Distribution network traveling wave fault point intelligent positioning method and system based on depth time sequence feature learning

Abstract

The invention provides a distribution network traveling wave fault point intelligent positioning method and system based on deep time sequence feature learning, and belongs to the technical field of intelligent power distribution detection based on deep learning. Firstly, a high-speed traveling wave sensor is arranged on a distribution line, multi-dimensional three-phase voltage and current signals are collected, and a data set of fault types, phases, grades and positions is constructed; then, fault state features are extracted through topology perception normalization, symmetric component mapping and two-channel time sequence modeling, and accurate recognition of fault types, related phases and grades is achieved through an attention mechanism; furthermore, a fault intelligent positioning model based on a topological graph is constructed, a tower sensing weight and a line information bearing weight are introduced, and space-time embedding is extracted through a graph attention network and an echo state network, so that line fault classification and accurate positioning are realized. Finally, through combination of off-line model training and on-line system deployment, real-time identification and positioning of power distribution network faults are realized, and accuracy, robustness and response speed of fault diagnosis are effectively improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent power distribution detection based on deep learning, and particularly relates to a power distribution network traveling wave fault point intelligent positioning method and system based on deep time sequence feature learning. BACKGROUND

[0002] In the power system, the distribution network, as an important link connecting the transmission and distribution system and the terminal user, its operation safety is directly related to the stable supply of electric energy and the reliability of power use on the user side. With the increasing of new energy access, load volatility and the increasingly prominent problem of equipment aging, the distribution network faces more and more uncertainties and potential risks in the operation process, especially the frequency and complexity of sudden faults are rising. As a key technical means of fault diagnosis and rapid response of the distribution network, the traveling wave fault positioning can judge the location and nature of the fault in a very short time, and is the core support to improve the fault disposal efficiency, shorten the power supply recovery time and guarantee the system resilience. However, due to the adaptability of traditional methods to noise interference, signal propagation characteristics change and complex network structure, the accuracy and robustness of current traveling wave fault positioning are still difficult to meet the operation needs of high reliability distribution system. Therefore, it is urgent to develop a new type of fault identification and positioning method with high accuracy, strong adaptability and real-time response capability to better support the safe operation and management of intelligent distribution system.

[0003] At present, the traveling wave fault positioning and identification of the distribution network mainly includes the following methods: (1) Time domain difference positioning method based on traveling wave signal: this method measures the time difference of traveling wave signal arriving at multiple measuring points after the fault occurs, and calculates the position of the fault point in the distribution line. Its principle is simple and intuitive, and the hardware implementation threshold is low. However, this method is highly dependent on the synchronization sampling accuracy and the clarity of the initial traveling wave of fault, and is easily affected by noise interference, channel delay error and weak fault characteristics, resulting in large positioning error, especially in complex distributed power grid, the accuracy decreases significantly; (2) Time-frequency analysis method based on wavelet transform and energy jump identification: the method extracts the sudden change characteristics in the traveling wave signal by means of wavelet transform, takes the energy jump point as the fault feature and combines with the physical model for positioning. It has good time-frequency resolution in multiple scales and is suitable for processing non-stationary signals. However, this method is sensitive to the selection of wavelet basis, and the feature extraction parameters depend on manual setting; the response to weak disturbance and part of high resistance fault is not obvious, which may lead to misidentification or missed identification; in addition, the stability of this method is insufficient under high frequency noise, which is difficult to adapt to complex scene generalization; (3) End-to-end identification method based on deep learning model such as LSTM and CNN: Deep learning method can learn the nonlinear mapping relationship between traveling wave and fault type / location through large-scale data, such as LSTM for capturing time sequence characteristics, CNN for local mutation identification, forming an end-to-end fault identification framework. However, this method relies heavily on large-scale labeled samples, and the generalization ability is limited by the distribution of training data; in addition, the model has poor interpretability, and it is difficult to trace the key decision features, which limits its reliable deployment in high-security power systems.

[0004] In summary, the current distribution network traveling wave fault identification and positioning method has achieved application effect in certain scenarios, but generally has problems such as strong precision dependence, poor anti-interference, and insufficient adaptability to complex faults. SUMMARY

[0005] To solve the above problems, the first aspect of the application provides a distribution network traveling wave fault point intelligent positioning method based on deep time sequence feature learning, including the following processes: S1, high-speed traveling wave sensors are arranged at each monitoring tower of the distribution line to collect multi-dimensional complete traveling wave signals of three-phase voltage and current; S2, the complete traveling wave signal is input into the trained distribution network traveling wave fault state identification model, first, topological perception normalization processing is performed to suppress amplitude drift; then, time sequence features are extracted by using symmetric component learnable mapping and structure-time double mask mechanism; combined with global attention enhancement and multi-task prediction branch, the fault type, involved phase and fault level are output; S3, a fault intelligent positioning model is constructed, the complete traveling wave signal and the tower physical connection relationship are constructed into a graph structure data, the node features are traveling wave signals, and the tower perception weight and line information bearing weight are introduced to reflect the node activity and channel strength; the fault state identification result output by S2 is used to optimize the weight; based on the graph sequence structure, the traveling wave driven graph perception modeling is performed, and the multi-layer graph attention network extracts the structure embedding features of the integrated space-time information; finally, the double-layer echo state network is used to output the fault classification result and the fault positioning result.

