Power distribution network communication fault diagnosis method

Abstract

The invention discloses a power distribution network communication fault diagnosis method, and relates to the technical field of power distribution network communication, and the method comprises the steps: dividing a multi-dimensional data variable into a plurality of data groups, forming a neighborhood graph, and constructing a similarity matrix; obtaining a low-dimensional data variable based on the feature value and the feature vector of the similarity matrix; calculating a corresponding influence weight and correlation between any two variables based on the low-dimensional data variables; performing preliminary grouping on the low-dimensional data variables according to the correlation, and dividing manifold groups according to the total group weight ratio of each group; acquiring a real-time data group, calculating a manifold coordinate and determining a manifold group, calculating a final credibility according to the manifold group, and determining fault data according to the final credibility; the influence degree of dominant paths of different factors is accurately measured; according to the manifold group condition to which the real-time data group belongs, different methods are adopted to calculate the final trust degree so as to determine the fault data, and the problems of high calculation complexity and limitation and accuracy which are difficult to understand in a traditional analysis method are solved.

Description

Technical Field

[0001] This invention relates to the field of power distribution network communication technology, and in particular to a method for diagnosing communication faults in power distribution networks. Background Technology

[0002] In the field of power system fault diagnosis, the complexity and multi-factor coupling characteristics of distribution networks have long constrained the accuracy and timeliness of fault analysis. Traditional diagnostic methods are usually based on single-dimensional data or simple threshold judgments, which are difficult to effectively handle the interaction of multi-dimensional variables such as communication delay, current overload, and protection status.

[0003] Traditional methods have significant limitations. Redundant information and noise in high-dimensional data lead to a dramatic increase in computational complexity and make intuitive visualization and analysis difficult. Existing technologies lack the ability to identify interactions between dynamic forces, making it impossible to identify hidden faults, which can easily escalate from local faults into large-scale power outages. This not only affects the accuracy of fault location but also poses a serious threat to the safe and stable operation of the power system. Summary of the Invention

[0004] This application provides a method for diagnosing communication faults in power distribution networks, which solves the problems of high computational complexity and inability to accurately identify potential faults in existing technologies, and achieves the technical effects of quantifying interaction effects and improving the accuracy of prediction and diagnosis.

[0005] This application provides a method for diagnosing communication faults in a power distribution network, the method comprising:

[0006] S1: Obtain multidimensional data variables that characterize the power grid operation status, divide the multidimensional data variables into several data groups and form a neighborhood graph; calculate the manifold distance between two adjacent edge neighbor groups in the neighborhood graph and construct a similarity matrix; determine the low-dimensional embedding dimension and low-dimensional spatial coordinate axis based on the eigenvalues ​​and eigenvectors of the similarity matrix, and then obtain the low-dimensional data variables.

[0007] S2: Obtain the initial weight of each variable in the low-dimensional data variables and calculate the correlation between any two variables; perform preliminary grouping of the low-dimensional data variables according to the initial weight and correlation, and divide the manifold groups according to the total weight ratio of each group;

[0008] S3: Obtain real-time data sets, calculate their coordinate positions in the low-dimensional space coordinate axes based on the similarity matrix and determine their manifold group, calculate the final confidence level based on the manifold group, and determine the fault data based on the final confidence level.

[0009] If the manifold group to which it belongs is a single manifold group, then the trust score of the corresponding manifold group is calculated as the final trust score.

[0010] If a manifold belongs to two manifold groups, the independent confidence level of each manifold group is calculated, and the dynamic weighting coefficient is determined based on the historical diagnostic accuracy of each manifold group to obtain the biased confidence level as the final confidence level.

[0011] If a sample belongs to three manifold groups, the confidence scores of the samples from the three manifold groups in the time range before and after the real-time data group are obtained. A decision matrix is ​​constructed to obtain the information entropy weight, positive ideal value, and negative ideal value. The distance between each sample and the positive ideal value and the negative ideal value is calculated to obtain the relative proximity as the final confidence score.

[0012] Furthermore, the method also includes: S31: If the user belongs to three manifold groups simultaneously, calculate the instantaneous rate of change of the trust level of each manifold group at each time point according to the decision matrix; set the sliding window size to N, and obtain the rate of change values ​​at N time points within the sliding window to form three change matrices;

[0013] Calculate the correlation matrix between the three change matrices based on the rate of change data within the sliding window;

[0014] Eigenvalue decomposition is performed based on the correlation matrix. The largest eigenvalue is compared with a pre-set interaction threshold. If the largest eigenvalue is greater than the interaction threshold, the current time point is marked as a strong interaction time point, and the largest eigenvalue is the manifold synergy factor. All strong interaction time points and manifold synergy factors are counted to establish a time variation curve.

[0015] S32: Based on the time variation curve, select the manifold group corresponding to the largest manifold synergy factor as the suspected source, calculate the propagation speed and pre-propagation path of the suspected source according to the fault propagation model, determine the affected line and affected area, and generate a fault report.

[0016] Furthermore, based on the eigenvalues ​​and eigenvectors of the similarity matrix, the low-dimensional embedding dimension and low-dimensional spatial coordinate axes are determined, thereby obtaining low-dimensional data variables. This includes: calculating the sum of all eigenvalues, calculating the proportion of each eigenvalue to the sum as the contribution rate of each eigenvalue, arranging all eigenvalues ​​in descending order, and accumulating the contribution rates sequentially from the first eigenvalue to obtain the cumulative sum, with a pre-set contribution threshold; when the cumulative sum first exceeds the contribution threshold, the accumulation stops, and the number of eigenvalues ​​participating in the accumulation is recorded.

[0017] Set the number of feature values ​​to the low-dimensional embedding dimension, and use the corresponding feature vectors as low-dimensional feature vectors to construct low-dimensional space coordinate axes.

[0018] Perform an inner product operation between each data set and the i-th low-dimensional feature vector to obtain its coordinate value on the i-th coordinate axis in the low-dimensional space; determine the low-dimensional embedding point based on the coordinate value, map the data set to the corresponding low-dimensional embedding point, and form a low-dimensional data variable after all embeddings are completed.

[0019] Furthermore, the correlation matrix includes a linear correlation matrix and a time-delay correlation matrix; the matrix elements of the linear correlation matrix are the covariance of the rates of change among the manifold groups, which are used to quantify the correlation strength between different manifold groups.

[0020] Establishing a time-delay correlation matrix includes: estimating the probability distribution of the rate of change sequence for each manifold group; constructing a nonlinear correlation matrix based on the probability distribution estimate and mutual information values, where the matrix elements are the mutual information values ​​between any two manifold groups, and the diagonal elements are 1; calculating the time-delay mutual information, calculating the time-delay mutual information for each pair of manifold group rate of change sequences within the time delay range, finding the time delay value that maximizes the mutual information, and establishing a time-delay correlation matrix based on the nonlinear correlation matrix and the time delay value; and fusing the analysis results of the linear correlation matrix and the time-delay correlation matrix to obtain the manifold synergy factor.

