Active power distribution network fault section identification method of characteristic current signal energy

Abstract

The invention provides an active power distribution network fault section identification method based on characteristic current signal energy, and relates to the technical field of fault identification, and the method comprises the steps: carrying out the analysis of a preliminary fault region through employing a distributed optimization algorithm according to a preliminary positioning result, dividing a fault positioning task into tasks corresponding to a plurality of sub-regions according to the topological structure of a power distribution network, and carrying out the calculation of the sub-regions; a task corresponding to each sub-region is allocated to a distributed node, and each distributed node independently executes optimization calculation of node current signal data in the sub-region according to the task corresponding to each sub-region to obtain a target value corresponding to each sub-region; and calculating the weight of each sub-region, carrying out weighted average on the target values corresponding to all the sub-regions, and generating a global final fault positioning result. According to the method, the abnormal nodes are screened out through global energy analysis and positioned, the energy value of each node is calculated, the area formed by connecting a plurality of nodes is identified, and errors caused by independent calculation of the nodes in a traditional method are avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of fault identification, and in particular to a method for identifying a fault section of an active distribution network based on characteristic current signal energy. Background Art

[0002] Distribution network fault diagnosis and location are key issues in power system operation and maintenance. Especially in large-scale power networks, fast and accurate location of fault sections is crucial to improving power supply reliability and reducing power outage time. Conventional fault detection methods mainly rely on current and voltage-based protection devices, but these methods have certain limitations in accuracy and response speed in complex network structures, especially in multi-fault scenarios. Specifically, the following limitations cause problems:

[0003] The problem of fault location accuracy: conventional methods are difficult to achieve accurate fault section location, especially when the distribution network structure is complex and the fault types are diverse, misjudgment or missed judgment is prone to occur.

[0004] Real-time and response speed issues,In conventional methods, the fault location process is slow, resulting in a delay in fault recovery time.

[0005] Due to insufficient data processing and analysis capabilities, conventional technical solutions are difficult to process the massive current signal data in the distribution network efficiently and in real time and to obtain accurate fault judgment results. Summary of the invention

[0006] The present invention provides a method for identifying fault sections of an active distribution network based on characteristic current signal energy, aiming to solve the problem of fault location accuracy in related technologies, avoid real-time and response speed problems, and further circumvent the problem of insufficient data processing and analysis capabilities.

[0007] In order to achieve the above objectives, this application provides the following technical solutions:

[0008] A method for identifying fault sections of an active distribution network based on characteristic current signal energy, the method comprising the following steps:

[0009] The original current signal data of each node in the distribution network is obtained and preprocessed.

[0010] According to the preprocessed current signal data, the global energy analysis method is used to calculate the energy value of each node, identify the area composed of multiple nodes, connect the nodes with abnormal energy value change rate in the area, and obtain the preliminary fault area.

[0011] Local features are extracted from the preprocessed current signal data in the preliminary fault area to obtain local features, and weights are dynamically adjusted according to the correlation between the local features and the fault to form a dynamically adjusted multidimensional feature vector.

[0012] The dynamically adjusted multidimensional feature vector is subjected to standardization processing to obtain the standardized multidimensional feature vector, and the principal component analysis is used to reduce the dimension of the standardized multidimensional feature vector to obtain the reduced-dimensional multidimensional feature vector.

[0013] A deep neural network is used to fuse and learn the multi-dimensional feature vectors after dimensionality reduction, and output the fault type and preliminary positioning results.

[0014] According to the preliminary positioning results, a distributed optimization algorithm is used to analyze the preliminary fault area, and the fault location task is divided into tasks corresponding to multiple sub-areas according to the topological structure of the distribution network. The tasks corresponding to each sub-area are assigned to distributed nodes. Each distributed node independently performs the optimization calculation of the node current signal data in the sub-area according to the tasks corresponding to each sub-area, and obtains the target value corresponding to each sub-area.

[0015] The weight of each sub-area is calculated, and the target values ​​corresponding to all sub-areas are weighted averaged to generate the final global fault location result.

[0016] According to the final fault location result, the fault section in the distribution network topology is matched.

[0017] As a preferred solution of the present invention, the original current signal data of each node in the distribution network is obtained and preprocessed. The specific steps are as follows:

[0018] The original current signal data of each node is obtained based on the current sensor preset at the distribution network node.

[0019] The high-frequency noise caused by external environmental interference and sensor errors is removed from the original current signal data of each node.

[0020] The denoised current signal data is normalized, and the normalized current signal data is segmented according to a time window to obtain preprocessed current signal data.