[0006] Preferably, the construction method of the data set for training the model includes: High-speed traveling wave sensors are arranged at each monitoring tower of the distribution line to collect continuous traveling wave signals before and after the fault occurs, and complete traveling wave signals of the fault occurrence duration are obtained , wherein, represents the traveling wave signal collected by the i-th monitoring tower in the duration, and ; and the traveling wave signal collected by any monitoring tower , wherein, , wherein, represents the traveling wave signal collected by the i-th monitoring tower in the duration, and One monitoring tower Traveling wave characteristic data acquired at specific time points, including three-phase voltage , , and three-phase current , , .

[0007] Preferably, in the power distribution network traveling wave fault state identification model, topology-sensing normalization processing is performed to suppress amplitude drift. The specific process is as follows: against The Middle One monitoring tower Traveling wave characteristic data acquired at specific time points First, determine the first line based on the line topology connection. The set of neighboring towers of a monitoring tower Calculate the neighborhood tower set in Local mean of traveling wave characteristic data at time points and standard deviation The Z-score normalization method was used to achieve local normalization based on topological neighborhood; topological sensing normalization was performed on all time-point data of all monitoring towers to obtain the processed data. .

[0008] Preferably, in the power distribution network traveling wave fault state identification model, the symmetric component learnable mapping is based on a linear mapping layer, which maps the three-phase signals [A,B,C] to [zero sequence, positive sequence, negative sequence] representations. This mapping layer uses the standard transformation matrix of the symmetric components as the initial weights for the linear transformation, preserves the original phase information through the residual structure, and allows the mapping weights to be optimized during training to obtain structured features. .

[0009] Preferably, in the power distribution network traveling wave fault state identification model, the structure-time dual masking mechanism is introduced to extract time-series features as follows: Structural features Robust temporal features are obtained by applying random time-axis masks across time-mapping layers to mask certain time points. By applying a monitoring point mask across the monitoring point mask layer, a scenario with missing nodes is simulated, and structural mask features are obtained. ;after, and State-driven modeling and nonlinear transformation are performed through DSS layers respectively, and layer normalization and ReLU activation are combined to output the results. and ; The above structure is repeated and stacked five layers, and the input of each layer of the rear layer is the two-channel output result of the upper layer, and finally the multi-layer fusion features are spliced .

[0010] Preferably, the multi-task prediction branch in the power distribution network traveling wave fault state identification model is specifically: Aiming at the fusion semantic representation , three task branches are designed to make the following predictions: Fault type prediction head: composed of full connection layer, Dropout layer and Softmax classification layer, to obtain fault type identification result , including no fault, single-phase ground fault, two-phase short circuit fault, two-phase ground short circuit fault, three-phase short circuit fault and three-phase ground short circuit fault; Phase identification prediction head: composed of full connection layer, Dropout layer and Sigmoid activation function, to obtain phase identification result , including A, B and C three phases; Fault severity level prediction head: composed of full connection layer, Dropout layer and Sigmoid activation function, to obtain fault level identification result , divided into grade I, grade II and grade III.

[0011] Preferably, in the fault intelligent positioning model, the complete traveling wave signal and the tower physical connection relationship are constructed into a graph structure data, specifically: Based on the complete traveling wave signal , first, the basic graph structure is constructed according to the physical connection relationship between each monitoring tower in the power distribution network; specifically, each monitoring tower is taken as a node, and the physical connection line between the towers is taken as an edge; the traveling wave feature of each tower in is taken as the node feature, and a graph sequence structure data with a length of is obtained ; In addition, two key mapping parameters are introduced in the original graph structure: Tower perception weight : used to measure the response strength of different tower nodes to the traveling wave disturbance signal, and can dynamically reflect the local node activity under the difference of fault type; Line information bearing weight : used to measure the strength of the graph edge in the fault signal propagation path, and to strengthen the representation ability of the dominant propagation path and the key channel; Two weight parameters and are assigned the same value for all nodes and edges at initialization, and are used as structure variables to be optimized, so that the graph structure can dynamically adjust the topology structure perception under different fault modes.

[0012] Preferably, in the intelligent fault positioning model, the fault state identification result output by S2 is used to optimize the weight, specifically: Define decision variables; tower perception weight : dimension , is the number of monitored towers, wherein represents the response strength weight of the th tower to the fault disturbance signal, and the value range is ; the higher the weight, the stronger the perception ability of the tower to the traveling wave signal under the current fault type; the line information bearing weight : dimension , is the number of lines between towers, wherein represents the bearing strength weight of the th line to the fault signal propagation, and the value range is , the higher the weight, the more dominant the line is as the main propagation path of the traveling wave signal under the current fault type; Establish the objective function; the objective function aims to maximize the fault type sparsity, tower perception weight smoothness and line information bearing weight, and comprehensively considers the fault type adaptability and structural smoothness constraints: ; ; ; ; wherein and are determined by the fault level identification result output by S2 , specifically as follows: ; ; and are used to control the sparsity constraint, indicating that the higher the fault level, the weaker the sparsity constraint; is the numerical representation of , takes the values 1, 2, 3, corresponding to levels I, II and III in respectively; is used to constrain the phase alignment under multi-phase fault, when the node in is a multi-phase fault , the node in The fault phase indication vector covers the phase code of each tower node, and is calculated according to The calculation shows that a certain phase fault is recorded as 1, and vice versa, if the three phases of the th node are all faulty ; is the reciprocal matrix of the line length, which is used to strengthen the weight of the short path line; is the weight of the path constraint on the fault type complexity, and the path constraint is smaller when the fault type is more complex, and the values of corresponding to no fault, single-phase ground fault, two-phase short-circuit fault, two-phase ground short-circuit fault, three-phase short-circuit fault and three-phase ground short-circuit fault are 0, 0.1, 0.2, 0.3, 0.4 and 0.5 respectively; The objective function is solved based on the physical constraint enhanced adaptive gradient descent algorithm to obtain optimal tower perception weight and optimal line information bearing weight .

