A distribution network fault research and judgment method and system based on multi-dimensional data fusion

Abstract

The application discloses a distribution network fault research and judgment method and system based on multi-dimensional data fusion, and relates to the technical field of distribution network fault diagnosis. The method comprises the following steps: collecting electrical quantities, switch state quantities, environmental parameters and repair text from a plurality of information sources of a distribution network, performing space-time synchronization and topological correlation, calculating the space-time coupling strength value of each information source as the field strength weight of the node, and generating a multi-dimensional fusion data field; extracting the fault energy fingerprint of each information source and calculating the deviation degree, and screening an abnormal information source subset; mapping the abnormal information source to a topological node, calculating the convergence density value, and determining the fault disturbance center node; calculating the attenuation gradient value along the topological branch, comparing with a preset fault type characteristic template, determining the fault section and the fault type; outputting the research and judgment result, correlatively storing the characteristic data, updating the normal operation benchmark and the characteristic template, and completing the closed loop processing. The application can effectively fuse multi-source information, accurately locate the fault section and identify the fault type.

Description

Technical Field

[0001] This invention relates to the field of distribution network fault diagnosis and assessment technology, and in particular to a distribution network fault assessment method and system based on multi-dimensional data fusion. Background Technology

[0002] As a critical link in the power system that directly faces users, the reliability of the distribution network directly affects the normal electricity consumption of social production and residents' lives. With the continuous expansion of the distribution network, the large-scale integration of distributed power sources, and the continuous improvement of users' requirements for power quality, the operating environment of the distribution network is becoming increasingly complex, and the types of faults are becoming more diversified, which places higher demands on the rapid and accurate assessment of faults.

[0003] Currently, fault diagnosis in distribution networks mainly relies on a combination of alarms triggered by excessive electrical quantity thresholds reported by feeder terminal devices and manual line inspection. However, traditional methods have the following shortcomings: First, relying on a single electrical quantity threshold is prone to false alarms when faced with fluctuations in line parameters caused by changes in ambient temperature. Second, the integration of distributed power sources in the distribution network alters the amplitude and direction of fault currents, reducing the accuracy of traditional fault location methods based on fixed thresholds. Third, existing methods lack effective integration and utilization of multi-source heterogeneous information; electrical quantities, switch status, environmental parameters, and user repair reports are analyzed independently, failing to form a collaborative judgment capability. Fourth, the lack of an adaptive closed-loop optimization mechanism makes it difficult to continuously improve the accuracy of fault judgment with the accumulation of operational data.

[0004] Therefore, there is an urgent need for a fault assessment method that can fully integrate multi-source information from the distribution network, adaptively track system state changes, and possess closed-loop optimization capabilities. Summary of the Invention

[0005] This invention proposes a method for analyzing distribution network faults based on multi-dimensional data fusion, comprising: S1. Collect electrical quantities, switch status quantities, environmental parameters and repair reports from multiple information sources in the distribution network, aggregate them into multi-dimensional raw data, perform spatiotemporal synchronization and topology association on the multi-dimensional raw data, calculate the spatiotemporal coupling strength value of each information source on the topology node, assign the spatiotemporal coupling strength value as the field strength weight to the corresponding node, and generate a multi-dimensional fused data field. S2. Based on the multi-dimensional fusion data field, extract the fault energy fingerprint of each information source under the influence of field strength weight, calculate the deviation of the fault energy fingerprint from the normal operation benchmark, and select the subset of abnormal information sources based on the deviation of the fault energy fingerprint. S3. Map the subset of abnormal information sources to the topological nodes of the multidimensional fused data field, calculate the convergence density value of the fault energy fingerprint deviation on each node, and determine the fault disturbance center node based on the peak position of the convergence density value. S4. Taking the fault disturbance center node as the origin, calculate the attenuation gradient value of the fault energy fingerprint deviation along the topological branch of the multidimensional fused data field. In the calculation of the attenuation gradient value, a three-phase unbalance amplification coefficient is introduced to enhance the gradient response of asymmetric faults. The attenuation gradient value is compared with the preset fault type feature template in terms of spatial attenuation mode, and the initial judgment result is verified by the real-time electrical quantity feature quantity in the fault section to determine the fault section and fault type. S5. The system outputs the fault section and fault type, and stores the attenuation gradient value in association with the fault section and fault type in the multi-dimensional fusion data field. Based on the association storage results, the normal operation benchmark and the preset fault type feature template are updated to complete the closed-loop processing of the distribution network fault judgment method.

[0006] The above-described method for distribution network fault assessment based on multidimensional data fusion involves collecting electrical quantities, switch status quantities, environmental parameters, and repair reports from multiple information sources in the distribution network, aggregating them into multidimensional raw data, performing spatiotemporal synchronization and topology association on the multidimensional raw data, calculating the spatiotemporal coupling strength value of each information source at the topology nodes, and assigning the spatiotemporal coupling strength value as a field strength weight to the corresponding nodes to generate a multidimensional fused data field. The method includes the following sub-steps: Electrical quantities, switch status quantities, environmental parameters, and repair reports are collected through voltage transformers, current transformers, distribution automation terminals, meteorological monitoring stations, and the 95598 customer service system. The data is then time-series aligned according to a unified timestamp and aggregated into multi-dimensional raw data. Based on the geographic information system of the power distribution network, an association matrix is ​​established between each topology node, and the time-aligned information source data is associated with the corresponding topology node to complete the spatiotemporal synchronization and topology association. The coupling coefficients between different information source data on each topology node are calculated, and the spatiotemporal coupling strength value is obtained by weighted summation. The spatiotemporal coupling strength value is used as the field strength weight and assigned to the corresponding node to construct a multidimensional fusion data field covering the distribution network topology.

[0007] The above-described method for analyzing distribution network faults based on multi-dimensional data fusion includes the following sub-steps: extracting fault energy fingerprints from each information source under the influence of field strength weights based on a multi-dimensional fused data field; calculating the deviation of the fault energy fingerprints from the normal operating baseline; and selecting a subset of abnormal information sources based on the fault energy fingerprint deviation. The data from each information source in the multidimensional fusion data field are weighted by field strength weight, and the fundamental amplitude and harmonic distortion rate of electrical quantities, the operation frequency of switch status quantities, the temperature and humidity change rate of environmental parameters, and the semantic coding value of the repair text are extracted to construct a fault energy fingerprint. The historical average values ​​of various fault energy fingerprints of each topology node under normal operating conditions are obtained as the normal operating benchmark. The Mahalanobis distance between the current fault energy fingerprint and the normal operating benchmark is calculated as the fault energy fingerprint deviation. The deviation of the fault energy fingerprint of each information source is compared with the preset deviation threshold, and information sources with a deviation greater than the preset deviation threshold are selected and classified into the abnormal information source subset.

