Power distribution network fault analyzing and positioning method and system based on artificial intelligence

Abstract

The invention discloses a power distribution network fault analysis and positioning method and system based on artificial intelligence, and relates to the technical field of intelligent power grids, and the method comprises the steps: collecting the multi-modal data of a power distribution network through a distributed multi-pulse source and D-PMU equipment, carrying out the preprocessing, and extracting the features of a transient signal through wavelet transform; constructing a feature graph based on the extracted transient signal features, analyzing the space-time association strength between nodes according to the feature graph, and performing dynamic adjustment and weighted fusion of the space-time association weight; positioning and classifying fault points by using a graph attention network model based on the weighted fusion feature graph; according to the method, the time-space correlation strength among the nodes is analyzed and dynamically adjusted, so that the accuracy and flexibility of power distribution network fault point recognition are effectively improved, and the fault points are positioned and classified by using the graph attention network model based on the weighted fusion fault feature graph, so that the fault point recognition accuracy is improved. And the automation degree and the accuracy of fault response are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of smart grid technology, and in particular to a distribution network fault analysis and positioning method and system based on artificial intelligence. Background Art

[0002] With the development of smart grids, the automation and intelligence of distribution networks are constantly improving, especially in the field of fault analysis and location, where related technologies have been widely studied and applied. Traditional distribution network fault detection methods mainly rely on data collected by sensors and monitoring equipment, and identify faults through artificial rules or threshold-based algorithms. Although these methods played an important role in the early stages, with the increasing complexity of distribution networks, traditional methods have gradually exposed many shortcomings. For example, traditional monitoring methods lack timeliness and flexibility, and cannot respond to rapidly changing electrical conditions in distribution networks in a timely manner, thus affecting the stability and reliability of the power system.

[0003] Existing spatiotemporal correlation analysis methods often rely on static models and fail to effectively consider the dynamic correlation between nodes, resulting in poor results in fault location and classification. Summary of the invention

[0004] In view of the problems existing in the above-mentioned existing distribution network fault analysis and positioning methods and systems based on artificial intelligence, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is that the existing spatiotemporal correlation analysis methods often rely on static models and fail to effectively consider the dynamic correlation relationship between nodes, resulting in poor results in fault location and classification.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a distribution network fault analysis and location method based on artificial intelligence, which comprises:

[0007] The multimodal data of the distribution network is collected by distributed multi-pulse source and D-PMU equipment for preprocessing, and the transient signal characteristics are extracted using wavelet transform;

[0008] A feature graph is constructed based on the extracted transient signal features, and the spatiotemporal correlation strength between nodes is analyzed according to the feature graph, and the spatiotemporal correlation weights are dynamically adjusted and weighted fused;

[0009] Based on the weighted fusion feature graph, the graph attention network model is used to locate and classify the fault points;

[0010] Based on the results of fault point location and classification, formulate and implement corresponding strategies.

[0011] As a preferred solution of the distribution network fault analysis and positioning method based on artificial intelligence described in the present invention, wherein: the multimodal data of the distribution network collected by the distributed multi-pulse source and the D-PMU device includes:

[0012] Distributed multi-pulse sources and D-PMU devices are deployed at key nodes of the distribution network to collect the fault voltage V at key nodes through distributed multi-pulse sources. i (t) and current I i (t) data, the voltage and current signals of the nodes are measured by D-PMU equipment and data is collected synchronously according to GPS time synchronization. The voltage V is removed from the collected data using an adaptive filtering algorithm. i (t), current I i The high-frequency noise of (t) is measured by using the D-PMU device to obtain the final voltage and current data.

[0013] As a preferred solution of the distribution network fault analysis and location method based on artificial intelligence described in the present invention, wherein: the use of wavelet transform to extract transient signal features refers to using Daubechies wavelet to perform discrete wavelet transform on the final data after preprocessing, and performing wavelet decomposition on the voltage and current through discrete wavelet transform:

[0014]

[0015] Where x(t) represents the final signal of input voltage and current, represents the Daubechies wavelet function, a, b represent the scale and translation parameters of the wavelet respectively, W(a, b) represents the time-frequency coefficient after wavelet transformation, N represents the total number of signals, t represents the time index of the input signal, and u represents the time index in the wavelet basis function;

[0016] Extract the high-frequency component D1 from the decomposed signal as the transient feature:

[0017]

[0018] In the formula, represents the highest frequency component of the voltage signal extracted by wavelet transform, represents the highest frequency component of the current signal extracted by wavelet transform, and Represent the transient characteristics of voltage and current respectively.

