Distribution network line distributed signal processing method, system, equipment and medium

Abstract

The invention discloses a distribution network line distributed signal processing method, system and device and a medium, and belongs to the technical field of distribution network lines, and the method comprises the steps: collecting node data of a power distribution network, carrying out the time alignment processing, generating a monitoring data set, and determining nodes with abnormal data in a power grid; decomposing and extracting characteristic data representing change characteristics; updating the feature data of each abnormal node based on distributed feature interaction; and generating a fault identifier according to the updated feature data, and determining a fault area. The power grid area is dynamically divided according to the monitoring range of the protection device, the protection area matrix is formed, dynamic adjustment of the protection area is achieved, the impedance and current data of all lines in the protection area are monitored in real time, abnormal lines are rapidly recognized, the fault detection precision and sensitivity are improved, and the fault detection efficiency is improved. A real-time signal analysis framework is constructed through EMD and Hilbert transformation, and rapid decomposition and feature extraction can be carried out on abnormal signals collected by edge nodes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of distribution network lines, and particularly relates to a distribution network line distributed signal processing method, system, device and medium. BACKGROUND

[0002] The distribution network is the part closest to the user in the power system, and its operation stability is directly related to the safety and reliability of power supply. With the continuous growth of power demand and the expansion of the scale of distributed energy access, the traditional distribution network is developing towards intelligence and digitization, and the distribution network fault monitoring and signal processing technology has made certain progress. The technical path with real-time signal collection, distributed analysis and fault location as the core has become a research hotspot. Through signal analysis methods such as wavelet transform and short-time Fourier transform, signal characteristics can be extracted for pattern recognition. In addition, with the rapid development of machine learning and deep learning technology, graph convolutional neural networks combined with distribution network topology information have gradually become one of the important tools for identifying distribution network fault characteristics, and distributed communication protocols such as Gossip algorithm are also applied to distributed node feature updating and information fusion due to their lightweight and high efficiency.

[0003] In the aspect of signal feature extraction, although traditional methods such as wavelet transform and short-time Fourier transform have certain time-frequency analysis ability, they are insufficient in describing the dynamic characteristics of nonlinear and non-stationary signals. However, the operation environment of the distribution network is complex, and abnormal signals often have significant nonlinear characteristics, which makes traditional methods unable to meet actual needs. Secondly, in the aspect of distributed feature fusion and analysis, existing methods usually rely on centralized processing architecture or simple distributed protocols, resulting in low processing efficiency, especially in large-scale node and dynamic fault scenarios. SUMMARY

[0004] In view of the above problems, the present application is proposed.

[0005] Therefore, the problem to be solved by the present application is that in the aspect of signal feature extraction, although traditional methods such as wavelet transform and short-time Fourier transform have certain time-frequency analysis ability, they are insufficient in describing the dynamic characteristics of nonlinear and non-stationary signals. However, the operation environment of the distribution network is complex, and abnormal signals often have significant nonlinear characteristics, which makes traditional methods unable to meet actual needs. Secondly, in the aspect of distributed feature fusion and analysis, existing methods usually rely on centralized processing architecture or simple distributed protocols, resulting in low processing efficiency, especially in large-scale node and dynamic fault scenarios.

[0006] To solve the above technical problems, the application provides the following technical scheme: a distribution network line distributed signal processing method, which comprises collecting node data of a power distribution network and performing time alignment processing to generate a unified reference monitoring data set and determining nodes with abnormal data in the power grid;

[0007] The abnormal data is decomposed and feature data representing change characteristics is extracted;

[0008] The feature data of each abnormal node is updated based on distributed feature interaction;

[0009] A fault identifier is generated according to the updated feature data, and a fault area is determined.

[0010] As a preferred scheme of the distribution network line distributed signal processing method, the collecting node data of the power distribution network and performing time alignment processing to generate a unified reference monitoring data set and determining nodes with abnormal data in the power grid comprises,

[0011] The node data in the power grid area is monitored in real time based on a preset power parameter threshold, and the node set in the power grid area is dynamically adjusted according to the monitoring result, to assist in identifying nodes with abnormal data.

[0012] As a preferred scheme of the distribution network line distributed signal processing method, the decomposing abnormal data and extracting feature data representing change characteristics comprises,

[0013] The abnormal data is split into several component data according to time sequence change characteristics, and effective component data is selected based on a preset rule for feature extraction.

