Power distribution terminal joint debugging data intelligent analysis and diagnosis method and system, and related equipment

Abstract

This application relates to the field of power distribution terminal data analysis, and particularly to a method, system, and related equipment for intelligent analysis and diagnosis of power distribution terminal commissioning data. The method includes: collecting commissioning parameters during the power distribution terminal commissioning process; performing distributed computation processing on the commissioning parameters based on edge computing to generate structured data; constructing a node feature matrix from the structured data, iteratively updating the correlation weights between nodes through a multi-layer attention mechanism, and outputting feature-enhanced correlation data; constructing an integer programming model including equipment state variables and commissioning link variables using the correlation data as constraints, and solving for a set of potential abnormal variables; calculating the mean squared error between the set of potential abnormal variables and the standard variable set, and assigning weights according to parameter importance to obtain a comprehensive evaluation coefficient; and converting the comprehensive evaluation coefficient into a visualized diagnostic result. This application significantly improves the intelligence level and practicality of power distribution terminal commissioning data analysis and diagnosis.

Description

Technical Field

[0001] This application relates to the field of power distribution terminal data analysis technology, and in particular to a method, system and related equipment for intelligent analysis and diagnosis of power distribution terminal commissioning data. Background Technology

[0002] The statements in this section are merely background information related to this application and do not necessarily constitute prior art.

[0003] With the continuous advancement of smart grid construction, distribution terminals, as key nodes in power grid operation monitoring and control, play a crucial role in ensuring the safe and stable operation of the power grid through efficient and accurate commissioning processes. Distribution terminal commissioning involves real-time interaction of multiple parameters such as voltage, current, and power. It involves a large number of devices, complex links, and high dynamism, posing challenges to traditional centralized data processing methods, such as large transmission delays and significant bandwidth pressure.

[0004] Existing technologies for processing and diagnosing distribution terminal commissioning data have two significant drawbacks. Firstly, their data correlation analysis capabilities are insufficient, often relying on single-parameter threshold judgments or simple statistical methods. This fails to fully capture the complex relationships between devices and parameters, resulting in low sensitivity to latent anomalies and a high likelihood of missed diagnoses. Secondly, the comprehensive evaluation mechanism for diagnostic results is inadequate, often depending on a single indicator for anomaly determination. It fails to consider the weight differences of different parameters during commissioning, making it difficult to comprehensively reflect the overall status of the commissioning system. This results in insufficient reliability and specificity of diagnostic results, making it difficult to meet the refined operation and maintenance needs of complex distribution networks. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, this application provides a method, system, storage medium, and electronic device for intelligent analysis and diagnosis of distribution terminal commissioning data. This method can achieve real-time and efficient data processing, improve the depth and accuracy of anomaly identification, and ensure the reliability and availability of diagnostic results, thereby significantly improving the intelligence level and practicality of distribution terminal commissioning data analysis and diagnosis.

[0006] To achieve the above objectives, this application provides the following technical solution:

[0007] Firstly, a method for intelligent analysis and diagnosis of distribution terminal commissioning data is provided, the method comprising:

[0008] Real-time acquisition of commissioning parameters during the commissioning process of power distribution terminals;

[0009] The collected joint debugging parameters are processed in a distributed manner based on edge computing to generate structured data;

[0010] The structured data is constructed into a node feature matrix, and the association weights between nodes are iteratively updated through the multi-layer attention mechanism of the graph attention network to output the feature-enhanced association data.

[0011] Using the associated data as constraints, an integer programming model including equipment status variables and joint debugging link variables is constructed, and the integer programming model is iteratively solved to obtain a set of potential abnormal variables;

[0012] Calculate the mean squared error between the set of potential outlier variables and the set of standard variables, and assign weights according to the importance of the parameters to obtain a comprehensive evaluation coefficient;

[0013] The comprehensive evaluation coefficients are then converted into visual diagnostic results.

[0014] Furthermore, the joint debugging parameters include voltage, current, power, frequency, switch status, and communication signal strength parameters; the distributed computing processing of the collected joint debugging parameters based on edge computing to generate structured data includes: receiving the collected joint debugging parameters, processing them into segments according to a preset time window, performing distributed parallel computing on each segment of data, extracting parameter change rate and fluctuation characteristics, and generating structured data.

[0015] Furthermore, the step of constructing the structured data into a node feature matrix, iteratively updating the association weights between nodes through a multi-layer attention mechanism of a graph attention network, and outputting feature-enhanced associated data includes: constructing the structured data into a graph structure with devices as nodes and parameter correlations as edges, learning the dependencies between nodes through a multi-layer attention mechanism of a graph attention network, updating the node feature representations, and outputting feature-enhanced associated data.

[0016] Furthermore, the process of constructing a node feature matrix from the structured data and iteratively updating the inter-node association weights through a multi-layer attention mechanism of a graph attention network to output enhanced associated data includes: converting voltage and current parameters in the structured data into high-dimensional feature vectors; assigning different dimensional weights to different parameters through a linear transformation matrix to form initial node features; converting the connection status between devices in the joint commissioning link into adjacency matrix element values ​​based on the physical connection relationship and parameter correlation of the distribution terminal, setting the element values ​​corresponding to highly correlated device nodes to non-zero values; performing nonlinear transformation on the node features using a multi-layer perceptron; calculating initial weights by comparing the similarity of different node features; and then normalizing the weights of different node pairs using a softmax function; weighted summing of neighboring node features according to the magnitude of the attention weights; and generating new node features by combining the node's own features with an activation function to complete one feature aggregation iteration.

[0017] Furthermore, the process of constructing an integer programming model, including equipment state variables and joint commissioning link variables, using the associated data as constraints, and iteratively solving the integer programming model to obtain a set of potential abnormal variables, includes: converting the equipment operating status and link connectivity during the joint commissioning process of the power distribution terminal into binary decision variables and clarifying the physical meaning of the variable values; converting the voltage fluctuation range and power balance relationship into mathematical inequalities based on the allowable range of joint commissioning parameters and the logical relationship between equipment, forming a set of constraints; constructing an objective function with the goal of minimizing the total number of abnormal variables or maximizing the accuracy of abnormal identification, and using the degree of abnormal influence of different variables as the coefficients of the objective function; and solving the constructed integer programming model using the branch and bound method, gradually narrowing the search range by continuously dividing the solution space and calculating the lower bound value, and finally obtaining the optimal set of abnormal variables that satisfies the constraints.

[0018] Furthermore, the calculation of the mean squared error between the potential abnormal variable set and the standard variable set, and the allocation of weights according to parameter importance to obtain a comprehensive evaluation coefficient, includes: storing different parameter values ​​from historical normal commissioning processes, classifying them according to distribution terminal type and commissioning scenario to form standard variable sets for different scenarios; performing a difference operation between each parameter value in the potential abnormal variable set and the average value of the corresponding parameter in the standard variable set to obtain the absolute deviation value of each parameter; squaring the deviation values ​​of all parameters and then calculating the arithmetic mean to obtain the initial mean squared error evaluation coefficient; assigning different weights to the deviation values ​​of different parameters according to their importance in commissioning, recalculating the weighted mean squared error value as the final comprehensive evaluation coefficient.

[0019] Furthermore, the process of converting the comprehensive evaluation coefficient into a visual diagnostic result includes: comparing the comprehensive evaluation coefficient with a preset threshold; if the threshold is exceeded, it is marked as abnormal; otherwise, it is judged as normal; integrating the abnormal judgment result with the corresponding parameter change curve and abnormal variable location information, converting it into a visual diagnostic result and outputting it.

[0020] Furthermore, the update process of the node feature matrix satisfies:

[0021]

[0022] in, For the first Layer nodes eigenvectors, For activation function, For nodes The set of neighboring nodes, For the first Layer nodes For nodes Attention weights For the first The weight matrix of the layer, For the first Layer nodes eigenvectors.

[0023] Furthermore, when iteratively updating the association weights between nodes, the weight calculation satisfies:

[0024]

[0025] in, For nodes For nodes Attention weights, with LeakyReLU as the activation function. For attention weight vectors, This is the weight matrix. They are nodes eigenvectors, This is a vector concatenation operation. For nodes The set of neighboring nodes, where exp represents the natural exponential function and T represents the transpose of the matrix.

[0026] Furthermore, the feature enhancement process satisfies:

[0027]

[0028] in, For the first Nodes in the next iteration Enhanced features, For the first In the nth iteration The weights of each convolutional kernel, For the first Each convolution operation, For the first Nodes in the next iteration The original characteristics, This represents the number of convolution kernels.

[0029] Furthermore, the layer number adjustment of the multi-layer attention mechanism satisfies:

[0030]

[0031] in, To determine the optimal number of layers, For the first Layer nodes The output characteristics, For nodes Target characteristics, for Norm, The total number of nodes. Variables that represent the minimum value The value of .

[0032] Furthermore, the objective function and constraints of the integer programming model are as follows:

[0033]

[0034] in, For device state variables, For the joint debugging link variables, This is the equipment anomaly cost coefficient. This is the link anomaly cost coefficient. For equipment With Link The correlation coefficient, For link The autocorrelation coefficient, For equipment The anomaly detection threshold For the total number of devices, This represents the total number of links.