[0008] The above-described method for analyzing distribution network faults based on multi-dimensional data fusion includes the following sub-steps: mapping a subset of abnormal information sources to topological nodes of the multi-dimensional fused data field, calculating the convergence density value of the fault energy fingerprint deviation on each node, and determining the fault disturbance center node based on the peak position of the convergence density value. A subset of abnormal information sources is mapped to the corresponding topological nodes of the multidimensional fused data field, and the number of abnormal information sources and the deviation value of fault energy fingerprint on each node are counted. Taking each topology node as the center, extend outward along the topology branch by a switch interval as the preset spatial neighborhood radius, and count the sum of the fault energy fingerprint deviation of all abnormal information sources within this neighborhood radius as the aggregation density value of the node. Iterate through the aggregation density values ​​of all topology nodes and select the topology node with the largest aggregation density value as the fault disturbance center node.

[0009] The above-described method for distribution network fault assessment based on multi-dimensional data fusion includes the following steps: Taking the fault disturbance center node as the origin, the attenuation gradient value of the fault energy fingerprint deviation is calculated along the topological branch of the multi-dimensional fused data field. A three-phase imbalance amplification coefficient is introduced in the attenuation gradient value calculation to enhance the gradient response of asymmetric faults. The attenuation gradient value is compared with a preset fault type feature template using spatial attenuation patterns, and the initial judgment result is verified using real-time electrical quantity characteristics within the fault section to determine the fault section and fault type. Taking the fault disturbance center node as the origin, the adjacent nodes are searched outward level by level along each topological branch direction. The ratio of the difference in fault energy fingerprint deviation between adjacent level nodes to the line impedance value is calculated. The three-phase unbalance amplification factor is introduced to correct the calculation results, and the attenuation gradient value in the branch direction is obtained to construct the attenuation gradient field. The correlation coefficient is calculated between the attenuation gradient value sequence of each branch direction and the standard attenuation curve in the preset fault type feature template, and the fault type with the highest correlation coefficient is selected as the fault type to be judged. Along the topological branch where the attenuation gradient value changes most drastically, the continuous node interval where the attenuation gradient value exceeds the preset gradient threshold is identified as the fault section.

[0010] The distribution network fault assessment method based on multi-dimensional data fusion, as described above, involves the system outputting the fault section and fault type, and storing the attenuation gradient value in a multi-dimensional fused data field in association with the fault section and fault type. Based on the association storage results, the system updates the normal operation benchmark and the preset fault type feature template, completing the closed-loop processing of the distribution network fault assessment method. This includes the following sub-steps: The fault section, fault type and attenuation gradient value sequence are used to generate a fault record data packet, which is then output to the distribution automation master station. The fault record data packets are associated and stored under the corresponding fault type category in the fault sample library; When the number of samples of the same fault type reaches the preset sample number threshold, the statistical mean curve of the decay gradient value sequence under the fault type is recalculated, the standard decay curve in the preset fault type feature template is updated, and the mean value of the fault energy fingerprint within the preset time window before the occurrence of this fault is used as the new normal operation benchmark sample and updated by moving average.

[0011] The above-described method for distribution network fault assessment based on multi-dimensional data fusion includes the following sub-steps: The correlation coefficient is calculated between the attenuation gradient value sequence of each branch direction and the standard attenuation curve in the preset fault type feature template. The fault type with the highest correlation coefficient is selected as the fault type for assessment. Acquire real-time electrical quantity data of each topology node in the fault section, calculate the actual fault current and fault voltage characteristics, and compare and verify them with the standard fault electrical characteristics corresponding to the fault type. If the comparison and verification results are consistent, the determined fault type is confirmed as the final fault type; if they are inconsistent, the fault type with the second highest correlation coefficient is selected as the corrected fault type, and the comparison and verification are repeated until the final fault type is confirmed. The final confirmed fault type will be associated with the faulty section and output.

[0012] This invention also proposes a distribution network fault assessment system based on multi-dimensional data fusion, comprising: Data acquisition and fusion module: Collects electrical quantities, switch status quantities, environmental parameters and repair reports from multiple information sources in the distribution network, aggregates them into multi-dimensional raw data, performs spatiotemporal synchronization and topology association on the multi-dimensional raw data, calculates the spatiotemporal coupling strength value of each information source on the topology node, assigns the spatiotemporal coupling strength value as the field strength weight to the corresponding node, and generates a multi-dimensional fused data field. Fault Feature Extraction Module: Based on a multi-dimensional fused data field, extract the fault energy fingerprint of each information source under the influence of field strength weight, calculate the deviation of the fault energy fingerprint from the normal operation benchmark, and filter out a subset of abnormal information sources based on the fault energy fingerprint deviation. Fault disturbance localization module: Maps a subset of abnormal information sources to the topological nodes of the multidimensional fused data field, calculates the convergence density value of the fault energy fingerprint deviation on each node, and determines the fault disturbance center node based on the peak position of the convergence density value; Fault analysis module: Taking the fault disturbance center node as the origin, the attenuation gradient value of the fault energy fingerprint deviation is calculated along the topological branch of the multidimensional fused data field. The attenuation gradient value calculation introduces the three-phase unbalance amplification coefficient to enhance the gradient response of asymmetric faults. The attenuation gradient value is compared with the preset fault type feature template for spatial attenuation mode, and the initial judgment result is verified by the real-time electrical quantity feature quantity in the fault section to determine the fault section and fault type. Output and Closed-Loop Update Module: The system outputs the fault section and fault type, and associates the attenuation gradient value with the fault section and fault type to the multi-dimensional fusion data field. Based on the associated storage results, it updates the normal operation benchmark and the preset fault type feature template to complete the closed-loop processing of the distribution network fault judgment method.

[0013] The beneficial effects achieved by this invention are as follows: By fusing information from multiple sources, including voltage transformers, current transformers, distribution automation terminals, meteorological monitoring stations, and the 95598 customer service system, a multi-dimensional fused data field is constructed. This achieves spatiotemporal coupling and topological association of electrical quantities, switch status, environmental parameters, and user repair information, overcoming the limitations of fault analysis from a single information source and improving the comprehensiveness and reliability of fault detection. Based on field strength weighting, fault energy fingerprints are extracted and deviations are calculated. The fault disturbance center is located by converging density peaks. The attenuation gradient value is calculated along the topological branch and compared with the spatial attenuation pattern of the preset fault type feature template to achieve accurate judgment of fault sections and fault types, effectively reducing false alarm and false alarm rates. The judgment results are associated and stored in a fault sample library. The normal operation benchmark and fault type feature template are updated based on sample accumulation, forming a closed-loop self-learning mechanism. This allows the system to continuously optimize with the operating experience of the distribution network, improving its adaptability to complex operating conditions and new faults. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0015] Figure 1 This is a flowchart of a distribution network fault assessment method based on multi-dimensional data fusion provided in an embodiment of this application.