[0021] As a preferred solution of the present invention, based on the preprocessed current signal data, a global energy analysis method is used to calculate the energy value of each node, identify the area composed of multiple nodes, connect the nodes with abnormal energy value change rate in the area, and obtain the preliminary fault area. The specific steps are as follows:

[0022] The global energy analysis method is used to perform time integration on the preprocessed current signal data of each node, and the energy value of the current signal data of each node in the time window is calculated; based on the energy values ​​of multiple continuous time windows, the energy value change rate of each node is calculated.

[0023] It is determined whether the energy value change rate of each node in the area meets the abnormal determination condition; if the abnormal determination condition is met, it means that the node is a node with abnormal energy value change rate.

[0024] Based on the topological structure of the distribution network, the nodes with abnormal energy value change rates in the area are connected to obtain the preliminary fault area.

[0025] As a preferred solution of the present invention, local feature extraction is performed on the preprocessed current signal data in the preliminary fault area to obtain local features, and the weights are dynamically adjusted according to the correlation between the local features and the fault to form a dynamically adjusted multidimensional feature vector. The specific steps are as follows:

[0026] The local features are extracted from the preprocessed current signal data in the preliminary fault area to obtain multiple local features. The correlation analysis between the multiple local features and the fault is performed to calculate the dynamic adjustment weight of the correlation corresponding to each local feature. The calculation formula is:

[0027]

[0028] In the formula, MI(F i , L) represents the mutual information value between the i-th local feature Fi and the fault L, which is used to measure the i-th local feature F i Correlation with fault L; W fi Indicates the dynamic adjustment weight of the correlation corresponding to the i-th local feature.

[0029] The weights are dynamically adjusted according to the correlation corresponding to each local feature, and the weights w after dynamic adjustment of each local feature are calculated. i , the formula is:

[0030]

[0031] In the formula, w i represents the weight of the i-th local feature after dynamic adjustment, represents the dynamic adjustment weight of the correlation corresponding to the i-th local feature, and Z represents the total number of local features.

[0032] The local features and the dynamically adjusted weights of the features are weighted to obtain multiple weighted features, which are expressed as:

[0033] F i ′=w i ·Fi

[0034] In the formula, F i ′ represents the i-th weighted feature after weighting, F i represents the original i-th local feature, w i is the dynamic adjustment weight of the i-th local feature.

[0035] All weighted features are combined to form a dynamically adjusted multidimensional feature vector, which is expressed as:

[0036] F adjusted =[F1′, F2′, … F Z ′]

[0037] In the formula, F i ′ represents the i-th weighted feature after weighting, F adjusted Represents a dynamically adjusted multidimensional feature vector.

[0038] As a preferred solution of the present invention, a standardized processing is performed on the dynamically adjusted multidimensional feature vector to obtain a standardized multidimensional feature vector, and a principal component analysis is used to reduce the dimension of the standardized multidimensional feature vector to obtain a reduced multidimensional feature vector. The specific steps are:

[0039] The dynamically adjusted multidimensional feature vector is standardized to eliminate the dimension and range differences of the feature vector, and all standardized feature vectors are combined to form a standardized multidimensional feature vector.

[0040] The principal component analysis is performed on the standardized multidimensional feature vector, the main components are selected through eigenvalue decomposition, and the dimension reduction matrix is ​​generated.

[0041] The dimension reduction matrix is ​​used to perform linear transformation on the standardized multidimensional feature vector to generate a reduced-dimensional multidimensional feature vector.

[0042] As a preferred solution of the present invention, the deep neural network includes an input layer, a hidden layer and an output layer; the input layer inputs the multidimensional feature vector after dimensionality reduction; the hidden layer realizes the fusion and learning between multidimensional features through a nonlinear activation function; the output layer completes the fault type classification and the output of the preliminary positioning results.

[0043] As a preferred solution of the present invention, each distributed node independently performs the optimization calculation of the node current signal data in the sub-region according to the tasks corresponding to each sub-region, and obtains the target value corresponding to each sub-region. The specific calculation formula is:

[0044]

[0045] Where, Tk represents the target value corresponding to the kth sub-region, n k represents the total number of nodes in the kth sub-region, and They represent the actual current value and predicted current value of the i-th node respectively.

[0046] As a preferred solution of the present invention, the weight of each sub-area is calculated, and the target values ​​corresponding to all sub-areas are weighted averaged to generate a global final fault location result. The specific steps are as follows:

[0047] The weight of each sub-region is calculated based on the target value corresponding to each sub-region. The calculation formula is:

[0048]

[0049] Where W k represents the weight of the kth sub-region, K represents the total number of sub-regions, T j represents the target value of the jth sub-region, T k Represents the target value of the kth sub-region.