[0014] As a preferred scheme of the distribution network line distributed signal processing method, the updating feature data of each abnormal node based on distributed feature interaction comprises,

[0015] A feature propagation structure is constructed based on the communication relationship between abnormal nodes, and feature data updating is realized through feature exchange between nodes in several rounds of interaction.

[0016] As a preferred scheme of the distribution network line distributed signal processing method, the determining nodes with abnormal data in the power grid further comprises,

[0017] The node set is defined according to the power distribution network, and the connection between all nodes is defined as a line set, a connection matrix A is defined according to the monitoring line of the protection device, wherein if the kth protection device monitors the jth line, it is represented as 1 in the connection matrix A, otherwise it is 0;

[0018] A correlation matrix C is defined according to the lines belonging to the grid area, wherein if the jth line belongs to the hth area, 1 is represented in the correlation matrix C, otherwise 0 is represented;

[0019] The connection matrix and the correlation matrix are multiplied by matrix multiplication to obtain a dynamic matrix of the protection area and the protection device, represented as:

[0020]

[0021] Wherein m represents the total number of lines, wherein N kh A relationship value indicating whether the kth protection device monitors the hth protection area, A kj A relationship value indicating whether the kth protection device monitors the jth line, C jh A relationship value indicating whether the jth line belongs to the hth protection area;

[0022] The impedance and current of all lines in the protection area are monitored in real time, the safe impedance threshold and the safe current threshold are determined based on the historical data of the lines, and if the impedance and current data of the lines are greater than or equal to the safe impedance threshold and the safe current threshold, the lines are removed from the protection area;

[0023] Based on the updated voltage and current data of the protection area lines, discrete wavelet transform is applied, and the energy distribution of the voltage and current data of the protection area lines is calculated based on the wavelet transform, and the sum of the historical energy mean and twice the standard deviation of the protection area is used as the area threshold, and if the maximum energy distribution in the protection area is greater than the area threshold, it indicates that there is an abnormal signal in the area.

[0024] The preferred technical scheme constructs the connection matrix and the correlation matrix and performs matrix operation to dynamically generate the corresponding relationship matrix of the protection area and the protection device, which can realize fast calculation and update of the relationship between the grid area and the monitoring device, so that the grid area division can adapt to the changes of the grid topology and the monitoring requirements in real time, and the flexibility and adaptability of the grid monitoring structure are improved. Through real-time monitoring of the impedance and current of the lines in the protection area, and combining with the historical threshold to dynamically remove the lines that do not meet the safety condition, the interference of non-fault lines on subsequent abnormal judgment is effectively avoided, and the accuracy and timeliness of abnormal data node confirmation are improved. The application of discrete wavelet transform combined with the statistical threshold of regional energy distribution can accurately identify the abnormal signals that may exist in the grid area, enhance the sensitivity and identification ability of multi-scale power fluctuations, and thus improve the reliability and precision of distribution network fault monitoring under complex operating conditions.

[0025] As a preferred scheme of the power distribution network line distributed signal processing method, the method comprises the following steps:

[0026] Based on each abnormal node updating feature data and node relationship construction feature data matrix and adjacent relationship matrix, for generating each abnormal node fault identification and regional division result as a kind of preferred scheme of the distributed signal processing method of the distribution network line of the application, wherein: the abnormal data are carried out multi-level decomposition processing, abnormal data are split into component data capable of reflecting different time sequence change characteristics, and in the decomposition process, each layer component data is verified based on the preset decomposition rule,

[0027] Based on the abnormal signal judged in the protection area, EMD is used for decomposition, and the difference between the abnormal signal and the local average value is extracted as the candidate IMF, which is expressed as:

[0028] h1(t)=f(t)-m(t);

[0029] Wherein, h1(t) represents the first layer candidate IMF of time t, f(t) represents the abnormal signal of time t, m(t) represents the local average value of time t, and the sum of the upper envelope and the lower envelope generated by the spline interpolation of the local extreme point is divided by 2 to obtain;

[0030] The next layer IMF is extracted from the residual signal recursively, which is expressed as:

[0031] r1(t)=f(t)-h1(t)

[0032] h i (t)=r i (t)-m i (t)

[0033] Wherein, r1(t) represents the residual signal of the first layer time t, r n (t) represents the residual signal of the i-th layer time t, h n (t) represents the candidate IMF of the i-th layer time t, and m n (t) represents the local average value of the i-th layer time t.