[0013] Preferably, in the fault intelligent positioning model, the traveling wave driven graph perception modeling is carried out based on the graph sequence structure, the multi-layer graph attention network extracts and fuses the structure embedding features of the space-time information, and the specific process is as follows: The first time point graph data is processed by using a graph attention network enhancement layer, the local structure relationship between nodes is fused, and the fault semantic embedding features of the first time point are extracted ; Then, the features are spliced with the graph structure input of the next time point , input into the second layer GATv2Conv, cross-time information transmission and linkage modeling are realized, and the fault semantic embedding features of the second time point are extracted ; all time point graph structures are processed in sequence to form all fault semantic embedding features containing time evolution ; The above structure is stacked with 4 layers of GATv2Conv, and finally the output results of all time point node embedding are spliced to form a complete line diagnosis sequence representation .

[0014] The second aspect of the application provides a distribution network traveling wave fault point intelligent positioning system, characterized by comprising high-speed traveling wave sensors arranged at each monitoring tower of a distribution line and a power grid monitoring center and an edge computing node, and the power grid monitoring center and the edge computing node are deployed with the distribution network traveling wave fault state identification model and the fault intelligent positioning model as described in the first aspect.

[0015] Compared with the prior art, the application has the following beneficial effects: (1) Multi-task fault state identification mechanism for complex distribution network scenarios: In view of the problem that the traveling wave signal is easily disturbed by branch reflection and noise in the complex distribution network, the application proposes a feature extraction method combining topology-aware normalization and symmetrical component learnable mapping, and realizes the joint identification of fault type, involved phase and fault level through double-channel time sequence modeling and attention aggregation, thereby improving the diagnostic accuracy and stability in multi-interference scenarios; (2) Topology graph modeling method fusing dynamic weights: In view of the differences in propagation path and tower response of different fault types, the application introduces tower-aware weights and line information bearing weights in topology graph modeling, and dynamically optimizes them combined with fault identification results, so that the graph structure can adaptively reflect the key nodes and dominant paths under fault mode, thereby enhancing the adaptability of the positioning model to complex topology and multiple types of faults; (3) Fault location framework driven by three graphs and space-time: In view of the task requirement of modeling the traveling wave time sequence evolution law and the topology space dependence for distribution network fault location, the application uses graph attention network for structure relationship modeling, and combines echo state network to depict time sequence dynamics, forming a graph space-time joint feature representation, thereby realizing high-precision positioning of line fault location while ensuring real-time. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The figure is the overall technical route flowchart of the application.

[0017] Figure 2 The figure is the structure diagram of the distribution network traveling wave fault state identification model of the application.

[0018] Figure 3 The figure is the network structure diagram of the fault intelligent positioning model based on topology graph of the application.

[0019] Figure 4 The figure is the fault line identification performance index comparison chart in the embodiment of the application.

[0020] Figure 5 The figure is the fault location positioning heat map in the embodiment of the application.

[0021] Figure 6 The figure is the line information bearing weight heat map in the embodiment of the application.

[0022] Figure 7 The figure is the tower node feature response value distribution chart in the embodiment of the application. DETAILED DESCRIPTION

[0023] The application provides a distribution network traveling wave fault point intelligent positioning method and system based on deep time sequence feature learning. The core idea is to identify the fault type first and then locate the fault position to improve the diagnosis accuracy and robustness. Specifically, first, high-speed traveling wave sensors are arranged on the distribution line to collect multi-dimensional three-phase voltage and current signals, and a data set of fault type, phase, level and position is constructed; then, fault state features are extracted through topology perception normalization, symmetric component mapping and double-channel time sequence modeling, and the attention mechanism is used to accurately identify the fault type, involved phase and level; further, a fault intelligent positioning model based on a topology graph is constructed, the tower perception weight and line information bearing weight are introduced, the space-time embedding is extracted through a graph attention network and an echo state network, thereby realizing the accurate identification of the line fault classification and positioning; finally, through the combination of offline model training and online system deployment, the real-time identification and positioning of the distribution network fault are realized, and the accuracy, robustness and response speed of the fault diagnosis are effectively improved. The overall implementation process is as shown in Figure 1

[0024] S1, distribution network traveling wave signal acquisition and data set construction: a plurality of monitoring towers are arranged in the distribution line to collect three-phase voltage and current multi-dimensional traveling wave signals to form complete traveling wave feature data; continuous signals before and after the fault are obtained through a high-speed sampling device, and a traveling wave fault identification data set is constructed, and the fault type, involved phase and fault level are labeled; on this basis, the fault position is further calibrated, a traveling wave fault position positioning data set is constructed, and structured data support is provided for model training.

[0025] S2, distribution network traveling wave fault state identification model construction: in view of the problem that the traveling wave signal is affected by reflection and interference, the application designs a fault state identification module. First, topology perception normalization processing is performed to suppress amplitude drift; then, the symmetric component learnable mapping and structure-time double mask mechanism are used to extract time sequence features; combined with global attention enhancement and multi-task prediction branch, the fault type, involved phase and fault level are accurately identified to provide stable input for subsequent positioning.

[0026] S3, construction of fault intelligent positioning model based on topology graph: the complete traveling wave signal and the physical connection relationship of the tower are constructed into a graph structure data, the node features are the traveling wave signals, and the tower perception weight and line information bearing weight are introduced to reflect the node activity and channel strength; the fault state identification result is used to optimize the weight; based on the graph sequence structure, the traveling wave driven graph perception modeling is performed, the multi-layer graph attention network extracts the structure embedding features of the integrated time and space information; finally, the double-layer echo state network is used for fault classification and positioning to realize the accurate identification of the fault in the line.