[0021] The probability distribution estimation includes: using an adaptive interval discretization method to calculate the percentile of each rate of change sequence, discretizing the continuous values ​​into 6 levels based on the percentiles, calculating the joint probability distribution, and for any two manifold groups of rate of change sequences, obtaining a two-dimensional joint probability distribution matrix and marginal probability distribution based on the joint probability distribution and the rate of change sequences.

[0022] Furthermore, the propagation speed and pre-propagation path of the suspected source are calculated based on the fault propagation model, including:

[0023] Based on the power grid topology, starting from the fault source node, the short-circuit current distribution is calculated to determine the affected downstream nodes and lines. Impedance matrices are used for power flow calculations to identify potentially overloaded components. Protection signal transmission is simulated, and the time from the fault source to each protection device is calculated by combining real-time communication delay and message loss rate. Protection actions and circuit breaker operations are simulated, incorporating real-time circuit breaker operating times and protection status. Fault propagation is iteratively simulated until no new nodes are affected or the propagation time exceeds the maximum allowable time. After iteration, the propagation speed and pre-propagation path are obtained.

[0024] Furthermore, a fault propagation model is established, including: the fault propagation model includes an electrical propagation layer, an information propagation layer, and a control propagation layer; fault propagation model parameters are predefined, including fault source nodes, propagation speed, and propagation path; based on the determined fault source, the propagation speed is calculated in the fault propagation model according to the equipment connection status of the relevant lines, and the propagation path is determined; the propagation speed includes information transmission speed and control transmission speed.

[0025] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0026] By manifold partitioning and decoupling, complex problems are decomposed into multiple relatively independent sub-problems, and the impact of each factor on the system is analyzed separately, thereby more accurately and quickly determining the cause of the fault and the system state, ensuring the safe and stable operation of the power system; multidimensional data variables are mapped to a low-dimensional space;

[0027] Preliminary grouping is performed based on the influence weights and correlations of low-dimensional data variables, and manifold groups are divided according to the weight ratio of the total group to accurately measure the influence of different factors on the dominant path. Based on the manifold group to which the real-time data group belongs, different methods are used to calculate the final confidence level, thereby identifying fault data and solving the limitations and inaccuracies of traditional analysis methods, such as high computational complexity and difficulty in understanding. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of a power distribution network communication fault diagnosis method according to an embodiment of the present invention. Detailed Implementation

[0029] To facilitate understanding of the present invention, a more complete description of this application will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the present invention.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention; the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0031] Example 1: In the prior art, multiple factors interact and influence each other in power systems, making it difficult for traditional analysis methods to accurately determine the cause of faults and the system state. In this example, through manifold partitioning and decoupling, the complex problem can be decomposed into multiple relatively independent sub-problems, and the impact of factors such as communication interference, power faults, and protection interactions on the system can be analyzed separately, thereby improving the accuracy and timeliness of fault diagnosis and ensuring the safe and stable operation of the power system.

[0032] like Figure 1 As shown, a method for diagnosing communication faults in a power distribution network includes:

[0033] S1: Obtain multidimensional data variables that characterize the power grid operation status, divide the multidimensional data variables into several data groups and form a neighborhood graph; calculate the manifold distance between two adjacent edge neighbor groups in the neighborhood graph and construct a similarity matrix; determine the low-dimensional embedding dimension and low-dimensional spatial coordinate axis based on the eigenvalues ​​and eigenvectors of the similarity matrix, and then obtain the low-dimensional data variables; the neighborhood graph includes a core data group and several edge neighbor groups.

[0034] In some embodiments, the multidimensional data variables include communication delay, message loss rate, primary protection status, backup protection status, circuit breaker operating time, and current overload rate. Each multidimensional data variable is grouped with its corresponding time point to form a data set, and each data set contains the corresponding time point and the obtained complete multidimensional data variable values.

[0035] The communication delay refers to the time difference in data transmission from the sending end to the receiving end in the power system's communication network. Specifically, it is achieved by installing high-precision time synchronization devices at both the sending and receiving ends of the distribution network communication link. The transmission time difference is calculated by comparing the sending and receiving timestamps, resulting in the transmission delay data variable. The communication delay represents the time elapsed for data transmission when different devices in the power system exchange information, measured in milliseconds (ms), ranging from 0 to 500 ms. Communication delay affects the fault detection and response speed of protection devices; excessive delay may prevent timely action by protection devices, potentially expanding the scope of the fault.

[0036] Deploying network probes or protocol analyzers on communication gateways or routers allows for the statistical analysis of the total number of messages sent and successfully received per unit time, calculating the loss ratio to obtain the message loss rate. The message loss rate is calculated by statistically analyzing the number of messages sent and successfully received within a pre-set time period. The time period needs to be pre-set based on actual conditions and historical experimental data, such as one month or three months. In power communication, the percentage of lost messages out of the total number of sent messages ranges from 0 to 100%. An excessively high message loss rate can affect the accuracy of information obtained by protection devices, potentially leading to malfunctions or failures to operate.

[0037] The main protection system is a primary protection device for power equipment. Its status is determined through the device's own monitoring functions and operational status feedback. Specifically, the protection device (such as a digital relay) uses its self-test function to monitor hardware status and fault signals in real time, and uploads status information (normal / failure to operate / false operation) to the monitoring system via a communication module to determine the main protection status. The protection device monitors its own hardware and software status in real time, as well as whether it has detected fault signals, and uploads this information to the monitoring system. The main protection status represents the current operating state of the main protection device, which is divided into three situations: normal, failure to operate, and false operation. Normal means that the protection device can correctly detect faults and operate according to design requirements; failure to operate means that the protection device fails to operate when it should; false operation means that the protection device operates incorrectly when there is no fault.

[0038] Backup protection serves as a backup to the main protection. Its status monitoring method is similar to that of the main protection. The backup protection device monitors its operating status through internal sensors and simultaneously receives action information from the main protection as a basis for judgment. The status data is transmitted to a central database via a communication link to determine the backup protection status. The backup protection status indicates the current operating state of the backup protection device and is categorized into three types: normal, failure to operate, and maloperation. When the main protection fails, the backup protection should operate to clear the fault; if the backup protection fails to operate when it should, it is considered a failure to operate; if it operates when it should not, it is considered a maloperation.