[0016] Figure 2 This is a schematic diagram of a distribution network fault assessment system based on multi-dimensional data fusion provided in an embodiment of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1

[0019] like Figure 1 As shown in the figure, an embodiment of this application provides a method for analyzing distribution network faults based on multi-dimensional data fusion, comprising: Step S1: Collect electrical quantities, switch status quantities, environmental parameters and repair reports from multiple information sources in the distribution network, aggregate them into multi-dimensional raw data, perform spatiotemporal synchronization and topology association on the multi-dimensional raw data, calculate the spatiotemporal coupling strength value of each information source on the topology node, assign the spatiotemporal coupling strength value as the field strength weight to the corresponding node, and generate a multi-dimensional fused data field. Specifically, electrical quantities, switch status quantities, environmental parameters, and repair reports are collected using voltage transformers, current transformers, distribution automation terminals, meteorological monitoring stations, and the 95598 customer service system. These data are then aggregated into multidimensional raw data after being aligned according to a unified timestamp sequence. Based on the topology node association matrix established by the distribution network geographic information system, the data from each information source is associated with the corresponding node, achieving spatiotemporal synchronization and topology association. The coupling coefficients between different information source data on each topology node are calculated and weighted to obtain the spatiotemporal coupling strength value, which is then assigned as the field strength weight to the corresponding node, constructing a multidimensional fused data field covering the distribution network topology. This process includes the following sub-steps: Step S11: Collect electrical quantities, switch status quantities, environmental parameters and repair reports through voltage transformers, current transformers, distribution automation terminals, meteorological monitoring stations and the 95598 customer service system, align them according to a unified timestamp, and aggregate them into multi-dimensional raw data. Real-time data on three-phase voltage RMS, three-phase current RMS, active power, and reactive power are collected via voltage and current transformers deployed at the terminals of each feeder in the distribution network. Circuit breaker and disconnector status signals, as well as protection device operation signals, are collected through distribution automation terminals to generate switch status data. Environmental parameter data is generated by collecting ambient temperature, relative humidity, wind speed, and rainfall data from meteorological monitoring stations distributed throughout the distribution network. Repair requests are received through the 95598 customer service system, and the repair address and fault description text are extracted to generate repair request text data.

[0020] The sampling frequencies of the four data sources mentioned above differ: electrical quantity data is sampled at 128 points per second, switch status data is acquired through event-triggered acquisition, environmental parameter data is acquired periodically every 10 minutes, and repair report text data is received in real time. Using the BeiDou satellite timing signal as a unified clock reference, millisecond-level timestamps are added to the data collected from each data source. An adaptive interpolation method is used to resample non-uniformly spaced data to a unified time grid, eliminating time misalignment caused by differences in sampling frequency. Data from various data points within the same time segment are then horizontally stitched together according to equipment identification to form multidimensional raw data.

[0021] Step S12: Establish an association matrix between each topology node based on the distribution network geographic information system, associate the time-aligned information source data with the corresponding topology node, and complete the spatiotemporal synchronization and topology association. Based on the network model provided by the distribution network geographic information system, the topological connections of all feeders, switching stations, distribution transformers, and tie switches are extracted, and an adjacency matrix of node connectivity is established. The spatial coordinates of each measuring device in the geographic information system are matched with the coordinates of the topological nodes, mapping each measuring point to its corresponding topological node number. For cases where multiple measuring sources are connected to the same topological node, master-slave weights are assigned according to the accuracy level of each measuring device, with higher accuracy levels receiving greater weights. The time-aligned multidimensional raw data from step S11 is then assigned to the corresponding nodes according to the mapping relationship between device identifiers and topological node numbers, forming a structured dataset indexed by topological nodes.

[0022] Step S13: Calculate the coupling coefficient between different information source data on each topology node, sum the weighted values ​​to obtain the spatiotemporal coupling strength value, assign the spatiotemporal coupling strength value as the field strength weight to the corresponding node, and construct a multi-dimensional fusion data field covering the distribution network topology. For each topology node, continuous sampling sequences of all information sources on that node within the same time window are extracted. For any two information source sampling sequences, the transmission delay between them is estimated using a cross-correlation function. Based on the estimated delay, one sequence is compensated for to eliminate misalignment caused by differences in data transmission paths. After delay compensation, the Pearson correlation coefficient and normalized mutual information value of the two information source sequences are calculated, and their product is used as the coupling coefficient of the combination. The coupling coefficients of all information source combinations on that node are multiplied by a preset information source type importance weighting factor and then summed to obtain the spatiotemporal coupling strength value of that node. The spatiotemporal coupling strength value of the node is expressed by the following formula:

[0023] in, This represents the spatiotemporal coupling strength value; Indicates mapping to nodes The total number of valid information sources; This represents the index number of any two different information sources on a node; This represents the prior importance weighting factor for information source type; This represents the aligned sequence after cross-correlation delay compensation; This represents the Fast Fourier Transform operator. express The conjugate of the complex number; the denominator uses the product of magnitudes. This is to avoid the problem of an excessively small denominator caused by phase reversal; This represents the lower and upper limits of the effective frequency band integration. This represents the bandwidth normalization coefficient; The exponential compression factor, representing the frequency domain coherence contribution, is set to 0.8. The mutual information of the two sequences after time delay compensation is calculated by estimating the joint probability distribution using the two-dimensional histogram method. These represent the marginal information entropy of the two sequences, respectively.

[0024] The spatiotemporal coupling strength value is assigned as the field strength weight to the current topology node. Based on the electrical distance between nodes, the inverse distance weighted interpolation method is used to assign field strength weights to virtual nodes that are not directly connected to measurement equipment, thereby generating a continuously distributed multidimensional fusion data field covering the entire distribution network topology.

[0025] Step S2: Based on the multi-dimensional fusion data field, extract the fault energy fingerprint of each information source under the field strength weight, calculate the fault energy fingerprint deviation from the normal operation benchmark, and select the abnormal information source subset based on the fault energy fingerprint deviation. Specifically, the data from each information source in the multi-dimensional fused data field are weighted using field strength weights to extract the fundamental amplitude and harmonic distortion rate of electrical quantities, the frequency of operation of switch status quantities, the temperature and humidity change rate of environmental parameters, and the semantic encoding value of the repair report text, thus constructing a fault energy fingerprint. The historical average values ​​of electrical fault energy fingerprints, switch fault energy fingerprints, environmental fault energy fingerprints, and repair report fault energy fingerprints for each topology node under normal operating conditions are obtained and combined to form a normal operating benchmark. The Mahalanobis distance between the fault energy fingerprint of each information source at the current moment and the normal operating benchmark is calculated, and this Mahalanobis distance is used as the fault energy fingerprint deviation. The fault energy fingerprint deviation of each information source on each topology node is compared with a preset deviation threshold. Information sources with fault energy fingerprint deviations greater than the preset deviation threshold are selected and classified into an abnormal information source subset. This process includes the following sub-steps: Step S21: Weight the data from each information source in the multidimensional fusion data field with field strength weight, extract the fundamental amplitude and harmonic distortion rate of electrical quantities, the frequency of operation of switch status quantities, the temperature and humidity change rate of environmental parameters, and the semantic encoding value of the repair text, and construct the fault energy fingerprint. The semantic encoding value of the repair report text is generated in the following way: the repair report text is segmented into words, a pre-trained language model in the power field is used to extract semantic feature vectors, the semantic feature vectors are reduced in dimensionality, and the values ​​of the first three principal components are extracted as semantic encoding values, which correspond to the implicit features of fault phenomenon, fault location and urgency, respectively.