[0019] As a preferred solution of the distribution network fault analysis and location method based on artificial intelligence described in the present invention, wherein: the construction of the feature graph based on the extracted transient signal features refers to constructing the feature vector of node i based on the transient voltage and current features combined with the transient features extracted by wavelet transform:

[0020]

[0021] In the formula, represents the pre-processed voltage signal of node i, represents the preprocessed current signal of node i, represents the voltage transient characteristics extracted by wavelet transform, represents the transient characteristics of current extracted by wavelet transform, g i (t) represents the feature vector of node i at time t;

[0022] Construct edge features based on the physical connection properties of adjacent nodes i and j:

[0023] e ij =[R ij , L ij , d ij ],

[0024] Where, L ij represents the inductance between node i and node j, d ij represents the length of the line between node i and node j, R ij represents the resistance between node i and node j, e ij Represents the edge feature vector between node i and node j;

[0025] The feature graph is constructed by integrating the feature vectors of all nodes and edges:

[0026] Z(t)={g1(t), g2(t),...,g i (t)},

[0027] E(t)={e 12 , e 13 , ..., e ij},

[0028] G(t)=(Z(t),E(t)),

[0029] In the formula, Z(t) represents the set of all nodes at time t, E(t) represents the set of all edges at time t, G(t) represents the feature graph at time t, and t represents the time variable.

[0030] As a preferred solution of the distribution network fault analysis and positioning method based on artificial intelligence of the present invention, wherein: the analysis of the spatiotemporal correlation strength between nodes according to the characteristic graph and the dynamic adjustment and weighted fusion of the spatiotemporal correlation weights refer to the calculation of the spatiotemporal correlation strength between node i and node j:

[0031]

[0032] Where, d ij represents the length of the line between nodes i and j, t i and t j represents the timestamp of nodes i and j, Δt represents the time step, S ij (t) represents the spatiotemporal correlation strength between nodes i and j, X i and X j represents the spatial coordinates of nodes i and j, |x i -X j | 2 Represents the square of the Euclidean distance between nodes i and j in spatial coordinates;

[0033] According to the spatiotemporal correlation strength S between node i and its neighbor node j ij (t), dynamically adjust the node weight:

[0034]

[0035] Where N(i) represents the set of neighbor nodes of node i, S ij (t) represents the spatiotemporal correlation strength between node i and its neighbor node j, w i (t) represents the association weight of node i, S ik (t) represents the spatiotemporal correlation strength between node i and its neighbor node k;

[0036] According to the node association weight w i (t), the feature vector g for node i i (t) is adjusted to obtain the weighted fusion feature map G fus (t).

[0037] As a preferred solution of the distribution network fault analysis and location method based on artificial intelligence described in the present invention, wherein: the feature graph based on weighted fusion, using the graph attention network model to locate and classify the fault point refers to constructing the architecture of the graph attention network, including an input layer, an attention mechanism layer, and a hidden layer;

[0038] Initialize the weight matrix and trainable parameters in the model, and divide the signal data into training set, validation set and test set;

[0039] By inputting the feature map into the model, the model forward propagates to calculate the new feature representation of each node, and calculates the attention weight between the node and its neighboring nodes. The cross entropy loss is used to calculate the difference between the predicted result and the actual label. According to the gradient of the loss function, the model parameters are updated through back propagation. The forward propagation, loss calculation, and back propagation steps are repeated until the model converges or reaches the predetermined training round.

[0040] By weighting the fused feature map G fus (t) is input into the trained graph attention network model, and the attention weight between node i and its neighbor node i is calculated according to the graph attention network model:

[0041] y ij =LeakyReLU(a T [Wg i ||Wg j ]),

[0042]

[0043] Where W represents the linear transformation weight matrix, g i and g j Represents the feature vector of node i and its neighbor node j, a T represents a trainable attention vector, LeakyReLU() represents the activation function of the linear rectification unit, [Wg i ||Wg j ] represents the concatenation of the feature vectors of node i and its neighbor node j, α ij represents the attention weight between node i and its neighbor node j, N(i) represents the set of neighbor nodes of node i, and y ij represents the similarity score between node i and node j, y ik Represents the similarity score between node i and node k;

[0044] According to the attention weight and the rate of change vector of node features, the faulty node is located:

[0045]

[0046] U ult = arg max i (||Δg i (t)||·α ij ),

[0047] In the formula, Δg i (t) represents the rate of change of the characteristic vector of node i when a fault occurs, g i (t) represents the feature vector of node i at time t, g i(t0) represents the feature vector of node i at the initial time t0, ||Δg i (t)|| represents the rate of change vector Δg i The norm of (t), U ult Indicates a faulty node;

[0048] By weighting the feature vector of node i at time t using the weighted summation method, we can obtain the weighted fusion fault feature vector g′ of node i. i :

[0049] g′ i =∑ j∈N(i) S ij ·g i (t),

[0050] Where N(i) represents the set of neighbor nodes of node i, g i (t) represents the feature vector of node i at time t;

[0051] Use the softmax function to calculate the fault feature vector g′ of node i i Classification is performed and the output matrix W is out Generate the probability distribution of fault type P(f i ):

[0052] P(f i )=softmax(W out g′ i ),

[0053] The fault types include short circuit fault, ground fault, disconnection fault and overload fault;

[0054] According to the probabilities of different fault types, the fault type with the highest probability is selected as the final classification result of node i.