[0034] h n (t) is verified, if the difference between the zero crossing point number and the extreme point number of h n (t) is less than or equal to 1, and the local average value is not 0, then h n (t) is taken as the IMF of the n-th layer time t.

[0035] Each candidate IMF is subjected to Hilbert transform to calculate the instantaneous frequency and instantaneous amplitude, which is expressed as:

[0036]

[0037] Wherein, H i (t) represents the original signal IMFi transformed signal of (t), IMF i (t) represents the IMF value of the i-th time variable t, t represents an independent time variable in the Hilbert transform, A i (t) represents the instantaneous frequency of the i-th time t, f i (t) represents the instantaneous amplitude of the i-th time t, IMF i (t) represents the IMF value of the i-th time t, and represents the transformed instantaneous phase angle of the i-th time t.

[0038] The instantaneous frequency and the instantaneous amplitude calculated for each abnormal signal are counted and combined into a state value corresponding to the edge node, which is represented as:

[0039] x i (t) = [f i (t), A i (t)];

[0040] Wherein, x i (t) represents the feature vector of the abnormal node i at time t.

[0041] The beneficial effects of the preferred technical solution are: by performing multi-level decomposition processing on abnormal data, the abnormal data is gradually disassembled into component data reflecting different time sequence change characteristics, which can effectively distinguish information of different frequency bands and different change rates in the signal, and improve the description ability of the dynamic characteristics of the nonlinear and non-stationary signal; in the decomposition process, the component data is verified for effectiveness based on the preset decomposition rule, ensuring that each layer of component extracted meets the characteristic requirements of the real signal component, avoiding noise or pseudo-component interference; further based on the Hilbert transform of each component data, the instantaneous frequency and the instantaneous amplitude characteristics are obtained, which can refine the local characteristics reflecting the change of the signal with time, and help to improve the accuracy and robustness of abnormal state recognition, and are especially suitable for working condition scenes in which the instantaneous fluctuation is significant and the local abnormality is difficult to distinguish.

[0042] The present application provides a kind of distribution network line distributed signal processing system.

[0043] To solve the above technical problems, the present application provides the following technical scheme: a kind of distribution network line distributed signal processing system, comprising: collection data module, extraction feature module, update module and output module;

[0044] The collection data module is to collect the node data of distribution network and carry out time alignment processing, generates the monitoring data set of uniform reference, determines the node of abnormal data in power grid;

[0045] The extraction feature module is to decompose and extract the feature data representing change characteristics for abnormal data;

[0046] The update module updates the feature data of each abnormal node based on distributed feature interaction;

[0047] The output module generates a fault identification according to the updated feature data and determines a fault area.

[0048] The application provides a computer device, including a memory and a processor, the memory stores a computer program, characterized in that the processor implements the steps of the distribution network line distributed signal processing method when executing the computer program.

[0049] The application provides a computer readable storage medium, which stores a computer program, characterized in that the computer program is executed by a processor to implement the steps of the distribution network line distributed signal processing method.

[0050] The application has the beneficial effects that: the protection device dynamically divides the power grid area according to the monitoring range, forms a protection area matrix, realizes dynamic adjustment of the protection area, realizes rapid identification of abnormal lines through real-time monitoring of the impedance and current data of all lines in the protection area, improves the accuracy and sensitivity of fault detection, constructs a real-time signal analysis framework through EMD and Hilbert transformation, can quickly decompose and extract features of abnormal signals collected by edge nodes, generates state values through edge nodes, supports distributed collaborative analysis and rapid response, supports intelligent distributed distribution network fault processing, reduces operation cost and manual intervention demand, through the Gossip algorithm, the abnormal node randomly selects neighbor nodes to exchange feature values, and constantly updates its own feature vector, provides an efficient distributed feature propagation and fusion mechanism, through the graph convolutional neural network, the feature information and network topology structure of neighbor nodes can be effectively integrated, the global topology characteristics are captured on the basis of local node information, and the comprehensiveness of the classification result is ensured. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0052] Figure 1 A general flowchart of a distribution network line distributed signal processing method provided by an embodiment of the application. DETAILED DESCRIPTION

[0053] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the protection scope of the present application.

[0054] Embodiment 1 is an embodiment of the present application, which provides a distribution network line distributed signal processing method, comprising:

[0055] S1, collecting node data of the power distribution network and performing time alignment processing to generate a unified reference monitoring data set and determine the nodes with abnormal data in the power grid.