[0026] Based on the differences in the contribution of different information sources to fault characterization in the multidimensional fusion data field, a field strength weight is used to apply differentiated weights to the data from each information source. On the basis of weighted processing, the fundamental frequency amplitude and the distortion ratio of each harmonic relative to the fundamental frequency are extracted from the electrical measurement channel to characterize the degree of distortion of the fault current or voltage waveform. The cumulative number of opening and closing actions of circuit breakers and disconnectors within a unit time window is statistically analyzed from the switching quantity channel to reflect the response strength of the switching equipment under fault impact. The rate of change of temperature and humidity within a unit time period is calculated from the environmental monitoring channel to eliminate the slow-changing components caused by diurnal temperature differences and seasonal fluctuations, highlighting the abrupt change trend of environmental parameters at the moment of fault. The semantic encoding values ​​of fault description keywords are extracted from the repair report text channel to obtain the implicit fault type and location information. The feature quantities of the above channels after field strength weighting are concatenated in a preset dimensional order to form a fault energy fingerprint vector.

[0027] The feature quantities of each channel after field strength weighting are concatenated in a preset dimension order to form a fault energy fingerprint vector.

[0028] Step S22: Obtain the historical average value of various fault energy fingerprints of each topology node under normal operating conditions as the normal operating benchmark, and calculate the Mahalanobis distance between the current fault energy fingerprint and the normal operating benchmark as the fault energy fingerprint deviation. The historical average values ​​of various fault energy fingerprints for each topology node under normal operating conditions are obtained as the normal operating baseline. The Mahalanobis distance between the current fault energy fingerprint and the normal operating baseline is calculated as the fault energy fingerprint deviation. The fault energy fingerprint deviation is expressed by the following formula:

[0029] in, Represents a node Current fault energy fingerprint deviation; This represents the fault energy fingerprint feature vector extracted at the current moment, with dimension [missing information]. Specifically, it includes the fundamental amplitude of electrical quantities, the third harmonic distortion rate, the fifth harmonic distortion rate, the frequency of switch operation, the rate of temperature rise, the rate of humidity change, and the first three principal components of the repair semantic code. This represents the baseline mean vector for normal operation, obtained by the moving average of the most recent 200 historical windows before the failure occurred; This represents a dimensionally normalized diagonal matrix, obtained by preprocessing various feature quantities using the Z-score standardization method. The diagonal elements are the reciprocals of the standard deviations of each feature quantity. The covariance matrix representing the eigenvectors under the historical normal operating benchmark is obtained by calculating the covariance of the most recent 200 historical samples one dimension at a time. This represents the ridge regression regularization coefficient, which is 1% of the average value of the main diagonal of the covariance matrix. It is used to prevent the covariance matrix from being ill-conditioned, which could lead to failure inversion. This represents the d-order identity matrix.

[0030] Step S23: Compare the fault energy fingerprint deviation of each information source with the preset deviation threshold, and filter out information sources with a deviation greater than the preset deviation threshold, and classify them into the abnormal information source subset. After independently calculating the fault energy fingerprint deviation for each information source, a differentiated deviation threshold is set for each information source. The average value of the fault energy fingerprint deviation calculated for each information source during its historical normal operation period, plus three times the standard deviation, is used as the dynamic deviation threshold for that information source. The deviation calculation result for each information source is compared numerically with its corresponding dynamic deviation threshold one by one. When the deviation of an information source exceeds its dynamic threshold, the information source is marked as abnormal, and all information sources in abnormal states are grouped into the abnormal information source subset.

[0031] Step S3: Map the subset of abnormal information sources to the topology nodes of the multidimensional fused data field, calculate the convergence density value of the fault energy fingerprint deviation on each node, and determine the fault disturbance center node based on the peak position of the convergence density value. Specifically, a subset of abnormal information sources is mapped to corresponding topological nodes in the multidimensional fused data field, and the number of abnormal information sources and the fault energy fingerprint deviation value on each node are counted. Taking each topological node as the center and extending outward along the topological branch by a switch interval as a preset spatial neighborhood radius, the sum of the fault energy fingerprint deviation values ​​of all abnormal information sources within this neighborhood radius is calculated as the aggregation density value of that node. All topological nodes are traversed, and the node with the largest aggregation density value is selected as the fault disturbance center node, specifically including the following sub-steps: Step S31: Map the subset of abnormal information sources to the corresponding topological nodes of the multidimensional fused data field, and count the number of abnormal information sources and the fault energy fingerprint deviation value on each node. Based on the topology node identifiers carried by each element in the subset of abnormal information sources, a mapping relationship is established between the information sources and the nodes of the multidimensional fused data field. For all abnormal information sources mapped to the same topology node, their number is counted one by one, and the fault energy fingerprint deviation value corresponding to each abnormal information source is read from the multidimensional fused data field and recorded as a set of deviation observation values ​​for that node.

[0032] Step S32: Taking each topology node as the center, extend outward along the topology branch by a switch interval as the preset spatial neighborhood radius, and calculate the sum of the fault energy fingerprint deviation of all abnormal information sources within the neighborhood radius as the aggregation density value of the node. For each topology node, centered on its location, trace outwards along each feeder branch directly connected to that node in the distribution topology to the next adjacent switchgear. Include the line segment covered by that switchgear and the terminal nodes under its jurisdiction within its neighborhood, forming a spatial neighborhood with the central node as the starting point and the adjacent switchgear as the boundary. Within this neighborhood, gather all anomaly information sources mapped to each topology node within the neighborhood, sum the fault energy fingerprint deviations of all these information sources, and use the sum as the aggregation density value of the central node.

[0033] Step S33: Traverse the aggregation density values ​​of all topology nodes and select the topology node with the largest aggregation density value as the fault disturbance center node. The convergence density values ​​of each node are compared sequentially according to their topology node numbers. The current maximum value and its corresponding node number are recorded. After all nodes have been traversed, the topology node corresponding to the maximum value is determined as the fault disturbance center node.

