An alarm analysis method and system for power transformation equipment fault tripping

Abstract

This invention relates to the field of power supply and transformation equipment tripping analysis technology, specifically to an alarm analysis method and system for power supply and transformation equipment fault tripping. The method includes the following steps: acquiring operating data of the power supply and transformation equipment, including historical operating data and real-time operating data; preprocessing the operating data of the power supply and transformation equipment; extracting features from the historical operating data of the power supply and transformation equipment based on kernel functions, and simultaneously fusing the extracted feature vectors of the power supply and transformation equipment using a random forest algorithm, and constructing a power supply and transformation equipment tripping type fusion regression model based on a combined averaging strategy; inputting the real-time operating data of the power supply and transformation equipment into the power supply and transformation equipment tripping type fusion regression model for calculation, and outputting the tripping fault analysis results of the power supply and transformation equipment.

Description

Technical Field

[0001] This invention relates to the field of power supply and transformation equipment tripping analysis technology, and more specifically, to an alarm analysis method and system for power supply and transformation equipment fault tripping. Background Technology

[0002] With the continuous expansion of power system scale and the in-depth advancement of smart grid construction, power supply and transformation equipment, as a core component of the power system, directly affects the reliability of the entire power system through its safe and stable operation. Currently, traditional fault early warning methods for power supply and transformation equipment mainly suffer from the following problems: conventional feature extraction methods, such as principal component analysis and wavelet transform, are difficult to effectively handle high-dimensional nonlinear data, resulting in extracted features that may not accurately reflect the equipment status; furthermore, single classification or regression models, such as support vector machines and neural networks, are prone to overfitting, and the analysis results are not stable or reliable enough. When facing complex operating environments of power supply and transformation equipment, the accuracy and real-time performance of early warnings are difficult to guarantee. Based on these issues, we have designed an alarm analysis method and system for power supply and transformation equipment fault tripping. Summary of the Invention

[0003] The purpose of this invention is to provide an alarm analysis method and system for power supply and transformation equipment fault tripping. It uses kernel functions to achieve nonlinear feature extraction, which can more accurately capture key feature information in the operation data of power supply and transformation equipment. In addition, this invention uses random forest for data fusion, which not only improves the generalization ability of the model, but also effectively reduces the overfitting risk of a single decision tree, and can significantly improve the accuracy and reliability of early warning.

[0004] This invention is achieved through the following technical solution:

[0005] An alarm analysis method for power supply equipment fault tripping, the method comprising the following steps:

[0006] Acquire operational data of power supply and transformation equipment, including historical and real-time operational data;

[0007] Preprocessing of power supply and transformation equipment operation data;

[0008] Feature extraction is performed on historical operating data of power supply and transformation equipment based on kernel function. At the same time, the extracted feature vectors of power supply and transformation equipment are fused by combining random forest algorithm. Finally, a fusion regression model of tripping type of power supply and transformation equipment is constructed based on the combined averaging strategy.

[0009] The real-time operating data of power supply and transformation equipment is input into the tripping type fusion regression model of power supply and transformation equipment for calculation, and the tripping fault analysis results of power supply and transformation equipment are output.

[0010] Optionally, the preprocessing of the power supply and transformation equipment operation data includes: data cleaning of the power supply and transformation equipment operation data, and data transformation of the power supply and transformation equipment operation data.

[0011] Optionally, the feature extraction based on kernel functions of historical operating data of power supply and transformation equipment specifically includes:

[0012] A feature matrix is ​​constructed based on the preprocessed historical operating data of power supply and transformation equipment;

[0013] Calculate the mean of each feature in the feature matrix, and then decenter each feature to obtain the decentralized feature matrix.

[0014] The eigenvectors are mapped to the first-dimensional space using a Gaussian kernel function, the kernel matrix is ​​centered, and its covariance matrix is ​​calculated.

[0015] Solve the covariance matrix, calculate the variance contribution rate of each eigenvalue, calculate the cumulative contribution rate based on the variance contribution rate of each eigenvalue, and select multiple principal components based on the cumulative contribution rate;

[0016] The selected principal components are mapped to the second-dimensional space, and the power supply and transformation equipment feature vectors in the historical operation data of the power supply and transformation equipment are extracted as the final output.