[0056] S2, decomposing the abnormal data and extracting feature data representing the change characteristics.

[0057] S3, updating the feature data of each abnormal node based on distributed feature interaction.

[0058] S4, generating a fault identification according to the updated feature data and determining a fault area.

[0059] In the process of power distribution network operation state monitoring and fault identification, due to asynchronous sampling and time sequence deviation in multi-node data collection, the data of different nodes cannot be unified in time domain, which leads to insufficient accuracy of data comparison and analysis, and it is difficult to identify the nodes with abnormal data in time. At the same time, for the processing of abnormal data, due to the lack of efficient decomposition and characterization ability of nonlinear and non-stationary characteristics, the feature information in different scale change components cannot be effectively extracted, and the local change characteristics are easy to be missed, which affects the integrity of feature expression. In terms of feature interaction and fusion, the feature information between each abnormal node is insufficient in the process of distributed processing, and the node feature data is easy to form an isolated state, which makes it difficult to realize global consistency while ensuring the advantages of distributed architecture, and limits the comprehensiveness of subsequent fault state identification. Further, when generating a fault identification and dividing a fault area based on feature data, the existing method has limited ability in feature utilization and node state association mining, and cannot fully combine feature data and node topology relationship to realize fault identification generation and accurate fault area division, which affects the accuracy and response efficiency of fault warning and disposal.

[0060] The application realizes accurate time sequence consistency of multi-node data, effectively avoids the problem of inaccurate identification of abnormal nodes caused by inconsistent time domain of data, by deploying edge nodes at key nodes of the power distribution network and collecting real-time data, and generating a unified reference monitoring data set in combination with time alignment processing; the application effectively realizes decomposition and representation of different scale change components in abnormal data, improves the feature extraction capability of nonlinear and non-stationary signals, and ensures complete expression of local change characteristics, by performing multi-level decomposition processing on abnormal data, extracting characteristic data such as instantaneous frequency and instantaneous amplitude based on empirical mode decomposition and Hilbert transform; in the feature interaction process, the application realizes distributed interaction and fusion of abnormal node feature data based on the Gossip algorithm, avoids isolation of node feature information, and enhances the global consistency of feature data; in terms of fault identification and region division, the application constructs a deep learning model based on a graph convolutional neural network, generates a fault identifier and divides a fault region in combination with node feature data and an adjacency relationship matrix, improves the accuracy of fault state identification and the precision of region positioning, and effectively supports fault early warning and rapid disposal in a complex distribution network environment.

[0061] Embodiment 2, which is an embodiment of the application, provides a distributed signal processing method for a distribution network based on the previous embodiment, comprising:

[0062] S1, collecting node data of the power distribution network and performing time alignment processing to generate a unified reference monitoring data set, and determining nodes with abnormal data in the power grid.

[0063] Edge collection nodes are arranged at multiple key positions of the power distribution network, each edge collection node collects power data corresponding to the node position, and performs time synchronization processing on the collected data to generate a monitoring data set corresponding to a unified time reference;

[0064] A mapping relationship between power grid regions and monitoring nodes is established based on the monitoring data set, and a monitoring node set corresponding to each power grid region is determined;

[0065] The node data in each power grid region is monitored in real time based on the mapping relationship between the power grid regions and the monitoring nodes, and the node set in the power grid region is dynamically adjusted according to a preset power parameter threshold;

[0066] Characteristic data representing the operating state of the power grid is extracted based on the node data in the dynamically adjusted power grid region, and the power grid nodes with abnormal data are identified according to the characteristic data.

[0067] Edge nodes are arranged at key positions of the power distribution network, and data is collected by sensors, including real-time current data collected by a current sensor and real-time voltage data collected by a voltage sensor;

[0068] The synchronization of the sampling events of the edge nodes is completed by using the IEEE 1588 precision time protocol (PTP).

[0069] Based on the distribution network topology, the protection device range is set based on the key nodes of the distribution network to divide the power grid area.

[0070] According to the definition of the node set and the connection between all nodes as a line set, the connection matrix A is defined according to the monitoring line of the protection device, wherein if the kth protection device monitors the jth line, it is represented as 1 in the connection matrix A, otherwise it is 0.

[0071] According to the line belonging to the power grid area, the correlation matrix C is defined, wherein if the jth line belongs to the hth area, it is represented as 1 in the correlation matrix C, otherwise it is 0.