[0035] Furthermore, the process of solving the set of potential abnormal variables satisfies:

[0036]

[0037] in, Used as an identifier for abnormal variables (1 indicates that it belongs to the abnormal set, and 0 indicates that it does not belong to the abnormal set). For variables Abnormal contribution For variables With variables The mutual exclusion coefficients, For variables The abnormal tolerance threshold This represents the total number of variables.

[0038] Furthermore, the constraint update of the integer programming model satisfies:

[0039]

[0040] in, For variables With variables The constraint coefficient correction amount, As a correction factor, The current value of the variable. , This represents the historical average value of the variable.

[0041] Furthermore, the calculation of the comprehensive evaluation coefficient satisfies:

[0042]

[0043] Wherein, MSE is the mean squared error comprehensive evaluation coefficient. For the total number of variables, For the first One potential outlier variable value, For the first Standard variable values.

[0044] Furthermore, the construction of the standard variable set satisfies:

[0045]

[0046] in, For the first One standard variable value, This represents the historical normal sample size. For the first The weight coefficients of each sample, For the first In the nth sample Each variable value.

[0047] Furthermore, the correction of the comprehensive evaluation coefficient satisfies:

[0048]

[0049] in, This is the corrected mean squared error comprehensive evaluation coefficient, where MSE is the original mean squared error value. For correction factor, For the first One potential outlier variable value, For the first One standard variable value, This represents the total number of variables.

[0050] Secondly, a smart analysis and diagnosis system for distribution terminal commissioning data is also provided. The system includes: a data acquisition module for real-time acquisition of commissioning parameters during the commissioning process of the distribution terminal; an edge computing processing module for distributed computing processing of the acquired commissioning parameters based on edge computing to generate structured data; a graph attention network analysis module for constructing a node feature matrix from the structured data, iteratively updating the association weights between nodes through a multi-layer attention mechanism of the graph attention network, and outputting feature-enhanced association data; a hybrid integer programming diagnosis module for constructing an integer programming model including equipment state variables and commissioning link variables using the association data as constraints, and iteratively solving the integer programming model to obtain a set of potential abnormal variables; a mean square error comprehensive evaluation module for calculating the mean square error between the set of potential abnormal variables and the standard variable set, and assigning weights according to parameter importance to obtain a comprehensive evaluation coefficient; and a result output module for converting the comprehensive evaluation coefficient into a visualized diagnostic result.

[0051] Based on the same inventive concept, this application also provides a computer-readable storage medium storing one or more programs, which, when executed, can realize the intelligent analysis and diagnosis method for power distribution terminal commissioning data as described above.

[0052] Based on the same inventive concept, this application also provides an electronic device, including a processor, a communication interface, a computer-readable storage medium as described above, and a communication bus; wherein the processor, the communication interface, and the computer-readable storage medium communicate with each other through the communication bus; the processor is used to execute a program stored in the computer-readable storage medium.

[0053] Compared with the prior art, the beneficial effects of this application are as follows:

[0054] By performing distributed computation processing on the collected joint commissioning parameters based on edge computing, massive and high-frequency joint commissioning parameters (voltage, current, power, etc.) can be processed locally at the terminal side close to the data source. By combining graph attention networks and mixed integer programming models, it is possible to deeply mine and accurately locate implicit and complex anomalies. By calculating the mean square error and assigning weights according to the importance of parameters to obtain a comprehensive evaluation coefficient, a quantitative, reliable, and intuitive diagnostic result can be output. This significantly improves the intelligence level and practicality of power distribution terminal joint commissioning data analysis and diagnosis.

[0055] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings.

[0056] The present application will be further described below with reference to the accompanying drawings. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of this application 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 of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a flowchart illustrating an embodiment of the intelligent analysis and diagnosis method for power distribution terminal commissioning data according to this application;

[0059] Figure 2 This is a schematic diagram of the structure of a power distribution terminal joint commissioning data intelligent analysis and diagnosis system according to an embodiment of this application;

[0060] Figure 3 This is a schematic diagram of the overall process of a method for intelligent analysis and diagnosis of power distribution terminal commissioning data according to an embodiment of this application;

[0061] Figure 4 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0063] like Figure 1 As shown, one embodiment of this application provides a method for intelligent analysis and diagnosis of distribution terminal commissioning data (i.e., a method for intelligent analysis and diagnosis of distribution terminal commissioning data based on edge computing), the method comprising:

[0064] Real-time acquisition of commissioning parameters during the commissioning process of power distribution terminals;

[0065] The collected joint debugging parameters are processed in a distributed manner based on edge computing to generate structured data;

[0066] The structured data is constructed into a node feature matrix, and the association weights between nodes are iteratively updated through the multi-layer attention mechanism of the graph attention network to output the feature-enhanced association data.

[0067] Using the associated data as constraints, an integer programming model including equipment status variables and joint debugging link variables is constructed, and the integer programming model is iteratively solved to obtain a set of potential abnormal variables;

[0068] Calculate the mean squared error between the set of potential outlier variables and the set of standard variables, and assign weights according to the importance of the parameters to obtain a comprehensive evaluation coefficient;

[0069] The comprehensive evaluation coefficients are then converted into visual diagnostic results.

[0070] The above technical solution utilizes edge computing to perform distributed computation processing on the collected joint commissioning parameters, enabling the local processing of massive, high-frequency joint commissioning parameters (voltage, current, power, etc.) at the terminal side close to the data source. By combining graph attention networks and mixed integer programming models, it can achieve in-depth mining and precise location of implicit and complex anomalies. By calculating the mean square error and assigning weights according to the importance of parameters to obtain a comprehensive evaluation coefficient, it can ultimately output a quantitative, reliable, and intuitive diagnostic result. This solution significantly improves the intelligence level and practicality of power distribution terminal joint commissioning data analysis and diagnosis.

[0071] As a preferred technical solution, the joint debugging parameters include voltage, current, power, frequency, switch status, and communication signal strength parameters; the distributed computing processing of the collected joint debugging parameters based on edge computing to generate structured data includes: receiving the collected joint debugging parameters, processing them into segments according to a preset time window, performing distributed parallel computing on each segment of data, extracting parameter change rate and fluctuation characteristics, and generating structured data.

[0072] As a preferred technical solution, the method of constructing the structured data into a node feature matrix, iteratively updating the association weights between nodes through the multi-layer attention mechanism of the graph attention network, and outputting the feature-enhanced association data includes: constructing the structured data into a graph structure with devices as nodes and parameter correlations as edges, learning the dependencies between nodes through the multi-layer attention mechanism of the graph attention network, updating the node feature representation, and outputting the feature-enhanced association data.

[0073] As a preferred technical solution, the process of converting the comprehensive evaluation coefficient into a visual diagnostic result includes: comparing the comprehensive evaluation coefficient with a preset threshold; if the threshold is exceeded, it is marked as abnormal; otherwise, it is judged as normal; integrating the abnormal judgment result with the corresponding parameter change curve and abnormal variable location information, converting it into a visual diagnostic result and outputting it.

[0074] This application embodiment is based on edge computing technology, which enables data processing and analysis on the terminal side. By adopting intelligent algorithms such as graph attention networks and mixed integer programming, it provides effective technical support for in-depth mining and anomaly diagnosis of joint debugging data.

[0075] like Figure 2 As shown, one embodiment of this application also provides a power distribution terminal commissioning data intelligent analysis and diagnosis system (i.e., a power distribution terminal commissioning data intelligent analysis and diagnosis system based on edge computing), the system comprising:

[0076] The data acquisition module is used to collect the commissioning parameters of the power distribution terminal in real time during the commissioning process.

[0077] The edge computing processing module is used to perform distributed computing processing on the collected joint debugging parameters based on edge computing to generate structured data;

[0078] The graph attention network analysis module is used to construct the structured data into a node feature matrix, iteratively update the association weights between nodes through the multi-layer attention mechanism of the graph attention network, and output the feature-enhanced association data.

[0079] The hybrid integer programming diagnostic module is used to construct an integer programming model including equipment state variables and joint debugging link variables using the associated data as constraints, and to iteratively solve the integer programming model to obtain a set of potential abnormal variables.

[0080] The mean square error comprehensive evaluation module is used to calculate the mean square error between the set of potential outlier variables and the set of standard variables, and to obtain the comprehensive evaluation coefficient by assigning weights according to the importance of the parameters.

[0081] The results output module is used to convert the comprehensive evaluation coefficients into visual diagnostic results.

[0082] The aforementioned intelligent analysis and diagnosis system for distribution terminal commissioning data improves the real-time performance of data processing through edge computing, mines correlations through graph attention networks, locates anomalies using mixed integer programming, and enhances reliability by combining mean square error assessment. It is suitable for distribution terminal commissioning analysis and diagnosis.

[0083] To make the above-mentioned intelligent analysis and diagnosis methods and systems for distribution terminal commissioning data clearer, further explanation is provided below.

[0084] The data acquisition module is connected to the power distribution terminal and is used to acquire voltage, current, power, frequency, switch status and communication signal strength parameters in real time during the power distribution terminal commissioning process.

