technology and intelligent algorithms to build a distribution network feeder fault automatic isolation and self-healing system has become one of the core directions of the intelligent upgrading of distribution networks.

[0004] Existing technologies mostly adopt a fragmented data collection and utilization mode: some schemes only rely on single electrical quantity data such as fault current and voltage, ignoring key spatial and state information such as real-time topology structure and section node state of the distribution network; another part of the schemes collects multi-dimensional data, but does not establish a correlation mapping between the data, resulting in a lack of basis for matching the fault section and the surrounding associated sections.

[0005] Existing technologies mostly rely on simple logic such as threshold comparison to judge the degree of fault impact, and do not build a quantitative evaluation system that integrates multiple factors such as fault strength and duration, resulting in strong subjectivity and low accuracy in judging the degree of section impact.

[0006] For example, the invention patent CN110571802A discloses a distribution network self-healing control method and device based on a rough set dynamic programming model, which includes: establishing a multi-objective optimization distribution network self-healing control model; using rough set theory to reduce data knowledge of the power grid data; performing fault positioning through evaluation function calculation and non-fault interval search; and quickly obtaining an optimal self-healing scheme of the intelligent distribution network to control fault self-healing. By directly analyzing and reasoning the data, implicit knowledge and potential laws are discovered, and the system self-healing recovery speed is improved.

[0007] For example, the invention patent announcement with publication number: CN116559598A provides a kind of intelligent distribution network fault location method and system, it is related to distribution network fault isolation technical field, comprising: according to the real-time electrical parameter of distribution network, environmental parameter and historical fault data, dynamic factor is automatically adjusted;The dynamic factor is fused with historical fault data and topological structure to form a multidimensional fault location model;According to the multidimensional fault location model, the fault position is predicted to obtain the prediction result, and the confidence of the prediction result is judged;According to the prediction result and the confidence corresponding to the prediction result, the fault position of distribution network is determined.By forming multidimensional fault location model, the accuracy of fault location is effectively improved, by setting automatic adjustment dynamic factor can respond to the real-time change of distribution network operating state, fault location automation can also be realized, reduce the dependence of artificial judgment, improve the efficiency and response speed of fault handling.

[0008] In the above disclosed technical solution, at least the following technical problems exist:

[0009] Although CN110571802A proposes to use rough set theory to reduce data knowledge of power grid data, and then fault location, but does not explain how to quantify the influence degree of fault on different sections, and the judgment basis is subjective.In addition, the existing technology uses dynamic programming, which is essentially to solve the optimal solution problem of multi-stage decision-making, but the core of distribution network fault isolation is classification decision based on real-time topology and characteristics, and the adaptability of dynamic programming is low.CN116559598A proposes a multidimensional fault location model that can accurately locate faults, but there are obvious deficiencies when directly applied to distribution network fault location.This method can only determine the specific node where the fault occurs, cannot judge the range of influence and the degree of influence of the fault on the surrounding sections, and only judges the reliability of the positioning result by confidence, without quantifying the degree of harm of the fault itself.

[0010] To solve the above problems, the present application provides a solution. SUMMARY

[0011] To overcome the above-mentioned defects of the prior art, embodiments of the present application provide a distribution network feeder fault automatic isolation and self-healing method, device and equipment, which calculates the fault severity, selects the affected section according to the reduction algorithm, and generates an isolation strategy using a sparse classification model, solving the problem that traditional methods cannot accurately identify the range of fault influence and efficiently generate optimal isolation and self-healing solutions.

[0012] To achieve the above-mentioned purposes, the present application provides the following technical solutions:

[0013] The method for automatically isolating and self-healing distribution network feeder faults, characterized in that it comprises the following steps: obtaining a set of adjacent sections based on obtained power grid operation data; generating a fusion reduction basis according to obtained fault severity and adjacent section attributes; screening an impact section from the set of adjacent sections based on the fusion reduction basis; and obtaining an isolation strategy through sparse classification according to the impact section and the power grid topology, and performing self-healing.

[0014] In a preferred embodiment, the set of adjacent sections is obtained based on the obtained power grid operation data, specifically: the physical connection paths between sections are determined according to the real-time topology structure of the distribution network; the section that has failed is determined according to the fault section identifier; and the sections connected to the fault section in the real-time topology structure of the distribution network are aggregated to form the set of adjacent sections, with the fault section as the starting point.