[0034] Step S4: Taking the fault disturbance center node as the origin, calculate the attenuation gradient value of the fault energy fingerprint deviation along the topological branch of the multidimensional fused data field. The three-phase unbalance amplification coefficient is introduced in the calculation of the attenuation gradient value to enhance the gradient response of the asymmetric fault. The attenuation gradient value is compared with the preset fault type feature template in terms of spatial attenuation mode, and the initial judgment result is verified by the real-time electrical quantity feature quantity in the fault section to determine the fault section and fault type. Specifically, taking the fault disturbance center node as the origin, adjacent nodes are searched level by level along the topology branches. The ratio of the difference in fault energy fingerprint deviation between adjacent level nodes to the line impedance is calculated to obtain the attenuation gradient and construct the gradient field. The correlation coefficient of each branch gradient sequence is calculated with the standard attenuation curve. The fault type with the highest correlation coefficient is selected as the judgment type. The fault current and voltage characteristics corresponding to the real-time electrical quantity data are compared and verified. If they match, it is confirmed. Otherwise, the second highest type is selected and the verification is repeated until it is confirmed. The final type is associated with the fault section. Along the branch with the most drastic gradient change, the continuous node interval with gradient value exceeding the threshold is defined as the fault section. The specific steps include the following: Step S41: Taking the fault disturbance center node as the origin, search for adjacent nodes outward step by step along each topological branch direction, calculate the ratio of the difference in fault energy fingerprint deviation between adjacent level nodes to the line impedance value, introduce the three-phase unbalance amplification factor to correct the calculation results, obtain the attenuation gradient value in the branch direction, and construct the attenuation gradient field. The search is expanded outwards level by level along the distribution network topology, with the fault disturbance center node as the origin. At each search level, all nodes to be searched are identified, and their line impedance parameters relative to their parent node at the next higher level are recorded. For each pair of adjacent nodes, their fault energy fingerprint deviation in the multidimensional fused data field is read. The deviation of the node at the next higher level is subtracted from the deviation of the node at the previous higher level, and the difference is divided by the impedance value of the corresponding line to obtain the attenuation gradient of that branch direction on that line segment. The following formula represents the attenuation gradient of the node at the next higher level. To the lower-level node The decay gradient value:

[0035] in, Indicates from node To the node The decay gradient value; Representing nodes respectively and nodes Fault energy fingerprint deviation; Represented as nodes and The equivalent complex impedance modulus of the line segment is calculated from the conductor type and length in the distribution network GIS model, with a value of 0.15Ω for typical urban cable lines; These represent the distances from the fault disturbance center node to the node. and The topological path hierarchy depth; The empirical attenuation coefficient is set to 0.12, which comprehensively reflects the propagation attenuation of transient high-frequency components in a multi-branch topology, the filtering effect of transformer windings on high-frequency components, and the discharge effect of line distributed capacitance. This represents the three-phase unbalance amplification factor, set to 0.5, used to enhance the gradient response to asymmetrical faults such as single-phase grounding. These represent the instantaneous peak values ​​of the three-phase currents during the fault transient process, provided by the real-time electrical quantities collected in step S1. This represents the average value of the peak values ​​of the three-phase currents. This indicates the elimination of the zero constant.

[0036] Repeat the above calculations for each branch and each level until all topological nodes in the network are covered, thus constructing a complete decay gradient field.

[0037] Step S42: Calculate the correlation coefficient between the attenuation gradient value sequence of each branch direction and the standard attenuation curve in the preset fault type feature template, and select the fault type with the highest correlation coefficient as the fault type to be judged. Step S421: Obtain real-time electrical quantity data of each topology node in the fault section, calculate the actual fault current and fault voltage characteristics, and compare and verify them with the standard fault electrical characteristics corresponding to the fault type. After obtaining the preliminary assessment of the fault type, all topological nodes within the fault section corresponding to that fault type are located, and real-time electrical quantity data of three-phase voltage and three-phase current are collected for these nodes at the moment of the fault occurrence and within several cycles before and after. After filtering the collected electrical quantity data, the fundamental component and characteristic subharmonic components are extracted respectively, and the amplitude and phase of the actual fault current and the amplitude and phase of the actual fault voltage are calculated to form the actual fault current and voltage characteristic vector. Simultaneously, the standard fault electrical characteristic vector corresponding to this assessed fault type is retrieved from a preset fault type feature template library to provide a benchmark reference for subsequent comparison and verification.

[0038] Step S422: If the comparison and verification results are consistent, the determined fault type is confirmed as the final fault type; if they are inconsistent, the fault type with the second highest correlation coefficient is selected as the corrected fault type, and the comparison and verification are repeated until the final fault type is confirmed. The actual fault feature vector is compared dimension by dimension with the standard fault electrical feature vector, and the relative deviation value of each dimension is calculated. If the relative deviation values ​​of all dimensions are within the preset allowable deviation range, the comparison verification results are considered consistent, and the current fault type is confirmed as the final fault type. If the relative deviation value of any dimension exceeds the allowable deviation range, the comparison verification results are considered inconsistent. In this case, the fault type corresponding to the second highest correlation coefficient is selected from the correlation coefficient ranking list as the corrected fault type, and the standard fault electrical feature vector corresponding to this type is re-extracted and compared again. The above process is repeated until a fault type passes the comparison verification and is confirmed as the final fault type.

[0039] Step S423: Associate the confirmed final fault type with the fault section and output the result; The fault type, confirmed through step-by-step comparison and verification, is then associated and bound with the fault section identified in the preceding steps to form a complete assessment result containing fault section location information and fault type attribute information. The associated result is then packaged in a unified data format and output to the result display module and alarm push module of the distribution network fault assessment system. At the same time, key intermediate variables and judgment criteria during the association process are recorded and stored in the system log for subsequent review and analysis.

[0040] Step S43: Along the direction of the topological branch where the attenuation gradient value changes most drastically, the continuous node interval where the attenuation gradient value exceeds the preset gradient threshold is determined as the fault section. In the constructed attenuation gradient field, the variation pattern of the attenuation gradient value is scanned one by one along the direction of each topological branch. Starting from the fault disturbance center node, the attenuation gradient values ​​of adjacent nodes are compared level by level along each branch, and the difference between adjacent gradient values ​​is calculated. The direction with the largest absolute value of the difference is marked as the direction of the most drastic change in that branch. In the direction of the most drastic change, an attenuation gradient threshold is set, and the search starts from the center node and moves outward. The node intervals where the attenuation gradient value continuously exceeds the threshold are extracted as candidate fault segments. If the gradient value falls back below the threshold in this direction, the search for that branch is interrupted, and the continuous node intervals already searched are determined as the final fault segments.