[0039] Circuit breakers are crucial devices in power systems used to disconnect and connect circuits. Their operating time can be measured by installing a time measuring device in the circuit breaker's control circuit. When the circuit breaker receives an opening or closing signal, the measuring device starts timing and stops timing when the circuit breaker completes its operation. Specifically, a high-resolution timer (such as a millisecond-level digital timing module) is installed in the circuit breaker's control circuit to start timing from the receipt of the opening / closing signal until the mechanical action is completed (timing stops when a change in the state of auxiliary contacts is triggered), thus determining the circuit breaker's operating time.

[0040] The circuit breaker operating time refers to the time required for the circuit breaker to complete the operation from receiving the operating signal. The operation generally includes two types: opening and closing, and the unit is milliseconds (ms), ranging from 0 to 200 ms. Excessive circuit breaker operating time will affect the fault clearing speed and impact the stable operation of the power system.

[0041] Current transformers are installed on power lines or equipment to convert actual current signals into measurable electrical signals. Current measuring instruments or protection devices then measure the current in real time. The measured actual current value is compared with the rated current value of the equipment or line to calculate the current overload rate. The current overload rate represents the degree to which the actual current exceeds the rated current, expressed as a percentage, ranging from 0% to 500%. The current overload rate reflects the load condition of power equipment or lines; an excessively high overload rate can lead to equipment overheating, damage, and even fires and other safety accidents.

[0042] In some embodiments, all multidimensional data variables within a preset period are obtained, and each multidimensional data variable is grouped with its corresponding time node to form a data set. A core data set is pre-determined, and T edge neighboring groups of the core data set in the high-dimensional space are found. With the core data set as the center, the T edge neighboring groups are connected to form a neighborhood graph. The manifold distance between two adjacent edge neighboring groups is calculated as the weight value of the edge in the neighborhood graph. A similarity matrix is ​​constructed based on the manifold distance. The low-dimensional embedding dimension and low-dimensional space coordinate axis are determined based on the eigenvalues ​​and eigenvectors of the similarity matrix, thereby obtaining the low-dimensional data variables.

[0043] Based on the eigenvalues ​​and eigenvectors of the similarity matrix, the low-dimensional embedding dimension and low-dimensional spatial coordinate axes are determined, thus obtaining low-dimensional data variables. This includes: calculating the sum of all eigenvalues ​​and calculating the proportion of each eigenvalue in the sum as the contribution rate of each eigenvalue. All eigenvalues ​​are arranged in descending order, and the contribution rates are accumulated sequentially from the first eigenvalue to obtain a cumulative sum, with a pre-set contribution threshold. When the cumulative sum first exceeds the contribution threshold, accumulation stops, and the number of eigenvalues ​​involved in the accumulation is recorded. The number of eigenvalues ​​is set as the low-dimensional embedding dimension, and the corresponding eigenvectors are used as low-dimensional eigenvectors to construct the low-dimensional spatial coordinate axes. Each data set is inner-producted with the i-th low-dimensional eigenvector to obtain the coordinate value on the i-th coordinate axis in the low-dimensional space. The low-dimensional embedding point is determined based on the coordinate value, and the data set is mapped to the corresponding low-dimensional embedding point. After all embeddings are completed, low-dimensional data variables are formed.

[0044] For example, the sequence of eigenvalues ​​is: The corresponding contribution rate The contribution rate is accumulated starting from the first eigenvalue, i.e. when When the contribution threshold is exceeded for the first time, the value of k becomes the determined low-dimensional embedding dimension. This means that selecting the feature vectors corresponding to the first k feature values ​​to construct the low-dimensional space can preserve the main features and structure of the high-dimensional data. Let the selected low-dimensional feature vectors be... For each high-dimensional data set, its coordinates in the low-dimensional space are the coordinates of the data set and the i-th low-dimensional feature vector. By performing an inner product of (i=1,2,...,k), we obtain the coordinate value on the i-th coordinate axis in the low-dimensional space. By calculating all k coordinate values, we determine the low-dimensional embedding point based on the coordinate value and map the data group to the corresponding low-dimensional embedding point. This allows us to map the high-dimensional data group to the low-dimensional space formed by the selected low-dimensional feature vectors. After all embedding is completed, a low-dimensional data variable is formed.

[0045] The shortest path algorithm is used to calculate the shortest path distance between two adjacent edge neighbor groups on the neighborhood graph as the manifold distance. The edge neighbor group is the remaining data group that is spatially closest to the data group within the high-dimensional space region. During graph construction, scaling is used to treat each data group as a point, and the distance within a group is calculated directly between the two points. T is a pre-set fixed parameter, dynamically adjusted according to data characteristics and density; the higher the data density, the larger the T value. Preferably, a base T value is pre-set based on historical experimental data. The manifold distance is the distance between two data groups in the neighborhood graph. A manifold is a geometric object with local Euclidean spatial properties. The manifold distance considers the local structure and paths of the data group on the manifold. By calculating the shortest path along the manifold to determine the distance between two points, it can better capture the inherent geometric structure of the data in high-dimensional space. The elements of the similarity matrix represent the similarity between two data groups, and a Gaussian kernel function is used to convert the manifold distance into similarity.

[0046] In some embodiments, for a matrix A, if there exist non-zero vectors and scalars such that the product of A and the non-zero vectors equals the product of the scalars and the non-zero vectors, then the scalar is called an eigenvalue of matrix A, and the non-zero vector is called an eigenvector corresponding to the eigenvalue. Eigenvectors represent the direction and degree of change of data in different dimensions, while eigenvalues ​​reflect the intensity of data change in that direction.

[0047] Low-dimensional embedding is the process of mapping high-dimensional data to a low-dimensional space. High-dimensional data often contains a large amount of redundant information and noise, and directly processing high-dimensional data will face problems such as high computational complexity, difficulty in understanding and visualization. The goal of low-dimensional embedding is to find a low-dimensional representation that can preserve the main features and structure of high-dimensional data in this low-dimensional space, while reducing the dimensionality of the data.

[0048] Low-dimensional feature vectors are selected from the k eigenvectors (where k is a defined low-dimensional embedding dimension) with larger eigenvalues ​​after performing eigenvalue decomposition on the similarity matrix. These feature vectors form a coordinate axis in a low-dimensional space, which can map high-dimensional data to this low-dimensional space while preserving the main features and structure of the high-dimensional data, reducing the dimensionality of the data, and thus forming low-dimensional data variables.

[0049] S2: Obtain the initial weight of each variable in the low-dimensional data variables and calculate the correlation between any two variables; perform preliminary grouping of the low-dimensional data variables according to the initial weight and correlation, and divide the manifold groups according to the total weight ratio of each group.