[0015] In a preferred embodiment, the fusion reduction basis is generated according to the obtained fault severity and adjacent section attributes, specifically: the conditional attributes are constructed by obtaining the topological position, adjacent relationship and operating state of the adjacent section; the fault severity is calculated according to the obtained fault data, and the decision attributes corresponding to the conditional attributes are established according to the fault severity; and the fusion reduction basis is generated by a reduction algorithm based on the conditional attributes and the decision attributes.

[0016] In a preferred embodiment, the fusion reduction basis is generated by a reduction algorithm based on the conditional attributes and the decision attributes, specifically: the attribute dependency of the conditional attributes on the decision attributes is calculated; redundant attributes are removed by using a heuristic search algorithm based on the attribute dependency to obtain the reduced conditional attributes; and the rule set is generated as the fusion reduction basis according to the reduced conditional attributes and the decision attributes.

[0017] In a preferred embodiment, the attribute dependency of the conditional attributes on the decision attributes is calculated, specifically: the conditional equivalence classes and the decision equivalence classes are obtained by performing equivalence class division on the conditional attributes and the decision attributes respectively, with samples of the same attribute value being classified into a class; and the attribute dependency is calculated according to the conditional equivalence classes and the decision equivalence classes.

[0018] In a preferred embodiment, the fault severity is calculated, and the decision attributes corresponding to the conditional attributes are established according to the fault severity, specifically: the amplitude standardized index is obtained by performing standardization processing on the fault current amplitude; the time weight index is obtained by performing weighted processing on the fault current duration; the fault severity is obtained by linearly combining the amplitude standardized index and the time weight index; and the decision attributes corresponding to the adjacent section attributes are established by dividing the fault impact level according to the fault severity.

[0019] In a preferred embodiment, the basis for the fusion summary is used to screen the impact section from the adjacent section set, specifically: obtaining the condition attribute of each section in the adjacent section set; comparing the condition attribute of the section with the basis for the fusion summary to find the target basis that fits the section condition attribute; determining the decision attribute of the corresponding adjacent section according to the target basis.

[0020] In a preferred embodiment, the sparse classification is performed according to the impact section and the power grid topology, and self-healing is performed, specifically: based on the impact section and the real-time topology structure of the power distribution network, feature parameters are extracted and an isolation feature vector is constructed; a sparse classification model is constructed to output an optimal isolation strategy; the optimal isolation strategy is substituted into the real-time topology structure for verification.

[0021] In a preferred embodiment, the sparse classification model is constructed to output the optimal isolation strategy: feature parameters are extracted from the impact section and the real-time topology structure of the power distribution network and are quantified as an isolation feature vector; an isolation classifier is constructed, and model parameter sparsification is realized by introducing a regularization term into the optimization objective of the classifier; the optimization objective with the regularization term is converted into a linear programming problem, and the optimal model parameters are solved; based on the optimal model parameters, a final sparse classification model is constructed, the isolation feature vector of the impact section is input into the sparse classification model, and the optimal isolation strategy is output.

[0022] In a preferred embodiment, after the isolation state is obtained, self-healing is performed, specifically: operation information is extracted according to the isolation strategy, and control instructions are issued according to the operation information to control the corresponding circuit breakers and switches; after the control instructions are completed, the power distribution network topology structure model is updated according to the switch state; based on the updated topology structure model, a recovery path candidate set is constructed, and the shortest path algorithm is used to screen the optimal recovery path; according to the optimal recovery path, a specific recovery scheme is generated.

[0023] In a preferred embodiment, the sparse classification model includes an isolation classifier, an optimization objective, and an isolation feature vector.

[0024] A power distribution network feeder fault automatic isolation and self-healing device includes a real-time data acquisition and integration module for obtaining power grid operation data according to a power distribution network real-time operation report; a fault impact section analysis module for analyzing the association between a fault data set and a power distribution network real-time topology structure by a reduction algorithm to generate a fusion summary basis; a fault impact section screening module for screening an impact section from an adjacent section set according to the fusion summary basis; and a fault isolation and self-healing execution module for constructing a sparse classification model according to the impact section and the power distribution network real-time topology structure, optimizing an isolation classifier by a linear programming algorithm, obtaining an isolation strategy and executing it to generate a recovery scheme.