[0041] Step S5: The system outputs the fault section and fault type, and stores the attenuation gradient value in association with the fault section and fault type in the multi-dimensional fusion data field. Based on the association storage result, the normal operation benchmark and the preset fault type feature template are updated to complete the closed-loop processing of the distribution network fault judgment method. Specifically, the fault section, fault type, and attenuation gradient value sequence are used to generate a fault record data package and output to the distribution automation master station. Simultaneously, the data is stored in the fault sample library under the corresponding fault type category. When the number of samples of the same fault type reaches a preset threshold, the statistical mean curve of the attenuation gradient value sequence under that type is recalculated, the standard attenuation curve in the preset fault type feature template is updated, and the average fault energy fingerprint within a preset time window before the fault occurs is used as a new normal operation benchmark sample for moving average updating. This achieves dynamic maintenance and closed-loop feedback of the multi-dimensional fused data field, specifically including the following sub-steps: Step S51: Generate a fault record data packet from the fault section, fault type, and attenuation gradient value sequence, and output it to the distribution automation master station; The system combines the fault segment and fault type output by the fault analysis module with the attenuation gradient value sequence calculated during the fault occurrence to generate a standardized fault record data packet. The data packet uses a compact binary format, containing header information and body data. The header information records the fault occurrence time, data packet length, and checksum. The body data includes the fault segment node number sequence, fault type code, and attenuation gradient value array. The system uploads the data packet to the distribution automation master station in a high-priority queue. Upon receiving the packet, the master station performs a cyclic redundancy check. If no errors are found, the data packet is decompressed and temporarily stored in the receiving buffer, awaiting subsequent data entry.

[0042] Step S52: Associate and store the fault record data packet under the corresponding fault type category in the fault sample library; The main station locates the corresponding type directory in the fault sample database based on the fault type code in the data packet. If the type directory has not yet been created, the system automatically creates the directory and initializes the metadata file, recording the type name, creation time, current sample count, and feature version number. Subsequently, the system generates a unique sample number for this fault, writes the complete fault record data packet to a separate storage file in that directory, and appends an index record to the metadata file, including the sample number, entry time, associated fault segment hash value, and the offset of the decay gradient value sequence in the file. After the index is updated, the system extracts the decay gradient value sequence for this fault and copies it to the historical sequence temporary storage area for this type. The temporary storage area uses a circular queue structure, retaining only the latest preset number of sample sequences to support fast batch reading and efficient use of storage space.

[0043] Step S53: When the number of samples of the same fault type reaches the preset sample number threshold, recalculate the statistical mean curve of the decay gradient value sequence under the fault type, update the standard decay curve in the preset fault type feature template, and at the same time use the average fault energy fingerprint within the preset time window before the occurrence of this fault as the new normal operation benchmark sample and perform a moving average update. After each data entry operation, the system checks whether the sample count under the fault type directory has reached the preset sample quantity threshold. If it has not, the update process is not triggered; if it has, the feature template update process is triggered. First, the attenuation gradient value sequence of all historical samples of this type is read from the temporary storage area. Since the number of sampling points for each fault may vary due to fluctuations in fault duration and sampling frequency, the system uses cubic spline interpolation to uniformly resample all sequences to a fixed length. The following formula represents the updated standard attenuation curve at the [missing information - likely a specific point or value]. New mean at each standardized location:

[0044] in, This indicates the updated standard decay curve at the [number]th [time]. A standardized location; This indicates that the historical template retains a weight, with an initial value of 0.7. When the cumulative number of updates for this fault type exceeds 20, it will be adjusted to 0.9 to enhance the stability of historical experience. This represents the old mean value of the original preset standard attenuation curve at the same position; This represents the total number of historical samples used when the update is triggered; in this embodiment, it is set to 50. Indicates the historical sample index number; Indicates the first The confidence weight of each sample is determined based on the signal-to-noise ratio of the decay gradient sequence of that sample. Calculated; Indicates the first The original decay gradient sequence of each historical sample is interpolated using cubic spline interpolation at the standard interpolation nodes. Resampled values.

[0045] After resampling, the arithmetic mean curve and standard deviation curve of all samples at each position along the length are calculated point by point. The mean curve is used as the new standard decay curve, and the mean curve plus or minus twice the standard deviation curve is used as the confidence band. The corresponding curve in the original preset fault type feature template is replaced with the newly generated standard decay curve and confidence band.

[0046] Simultaneously, for updating the normal operation benchmark, the system extracts the fault energy fingerprint vector within a fixed time window from the continuous operational data prior to the current fault, calculates the mean of each dimension within that time window, and obtains the current measurement vector. A sliding weighted average strategy is used to fuse the current measurement vector with the historical benchmark vector. The updated benchmark vector equals the historical benchmark vector multiplied by a first weight coefficient plus the current measurement vector multiplied by a second weight coefficient, where the sum of the first and second weight coefficients is 1. The value of the second weight coefficient is dynamically adjusted based on the number of historical benchmark updates; fewer updates result in a larger second weight coefficient to accelerate convergence, while more updates result in a smaller second weight coefficient to maintain stability. The updated benchmark vector and its update timestamp are stored in the normal operation benchmark database, and the decay gradient value sequence of this fault and the feature version number of the fault energy fingerprint vector are synchronously updated in the metadata file.