[0050] For low-dimensional data variables, based on the analysis of the impact of different factors in the power system, the index used to divide the manifold is determined as the weight ratio of each variable in the manifold. The weight ratio reflects the importance of the variable in a specific manifold. Specifically, historical fault data is collected, and the numerical changes of each low-dimensional data variable before and after the fault occur are recorded; the Pearson correlation coefficient is used to calculate the linear correlation strength between variables, and nonlinear correlation is analyzed through mutual information; the loading of each variable on the principal components is determined; the initial weight is used to measure the degree of influence of each variable in the data set on communication stability. The larger the initial weight, the greater the influence on communication stability (e.g., the influence coefficient of communication delay on the protection system is 0.7, and the influence coefficient of message loss rate is 0.8).

[0051] Calculating the correlation between any two variables includes: obtaining low-dimensional data variables, performing data cleaning and normalization, generating a linear matrix based on the Pearson correlation coefficient, and calculating the linear correlation; generating a nonlinear matrix based on mutual information and calculating the nonlinear correlation; weighted summing of the linear and nonlinear correlations to calculate the comprehensive correlation; determining the correlation threshold for the comprehensive correlation based on the initial weights, and performing preliminary grouping based on the correlation threshold. The larger the initial weights, the stricter the correlation classification criteria, and the larger the correlation threshold. For example, if the impact coefficient of communication delay on the protection system is 0.7 and the impact coefficient of message loss rate is 0.8, the comprehensive correlation threshold for calculating communication delay and message loss rate should be greater than the average of their initial weights, i.e., 0.75. If the comprehensive correlation is greater than 0.75, it is marked as strongly correlated.

[0052] The initial grouping includes three groups. The first group consists of communication delay and message loss rate. Communication delay and message loss rate are highly correlated and jointly affect the reliability of the protection system. The second group consists of current overload rate and circuit breaker operating time. Current overload rate and circuit breaker operating time are strongly positively correlated and directly determine the speed of power fault propagation. The third group consists of main protection status and backup protection status. The state variables of the protection device (main / backup) and other variables have weak correlation (r<0.3) but have a decisive impact on stability. Manifold groups are divided according to the total weight ratio of each group.

[0053] The manifold groups are divided based on the total weight ratio of each group. This process includes: extracting all data groups from the low-dimensional data variables; for each data group within a group, calculating the weight ratio of each variable within that group and comparing it to a corresponding set threshold to divide the manifold groups. The initial group contains more data groups than the manifold groups contain; the manifold groups are a further selection process for each initial group.

[0054] Specifically: The weights of communication delay and message loss rate within each data group are calculated as a percentage of all variable combinations to measure the dominance of variable values ​​in each data group. If the sum of the weights exceeds a first threshold, the corresponding data group is assigned to the first manifold group. The first threshold is obtained through analysis and correlation calculation of historical data, or predetermined based on historical experimental data and expert analysis; in this application, the first threshold is set to 70%. The first manifold group is used to measure the influence of the dominant communication interference path. For example, suppose a data group has n variables. (The value of n is the same as the number of pre-acquired multidimensional data variables; here n is 6, representing 6 types of variable data.) Assume the communication delay is... Message loss rate ,calculate Weighting of all variables Normalize each variable and calculate its proportion in the sum of all variable values. The normalized variable value is... ,but , Calculate the sum of the weighted proportions of communication latency and message loss rate. ,like If the value is greater than the first threshold, the data group is assigned to the first manifold group.

[0055] The weighted proportions of current overload rate and circuit breaker operating time in all variable combinations are calculated for each data group. If the sum of these weighted proportions exceeds a second threshold, the corresponding data group is assigned to the second manifold group. The second threshold is obtained through analysis and correlation calculation of historical data, or predetermined based on historical experimental data and expert analysis. In this application, the second threshold is set to 75%, meaning that when the sum of the weighted proportions of these two variables exceeds 75%, the data group containing them is assigned to the second manifold group. The second manifold group is used to measure the impact of the dominant power fault path.

[0056] The weighted proportions of the primary protection state and backup protection state within each data group are calculated among all variable combinations. If the sum of these weighted proportions exceeds a third threshold, the corresponding data group is assigned to a third manifold group. The third threshold is obtained through analysis and correlation calculation of historical data, or predetermined based on historical experimental data and expert analysis. In this application, the third threshold is set to 80%, meaning that when the sum of the weighted proportions of these two variables exceeds 80%, the data group containing them is assigned to the third manifold group. The third manifold group is used to measure the impact of protection interaction paths. The first threshold is less than the second threshold, and the second threshold is less than the third threshold.

[0057] S3: Obtain real-time data sets, calculate their coordinate positions in the low-dimensional space coordinate axes based on the similarity matrix and determine their manifold groups, calculate the final confidence level based on the manifold group, and determine the fault data based on the final confidence level.

[0058] In some embodiments, a real-time data set refers to a collection of multidimensional data variables (including the values ​​of six variables and the timestamp of acquisition) collected in real time from the distribution network system for fault diagnosis. Real-time data variables include communication delay, message loss rate, main protection status, backup protection status, circuit breaker operating time, and current overload rate. These variables are acquired in real time through sensors (such as current transformers and timestamp devices) installed on grid equipment or through a communication monitoring system, and the data is transmitted to the diagnostic system in time-series format. Real-time data sets are continuously acquired, forming a real-time data stream. Each data set is processed independently, but a temporal context (such as a sliding window) is considered for subsequent analysis. Based on the similarity matrix and low-dimensional spatial coordinate axes generated in steps S1 and S2, the similarity between the real-time data set and the low-dimensional data variables is calculated to obtain a new real-time similarity matrix. The specific coordinate position in the low-dimensional spatial coordinate axes is determined based on eigenvalues ​​and eigenvectors, and compared with the pre-formed low-dimensional data variables to determine the manifold group to which it belongs.

[0059] Specifically, based on the multidimensional data variables of the real-time data group, its weight proportion in the three manifold groups is calculated. Combined with the pre-set corresponding thresholds, the manifold group to which it belongs is determined. Fuzzy manifold membership is used, allowing a data group to belong to multiple manifold groups with different probabilities. If it belongs to a single manifold group, the trust score of the corresponding manifold group is directly calculated. The fault data is determined based on the trust score. The fault data includes the fault cause, fault type, fault occurrence time, fault impact range, and optimal solution. The optimal solution is automatically generated by matching similar faults and historical solutions in historical analysis data.

[0060] In some embodiments, calculating the final trust score based on the manifold group to which the manifold group belongs, and determining the fault data based on the final trust score, includes: if the manifold group to which the manifold group belongs is a single manifold group, then calculating the trust score of the corresponding manifold group as the final trust score.

[0061] If a manifold belongs to two manifold groups, the independent confidence level of each manifold group is calculated, and the dynamic weighting coefficient is determined based on the historical diagnostic accuracy of each manifold group to obtain the biased confidence level as the final confidence level.