[0025] An electronic device, comprising: at least one processor; and an input / output interface connected in communication with the at least one processor; a memory connected in communication with the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the power distribution feeder fault automatic isolation and self-healing method.

[0026] The power distribution feeder fault automatic isolation and self-healing method, device and equipment provided by the application have the following technical effects and advantages:

[0027] The power distribution feeder fault automatic isolation and self-healing method, device and equipment provided by the application have the following technical effects and advantages: BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 The power distribution feeder fault automatic isolation and self-healing method flowchart provided by the embodiment of the application.

[0029] Figure 2 The power distribution feeder fault automatic isolation and self-healing device structure diagram provided by the embodiment of the application.

[0030] Figure 3 The power distribution feeder fault automatic isolation and self-healing device structure diagram provided by the embodiment of the application. DETAILED DESCRIPTION

[0031] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0032] It is to be noted that the relative terms such as first and second and the like are used herein solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element preceded by "comprises... " does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that comprises the stated elements.

[0033] Embodiment 1, Figure 1 A power distribution network feeder fault automatic isolation and self-healing method is given, comprising the following steps:

[0034] S1, obtaining a set of adjacent sections based on the obtained power grid operation data;

[0035] S2, generating a fusion contract basis according to the obtained fault severity and adjacent section attributes;

[0036] S3, screening an impact section from the set of adjacent sections based on the fusion contract basis;

[0037] S4, obtaining an isolation strategy through sparse classification according to the impact section and the power grid topology, and performing self-healing.

[0038] The power distribution network feeder fault automatic isolation and self-healing method provided by the application effectively solves the problem that the traditional method cannot accurately identify the fault impact range and efficiently generate the optimal isolation and self-healing scheme.

[0039] S1, obtaining a set of adjacent sections based on the obtained power grid operation data;

[0040] In this embodiment, the set of adjacent sections is obtained based on the obtained power grid operation data, specifically:

[0041] According to the real-time topology structure of the distribution network, the physical connection path between the sections is determined;

[0042] According to the fault section identifier, the section where the fault occurs is determined;

[0043] Taking the fault section as the starting point, the sections connected to the fault section in the real-time topology structure of the distribution network are summarized to form a set of adjacent sections.

[0044] In this embodiment, the power grid operation data includes a fault data set, a real-time topology structure of the distribution network, real-time electrical quantity waveforms, and section node states.

[0045] In this embodiment, the fault data set includes fault section identification, fault current amplitude and fault current duration.

[0046] It should be noted that, in this embodiment, by integrating multi-dimensional data such as fault data set and real-time topology, the adjacent section of the fault is locked, not only solving the problem of scattered and insufficient correlation of traditional data, but also reducing the omission or misjudgment of adjacent section determination caused by data fragmentation.

[0047] S2, generating a fusion reduction basis according to the obtained fault severity and adjacent section attribute;

[0048] In this embodiment, the fusion reduction basis is generated according to the obtained fault severity and adjacent section attribute, specifically:

[0049] Obtain the topology position, adjacent relationship and operating state of the adjacent section to construct the condition attribute;

[0050] According to the obtained fault data, the fault severity is calculated, and the decision attribute corresponding to the condition attribute is established according to the fault severity;

[0051] Based on the condition attribute and the decision attribute, a fusion reduction basis is generated by a reduction algorithm.

[0052] In this embodiment, the fusion reduction basis is generated based on the condition attribute and the decision attribute by a reduction algorithm, specifically:

[0053] Calculate the attribute dependency of the condition attribute on the decision attribute;

[0054] Based on the attribute dependency, a heuristic search algorithm is used to eliminate redundant attributes to obtain the reduced condition attribute;

[0055] According to the reduced condition attribute and the decision attribute, a rule set is generated as the fusion reduction basis.