[0085] Specifically, the data acquisition module is the foundation for the system to obtain raw information, and its technical parameters directly affect the accuracy of subsequent analysis. This module supports voltage acquisition ranging from 0-380V, current acquisition ranging from 0-500A, power measurement accuracy up to ±0.5%, frequency acquisition ranging from 45-55Hz, switch status recognition response time not exceeding 10ms, and communication signal strength acquisition resolution of 1dBm. These parameters cover common electrical and status quantities encountered during the commissioning of power distribution terminals, comprehensively reflecting the terminal's operating status. Its significance lies in providing complete and accurate raw data for the entire system, serving as a prerequisite for subsequent edge computing processing, network analysis, diagnostic evaluation, and other stages. If the acquired data is biased or missing, it will directly affect the system's judgment of the terminal's commissioning status.

[0086] In practical implementation, the data acquisition module acquires data through sensors deployed at key nodes of the power distribution terminal. Voltage sensors utilize the Hall effect principle to monitor three-phase voltage values ​​in real time, sampling every 50ms and converting the data into digital signals. Current sensors employ Rogowski coil technology to sense AC current in the line, with a sampling frequency consistent with the voltage sensors to ensure time synchronization of voltage and current data. Power parameters are calculated in real time from voltage and current data, updated every 100ms. Frequency parameters are calculated by detecting the zero-crossing point of the voltage signal, with an accuracy controlled within ±0.01Hz. Switch status is acquired through auxiliary contact signals connected to the terminal; when a switch status changes, the module identifies the status and records the change time within 10ms. Communication signal strength is acquired through the terminal's built-in wireless module, obtaining the current signal strength value every 200ms, ranging from -50dBm to -110dBm, used to determine the communication quality between the terminal and the dispatch center. All acquired data is transmitted to the edge computing processing module via an internal bus, with CRC checks used during transmission to ensure data integrity.

[0087] The edge computing processing module is connected to the data acquisition module, and performs distributed computing processing on the acquired parameters to generate structured data.

[0088] Specifically, the edge computing processing module undertakes the initial processing of the raw collected data. Its core technical parameters include a data processing latency of no more than 500ms, a maximum supported concurrent data volume of 1000 records / second, a data storage capacity of 10GB, and the ability to expand the number of distributed computing nodes from 1 to 10 depending on the scale of the terminals. This module uses a low-power processor with an operating temperature range of -40℃ to 70℃, adapting to the complex environment of the power distribution terminal. Its significance lies in reducing the amount of data transmitted to the cloud by processing data locally at the terminal, thus reducing network bandwidth pressure and accelerating data processing speed. This provides a guarantee for real-time analysis and diagnosis, avoiding the latency issues that may occur in centralized processing.

[0089] In practice, after receiving the raw data from the data acquisition module, the edge computing processing module first cleans the data, removing outliers that are clearly outside the reasonable range, such as data with voltages exceeding 400V or falling below 0V. Then, the cleaned data undergoes format conversion, unifying the heterogeneous data output from different sensors into JSON format, which includes fields such as data type, acquisition time, numerical value, and sensor number. Next, according to the regional division of the power distribution terminals, the data is distributed to different distributed computing nodes, each responsible for processing terminal data within a specific area. The computing nodes extract time-domain features from the data, including the maximum, minimum, and average voltage values ​​every 5 minutes, the effective value of the current, and the power fluctuation amplitude. Simultaneously, the switch state change data is sequenced, recording the time points and durations of state changes. The processed structured data is stored in a local cache, retaining data from the most recent 24 hours; data exceeding this period is compressed and backed up to external storage. Finally, the module transmits the structured data to the graph attention network analysis module at preset time intervals (e.g., 100ms).

[0090] The graph attention network analysis module is connected to the output of the edge computing processing module. It constructs structured data into a node feature matrix, iteratively updates the association weights between nodes through a multi-layer graph attention mechanism, and outputs feature-enhanced association data.

[0091] Specifically, the key technical parameters of the graph attention network analysis module include a network layer count of 3-5 layers, a feature dimension of 64-256 for each node, a computational precision of 6 decimal places for the attention mechanism, 1000-5000 model training iterations, and a processing time of no more than 2 seconds per iteration. This module supports a maximum of 1000 nodes, capable of covering medium to large-scale power distribution terminal commissioning networks. Its significance lies in its ability to fully explore the complex relationships between various devices and parameters in a power distribution terminal by constructing a graph structure model. This overcomes the limitation of traditional analysis methods that ignore device correlations, improves the ability to extract implicit correlation features, and provides richer feature evidence for subsequent anomaly diagnosis.

[0092] During implementation, the graph attention network analysis module receives structured data from the edge computing processing module and first constructs the graph structure. Using devices in the power distribution terminal (such as circuit breakers, transformers, and switchgear) as nodes, and the physical connections and parameter correlations between devices as edges, an initial graph model is built. The node feature vector consists of the time-domain features of parameters such as voltage, current, power, and frequency corresponding to the device, with a dimension of 64 dimensions determined by the number of parameters. Next, the network performs multi-layer attention calculations. The first layer performs a linear transformation on the node features, mapping them to a new space using a weight matrix. Then, it calculates the attention weights between each node and its neighbors; the weights depend on the similarity of the node features—higher similarity results in larger weights. In the second layer, the features of neighboring nodes are weighted and summed based on the attention weights calculated in the first layer. Combined with its own node features, a new node feature vector is generated through an activation function, increasing the dimension to 128 dimensions. The third layer repeats the above process, further strengthening the correlation features between nodes and outputting 256-dimensional enhanced correlation data. Throughout the computation process, the network continuously adjusts the weight matrix through backpropagation until the number of iterations reaches 3000 or the loss function value falls below a preset threshold (e.g., 0.001). Finally, the module transmits the processed correlation data to the mixed-integer programming diagnostic module.

[0093] The input of the hybrid integer programming diagnostic module is connected to the output of the graph attention network analysis module. Using associated data as constraints, an integer programming model containing equipment state variables and joint debugging link variables is constructed, and the potential abnormal variable set is obtained by solving the model.

[0094] Specifically, the core parameters of the mixed-integer programming diagnostic module include a maximum of 500 variables (300 equipment status variables and 200 joint commissioning link variables), 1000-2000 constraints, a model solution time of no more than 10 seconds, and a solution accuracy of 4 decimal places. The module supports objective function types including minimizing the total number of abnormal variables and maximizing anomaly identification coverage, which can be selected according to the joint commissioning requirements. Its significance lies in the fact that by constructing a mathematical programming model, it can accurately locate potential abnormal variables in the power distribution terminal joint commissioning process under multiple constraints, overcoming the limitations of traditional diagnostic methods that rely on experience-based judgment or single parameter thresholds, and improving the scientific rigor and accuracy of anomaly diagnosis.

[0095] In practical implementation, after obtaining the associated data output by the graph attention network analysis module, the mixed integer programming diagnostic module first defines variables. The operating status of each power distribution terminal device is defined as a binary variable (1 indicates abnormality, 0 indicates normal), such as circuit breaker status variables and transformer status variables. The connection quality of the joint commissioning link is defined as another type of binary variable (1 indicates abnormality, 0 indicates normal), such as communication link variables and power transmission link variables. Next, constraints are constructed based on the parameter relationships in the associated data. For example, the power loss of a power transmission link must be less than the maximum allowable loss value of that link; this constraint is expressed through a functional relationship between parameters such as link current and resistance. Another example is that the switching state of a circuit breaker and the current value of its corresponding line must satisfy logical consistency; when the circuit breaker is in the open state, the line current should be 0. Then, an objective function is constructed to minimize the weighted sum of abnormal variables. The weights of the device status variables are set according to the importance of the device (e.g., the weight of the main transformer is 5, and the weight of the branch circuit breaker is 2), and the weights of the joint commissioning link variables are set according to the transmission capacity of the link (e.g., the weight of the main link is 4, and the weight of the branch link is 1). Next, the branch and bound algorithm is used to solve the model. First, the range of variable values ​​is determined. Then, the solution space is continuously divided, and the lower bound of the objective function in each subspace is calculated. Subspaces that cannot contain the optimal solution are gradually eliminated until the optimal solution that satisfies all constraints is found, i.e., the set of potential outliers. Finally, the set of potential outliers is transferred to the mean squared error comprehensive evaluation module.

[0096] The input of the mean square error comprehensive evaluation module is connected to the output of the mixed integer programming diagnostic module. It calculates the mean square error between the set of potential abnormal variables and the set of standard variables, and generates comprehensive evaluation coefficients.

[0097] Specifically, the technical parameters of the mean square error comprehensive evaluation module include: evaluation coefficient calculation accuracy to 6 decimal places; a standard variable set sample size of no less than 1000 groups; parameter weight adjustment range of 0.1-1.0; and evaluation result output time not exceeding 5 seconds. This module supports the establishment of standard variable sets for different types of distribution terminals (such as pole-mounted switch terminals and ring main unit terminals), improving the targeting of the evaluation. Its significance lies in the fact that by calculating the mean square error between the potential abnormal variable set and the standard variable set, the degree of abnormality can be quantitatively assessed. Combined with the differences in parameter weights, it comprehensively reflects the overall status of the distribution terminal's joint commissioning, providing an objective and comparable evaluation basis for subsequent result output.