[0017] Optionally, the extraction of power supply and transformation equipment feature vectors using the random forest algorithm is fused into data, specifically as follows:

[0018] The data in the feature vector of the power supply and transformation equipment is used as the original dataset, and the Bootstrap sampling method is used to randomly sample the original dataset with replacement according to a set ratio, and this process is repeated multiple times to generate multiple training sample sets.

[0019] For each training sample set, construct a CART decision tree:

[0020] Iterate through each feature and its segmentation point in the training sample set;

[0021] The training sample set is divided into a left sub-region and a right sub-region according to the segmentation criteria;

[0022] Calculate the optimal regression values ​​for the left and right sub-regions respectively, and calculate the segmentation loss function based on the optimal regression values;

[0023] Select the feature and split point that minimizes the loss function as the optimal splitting scheme for the current node;

[0024] Repeat the above segmentation process for the left and right sub-regions until the maximum number of iterations is reached to generate a complete decision tree. Test each decision tree with an unsampled set, combine each decision tree into a random forest model, and complete the training process of the random forest model.

[0025] Optionally, the step of constructing a fusion regression model for power supply and substation equipment tripping types based on a combined average strategy is as follows:

[0026] For each trip type:

[0027] Collect the analysis results of each decision tree for the same input;

[0028] Combine and average the results of each analysis;

[0029] The averaged result is used as the final analysis value for this trip type;

[0030] Summarize all tripping types and complete the construction of a fusion regression model for tripping types of power supply and transformation equipment.

[0031] Optionally, the loss function is specifically calculated using the following formula:

[0032]

[0033] Where Y is the loss function value, R1 and R2 are the number of samples in the left and right sub-regions, respectively, j is the feature index, s is the value of the split point, and x is the value of the split point. i y i These are the feature vector and the true value of the i-th sample, respectively.

[0034] Optionally, the calculation formula for the power supply and transformation equipment tripping type fusion regression model is as follows:

[0035]

[0036] Where G(X) is the final analysis value of the fusion model for input X, N is the total number of decision trees, and f b (X) is the analysis value of the b-th decision tree for the input X, w b Let be the weight coefficients of the b-th decision tree, ∈ b Let denot be the analysis error of the b-th decision tree on the training samples, and α be the error decay coefficient.

[0037] An alarm analysis system for power supply equipment fault tripping includes:

[0038] The data acquisition module acquires the operating data of the power supply and transformation equipment, including historical operating data and real-time operating data;

[0039] The preprocessing module preprocesses the operating data of power supply and transformation equipment;

[0040] The feature fusion module extracts features from historical operating data of power supply and transformation equipment based on kernel functions. It also combines the extracted feature vectors of power supply and transformation equipment with random forest algorithm to fuse the data, and builds a fusion regression model of power supply and transformation equipment tripping type based on the combined averaging strategy.

[0041] The tripping type early warning module inputs real-time operating data of power supply and transformation equipment into the tripping type fusion regression model of power supply and transformation equipment for calculation, and outputs the tripping fault analysis results of power supply and transformation equipment.

[0042] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0043] This invention uses kernel functions to extract nonlinear features, which can more accurately capture key feature information in the operation data of power supply and transformation equipment. In addition, this invention uses random forest for data fusion, which not only improves the generalization ability of the model, but also effectively reduces the overfitting risk of a single decision tree, and can significantly improve the accuracy and reliability of early warning. Attached Figure Description

[0044] Figure 1 A flowchart illustrating an alarm analysis method for power supply equipment fault tripping provided by the present invention;

[0045] Figure 2 This is a schematic diagram of data processing provided by the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, what is described is only a part of this invention, not all of it. The components of this invention, typically described and shown in the accompanying drawings, can be arranged and designed in various different configurations.

[0047] like Figure 1 As shown, the present invention provides one embodiment: an alarm analysis method for power supply equipment fault tripping, the method comprising the following steps:

[0048] Acquire operational data of power supply and transformation equipment, including historical and real-time operational data;

[0049] Preprocessing of power supply and transformation equipment operation data;

[0050] Feature extraction is performed on historical operating data of power supply and transformation equipment based on kernel function. At the same time, the extracted feature vectors of power supply and transformation equipment are fused by combining random forest algorithm. Finally, a fusion regression model of tripping type of power supply and transformation equipment is constructed based on the combined averaging strategy.

[0051] The real-time operating data of power supply and transformation equipment is input into the tripping type fusion regression model of power supply and transformation equipment for calculation, and the tripping fault analysis results of power supply and transformation equipment are output.