[0072] The connection matrix and the correlation matrix are multiplied by matrix multiplication to obtain the dynamic matrix of the protection area and the protection device, which is represented as:

[0073]

[0074] Wherein, m represents the total number of lines, wherein N kh The relationship value of whether the kth protection device monitors the hth protection area is represented as A kj The relationship value of whether the kth protection device monitors the jth line is represented as C jh The relationship value of whether the jth line belongs to the hth protection area is represented as C.

[0075] The impedance and current of all lines in the protection area are monitored in real time, the safe impedance threshold and the safe current threshold are determined based on the historical data of the line, and if the impedance and current data of the line are greater than or equal to the safe impedance threshold and the safe current threshold, the line is removed from the protection area.

[0076] Based on the updated voltage and current data of the protection area line, the discrete wavelet transform is applied, and the energy distribution of the voltage and current data of the protection area line is calculated based on the wavelet transform, and the sum of the historical energy mean and twice the standard deviation of the protection area is used as the area threshold, and if the maximum energy distribution in the protection area is greater than the area threshold, it indicates that there is an abnormal signal in the area.

[0077] S2, decompose the abnormal data and extract feature data representing change characteristics.

[0078] The feature data representing change characteristics is extracted in the preferred embodiment of the present application: using empirical mode decomposition and calculating the instantaneous frequency and instantaneous amplitude by Hilbert transform as features.

[0079] Based on the abnormal signal judged in the protection area, the experience mode decomposition (EMD) is used for decomposition, and the difference between the abnormal signal and the local average value is extracted as the candidate IMF, which is expressed as:

[0080] h1(t) = f(t) - m(t);

[0081] where h1(t) represents the first layer candidate IMF of time t, f(t) represents the abnormal signal of time t, and m(t) represents the local average value of time t, which is calculated by the sum of the upper envelope and the lower envelope generated by the spline interpolation of the local extreme points divided by 2;

[0082] The next layer IMF is recursively extracted from the residual signal, which is expressed as:

[0083] r1(t) = f(t) - h1(t);

[0084] h i (t) = r i (t) - m i (t);

[0085] where r1(t) represents the residual signal of time t of the first layer, r n (t) represents the residual signal of time t of the i-th layer, h n (t) represents the candidate IMF of time t of the i-th layer, and m n (t) represents the local average value of time t of the i-th layer.

[0086] h n (t) is verified, if the difference between the zero crossing point number and the extreme point number of h n (t) is less than or equal to 1, and the local average value is not 0, then h n (t) is taken as the IMF of time t of the n-th layer.

[0087] Each candidate IMF is subjected to Hilbert transform to calculate the instantaneous frequency and the instantaneous amplitude, which is expressed as:

[0088]

[0089] where H i (t) represents the transformed signal of the original signal IMF i (τ), IMF i (τ) represents the IMF value of the i-th time variable τ, τ represents the independent time variable in Hilbert transform, A i (t) represents the instantaneous frequency of the i-th time t, and f i (t) represents the instantaneous amplitude of the i-th time t, IMF i(t) represents the IMF value of the i-th time t, and φi(t) represents the transformed instantaneous phase angle of the i-th time t;

[0090] The instantaneous frequency and the instantaneous amplitude calculated for each abnormal signal are counted and combined into a state value corresponding to the edge node, which is represented as:

[0091] x i (t) = [f i (t), A i (t)]

[0092] where x i (t) represents the feature vector of the abnormal node i at time t.

[0093] In an optional embodiment of the present application, the feature data representing the change characteristics is extracted by using a wavelet packet decomposition method on the abnormal data, decomposing the abnormal signal into sub-signals containing different frequency components layer by layer according to the frequency band, and calculating the energy distribution and the main frequency characteristics of each sub-signal to form the feature data reflecting the change characteristics of the abnormal signal.

[0094] In an optional embodiment of the present application, the feature data representing the change characteristics is extracted by performing time-frequency analysis on the abnormal data based on short-time Fourier transform (STFT), extracting the amplitude spectrum distribution and the energy spectrum change characteristics of the abnormal signal in the time domain and the frequency domain, and forming the feature data reflecting the dynamic change law of the abnormal data.