[0062] If a sample belongs to three manifold groups, the confidence scores of the samples from the three manifold groups in the time range before and after the real-time data group are obtained. A decision matrix is ​​constructed to obtain the information entropy weight, positive ideal value, and negative ideal value. The distance between each sample and the positive ideal value and the negative ideal value is calculated to obtain the relative proximity as the final confidence score.

[0063] In some embodiments, if a user belongs to two manifold groups simultaneously, the independent confidence score for each manifold group is calculated, and a dynamic weighting coefficient is determined based on the historical diagnostic accuracy of each manifold group to obtain the biased confidence score. The calculation formula is as follows:

[0064] ;

[0065] in, It prioritizes the degree of trust and normalizes the initially calculated values, mapping them to the (0,1) interval; It is the independent confidence degree of the i-th manifold set; it is a minimal non-zero constant; Indicates the product of multiple terms; prevents zero values; is the dynamic weighting coefficient of the i-th manifold group, calculated using the following formula:

[0066] ;

[0067] in, Let be the accuracy of the diagnosis in the past 100 times for the i-th manifold group, and n be the total number of manifold groups, which is three manifold groups in this application.

[0068] In some embodiments, if a sample belongs to three manifold groups simultaneously, the confidence scores of the three manifold groups for several sample data groups within the time range before and after the real-time data group are obtained. A decision matrix is ​​constructed to obtain the information entropy weights, positive ideal values, and negative ideal values. The distance between each sample and the positive ideal value and the negative ideal value is calculated to obtain the relative proximity as the final confidence score. For the independent confidence scores of n manifold groups at m time sample points, an m×n decision matrix is ​​constructed to obtain the information entropy weights. The calculation formula is as follows:

[0069] ;

[0070] ;

[0071] in, It is the information entropy of the j-th manifold group, used to measure the degree of data dispersion; It is the confidence level of the i-th time sample point in the j-th manifold group; yes The standardized value;

[0072] ;

[0073] in, It is the entropy weight of the j-th manifold group. The smaller The larger;

[0074] ;

[0075] ;

[0076] in, It is the ideal solution, representing the maximum confidence level of the j-th manifold set; It is a negative ideal solution, representing the minimum confidence level of the j-th manifold set;

[0077] ;

[0078] ;

[0079] in, It is the distance between the i-th time sample and the positive ideal solution; It is the distance between the i-th time sample and the negative ideal solution;

[0080] ;

[0081] in, This is the relative proximity to the i-th time sample, used to determine whether it is closer to the positive or negative ideal solution. A relative proximity score closer to 1 indicates a greater distance from the negative ideal solution, meaning it is closer to the positive ideal solution and indicates a high fault risk; conversely, a relative proximity score closer to 0 indicates a greater distance from the positive ideal solution, meaning it is closer to the negative ideal solution and indicates a normal state. The relative proximity scores are generated into a sequence, and the fault types of samples at different time points are statistically analyzed to obtain the closest fault type and its corresponding fault data.

[0082] In this implementation, the trust level refers to the judgment indicator between the current real-time data and the fault. The higher the trust level, the more accurate the judgment, i.e., a fault state; the lower the trust level, the more inaccurate the judgment, i.e., a normal state. Because It is based on the maximum confidence level of each manifold group in historical data (representing the most faulty state). It is the minimum value (representing the most normal state); check the different manifold groups at the current time point, and which one has a higher trust level indicates that there is a problem in the corresponding field.

[0083] By decoupling the characteristics and performing independent analysis when multiple characteristics overlap, samples from several time points of different manifold groups are selected. Here, the time samples selected are the time range before and after the real-time data group, namely the historical time samples and the estimated time samples, with a ratio of 8:2. The estimated time samples are composed of three different estimation results. When estimating, the actual cause of the current failure is assumed first. For example, if it belongs to the first manifold group, the characteristics of the data group in the subsequent time range will be estimated based on the failure characteristics and data characteristics of the first manifold group. The estimated time range is set according to the specific actual situation, and this application does not impose specific restrictions here.

[0084] Preferably, based on the obtained final trust level, multiple threshold levels are pre-set, including normal range, alert state, warning state, and fault state. Each state corresponds to a different threshold range, and the thresholds are dynamically adjusted according to the real-time operating status (real-time load rate, etc.). The degree of conformity between each diagnostic result and the actual situation is recorded, the historical accuracy of each manifold group is updated, and the entropy weight is recalculated to ensure the timeliness of weight allocation. For example, a single manifold trust level greater than 0.7 indicates a yellow warning; a multi-manifold collaborative trust level greater than 0.85 indicates a red warning; and the final decision is made in conjunction with the status of the protection device.

[0085] In this embodiment, by manifold partitioning and decoupling, the complex problem is decomposed into multiple relatively independent sub-problems, and the impact of each factor on the system is analyzed separately, thereby more accurately and quickly determining the cause of the fault and the system status, and ensuring the safe and stable operation of the power system; multidimensional data variables are mapped to a low-dimensional space to generate low-dimensional data variables, reducing the data dimensionality and solving the problems of high computational complexity, difficulty in understanding and visualization of high-dimensional data.

[0086] Preliminary grouping is performed based on the influence weights and correlations of low-dimensional data variables, and manifold groups are divided according to the total group weight ratio to accurately measure the influence of different factors on the dominant path. Based on the manifold group to which the real-time data group belongs, different methods are used to calculate the final confidence level, thereby determining the fault data, including the fault cause, type, occurrence time, impact range and optimal measures, to solve the limitations and inaccuracies of traditional analysis methods, such as high computational complexity and difficulty in understanding.

[0087] Example 2: The above examples improve data processing efficiency and fault identification accuracy through manifold partitioning and decoupling. However, when the first, second, and third manifold groups are all within the normal range, it is impossible to identify abnormal situations caused by the interaction dynamics among them; moreover, when a distribution network fault propagates along electrical connections to the transmission layer, the quantitative prediction effect of propagation path and speed is reduced. This example makes further improvements based on the above.

[0088] The method further includes: S31: If the user belongs to three manifold groups simultaneously, calculate the instantaneous rate of change of the trust level of each manifold group at each time point according to the decision matrix; set the sliding window size to N, and obtain the rate of change values ​​of N time points within the sliding window to form three change matrices; calculate the correlation matrix between the three change matrices based on the rate of change data within the sliding window; perform eigenvalue decomposition based on the correlation matrix, compare the largest eigenvalue with a pre-set interaction threshold, and if the largest eigenvalue is greater than the interaction threshold, mark the current time point as a strong interaction time point, and the largest eigenvalue as the manifold collaboration factor; statistically analyze all strong interaction time points and manifold collaboration factors to establish a time change curve.