[0052] In this embodiment, the data currently used for tripping cause analysis mainly comes from fault report data from the SCADA system, equipment operation data, setting data, live-line derivation conclusion data, and train operation data from the train dispatching and command system. This data may be affected by factors such as user operations, the quality of the acquisition device, the environment, and the data transmission method, resulting in a large amount of interference data. Furthermore, due to differences in device manufacturers and data sources, the types and formats of the data vary, which poses difficulties for traction substation feeder switch tripping analysis. Therefore, to improve the reliability and usability of the analysis data and reduce the complexity of the tripping analysis data, this invention proposes data processing for large amounts of high-dimensional heterogeneous data to improve the quality of the tripping analysis data.

[0053] like Figure 2 As shown, the preprocessing of power supply and transformation equipment operation data includes: data cleaning of power supply and transformation equipment operation data, and data transformation of power supply and transformation equipment operation data.

[0054] Specifically, data processing mainly consists of four steps:

[0055] Data cleaning: This involves cleaning the received data to eliminate any inconsistencies, errors, or invalid data. It ensures the accuracy and consistency of the data. Data cleaning mainly includes the following:

[0056] Remove report data that is irrelevant to or duplicates the switch tripping data, and retain only the fault reports generated by the feeder protection device and the fault location device;

[0057] Remove fault reports containing experimental information;

[0058] Clean up relevant equipment operation data caused by communication equipment failure and protection device malfunction;

[0059] Clear incomplete equipment operation data (such as missing essential data like unit and equipment information);

[0060] Clear setting data that has missing ratio coefficients or incomplete setting description information;

[0061] Clear train operation data that is unrelated to the current switch tripping event, that is, exclude train operation data of other lines that have no substation relationship with the line where the current tripping switch is located;

[0062] Missing equipment operation telemetry data are filled using linear interpolation or Newton interpolation, and missing equipment operation telemetry data are filled using historical values. Principal component analysis is used to extract trip-related features based on a large amount of historical trip fault report data.

[0063] Data transformation: This involves converting cleaned data in different formats into a unified form that facilitates analysis and management, and performing a uniform type conversion on relevant attribute values ​​to ensure that the transformed data has a consistent data format and numerical type. Data transformation mainly includes the following:

[0064] Convert telemetry values ​​with units in the fault report to values ​​with uniform units (based on the units required for analysis), and convert the description values ​​of remote signaling points to coded values.

[0065] Convert the constant quadratic value into a linear value according to the corresponding ratio coefficient;

[0066] Convert the live-line derivation conclusions into remote signaling codes;

[0067] Convert the telemetry values ​​in the equipment operation data to a unified unit (based on the unit required for analysis).

[0068] Data Fusion: The transformed data is categorized according to time and equipment dimensions, and then assembled into corresponding models based on these categories. First, fault report data, equipment operation data, setpoints, live-line derivations, and train operation data are fused by equipment. Kernel Principal Component Analysis (KPCA) is employed, introducing a kernel function to map the inner product results of the multi-source data in the low-dimensional space, transforming nonlinearity into linear operations in the high-dimensional space. Based on the maximum kernel principal component and cumulative variance contribution rate, feature vectors within a 2-second range before and after the feeder switch tripping time are effectively extracted. Second, the Random Forest Algorithm is used to learn the patterns of historical feature vector data through multiple decision trees. A combination strategy and ensemble method are employed to fuse the feature results from all decision tree analyses, generating a comprehensive data model.

[0069] In the above specific implementation, the feature extraction based on the kernel function of historical operating data of power supply and transformation equipment is specifically as follows:

[0070] A feature matrix is ​​constructed based on the preprocessed historical operating data of power supply and transformation equipment;

[0071] Calculate the mean of each feature in the feature matrix, and then decenter each feature to obtain the decentralized feature matrix.

[0072] The eigenvectors are mapped to the first-dimensional space using a Gaussian kernel function, the kernel matrix is ​​centered, and its covariance matrix is ​​calculated.

[0073] Solve the covariance matrix, calculate the variance contribution rate of each eigenvalue, calculate the cumulative contribution rate based on the variance contribution rate of each eigenvalue, and select multiple principal components based on the cumulative contribution rate;

[0074] The selected principal components are mapped to the second-dimensional space, and the power supply and transformation equipment feature vectors in the historical operation data of the power supply and transformation equipment are extracted as the final output.