[0056] In this embodiment, the fault severity is calculated, and the decision attribute corresponding to the condition attribute is established according to the fault severity, specifically:

[0057] The fault current amplitude is standardized to obtain an amplitude standardized index;

[0058] The fault current duration is weighted to obtain a time weight index;

[0059] The amplitude standardized index and the time weight index are linearly combined to obtain the fault severity;

[0060] According to the fault severity, the fault influence level is divided, and the decision attribute corresponding to the adjacent section attribute is established.

[0061] In this embodiment, the attribute dependency degree of the calculation condition attribute to the decision attribute is specifically:

[0062] The condition attribute and the decision attribute are respectively divided into equivalence classes, and samples with the same attribute value are classified into a class to obtain condition equivalence classes and decision equivalence classes;

[0063] The attribute dependency degree is calculated according to the condition equivalence classes and the decision equivalence classes.

[0064] In this embodiment, Table 1 shows three adjacent section samples:

[0065] Table 1

[0066]

[0067] The condition attribute is divided according to the topological position and the load rate interval, and then the condition equivalence classes are obtained:

[0068] C1={A1} (topological position=main line and load rate interval=M)

[0069] C2={A2, A3} (topological position=branch line and load rate interval=L).

[0070] Decision equivalence classes:

[0071] D1={A2} (decision attribute=1)

[0072] D2={A3} (decision attribute=2)

[0073] D3={A1} (decision attribute=3).

[0074] It should be noted that if all samples in a condition equivalence class belong to the same decision equivalence class, that is, the decision result can be determined by 100% through this condition grouping, then the samples in the condition equivalence class are counted into the positive domain, otherwise, if the samples in a condition equivalence class are scattered in multiple decision equivalence classes and the decision result cannot be determined, then they are not counted into the positive domain.

[0075] It should be noted that the positive domain determines which section fault impact degree can be directly locked by the key attribute, and provides a quantitative basis for subsequent removal of redundant attributes and extraction of fault decision rules. The larger the positive domain size is, the stronger the certainty of the current condition attribute in determining the fault severity is, and the higher the guidance value of the section affected by the screening is.

[0076] In this embodiment, the attribute dependency degree is specifically as follows:

[0077]

[0078] is the set of all adjacent section samples, For With respect to The positive domain of For condition equivalence class, For decision equivalence class, For attribute dependency, the value range is [0, 1].

[0079] In this embodiment, based on the attribute dependency, the heuristic search algorithm is used to eliminate redundant attributes, and the reduced condition attributes are obtained, which are specifically:

[0080] The attribute dependency of the complete attribute set on the decision attribute is calculated to obtain the benchmark dependency;

[0081] The empty set is used as the initial reduction set, and the heuristic search is used to iteratively select the attribute that can maximize the dependency to join the reduction set;

[0082] The attribute set obtained by iteration screening is subjected to redundancy test, and redundant attributes are eliminated.

[0083] In this embodiment, the empty set is used as the initial reduction set, and the heuristic search is used to iteratively select the attribute that can maximize the dependency to join the reduction set, which is specifically:

[0084] In each iteration, from the current unselected attributes, the attribute that can maximize the combination dependency of the current reduction set and the condition attribute is selected, and is added to the reduction set;

[0085] Repeat the process until the attribute dependency of the reduction set reaches or is equal to the benchmark dependency, indicating that the current reduction set has the same decision-making ability as the complete attribute set.

[0086] In this embodiment, the reduction set obtained by iteration screening is subjected to redundancy test, and redundant attributes are eliminated, which is specifically:

[0087] Remove one attribute in the reduction set one by one, and check whether the dependency of the reduction set after removing the attribute still maintains the benchmark dependency level;

[0088] If the dependency does not decrease after removing the attribute, the attribute is a redundant attribute and is eliminated, and if the dependency decreases, the attribute is a key attribute and is retained in the reduction set.

[0089] In this embodiment, the heuristic search algorithm is a greedy selection strategy, which selects the attribute that can maximize the dependency in each iteration, ensuring that the same decision-making ability as the complete attribute set is achieved with the least attributes.

[0090] In this embodiment, the fault severity formula is specifically:

[0091]

[0092] wherein, and are set weight coefficients, both of which are 0.5 in this embodiment, is a fault severity, is a current amplitude index, is a time weight index.

[0093] In this embodiment, the current amplitude index is a normalized fault current amplitude, and the specific formula is:

[0094]

[0095] wherein, is a measured fault current amplitude, is a minimum fault current value in the data, is a maximum fault current value in the data.