[0050] The target values ​​and weights corresponding to all sub-areas are aggregated, and the final global fault location result is generated by weighted average calculation. The calculation formula is:

[0051]

[0052] Where, T k represents the target value corresponding to the kth sub-region, R represents the final fault location result, K represents the total number of sub-regions, and W k Indicates the weight of the kth sub-area. The final fault location result is obtained based on the calculation. k indicates the sub-area number currently being calculated.

[0053] As a preferred solution of the present invention, according to the final fault location result, the fault section in the distribution network topology is matched, and the specific steps are as follows:

[0054] The final fault location result is compared with the target value T corresponding to each area in the distribution network topology. i Perform comparison calculation, the comparison calculation formula is:

[0055]

[0056] Where M i represents the matching degree between the final fault location result and the ith region, T i represents the characteristic target value of the i-th region, which is obtained by counting the node data in the region; σ i Represents the standard deviation of the target value of the ith region, which is used to reflect the consistency of the nodes in the region.

[0057] Sort all calculated matching degrees and select the area with the smallest matching degree. The calculation formula is:

[0058] i min =argmin i∈[1，N] M i

[0059] In the formula, i min Indicates the area with the smallest matching degree, M i represents the matching degree between the final fault location result and the i-th region, min represents the minimum operator, and i∈[1,N] represents that i is the index in the set of integers from 1 to N.

[0060] The area i with the smallest matching degree min Determine the faulty section and record the corresponding section number, node number and location coordinates of the faulty section.

[0061] The beneficial effects of the present application are as follows: by using a global energy analysis method, the energy value of each node is calculated and the area composed of multiple nodes is identified, thus avoiding the error caused by independent calculation of nodes in the traditional method. The nodes with abnormal energy value change rates of each node in the area are connected to screen out the initial fault area, improve the accuracy of fault location, and reduce the occurrence of misjudgment and missed judgment; the fault location task is decomposed into multiple subtasks by using a distributed optimization algorithm, and the optimization calculation is performed in parallel in the distributed nodes to shorten the time of fault location; compared with the disadvantage of the slow fault location process in the conventional method, the present invention improves the real-time response capability and reduces the fault recovery time through parallel calculation.

[0062] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0064] Figure 1 A flow chart of a method for identifying a fault section in an active distribution network based on characteristic current signal energy provided in an embodiment of the present application. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0066] See also Figure 1 , Figure 1 A flowchart of a method for identifying a fault section in an active distribution network based on characteristic current signal energy is provided for an embodiment of the present application.

[0067] In this embodiment, a method for identifying a fault section of an active power distribution network based on characteristic current signal energy includes step S101, step S102, step S103, step S104, step S105, step S106 and step S107.

[0068] Step S101: obtaining the original current signal data of each node in the distribution network and preprocessing the original current signal data. The specific steps are as follows:

[0069] The original current signal data of each node is obtained based on the current sensor preset at the distribution network node.

[0070] The high-frequency noise caused by external environmental interference and sensor errors is removed from the original current signal data of each node.

[0071] The denoised current signal data is normalized, and the normalized current signal data is segmented according to a time window to obtain preprocessed current signal data.

[0072] It should be noted that the acquired raw current signal data includes transient and steady-state data of the three-phase current. The raw current signal data is sampled at fixed time intervals, and the sampling frequency is higher than the upper limit of the characteristic frequency of the fault signal.

[0073] It should be further explained that the denoising process uses a wavelet transform method to decompose the multi-scale components of the original current signal data and filter out high-frequency noise. High-frequency noise is defined as: current signal data outside the range of 1kHz to 10kHz is regarded as high-frequency noise.

[0074] It should be further explained that the normalization formula is:

[0075]

[0076] Where, I′ is the normalized current signal data, I min and I max They are respectively the minimum and maximum values ​​of the current node current signal data.

[0077] It should be further explained that the segment length is dynamically set according to the characteristic period of the fault current signal data.

[0078] Step S102: Based on the preprocessed current signal data, a global energy analysis method is used to calculate the energy value of each node, identify an area formed by connecting multiple nodes, connect the nodes with abnormal energy value change rates in the area, and obtain a preliminary fault area. The specific steps are as follows:

[0079] The global energy analysis method is used to perform time integration on the preprocessed current signal data of each node, and the energy value of the current signal data of each node in the time window is calculated; based on the energy values ​​of multiple continuous time windows, the energy value change rate of each node is calculated.