[0027] ​S4, model training and system deployment: based on the dataset constructed in S1 and the model designed in S2 and S3, offline training is performed to optimize the prediction error of fault type, involved phase, fault level and fault location; after training, the fault state identification model and the intelligent positioning model are modularly deployed to the monitoring center and the edge node, and real-time collection of traveling wave signals is performed through high-speed sensors; in the online stage, fault state identification is performed first, then the physical constraint enhanced adaptive gradient descent algorithm is used to update the tower perception weight and the line information bearing weight in real time, and finally the fault intelligent positioning model is input to output the specific location of the fault on the line, thereby realizing fast and accurate online fault positioning.

[0028] The application will be further described below in conjunction with specific embodiments.

[0029] I. Traveling wave signal collection and dataset construction of distribution network In engineering practice, different types of faults (such as single-phase grounding, two-phase short circuit and three-phase short circuit) have significant differences in signal propagation path, disturbance amplitude and duration characteristics. If unified modeling is performed, it is easy to cause feature mixing and identification ambiguity, thereby reducing the positioning accuracy. In order to improve the overall performance of fault diagnosis of distribution network, the application adopts a phased modeling strategy of “first identifying fault type and then locating fault position”, and by constructing independent fault type identification and fault positioning datasets, the model training target is clear and the feature extraction is stable, thereby effectively enhancing the accuracy and robustness of fault positioning. Therefore, a dataset containing two tasks, i.e. a traveling wave fault identification dataset and a traveling wave fault position positioning dataset, needs to be constructed, which includes the following steps: Distribution network traveling wave signal feature selection: a plurality of monitoring towers are pre-selected in the distribution line, and multi-dimensional features including voltage and current are selected as input signals. Specifically, each monitoring point collects traveling wave signals of the following six channels: three-phase voltage 、 、 , and three-phase current 、 、 Therefore, the complete traveling wave feature data of the distribution network traveling wave signal is .

[0030] Distribution network traveling wave input data collection: high-speed traveling wave sampling devices are arranged at each monitoring tower of the distribution line, and continuous traveling wave signals before and after the fault occur are collected at a high frequency to obtain complete traveling wave signals in the fault duration period, wherein represents the traveling wave signal collected by the th monitoring tower in the duration period, and ; and the traveling wave signal collected by any monitoring tower, wherein represents the traveling wave signal collected by the a monitoring tower at a time point collects traveling wave characteristic data (including three-phase voltage and current).

[0031] Traveling wave fault identification dataset construction: based on the collected traveling wave input data of the distribution network, the fault type label corresponding to the input data is labeled (including: no fault, single-phase ground fault, two-phase short circuit fault, two-phase ground short circuit fault, three-phase short circuit fault, three-phase ground short circuit fault, a total of six); At the same time, the fault involved phase (including A, B, C three-phase), and the fault severity level (divided into level I, level II, level III according to GB / T 30137-2013, and level I corresponds to the highest severity); Therefore, the traveling wave fault identification label ; And, with the complete traveling wave signal as input, the traveling wave fault identification label as output, a set of traveling wave fault identification dataset is obtained, and extensive data collection is carried out based on the method, so as to obtain a complete traveling wave fault identification dataset.

[0032] Traveling wave fault location positioning dataset construction: after completing the dataset construction of the fault identification task, the fault location information label is further labeled; Specifically, the fault positioning label is added to the line between the monitoring towers and ; Wherein, if the line between the detection tower occurs a fault, then , otherwise 0; represents the specific position of the fault in the line, the starting point is defined by the current direction, represents that the fault occurs at the starting tower, represents that the fault occurs at the end tower, then ; If no fault occurs, then is set to a preset maximum value; Therefore, based on the method, the fault location of all lines is labeled, and the final traveling wave fault location positioning label is ; With the complete traveling wave signal as input, the traveling wave fault location positioning label as output, a set of traveling wave fault location positioning dataset is obtained, and extensive data collection is carried out based on the method, so as to obtain a complete traveling wave fault location positioning dataset.

[0033] II. Design of distribution network traveling wave fault state identification model Because traveling wave signals in complex power distribution networks are often affected by branch reflections and environmental interference, direct fault location inference can easily lead to model location deviations and convergence instability. Therefore, this invention proposes introducing a traveling wave fault state identification module before fault location. By pre-identifying the global disturbance patterns of the traveling wave, a clearer input representation and a more stable optimization boundary are provided for subsequent location, ensuring that the overall method possesses the advantages of phased collaborative modeling.

[0034] The input data for this model is a complete traveling wave signal. The output includes fault type identification results, phase identification results, and fault level identification results. , structure as Figure 2 As shown.

[0035] Topology-sensing normalization processing: To suppress amplitude drift caused by load differences at different measurement points, this invention proposes a topology-sensing normalization method for traveling wave signals. Normalization is performed to suppress modeling bias caused by differences in amplitude at different measurement points; Specifically, targeting The Middle One monitoring tower Traveling wave characteristic data acquired at specific time points The following processing is performed on it: First, determine the first line based on the line topology connection. The set of neighboring towers of a monitoring tower Calculate the neighborhood tower set in Local mean of traveling wave characteristic data at time points and standard deviation And the Z-score normalization method is used to achieve local normalization based on topological neighborhood; Based on this method, topology-sensing normalization is performed on all time-point data from all monitoring towers to obtain the processed data. .

