Single-phase fault positioning and judging method based on fuzzy control and dynamic topology

Abstract

The invention discloses a single-phase fault positioning judgment method based on fuzzy control and dynamic topology, and the method comprises the steps: S1, constructing a dynamic graph structure of a power system, and extracting a target feature vector and a correlation feature set from the dynamic graph structure; s2, performing weighting processing on the target feature vector and the associated feature set through a cross attention mechanism to obtain weighted target features fused with associated information; s3, performing fuzzy control fault reasoning on the weighted target features to obtain initial fault possibilities of the candidate lines; and S4, verifying the initial fault possibility based on an electrical quantity rule constraint to obtain a final fault possibility of the candidate line. Therefore, the method has the advantages of adapting to the dynamic topology of the power grid, effectively fusing multi-source fault features, having excellent complex scene positioning reliability and the like.

Description

Technical Field

[0001] This invention relates to the field of power electronics technology, specifically to a method for single-phase fault location and judgment based on fuzzy control and dynamic topology. Background Technology

[0002] During the operation of the power distribution network, single-phase grounding faults are the most frequent type of fault due to factors such as line aging, meteorological disasters, external damage, and insulation failure.

[0003] Traditional fault location methods are represented by analysis methods based on steady-state electrical quantities, including the zero-sequence current amplitude comparison method, the zero-sequence power direction method, and the impedance method. Among these, the zero-sequence current amplitude comparison method relies on the difference in zero-sequence current amplitude between the faulty and non-faulty lines. However, in low-current grounding systems, the steady-state zero-sequence current amplitude of a single-phase ground fault is extremely small and easily affected by factors such as line distributed capacitance and load imbalance, resulting in insignificant amplitude differences and low location sensitivity. The zero-sequence power direction method distinguishes faulty lines by judging the phase relationship between zero-sequence current and zero-sequence voltage. However, in high-resistance grounding scenarios, the phase characteristics are ambiguous, easily leading to misjudgments. The impedance method locates the fault location based on the impedance calculation from the fault point to the measurement point. However, the distribution network topology is complex with many branches, and the uncertainty of grounding resistance severely affects the accuracy of impedance calculation, resulting in large location errors. Summary of the Invention

[0004] The present invention aims to at least partially solve the technical problems in the above-mentioned technologies.

[0005] Therefore, this invention discloses a single-phase fault location and judgment method based on fuzzy control and dynamic topology, comprising:

[0006] S1: Construct a dynamic graph structure of the power system, and extract the target feature vector and associated feature set from the dynamic graph structure;

[0007] S2: The target feature vector and the associated feature set are weighted by a cross-attention mechanism to obtain a weighted target feature that integrates associated information;

[0008] S3: Perform fuzzy control fault reasoning on the weighted target features to obtain the initial fault probability of the candidate line;

[0009] S4: Verify the initial fault probability based on electrical quantity regularity constraints to obtain the final fault probability of the candidate line.

[0010] The single-phase fault location and judgment method based on fuzzy control and dynamic topology disclosed in this invention has the advantages of adapting to the dynamic topology of the power grid, effectively integrating multi-source fault characteristics, and excellent location reliability in complex scenarios.

[0011] In addition, the single-phase fault location and judgment method based on fuzzy control and dynamic topology disclosed in this invention may also have the following additional technical features:

[0012] Further, in step S1, constructing the dynamic graph structure of the power system includes:

[0013] The primary equipment in the power system is mapped as a primary equipment node, the three-phase wires in the power system are mapped as main line edges, and each phase of the three-phase wires is mapped as a branch line edge. The main line edge and the branch line edge form a master-slave relationship.

[0014] The static attributes and dynamic operating status of the primary equipment node are collected, as are the static parameters and dynamic features of the main edge and the branch edge. Based on the dynamic operating status, the active status of the primary equipment node, the main edge, and the branch edge is updated to form a dynamic graph structure.

[0015] Further, in step S1, extracting the target feature vector and associated feature set from the dynamic graph structure includes:

[0016] The active main edge in the dynamic graph structure is taken as the target active main edge, and the static parameters and dynamic features of the target active main edge, as well as the static parameters and dynamic features of the associated active branch edge, are taken as the target feature vector.

[0017] The dynamic characteristics of the active primary device nodes at both ends of the target active main line edge, as well as the static parameters and dynamic characteristics of the active main line edges adjacent to the target active main line edge, are used as the associated feature set.

[0018] Further, in step S2, the cross-attention mechanism includes:

[0019] The target feature vector is used as the query feature, and the set of associated features is used as the key feature and the value feature. A linear transformation is performed on the query feature, the key feature and the value feature respectively to make the dimensions of the query feature, the key feature and the value feature the same.

[0020] Further, in step S2, the target feature vector and the associated feature set are weighted, including:

[0021] The similarity between the query feature and the key feature is calculated, and the similarity is normalized to obtain the attention weight. The value feature is then weighted and summed using the attention weight to obtain the weighted target feature that integrates the associated information.

[0022] Further, in step S3, the fuzzy control fault reasoning includes:

[0023] The weighted target features are preprocessed;

[0024] Sensitive core features are selected from the preprocessed weighted target features;

[0025] The sensitive core features are converted into fuzzy linguistic variables;

[0026] Define the fuzzy set and membership function of the fuzzy linguistic variable.

[0027] Furthermore, in step S3, the fuzzy control fault reasoning also includes:

[0028] A fuzzy rule base with attention weight threshold and the fuzzy language variable is constructed. Rule matching and trigger intensity calculation are performed on the fuzzy rule base. The matching results are defuzzified using the centroid method to obtain the initial fault probability of the candidate line.