[0080] It should be noted that the global energy analysis method is used to perform time integration on the preprocessed current signal data of each node, and the energy value of the current signal data of each node in the time window is calculated. The calculation formula is:

[0081]

[0082] In the formula, E i represents the energy value of the current signal data of the i-th node in the time window [t0, t1], I i (t) represents the amplitude of the current signal data of the i-th node at time t, t0 and t1 represent the start time and end time of the time window in the global energy analysis, respectively, and dt represents the small time increment of the integration, that is, the change amplitude of the time variable t.

[0083] It is determined whether the energy value change rate of each node in the area meets the abnormal determination condition; if the abnormal determination condition is met, it means that the node is a node with abnormal energy value change rate.

[0084] It should be noted that the energy value change rate of each node is established based on the time-integrated energy values ​​of multiple continuous time windows. The calculation formula for the node energy value change rate is:

[0085]

[0086] Where ΔE i represents the rate of change of the energy value of the i-th node in the continuous time windows j and j+1, E i,j and E i,j+1 They represent the time-integrated energy values ​​of the i-th node in time windows j and j+1 respectively.

[0087] According to the calculation results of the node energy value change rate, the abnormal nodes are judged and the energy value change rate ΔE of each node is determined. i Whether the abnormality determination conditions are met; if the abnormality determination conditions are met, it means that the node is a node with abnormal energy value change rate; if the abnormality determination conditions are not met, it means that the node is a normal node; the abnormality determination conditions are:

[0088]

[0089] Where ΔE i represents the rate of change of the energy value of the i-th node, represents the average energy value change rate of all nodes in the region, σ ΔE It represents the standard deviation of the rate of change of energy values ​​of all nodes in the region. k represents the hyperparameter used to adjust the sensitivity of abnormal judgment, and its value range is set to 1.5 to 3.

[0090] It should be noted that when the energy value change rate of the node ΔE i Exceeds the average energy value change rate of all nodes in the area k times the standard deviation σ ΔE , the node is judged as a node with abnormal energy value change rate.

[0091] Based on the topological structure of the distribution network, the area connected by multiple nodes with abnormal energy value change rates is identified as the preliminary fault area.

[0092] It should be further explained that, through the topological structure of the distribution network, that is, the connection relationship between nodes and lines, the various nodes in the distribution network are connected into a graph; the connection between nodes reflects the physical connection between power lines and transformers; the distribution network topology is represented by a network graph, the nodes represent the various elements in the distribution network, and the edges represent the power transmission lines between them; for each node with an abnormal energy value change rate, based on the node connection relationship in the distribution network topological structure, find out which adjacent nodes the node is connected to.

[0093] Specifically, the connection between nodes is represented according to the adjacency matrix or adjacency list of the graph; if node i and node j are connected, and they are both nodes with abnormal energy value change rates, then it is considered that the two nodes form a part of a candidate fault segment; and based on this rule, the connections between all abnormal nodes and their adjacent nodes are traversed to identify a group of connected abnormal nodes to form a preliminary fault area; here, a connected subgraph composed of all connected nodes with abnormal energy value change rates is identified through breadth-first search, and the connected subgraph is the preliminary fault area; breadth-first search starts from a node and traverses all its directly or indirectly connected abnormal nodes layer by layer.

[0094] Step S103, extracting local features from the preprocessed current signal data in the preliminary fault area to obtain local features, dynamically adjusting weights according to the correlation between the local features and the fault, and forming a dynamically adjusted multidimensional feature vector. The specific steps are as follows:

[0095] The local features are extracted from the preprocessed current signal data in the preliminary fault area to obtain multiple local features. The correlation analysis between the multiple local features and the fault is performed to calculate the dynamic adjustment weight of the correlation corresponding to each local feature. The calculation formula is:

[0096]

[0097] In the formula, MI(F i , L) represents the mutual information value between the i-th local feature Fi and the fault L, which is used to measure the i-th local feature F i Correlation with fault L; W fi Indicates the dynamic adjustment weight of the correlation corresponding to the i-th local feature.

[0098] The weights are dynamically adjusted according to the correlation corresponding to each local feature, and the weights w after dynamic adjustment of each local feature are calculated. i , the formula is:

[0099]

[0100] In the formula, w i represents the weight of the i-th local feature after dynamic adjustment, represents the dynamic adjustment weight of the correlation corresponding to the i-th local feature, and Z represents the total number of local features;

[0101] The local features and the dynamically adjusted weights of the features are weighted to obtain multiple weighted features, which are expressed as:

[0102] F i ′=w i ·F i

[0103] In the formula, F i ′ represents the i-th weighted feature after weighting, F i represents the original i-th local feature, w i is the dynamic adjustment weight of the i-th local feature.