[0036] Symmetric component learnable mapping: The symmetric component mapping layer, based on a linear mapping layer, maps the three-phase signal [A,B,C] to [zero-sequence, positive-sequence, negative-sequence] representation, thereby enhancing the structural decoupling of the electrophysical semantics. Furthermore, this mapping layer uses the standard transformation matrix of the symmetric components as the initial weights for the linear transformation, preserves the original phase information through the residual structure, and allows the mapping weights to be optimized during training to obtain structured features. .

[0037] Dual-channel temporal modeling based on DSS: In order to improve the robustness of the model to fault signals, this invention introduces a structure-time dual masking mechanism and integrates a deep separable state space (DSS) layer to extract temporal evolution features. Specifically, structural features Robust temporal features are obtained by applying time-axis random masks through the cross-time mask layer to mask part of the time points Structural mask features are obtained by applying monitoring point masks through the cross-monitoring point mask layer to simulate node missing scenarios After that, and State-driven modeling and nonlinear transformation are performed through the DSS layer, combined with layer normalization and ReLU activation, to output and respectively; This structure is repeated five times, and the input of each layer of the latter is the two-channel output results of the upper layer, finally forming a multi-layer fusion feature , which has both context awareness and structural consistency modeling capabilities.

[0038] Global awareness and attention enhancement fusion: the global average pooling is used to obtain the global semantic features, which are sent to the fault awareness attention aggregation layer. This layer dynamically assigns weights to the key pole tower nodes through the multi-head attention mechanism, thereby strengthening the modeling of the fault influence path, and outputs the feature .

[0039] Subsequently two-layer full connection and ReLU activation are used for dimension reduction to generate the final fusion semantic representation .

[0040] Multi-task prediction output: for the fusion semantic representation , three task branches are designed to make the following predictions: (1) Fault type prediction head: composed of fully connected layers, Dropout layers and Softmax classification layers, to obtain fault type identification results ; (2) Phase recognition prediction head: composed of fully connected layers, Dropout layers and Sigmoid activation functions, to obtain phase recognition results ; (3) Fault severity level prediction head: composed of fully connected layers, Dropout layers and Sigmoid activation functions, to obtain fault level identification results .

[0041] III. Topology-based intelligent fault positioning model design In order to solve the uncertainty of traveling wave signal propagation path in complex distribution network, and the influence of structural dynamic change on fault accurate positioning, the application designs a dynamic atlas driven intelligent fault positioning model, which fully integrates the distribution network topology connection, traveling wave disturbance characteristics and fault type identification result, realizes the accurate identification of fault position and fault line; the network structure of the intelligent fault positioning model based on topology graph is as shown in Figure 3 .

[0042] 1. Topology prior based graph structure data construction: based on complete traveling wave signal , first, the basic graph structure is constructed according to the physical connection relationship between each monitoring tower in the distribution network; specifically, each monitoring tower is taken as a node (Node), and the physical connection line between the towers is taken as an edge (Edge); the traveling wave characteristics of each tower in the traveling wave signal are taken as node characteristics, so that the graph sequence structure data with a length of is obtained; In addition, considering the difference of different fault types in the propagation process on the response intensity and path dependence degree of the tower, the application introduces two kinds of key graphing parameters in the original graph structure: tower perception weight : used to measure the response intensity of different tower nodes to the traveling wave disturbance signal, and can dynamically reflect the local node activity under the difference of fault type; line information bearing weight : used to measure the intensity of the graph edge in the fault signal propagation path, and to strengthen the representation ability of the dominant propagation path and the key channel; The above two weight parameters and are assigned the same value to all nodes and edges at the initialization, and are used as structure variables to be optimized, so that the graph structure can dynamically adjust the topology structure perception under different fault modes, thereby enhancing the adaptability of the model to complex space-time disturbance.

[0043] 2. Optimization of tower perception weight and line information bearing weight; in order to improve the perception ability of the graph structure under different fault modes, reduce the fault positioning error, and improve the adaptability of the model to complex fault scenarios. The fault type identification result , the phase identification result and the fault level identification result output by the distribution network traveling wave fault state identification model are used to optimize the tower perception weight and the line information bearing weight of the dynamic topology graph modeling, specifically as follows: Define decision variable: tower perception weight : dimension , To monitor the number of towers, among which Indicates the first The response strength weight of each tower to fault disturbance signals, with a value range of [value range missing]. A higher weight indicates a stronger ability of the tower to detect traveling wave signals under the current fault type. Line information carrying weight. Dimensions are , The number of lines between towers, of which Indicates the first The carrying strength weight of each line for fault signal propagation, with a value range of [value range missing]. A higher weight indicates that the line is the dominant propagation path for traveling wave signals under the current fault type.