[0029] Furthermore, in step S4, the electrical quantity regularity constraint includes:

[0030] Kirchhoff's current law constraints, impedance and current magnitude constraints, and positive correlation constraints between transient energy and steady-state current for zero-sequence networks;

[0031] The deviation rate is calculated based on the electrical quantity law constraints, and the initial fault probability is processed according to the deviation rate to obtain the final fault probability.

[0032] Furthermore, this single-phase fault location and judgment method based on fuzzy control and dynamic topology also includes:

[0033] S5: Verify the activity status and connectivity of the candidate lines through a search algorithm, and determine whether the candidate lines satisfy the topological logic.

[0034] Furthermore, this single-phase fault location and judgment method based on fuzzy control and dynamic topology also includes:

[0035] S6: Select the active main line edge with the largest final fault probability that satisfies the preset threshold among the candidate lines that satisfy the topology logic as the fault line, and select the active branch edge with the largest zero-sequence current that satisfies the impedance and current amplitude constraints among the fault lines as the fault phase.

[0036] Additional features and advantages of this invention will be set forth in the description which follows, or may be learned by practicing the invention. Attached Figure Description

[0037] The technical solution and beneficial effects of the present invention will become apparent and readily understood from the following description in conjunction with the accompanying drawings, wherein:

[0038] Figure 1 This is a flowchart of a single-phase fault location and judgment method based on fuzzy control and dynamic topology according to the present invention.

[0039] Figure 2 This is a flowchart of the dynamic graph construction process for the single-phase fault location and judgment method based on fuzzy control and dynamic topology of the present invention.

[0040] Figure 3 This is a schematic diagram of dynamic graph construction for the single-phase fault location and judgment method based on fuzzy control and dynamic topology of the present invention;

[0041] Figure 4 This is a flowchart of the cross-attention mechanism in the single-phase fault location and judgment method based on fuzzy control and dynamic topology of the present invention;

[0042] Figure 5 This is a schematic diagram of the cross-attention mechanism of the single-phase fault location and judgment method based on fuzzy control and dynamic topology of the present invention;

[0043] Figure 6 This is a flowchart of the fuzzy control fault reasoning process of the single-phase fault location and judgment method based on fuzzy control and dynamic topology of the present invention.

[0044] Figure 7 This is a membership function diagram of the single-phase fault location and judgment method based on fuzzy control and dynamic topology of the present invention. Detailed Implementation

[0045] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0046] The single-phase fault location and judgment method based on fuzzy control and dynamic topology disclosed in this invention will now be described with reference to the accompanying drawings.

[0047] This single-phase fault location and judgment method based on fuzzy control and dynamic topology, such as Figure 1 As shown, it includes:

[0048] S1: Construct a dynamic graph structure of the power system and extract the target feature vector and associated feature set from the dynamic graph structure;

[0049] S2: The target feature vector and the associated feature set are weighted by a cross-attention mechanism to obtain a weighted target feature that integrates associated information;

[0050] S3: Perform fuzzy control fault reasoning on the weighted target features to obtain the initial fault probability of the candidate line;

[0051] S4: Verify the initial fault probability based on electrical quantity regularity constraints to obtain the final fault probability of the candidate line.

[0052] In one embodiment of the present invention, such as Figure 2 As shown:

[0053] In step S1, the dynamic graph structure of the power system is constructed, including:

[0054] The primary equipment in the power system is mapped to primary equipment nodes, the three-phase wires in the power system are mapped to main line edges, and each phase of the three-phase wires is mapped to a branch line edge. The main line edge and the branch line edge form a master-slave relationship.

[0055] The static attributes and dynamic operating status of primary equipment nodes are collected, as well as the static parameters and dynamic characteristics of main and branch edges. Based on the dynamic operating status, the active status of primary equipment nodes, main edges, and branch edges is updated to form a dynamic graph structure.

[0056] Extracting target feature vectors and associated feature sets from dynamic graph structures, including:

[0057] The active main edge in the dynamic graph structure is taken as the target active main edge, and the static parameters and dynamic features of the target active main edge, as well as the static parameters and dynamic features of the associated active branch edges, are taken as the target feature vector.

[0058] The dynamic characteristics of the active primary device nodes at both ends of the target active main line edge, as well as the static parameters and dynamic characteristics of the active main line edges adjacent to the target active main line edge, are used as the associated feature set.

[0059] Specifically, the static attributes of primary equipment nodes include equipment type, rated voltage, rated current, physical location coordinates, and unique identifier ID. The dynamic operating status is divided into two states: running and shutdown. These are collected in real time through the power system's SCADA system or RTU remote terminal unit. The collected signals need to carry the equipment's unique identifier ID and status change timestamp.

[0060] The static parameters of the main line side include the main line length, the main line positive sequence impedance, the main line zero sequence impedance, and the unique identifier ID. The dynamic characteristics include the main line transient energy.

[0061] The static parameters of the branch include the positive sequence impedance, the zero sequence impedance, the A, B, C phase identifiers and the unique phase identifier ID. The dynamic characteristics include the zero sequence current, the zero sequence voltage, and the high-frequency component of the transient current of the branch.

[0062] The dynamic graph structure update adopts two mechanisms: event triggering and timed triggering. Event triggering is triggered immediately when the SCADA system or RTU remote terminal unit uploads a signal indicating a change in the operating status of a primary device. This updates the dynamic operating status of the corresponding primary device node to running or stopped, and synchronously updates the active status of the corresponding main line edge and branch line edge. Specifically, the main line edge and its corresponding branch line edge are active only when both primary device nodes at both ends of the main line edge are in a running state; otherwise, they are in an inactive state. Timed triggering is set to a fixed time period of 1 second to periodically synchronize the current operating status of all primary devices.