[0104] All weighted features are combined to form a dynamically adjusted multidimensional feature vector, which is expressed as:

[0105] F adjusted =[F1′, F2′, … F Z ′]

[0106] In the formula, F i ′ represents the i-th weighted feature after weighting, F adjusted Represents a dynamically adjusted multidimensional feature vector.

[0107] It should be noted that local features include time domain features, frequency domain features and energy features.

[0108] Time domain feature extraction, statistical analysis is performed on the current signal of the preliminary fault area, and the mean, variance and peak factor of the signal are extracted as time domain features. The time domain features are used to characterize the amplitude fluctuation characteristics of the signal.

[0109] Frequency domain feature extraction, converting the current signal from the time domain to the frequency domain through fast Fourier transform, extracting the main frequency amplitude, harmonic content and frequency center as frequency domain features, and the frequency domain features are used to analyze the frequency component distribution of the signal.

[0110] Energy feature extraction calculates the energy of the current signal in the time domain and frequency domain, and obtains the total signal energy and the proportion of high-frequency energy as energy features. The energy features are used to reflect the distribution and changes of signal energy.

[0111] Step S104: performing standardization processing on the dynamically adjusted multidimensional feature vector to obtain the standardized multidimensional feature vector, and performing dimensionality reduction on the standardized multidimensional feature vector using principal component analysis to obtain the reduced dimensionality multidimensional feature vector. The specific steps are:

[0112] The dynamically adjusted multidimensional feature vector is standardized to eliminate the dimension and range differences of the feature vector, and all standardized feature vectors are combined to form a standardized multidimensional feature vector.

[0113] It should be noted that the standardization process uses z-score standardization. According to this standardization step, all features are ensured to be in the same range and provide a benchmark for subsequent dimensionality reduction processing to avoid specific features from having too much impact on the analysis results due to dimensional differences.

[0114] It should be further explained that the standardized feature vectors are combined to form a new multi-dimensional feature vector, which contains the standardized feature vectors of each node in each dimension and can fully reflect the state characteristics of the node. The standardized multi-dimensional feature vector is expressed as:

[0115] X standard =[Z1,Z2,…,Z n ]

[0116] Where Z1, Z2, …, Z n They are the standardized feature vectors, n is the total number of features, X standard Represents the normalized feature vector.

[0117] After obtaining the standardized eigenvectors, the covariance matrix is ​​calculated to provide input for principal component analysis.

[0118] The principal component analysis is performed on the standardized multidimensional feature vector, the main components are selected through eigenvalue decomposition, and the dimension reduction matrix is ​​generated.

[0119] It should be noted that the principal component analysis method is used to perform eigenvalue decomposition on the covariance matrix to obtain the decomposed eigenvalues ​​and eigenvectors. The eigenvalues ​​reflect the variance contribution of each principal component, and the eigenvectors represent the direction of the corresponding principal component. The main components whose cumulative contribution rate of eigenvalues ​​reaches 90% to 95% are selected to construct the dimensionality reduction matrix P.

[0120] The dimension reduction matrix is ​​used to perform linear transformation on the standardized multidimensional feature vector to generate a reduced-dimensional multidimensional feature vector.

[0121] It should be noted that the dimensionality reduction matrix is ​​used to perform linear transformation on the standardized multidimensional feature vector to generate a multidimensional feature vector after dimensionality reduction, specifically:

[0122] X reduced =P T X standard

[0123] Where P is the dimension reduction matrix, which is constructed by selecting the main components from the eigenvectors of the covariance matrix; X standard represents the standardized multidimensional feature vector, X reduced Represents the multidimensional feature vector after dimensionality reduction.

[0124] Step S105, the deep neural network includes an input layer, a hidden layer and an output layer; the input layer inputs the multi-dimensional feature vector after dimensionality reduction; the hidden layer realizes the fusion and learning between multi-dimensional features through a nonlinear activation function; the output layer simultaneously completes the output of fault type classification and preliminary positioning results, and the specific steps are as follows:

[0125] The multi-dimensional feature vector obtained after dimension reduction in step S104 is used as input data of the deep neural network, and the input data is in the form of:

[0126] X reduced =[X1, X2, …X k ]

[0127] Where, X reduced Represents the multi-dimensional feature vector after dimensionality reduction of the input, X1,X2,…,X k Represents the k-dimensional feature vector obtained after dimensionality reduction, where k is the dimension of the feature vector after dimensionality reduction.