[0089] The rate of change is calculated based on the instantaneous rate of change at all time points within the sliding window. Assuming the confidence level of the first manifold group is [0.5, 0.7, 0.6] at three consecutive time points, the rate of change at the middle time point is (0.6 - 0.5) / 2 = 0.05. A positive rate of change indicates an increase in confidence, while a negative value indicates a decrease.

[0090] The change matrix is ​​a 3xN matrix, with each row corresponding to a change rate sequence of a manifold group. The correlation matrix includes a linear correlation matrix and a time-delay correlation matrix. The linear correlation matrix is ​​a 3x3 matrix, and its elements represent the covariance of the change rates between manifold groups, used to quantify the correlation strength between different manifold groups. The change rate is used to reflect the dynamic trend of trust, with the positive or negative sign indicating the direction of change and the absolute value indicating the magnitude of change.

[0091] This embodiment captures linear correlations based on the covariance relationship of rates of change. In this embodiment, the correlation matrix includes a linear correlation matrix and a time-delay correlation matrix. The time-delay correlation matrix provides a nonlinear correlation measurement method based on mutual information. Specifically: the probability distribution of the rate of change sequence for each manifold group is estimated. An adaptive interval discretization method is used to calculate the percentiles (5%, 25%, 50%) of each rate of change sequence. Based on the percentiles, the continuous values ​​are discretized into 6 levels to ensure that each level contains enough data sets to guarantee statistical significance. The joint probability distribution is calculated. For the rate of change sequences of any two manifold groups, a two-dimensional joint probability distribution matrix is ​​obtained based on the joint probability distribution and the rate of change sequences. At the same time, the marginal probability distribution is calculated. The mutual information value is calculated using the KL divergence formula, and the Miller-Madow correction method is used to eliminate the bias caused by the limited sample size. A nonlinear correlation matrix is ​​constructed based on the probability distribution estimation and the mutual information value. A 3×3 nonlinear correlation matrix is ​​constructed, where the matrix elements are the mutual information values ​​between any two manifold groups, and the diagonal elements are kept to be 1 (each variable is fully correlated with itself).

[0092] Interactions in power systems often exhibit time delay effects, necessitating time-delay nonlinear correlation analysis. This includes: calculating time-delay mutual information by performing time-delay mutual information calculations on the rate of change sequences of each pair of manifolds within the time delay range, identifying the time delay value (a specific time value, such as 0.2 seconds) that maximizes the mutual information, reflecting the most significant time delay in the interaction between the two manifolds; establishing a time-delay correlation matrix, which is a matrix composed of elements obtained by multiplying each element of the nonlinear correlation matrix by the time delay value; performing eigenvalue decomposition on the nonlinear correlation matrix to obtain the largest eigenvalue, reflecting the overall nonlinear correlation strength of the system, and obtaining the synergy factor.

[0093] The results of linear correlation matrix and time-delay correlation matrix analysis are fused to obtain manifold synergy factor, which is then fused with fault propagation model to obtain propagation speed and pre-propagation path, determine affected lines and affected areas, and generate fault reports; cross-validation is used to evaluate the accuracy of nonlinear correlation detection, and detection sensitivity is tested by injecting known nonlinear fault modes, and the results of linear and nonlinear analysis are combined.

[0094] In some embodiments, the strong interaction time point refers to a point where the trust levels of the three manifold groups are normal, but the correlation and interaction effects between the data of the three manifold groups exceed the interaction threshold. Emergent failures are highly likely to occur at this time point, hence it is marked. The largest value among the three correlation feature values ​​is used as the manifold co-factor, and combined with all the strong interaction time points and the manifold co-factor, a time variation curve is established to represent the dynamic changes in the interaction relationships between the manifold groups within a certain time range. The time variation curve is plotted with the strong interaction time point on the x-axis and the manifold co-factor on the y-axis. The time range can be dynamically adjusted according to specific circumstances, and each time range includes at least five strong interaction time points.

[0095] S32: Based on the time variation curve, select the manifold group corresponding to the largest manifold synergy factor as the suspected source, calculate the propagation speed and pre-propagation path of the suspected source according to the fault propagation model, determine the affected line and affected area, and generate a fault report.

[0096] The fault report includes the critical equipment identified based on the affected lines and areas, the fault propagation speed, the fault type, the risk level, and the estimated time of occurrence; the estimated time of occurrence refers to the time from the current time, based on the propagation speed, to predict when the fault will reach the critical equipment.

[0097] In some embodiments, calculating the propagation velocity and pre-propagation path of a suspected source based on a fault propagation model includes: calculating the short-circuit current distribution starting from the fault source node based on the grid topology, and identifying affected downstream nodes and lines; performing power flow calculations using an impedance matrix to identify potentially overloaded components; simulating protection signal transmission and calculating the time from the fault source to each protection device by incorporating real-time communication delays and message loss rates; simulating protection actions and circuit breaker operations, incorporating real-time circuit breaker operating times and protection states (normal, non-operation, maloperation); if protection fails to operate, the fault continues to propagate; iteratively simulating fault propagation until no new nodes are affected or the propagation time exceeds the maximum allowable time (e.g., 30 seconds); after iteration, the propagation velocity and pre-propagation path are obtained.

[0098] A fault propagation model is established, comprising: an electrical propagation layer, an information propagation layer, and a control propagation layer; predefined fault propagation model parameters, including fault source nodes, propagation speed, and propagation path; and, based on a determined fault source, calculating the propagation speed and determining the propagation path according to the equipment connection status of relevant lines within the fault propagation model. The propagation speed includes both information transmission speed and control transmission speed.

[0099] The electrical propagation layer, based on the power grid topology, calculates the short-circuit current distribution starting from the fault source node, determining the affected downstream nodes and lines; it uses an impedance matrix for power flow calculation to identify potentially overloaded components. Specifically, it constructs full-network topology data, including: a bus node table (node ​​number, voltage level, geographical location), a line parameter table (line number, starting node, ending node, positive / zero sequence impedance, capacitance to ground), and a transformer parameter table (turns ratio, winding connection method, short-circuit impedance). It establishes an association matrix between nodes and branches, marking the location and current opening / closing status of all circuit breakers. The impedance matrix construction includes: using the node admittance matrix method: calculating the full-network node admittance matrix based on line impedance and transformer parameters, performing equivalent conversion for multi-voltage level networks to form a unified impedance matrix. The admittance matrix undergoes LU decomposition preprocessing to improve computational efficiency. Short-circuit current distribution calculation includes: simulating three-phase short circuits, single-phase grounding, and other fault types for each potential fault source node (nodes corresponding to strong interaction time points). A unit fault current is injected at the fault point, the node voltage equation is solved, the current in each branch is calculated, a fault current distribution map is generated, and the overcurrent multiple and direction of each line are marked. Then, the order of influence is determined, including: based on electrical distance ranking, calculating the electrical distance (impedance magnitude) from the fault point to each node, and ranking the influence order from smallest to largest distance. Overcurrent protection coordination is analyzed, and the sequence of protection actions is determined based on current amplitude and time characteristics.