[0055] As a preferred solution of the distribution network fault analysis and location method based on artificial intelligence of the present invention, wherein: the formulation and implementation of corresponding strategies based on the results of fault point location and classification refers to automatically generating a fault isolation strategy according to the fault type and location, using the weighted fusion fault feature map G fus (t) and the weighted edge eigenvector Determine the isolation line:

[0056]

[0057] In the formula, C iso (U ult ) represents the isolated line, N(i) represents the set of neighbor nodes of node i, U ult Indicates a faulty node;

[0058] After the strategy is generated, an execution command is issued to execute the isolation command, cut off the physical connection between the faulty node and the adjacent nodes, and restore the power supply to the non-faulty area through the backup line;

[0059] After the command is executed, the voltage and current of the faulty node are monitored in real time, and a processing report is generated after the fault processing is completed, and the data is stored in the database.

[0060] Another object of the present invention is to provide a distribution network fault analysis and location system based on artificial intelligence, which comprises:

[0061] The acquisition and feature extraction module collects multi-modal data of the distribution network through distributed multi-pulse sources and D-PMU devices for pre-processing, and uses wavelet transform to extract transient signal features;

[0062] The analysis and fusion module builds a fault feature map based on the extracted transient signal features, analyzes the spatiotemporal correlation strength between nodes according to the feature map, and dynamically adjusts and weighted-fuses the spatiotemporal correlation weights;

[0063] The positioning and classification module locates and classifies fault points based on the weighted fusion feature graph and uses the graph attention network model;

[0064] The strategy module formulates and implements corresponding strategies based on the results of fault point location and classification.

[0065] A computer device comprises: a memory and a processor; the memory stores a computer program, and the processor implements the steps of a distribution network fault analysis and positioning method based on artificial intelligence when executing the computer program.

[0066] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a distribution network fault analysis and location method based on artificial intelligence.

[0067] The beneficial effects of the present invention are as follows: the present invention effectively improves the accuracy and flexibility of fault point identification in the distribution network by analyzing the spatiotemporal correlation strength between nodes and making dynamic adjustments, and based on the weighted fusion fault feature graph, uses a graph attention network model to locate and classify fault points, thereby greatly improving the automation and accuracy of fault response. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0069] Figure 1 The figure is a flow chart of the distribution network fault analysis and location method based on artificial intelligence.

[0070] Figure 2 This is a structural diagram of the distribution network fault analysis and positioning system based on artificial intelligence. DETAILED DESCRIPTION

[0071] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0072] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0073] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0074] Example 1

[0075] Reference Figure 1 , which is the first embodiment of the present invention, and which provides a distribution network fault analysis and location method based on artificial intelligence. The distribution network fault analysis and location method based on artificial intelligence includes:

[0076] S1, collect multi-modal data of the distribution network through distributed multi-pulse source and D-PMU equipment for pre-processing, and use wavelet transform to extract transient signal characteristics;

[0077] Specifically, the multi-modal data of the distribution network collected by the distributed multi-pulse source and D-PMU equipment include:

[0078] Distributed multi-pulse sources and D-PMU devices are deployed at key nodes of the distribution network to collect the fault voltage V at key nodes through distributed multi-pulse sources. i (t) and current I i (t) Data, the voltage and current signals of the nodes are measured by D-PMU equipment and data is collected synchronously according to GPS time synchronization, and the collected data is pre-processed;

[0079] The pre-processing refers to using an adaptive filtering algorithm to remove the voltage V i (t), current Ii (t) high frequency noise:

[0080]

[0081] In the formula, represents the filtered signal, h(tT) represents the impulse response function, X i (T) represents the original signal at time T, t represents the current time, and T represents the integrated variable;

[0082] According to the statistical distribution of historical data, the outlier threshold is set to Q, and the filtered signal Compare with the outlier value Q. If the filtered signal When it is greater than or equal to the abnormal value Q, it means that the filtered signal If the signal after filtering is abnormal, the abnormal value needs to be deleted. When it is less than the abnormal value Q, it means that the filtered signal If it is normal, the filtered signal reserve;

[0083] Signal for reservation To verify time synchronization:

[0084] ΔB i (t)=|B i (t)-A syn (t)|,

[0085] In the formula, ΔB i (t) represents the deviation between the data timestamp of node i and the global synchronization, B i (t) represents the data timestamp of node i, A syn (t) represents the global GPS synchronization time;

[0086] According to the requirements of time synchronization accuracy, the normal deviation threshold is set to L, and the calculated deviation ΔB i (t) Compare with the normal deviation L. If the calculated deviation ΔB i When (t) is greater than the normal deviation L, it is asynchronous data and will be eliminated;

[0087] After time synchronization verification, the final data is obtained

[0088] By deploying distributed multi-pulse sources and D-PMU devices at key nodes of the distribution network, multi-modal data of the distribution network, including voltage and current, can be collected in real time and accurately, thus providing a high-quality raw data basis for subsequent fault analysis. And the use of adaptive filtering algorithms can effectively remove high-frequency noise in voltage and current signals, thereby improving the signal-to-noise ratio of the signal. It ensures that the obtained data is more reliable in subsequent analysis and avoids misjudgment caused by noise. By setting the outlier threshold Q, possible abnormal data can be effectively identified and eliminated. And after time synchronization verification, only data that meets the normal deviation threshold W requirements will be retained. By combining the high-quality data obtained, the subsequent fault analysis and location process will have higher accuracy and reliability.

[0089] Furthermore, using wavelet transform to extract transient signal features refers to using Daubechies wavelet to perform discrete wavelet transform on the final data after preprocessing, and performing wavelet decomposition on voltage and current through discrete wavelet transform:

[0090]

[0091] Where x(t) represents the final signal of input voltage and current, represents the Daubechies wavelet function, a, b represent the scale and translation parameters of the wavelet, W(a, b) represents the time-frequency coefficient after wavelet transformation, N represents the total number of signals, t represents the time index of the input signal, and u represents the time index in the wavelet basis function;

[0092] Extract the high-frequency component D1 from the decomposed signal as the transient feature:

[0093]

[0094] In the formula, represents the highest frequency component of the voltage signal extracted by wavelet transform, represents the highest frequency component of the current signal extracted by wavelet transform, and Represent the transient characteristics of voltage and current respectively.

[0095] By using Daubechies wavelet for discrete wavelet transform, the voltage and current signals can be efficiently decomposed and the high-frequency components of transient fault signals can be extracted. This enables the present invention to accurately capture the rapid changes when a fault occurs, providing key data support for subsequent analysis. The multi-resolution analysis characteristics of wavelet transform enable the present invention to flexibly respond to signals with different frequency characteristics. Selecting appropriate wavelet scale and translation parameters according to the frequency characteristics and time resolution requirements of the signal can meet the analysis requirements in specific application scenarios, such as improving the time resolution to capture transient faults. By extracting high-frequency components as transient fault features, the impact of background noise can be effectively reduced, thereby improving the accuracy of fault signal identification. The extracted transient fault features provide an effective information basis for subsequent fault diagnosis and location algorithms. Combined with machine learning or deep learning technology, the intelligence level of fault detection can be further improved, helping operation and maintenance personnel to quickly respond to potential faults.

[0096] S2. Construct a feature graph based on the extracted transient signal features, analyze the spatiotemporal correlation strength between nodes according to the feature graph, and dynamically adjust and weightedly fuse the spatiotemporal correlation weights;

[0097] Specifically, constructing a feature map based on the extracted transient signal features means constructing a feature vector of node i based on the transient voltage and current features combined with the transient features extracted by wavelet transform:

[0098]

[0099] In the formula, represents the pre-processed voltage signal of node i, represents the preprocessed current signal of node i, represents the voltage transient characteristics extracted by wavelet transform, represents the transient characteristics of current extracted by wavelet transform, g i (t) represents the feature vector of node i at time t;

[0100] Construct edge features based on the physical connection properties of adjacent nodes i and j:

[0101] e ij =[R ij , L ij , d ij ],

[0102] Where, L ij represents the inductance between node i and node j, d ij represents the length of the line between node i and node j, R i represents the resistance between node i and node j, e ij Represents the edge feature vector between node i and node j;

[0103] The feature graph is constructed by integrating the feature vectors of all nodes and edges:

[0104] Z(t)={g1(t), g2(t),...,g i (t)},

[0105] E(t)={e 12 , e 13 , ..., e ij},

[0106] G(t)=(Z(t),E(t)),

[0107] Where Z(t) represents the set of all nodes at time t, E(t) represents the set of all edges at time t, G(t) represents the fault feature graph at time t, and t represents the time variable.