[0047] Example 2

[0048] like Figure 2 As shown, Embodiment 2 of this application provides a distribution network fault assessment system based on multi-dimensional data fusion, comprising: Data Acquisition and Fusion Module 21: Collects electrical quantities, switch status quantities, environmental parameters, and maintenance reports from multiple information sources in the distribution network, aggregates them into multi-dimensional raw data, performs spatiotemporal synchronization and topology association on the multi-dimensional raw data, calculates the spatiotemporal coupling strength value of each information source on the topology nodes, assigns the spatiotemporal coupling strength value as the field strength weight to the corresponding nodes, and generates a multi-dimensional fused data field; including the following sub-modules: Multi-source data acquisition and time-series alignment submodule 211: Collects electrical quantities, switch status quantities, environmental parameters and repair texts through voltage transformers, current transformers, distribution automation terminals, meteorological monitoring stations and 95598 customer service system, performs time-series alignment according to a unified timestamp, and aggregates them into multi-dimensional raw data; Topology Association Submodule 212: Based on the distribution network geographic information system, establish an association matrix between each topology node, associate the time-aligned information source data with the corresponding topology node, and complete spatiotemporal synchronization and topology association; Multidimensional fusion data field construction submodule 213: Calculate the coupling coefficient between different information source data on each topology node, obtain the spatiotemporal coupling strength value by weighted summation, and assign the spatiotemporal coupling strength value as the field strength weight to the corresponding node to construct a multidimensional fusion data field covering the distribution network topology. Fault Feature Extraction Module 22: Based on a multi-dimensional fused data field, it extracts the fault energy fingerprint of each information source under the influence of field strength weights, calculates the deviation of the fault energy fingerprint from the normal operating baseline, and filters out a subset of abnormal information sources based on the fault energy fingerprint deviation; including the following sub-modules: Fault energy fingerprint construction submodule 221: The data of each information source in the multi-dimensional fusion data field are weighted by field strength weight, and the fundamental amplitude and harmonic distortion rate of electrical quantities, the action frequency of switch status quantities, the temperature and humidity change rate of environmental parameters, and the semantic encoding value of the repair text are extracted to construct the fault energy fingerprint. Deviation calculation submodule 222: Obtain the historical average value of various fault energy fingerprints of each topology node under normal operating conditions as the normal operating benchmark, calculate the Mahalanobis distance between the current fault energy fingerprint and the normal operating benchmark, and use it as the fault energy fingerprint deviation. Abnormal information source filtering submodule 223: compares the fault energy fingerprint deviation of each information source with the preset deviation threshold, filters out information sources with a deviation greater than the preset deviation threshold, and classifies them into the abnormal information source subset; Fault Disturbance Location Module 23: Maps a subset of abnormal information sources to topological nodes of the multidimensional fused data field, calculates the convergence density value of the fault energy fingerprint deviation on each node, and determines the fault disturbance center node based on the peak position of the convergence density value; includes the following sub-modules: Anomaly information source mapping submodule 231: Maps a subset of anomaly information sources to the corresponding topology nodes of the multidimensional fused data field, and counts the number of anomaly information sources and the fault energy fingerprint deviation value on each node; Aggregation density calculation submodule 232: Taking each topology node as the center and extending outward along the topology branch by a switch interval as the preset spatial neighborhood radius, the sum of the fault energy fingerprint deviation of all abnormal information sources within the neighborhood radius is calculated as the aggregation density value of the node. Fault Disturbance Center Determination Submodule 233: Traverse the convergence density values ​​of all topology nodes and select the topology node with the largest convergence density value as the fault disturbance center node; Fault Analysis Module 24: Taking the fault disturbance center node as the origin, it calculates the attenuation gradient value of the fault energy fingerprint deviation along the topological branch of the multidimensional fused data field. A three-phase unbalance amplification coefficient is introduced in the attenuation gradient value calculation to enhance the gradient response of asymmetric faults. The attenuation gradient value is compared with a preset fault type feature template using spatial attenuation patterns, and the initial judgment result is verified using real-time electrical quantity characteristics within the fault section to determine the fault section and fault type. It includes the following sub-modules: Attenuation gradient field construction submodule 241: Taking the fault disturbance center node as the origin, search for adjacent nodes outward step by step along each topology branch direction, calculate the ratio of the difference in fault energy fingerprint deviation between adjacent level nodes to the line impedance value, introduce the three-phase unbalance amplification factor to correct the calculation results, obtain the attenuation gradient value in the branch direction, and construct the attenuation gradient field. Fault type judgment submodule 242: Calculate the correlation coefficient between the attenuation gradient value sequence of each branch direction and the standard attenuation curve in the preset fault type feature template, and select the fault type with the highest correlation coefficient as the fault type to be judged. Fault Section Determination Submodule 243: Along the direction of the topological branch where the attenuation gradient value changes most drastically, the continuous node interval where the attenuation gradient value exceeds the preset gradient threshold is determined as the fault section. Output and Closed-Loop Update Module 25: The system outputs the fault section and fault type, and associates the attenuation gradient value with the fault section and fault type to the multi-dimensional fused data field. Based on the associated storage results, it updates the normal operation benchmark and the preset fault type feature template, completing the closed-loop processing of the distribution network fault judgment method; including the following sub-modules: Fault Result Output Submodule 251: Generates fault record data packets from fault section, fault type and attenuation gradient value sequence, and outputs them to the distribution automation master station; Fault record storage submodule 252: Associates and stores fault record data packets under the corresponding fault type category in the fault sample library; Benchmark and Template Update Submodule 253: When the number of samples of the same fault type reaches the preset sample number threshold, the statistical mean curve of the decay gradient value sequence under the fault type is recalculated, the standard decay curve in the preset fault type feature template is updated, and the mean value of the fault energy fingerprint within the preset time window before the occurrence of the fault is used as the new normal operation benchmark sample for moving average update.

Claims

1. A method for analyzing distribution network faults based on multi-dimensional data fusion, characterized in that, include: S1. Collect electrical quantities, switch status quantities, environmental parameters and repair reports from multiple information sources in the distribution network, aggregate them into multi-dimensional raw data, perform spatiotemporal synchronization and topology association on the multi-dimensional raw data, calculate the spatiotemporal coupling strength value of each information source on the topology node, assign the spatiotemporal coupling strength value as the field strength weight to the corresponding node, and generate a multi-dimensional fused data field. S2. Based on the multi-dimensional fusion data field, extract the fault energy fingerprint of each information source under the influence of field strength weight, calculate the deviation of the fault energy fingerprint from the normal operation benchmark, and select the subset of abnormal information sources based on the deviation of the fault energy fingerprint. S3. Map the subset of abnormal information sources to the topological nodes of the multidimensional fused data field, calculate the convergence density value of the fault energy fingerprint deviation on each node, and determine the fault disturbance center node based on the peak position of the convergence density value. S4. Taking the fault disturbance center node as the origin, calculate the attenuation gradient value of the fault energy fingerprint deviation along the topological branch of the multidimensional fused data field. In the calculation of the attenuation gradient value, a three-phase unbalance amplification coefficient is introduced to enhance the gradient response of asymmetric faults. The attenuation gradient value is compared with the preset fault type feature template in terms of spatial attenuation mode, and the initial judgment result is verified by the real-time electrical quantity feature quantity in the fault section to determine the fault section and fault type. S5. The system outputs the fault section and fault type, and stores the attenuation gradient value in association with the fault section and fault type in the multi-dimensional fusion data field. Based on the association storage results, the normal operation benchmark and the preset fault type feature template are updated to complete the closed-loop processing of the distribution network fault judgment method.

2. The method for analyzing distribution network faults based on multi-dimensional data fusion according to claim 1, characterized in that, Electrical quantities, switch status quantities, environmental parameters, and maintenance reports are collected from multiple information sources in the distribution network and aggregated into multidimensional raw data. Spatiotemporal synchronization and topology association are performed on the multidimensional raw data, and the spatiotemporal coupling strength value of each information source on the topology node is calculated. The spatiotemporal coupling strength value is used as the field strength weight and assigned to the corresponding node to generate a multidimensional fused data field, including the following sub-steps: Electrical quantities, switch status quantities, environmental parameters, and repair reports are collected through voltage transformers, current transformers, distribution automation terminals, meteorological monitoring stations, and the 95598 customer service system. The data is then time-series aligned according to a unified timestamp and aggregated into multi-dimensional raw data. Based on the geographic information system of the power distribution network, an association matrix is ​​established between each topology node, and the time-aligned information source data is associated with the corresponding topology node to complete the spatiotemporal synchronization and topology association. The coupling coefficients between different information source data on each topology node are calculated, and the spatiotemporal coupling strength value is obtained by weighted summation. The spatiotemporal coupling strength value is used as the field strength weight and assigned to the corresponding node to construct a multidimensional fusion data field covering the distribution network topology.