[0095] The beneficial effects of the preferred embodiment are: by using empirical mode decomposition to decompose the abnormal data layer by layer, the abnormal signal is effectively decomposed into different scale intrinsic mode function components, so that the non-stationary and nonlinear change characteristics contained in the signal can be accurately separated and characterized; by verifying the effectiveness of the candidate components based on the difference between the zero-crossing number and the extreme point number and the local average value during the decomposition process, the interference of false components or noise components can be avoided, and the authenticity and stability of the extracted features can be ensured; further performing Hilbert transform on the component data to obtain the instantaneous frequency and the instantaneous amplitude, so that the extracted feature data can reflect the local dynamic characteristics of the abnormal signal, improve the expression ability of the time series change characteristics, and enhance the fine characterization effect of the abnormal state.

[0096] S3, updating the feature data of each abnormal node based on distributed feature interaction.

[0097] In the preferred embodiment of the present application, the feature data of each abnormal node is updated based on the Gossip algorithm.

[0098] The abnormal nodes are constructed into a node set, the maximum communication distance of the historical data is taken as a distance threshold, and the edges between the abnormal nodes with a communication distance less than the distance threshold are constructed into an edge set of the abnormal nodes.

[0099] A feature propagation subgraph is constructed based on the node set and the edge set, and a deep search DFS is used to determine whether the feature propagation subgraph is connected;

[0100] Based on the Gossip algorithm, a neighbor node is randomly selected for feature value exchange for each abnormal node, which is represented as:

[0101] x i (t+1)=x i (t)+∈·(x j (t)-x i (t))

[0102] where x i (t+1) represents the updated feature vector of node i at time t+1, ∈ represents the learning rate, and x j (t) represents the feature vector of the jth neighbor node.

[0103] When the change of the updated feature vector of the node is no longer obvious, the iteration update is stopped, and the updated feature vector of each node is finally updated.

[0104] A feature matrix is constructed based on the updated feature vector of each node, an adjacency matrix is constructed based on the edge connection between the nodes in the propagation subgraph, and the adjacency matrix is normalized to obtain a normalized adjacency matrix.

[0105] A deep learning model is constructed based on a graph convolutional neural network GCN, including an input layer, a graph convolutional layer, a classification layer, and an output layer.

[0106] The input layer inputs the feature matrix and the normalized adjacency matrix, the graph convolutional layer updates the current node feature by weighting the features of the neighbor nodes, the classification layer maps the output features of the graph convolutional layer to the classification label space, and the output layer outputs the classification label probability of the node.

[0107] The deep learning model is trained using the training set, a cross-entropy loss function is selected to calculate the difference between the predicted classification label probability of the model and the actual label, an Adam optimizer is used for gradient descent optimization to update the weights of the model, and the iteration is stopped when the loss of the model no longer obviously decreases in the continuous iteration process, and the model parameters are output.

[0108] The trained deep learning model is used to input the feature matrix and the normalized adjacency matrix to obtain the fault class probability of each node, and the fault class with the maximum probability is selected as the fault label of the node.

[0109] In an optional embodiment of the application for updating the feature data of each abnormal node, a weighted average feature fusion method is used. First, a neighbor node set is constructed for each abnormal node, and a communication link distance or a preset importance weight is defined for each node in the set. When updating the feature data, each abnormal node performs a weighted summation on the neighbor node feature vectors according to the weights of the neighbor nodes, and performs a proportional fusion on the feature vectors of the abnormal node and the neighbor nodes to form an updated feature vector. The process is repeated in each iteration until the variation amplitude of the updated feature vector of each node is lower than a preset convergence threshold.

[0110] In an optional embodiment of the application for updating the feature data of each abnormal node, a local cluster center feature extraction method is used. The entire abnormal node set is divided into several local clusters, and the node with the highest communication density or the optimal position in each cluster is selected as the cluster center node. Each abnormal node in the cluster uploads its feature vector to the cluster center node, which performs feature aggregation processing on the feature vectors in the cluster to generate an aggregated feature vector, and then distributes the aggregated feature vector to each node in the cluster. Each node updates its feature vector accordingly. The process can be executed in multiple rounds in the cluster until the feature update converges.