[0128] It should be noted that the input layer receives the multi-dimensional feature vector after dimensionality reduction, and these input data are passed to the hidden layer through the neural network. In the input layer, each node represents a feature value, and these feature values ​​are passed to the hidden layer of the neural network; the hidden layer fuses the multi-dimensional features of the input through a nonlinear activation function, and the activation function uses ReLU. Through multi-layer hidden layer processing, the neural network can learn the complex nonlinear relationship between the input features; and through the activation function, the neural network can capture the correlation between different features and optimize its internal representation, making subsequent classification and regression tasks more accurate; the output layer is responsible for two tasks. One is to use the soft max function classification method for multi-category classification and output the fault category; the second is to output the fault location, that is, the identification of the fault section. This part of the task is a regression task, and the output is a continuous value, indicating the specific location or relative position of the fault.

[0129] It should be further explained that the deep neural network is trained using labeled data, where the labels include classification labels of fault types and regression labels of fault locations; the training data set contains multiple samples, each of which consists of an input feature vector and corresponding fault type and fault location labels.

[0130] After training is completed, the neural network outputs the prediction result of the fault type based on the input dimensionality reduction feature vector, and obtains the specific category of the fault through the output of the classification layer; at the same time, the neural network outputs the preliminary positioning result of the fault through the regression task, that is, the fault segment. This result is represented as a specific segment identifier for subsequent fault processing.

[0131] Step S106: According to the preliminary positioning result, a distributed optimization algorithm is used to analyze the preliminary fault area, and the fault location task is divided into tasks corresponding to multiple sub-areas according to the topological structure of the distribution network, and the tasks corresponding to each sub-area are assigned to distributed nodes. Each distributed node independently performs the optimization calculation of the node current signal data in the sub-area according to the tasks corresponding to each sub-area, and obtains the target value corresponding to each sub-area. The specific calculation formula is:

[0132]

[0133] Where, T k represents the target value corresponding to the kth sub-region, n k represents the total number of nodes in the kth sub-region, and They represent the actual current value and predicted current value of the i-th node respectively.

[0134] It should be noted that according to the preliminary positioning results, when using the distributed optimization algorithm, the basis for dividing the fault area into sub-areas is the topological structure of the distribution network, the physical location of the node and the current signal characteristics, including:

[0135] According to the electrical connection relationship of the distribution network, the nodes in the area are divided into sub-areas to reduce the cross-regional computing burden; according to the distribution characteristics of the nodes in the fault area, the nodes close to the fault center are divided into separate sub-areas to improve the efficiency of distributed optimization; if the current signals of some nodes fluctuate violently, they are classified as independent sub-areas to more accurately analyze the impact of independent sub-areas on fault location.

[0136] It should be noted that in step S105, the prediction results and preliminary positioning results of the fault type are obtained through the deep neural network; according to the preliminary positioning results, the analysis task of the fault section is further decomposed into tasks corresponding to multiple sub-areas; the initial fault area of ​​the distribution network is divided into multiple sub-areas; each sub-area corresponds to a distributed node for further optimization calculation; the division of each sub-area is based on the spatial distribution of the fault area and the relative position of the fault point; if the preliminary positioning result shows that the fault is located in the central area of ​​a specific section, it is divided into multiple small areas according to the characteristics of the area to make the distributed computing more efficient.

[0137] Step S107: Calculate the weight of each sub-area, perform weighted average on the target values ​​corresponding to all sub-areas, and generate a global final fault location result. The specific steps are as follows:

[0138] The weight of each sub-region is calculated based on the target value corresponding to each sub-region. The calculation formula is:

[0139]

[0140] Where W k represents the weight of the kth sub-region, K represents the total number of sub-regions, T j represents the target value of the jth sub-region, T k Represents the target value of the kth sub-region.

[0141] The target values ​​and weights corresponding to all sub-areas are aggregated, and the final global fault location result is generated by weighted average calculation. The calculation formula is:

[0142]

[0143] Where, T k represents the target value corresponding to the kth sub-region, R represents the final fault location result, K represents the total number of sub-regions, and W kIndicates the weight of the kth sub-area. The final fault location result is obtained based on the calculation. k indicates the sub-area number currently being calculated.

[0144] Step S108: Match the fault section in the distribution network topology according to the final fault location result. The specific steps are as follows:

[0145] The final fault location result is compared with the target value T corresponding to each area in the distribution network topology. i Perform comparison calculation, the comparison calculation formula is:

[0146]

[0147] Where M i represents the matching degree between the final fault location result and the ith region, T i represents the characteristic target value of the i-th region, which is obtained by counting the node data in the region; σ i Represents the standard deviation of the target value of the ith region, which is used to reflect the consistency of the nodes in the region.