[0098] In the specific implementation process, after receiving the set of potential abnormal variables output by the mixed integer programming diagnostic module, the mean square error comprehensive evaluation module first calls the preset standard variable set. The standard variable set is constructed by collecting parameter data from similar distribution terminals under normal commissioning conditions over the past year. It includes the normal range and typical values ​​of parameters such as voltage, current, and power, with more than 1000 sets of sample data for each parameter. Next, each parameter in the potential abnormal variable set is matched with the corresponding parameter in the standard variable set; for example, the circuit breaker current value to be evaluated is compared with the normal current value of the circuit breaker in the standard variable set. Then, the deviation value of each parameter is calculated, which is the difference between the potential abnormal variable value and the average value of the corresponding parameter in the standard variable set. Afterwards, a weight is assigned to each deviation value according to the importance of the parameter; for example, the weight of the voltage parameter is 0.8, the weight of the frequency parameter is 0.6, and the weight of the switch status parameter is 1.0. Finally, the deviation value of each parameter is squared and multiplied by the corresponding weight, and then the weighted average of all parameters is calculated to obtain the mean square error comprehensive evaluation coefficient. If a parameter is not present in the set of potential outliers, its deviation value is counted as 0. After calculation, the comprehensive evaluation coefficient is transmitted to the result output module. In practical applications, if the standard variable set needs to be updated, incremental training can be performed periodically with new normal joint debugging data to ensure the timeliness of the standard.

[0099] The input end of the result output module is connected to the output end of the mean square error comprehensive evaluation module, and the comprehensive evaluation coefficient is converted into a visual diagnostic result.

[0100] Specifically, key parameters of the results output module include a refresh rate of 1 time per second for visualization results, supported output formats such as charts (line charts, bar charts, pie charts), text reports, and alarm signals, a display resolution of 1920×1080, and an alarm signal response time of no more than 100ms. This module can be connected to display devices including local monitoring screens, remote terminals, and mobile devices, demonstrating strong compatibility. Its significance lies in transforming the abstract evaluation coefficients generated by the mean square error comprehensive evaluation module into intuitive and easy-to-understand visualization results, facilitating maintenance personnel to quickly understand abnormal conditions in the power distribution terminal commissioning, providing a clear basis for decision-making, and improving maintenance efficiency.

[0101] In practice, after receiving the mean squared error comprehensive evaluation coefficient, the result output module first interprets the evaluation coefficient. When the evaluation coefficient is less than 0.1, the joint debugging status is judged to be normal; when the evaluation coefficient is between 0.1 and 0.5, it is judged to be slightly abnormal; when the evaluation coefficient is greater than 0.5, it is judged to be seriously abnormal. Subsequently, corresponding visualization content is generated based on the judgment results. For the normal state, a real-time monitoring line graph of the joint debugging parameters is generated, showing the curves of voltage, current, power and other parameters changing over time. The curves are distinguished by different colors, with the horizontal axis representing time (accurate to the second) and the vertical axis representing the parameter value (with specific units labeled). For slightly abnormal, in addition to the real-time curve, the time period of the abnormal parameter is marked in yellow in the graph, and a text report is generated, listing the name of the abnormal parameter, the time of occurrence, the deviation value, and other information. For seriously abnormal, in addition to the above, a red alarm signal is triggered, flashing on the display interface, and an alarm message is sent to the mobile device of the maintenance personnel through the communication interface, including the location of the abnormality, the level of the abnormality, and the suggested handling measures. At the same time, the diagnostic results are stored in the local database, retaining records for the most recent 3 months for easy subsequent query and traceability. During the output process, the module will monitor the connection status of the display device in real time. If the connection is interrupted, the result will be cached automatically and retransmitted after the connection is restored.

[0102] As a preferred technical solution, the node feature matrix update process in the graph attention network analysis module satisfies:

[0103]

[0104] in, For the first Layer nodes eigenvectors, For activation function, For nodes The set of neighboring nodes, For the first Layer nodes For nodes Attention weights For the first The weight matrix of the layer, For the first Layer nodes eigenvectors.

[0105] As a preferred technical solution, the abnormal variable identification model in the mixed integer programming diagnostic module satisfies:

[0106]

[0107] in, This is a device status variable (1 indicates abnormal, 0 indicates normal). For the joint debugging link variable (1 indicates abnormal, 0 indicates normal). This is the equipment anomaly cost coefficient. This is the link anomaly cost coefficient. For equipment With Link The correlation coefficient, For link The autocorrelation coefficient, For equipment The anomaly detection threshold For the total number of devices, This represents the total number of links.

[0108] Specifically, in the graph attention network analysis module, the update process of the node feature matrix relies on multi-layer iterative computation. The ReLU activation function is used to effectively handle non-linear feature transformations. The dimension of the weight matrix in each layer is dynamically adjusted according to the dimension of the node features. For example, when the input features are 64-dimensional, the weight matrix is ​​set to 128×64 to increase the feature dimension. The attention weights are calculated based on the similarity of node features, using a multi-layer perceptron for non-linear transformation. The initial learning rate during the transformation is set to 0.001, and it decays exponentially with the number of iterations to ensure model convergence and stability.

[0109] In the mixed-integer programming diagnostic module, the objective function coefficients of the anomaly variable identification model are set according to the importance levels of equipment and links. The equipment anomaly cost coefficient ranges from 1 to 10, with the coefficient for primary equipment such as transformers set to 10 and for secondary equipment such as branch switches set to 1. The link anomaly cost coefficient ranges from 1 to 5, with the coefficient for backbone communication links set to 5 and for branch links set to 1. The correlation coefficients in the constraints are obtained through historical data statistics, ranging from 0 to 1, reflecting the degree of dependence between equipment and links. For example, the correlation coefficient between transformers and the main transmission link is set to 0.8. The anomaly judgment threshold is determined based on the rated parameters of the equipment. For example, the voltage anomaly judgment threshold is set to ±10% of the rated value. During implementation, the node features are first iterated and updated 3-5 times using a graph attention network. After each iteration, the error is calculated using a loss function. Iteration stops when the error is less than 0.0001. Then, the output correlation data is input into the mixed-integer programming model, and the branch and bound method is used to solve the problem. The time of each iteration is controlled within 2 seconds to ensure that the optimal solution, i.e., the set of potential anomaly variables, is obtained within 10 seconds. The entire process is processed in parallel by distributed computing nodes to improve computational efficiency.

[0110] As a preferred technical solution, when the graph attention network analysis module iteratively updates the association weights between nodes, the weight calculation satisfies:

[0111]

[0112] in, For nodes For nodes Attention weights, with LeakyReLU as the activation function. For attention weight vectors, This is the weight matrix. They are nodes eigenvectors, This is a vector concatenation operation. For nodes The set of neighboring nodes, where exp represents the natural exponential function and T represents the transpose of the matrix.

[0113] As a preferred technical solution, the calculation of the comprehensive evaluation coefficients in the mean square error comprehensive evaluation module satisfies the following:

[0114]

[0115] Wherein, MSE is the mean squared error comprehensive evaluation coefficient. For the total number of variables, For the first One potential outlier variable value, For the first Standard variable values.

[0116] Specifically, in the graph attention network, the attention weights are calculated using the LeakyReLU activation function with a negative slope of 0.2, which effectively handles negative features. The dimension of the attention weight vector is consistent with the dimension of the node features; for example, when the node features are 128-dimensional, the attention weight vector is set to 128-dimensional. The weight matrix is ​​initialized using the Xavier initialization method to ensure gradient stability during forward and backward propagation. The determination of the node neighbor set is based on the physical connectivity of devices and parameter correlation, with a correlation threshold set to 0.5. That is, when the parameter correlation coefficient between two nodes is greater than 0.5, they are included in the neighbor set.

[0117] In the mean squared error (MSE) comprehensive evaluation module, the total number of variables is determined based on the number of integrated testing devices, generally covering 50-200 variables, including electrical parameters such as voltage, current, and power, and status parameters such as switch status and communication quality. Standard variable values ​​are calculated using the arithmetic mean of historical normal data, with at least 500 historical samples for each variable to ensure the stability of the average. During implementation, the graph attention network first calculates the original attention weights for all node pairs, then normalizes them using the softmax function. The sum of the normalized weights is 1, ensuring the rationality of weight allocation. Subsequently, the MSE comprehensive evaluation module calculates the difference between each variable value in the potential abnormal variable set and its corresponding standard variable value. The sum of the squares of these differences divided by the total number of variables yields the MSE value. 64-bit floating-point arithmetic is used in the calculation to ensure precision to 6 decimal places. The final MSE value serves as the comprehensive evaluation coefficient, providing a quantitative basis for judging the degree of anomaly.

[0118] As a preferred technical solution, the process of solving the potential abnormal variable set in the mixed integer programming diagnostic module satisfies:

[0119]

[0120] in, Used as an identifier for abnormal variables (1 indicates that it belongs to the abnormal set, and 0 indicates that it does not belong to the abnormal set). For variables Abnormal contribution For variables With variables The mutual exclusion coefficients, For variables The abnormal tolerance threshold This represents the total number of variables.

[0121] As a preferred technical solution, the feature enhancement processing in the graph attention network analysis module satisfies the following:

[0122]

[0123] in, For the first Nodes in the next iteration Enhanced features, For the first In the nth iteration The weights of each convolutional kernel, For the first Each convolution operation, For the first Nodes in the next iteration The original characteristics, This represents the number of convolution kernels.