[0063] The static parameters and dynamic characteristics of the active main line edge, as well as the static parameters and dynamic characteristics of the associated active branch edges, are used as the target feature vector. In this embodiment, the target feature vector specifically consists of the main line length, main line positive sequence impedance, main line zero sequence impedance, main line transient energy, A-phase branch positive sequence impedance, A-phase branch zero sequence impedance, A-phase branch zero sequence current, A-phase branch zero sequence voltage, A-phase branch transient current high-frequency component, B-phase branch positive sequence impedance, B-phase branch zero sequence impedance, B-phase branch zero sequence current, B-phase branch zero sequence voltage, B-phase branch transient current high-frequency component, C-phase branch positive sequence impedance, C-phase branch zero sequence impedance, C-phase branch zero sequence current, C-phase branch zero sequence voltage, and C-phase branch transient current high-frequency component.

[0064] The associated feature set consists of two parts. One part specifically includes the bus zero-sequence voltage of the primary equipment nodes at both ends of the target active main line edge. The other part specifically includes the main line length, main line positive-sequence impedance, main line zero-sequence impedance, main line transient energy of all active main line edges adjacent to the target active main line edge, as well as the associated branch edge's A-phase branch zero-sequence current, A-phase branch zero-sequence voltage, A-phase branch transient current high-frequency component, B-phase branch zero-sequence current, B-phase branch zero-sequence voltage, B-phase branch transient current high-frequency component, C-phase branch zero-sequence current, C-phase branch zero-sequence voltage, and C-phase branch transient current high-frequency component.

[0065] like Figure 3 As shown, DOT1 to DOT10 are primary equipment nodes formed by mapping primary equipment in the power system. Nodes marked with black dots represent active nodes in operation, while white nodes represent inactive nodes in shutdown state.

[0066] L101 to L201 are the main line sides formed by the mapping of three-phase wires. The solid main line side is active because the primary equipment nodes associated with both ends are active. Therefore, the main line side and the corresponding branch line side are also active. The dashed main line side is inactive because at least one of the primary equipment nodes associated with it is inactive. Therefore, it is inactive as a whole and does not participate in subsequent feature extraction and fault judgment.

[0067] Specifically, L109 between DOT3 and DOT8 is the main line edge, and the L109A, L109B, and L109C below it correspond to phases A, B, and C of the three-phase wire, respectively, which are the branch lines formed by mapping. The three of them form a clear master-slave relationship with the main line edge L109. The active state of the main line edge directly determines the active attribute of the branch line edge.

[0068] Taking the active mainline edge L109 as an example, its target feature vector specifically includes the mainline length, mainline positive sequence impedance, mainline zero sequence impedance, mainline transient energy, and the branch positive sequence impedance, branch zero sequence impedance, zero sequence current, zero sequence voltage, and high-frequency components of transient current of L109A, L109B, and L109C respectively.

[0069] Its associated feature set covers two parts: first, the bus zero-sequence voltage of the two active nodes DOT3 and DOT8; and second, the main line length, main line positive-sequence impedance, main line zero-sequence impedance, main line transient energy of the active main line side adjacent to L109, as well as the A, B, and C phase zero-sequence current, zero-sequence voltage, and high-frequency components of transient current of the associated branch sides of these adjacent main line sides.

[0070] For the remaining technical details of this embodiment, please refer to the above technical details, which will not be repeated here.

[0071] In another embodiment of the invention, such as Figure 4 As shown:

[0072] In step S2, the cross-attention mechanism includes:

[0073] The target feature vector is used as the query feature, and the associated feature set is used as the key feature and value feature. A linear transformation is performed on the query feature, key feature and value feature respectively to make the query feature, key feature and value feature have the same dimension.

[0074] The target feature vector and the associated feature set are weighted, including:

[0075] Calculate the similarity between query features and key features, normalize the similarity to obtain attention weights, and use the attention weights to perform a weighted summation of value features to obtain weighted target features that fuse related information.

[0076] Specifically, such as Figure 5 As shown, firstly, linear transformations are performed on the query features, key features, and value features respectively to make the dimensions of the query features, key features, and value features the same.

[0077] The similarity between query features and key features is obtained by performing a dot product operation to obtain a similarity matrix. Each element in the similarity matrix represents the degree of correlation between the corresponding query features and key features.

[0078] It should be noted that in this embodiment, in order to avoid the gradient vanishing due to excessively large values ​​in the similarity matrix, the similarity matrix can be scaled using a scaling factor.

[0079] Next, the similarity matrix is ​​normalized using an activation function to obtain the attention weight matrix. Based on the attention weight matrix, the value features are weighted and summed to obtain the weighted target features.

[0080] In addition, it should be noted that in this embodiment, there is strong electrical coupling between zero-sequence voltage and zero-sequence current. Therefore, the sum of the correlation weights of zero-sequence voltage and zero-sequence current should be greater than or equal to 0.5. If the sum of the correlation weights of zero-sequence voltage and zero-sequence current is less than 0.5, then the ratio of 0.5 to the sum of correlation weights is multiplied by the sum of correlation weights to obtain the final sum of correlation weights.

[0081] like Figure 3 As shown, firstly, the target feature vector corresponding to L109 is used as the query feature. This vector specifically includes the static parameters and dynamic features of L109, as well as the static parameters and dynamic features of L109A, B, and C. The initial dimension of this query feature is 19.

[0082] The associated feature set is used as the key feature and value feature. The associated feature set specifically includes two parts: one part is the dynamic features of the active nodes DOT3 and DOT8 at both ends of L109, and the other part is the static parameters and dynamic features of the active main line edges L107 and L108 adjacent to L109, as well as the dynamic features of the associated branch edges of L107 and L108. After being sorted, the associated feature set forms two key-value pairs, and the initial dimension of each key feature and value feature is 16 dimensions.

[0083] Next, linear transformations are performed on the query features, key features, and value features respectively. Through three independent fully connected layers, the query features are mapped from 19 dimensions to 64 dimensions, and the key features and value features are mapped from 16 dimensions to 64 dimensions respectively, ensuring that the three dimensions are consistent and meet the input requirements for similarity calculation.