[0148] Sort all calculated matching degrees and select the area with the smallest matching degree. The calculation formula is:

[0149] i min =argmin i∈[1，N] M i

[0150] In the formula, i min Indicates the area with the smallest matching degree, M i represents the matching degree between the final fault location result and the i-th region, min represents the minimum operator, and i∈[1,N] represents that i is the index in the set of integers from 1 to N.

[0151] The area i with the smallest matching degree min Determine the faulty section and record the corresponding section number, node number and location coordinates of the faulty section.

[0152] So far, a method for identifying fault sections of active distribution networks based on characteristic current signal energy has been completed.

[0153] The above is only a specific implementation of the present invention, but the protection scope of the embodiments of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the embodiments of the present invention, which should be included in the protection scope of the embodiments of the present invention. Therefore, the protection scope of the embodiments of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for identifying fault sections of an active distribution network based on characteristic current signal energy, characterized in that: The method comprises the following steps: Obtaining the original current signal data of each node in the distribution network and preprocessing the original current signal data; According to the preprocessed current signal data, the global energy analysis method is used to calculate the energy value of each node, identify the area composed of multiple nodes, connect the nodes with abnormal energy value change rate in the area, and obtain the preliminary fault area; Extract local features from the preprocessed current signal data in the preliminary fault area to obtain local features, dynamically adjust weights according to the correlation between the local features and the fault, and form a dynamically adjusted multi-dimensional feature vector; Performing standardization processing on the dynamically adjusted multidimensional feature vector to obtain a standardized multidimensional feature vector, and performing dimensionality reduction on the standardized multidimensional feature vector using principal component analysis to obtain a reduced dimensional multidimensional feature vector; A deep neural network is used to fuse and learn the multi-dimensional feature vectors after dimensionality reduction, and the fault type and preliminary location results are output; According to the preliminary positioning results, a distributed optimization algorithm is used to analyze the preliminary fault area, and the fault location task is divided into tasks corresponding to multiple sub-areas according to the topological structure of the distribution network. The tasks corresponding to each sub-area are assigned to distributed nodes. Each distributed node independently performs the optimization calculation of the node current signal data in the sub-area according to the tasks corresponding to each sub-area, and obtains the target value corresponding to each sub-area. Calculate the weight of each sub-area, take the weighted average of the target values ​​corresponding to all sub-areas, and generate the final global fault location result; According to the final fault location result, the fault section in the distribution network topology is matched.

2. The method for identifying fault sections of an active distribution network based on characteristic current signal energy according to claim 1, characterized in that: Obtain the original current signal data of each node in the distribution network and preprocess the original current signal data. The specific steps are as follows: The original current signal data of each node is obtained based on the current sensor preset at the distribution network node; The high-frequency noise caused by external environmental interference and sensor error is removed from the original current signal data of each node; The denoised current signal data is normalized, and the normalized current signal data is segmented according to a time window to obtain preprocessed current signal data.

3. The method for identifying fault sections of an active distribution network based on characteristic current signal energy according to claim 2, characterized in that: According to the preprocessed current signal data, the global energy analysis method is used to calculate the energy value of each node, identify the area composed of multiple nodes, connect the nodes with abnormal energy value change rate in the area, and obtain the preliminary fault area. The specific steps are as follows: The global energy analysis method is used to perform time integration on the preprocessed current signal data of each node, and the energy value of the current signal data of each node within the time window is calculated; Based on the energy values ​​of multiple continuous time windows, the energy value change rate of each node is calculated; Determine whether the energy value change rate of each node in the area meets the abnormal judgment condition; If the abnormality judgment condition is met, it means that the node is a node with abnormal energy value change rate; Based on the topological structure of the distribution network, the area connected by multiple nodes with abnormal energy value change rates is identified as the preliminary fault area.