[0108] By combining transient voltage and current features and transient high-frequency features extracted by wavelet transform, the constructed feature vector can fully reflect the node status. This comprehensive feature extraction method provides more abundant information for fault diagnosis and helps to improve the accuracy of fault identification. And by constructing edge features through physical properties such as inductance, line length and resistance, the connection relationship between nodes can be accurately described. The introduction of physical connection features enhances the expressiveness of the fault feature graph, making fault analysis more scientific and reasonable. Moreover, the present invention can dynamically update the fault feature graph according to the real-time collected data. Thereby ensuring that the fault analysis is always based on the latest data, it helps to quickly respond to possible faults in the power system. The inductance, length and resistance can be obtained through the power system configuration data. By integrating the feature vectors of all nodes and edges, the constructed fault feature graph can effectively support the subsequent fault location algorithm. And by combining machine learning and data mining technology, the analysis based on the fault feature graph can provide support for intelligent decision-making of the power system. This intelligent decision-making capability can automatically identify potential faults and recommend corresponding maintenance strategies.

[0109] Furthermore, the temporal and spatial correlation strength between nodes is analyzed according to the feature graph, and the temporal and spatial correlation weights are dynamically adjusted and weighted fused, which means calculating the temporal and spatial correlation strength between node i and node j:

[0110]

[0111] Where, d ij represents the length of the line between nodes i and j, t i and t j represents the timestamp of nodes i and j, Δt represents the time step, S ij (t) represents the spatiotemporal correlation strength between nodes i and j, X i and X jrepresents the spatial coordinates of nodes i and j, |X i -X j | 2 Represents the square of the Euclidean distance between nodes i and j in spatial coordinates;

[0112] In the distribution network, the line length may not be consistent with the geometric distance. The line length may bend or detour due to actual obstacles. Therefore, the line length d ij Normalization can be performed to convert the geometric distance into a scale that better matches the physical layout of the power grid;

[0113] According to the spatiotemporal correlation strength S between node i and its neighbor node j ij (t), dynamically adjust the weight of the node:

[0114]

[0115] Where N(i) represents the set of neighbor nodes of node i, S ij (t) represents the spatiotemporal correlation strength between node i and its neighbor node j, w i (t) represents the association weight of node i, S ik (t) represents the spatiotemporal correlation strength between node i and its neighbor node k;

[0116] According to the node association weight w i (t), the feature vector g for node i i (t) is adjusted to obtain the weighted fusion feature map G fus (t).

[0117] By analyzing the temporal and spatial correlation strength between nodes, the fault point can be identified more accurately. In this way, the dynamic relationship between nodes can be effectively captured, thereby improving the accuracy of fault identification. The spatial coordinates are obtained according to the geographical layout of the distribution network. The mechanism of dynamically adjusting the node weights ensures that the fault characteristic diagram can reflect the operating status of the distribution network in real time. The real-time updated characteristic diagram enables the fault analysis process to respond quickly to system changes, improving the real-time and adaptability of the system. By weighting the characteristic vectors of the fusion nodes, the importance of each node can be comprehensively considered to obtain a more accurate fault characteristic diagram. Thereby improving the utilization efficiency of information, and the present invention can be effectively applied to complex distribution network environments, taking into account the spatial layout and time synchronization between nodes in the network, so that it has efficient fault analysis capabilities in a variety of application scenarios.

[0118] S3, based on the weighted fusion feature graph, uses the graph attention network model to locate and classify the fault points;

[0119] Specifically, based on the weighted fusion feature graph, using the graph attention network model to locate and classify the fault point refers to constructing the architecture of the graph attention network, including an input layer, an attention mechanism layer, and a hidden layer;

[0120] Initialize the weight matrix and trainable parameters in the model, and divide the signal data into training set, validation set and test set;

[0121] By inputting the feature map into the model, the model forward propagates to calculate the new feature representation of each node, and calculates the attention weight between the node and its neighboring nodes. The cross entropy loss is used to calculate the difference between the predicted result and the actual label. According to the gradient of the loss function, the model parameters are updated through back propagation. The forward propagation, loss calculation, and back propagation steps are repeated until the model converges or reaches the predetermined training round.

[0122] By weighting the fused feature map G fus (t) is input into the trained graph attention network model, and the attention weight between node i and neighbor node j is calculated according to the graph attention network model:

[0123] y ij =LeakyReLU(a T |Wg i ||Wg j |),

[0124]

[0125] Where W represents the linear transformation weight matrix, g i and g j Represents the feature vector of node i and its neighbor node j, a T represents a trainable attention vector, LeakyReLU() represents the activation function of the linear rectification unit, [Wg i ||Wg j ] represents the concatenation of the feature vectors of node i and its neighbor node j, α ij represents the attention weight between node i and its neighbor node j, N(i) represents the set of neighbor nodes of node i, and y ij represents the similarity score between node i and node j, y ik Represents the similarity score between node i and node k;

[0126] According to the attention weight and the rate of change vector of node features, the faulty node is located:

[0127]

[0128] U ult = arg max i (||Δgi (t)||·α ij ),

[0129] In the formula, Δg i (t) represents the rate of change of the characteristic vector of node i when a fault occurs, g i (t) represents the feature vector of node i at time t, g i (t0) represents the feature vector of node i at the initial time t0, ||Δg i (t)|| represents the rate of change vector Δg i The norm of (t), U ult Indicates a faulty node;

[0130] By transforming the rate of change vector Δg i (t) is converted into a norm. The norm of a vector can provide a scalar value, so that the nodes can be compared and the node with the largest change rate can be found, thereby ensuring that the vector change rate information is effectively utilized;

[0131] By weighting the feature vector of node i at time t using the weighted summation method, we can obtain the weighted fusion fault feature vector g′ of node i. i :

[0132] g′ i =∑ j∈N(i) S ij ·g i (t),

[0133] Where N(i) represents the set of neighbor nodes of node i, g i (t) represents the feature vector of node i at time t;

[0134] Use the softmax function to calculate the fault feature vector g′ of node i i Classification is performed and the output matrix W is out Generate the probability distribution of fault type P(f i ):

[0135] P(f i )=softmax(W out g i ),

[0136] The fault types include short circuit fault, ground fault, disconnection fault and overload fault;

[0137] According to the probabilities of different fault types, the fault type with the highest probability is selected as the final classification result of node i.

[0138] The calculation of the rate of change reflects the dynamic characteristics of the node in a specific time period. Especially when a fault occurs, the rate of change of the node characteristics will increase or decrease significantly. The introduction of the rate of change can effectively capture the abnormal fluctuations of the node, help locate the faulty node in the time series, and lock the spatiotemporal characteristics of the fault source through the analysis of dynamic changes.

[0139] By constructing a graph attention network model, the present invention can accurately calculate the attention weights between a node and its neighboring nodes according to the fault feature graph. The graph-based model can more effectively capture the complex relationships between nodes and improve the accuracy of locating the source of the fault. During the training process, the model updates parameters through back propagation so that node features can be dynamically adjusted as the data changes. By using the softmax function to classify node features, a clear probability distribution of fault types can be generated. By selecting the fault type with the highest probability as the final classification result, this method can effectively reduce the risk of misclassification and improve the accuracy of fault classification. The present invention can identify multiple fault types such as short circuit faults, ground faults, line break faults and overload faults, and has strong practicality.

[0140] S4. Based on the results of fault location and classification, formulate and implement corresponding strategies;

[0141] Specifically, based on the results of fault point location and classification, formulate and implement corresponding strategies, which refers to automatically generating fault isolation strategies according to fault type and location, using the weighted fusion fault feature map G fus (t) and the weighted edge eigenvector Determine the isolation line:

[0142]

[0143] In the formula, C iso (U ult ) represents the isolated line, N(i) represents the set of neighbor nodes of node i, U ult Indicates a faulty node;

[0144] In the distribution network, the edge between nodes is not only a representation of a connection, but also contains a variety of physical interaction information. It can capture the correlation between nodes more accurately, and the edge feature vector Able to provide faulty node U ult The degree of fault impact between its neighbors. When a fault occurs, the change in edge characteristics can often reveal the different impacts of the fault on different neighboring nodes, thereby better evaluating the propagation path and scope of the fault;

[0145] After the strategy is generated, an execution command is issued to execute the isolation command, cut off the physical connection between the faulty node and the adjacent nodes, and restore the power supply to the non-faulty area through the backup line;

[0146] After the command is executed, the voltage and current of the faulty node are monitored in real time, and a processing report is generated after the fault processing is completed, and the data is stored in the database.

[0147] Automatically generate fault isolation strategies based on fault type and location, reducing the need for manual intervention. Improve the timeliness of fault handling. By using edge feature vectors to determine the isolation line, the source of the fault and its neighboring nodes can be more accurately identified, thereby effectively preventing the spread of the fault. After the fault is isolated, the use of alternative power supply paths to restore power to non-faulty areas helps ensure the continuity of power supply. Real-time monitoring of the voltage and current at the source of the fault can obtain the system operation status in a timely manner. After the fault is handled, a detailed processing report is generated and the relevant data is stored in the historical database. This approach not only provides valuable data resources for subsequent fault analysis, but can also be used to optimize future fault detection and response strategies.