[0084] Then, the similarity between the query feature and the key feature is calculated. The dot product operation is used to multiply the 64-dimensional query feature and the 64-dimensional key feature element by element and then sum them to obtain the degree of association between each key feature and the query feature.

[0085] To avoid gradient vanishing due to excessively large values, a scaling factor is introduced, dividing all dot product results by 8 to obtain a scaled similarity matrix.

[0086] The similarity matrix was then normalized. The SoftMax activation function was used to calculate the attention weights corresponding to each key feature. At this point, the sum of the association weights of zero-sequence voltage and zero-sequence current was checked. The sum of the zero-sequence voltage weights of DOT3 and DOT8 buses and the zero-sequence current weights of L109A, B, and C was evaluated to determine if it satisfied the strong electrical coupling constraint. If so, no weight adjustment was needed.

[0087] Finally, the value features are weighted and summed using attention weights. Each value feature is multiplied by its corresponding attention weight, and then all weighted value features are summed element-wise to obtain a weighted target feature that incorporates correlation information. This vector retains the core fault features of L109 and L109A, B, and C themselves, while also incorporating the operating status of adjacent lines and the electrical characteristics of terminal equipment.

[0088] For the remaining technical details of this embodiment, please refer to the above technical details, which will not be repeated here.

[0089] In another embodiment of the invention, such as Figure 6 As shown:

[0090] In step S3, fuzzy control fault reasoning includes:

[0091] Preprocess the weighted target features;

[0092] Select sensitive core features from the preprocessed weighted target features;

[0093] Convert sensitive core features into fuzzy linguistic variables;

[0094] Define the fuzzy set and membership function for fuzzy linguistic variables.

[0095] A fuzzy rule base with attention weight thresholds and fuzzy linguistic variables is constructed. Rule matching and trigger intensity calculation are performed on the fuzzy rule base. The matching results are defuzzified using the centroid method to obtain the initial fault probability of the candidate lines.

[0096] Specifically, four core features most sensitive to single-phase grounding faults are selected from the weighted target features and used as inputs for fuzzy inference: transient energy on the main line, zero-sequence current on the branch line, zero-sequence voltage on the branch line, and high-frequency component of transient current on the branch line.

[0097] It should be noted that, in this embodiment, in order to facilitate the subsequent data processing and use of the high-frequency components of the transient energy, zero-sequence current, zero-sequence voltage, and transient current of the main line, the high-frequency components of the transient energy, zero-sequence current, zero-sequence voltage, and transient current of the main line can be linearly normalized within the interval [0,2].

[0098] Next, the fuzzy sets in Table 1 are converted into fuzzy linguistic variables. The contents of Table 1 are as follows:

[0099] Fuzzy set abbreviation Fuzzy linguistic variables Physical meaning Applicable Scenarios VS extremely small Characteristic values ​​are close to normal operating levels with no signs of faults. Non-faulty lines S smaller Slight deviations in characteristic values ​​from normal may be due to induced interference. Non-faulty line Faulty line remote end M medium The characteristic value deviates significantly from the normal range, suggesting a possible malfunction. High-resistance grounding fault in the middle section of the fault line L Larger Characteristic values ​​that deviate significantly from normal values ​​indicate a high probability of failure. Low-resistance grounding fault near the fault line VL great Characteristic values ​​far exceeding normal levels confirm fault correlation Metallic grounding fault near the fault point

[0100] Table 1

[0101] It should also be noted that in this embodiment, the domain of the membership function is normalized to the interval [0,2], such as... Figure 7 As shown, the specific formula is:

[0102] (1) Membership function of transient energy of principal edge

[0103] ;

[0104] ;

[0105] ;

[0106] ;

[0107] .

[0108] (2) Membership function of zero-sequence current on branch side

[0109] ;

[0110] ;

[0111] ;

[0112] ;

[0113] .

[0114] (3) Membership function of zero-sequence voltage on branch side

[0115] ;

[0116] ;

[0117] ;

[0118] ;

[0119] .

[0120] (4) Membership function of high-frequency components of transient current on branch side

[0121] ;

[0122] ;

[0123] ;

[0124] ;

[0125] .

[0126] The fuzzy rule base contains 25 rules. The antecedent of each rule is an attention weight (threshold) and a fuzzy linguistic variable, and the consequent is the probability of failure. See Table 2 for details.