[0096] In this embodiment, the time weight index is a fault current duration after weighting processing, and the specific formula is:

[0097]

[0098] wherein, is a measured fault duration, is a preset attenuation coefficient, and the general range is [0.1, 0.3].

[0099] In this embodiment, the fault influence level is divided according to the fault severity, and the decision attribute corresponding to the attribute of the adjacent section is established, which is specifically:

[0100] The fault severity is divided into several levels, and the initial threshold value of each level is set;

[0101] Based on historical fault data, the threshold value is adjusted by using a boundary optimization algorithm;

[0102] The fault severity of each adjacent section is mapped to the corresponding level as a decision attribute.

[0103] It should be noted that the boundary optimization algorithm in this embodiment is a particle swarm algorithm, which is a prior art and will not be repeated here.

[0104] In this embodiment, three adjacent section samples are obtained, each sample has three kinds of data, i.e., topological position, load rate interval and fault severity level, , The fault influence level is divided into three levels, and the threshold value is 1 for the first level , 2 for the second level , and 3 for the third level Then, the adjacent section attribute containing the decision attribute is obtained, as shown in Table 2.

[0105] Table 2

[0106]

[0107] In this embodiment, the rule set is generated according to the reduced condition attribute and the decision attribute as the fusion reduction basis, specifically:

[0108] The initial rule is extracted from the corresponding relationship between the reduced condition equivalence class and the decision equivalence class. If a condition equivalence class (A) is completely contained in a decision equivalence class (B), an initial rule is generated, in which the condition attribute is A and the decision attribute is B.

[0109] The initial rule is optimized by removing duplicates, merging similar rules and screening confidence.

[0110] The optimized rule is arranged as a structured rule set, which is used as the fusion reduction basis for subsequent screening of the influence section.

[0111] In this embodiment, the initial rule is optimized by removing duplicates, merging similar rules and screening confidence, specifically:

[0112] If the condition attribute value combination of two rules is completely consistent and the decision result is the same, only one rule is retained and the remaining duplicates are deleted.

[0113] If the difference is only in the value of the continuous condition attribute, and the difference has no significant effect on the decision result, the rules are merged into a more generalized rule, and the specific numerical value is replaced by an interval range.

[0114] The confidence of each rule is calculated according to the historical data, a confidence threshold is set, and rules with a confidence lower than the threshold are removed.

[0115] In this embodiment, the specific calculation formula of the confidence is:

[0116]

[0117] In the formula, n represents the total number of samples in the historical sample set, n represents the total number of samples in the historical sample set, and n represents the number of samples in the historical sample set that meet the condition attribute and the decision attribute.

[0118] ​​​​​It should be noted that the embodiment calculates the fault severity and combines the reduction algorithm, quantifies the key parameters as decision attributes, removes redundant information to screen the affected sections, improves the decision efficiency, and avoids the subjective bias of artificial experience judgment based on the structured judgment logic of the rule set, further improves the consistency and reliability of the screening result, and lays a foundation for subsequent accurate fault isolation.

[0119] S3, screening the affected section from the adjacent section set based on the fusion reduction basis;

[0120] In the embodiment, the affected section is screened from the adjacent section set based on the fusion reduction basis, specifically:

[0121] Obtain the condition attribute of each section in the adjacent section set;

[0122] Compare the condition attribute of the section with the fusion reduction basis, and find the target basis that matches the condition attribute of the section;

[0123] Determine the decision attribute of the corresponding adjacent section according to the target basis.

[0124] It should be noted that the embodiment screens the affected section by removing redundant information based on the reduction basis, which not only improves the judgment accuracy, but also focuses on key attributes to reduce irrelevant information interference, making the screening of the affected section more in line with the fault diffusion law.

[0125] S4, according to the affected section and the power grid topology, obtain the isolation strategy through sparse classification, and perform self-healing;

[0126] In the embodiment, the isolation strategy is obtained through sparse classification based on the affected section and the power grid topology, and self-healing is performed, specifically:

[0127] Based on the affected section and the real-time topology structure of the distribution network, extract feature parameters and construct an isolation feature vector;

[0128] Construct a sparse classification model to output the optimal isolation strategy;

[0129] Substitute the optimal isolation strategy into the real-time topology structure for verification.