[0100] The information propagation layer simulates protection signal transmission, considering real-time communication delay and message loss rate, and calculates the time it takes for the signal to travel from the fault source to each protection device, i.e., the information propagation speed. Specifically, the communication system architecture is extracted, and communication link information between protection devices (fiber optic / microwave channels, routing nodes, transmission protocols) is collected. A communication topology diagram is constructed, with nodes representing protection devices and communication gateways, edges representing communication links, and transmission medium characteristics (bandwidth, bit error rate) labeled. The basic transmission delay and protocol processing delay are calculated, a retransmission mechanism is constructed, and the message loss detection time, retransmission count, and total retransmission time are determined. The comprehensive propagation time is calculated, and based on the comprehensive propagation time and transmission distance, the information transmission speed is obtained, which depends on the real-time communication delay and message loss rate (i.e., the first manifold group).

[0101] The control propagation layer simulates protection actions and circuit breaker operations, considering real-time circuit breaker operating time and protection status (normal, failure to operate, maloperation). If the protection fails to operate, the fault continues to propagate. Specifically, characteristic quantities are extracted, and action characteristics (amplitude, phase, harmonic content) are extracted from the fault current waveform. The response times of overcurrent protection, differential protection, and directional protection are calculated, and fixed delays are set according to the protection type. The mechanical operating time and arc extinguishing time are calculated based on the circuit breaker operation to obtain the total opening time. The control transmission speed is calculated, which depends on the circuit breaker operating time and protection response time (i.e., the second manifold group and the third manifold group). The total propagation speed and total propagation time are obtained based on the information transmission speed and control transmission speed.

[0102] A unified timeline is established for the three sub-layer models, with the fault initiation point as the zero point, synchronizing the time scales of the three models. A variable step size algorithm is set, with an initial step size of 0.1ms and a maximum step size of 5ms. The electrical layer outputs a set of nodes affected by the initial fault. The information layer filters the nodes in the initial fault-affected node set, calculating the information arrival time and removing nodes not covered due to communication failure. The control layer filters the remaining nodes, calculating the protection action time and removing nodes that could not be isolated due to protection failure, ultimately forming an effective propagation path. Based on the above data analysis, a chain reaction analysis is performed, iteratively simulating fault propagation. The iterative mechanism includes: calculating the impact of the initial fault point, treating affected nodes as new fault sources, and recalculating propagation. The termination condition is until no new nodes are affected or the propagation time exceeds the maximum allowable time, ultimately determining the propagation path and impact range. Parameter calibration methods include retransmission time calibration: grouping historical communication data according to message loss rate and statistically analyzing actual retransmission times; and protection response time calibration: comparing theoretical and actual action times, calculating correction coefficients, and recalculating calibration coefficients using a sliding window method (window length 30 days).

[0103] In this embodiment, the time variation curve is generated by analyzing all devices in a certain communication line. In specific cases, the analysis can be based on the power data of a certain device in different time ranges, or by analyzing and generating multiple power devices in the same area. The specific selection is made dynamically according to the actual situation.

[0104] In this embodiment, the instantaneous rate of change and correlation matrix are calculated; the correlation matrix between the change matrices is calculated, including the linear correlation matrix and the time-delay correlation matrix; the linear correlation matrix quantifies the correlation strength between different manifold groups, and the time-delay correlation matrix provides a nonlinear correlation measurement method based on mutual information. The calculation of manifold synergy factors and the establishment of time-varying curves can accurately identify abnormal situations arising from the interaction of the three manifold groups, and the dynamic changes in the interaction relationship are intuitively displayed through the time-varying curves; this solves the problem that the three manifold groups themselves are normal but their interactions are abnormal and cannot be identified.

[0105] Based on time-delay mutual information and the establishment of a time-delay correlation matrix, the results of correlation matrix analysis are integrated, and a fault propagation model is constructed. The manifold synergy factor is fused with the fault propagation model to obtain the propagation speed and pre-propagation path, determine the affected lines and affected areas, and generate fault reports. This improves the quantitative prediction effect of fault propagation paths and speeds, and can generate detailed fault reports, providing comprehensive and accurate information for fault handling.

[0106] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for diagnosing communication faults in a power distribution network, characterized in that, include: S1: Obtain multidimensional data variables that characterize the power grid operating status, divide the multidimensional data variables into several data groups and form a neighborhood graph; Calculate the manifold distance between two adjacent edge neighbor groups in the neighborhood graph and construct a similarity matrix; determine the low-dimensional embedding dimension and low-dimensional spatial coordinate axes based on the eigenvalues ​​and eigenvectors of the similarity matrix, and then obtain the low-dimensional data variables; S2: Obtain the initial weight of each variable in the low-dimensional data variables and calculate the correlation between any two variables; perform preliminary grouping of the low-dimensional data variables according to the initial weight and correlation, and divide the manifold groups according to the total weight ratio of each group; S3: Obtain real-time data sets, calculate their coordinate positions in the low-dimensional space coordinate axes based on the similarity matrix and determine their manifold group, calculate the final confidence level based on the manifold group, and determine the fault data based on the final confidence level. If the manifold group to which it belongs is a single manifold group, then the trust score of the corresponding manifold group is calculated as the final trust score. If a manifold belongs to two manifold groups, the independent confidence level of each manifold group is calculated, and the dynamic weighting coefficient is determined based on the historical diagnostic accuracy of each manifold group to obtain the biased confidence level as the final confidence level. If it belongs to three manifold groups at the same time, then obtain the confidence scores of the three manifold groups of several sample data groups in the time range before and after the real-time data group, construct the decision matrix, and obtain the information entropy weight, positive ideal value and negative ideal value. Calculate the distance between each sample and the positive ideal value and the negative ideal value, and use the relative proximity as the final confidence level.

2. The method for diagnosing communication faults in a power distribution network as described in claim 1, characterized in that, The method further includes: S31: If it belongs to three manifold groups at the same time, calculate the instantaneous rate of change of the trust degree of each manifold group at each time point according to the decision matrix; set the sliding window size to N, and obtain the rate of change values ​​of N time points within the sliding window to form three change matrices; Calculate the correlation matrix between the three change matrices based on the rate of change data within the sliding window; Eigenvalue decomposition is performed based on the correlation matrix. The largest eigenvalue is compared with a pre-set interaction threshold. If the largest eigenvalue is greater than the interaction threshold, the current time point is marked as a strong interaction time point, and the largest eigenvalue is the manifold synergy factor. All strong interaction time points and manifold synergy factors are counted to establish a time variation curve. S32: Based on the time variation curve, select the manifold group corresponding to the largest manifold synergy factor as the suspected source, calculate the propagation speed and pre-propagation path of the suspected source according to the fault propagation model, determine the affected line and affected area, and generate a fault report.