[0111] The current preferred embodiment has the following beneficial effects. Through the distributed feature interaction update based on the Gossip algorithm, each abnormal node can efficiently realize the random interaction and fusion of the feature data of the neighbor nodes while maintaining autonomous feature processing, so that the node feature vector gradually converges to a comprehensive feature vector containing neighborhood information, effectively avoiding the generation of feature isolation state and improving the feature consistency in a distributed environment. Further, a graph convolutional neural network deep model is constructed based on the updated feature vector, the weighted update of the node feature is completed in the graph convolution layer through the joint input of the feature matrix and the adjacency matrix, the perception ability of the node to the neighbor state is enhanced, and the accuracy of fault state recognition and the precision of region division are improved, which is particularly suitable for the distributed feature interaction of large-scale nodes in the distribution network and the complex fault classification requirements.

[0112] S4. Generate a fault label based on the updated feature data, and determine a fault region.

[0113] Divide the region based on the fault label, and perform visual marking.

[0114] Based on the fault label of each abnormal node, divide the fault region, group the regions according to the same fault label, and output a fault region node set.

[0115] And based on the color marking in the two-dimensional view of the distribution network line node according to the fault node set, for each fault area node, the average value of the position of all nodes in the area is calculated, and the center position is determined, the center position is marked in the two-dimensional graph, and the average instantaneous frequency and instantaneous amplitude of the area are marked.

[0116] The state information of all nodes is continuously monitored, and the change of the node fault label and the generation of new abnormal signals are recorded;

[0117] The average instantaneous frequency and instantaneous amplitude data of the real-time fault area are compared with the historical threshold value, if the average instantaneous frequency and instantaneous amplitude data of the current fault area are greater than or equal to the historical threshold value, it is judged that there is a higher risk, and a warning alarm is triggered;

[0118] The terminal maintenance personnel are reminded in real time through the sound and light alarm.

[0119] Embodiment 3 is an embodiment of the application, which provides a distribution network line distributed signal processing system, comprising a data collection module, a feature extraction module, an update module and an output module;

[0120] The data collection module is to collect node data of the distribution network and perform time alignment processing to generate a unified reference monitoring data set, and determine the nodes with abnormal data in the power grid;

[0121] The feature extraction module is to decompose and extract feature data representing the change characteristics of the abnormal data;

[0122] The update module is to update the feature data of each abnormal node based on the distributed feature interaction;

[0123] The output module is to generate a fault identification according to the updated feature data and determine a fault area.

[0124] The embodiment also provides an electronic device suitable for a distribution network line distributed signal processing method, comprising a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the distribution network line distributed signal processing method as described in the above embodiment.

[0125] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to realize the distribution network line distributed signal processing method as described in the above embodiment.

[0126] The storage medium proposed in the embodiment belongs to the same inventive concept as the method for implementing a distribution line distributed signal processing method proposed in the above embodiment. Technical details not described in the embodiment can be seen from the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.

[0127] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary general hardware, and of course can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH, a hard disk or an optical disk, etc., and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the method of each embodiment of the present application.

[0128] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A method for distributed signal processing of distribution network lines, characterized in that: The method comprises the following steps: Collecting node data of the power distribution network and performing time alignment processing to generate a unified reference monitoring data set, and determining nodes with abnormal data in the power grid; Decomposing the abnormal data and extracting feature data representing the change characteristics; Updating the feature data of each abnormal node based on distributed feature interaction; Generating a fault identification based on the updated feature data and determining the fault area.

2. The method of claim 1, wherein the method further comprises: The step of collecting node data of the power distribution network and performing time alignment processing to generate a unified reference monitoring data set, and determining nodes with abnormal data in the power grid comprises: Real-time monitoring of node data in the power grid area based on a preset power parameter threshold, and dynamically adjusting the node set in the power grid area according to the monitoring result to assist in identifying nodes with abnormal data.

3. The method of claim 2, wherein the method further comprises: The step of decomposing the abnormal data and extracting feature data representing the change characteristics comprises: Splitting the abnormal data into several component data according to the time sequence change characteristics, and filtering effective component data based on a preset rule for feature extraction.

4. The method of claim 3, wherein the method further comprises: The step of updating the feature data of each abnormal node based on distributed feature interaction comprises: Constructing a feature propagation structure based on the communication relationship between abnormal nodes, and updating the feature data through feature exchange between nodes in several rounds of interaction.

5. The method of claim 4, wherein: The step of generating a fault identification based on the updated feature data and determining the fault area comprises: Constructing a feature data matrix and an adjacency relationship matrix based on the updated feature data of each abnormal node and the node relationship, to generate the fault identification and area division result of each abnormal node.