[0044] The objective function aims to maximize the sparsity of fault types, the smoothness of tower sensing weights, and the weight of line information carrying capacity. Considering both fault type adaptability and structural smoothness constraints, the expression is as follows: ; ; ; ; in and Fault level identification results output by S2 Confirmed, details are as follows: ; ; and Used to control sparsity constraints, indicating that the higher the fault level, the weaker the sparsity constraints; for Numerical representation of The values ​​1, 2, and 3 correspond to respectively Levels I, II, and III; Used to constrain phase alignment under multi-phase faults, when When the node in the middle is a multi-phase fault , When the node in the middle is a single-phase fault ; The fault phase indication vector covers the phase encoding of each tower node, based on... The calculation shows that a fault in one phase is recorded as 1, and vice versa as 0. If the first phase is faulty... If all three phases of a node fail, then ; is the inverse matrix of line length, used to strengthen the weight of short path line; is the weight of path constraint to fault type complexity, the more complex the fault type, the smaller the path constraint, and the corresponding are defined as 0, 0.1, 0.2, 0.3, 0.4 and 0.5 respectively; The physical constraint enhanced adaptive gradient descent algorithm is designed to obtain the optimal tower perception weight and the optimal line information bearing weight , which are as follows: 1) Initialize the tower perception weight is , as the current tower perception weight, each dimension of which is a random number between 0 and 1, and initialize the line information bearing weight is , as the current line information bearing weight, each dimension of which is a random number between 0 and 1, and set the convergence threshold ; 2) Calculate , and using the fault type identification results , , , and output by the power grid traveling wave fault state identification model; 3) Calculate , and using the current tower perception weight and the current line information bearing weight and the calculated parameters, and then calculate the current objective function value ; 4) Perturb the numerical values of the current tower perception weight and the current line information bearing weight, randomly mutate some numerical values in the tower perception weight and the current line information bearing weight to obtain the mutated tower perception weight and the mutated line information bearing weight; 5) Exchange the numerical values of the random points in the mutated tower perception weight and the mutated line information bearing weight to obtain the crossed tower perception weight and the crossed line information bearing weight; 6) Input the crossed tower perception weight and the crossed line information bearing weight to calculate the next time objective function value ; 7) If If yes, output the cross-pole tower perception weight and the cross-line information bearing weight as the optimal pole tower perception weight and the optimal line information bearing weight respectively; otherwise, repeat steps 3) to 7). 8) Configure the graph structure using the optimal pole tower perception weight and the optimal line information bearing weight.

[0045] 3. Traveling wave driven graph perception modeling: based on graph sequence structure data Traveling wave driven graph perception modeling is performed to extract structure embedding features that fuse spatio-temporal information; Specifically, for the first time point graph data , a graph attention network enhancement layer (GATv2Conv) is used for processing to fuse local structure relationships between nodes and extract fault semantic embedding features at the first time point . Subsequently, the features are concatenated with the graph structure input at the next time point and input into the second layer GATv2Conv to realize cross-time information transmission and linkage modeling and extract fault semantic embedding features at the second time point ; based on this method, all time point graph structures are processed in turn to form all fault semantic embedding features containing time evolution . To improve the high-order structure perception ability of the model, the structure is stacked with 4 layers of GATv2Conv, and finally the output results of all time point node embeddings are concatenated to form a complete line diagnosis sequence representation for subsequent task reasoning.

[0046] 4. Line fault classification and positioning: based on line diagnosis sequence features The invention uses a double-layer echo state network (ESN) to process the sequence features to model the dynamic law of fault signals over time and obtain fault time series processing features ; then the following two task heads are used for fault diagnosis and positioning: (1) Fault time series processing features are classified by a Softmax classification layer to determine the fault type of the line between the tower and the pole to obtain the line fault classification result . (2) Fault time series processing features are calculated by a ReLU activation function to realize accurate positioning of the spatial position to obtain the fault position positioning result . Therefore, the fault intelligent positioning model based on the topology graph combines the actual topology structure of the power system and the dynamic traveling wave characteristics, and finally improves the fault positioning accuracy of different fault modes by identifying whether a specific tower line is faulty and the fault location.

[0047] IV. Model training and system deployment The present application carries out model training and system deployment, realizes the full-link closed loop from offline training to online inference, and thus guarantees the availability and stability of the method in the actual distribution network. Specifically, the following steps are included: Distribution network traveling wave fault state identification model training: based on the constructed distribution network traveling wave fault identification dataset and the constructed distribution network traveling wave fault state identification module, the joint loss function is used to simultaneously optimize the three tasks of fault type, involved phase and fault level; when the loss function converges, the training is terminated, and the final distribution network traveling wave fault state identification model is obtained.

[0048] Fault intelligent positioning model training based on topology graph: based on the constructed traveling wave fault location positioning dataset and the constructed fault intelligent positioning model based on topology graph, the structure parameters of the graph attention network layer and the echo state network layer of the model are trained, with the minimum line fault classification error and fault position deviation as the target, and when the loss function converges, the training is terminated, and the final fault intelligent positioning model is obtained.

[0049] Model system deployment: the trained distribution network traveling wave fault state identification model and fault intelligent positioning model are deployed to the power grid monitoring center and edge computing node in a modular way, and high-speed traveling wave sensors are arranged at each monitoring tower of the distribution line to realize real-time collection of complete traveling wave signal data.

[0050] Online fault state identification: the collected complete traveling wave signal data is input into the trained distribution network traveling wave fault state identification model to obtain fault type identification results, involved phase identification results and fault level identification results, thereby completing real-time diagnosis of the current fault state of the distribution network.

[0051] Online weight optimization: based on the obtained fault type identification results, involved phase identification results and fault level identification results, a physical constraint enhanced adaptive gradient descent algorithm designed by the fault intelligent positioning model is used to update the tower perception weight and line information bearing weight in the fault intelligent positioning model in real time, and the optimal weight configuration under the working condition is obtained; Fault positioning: after the weight update is completed, the graph structure data is constructed based on the complete traveling wave signal data, and the trained fault intelligent positioning model is input, and finally the specific position of the current fault in the line is output, realizing accurate positioning and rapid response to the distribution network fault.

[0052] V. Experimental Results and Analysis To verify the effectiveness of the proposed fault precise location method for power distribution networks, the present application constructs a multi-class fault sample set based on a typical power distribution network simulation scenario, covering different topological structures, disturbance types, and fault levels. The experimental objectives include: (1) evaluating the prediction performance of the present application model in fault line identification and fault location; (2) verifying the applicability and modeling advantages of the present method in dynamic graph structure modeling and topological feature extraction, highlighting its robustness and precision under complex spatio-temporal disturbance conditions.