[0148] Example 2

[0149] Reference Figure 2 , which is the second embodiment of the present invention, and which is different from the previous embodiment, provides a distribution network fault analysis and positioning system based on artificial intelligence, which includes:

[0150] The acquisition and feature extraction module collects multi-modal data of the distribution network through distributed multi-pulse sources and D-PMU devices for pre-processing, and uses wavelet transform to extract transient signal features;

[0151] The analysis and fusion module builds a fault feature map based on the extracted transient signal features, analyzes the spatiotemporal correlation strength between nodes according to the feature map, and dynamically adjusts and weighted-fuses the spatiotemporal correlation weights;

[0152] The positioning and classification module locates and classifies fault points based on the weighted fusion feature graph and uses the graph attention network model;

[0153] The strategy module formulates and implements corresponding strategies based on the results of fault point location and classification.

[0154] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0155] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0156] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0157] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

Claims

1. A distribution network fault analysis and location method based on artificial intelligence, characterized in that: include, The multimodal data of the distribution network is collected by distributed multi-pulse source and D-PMU equipment for preprocessing, and the transient signal characteristics are extracted using wavelet transform; A feature graph is constructed based on the extracted transient signal features, and the spatiotemporal correlation strength between nodes is analyzed according to the feature graph, and the spatiotemporal correlation weights are dynamically adjusted and weighted fused; Based on the weighted fusion feature graph, the graph attention network model is used to locate and classify the fault points; Based on the results of fault point location and classification, formulate and implement corresponding strategies.

2. The method for analyzing and locating distribution network faults based on artificial intelligence according to claim 1, characterized in that: The multi-modal data of the distribution network collected by the distributed multi-pulse source and the D-PMU device includes: Distributed multi-pulse sources and D-PMU devices are deployed at key nodes of the distribution network to collect the fault voltage V at key nodes through distributed multi-pulse sources. i (t) and current I i (t) data, the voltage and current signals of the nodes are measured by D-PMU equipment and data is collected synchronously according to GPS time synchronization. The voltage V is removed from the collected data using an adaptive filtering algorithm. i (t), current I i The high-frequency noise of (t) is measured by using the D-PMU device to obtain the final voltage and current data.

3. The method for analyzing and locating distribution network faults based on artificial intelligence according to claim 2, characterized in that: The use of wavelet transform to extract transient signal features refers to using Daubechies wavelet to perform discrete wavelet transform on the final data after preprocessing, and performing wavelet decomposition on the voltage and current through discrete wavelet transform: Where x(t) represents the final signal of input voltage and current, represents the Daubechies wavelet function, a, b represent the scale and translation parameters of the wavelet, W(a, b) represents the time-frequency coefficient after wavelet transformation, N represents the total number of signals, t represents the time index of the input signal, and u represents the time index in the wavelet basis function; Extract the high-frequency component D1 from the decomposed signal as the transient feature: In the formula, represents the highest frequency component of the voltage signal extracted by wavelet transform, represents the highest frequency component of the current signal extracted by wavelet transform, and Represent the transient characteristics of voltage and current respectively.

4. The method for analyzing and locating distribution network faults based on artificial intelligence according to claim 3, characterized in that: The construction of the feature graph based on the extracted transient signal features refers to constructing a feature vector of node i based on the transient voltage and current features combined with the transient features extracted by wavelet transform: In the formula, represents the pre-processed voltage signal of node i, represents the preprocessed current signal of node i, represents the voltage transient characteristics extracted by wavelet transform, represents the transient characteristics of current extracted by wavelet transform, g i (t) represents the feature vector of node i at time t; Construct edge features based on the physical connection properties of adjacent nodes i and j: e ij =[R ij ,L ij ,d ij ], Where, L ij represents the inductance between node i and node j, d ij represents the length of the line between node i and node j, R ij represents the resistance between node i and node j, e ij Represents the edge feature vector between node i and node j; The feature graph is constructed by integrating the feature vectors of all nodes and edges: Z(t)={g1(t),g2(t),...,g i (t)}, E(t)={and 12 ,And 13 ,...,And ij }, G(t)=(Z(t),E(t)), In the formula, Z(t) represents the set of all nodes at time t, E(t) represents the set of all edges at time t, G(t) represents the feature graph at time t, and t represents the time variable.

5. The method for analyzing and locating distribution network faults based on artificial intelligence according to claim 4, characterized in that: The analysis of the spatiotemporal correlation strength between nodes according to the feature graph and the dynamic adjustment and weighted fusion of the spatiotemporal correlation weights refers to the calculation of the spatiotemporal correlation strength between node i and node j: Where, d ij represents the length of the line between nodes i and j, t i and t j represents the timestamp of nodes i and j, Δt represents the time step, S ij (t) represents the spatiotemporal correlation strength between nodes i and j, X i and X j represents the spatial coordinates of nodes i and j, |X i -X j | 2 Represents the square of the Euclidean distance between nodes i and j in spatial coordinates; According to the spatiotemporal correlation strength S between node i and its neighbor node j ij (t), dynamically adjust the node weight: Where N(i) represents the set of neighbor nodes of node i, S ij (t) represents the spatiotemporal correlation strength between node i and its neighbor node j, w i (t) represents the association weight of node i, S ik (t) represents the spatiotemporal correlation strength between node i and its neighbor node k; According to the node association weight w i (t), the feature vector g for node i i (t) is adjusted to obtain the weighted fusion feature map G fus (t).