6. The method of claim 5, wherein: The step of determining nodes with abnormal data in the power grid further comprises: Defining a node set of the power distribution network and a connection set between all nodes as lines, and defining a connection matrix A based on the monitoring lines of the protection device, wherein if the kth protection device monitors the jth line, it is represented as 1 in the connection matrix A, otherwise as 0; Defining a correlation matrix C based on the lines belonging to the power grid area, wherein if the jth line belongs to the hth area, it is represented as 1 in the correlation matrix C, otherwise as 0; Multiplying the connection matrix and the correlation matrix by matrix multiplication to obtain a dynamic matrix of the protection area and the protection device, represented as: wherein m represents the total number of lines, wherein N kh represents the relationship value of whether the kth protection device monitors the hth protection area, A kj represents the relationship value of whether the kth protection device monitors the jth line, C jh represents the relationship value of whether the jth line belongs to the hth protection area; Real-time monitoring of the impedance and current of all lines in the protection area, determining the safe impedance threshold and the safe current threshold based on the historical data of the lines, and removing the line from the protection area if the impedance and current data of the line are greater than or equal to the safe impedance threshold and the safe current threshold; Applying discrete wavelet transform based on the voltage and current data of the updated protection area lines, calculating the energy distribution of the voltage and current data of the protection area lines based on the wavelet transform, and setting the sum of the historical energy mean and twice the standard deviation of the protection area as the area threshold, if the maximum energy distribution in the protection area is greater than the area threshold, it indicates that there is an abnormal signal in the area.

7. The distributed signal processing method for distribution network lines as described in claim 6, characterized in that: The step of performing multi-level decomposition processing on the abnormal data to split the abnormal data into component data reflecting different time sequence change characteristics, and verifying the effectiveness of each layer of component data based on a preset decomposition rule during the decomposition process comprises: Based on the abnormal signal judged in the protection area, empirical mode decomposition (EMD) is used for decomposition, and the difference between the abnormal signal and the local mean value is extracted as the candidate IMF, denoted as: h1(t)=f(t)-m(t) Where h1(t) represents the first layer candidate IMF of time t, f(t) represents the abnormal signal of time t, and m(t) represents the local mean value of time t, which is calculated by the sum of the upper envelope and the lower envelope generated by the spline interpolation of the local extreme points divided by 2; The next layer IMF is recursively extracted from the residual signal, denoted as: r1(t)=f(t)-h1(t) h i (t) = r i (t) - m i (t) wherein r1(t) represents the residual signal at time t of the 1st layer, r n (t) represents the residual signal at time t of the i-th layer, h n (t) represents the candidate IMF at time t of the i-th layer, m n (t) represents the local mean value at time t of the i-th layer; h n (t) is verified, if the difference between the number of zero-crossing points and the number of extreme points of h n (t) is less than or equal to 1, and the local average value is not 0, then h n (t) is taken as the IMF of the nth layer time t. The Hilbert transform is performed on each candidate IMF to calculate the instantaneous frequency and the instantaneous amplitude, denoted as: where H i (t) represents the transformed signal of the original signal IMF i (t), IMF i (t) represents the IMF value of the i-th time variable τ, τ represents the independent time variable in the Hilbert transform, A i (t) represents the instantaneous frequency of the i-th time t, f i (t) represents the instantaneous amplitude of the i-th time t, IMF i (t) represents the IMF value of the i-th time t, φ i (t) represents the transformed instantaneous phase angle of the i-th time t; The instantaneous frequency and the instantaneous amplitude calculated for each abnormal signal are counted and combined as the state value of the corresponding edge node, denoted as: x i (t) = [f i (t), A i (t)] where x i (t) denotes the feature vector of the abnormal node i at time t.

8. A distribution line distributed signal processing system applying the distribution line distributed signal processing method according to any one of claims 1 to 7, characterized by, Including: The data collection module, the feature extraction module, the update module, and the output module; The data collection module collects node data of the power distribution network and performs time alignment processing to generate a unified reference monitoring data set and determine the nodes with abnormal data in the power grid; The feature extraction module decomposes the abnormal data and extracts feature data representing the change characteristics; The update module updates the feature data of each abnormal node based on distributed feature interaction; The output module generates a fault identification according to the updated feature data and determines the fault area. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to realize the steps of the power distribution line distributed signal processing method of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the power distribution line distributed signal processing method of any one of claims 1 to 7.

Images