3. The method for analyzing distribution network faults based on multi-dimensional data fusion according to claim 1, characterized in that, Based on a multi-dimensional fused data field, fault energy fingerprints are extracted from each information source under the influence of field strength weights. The deviation of the fault energy fingerprint from the normal operating baseline is calculated. Based on the fault energy fingerprint deviation, a subset of abnormal information sources is selected, including the following sub-steps: The data from each information source in the multidimensional fusion data field are weighted by field strength weight, and the fundamental amplitude and harmonic distortion rate of electrical quantities, the operation frequency of switch status quantities, the temperature and humidity change rate of environmental parameters, and the semantic coding value of the repair text are extracted to construct a fault energy fingerprint. The historical average values ​​of various fault energy fingerprints of each topology node under normal operating conditions are obtained as the normal operating benchmark. The Mahalanobis distance between the current fault energy fingerprint and the normal operating benchmark is calculated as the fault energy fingerprint deviation. The deviation of the fault energy fingerprint of each information source is compared with the preset deviation threshold, and information sources with a deviation greater than the preset deviation threshold are selected and classified into the abnormal information source subset.

4. The method for analyzing distribution network faults based on multi-dimensional data fusion according to claim 1, characterized in that, Mapping a subset of abnormal information sources to topological nodes of the multidimensional fused data field, calculating the convergence density value of the fault energy fingerprint deviation on each node, and determining the fault disturbance center node based on the peak position of the convergence density value, includes the following sub-steps: A subset of abnormal information sources is mapped to the corresponding topological nodes of the multidimensional fused data field, and the number of abnormal information sources and the deviation value of fault energy fingerprint on each node are counted. Taking each topology node as the center, extend outward along the topology branch by a switch interval as the preset spatial neighborhood radius, and count the sum of the fault energy fingerprint deviation of all abnormal information sources within this neighborhood radius as the aggregation density value of the node. Iterate through the aggregation density values ​​of all topology nodes and select the topology node with the largest aggregation density value as the fault disturbance center node.

5. The method for analyzing distribution network faults based on multi-dimensional data fusion according to claim 1, characterized in that, Taking the fault disturbance center node as the origin, the attenuation gradient value of the fault energy fingerprint deviation is calculated along the topological branch of the multidimensional fused data field. The three-phase unbalance amplification coefficient is introduced in the calculation of the attenuation gradient value to enhance the gradient response of asymmetric faults. The attenuation gradient value is compared with the preset fault type feature template to determine the spatial attenuation pattern, and the initial judgment result is verified by the real-time electrical quantity feature quantity within the fault section to determine the fault section and fault type. This includes the following sub-steps: Taking the fault disturbance center node as the origin, the adjacent nodes are searched outward level by level along each topological branch direction. The ratio of the difference in fault energy fingerprint deviation between adjacent level nodes to the line impedance value is calculated. The three-phase unbalance amplification factor is introduced to correct the calculation results, and the attenuation gradient value in the branch direction is obtained to construct the attenuation gradient field. The correlation coefficient is calculated between the attenuation gradient value sequence of each branch direction and the standard attenuation curve in the preset fault type feature template, and the fault type with the highest correlation coefficient is selected as the fault type to be judged. Along the topological branch where the attenuation gradient value changes most drastically, the continuous node interval where the attenuation gradient value exceeds the preset gradient threshold is identified as the fault section.

6. The method for analyzing distribution network faults based on multi-dimensional data fusion according to claim 1, characterized in that, The system outputs the fault section and fault type, and associates the attenuation gradient value with the fault section and fault type, storing it in a multi-dimensional fused data field. Based on the associated storage results, it updates the normal operation benchmark and the preset fault type feature template, completing the closed-loop processing of the distribution network fault assessment method, including the following sub-steps: The fault section, fault type and attenuation gradient value sequence are used to generate a fault record data packet, which is then output to the distribution automation master station. The fault record data packets are associated and stored under the corresponding fault type category in the fault sample library; When the number of samples of the same fault type reaches the preset sample number threshold, the statistical mean curve of the decay gradient value sequence under the fault type is recalculated, the standard decay curve in the preset fault type feature template is updated, and the mean value of the fault energy fingerprint within the preset time window before the occurrence of this fault is used as the new normal operation benchmark sample and updated by moving average.

7. The method for analyzing distribution network faults based on multi-dimensional data fusion according to claim 5, characterized in that, The correlation coefficient is calculated between the attenuation gradient value sequence of each branch direction and the standard attenuation curve in the preset fault type feature template. The fault type with the highest correlation coefficient is selected as the fault type for judgment. This includes the following sub-steps: Acquire real-time electrical quantity data of each topology node in the fault section, calculate the actual fault current and fault voltage characteristics, and compare and verify them with the standard fault electrical characteristics corresponding to the fault type. If the comparison and verification results are consistent, then the fault type is confirmed as the final fault type. If there is a discrepancy, the fault type with the second highest correlation coefficient is selected as the corrected fault type, and the comparison and verification are repeated until the final fault type is confirmed. The final confirmed fault type will be associated with the faulty section and output.

8. A distribution network fault assessment system based on multi-dimensional data fusion, characterized in that, include: Data acquisition and fusion module: Collects electrical quantities, switch status quantities, environmental parameters and repair reports from multiple information sources in the distribution network, aggregates them into multi-dimensional raw data, performs spatiotemporal synchronization and topology association on the multi-dimensional raw data, calculates the spatiotemporal coupling strength value of each information source on the topology node, assigns the spatiotemporal coupling strength value as the field strength weight to the corresponding node, and generates a multi-dimensional fused data field. Fault Feature Extraction Module: Based on a multi-dimensional fused data field, extract the fault energy fingerprint of each information source under the influence of field strength weight, calculate the deviation of the fault energy fingerprint from the normal operation benchmark, and filter out a subset of abnormal information sources based on the fault energy fingerprint deviation. Fault disturbance localization module: Maps a subset of abnormal information sources to the topological nodes of the multidimensional fused data field, calculates the convergence density value of the fault energy fingerprint deviation on each node, and determines the fault disturbance center node based on the peak position of the convergence density value; Fault analysis module: Taking the fault disturbance center node as the origin, the attenuation gradient value of the fault energy fingerprint deviation is calculated along the topological branch of the multidimensional fused data field. The attenuation gradient value calculation introduces the three-phase unbalance amplification coefficient to enhance the gradient response of asymmetric faults. The attenuation gradient value is compared with the preset fault type feature template in terms of spatial attenuation mode, and the initial judgment result is verified by the real-time electrical quantity feature quantity in the fault section to determine the fault section and fault type. Output and Closed-Loop Update Module: The system outputs the fault section and fault type, and associates the attenuation gradient value with the fault section and fault type to the multi-dimensional fusion data field. Based on the associated storage results, it updates the normal operation benchmark and the preset fault type feature template to complete the closed-loop processing of the distribution network fault judgment method.

Images