[0127] Serial Number Priority Antecedent After 1 1 Transient energy of the main line side (fuzzy linguistic variable = VL and attention weight ≥ 0.7) and zero-sequence current of the branch line side (fuzzy linguistic variable = VL and attention weight ≥ 0.6). Extremely high 2 1 Transient energy of the main line side (fuzzy linguistic variable = L and attention weight ≥ 0.65) and zero-sequence voltage of the branch side (fuzzy linguistic variable = VL and attention weight ≥ 0.6). Extremely high 3 2 Transient energy of the main line side (fuzzy linguistic variable = L and attention weight ≥ 0.6), zero-sequence current of the branch line side (fuzzy linguistic variable = L and attention weight ≥ 0.55), zero-sequence voltage of the branch line side (fuzzy linguistic variable = L). Extremely high 4 2 High-frequency component of transient current on branch side (fuzzy linguistic variable = VL and attention weight ≥ 0.7) and transient energy on main side (fuzzy linguistic variable = L and attention weight ≥ 0.5) Extremely high 5 3 Transient energy of the main line (fuzzy linguistic variable = M and attention weight ≥ 0.55), zero-sequence current of the branch line (fuzzy linguistic variable = L and attention weight ≥ 0.5), zero-sequence voltage of the branch line (fuzzy linguistic variable = M). high 6 3 High-frequency component of transient current on branch side (fuzzy linguistic variable = L and attention weight ≥ 0.65) and zero-sequence voltage on branch side (fuzzy linguistic variable = L and attention weight ≥ 0.5) high 7 3 Transient energy of the main line side (fuzzy linguistic variable = L and attention weight ≥ 0.5), zero-sequence current of the branch line side (fuzzy linguistic variable = M and attention weight ≥ 0.45), high-frequency component of transient current of the branch line side (fuzzy linguistic variable = M). high 8 4 Transient energy of the main line side (fuzzy linguistic variable = M and attention weight ≥ 0.5), zero-sequence voltage of the branch line side (fuzzy linguistic variable = M and attention weight ≥ 0.45), high-frequency component of transient current of the branch line side (fuzzy linguistic variable = L). Medium and high 9 4 Branch-side zero-sequence current (fuzzy linguistic variable = M and attention weight ≥ 0.5), branch-side transient current high-frequency component (fuzzy linguistic variable = M and attention weight ≥ 0.5), main-side transient energy (fuzzy linguistic variable = M). Medium and high 10 4 Branch-side zero-sequence voltage (fuzzy linguistic variable = L and attention weight ≥ 0.55), main-side transient energy (fuzzy linguistic variable = M and attention weight ≥ 0.4), branch-side zero-sequence current (fuzzy linguistic variable = S). Medium and high 11 5 Transient energy of the main line (fuzzy linguistic variable = M and attention weight ≥ 0.45), zero-sequence current of the branch line (fuzzy linguistic variable = M and attention weight ≥ 0.4), zero-sequence voltage of the branch line (fuzzy linguistic variable = S). middle 12 5 High-frequency component of transient current on branch side (fuzzy linguistic variable = M and attention weight ≥ 0.5), zero-sequence voltage on branch side (fuzzy linguistic variable = M and attention weight ≥ 0.4), transient energy on main line side (fuzzy linguistic variable = S). middle 13 5 Transient energy of the main line side (fuzzy linguistic variable = S and attention weight ≥ 0.4), zero-sequence current of the branch line side (fuzzy linguistic variable = M and attention weight ≥ 0.45), high-frequency component of transient current of the branch line side (fuzzy linguistic variable = S). middle 14 6 Transient energy of the main line (fuzzy linguistic variable = S and attention weight ≥ 0.4), zero-sequence current of the branch line (fuzzy linguistic variable = S and attention weight ≥ 0.35), zero-sequence voltage of the branch line (fuzzy linguistic variable = M). medium to low 15 6 High-frequency component of transient current on branch side (fuzzy linguistic variable = S and attention weight ≥ 0.45), zero-sequence voltage on branch side (fuzzy linguistic variable = S and attention weight ≥ 0.35), transient energy on main line side (fuzzy linguistic variable = M). medium to low 16 6 Transient energy of the main line side (fuzzy linguistic variable = M and attention weight ≥ 0.35), zero-sequence current of the branch line side (fuzzy linguistic variable = S and attention weight ≥ 0.3), high-frequency component of transient current of the branch line side (fuzzy linguistic variable = S). medium to low 17 7 Transient energy of the main line side (fuzzy linguistic variable = S and attention weight ≥ 0.3), zero-sequence current of the branch line side (fuzzy linguistic variable = S and attention weight ≥ 0.3), zero-sequence voltage of the branch line side (fuzzy linguistic variable = S). Low 18 7 High-frequency component of transient current on branch side (fuzzy linguistic variable = S and attention weight ≥ 0.35), transient energy on main line side (fuzzy linguistic variable = VS and attention weight ≥ 0.25), zero-sequence voltage on branch side (fuzzy linguistic variable = S). Low 19 7 Branch-side zero-sequence voltage (fuzzy linguistic variable = S and attention weight ≥ 0.3), branch-side zero-sequence current (fuzzy linguistic variable = VS and attention weight ≥ 0.25), main-side transient energy (fuzzy linguistic variable = S). Low 20 8 Transient energy on the main line (fuzzy linguistic variable = VS), zero-sequence current on the branch line (fuzzy linguistic variable = VS), zero-sequence voltage on the branch line (fuzzy linguistic variable = VS), high-frequency component of transient current on the branch line (fuzzy linguistic variable = VS). Extremely low 21 8 Transient energy of the main line side (fuzzy linguistic variable = VS and attention weight < 0.2) and zero-sequence voltage of the branch side (fuzzy linguistic variable = VS and attention weight < 0.2). Extremely low 22 8 High-frequency component of transient current on branch side (fuzzy linguistic variable = VS and attention weight < 0.25) and zero-sequence voltage on branch side (fuzzy linguistic variable = VS and attention weight < 0.25) Extremely low 23 9 Transient energy of the main line side (fuzzy linguistic variable = L and attention weight ≥ 0.5) and zero-sequence current of the branch line side (fuzzy linguistic variable = VS and attention weight < 0.3). medium to low 24 9 Branch-side zero-sequence voltage (fuzzy linguistic variable = L and attention weight ≥ 0.5) and branch-side transient current high-frequency component (fuzzy linguistic variable = VS and attention weight < 0.3). medium to low 25 9 Branch-side zero-sequence current (fuzzy linguistic variable = L and attention weight ≥ 0.5) and main-side transient energy (fuzzy linguistic variable = VS and attention weight < 0.3). medium to low

[0128] Table 2

[0129] It should be noted that two checks are performed when matching rules. The first check is to check whether the attention weight meets the rule threshold, and the second check is to check the fuzzy state. Unless both are met, the rule is considered to be triggered.

[0130] The trigger strength is calculated using the minimum-maximum method. Specifically, the trigger strength is the product of the minimum membership value among the high-frequency components of the transient energy, zero-sequence current, zero-sequence voltage, and transient current of the main line side and the maximum attention weight among the high-frequency components of the transient energy, zero-sequence current, zero-sequence voltage, and transient current of the main line side and the branch side. Membership values ​​and attention weights not involved are all set to 1.