[0124] Specifically, in the mixed-integer programming diagnostic module, the number of abnormal variable identifiers is consistent with the total number of variables in the joint debugging system. The abnormal contribution rate is set according to the degree of influence of the variable during the joint debugging process, ranging from 0.1 to 1.0. The contribution rate of core variables, such as the main line current, is set to 1.0, and that of secondary variables, such as ambient temperature, is set to 0.1. The mutual exclusion coefficient reflects the repulsion relationship between variables, ranging from 0 to 1. When two variables cannot be abnormal at the same time, the mutual exclusion coefficient is set to 1. The abnormal tolerance threshold is determined according to the normal fluctuation range of the variable; for example, the tolerance threshold for voltage variables is set to 5% of the rated value.

[0125] In the graph attention network analysis module, the number of convolutional kernels is set according to the feature dimension, generally 16-64. Each convolutional kernel is 3×3 in size with a stride of 1, and local features are extracted through a sliding window. The convolutional kernel weights are initialized using a random normal distribution with a mean of 0 and a standard deviation of 0.01. During implementation, the mixed integer programming diagnostic module first constructs a model with the goal of maximizing the sum of anomaly contributions. The constraints include variable mutual exclusion relationships and anomaly tolerance ranges. When solving using the branch and bound method, the initial lower bound is set to 0, and the upper bound is set to the sum of the anomaly contributions of all variables. The upper and lower bounds are updated by continuously dividing the solution space until the difference between the upper and lower bounds is less than 0.01, at which point the solution stops. Simultaneously, the graph attention network analysis module performs multiple rounds of convolution operations on the node features. In each round, different convolutional kernels are used to extract different features. The convolution results are multiplied by the corresponding convolutional kernel weights and then summed to obtain enhanced features. The dimensions of the enhanced features are consistent with the original features to ensure compatibility with subsequent processing. The entire process is accelerated by GPU, and the iteration time for each round is controlled within 1 second.

[0126] As a preferred technical solution, the standard variable set construction in the mean square error comprehensive evaluation module satisfies the following:

[0127]

[0128] in, For the first One standard variable value, This represents the historical normal sample size. For the first The weight coefficients of each sample, For the first In the nth sample Each variable value.

[0129] As a preferred technical solution, in the mixed integer programming diagnostic module, the constraint update satisfies:

[0130]

[0131] in, For variables With variables The constraint coefficient correction amount, As a correction factor, The current value of the variable. , This represents the historical average value of the variable.

[0132] Specifically, in the mean squared error comprehensive evaluation module, the number of historical normal samples is determined based on the type of distribution terminal. For conventional terminals, the sample size is set to 1000 groups, while for newer terminals, the sample size can be expanded to 2000 groups to improve the accuracy of the standard. The sample weighting coefficient is set according to the sample collection time, with recent samples having a higher weight. For example, the weight of samples from the most recent 3 months is set to 1.0, the weight of samples from 3-6 months is set to 0.8, and the weight of samples from 6-12 months is set to 0.5. The influence of outdated data is reduced through weighted averaging.

[0133] In the mixed-integer programming diagnostic module, the correction factor is set according to the stability of the joint debugging system, ranging from 0.1 to 0.5. A larger value, such as 0.5, is used when the system stability is poor, and a smaller value, such as 0.1, is used when the stability is good. The historical average variable value is calculated from the normal data of the past 6 months and updated monthly to ensure it matches the current system state. During implementation, the mean square error comprehensive evaluation module first assigns weights to the historical samples by time, and then calculates the weighted average of each variable as the standard variable value. The calculation uses a weighted summation and then divides by the total weight to ensure the accuracy of the result. The mixed-integer programming diagnostic module calculates the deviation ratio between the current variable value and the historical average variable value in real time, and adjusts the constraint coefficients according to the deviation ratio and the correction factor. When the deviation ratio is greater than 1.5, the constraint coefficient increases by 0.2, and when the deviation ratio is less than 0.5, the constraint coefficient decreases by 0.1. By dynamically adjusting the constraints, the model becomes more adaptable to the real-time changes of the system, improving the flexibility and accuracy of anomaly identification.

[0134] As a preferred technical solution, in the graph attention network analysis module, the layer number adjustment of the multi-layer graph attention mechanism satisfies the following:

[0135]

[0136] in, To determine the optimal number of layers, For the first Layer nodes The output characteristics, For nodes Target characteristics, for Norm, The total number of nodes. Variables that represent the minimum value The value of .

[0137] As a preferred technical solution, in the mean square error comprehensive evaluation module, the evaluation coefficient correction satisfies:

[0138]

[0139] in, This is the corrected mean squared error comprehensive evaluation coefficient, where MSE is the original mean squared error value. For correction factor, For the first One potential outlier variable value, For the first One standard variable value, This represents the total number of variables.

[0140] Specifically, in the graph attention network analysis module, the total number of nodes is determined based on the scale of the power distribution terminal commissioning network: 50-100 nodes for small networks, 100-300 nodes for medium-sized networks, and 300-500 nodes for large networks. Target features are determined through a combination of expert annotation and historical normal data extraction to ensure they accurately reflect the normal state of the nodes. The L2 norm is calculated to six decimal places to guarantee the accuracy of error calculation.

[0141] In the mean square error (MSE) comprehensive evaluation module, correction coefficients are set according to the sensitivity of the parameters. Sensitive parameters, such as short-circuit current, have a correction coefficient of 0.3, while insensitive parameters, such as ambient humidity, have a correction coefficient of 0.1. The total number of variables is consistent with the number of parameters in the joint debugging system, covering multiple categories such as electrical parameters and status parameters. During implementation, the graph attention network analysis module first sets the candidate range for the number of network layers to 3-5 layers. Then, it calculates the sum of the L2 norms of the node output features and target features under different layer numbers, selecting the layer with the smallest sum as the optimal layer number. During the calculation process, the loss function value in each iteration must be less than 0.001 to ensure model convergence. The MSE comprehensive evaluation module, based on the original MSE value, corrects it according to the relative value of the variable deviation. The relative value is the ratio of the variable deviation to the maximum variable value. The corrected MSE value better reflects the relative severity of the anomaly. Floating-point arithmetic is used during the correction process to ensure the reliability of the results. The corrected evaluation coefficients are used for the final anomaly determination, improving the specificity and accuracy of the diagnostic results.

[0142] As a preferred technical solution, the graph attention network analysis module includes a feature mapping unit, an adjacency matrix construction unit, an attention weight calculation unit, and a feature aggregation unit. The feature mapping unit converts voltage and current parameters in structured data into high-dimensional feature vectors, and assigns different dimensional weights to different parameters through a linear transformation matrix to form initial node features. The adjacency matrix construction unit converts the connection status between devices in the joint commissioning link into adjacency matrix element values ​​based on the physical connection relationship and parameter correlation of the distribution terminal, setting the element values ​​corresponding to highly correlated device nodes to non-zero. The attention weight calculation unit uses a multilayer perceptron to perform nonlinear transformation on the node features, calculates the initial weights by comparing the similarity of different node features, and then normalizes them using a softmax function to obtain the attention weights of different node pairs. The feature aggregation unit performs a weighted summation of the neighboring node features according to the attention weights, and generates new node features by combining them with its own node features through an activation function, completing one feature aggregation iteration.

[0143] Specifically, the four units of the graph attention network analysis module are clearly defined in terms of technical parameters, implementation methods, and significance. The core parameters of the feature mapping unit include feature vector dimensions of 64-256, and the dimension of the linear transformation matrix matching the number of input parameters. For example, when the input contains 8 parameters, the matrix is ​​set to 64×8. The weight of each parameter in different dimensions is determined through model training, ranging from -1 to 1. The significance of this unit is to transform the original parameters into more representative high-dimensional features, laying the foundation for subsequent association analysis. In the adjacency matrix construction unit, the matrix dimension is consistent with the number of device nodes. For example, when there are 50 device nodes, the matrix is ​​50×50. The element values ​​corresponding to physical connections are set to 1, and those for non-connections are set to 0. Parameter correlation is calculated using the Pearson correlation coefficient; element values ​​with an absolute value greater than 0.6 are set as the correlation coefficient itself, otherwise they are set to 0. Its significance lies in quantifying the degree of association between devices, providing a basis for attention weight calculation. The attention weight calculation unit uses a multilayer perceptron with two hidden layers: 128 neurons in the first layer and 64 in the second. The activation function is LeakyReLU, and the initial weights are initialized using a normal distribution with a mean of 0 and a standard deviation of 0.01. The significance of this unit is to determine the importance of different node pairs through learning, highlighting key associations. In the feature aggregation unit, the activation function is ReLU, and the learning rate for each aggregation iteration is set to 0.001. The number of iterations is determined by the loss function value, stopping when the loss is less than 0.0001. Its significance lies in integrating neighbor node information to enhance the expressive power of node features. During implementation, the feature mapping unit first receives the structured data output by the edge computing processing module, inputs parameters such as voltage and current into the linear transformation matrix, and generates an initial feature vector; the adjacency matrix construction unit determines the connection relationship according to the physical topology of the power distribution terminal, and calculates the correlation coefficient between parameters to fill the matrix elements; the attention weight calculation unit inputs the node features into the multilayer perceptron, obtains the original weights, and normalizes them through the softmax function; the feature aggregation unit sums the features of neighboring nodes according to the normalized weights, combines them with its own features, and generates new features through an activation function. This process is repeated until the stopping condition is met. The entire process is accelerated by the GPU, and the iteration time of each round is controlled within 0.5 seconds to ensure efficient processing.