4. The method for identifying fault sections of an active distribution network based on characteristic current signal energy according to claim 1, characterized in that: The local features are extracted from the preprocessed current signal data in the preliminary fault area to obtain the local features. The weights are dynamically adjusted according to the correlation between the local features and the faults to form a dynamically adjusted multidimensional feature vector. The specific steps are as follows: The local features are extracted from the preprocessed current signal data in the preliminary fault area to obtain multiple local features. The correlation analysis between the multiple local features and the fault is performed to calculate the dynamic adjustment weight of the correlation corresponding to each local feature. The calculation formula is: In the formula, MI(F i , L) represents the mutual information value between the i-th local feature Fi and the fault L, which is used to measure the i-th local feature F i Correlation with fault L; W fi Indicates the dynamic adjustment weight of the correlation corresponding to the i-th local feature; The weights are dynamically adjusted according to the correlation corresponding to each local feature, and the weights w after dynamic adjustment of each local feature are calculated. i , the formula is: In the formula, w i represents the weight of the i-th local feature after dynamic adjustment, represents the dynamic adjustment weight of the correlation corresponding to the i-th local feature, and Z represents the total number of local features; The local features and the dynamically adjusted weights of the features are weighted to obtain multiple weighted features, which are expressed as: F i ′=w i ·F i In the formula, F i ′ represents the i-th weighted feature after weighting, F i represents the original i-th local feature, w i is the dynamic adjustment weight of the i-th local feature; All weighted features are combined to form a dynamically adjusted multidimensional feature vector, which is expressed as: F adjusted =[F1′,F2′,…F Z ′] In the formula, F i ′ represents the i-th weighted feature after weighting, F adjusted Represents a dynamically adjusted multidimensional feature vector.

5. The method for identifying fault sections of active distribution networks based on characteristic current signal energy according to claim 1, characterized in that: The dynamically adjusted multidimensional feature vector is standardized to obtain the standardized multidimensional feature vector, and the principal component analysis is used to reduce the dimension of the standardized multidimensional feature vector to obtain the reduced multidimensional feature vector. The specific steps are as follows: Standardizing the dynamically adjusted multidimensional feature vectors to eliminate dimension and range differences of the feature vectors, and combining all standardized feature vectors to form a standardized multidimensional feature vector; Perform principal component analysis on the standardized multidimensional eigenvectors, select the main components through eigenvalue decomposition, and generate a dimensionality reduction matrix; The dimension reduction matrix is ​​used to perform linear transformation on the standardized multidimensional feature vector to generate a multidimensional feature vector after dimension reduction.

6. The method for identifying fault sections of active distribution networks based on characteristic current signal energy according to claim 1, characterized in that: The deep neural network consists of an input layer, a hidden layer, and an output layer. The input layer inputs the multidimensional feature vector after dimensionality reduction. The hidden layer realizes the fusion and learning of multidimensional features through nonlinear activation functions. The output layer completes the fault type classification and outputs the preliminary positioning results.

7. The method for identifying fault sections of an active distribution network based on characteristic current signal energy according to claim 1, characterized in that: Each distributed node independently performs the optimization calculation of the node current signal data in the sub-region according to the tasks corresponding to each sub-region, and obtains the target value corresponding to each sub-region. The specific calculation formula is: Where, T k represents the target value corresponding to the kth sub-region, n k represents the total number of nodes in the kth sub-region, and They represent the actual current value and predicted current value of the i-th node respectively.

8. The method for identifying fault sections of an active distribution network based on characteristic current signal energy according to claim 7, characterized in that: Calculate the weight of each sub-area, take the weighted average of the target values ​​corresponding to all sub-areas, and generate the final global fault location result. The specific steps are as follows: The weight of each sub-region is calculated based on the target value corresponding to each sub-region. The calculation formula is: Where W k represents the weight of the kth sub-region, K represents the total number of sub-regions, T j represents the target value of the jth sub-region, T k represents the target value of the kth sub-region; The target values ​​and weights corresponding to all sub-areas are aggregated, and the final global fault location result is generated by weighted average calculation. The calculation formula is: Where, T k represents the target value corresponding to the kth sub-region, R represents the final fault location result, K represents the total number of sub-regions, and W k Indicates the weight of the kth sub-area. The final fault location result is obtained based on the calculation. k indicates the sub-area number currently being calculated.

9. The method for identifying fault sections of an active distribution network based on characteristic current signal energy according to claim 8, characterized in that: According to the final fault location result, match the fault section in the distribution network topology. The specific steps are as follows: The final fault location result is compared with the target value T corresponding to each area in the distribution network topology. i Perform comparison calculation, the comparison calculation formula is: Where M i represents the matching degree between the final fault location result and the ith region, T i represents the characteristic target value of the i-th region, which is obtained by counting the node data in the region; σ i represents the standard deviation of the target value of the i-th region, which is used to reflect the consistency of the nodes in the region; Sort all calculated matching degrees and select the area with the smallest matching degree. The calculation formula is: i min =argmin i∈ [1,N]M i In the formula, i min Indicates the area with the smallest matching degree, M i represents the matching degree between the final fault location result and the i-th region, min represents the minimum operator, i∈[1,N] represents that i is the index in the set of integers from 1 to N; The area i with the smallest matching degree min Determine the faulty section and record the corresponding section number, node number and location coordinates of the faulty section.

Images