6. The method for analyzing and locating distribution network faults based on artificial intelligence according to claim 5, characterized in that: The method of locating and classifying fault points using a graph attention network model based on a weighted fusion feature graph refers to constructing a graph attention network architecture, including an input layer, an attention mechanism layer, and a hidden layer; Initialize the weight matrix and trainable parameters in the model, and divide the signal data into training set, validation set and test set; By inputting the feature map into the model, the model forward propagates to calculate the new feature representation of each node, and calculates the attention weight between the node and its neighboring nodes. The cross entropy loss is used to calculate the difference between the predicted result and the actual label. According to the gradient of the loss function, the model parameters are updated through back propagation. The forward propagation, loss calculation, and back propagation steps are repeated until the model converges or reaches the predetermined training round. By weighting the fused feature map G fus (t) is input into the trained graph attention network model, and the attention weight between node i and neighbor node j is calculated according to the graph attention network model: and ij =LeakyReLU(a T [According to i ||According to j ]), Where W represents the linear transformation weight matrix, g i and g j Represents the feature vector of node i and its neighbor node j, a T represents a trainable attention vector, LeakyReLU() represents the activation function of the linear rectification unit, [Wg i |Wg j ] represents the concatenation of the feature vectors of node i and its neighbor node j, α ij represents the attention weight between node i and its neighbor node j, N(i) represents the set of neighbor nodes of node i, and y ij represents the similarity score between node i and node j, y ik Represents the similarity score between node i and node k; According to the attention weight and the rate of change vector of node features, the faulty node is located: U ult =arg max i (||Δg i (t)||·a ij ), In the formula, Δg i (t) represents the rate of change of the characteristic vector of node i when a fault occurs, g i (t) represents the feature vector of node i at time t, g i (t0) represents the feature vector of node i at the initial time t0, ||Δg i (t)|| represents the rate of change vector Δg i The norm of (t), U ult Indicates a faulty node; By using the weighted summation method to weight the feature vector of node i at time t, the weighted fusion fault feature vector g' of node i is obtained. i : g i =∑ j∈N(i) S ij ·g i (t), In the formula, N(i) represents the set of neighbor nodes of node i, g i (t) represents the feature vector of node i at time t; Use the softmax function to calculate the fault feature vector g' of node i i Classification is performed and the output matrix W is out Generate the probability distribution of fault type P(f i ): P(f i )=softmax(W out g i ), The fault types include short circuit fault, ground fault, disconnection fault and overload fault; According to the probabilities of different fault types, the fault type with the highest probability is selected as the final classification result of node i.

7. The method for analyzing and locating distribution network faults based on artificial intelligence according to claim 6, characterized in that: The formulating and implementing corresponding strategies based on the results of fault point location and classification refers to automatically generating fault isolation strategies according to the fault type and location, using the weighted fusion fault feature map G fus (t) and the weighted edge eigenvector Determine the isolation line: In the formula, C iso (U ult ) represents the isolated line, N(i) represents the set of neighbor nodes of node i, U ult Indicates a faulty node; After the strategy is generated, an execution command is issued to execute the isolation command, cut off the physical connection between the faulty node and the adjacent nodes, and restore the power supply to the non-faulty area through the backup line; After the command is executed, the voltage and current of the faulty node are monitored in real time, and a processing report is generated after the fault processing is completed, and the data is stored in the database.

8. An artificial intelligence-based distribution network fault analysis and positioning system based on the artificial intelligence-based distribution network fault analysis and positioning method according to any one of claims 1 to 7, characterized in that: include, The acquisition and feature extraction module collects multi-modal data of the distribution network through distributed multi-pulse sources and D-PMU devices for pre-processing, and uses wavelet transform to extract transient signal features; The analysis and fusion module constructs a feature graph based on the extracted transient signal features, analyzes the spatiotemporal correlation strength between nodes based on the feature graph, and dynamically adjusts and weighted-fuses the spatiotemporal correlation weights; The positioning and classification module locates and classifies fault points based on the weighted fusion feature graph and uses the graph attention network model; The strategy module formulates and implements corresponding strategies based on the results of fault point location and classification.

9. A computer device comprising: Memory and processor; The memory stores a computer program, characterized in that: when the processor executes the computer program, the steps of the distribution network fault analysis and positioning method based on artificial intelligence described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the distribution network fault analysis and location method based on artificial intelligence described in any one of claims 1 to 7 are implemented.

Images