3. The method for diagnosing communication faults in a power distribution network as described in claim 1, characterized in that, Based on the eigenvalues ​​and eigenvectors of the similarity matrix, the low-dimensional embedding dimension and low-dimensional spatial coordinate axes are determined, thereby obtaining low-dimensional data variables. This includes: calculating the sum of all eigenvalues, calculating the proportion of each eigenvalue to the sum as the contribution rate of each eigenvalue, arranging all eigenvalues ​​in descending order, and accumulating the contribution rates sequentially from the first eigenvalue to obtain the cumulative sum, with a pre-set contribution threshold; when the cumulative sum first exceeds the contribution threshold, the accumulation stops, and the number of eigenvalues ​​participating in the accumulation is recorded. Set the number of feature values ​​to the low-dimensional embedding dimension, and use the corresponding feature vectors as low-dimensional feature vectors to construct low-dimensional space coordinate axes. Perform an inner product operation between each data set and the i-th low-dimensional feature vector to obtain its coordinate value on the i-th coordinate axis in the low-dimensional space; determine the low-dimensional embedding point based on the coordinate value, map the data set to the corresponding low-dimensional embedding point, and form a low-dimensional data variable after all embeddings are completed.

4. The method for diagnosing communication faults in a power distribution network as described in claim 1, characterized in that, The initial weights are used to measure the degree of influence of each variable in the data set on communication stability; Calculating the correlation between any two variables includes: obtaining low-dimensional data variables, performing data cleaning and normalization, generating a linear matrix based on the Pearson correlation coefficient, and calculating the linear correlation; generating a nonlinear matrix based on mutual information and calculating the nonlinear correlation; weighted summing of the linear and nonlinear correlations to calculate the comprehensive correlation; determining the correlation threshold of the comprehensive correlation based on the initial weights, and performing preliminary grouping based on the correlation threshold.

5. The method for diagnosing communication faults in a power distribution network as described in claim 1, characterized in that, The manifold groups are divided according to the total weight ratio of each group, including: for all data groups in each group, the weight ratio of each variable in the data group is calculated and compared with the corresponding set threshold to divide the manifold groups; The initial grouping includes three groups: the first group includes communication delay and message loss rate; the second group includes current overload rate and circuit breaker operating time; the third group includes main protection status and backup protection status; and the manifold group includes the first manifold group, the second manifold group, and the third manifold group.

6. The method for diagnosing communication faults in a power distribution network as described in claim 5, characterized in that, Calculate the weight of communication delay and message loss rate in each data group among all variable combinations. If the sum of the weights is greater than the first threshold, then the corresponding data group is assigned to the first manifold group. Calculate the weight of current overload rate and circuit breaker operating time in all variable combinations within each data group. If the sum of the weights is greater than the second threshold, then the corresponding data group is assigned to the second manifold group. Calculate the weight ratio of the primary protection status and backup protection status in all variable combinations within each data group. If the sum of the weight ratios is greater than the third threshold, then the corresponding data group is assigned to the third manifold group. The first threshold is less than the second threshold, and the second threshold is less than the third threshold.

7. The method for diagnosing communication faults in a power distribution network as described in claim 2, characterized in that, The correlation matrix includes a linear correlation matrix and a time-delay correlation matrix; the matrix elements of the linear correlation matrix are the covariance of the rates of change among the manifold groups, which are used to quantify the correlation strength between different manifold groups. Establishing a time-delay correlation matrix includes: estimating the probability distribution of the rate of change sequence for each manifold group; constructing a nonlinear correlation matrix based on the probability distribution estimation and mutual information values; the matrix elements are the mutual information values ​​between any two manifold groups, and the diagonal elements are 1; Calculate the time-delay mutual information. For each pair of manifold change rate sequences, calculate the time-delay mutual information within the time delay range. Find the time delay value that maximizes the mutual information. Establish the time-delay correlation matrix based on the nonlinear correlation matrix and the time delay value. The results of the linear correlation matrix and the time-delay correlation matrix analysis are combined to obtain the manifold synergy factor; The probability distribution estimation includes: using an adaptive interval discretization method to calculate the percentile of each rate of change sequence, discretizing the continuous values ​​into 6 levels based on the percentiles, calculating the joint probability distribution, and for any two manifold groups of rate of change sequences, obtaining a two-dimensional joint probability distribution matrix and marginal probability distribution based on the joint probability distribution and the rate of change sequences.

8. The method for diagnosing communication faults in a power distribution network as described in claim 1, characterized in that, The multidimensional data variables include communication delay, message loss rate, main protection status, backup protection status, circuit breaker operating time, and current overload rate. The shortest path distance between two adjacent edge neighbor groups is calculated as the manifold distance on the neighborhood graph according to the shortest path algorithm. Dividing multidimensional data variables into several data groups and forming a neighborhood graph includes: pre-determining a core data group, finding several edge neighbor groups of the core data group in the high-dimensional space, and connecting several edge neighbor groups with the core data group as the center to form a neighborhood graph; the neighborhood graph includes a core data group and several edge neighbor groups; the edge neighbor groups are the remaining data groups that are spatially closest to the data group in the high-dimensional space region.

9. The method for diagnosing communication faults in a power distribution network as described in claim 2, characterized in that, The propagation speed and pre-propagation path of the suspected source are calculated based on the fault propagation model, including: Based on the power grid topology, starting from the fault source node, the short-circuit current distribution is calculated to determine the affected downstream nodes and lines; power flow calculation is performed using the impedance matrix to identify components that may be overloaded; protection signal transmission is simulated, and the time from the fault source to each protection device is calculated by combining real-time communication delay and message loss rate; protection actions and circuit breaker operations are simulated, combining real-time circuit breaker action time and protection status; fault propagation is iteratively simulated until no new nodes are affected or the propagation time exceeds the maximum allowable time. After the iteration is completed, the propagation speed and pre-propagation path are obtained.

10. The method for diagnosing communication faults in a power distribution network as described in claim 1, characterized in that, Establishing a fault propagation model includes: the fault propagation model includes an electrical propagation layer, an information propagation layer, and a control propagation layer; pre-defining fault propagation model parameters, including fault source nodes, propagation speed, and propagation path; based on the determined fault source, calculating the propagation speed and determining the propagation path in the fault propagation model according to the equipment connection status of the relevant lines; the propagation speed includes information transmission speed and control transmission speed.

Images