[0144] As a preferred technical solution, the mixed integer programming diagnostic module includes a variable definition unit, a constraint generation unit, an objective function construction unit, and a solution unit. The variable definition unit converts the equipment operating status and link connectivity during power distribution terminal commissioning into binary decision variables, clarifying the physical meaning of each variable value, such as a variable value of 1 corresponding to equipment anomalies. The constraint generation unit, based on the allowable range of commissioning parameters and the logical relationships between equipment, converts voltage fluctuation ranges and power balance relationships into mathematical inequalities, forming a set of constraints. The objective function construction unit aims to minimize the total number of abnormal variables or maximize the anomaly identification accuracy, using the degree of anomaly impact of different variables as the coefficients of the objective function. The solution unit uses the branch-and-bound method to solve the constructed mixed integer programming model, gradually narrowing the search range by continuously dividing the solution space and calculating the lower bound, ultimately obtaining the optimal set of abnormal variables that satisfies the constraints.

[0145] Specifically, the four units of the mixed-integer programming diagnostic module have clearly defined technical parameters, implementation significance, and operational procedures. In the variable definition unit, equipment operating status variables cover 30 types of equipment, including circuit breakers and transformers, while joint commissioning link variables include 15 types of links, such as communication links and power transmission links. Each variable takes only 0 or 1, corresponding to normal and abnormal conditions, respectively. The significance of this unit lies in transforming complex physical states into calculable mathematical variables, providing a foundation for model construction. In the constraint generation unit, voltage fluctuation range constraints are determined based on the equipment's rated voltage. For example, for equipment rated at 380V, the constraint is 342V to 418V. The power balance relationship is reflected by ensuring that the absolute value of the difference between the injected power and the outflow power at each node is less than 0.5kW. The number of various constraints is determined based on the system scale, generally including 200-500. Their significance lies in limiting the reasonable value range of variables to ensure the physical feasibility of the model solution. In the objective function construction unit, the weights of equipment anomaly variables are determined based on the impact range of equipment faults; the weight of the main transformer is set to 10, and that of the branch circuit breaker to 3. The weights of link anomaly variables are determined based on transmission capacity; the weight of the trunk link is set to 8, and that of the branch link to 2. The significance of this unit is to guide the model to prioritize the identification of anomalies with greater impact. In the branch and bound method used in the solution unit, the initial difference between the upper and lower bounds is set to 100, the upper and lower bounds are updated with a threshold of 5 after each branch, and the maximum number of iterations is set to 1000. This is significant in that it efficiently finds the optimal solution and ensures the accuracy of the diagnostic results. During implementation, the variable definition unit first numbers all devices and links, establishing a correspondence between variables and physical objects; the constraint generation unit, based on equipment manuals and operating specifications, transforms the allowable ranges of parameters such as voltage and power into inequalities, forming a constraint set; the objective function construction unit assigns weights according to the importance of devices and links, constructing a weighted sum form of the objective function; the solution unit initializes the variable value range, continuously divides the solution space using the branch and bound method, calculates the objective function value of each subspace, eliminates subspaces that cannot contain the optimal solution, until the solution that satisfies all constraints and minimizes the objective function value is found, i.e., the potential abnormal variable set. During the solution process, the calculation time of each branch is controlled within 1 second, and the total solution time does not exceed 10 seconds, ensuring real-time performance.

[0146] As a preferred technical solution, the mean square error comprehensive evaluation module includes a standard sample library unit, a difference calculation unit, a mean square error solution unit, and a coefficient correction unit. The standard sample library unit stores different parameter values ​​from historical normal commissioning processes, categorized by distribution terminal type and commissioning scenario, forming standard parameter sets (i.e., the aforementioned standard variable sets) for different scenarios. The difference calculation unit performs a difference operation between each parameter value in the potential abnormal variable set and the average value of the corresponding parameter in the standard sample library (i.e., the aforementioned standard variable set) to obtain the absolute deviation value of each parameter. The mean square error solution unit squares the deviation values ​​of all parameters and then calculates the arithmetic mean to obtain the initial mean square error evaluation coefficient. The coefficient correction unit assigns different weights to the deviation values ​​of different parameters according to their importance in the commissioning process, recalculates the weighted mean square error value, and uses it as the final comprehensive evaluation coefficient.

[0147] Specifically, the four units of the mean square deviation comprehensive assessment module have specific technical parameters, practical significance, and implementation steps. The standard sample library unit is divided into five categories based on the type of distribution terminal, such as pole-mounted switch terminals and ring main unit terminals. Each type of terminal contains 20 standard parameter sets, including voltage and current. The historical normal sample size for each parameter is 1000 sets, with a sample collection period of one year and 200 new samples updated monthly. The significance of this unit is to provide a benchmark reference and a basis for anomaly assessment. In the difference calculation unit, the parameter average is calculated using the arithmetic mean method. For continuous parameters such as voltage, the deviation value is retained to two decimal places; for discrete parameters such as switch status, the deviation value is 0 or 1 (1 if not both in the standard state). The significance of this unit is to quantify the difference between potential abnormal variables and standard values. In the mean square deviation solution unit, the squaring operation uses 64-bit floating-point precision, and the arithmetic mean is calculated to six decimal places. The significance of this unit is to synthesize the deviations of all parameters to obtain a quantitative index of the overall anomaly degree. In the coefficient correction unit, the importance weights of parameters are set according to the degree of fault impact: voltage parameter weight is 0.2, current is 0.18, power is 0.15, switch status is 0.12, communication signal strength is 0.1, and the remaining parameters total 0.25, with a total weight of 1. The significance of this unit is to highlight the impact of important parameters, making the evaluation results more in line with actual needs. During implementation, the standard sample library unit first calls the corresponding standard parameter set according to the terminal type to ensure the relevance of the evaluation; the difference calculation unit extracts the parameter values ​​from the potential abnormal variable set and calculates the deviation from the average value of the corresponding parameters in the standard sample library; the mean square error calculation unit squares all deviation values, sums them, and divides by the total number of parameters to obtain the initial mean square error; the coefficient correction unit multiplies the squared deviation of each parameter by its corresponding weight, sums them, and then divides by the total weight to obtain the final evaluation coefficient. Distributed computing is used during the calculation process to ensure efficiency, and the entire evaluation process is completed within 3 seconds, providing timely support for the result output.

[0148] Graph Attention Network (GAN) is a deep learning model based on graph-structured data. It captures the relationships between nodes in a graph through an attention mechanism and enhances node features. In this application, the model is implemented as follows: First, it receives structured data output from an edge computing processing module, treating various devices in the power distribution terminal as nodes in the graph, and the physical connections and parameter correlations between devices as edges. A node feature matrix is ​​constructed, with the matrix dimension determined by the number of devices and parameter types. For example, when there are 50 device nodes and 8 types of parameters, the matrix is ​​50×8. Next, iteratively updates the relationship weights between nodes using a multi-layer graph attention mechanism. Each layer includes linear transformation, attention weight calculation, and feature aggregation. The attention weights are calculated based on the similarity of node features, undergoing nonlinear transformation via a multi-layer perceptron and then normalized using a softmax function to ensure the total weight sum is 1. Then, after 3-5 layers of iterative calculation, the enhanced relationship data is output. The learning rate for each iteration is initially set to 0.001, decreasing exponentially with each iteration until the loss function value is less than 0.0001. The model aims to deeply explore the complex relationships between power distribution terminal equipment and parameters, especially implicit relationships, transforming raw data into more representative correlation features. This provides accurate feature basis for subsequent anomaly diagnosis. Its significance lies in overcoming the limitation of traditional analysis methods that ignore equipment correlations, improving the ability to extract characteristics of power distribution terminal commissioning status, and enabling subsequent diagnostic processes to more comprehensively consider the mutual influence of various factors, thereby improving the accuracy and reliability of the diagnosis.

[0149] Mixed-integer programming is a mathematical optimization model that solves for the optimal solution by constructing an objective function and constraints containing integer and continuous variables. In this application, the implementation process of this model is as follows: Using the associated data output by the graph attention network analysis module as constraints, the equipment status of the distribution terminal (such as the operating status of circuit breakers and transformers) and the connection status of the joint commissioning links (such as the connection status of communication links and power transmission links) are defined as binary integer variables, taking values ​​of 0 or 1, representing normal and abnormal respectively. Then, an objective function is constructed with the goal of minimizing the weighted sum of abnormal variables. The weights of the equipment status variables are set according to the importance of the equipment, with primary equipment having a weight of 10 and secondary equipment having a weight of 1-3; the weights of the joint commissioning link variables are set according to the transmission capacity, with main links having a weight of 5-8 and branch links having a weight of 1-2. Simultaneously, constraints are set according to the parameter relationships in the associated data. For example, the power loss of a power transmission link must be less than its maximum allowable loss value, and the switching status of a circuit breaker and the current value of its corresponding line must satisfy logical consistency, etc. The number of constraints is determined according to the system scale, generally 200-500. Subsequently, the branch and bound method was used to solve the model. The initial difference between the upper and lower bounds was set to 100. After each branch, the upper and lower bounds were updated until the difference was less than 5 or the maximum number of iterations (1000) was reached, resulting in a set of potential abnormal variables. The purpose of this model is to accurately locate potential abnormal variables during the commissioning of power distribution terminals, considering multiple constraints, and to identify which devices or links exhibit anomalies. Its significance lies in transforming the complex anomaly diagnosis problem into a mathematical optimization problem. By replacing traditional experience-based judgment with a systematic solution process, it improves the scientific rigor and accuracy of anomaly identification, providing clear anomaly targets for subsequent comprehensive evaluation.