[0053] 1. Fault classification and positioning prediction effect Firstly, the line identification and location ability of the present application model under different fault scenarios is compared and analyzed. Figure 4 The accuracy, macro-F1 value, and recall rate of the fault line identification task are shown, where model A is a graph convolutional neural network model, and model B is an LSTM model. Figure 5 The corresponding fault location result heat distribution map is shown.

[0054] Firstly, the line identification and location performance of the model proposed by the present application under different fault scenarios is compared and analyzed. Figure 4 The comparison results of three key performance indicators in the fault line identification task are shown, including accuracy, macro-F1 value, and recall rate, where model A is a structure-aware model based on graph convolutional neural network (GCN), and model B is a typical time series modeling model based on long short-term memory network (LSTM). Figure 5 The fault location heat distribution map under the typical fault scenario is presented to intuitively reflect the difference in fault point spatial recognition accuracy of different models.

[0055] From Figure 4 It can be seen that the present application method is superior to the comparison model in all evaluation indicators, with an accuracy of 97.3% and an F1 value of more than 96%, indicating that the collaborative architecture of "topology-enhanced graph construction + GATv2Conv graph modeling + echo state network (ESN) time series processing" constructed by the present application can effectively extract key disturbance path features and accurately map to fault types. Figure 5 Further, the fault point positioning heat distribution on the physical topology is shown, and the present application model can effectively focus on the fault tower and its adjacent nodes, reflecting the advantages of "line diagnosis sequence features" in fine-grained positioning. The overall results show that the model can achieve high-precision identification of fault types and spatial locations in power distribution networks under complex disturbance propagation.

[0056] 2. Graph modeling structure and node feature explainability analysis To analyze the contribution of key parameters to model performance and physical semantics during graph structure modeling, this paper further examines the learning process and spatial distribution of two types of structural variables: tower sensing weight and line information carrying weight. Figure 6 The heatmap of the edge weight space obtained after training is shown; Figure 7 The characteristic response values ​​of different tower nodes are distributed.

[0057] Depend on Figure 6 It is evident that, under different combinations of fault types and paths, the model can dynamically adjust the edge weight distribution, concentrating the disturbance intensity on the actual propagation path, thus improving the physical consistency and structural flexibility of graph modeling. Figure 7 The data shows that nodes with high perception weights are concentrated in multi-path convergence areas and topological boundaries. This indicates that "tower perception weights" can effectively measure the response capability of nodes to fault disturbances and are a key mediating feature for achieving interpretable fault location.

[0058] Experimental results show that the fault intelligent location method proposed in this invention achieves a unified processing path for fault type identification and spatial location inference through topology enhancement, weight optimization perception, and multi-channel graph time-series modeling. It has both high-precision prediction and structural interpretation capabilities, providing a solid algorithmic foundation and technical support for rapid fault diagnosis and fine control of intelligent power distribution systems.

[0059] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0060] While the specific embodiments of the present invention have been described above, they are not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for intelligent localization of traveling wave fault points in distribution networks based on deep temporal feature learning, characterized in that, The process includes the following: S1, high-speed traveling wave sensors are deployed on each monitoring tower of the power distribution line to collect multi-dimensional complete traveling wave signals of three-phase voltage and current; S2, the complete traveling wave signal is input into the trained distribution network traveling wave fault state identification model. First, topology-sensing normalization is performed to suppress amplitude drift. Then, symmetric component learnable mapping and structure-time dual masking mechanism are used to extract time-series features. By combining global attention enhancement and multi-task prediction branches, the fault type, involved phase, and fault level are output. S3 constructs an intelligent fault location model, which integrates the complete traveling wave signal with the physical connection relationship between the tower and the pole to create graph structure data. The node feature is the traveling wave signal, and the tower perception weight and the line information carrying weight are introduced to reflect the node activity and channel strength. The weights are optimized using the fault state identification results output by S2. Based on graph sequence structure, wave-driven graph perception modeling is performed, and multi-layer graph attention network extracts structural embedding features that fuse spatiotemporal information. Finally, the fault classification and fault location results are output using a two-layer echo state network.

2. The intelligent location method for traveling wave fault points in distribution networks based on deep temporal feature learning as described in claim 1, characterized in that: The datasets used to train models can be constructed in the following ways: High-speed traveling wave sensors are deployed on each monitoring tower of the power distribution line to collect continuous traveling wave signals before and after the fault occurs, thus obtaining the complete traveling wave signal for the duration of the fault. ,in, Indicates the first The traveling wave signals collected by each monitoring tower during this duration, and ; and the traveling wave signal collected by any monitoring tower ,in, Indicates the first One monitoring tower Traveling wave characteristic data acquired at specific time points, including three-phase voltage , , and three-phase current , , .

3. The intelligent location method for traveling wave fault points in distribution networks based on deep temporal feature learning as described in claim 1, characterized in that: In the power distribution network traveling wave fault state identification model, topology sensing normalization is performed to suppress amplitude drift. The specific process is as follows: against The Middle One monitoring tower Traveling wave characteristic data acquired at specific time points First, determine the first line based on the line topology connection. The set of neighboring towers of a monitoring tower Calculate the neighborhood tower set in Local mean of traveling wave characteristic data at time points and standard deviation The Z-score normalization method was used to achieve local normalization based on topological neighborhood; topological sensing normalization was performed on all time-point data of all monitoring towers to obtain the processed data. .