[0131] For defuzzification, the centroid method is used. First, a centroid value is assigned to each fuzzy linguistic variable, which is: extremely low = 0.1, low = 0.3, medium-low = 0.5, medium = 0.6, medium-high = 0.7, high = 0.8 and extremely high = 0.95. The initial fault probability is obtained by the ratio between the sum of the products of the trigger intensity and the centroid value and the sum of the trigger intensity.

[0132] like Figure 3 As shown, firstly, the weighted target features are preprocessed. For the indicators related to fault sensitivity in the weighted target features corresponding to L109, four core features are retained: main line transient energy, branch line zero-sequence current, branch line zero-sequence voltage, and high-frequency component of branch line transient current. Redundant static parameters are removed.

[0133] Meanwhile, perform a secondary normalization check on these 4 types of core features to ensure that their values all fall within the range of [0, 2]. For example, determine the sensitive core features and their values for input fuzzy inference as follows: the normalized value of the transient energy on the main line side of L109 is 1.3, the normalized value of the zero-sequence current on the branch line side of L109A is 1.6, the normalized value of the zero-sequence voltage on the branch line side of L109A is 1.5, and the normalized value of the high-frequency component of the transient current on the branch line side of L109A is 1.4. Another example is that the core feature values of phases B and C of L109 are both lower than 0.35, with low fault sensitivity, and they are not included in the core inference input for the time being.

[0134] Secondly, convert the sensitive core features into fuzzy language variables, and calculate the corresponding fuzzy language variables and membership degrees for each core feature respectively according to the fuzzy set definitions in Table 1 and Figure 7 the membership functions.

[0135] (1) Transient energy on the main line side (1.3):

[0136] Query the membership function of the transient energy on the main line side. Therefore, the membership degree of 1.3 is (1.3 - 1.0) / 0.4 = 0.75, and the fuzzy language variable is "L (Large)";

[0137] (2) Zero-sequence current on the branch line side of L109A (1.6):

[0138] Query the membership function of the zero-sequence current on the branch line side. For the interval 1.5 < x < 1.9, it corresponds to "VL (Very Large)", and the membership degree is (1.6 - 1.5) / 0.4 = 0.25. At the same time, it also satisfies the membership degree of "L (Large)" which is (1.9 - 1.6) / 0.4 = 0.75. Take the core fuzzy language variable as "VL (Very Large)";

[0139] (3) Zero-sequence voltage on the branch line side of L109A (1.5):

[0140] Query the membership function of the zero-sequence voltage on the branch line side. For the interval 1.45 < x < 1.85, it corresponds to "L (Large)", and the membership degree is (1.85 - 1.5) / 0.4 = 0.875, and the fuzzy language variable is "L (Large)";

[0141] (4) High-frequency component of the transient current on the branch line side of L109A (1.4):

[0142] Query the membership function of the high-frequency component of the transient current on the branch line side. For the interval 1.0 < x < 1.4, it corresponds to "L (Large)", and the membership degree is (1.4 - 1.0) / 0.4 = 1.0, and the fuzzy language variable is "L (Large)".

[0143] Next, define the fuzzy sets and membership functions of the fuzzy language variables, and adopt the 5 types of fuzzy sets of VS (Very Small), S (Small), M (Medium), L (Large), and VL (Very Large) preset in Table 1.

[0144] Subsequently, a fuzzy rule base with attention weight thresholds was constructed and rule matching was performed. Combining the attention weights output by the cross-attention mechanism (mainline transient energy weight 0.68, L109A branch zero-sequence current weight 0.65, L109A branch zero-sequence voltage weight 0.62, L109A branch transient current high-frequency component weight 0.60), valid triggering rules were matched from 25 fuzzy rules.

[0145] (1) Rule 2 (Priority 1):

[0146] The antecedent is "the transient energy of the main side (fuzzy linguistic variable = L and attention weight ≥ 0.65) + the zero-sequence voltage of the branch side (fuzzy linguistic variable = VL and attention weight ≥ 0.6)", and the consequent is "extremely high".

[0147] Verification: The transient energy fuzzy state of the main line side = L, weight 0.68 ≥ 0.65, the zero-sequence voltage fuzzy state of the branch side = L (approximately matches the VL threshold boundary, membership degree 0.875 close to the VL interval), weight 0.62 ≥ 0.6, which meets the rule triggering condition;

[0148] (2) Rule 3 (priority 2):

[0149] The antecedent is "transient energy on the main line side (L and ≥0.6) + zero-sequence current on the branch line side (L and ≥0.55) + zero-sequence voltage on the branch line side (L)", and the consequent is "extremely high".

[0150] Verification: The transient energy weight of the main line side is 0.68≥0.6, the fuzzy state of the zero-sequence current of the branch line side is VL (including the fuzzy feature of L) with a weight of 0.65≥0.55, and the fuzzy state of the zero-sequence voltage of the branch line side is L, which satisfies the triggering condition.

[0151] Then, using the min-max method, the trigger strength of rule 2 is min(mainline transient energy membership degree 0.75, branch zero-sequence voltage membership degree 0.875) × max(weights 0.68, 0.62) = 0.75 × 0.68 = 0.51, and the trigger strength of rule 3 is min(0.75, 0.75, 0.875) × max(0.68, 0.65, 1.0) = 0.75 × 0.68 = 0.51 (the attention weight not involved is taken as 1).

[0152] Finally, using the centroid method for defuzzification, a center value of 0.95 is assigned to the consequent of the fuzzy rule "extremely high" and 0.8 is assigned to "high". The center value of the triggering rule is weighted and summed with the triggering intensity, and then divided by the total sum of triggering intensities. That is, the initial fault probability = (0.95 × 0.51 + 0.95 × 0.51) / (0.51 + 0.51) = 0.95. Thus, the initial fault probability of the main line edge L109 and the branch line edge L109A is 0.95, which is significantly higher than the fault suspicion threshold.