[0150] The mean squared error (MSE) comprehensive evaluation model is a quantitative evaluation model based on MSE. It generates a comprehensive evaluation coefficient by calculating the MSE between a set of potential outlier variables and a set of standard variables, combined with parameter weights. In this application, the model is implemented as follows: First, a preset set of standard variables is called. This set is constructed from parameter data of the same type of distribution terminal under normal commissioning conditions over the past year. Each parameter contains more than 1000 sets of sample data. The standard variable value is calculated through a weighted average, with recent samples having higher weights (e.g., 1.0 for the last 3 months and 0.8 for 3-6 months). Then, the difference between each parameter value in the set of potential outlier variables output by the mixed integer programming diagnostic module and the corresponding parameter value in the set of standard variables is calculated to obtain the deviation value for each parameter. Next, a weight is assigned to each deviation value according to the importance of the parameter: key parameters such as voltage and current have a weight of 0.8-1.0, while secondary parameters such as ambient temperature have a weight of 0.1-0.3, with a total weight of 1. The deviation value of each parameter is then squared and multiplied by its corresponding weight to calculate the weighted average of all parameters, yielding the root mean square error comprehensive evaluation coefficient. The calculation accuracy reaches six decimal places. The model's function is to quantitatively assess the degree of anomaly in a set of potential abnormal variables, comprehensively reflecting the overall status of the distribution terminal's joint commissioning. Its significance lies in enabling maintenance personnel to intuitively understand the severity of anomalies through quantitative evaluation results. This overcomes the limitations of traditional diagnostic methods that rely on single indicators or qualitative descriptions, providing an objective and comparable basis for developing targeted handling measures, further enhancing the practicality and reliability of the entire system.

[0151] This application proposes an intelligent analysis and diagnosis system for distribution terminal commissioning data based on edge computing. This system integrates edge computing with models such as graph attention networks, mixed integer programming, and mean square error comprehensive evaluation, offering several beneficial effects. The edge computing processing module processes commissioning data locally at the terminal, significantly reducing transmission latency and bandwidth pressure, and improving real-time data processing. The graph attention network analysis module can deeply mine complex relationships between devices and parameters, capturing implicit relationships through a multi-layered attention mechanism, effectively improving the sensitivity to identify hidden anomalies and overcoming the shortcomings of existing technologies, such as insufficient data correlation analysis capabilities and susceptibility to missed diagnoses. The mixed integer programming diagnosis module constructs a model using correlated data as constraints, accurately locating abnormal variables. The mean square error comprehensive evaluation module combines parameter weight differences to generate a comprehensive evaluation result, improving the comprehensive evaluation mechanism of the diagnosis results. This solves the problem of existing technologies relying on single indicators and failing to comprehensively reflect the system status, making the diagnosis results more reliable and targeted, meeting the refined operation and maintenance needs of complex distribution networks, and significantly improving the efficiency and accuracy of distribution terminal commissioning.

[0152] Based on the aforementioned intelligent analysis and diagnosis system for distribution terminal commissioning data, this embodiment also provides a method for intelligent analysis and diagnosis of distribution terminal commissioning data, such as... Figure 3 As shown, the method includes the following steps:

[0153] Step S1: The edge computing processing module receives the power distribution terminal commissioning parameters transmitted by the data acquisition module, performs segmented processing according to a preset time window, performs distributed parallel computing on each segment of data, and extracts the parameter change rate and fluctuation characteristics.

[0154] Step S2: The graph attention network analysis module constructs the processed data into a graph structure with devices as nodes and parameter correlations as edges. It learns the dependencies between nodes through a multi-layer attention mechanism and updates the node feature representations.

[0155] Step S3: The mixed integer programming diagnostic module takes the updated node features as input, sets the normal operating parameter range of the equipment and link connection constraints, constructs a planning model containing integer variables, and obtains the suspected abnormal variables (i.e. the aforementioned potential abnormal variables) through iterative solution.

[0156] Step S4: The mean square error comprehensive evaluation module calls the historical data in the standard parameter library (i.e., the aforementioned standard variable set), calculates the mean square error between the suspected abnormal variable and the standard parameter, and assigns weights according to the importance of the parameters to obtain the comprehensive evaluation value (i.e., the aforementioned comprehensive evaluation coefficient).

[0157] Step S5: Compare the comprehensive evaluation value with the preset threshold. If it exceeds the threshold, it is marked as abnormal; otherwise, it is judged as normal.

[0158] Step S6: The result output module integrates the anomaly judgment result with the corresponding parameter change curve and anomaly variable location information, converts it into a diagnostic report format that the power distribution terminal can recognize, and transmits it to the terminal display interface.

[0159] The edge computing-based intelligent analysis and diagnosis system for distribution terminal commissioning data in this application embodiment offers significant advantages in data processing. The edge computing processing module completes the commissioning data processing at the terminal side, completely avoiding the latency caused by long-distance data transmission in traditional centralized processing. This greatly alleviates bandwidth pressure and significantly improves data processing speed and real-time performance. This feature precisely addresses the problem of low transmission efficiency in the centralized mode in existing technologies, enabling timely capture of dynamic parameter changes during commissioning. This provides efficient and timely data support for subsequent analysis and diagnosis, ensuring the smoothness of data processing throughout the entire commissioning process.

[0160] The system performs exceptionally well in anomaly identification and analysis. The graph attention network analysis module delves into the complex relationships between devices and parameters, accurately capturing implicit connections through a multi-layered attention mechanism, effectively improving the sensitivity of identifying latent anomalies. This design successfully solves the problem of missed diagnoses caused by insufficient data correlation analysis capabilities in existing technologies. Simultaneously, the mixed-integer programming diagnostic module uses correlated data as constraints to build models, accurately locating anomalous variables, making anomaly identification more targeted and accurate, and significantly improving the effectiveness of anomaly diagnosis.

[0161] The system's comprehensive evaluation mechanism is also noteworthy. The mean square error comprehensive evaluation module combines the weight differences of different parameters to generate a comprehensive evaluation result, improving the comprehensive evaluation system of diagnostic results and solving the shortcomings of existing technologies that rely on a single indicator and cannot fully reflect the system status. This makes the diagnostic results more reliable and more in line with actual operation and maintenance needs, significantly improving the efficiency and accuracy of distribution terminal commissioning, fully meeting the requirements of refined operation and maintenance of complex distribution networks, and providing strong support for the stable operation of distribution terminals.

[0162] Based on the same inventive concept, this application also provides a computer-readable storage medium storing one or more programs, which, when executed, can realize the intelligent analysis and diagnosis method for power distribution terminal commissioning data as described above.

[0163] Based on the same inventive concept, such as Figure 4 As shown, this application also provides an electronic device, including a processor, a communication interface, a computer-readable storage medium as described above, and a communication bus; wherein the processor, the communication interface, and the computer-readable storage medium communicate with each other through the communication bus; the processor is used to execute a program stored in the computer-readable storage medium.

[0164] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of systems or modules may be electrical, mechanical, or other forms.

[0165] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0166] If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0167] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0168] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0169] The parts not mentioned in the above embodiments are the same as or can be implemented using existing technologies, and will not be further described here.

[0170] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A power distribution terminal joint debugging data intelligent analysis and diagnosis method, characterized in that, The method includes: Real-time acquisition of commissioning parameters during the commissioning process of power distribution terminals; The collected joint debugging parameters are processed in a distributed manner based on edge computing to generate structured data; The structured data is constructed into a node feature matrix, and the association weights between nodes are iteratively updated through the multi-layer attention mechanism of the graph attention network to output the feature-enhanced association data. Using the associated data as constraints, an integer programming model including equipment status variables and joint debugging link variables is constructed, and the integer programming model is iteratively solved to obtain a set of potential abnormal variables; Calculate the mean squared error between the set of potential outlier variables and the set of standard variables, and assign weights according to the importance of the parameters to obtain a comprehensive evaluation coefficient; The comprehensive evaluation coefficients are then converted into visual diagnostic results. Specifically, the process involves constructing an integer programming model, using the associated data as constraints, including equipment state variables and joint debugging link variables. Iteratively solving this integer programming model yields a set of potential abnormal variables, including: The equipment operating status and link connectivity during the power distribution terminal commissioning process are converted into binary decision variables, and the physical meaning of the variable values ​​is clarified. Based on the allowable range of joint debugging parameters and the logical relationship between equipment, the voltage fluctuation range and power balance relationship are transformed into mathematical inequalities, forming a set of constraints; An objective function is constructed with the goal of minimizing the total number of abnormal variables or maximizing the accuracy of anomaly identification, and the degree of abnormal influence of different variables is used as the coefficient of the objective function. The branch and bound method is used to solve the constructed integer programming model. By continuously dividing the solution space and calculating the lower bound value, the search range is gradually narrowed, and finally the optimal set of abnormal variables that satisfy the constraints is obtained.