[0130] In the embodiment, the sparse classification model is constructed to output the optimal isolation strategy:

[0131] Extract feature parameters from the affected section and the real-time topology structure of the distribution network and quantify them as an isolation feature vector;

[0132] Construct an isolation classifier, and realize the sparsification of model parameters by introducing a regularization term into the optimization objective of the classifier;

[0133] The optimization objective with the regularization term is converted into a linear programming problem, and the optimal model parameters are obtained by solving the linear programming problem;

[0134] Based on the optimal model parameters, a final sparse classification model is constructed, and the isolation feature vector of the influence section is input into the sparse classification model to output an optimal isolation strategy.

[0135] In this embodiment, the extracted feature parameters are quantified into an isolation feature vector, specifically, the non-numerical features and numerical features extracted from the influence section and the real-time topology structure of the power distribution network are uniformly converted into numerical vector forms that can be used for sparse classification model calculation through standardization, encoding and mapping processing.

[0136] In this embodiment, the isolation classifier is constructed, and the model parameter sparsification is realized by introducing a regularization term into the optimization objective of the classifier, specifically:

[0137] Based on the influence section and the real-time topology structure of the power distribution network, a training sample is constructed;

[0138] The classification function is represented by a linear combination of isolation feature vectors, an L1 regularization term is introduced into the optimization objective of the classifier, and parameter sparsification is realized by constraining the L1 norm of the model parameters;

[0139] A regularization parameter is set to form an optimization objective with regularization constraints, and automatic screening of key features is performed.

[0140] In this embodiment, the specific formula of the output of the classifier is:

[0141]

[0142] is a feature weight vector, is a bias term, is the isolation feature vector of the i-th section.

[0143] In this embodiment, the specific formula of the classification loss function is:

[0144]

[0145] In the formula, is a classification loss function, is the total number of training samples, is the isolation decision label of the i-th section (1 indicates that isolation is needed, and -1 indicates that isolation is not needed), is the output obtained by inputting the isolation feature vector of the i-th section into the classifier.

[0146] ​​​In this embodiment, the optimization objective with regularization constraint is specifically a target function:

[0147]

[0148] wherein, is the minimum value of and , is a preset regularization parameter, is the sum of absolute values of all weight elements of the L1 norm of the feature weight vector.

[0149] It should be noted that, The greater the value is, the easier the weight is to be compressed to 0, and the more strict the feature screening is. The sparse induction feature of the regularization constraint can force the weight of the redundant feature to be 0.

[0150] In this embodiment, the regularization parameter is set to form the optimization objective with regularization constraint, and the automatic screening of the key features is specifically as follows:

[0151] The target function is solved, and the weight corresponding to the feature that has a small influence on the isolation decision is compressed to 0;

[0152] After the solving is completed, the feature corresponding to the non-zero weight is screened out;

[0153] The feature corresponding to the non-zero weight is retained to form a key feature set that has a large influence on the isolation decision.

[0154] In this embodiment, the optimization objective with the regularization term is converted into a linear programming problem, and the optimal model parameter is solved, and the conversion is specifically as follows:

[0155] An auxiliary variable is introduced to replace the absolute value term in the target function, and the L1 regularization constraint is converted into a linear constraint;

[0156] According to the decision boundary condition of the classifier, the nonlinear constraint of the original optimization problem is converted into a linear inequality constraint;

[0157] The target function and the linear constraint condition are integrated to form a linear programming problem;

[0158] A linear programming solving algorithm is used to solve the converted problem to obtain the optimal parameter of the isolation classifier.

[0159] In this embodiment, the target function and the linear constraint condition are integrated to form a linear programming problem, and the integration is specifically as follows:

[0160] A non-negative auxiliary variable is introduced to convert the absolute value constraint into a linear inequality;

[0161] The interval constraint of the isolation classifier is converted into a linear inequality by expanding the feature weight vector.

[0162] In the embodiment, the introduction of non-negative auxiliary variables converts the absolute value constraint into a linear inequality, specifically introducing auxiliary variables , adding and , where is a regularization term parameter, and when takes a value equivalent to , the original regularization constraint is converted to , and the absolute value nonlinear constraint is decomposed into a linear inequality constraint.