4. The intelligent location method for distribution network traveling wave fault points based on deep temporal feature learning as described in claim 1, characterized in that: In the power distribution network traveling wave fault state identification model, the symmetric component learnable mapping is based on a linear mapping layer, which maps the three-phase signals [A,B,C] to [zero sequence, positive sequence, negative sequence] representations. This mapping layer uses the standard transformation matrix of the symmetric components as the initial weights for the linear transformation, preserves the original phase information through the residual structure, and allows the mapping weights to be optimized during training to obtain structured features. .

5. The intelligent location method for traveling wave fault points in distribution networks based on deep temporal feature learning as described in claim 4, characterized in that: In the power distribution network traveling wave fault state identification model, the structure-time dual masking mechanism is introduced to extract time-series features as follows: Structural features Robust temporal features are obtained by applying random time-axis masks across time-mapping layers to mask certain time points. By applying a monitoring point mask across the monitoring point mask layer, a scenario with missing nodes is simulated, and structural mask features are obtained. ;after, and State-driven modeling and nonlinear transformation are performed through DSS layers respectively, and layer normalization and ReLU activation are combined to output the results. and ; The above structure is repeatedly stacked five times, with each subsequent layer taking the input of the two channels of the previous layer as input, and finally spliced ​​together to form a multi-layer fusion feature. .

6. The intelligent location method for traveling wave fault points in distribution networks based on deep temporal feature learning as described in claim 1, characterized in that: The multi-task prediction branch in the power distribution network traveling wave fault state identification model is specifically as follows: For fusion semantic representation The three task branches are designed to make the following predictions: Fault type prediction head: Consists of a fully connected layer, a Dropout layer, and a Softmax classification layer, which yields the fault type identification result. It includes six types: no fault, single-phase ground fault, two-phase short circuit fault, two-phase ground short circuit fault, three-phase short circuit fault, and three-phase ground short circuit fault. Phase recognition prediction head: Consists of a fully connected layer, a Dropout layer, and a sigmoid activation function, which yields the phase recognition results. It includes three phases: A, B, and C. Fault severity prediction head: Composed of a fully connected layer, a Dropout layer, and a Sigmoid activation function, it obtains the fault severity identification result. It is divided into Level I, Level II, and Level III.

7. The intelligent location method for traveling wave fault points in distribution networks based on deep temporal feature learning as described in claim 1, characterized in that: In the aforementioned intelligent fault location model, a graph structure data is constructed by linking the complete traveling wave signal with the physical connection relationship of the tower, specifically as follows: Based on complete traveling wave signal First, a basic diagram structure is constructed based on the physical connection relationships between the monitoring towers in the distribution network; specifically, each monitoring tower is used as a node, and the physical connection lines between the towers are used as edges; The traveling wave characteristics of each tower are used as nodal features to obtain a length of Graph sequence structure data ; In addition, two types of key mapping parameters are introduced into the original graph structure: Tower sensing weight It is used to measure the response intensity of different tower nodes to traveling wave disturbance signals, and can dynamically reflect the activity of local nodes under different fault types; Line information carrying weight Used to measure the strength of graph edges in the propagation path of fault signals, enhancing the characterization of dominant propagation paths and key channels; Two weight parameters and During initialization, all nodes and edges are assigned the same value and used as structural variables to be optimized, enabling the graph structure to dynamically adjust topology awareness under different fault modes.

8. The intelligent location method for distribution network traveling wave fault points based on deep temporal feature learning as described in claim 7, characterized in that: In the aforementioned intelligent fault location model, the weights are optimized using the fault state identification results output by S2, specifically as follows: Define decision variables; tower sensing weights Dimensions are , To monitor the number of towers, among which Indicates the first The response strength weight of each tower to fault disturbance signals, with a value range of [value range missing]. ; A higher weight indicates a stronger ability of the tower to detect traveling wave signals under the current fault type; line information carrying weight. Dimensions are , The number of lines between towers, of which Indicates the first The carrying strength weight of each line for fault signal propagation, with a value range of [value range missing]. The higher the weight, the more likely the line is the dominant propagation path of the traveling wave signal under the current fault type. Establish an objective function; the objective function aims to maximize the sparsity of fault types, the smoothness of tower sensing weights, and the weight of line information carrying capacity, while taking into account the constraints of fault type adaptability and structural smoothness. The objective function is solved using a physically constrained enhanced adaptive gradient descent algorithm to obtain the optimal tower sensing weights. and optimal line information carrying weight .

9. The intelligent location method for traveling wave fault points in distribution networks based on deep temporal feature learning as described in claim 8, characterized in that: In the aforementioned intelligent fault localization model, traveling wave-driven graph perception modeling is performed based on a graph sequence structure, and a multi-layer graph attention network extracts structural embedding features that fuse spatiotemporal information. The specific process is as follows: For the first time point plot data A graph attention network enhancement layer is used to process the data, fusing the local structural relationships between nodes to extract the fault semantic embedding features at the first time point. ; Subsequently, this feature With the graph structure input at the next time step The data is concatenated and input into the second layer, GATv2Conv, to achieve cross-time-time information transmission and linkage modeling, and to extract the fault semantic embedding features at the second time point. Process all in sequence A time-point graph structure is generated, forming all fault semantic embedding features that encompass temporal evolution. ; The above structure consists of four stacked GATv2Conv layers. Finally, the output results embedded in all time nodes are concatenated to form a complete line diagnostic sequence representation. .

10. A distribution network traveling wave fault point intelligent location system based on deep temporal feature learning, characterized in that: It includes high-speed traveling wave sensors deployed on each monitoring tower of the power distribution line and a power grid monitoring center and edge computing nodes, wherein the power grid monitoring center and edge computing nodes are equipped with a power distribution network traveling wave fault state identification model and a fault intelligent location model as described in any one of claims 1 to 9.

Images