2. The intelligent analysis and diagnosis method for distribution terminal commissioning data according to claim 1, characterized in that, The joint debugging parameters include voltage, current, power, frequency, switch status, and communication signal strength parameters; The distributed computation processing of the collected joint debugging parameters based on edge computing to generate structured data includes: The system receives and processes the collected parameters, divides them into segments according to a preset time window, performs distributed parallel computation on each segment, extracts the parameter change rate and fluctuation characteristics, and generates structured data.

3. The intelligent analysis and diagnosis method for distribution terminal commissioning data according to claim 1, characterized in that, The process of constructing a node feature matrix from the structured data, iteratively updating the association weights between nodes through a multi-layer attention mechanism of a graph attention network, and outputting feature-enhanced association data includes: The structured data is constructed into a graph structure with devices as nodes and parameter correlations as edges. The dependencies between nodes are learned through the multi-layer attention mechanism of the graph attention network, the node feature representations are updated, and the feature-enhanced associated data is output.

4. The intelligent analysis and diagnosis method for distribution terminal commissioning data according to claim 1, characterized in that, The process of constructing a node feature matrix from the structured data, iteratively updating the association weights between nodes through a multi-layer attention mechanism of a graph attention network, and outputting feature-enhanced association data includes: The voltage and current parameters in the structured data are converted into high-dimensional feature vectors. Different dimensional weights are assigned to different parameters through a linear transformation matrix to form the initial features of the nodes. Based on the physical connection relationship and parameter correlation of the power distribution terminal, the connection status between devices in the joint commissioning link is converted into the adjacency matrix element value, and the element value corresponding to the device node with high correlation is set to non-zero; A multilayer perceptron is used to perform nonlinear transformation on node features. Initial weights are calculated by comparing the similarity of different node features, and then the attention weights of different node pairs are obtained by normalization using the softmax function. The features of neighboring nodes are weighted and summed according to their attention weights, and new node features are generated by combining them with the node's own features through an activation function, thus completing one feature aggregation iteration.

5. The intelligent analysis and diagnosis method for distribution terminal commissioning data according to claim 1, characterized in that, The calculation of the mean squared error between the set of potential outliers and the set of standard variables, and the allocation of weights according to parameter importance to obtain a comprehensive evaluation coefficient, includes: Store different parameter values ​​from historical normal commissioning processes, classify them according to power distribution terminal type and commissioning scenario, and form standard variable sets for different scenarios; The absolute deviation value of each parameter is obtained by calculating the difference between the value of each parameter in the potential abnormal variable set and the average value of the corresponding parameter in the standard variable set. The initial root mean square error evaluation coefficient is obtained by squaring the deviation values ​​of all parameters and then calculating the arithmetic mean. Based on the importance of the parameters in the joint debugging, different weights are assigned to the deviation values ​​of different parameters, and the weighted mean square error is recalculated as the final comprehensive evaluation coefficient.

6. A method for intelligent analysis and diagnosis of distribution terminal commissioning data according to any one of claims 1-5, characterized in that, The process of converting the comprehensive evaluation coefficients into visual diagnostic results includes: The comprehensive evaluation coefficient is compared with a preset threshold. If it exceeds the threshold, it is marked as abnormal; otherwise, it is judged as normal. The anomaly detection results are integrated with the corresponding parameter change curves and anomaly variable location information, and then transformed into visual diagnostic results and output.

7. The intelligent analysis and diagnosis method for distribution terminal commissioning data according to claim 1, characterized in that, The update process of the node feature matrix satisfies: in, For the first Layer nodes eigenvectors, For activation function, For nodes The set of neighboring nodes, For the first Layer nodes For nodes Attention weights For the first The weight matrix of the layer, For the first Layer nodes eigenvectors.

8. The intelligent analysis and diagnosis method for distribution terminal commissioning data according to claim 1, characterized in that, When iteratively updating the association weights between nodes, the weight calculation satisfies: in, For nodes For nodes Attention weights, with LeakyReLU as the activation function. For attention weight vectors, This is the weight matrix. They are nodes eigenvectors, This is a vector concatenation operation. For nodes The set of neighboring nodes, where exp represents the natural exponential function and T represents the transpose of the matrix.

9. The intelligent analysis and diagnosis method for distribution terminal commissioning data according to claim 1, characterized in that, Feature enhancement processing satisfies: in, For the first Nodes in the next iteration Enhanced features, For the first In the nth iteration The weights of each convolutional kernel, For the first Each convolution operation, For the first Nodes in the next iteration The original characteristics, This represents the number of convolution kernels.

10. The intelligent analysis and diagnosis method for distribution terminal commissioning data according to claim 1, characterized in that, The layer number adjustment of the multi-layer attention mechanism satisfies: in, To determine the optimal number of layers, For the first Layer nodes The output characteristics, For nodes Target characteristics, for Norm, The total number of nodes. Variables that represent the minimum value The value of .

11. The intelligent analysis and diagnosis method for distribution terminal commissioning data according to claim 1, characterized in that, The objective function and constraints of the integer programming model are as follows: in, For device state variables, For the joint debugging link variables, This is the equipment anomaly cost coefficient. This is the link anomaly cost coefficient. For equipment With Link The correlation coefficient, For link The autocorrelation coefficient, For equipment The anomaly detection threshold For the total number of devices, This represents the total number of links.

12. The intelligent analysis and diagnosis method for distribution terminal commissioning data according to claim 1, characterized in that, The process of solving the set of potential abnormal variables satisfies: in, As an identifier for abnormal variables, A value of 1 indicates that it belongs to the abnormal set. A value of 0 indicates that it does not belong to the abnormal set. For variables Abnormal contribution For variables With variables The mutual exclusion coefficients, For variables The abnormal tolerance threshold This represents the total number of variables.

13. The intelligent analysis and diagnosis method for distribution terminal commissioning data according to claim 1, characterized in that, The constraint update of the integer programming model satisfies: in, For variables With variables The constraint coefficient correction amount, As a correction factor, The current value of the variable. , This represents the historical average value of the variable.

14. The intelligent analysis and diagnosis method for distribution terminal commissioning data according to claim 1, characterized in that, The calculation of the comprehensive evaluation coefficient satisfies: Wherein, MSE is the mean squared error comprehensive evaluation coefficient. For the total number of variables, For the first One potential outlier variable value, For the first Standard variable values.

15. The intelligent analysis and diagnosis method for distribution terminal commissioning data according to claim 1, characterized in that, The construction of the standard variable set satisfies: in, For the first One standard variable value, This represents the historical normal sample size. For the first The weight coefficients of each sample, For the first In the nth sample Each variable value.

16. The intelligent analysis and diagnosis method for distribution terminal commissioning data according to claim 1, characterized in that, The correction of the comprehensive evaluation coefficient satisfies: in, The corrected mean squared error (MSE) is the comprehensive evaluation coefficient, where MSE is the original mean squared error value. For correction factor, For the first One potential outlier variable value, For the first One standard variable value, This represents the total number of variables.

17. A power distribution terminal joint debugging data intelligent analysis and diagnosis system, characterized in that, The system includes: The data acquisition module is used to collect the commissioning parameters of the power distribution terminal in real time during the commissioning process. The edge computing processing module is used to perform distributed computing processing on the collected joint debugging parameters based on edge computing to generate structured data; The graph attention network analysis module is used to construct the structured data into a node feature matrix, iteratively update the association weights between nodes through the multi-layer attention mechanism of the graph attention network, and output the feature-enhanced association data. The mixed integer programming diagnostic module is used to construct an integer programming model including equipment state variables and joint debugging link variables using the associated data as constraints, and to iteratively solve the integer programming model to obtain a set of potential abnormal variables, including: The equipment operating status and link connectivity during the power distribution terminal commissioning process are converted into binary decision variables, and the physical meaning of the variable values ​​is clarified. Based on the allowable range of joint debugging parameters and the logical relationship between equipment, the voltage fluctuation range and power balance relationship are transformed into mathematical inequalities, forming a set of constraints; An objective function is constructed with the goal of minimizing the total number of abnormal variables or maximizing the accuracy of anomaly identification, and the degree of abnormal influence of different variables is used as the coefficient of the objective function. The branch and bound method is used to solve the constructed integer programming model. By continuously dividing the solution space and calculating the lower bound, the search range is gradually narrowed, and finally the optimal set of abnormal variables that satisfy the constraints is obtained. The mean square error comprehensive evaluation module is used to calculate the mean square error between the set of potential outlier variables and the set of standard variables, and to obtain the comprehensive evaluation coefficient by assigning weights according to the importance of the parameters. The results output module is used to convert the comprehensive evaluation coefficients into visual diagnostic results.

18. A computer-readable storage medium storing one or more programs, characterized in that, When one or more of these programs are executed, the intelligent analysis and diagnosis method for power distribution terminal commissioning data as described in any one of claims 1-16 is implemented.

19. An electronic device comprising a processor, a communication interface, the computer readable storage medium of claim 18 and a communication bus; wherein, The processor, communication interface, and computer-readable storage medium communicate with each other via a communication bus; Its features are, The processor is used to execute programs stored in a computer-readable storage medium.

Images