[0153] For the remaining technical details of this embodiment, please refer to the above technical details, which will not be repeated here.

[0154] In another embodiment of the invention:

[0155] In step S4, the electrical quantity regularity constraints include:

[0156] Kirchhoff's current law constraints, impedance and current magnitude constraints, and positive correlation constraints between transient energy and steady-state current for zero-sequence networks;

[0157] The deviation rate is calculated based on the constraints of electrical quantity laws, and the initial fault probability is processed according to the deviation rate to obtain the final fault probability.

[0158] Specifically, the calculation and verification logic is based on the constraints of three electrical quantities.

[0159] For the Kirchhoff current law constraint of zero-sequence network, under normal circumstances, the algebraic sum of the zero-sequence currents of all associated branches of the bus node where the fault point is located should be approximately 0. Therefore, the sum of the zero-sequence currents of the associated branches of the target active main line edge and the sum of the zero-sequence currents of other active branches associated with the node is calculated as the algebraic sum of the node's zero-sequence currents. At the same time, the ratio of the algebraic sum of the node's zero-sequence currents to the maximum zero-sequence current of the associated branches of the target active main line edge is calculated as the balance deviation rate.

[0160] For impedance and current amplitude constraints, under normal circumstances, the amplitude of the zero-sequence current on the branch side and the zero-sequence voltage on the branch side should satisfy Ohm's law. Therefore, the ratio of the zero-sequence voltage on the branch side to the zero-sequence impedance on the branch side is calculated as the theoretical zero-sequence current. At the same time, the ratio of the difference between the actual collected zero-sequence current on the branch side and the theoretical zero-sequence current to the theoretical zero-sequence current is calculated as the amplitude deviation rate.

[0161] For the positive correlation constraint between transient energy and steady-state current, under normal circumstances, the transient energy of the main line side and the steady-state value of the zero-sequence current of the branch line side should satisfy a positive correlation. Therefore, the ratio of the transient energy of the main line side to the steady-state value of the zero-sequence current of the branch line side is calculated as the energy-current matching degree.

[0162] The final failure probability is calculated as the product of the initial failure probability, 1 minus the balance deviation rate, 1 minus the amplitude deviation rate, and 1 minus the energy-current matching degree. If the final failure probability is less than 0.4, the corresponding target active main line edge is removed from the candidate line.

[0163] like Figure 3 As shown, during normal operation, the effective value deviation of the current on the three-phase branches of L09A, B, and C should be less than or equal to 5%, and the zero-sequence current should be less than or equal to 5% of the rated current. However, during a fault, such as when phase A is grounded, the phase A current increases significantly, while the phase B and C currents remain basically unchanged. Therefore, the zero-sequence current is approximately equal to the phase A fault current.

[0164] Secondly, the phase difference between zero-sequence current and zero-sequence voltage should be close to 180 degrees, with a deviation range of less than 15%.

[0165] The difference between the zero-sequence voltages of the DOT3 and DOT8 busbars should satisfy Ohm's law with the zero-sequence impedance and zero-sequence current of the L109 main line, with an error within 10%.

[0166] The transient energy of the main line edge L109 should be greater than the transient energy of the adjacent main line edges L107 and L108, and the energy ratio should be greater than 3 times.

[0167] Next, the measured electrical quantity data needs to be input and verified:

[0168] Assuming the measured effective current of L109A is 120A, L109B is 31A, and L109C is 30A, the three-phase deviation reaches 74% (far greater than 5%), the zero-sequence current is (120+31+30) / 3≈60.5A, and the zero-sequence current is approximately equal to 50.4% of the measured effective current of L109A, which satisfies the three-phase imbalance constraint.

[0169] The zero-sequence voltage of the L109A branch is 380V (0 degrees phase), the zero-sequence current is 60.5A (175 degrees phase), and the phase difference is 175 degrees, which is within the range of 180 degrees ± 15%, thus meeting the zero-sequence component phase constraint.

[0170] The zero-sequence voltage of DOT3 bus is 420V, the zero-sequence voltage of DOT8 bus is 40V, ΔU0=380V, the zero-sequence impedance of L109 main line is 3.84Ω, the theoretical value of ΔU0 is calculated to be 353V, the error is (380-353) / 353≈7.6%, which meets the constraints.

[0171] The transient energy of the main line edge L109 is 1.3 (normalized value), L107 is 0.4, and L108 is 0.38, with ratios of 3.25:1 and 3.42:1 respectively. All of them are ≥3:1, which is consistent with the characteristics of transient energy concentration.

[0172] Finally, based on the comprehensive constraint verification results: the four core electrical quantity rules are fully satisfied, the overall verification is passed, and the final output fault conclusion is "a single-phase grounding fault occurred on the A phase branch (L109A) of the DOT3-DOT8 main line side L109, with a fault probability of 0.95 and a confidence level of 'extremely high'".

[0173] For the remaining technical details of this embodiment, please refer to the above technical details, which will not be repeated here.

[0174] In addition, it should be noted that:

[0175] This single-phase fault location and judgment method based on fuzzy control and dynamic topology also includes:

[0176] S5: Verify the activity status and connectivity of candidate lines through a search algorithm to determine whether the candidate lines satisfy the topological logic.

[0177] Specifically, in this embodiment, a depth-first search algorithm or a breadth-first search algorithm can be used.

[0178] First, traverse the main line edges and associated branch edges corresponding to the candidate lines in the dynamic graph structure, determine their active state to be active, and ensure that all associated primary device nodes are in running state. If these conditions are not met, it is determined that the topology logic is not met.

[0179] Next, starting from the faulty bus node, traverse all reachable active nodes along the active edges to form a faulty connected subgraph. Determine whether the main line edge corresponding to the candidate line is completely contained in the faulty connected subgraph. If not, it is determined that the topology logic is not satisfied.