[0163] In the embodiment, the interval constraint of the isolated classifier is converted into a linear inequality by expanding the feature weight vector, specifically the constraint that the sparse classifier needs to satisfy the correct classification of the sample and has a preset interval. The original constraint is:

[0164]

[0165]

[0166] is the sample label, is the feature vector, is the bias, is the slack variable.

[0167] By expanding , the original constraint is converted into a linear inequality, and the specific formula after expansion is:

[0168]

[0169] In the formula, is the jth feature value of sample i.

[0170] It should be noted that the introduction of the L1 regularization term in the classifier optimization objective and the realization of parameter sparsification by constraining the L1 norm of the model parameter utilize the sparse induction characteristics of the L1 regularization to the parameter, forcing most parameters to become 0 and only retaining a few key parameters for classification decision.

[0171] In the embodiment, based on the updated topological structure model, a candidate set of recovery paths is constructed, and a shortest path algorithm is used to select the optimal recovery path, specifically:

[0172] Based on the updated topological structure model, all power loss sections and available power supply points are identified;

[0173] According to the topological connection relationship, all possible power supply paths from the available power supply points to each power loss section are generated to construct a candidate set of recovery paths;

[0174] constructing a candidate path optimization index, and assigning a weight to the candidate path;

[0175] adopting a shortest path algorithm to calculate a comprehensive weight value of each candidate path, screening out a path with the minimum weight value, and determining the screened path as an optimal recovery path.

[0176] In this embodiment, the candidate path optimization index includes path length, switch operation times, and load transfer amount.

[0177] In this embodiment, self-healing is performed after the isolation strategy is generated, and specifically,

[0178] According to the isolation strategy, operation information is extracted, and a control instruction is issued according to the operation information to control the corresponding circuit breaker and switch.

[0179] After the control instruction is completed, the power distribution network topology structure model is updated according to the switch state.

[0180] Based on the updated topology structure model, a candidate set of recovery paths is constructed, and a shortest path algorithm is adopted to screen an optimal recovery path.

[0181] According to the optimal recovery path, a specific recovery scheme is generated.

[0182] It should be noted that, in this embodiment, a sparse classification model is constructed based on the influence section and topology, the feature parameters are quantified as vectors, the sparsity is realized through regularization and converted into linear programming for solving, the optimal strategy is quickly generated, the influence on the non-fault area is reduced, and the pain points of traditional isolation operation complexity and slow response are solved.

[0183] Embodiment 2, Figure 2 A power distribution network feeder fault automatic isolation and self-healing device is given, which comprises:

[0184] A real-time data acquisition and integration module is configured to obtain power grid operation data according to a power distribution network real-time operation report.

[0185] A fault influence section analysis module is configured to analyze the association between a fault data set and a power distribution network real-time topology structure through a reduction algorithm, and generate a fusion reduction basis.

[0186] A fault influence section screening module is configured to screen an influence section from a set of adjacent sections according to the fusion reduction basis.

[0187] A fault isolation and self-healing execution module is configured to construct a sparse classification model according to the influence section and the power distribution network real-time topology structure, optimize an isolation classifier through a linear programming algorithm, obtain an isolation strategy and execute it, and generate a recovery scheme.

[0188] Embodiment 3, Figure 3This invention discloses an automatic isolation and self-healing device for distribution network feeder faults, the electronic device comprising:

[0189] At least one processor; and an input / output interface communicatively connected to the at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the automatic isolation and self-healing method for distribution network feeder faults.

[0190] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0191] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0192] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0193] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0194] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0195] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for automatic isolation and self-healing of distribution network feeder faults, characterized in that, Includes the following steps: Based on the acquired power grid operation data, a set of adjacent sections is obtained; Based on the obtained fault severity and adjacent segment attributes, a fusion reduction basis is generated; Influenced segments are selected from the set of adjacent segments based on the fusion reduction criteria; Based on the affected sections and power grid topology, isolation strategies are obtained through sparse classification and self-healing is performed.

2. The automatic isolation and self-healing method for distribution network feeder faults according to claim 1, characterized in that, The set of adjacent sections is obtained based on the acquired power grid operation data, specifically as follows: Based on the real-time topology of the distribution network, determine the physical connection paths between sections; Identify the faulty section based on the faulty section identifier; Starting from the faulty section, the segments connected to the faulty section in the real-time topology of the distribution network are summarized to form a set of adjacent segments.

3. The automatic isolation and self-healing method for distribution network feeder faults according to claim 2, characterized in that, The step of generating a fusion reduction basis based on the obtained fault severity and adjacent segment attributes is as follows: Obtain the topological location, adjacency relationship, and operational status of adjacent segments, and construct conditional attributes; Based on the acquired fault data, calculate the fault severity, and establish decision attributes corresponding to the condition attributes based on the fault severity. Based on conditional attributes and decision attributes, a reduction algorithm is used to generate a fusion reduction basis.

4. The automatic isolation and self-healing method for distribution network feeder faults according to claim 3, characterized in that, The process of generating a fusion reduction basis based on conditional attributes and decision attributes using a reduction algorithm is as follows: Calculate the attribute dependency of the conditional attribute on the decision attribute; Based on attribute dependency, a heuristic search algorithm is used to remove redundant attributes and obtain the reduced conditional attributes. Based on the reduced conditional and decisional attributes, a rule set is generated as the basis for fusion reduction.

5. The automatic isolation and self-healing method for distribution network feeder faults according to claim 4, characterized in that, The attribute dependency of the calculation condition attribute on the decision attribute is specifically as follows: Equivalence classes are partitioned for conditional attributes and decision attributes respectively, and samples with the same attribute value are grouped into one class to obtain conditional equivalence classes and decision equivalence classes; Calculate attribute dependencies based on conditional equivalence classes and decision equivalence classes.

6. The automatic isolation and self-healing method for distribution network feeder faults according to claim 5, characterized in that, The process of selecting the affected segments from the set of adjacent segments based on the fusion reduction criteria is as follows: Retrieve the conditional attributes of each segment in the set of adjacent segments; Compare the conditional attributes of the segment with the fusion and reduction criteria to find the target criteria that match the conditional attributes of the segment. The decision attributes of the corresponding adjacent segments are determined based on the target criteria.

7. The automatic isolation and self-healing method for distribution network feeder faults according to claim 6, characterized in that, The isolation strategy, based on the affected section and power grid topology, is obtained through sparse classification and then self-healed. Specifically: Based on the affected section and the real-time topology of the distribution network, feature parameters are extracted and isolated feature vectors are constructed. Construct a sparse classification model and output the optimal isolation strategy; The optimal isolation strategy was validated by applying it to the real-time topology.

8. The automatic isolation and self-healing method for distribution network feeder faults according to claim 7, characterized in that, The construction of the sparse classification model and the output of the optimal isolation strategy are as follows: Feature parameters are extracted from the affected section and the real-time topology of the distribution network and quantized into isolated feature vectors; Construct an isolation classifier and achieve sparsity of model parameters by introducing a regularization term into the optimization objective of the classifier; The optimization objective with regularization terms is transformed into a linear programming problem, and the optimal model parameters are obtained by solving it. The final sparse classification model is constructed based on the optimal model parameters. The isolation feature vector of the affected segment is input into the sparse classification model, and the optimal isolation strategy is output.

9. An apparatus using the automatic isolation and self-healing method for distribution network feeder faults as described in any one of claims 1-8, comprising: The real-time data acquisition and integration module is used to obtain power grid operation data based on the real-time operation report of the distribution network; The fault impact section analysis module is used to analyze the correlation between the fault dataset and the real-time topology of the distribution network through a reduction algorithm, and generate a fusion reduction basis. The fault-affected segment filtering module is used to filter the affected segments from the set of adjacent segments based on the fusion reduction criteria. The fault isolation and self-healing execution module is used to construct a sparse classification model based on the affected section and the real-time topology of the distribution network, optimize the isolation classifier through a linear programming algorithm, obtain the isolation strategy and execute it to generate a recovery plan.

10. An electronic device, characterized in that, The electronic device includes: At least one processor; and an input / output interface communicatively connected to the at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the automatic isolation and self-healing method for distribution network feeder faults as described in any one of claims 1 to 8.

Images