[0180] Finally, if a candidate line fails any of the above checks, the associated feature set of the line is re-extracted, and steps S2 to S4 are returned to recalculate the final probability of failure. If the topology logic is still not satisfied after recalculation, the line is completely ruled out as a potential failure.

[0181] In addition, it should be noted that:

[0182] This single-phase fault location and judgment method based on fuzzy control and dynamic topology also includes:

[0183] S6: Select the active main line edge with the largest final fault probability that satisfies the preset threshold among the candidate lines that meet the topology logic as the fault line, and select the active branch edge with the largest zero-sequence current that satisfies the impedance and current amplitude constraints among the fault lines as the fault phase.

[0184] Specifically, candidate lines with a final failure probability ≥ 0.8 are selected, and the active main line edge with the largest value is selected as the faulty line.

[0185] As for the determination of the faulty phase, first determine that the zero-sequence current of the branch side is the largest among the three associated branches of the faulty line, and then determine that the amplitude deviation rate of the branch side is less than 0.4. If both conditions are met, then the branch side is the faulty phase.

[0186] In summary, the single-phase fault location and judgment method based on fuzzy control and dynamic topology disclosed in this invention has advantages such as adapting to the dynamic topology of the power grid, effectively integrating multi-source fault characteristics, and excellent location reliability in complex scenarios.

[0187] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A single-phase fault location and judgment method based on fuzzy control and dynamic topology, characterized in that, include: S1: Construct a dynamic graph structure of the power system, and extract the target feature vector and associated feature set from the dynamic graph structure; S2: The target feature vector and the associated feature set are weighted by a cross-attention mechanism to obtain a weighted target feature that integrates associated information; S3: Perform fuzzy control fault reasoning on the weighted target features to obtain the initial fault probability of the candidate line; S4: Verify the initial fault probability based on electrical quantity regularity constraints to obtain the final fault probability of the candidate line.

2. The single-phase fault location and judgment method based on fuzzy control and dynamic topology as described in claim 1, characterized in that, In step S1, constructing the dynamic graph structure of the power system includes: The primary equipment in the power system is mapped as a primary equipment node, the three-phase wires in the power system are mapped as main line edges, and each phase of the three-phase wires is mapped as a branch line edge. The main line edge and the branch line edge form a master-slave relationship. The static attributes and dynamic operating status of the primary equipment node are collected, as are the static parameters and dynamic features of the main edge and the branch edge. Based on the dynamic operating status, the active status of the primary equipment node, the main edge, and the branch edge is updated to form a dynamic graph structure.

3. The single-phase fault location and judgment method based on fuzzy control and dynamic topology as described in claim 2, characterized in that, In step S1, the target feature vector and associated feature set are extracted from the dynamic graph structure, including: The active main edge in the dynamic graph structure is taken as the target active main edge, and the static parameters and dynamic features of the target active main edge, as well as the static parameters and dynamic features of the associated active branch edge, are taken as the target feature vector. The dynamic characteristics of the active primary device nodes at both ends of the target active main line edge, as well as the static parameters and dynamic characteristics of the active main line edges adjacent to the target active main line edge, are used as the associated feature set.

4. The single-phase fault location and judgment method based on fuzzy control and dynamic topology as described in claim 3, characterized in that, In step S2, the cross-attention mechanism includes: The target feature vector is used as the query feature, and the set of associated features is used as the key feature and the value feature. A linear transformation is performed on the query feature, the key feature and the value feature respectively to make the dimensions of the query feature, the key feature and the value feature the same.

5. The single-phase fault location and judgment method based on fuzzy control and dynamic topology as described in claim 4, characterized in that, In step S2, the target feature vector and the associated feature set are weighted, including: The similarity between the query feature and the key feature is calculated, and the similarity is normalized to obtain the attention weight. The value feature is then weighted and summed using the attention weight to obtain the weighted target feature that integrates the associated information.

6. The single-phase fault location and judgment method based on fuzzy control and dynamic topology as described in claim 5, characterized in that, In step S3, fuzzy control fault reasoning includes: The weighted target features are preprocessed; Sensitive core features are selected from the preprocessed weighted target features; The sensitive core features are converted into fuzzy linguistic variables; Define the fuzzy set and membership function of the fuzzy linguistic variable.

7. The single-phase fault location and judgment method based on fuzzy control and dynamic topology as described in claim 6, characterized in that, In step S3, the fuzzy control fault reasoning also includes: A fuzzy rule base with attention weight threshold and the fuzzy language variable is constructed. Rule matching and trigger intensity calculation are performed on the fuzzy rule base. The matching results are defuzzified using the centroid method to obtain the initial fault probability of the candidate line.

8. The single-phase fault location and judgment method based on fuzzy control and dynamic topology as described in claim 7, characterized in that, In step S4, the electrical quantity regularity constraints include: Kirchhoff's current law constraints, impedance and current magnitude constraints, and positive correlation constraints between transient energy and steady-state current for zero-sequence networks; The deviation rate is calculated based on the electrical quantity law constraints, and the initial fault probability is processed according to the deviation rate to obtain the final fault probability.

9. The single-phase fault location and judgment method based on fuzzy control and dynamic topology as described in claim 8, characterized in that, Also includes: S5: Verify the activity status and connectivity of the candidate lines through a search algorithm, and determine whether the candidate lines satisfy the topological logic.

10. The single-phase fault location and judgment method based on fuzzy control and dynamic topology as described in claim 9, characterized in that, Also includes: S6: Select the active main line edge with the largest final fault probability that satisfies the preset threshold among the candidate lines that satisfy the topology logic as the fault line, and select the active branch edge with the largest zero-sequence current that satisfies the impedance and current amplitude constraints among the fault lines as the fault phase.

Images