[0075] In implementation, fault report data (feeder measured voltage and current, feeder protection voltage and angle, current and angle, impedance, impedance angle, second harmonic content, frequency, fault distance, kilometer marker), equipment operation data (circuit breaker open / closed status, feeder current, feeder voltage), setting values, live-line derivation (disconnector energized status, circuit breaker energized status), and train operation data (locomotive speed, number of vehicles, load power consumption) are used as feature inputs for kernel principal component analysis (KPCA): input vector X = [X1 X2 … X n ], where X i = [x1 x2 … x m ] T n refers to the total number of feature types, and m refers to the total number of data points for each feature type.

[0076] The input vector x samples are passed through a Gaussian kernel function K(X). i X j Mapped to a higher-dimensional space, i.e.

[0077]

[0078] Centralized processing will be completed. Calculate the covariance matrix and perform principal component analysis in high-dimensional space:

[0079]

[0080] The eigenvalue is λ, and v is the eigenvector corresponding to λ. Each eigenvector is a subspace mapped by a nonlinear function. The largest eigenvector corresponding to the largest eigenvalue is the largest kernel principal component, which forms a new dimensionality-reduced space, where ai are the eigenvalue coefficients.

[0081]

[0082] Introducing the kernel matrix K, i.e. K ij =K(X) i X j), calculate the cumulative variance contribution rate, select the feature data that meets the expected percentage of cumulative contribution, and then output the feature data. n It is an identity matrix of n times n.

[0083] λn=(KI n K-KI n +I n KI n )

[0084] KPCA can output 2S fault report data before and after tripping (feeder measured voltage and current, impedance angle), equipment operation data (circuit breaker open / closed status, feeder current, feeder voltage), setting values, live derivation (circuit breaker live conclusion), and train operation data (number of vehicles) feature vectors.

[0085] In this embodiment, the data fusion of the extracted feature vectors of power supply and transformation equipment using the random forest algorithm specifically involves:

[0086] The data in the feature vector of the power supply and transformation equipment is used as the original dataset, and the Bootstrap sampling method is used to randomly sample the original dataset with replacement according to a set ratio, and this process is repeated multiple times to generate multiple training sample sets.

[0087] For each training sample set, construct a CART decision tree:

[0088] Iterate through each feature and its segmentation point in the training sample set;

[0089] The training sample set is divided into a left sub-region and a right sub-region according to the segmentation criteria;

[0090] Calculate the optimal regression values ​​for the left and right sub-regions respectively, and calculate the segmentation loss function based on the optimal regression values;

[0091] Select the feature and split point that minimizes the loss function as the optimal splitting scheme for the current node;

[0092] Repeat the above segmentation process for the left and right sub-regions until the maximum number of iterations is reached to generate a complete decision tree. Test each decision tree with an unsampled set and combine each decision tree into a random forest model.

[0093] For each trip type:

[0094] Collect the analysis results of each decision tree for the same input;

[0095] Combine and average the results of each analysis;

[0096] The averaged result is used as the final analysis value for this trip type;

[0097] Summarize all tripping types and complete the construction of a fusion regression model for tripping types of power supply and transformation equipment.

[0098] In implementation, the p feature vector data corresponding to each tripping type of the feeder switch from the KPCA output are used, and the k data points of each feature are used as the original dataset Y = [Y1 Y2 … Y]. n The system uses Bootstrap to randomly select samples with replacement at a certain number of times and a ratio of 2 / 3. n represents the number of tripping types.

[0099]

[0100] Create a training set of N sample decision trees, divide it into two parts R1 and R2 by feature splitting, and train to generate N CART regression trees. Find the optimal feature according to the principle of minimizing loss Y. Where C1 is the regression value of N1 regression trees and C2 is the regression value of N2 regression trees.

[0101]

[0102] Where Y is the loss function value, R1 and R2 are the number of samples in the left and right sub-regions, respectively, j is the feature index, s is the value of the split point, and x is the value of the split point. i y i These are the feature vector and the true value of the i-th sample, respectively.

[0103] For the remaining 1 / 3 of the samples, tests were performed and OOB error was calculated.

[0104] A combined averaging strategy was adopted to construct the data fusion regression model required for each type of tripping.

[0105]

[0106] Where G(X) is the final analysis value of the fusion model for input X, N is the total number of decision trees, and f b (X) is the analysis value of the b-th decision tree for the input X, w b Let be the weight coefficients of the b-th decision tree, ∈ b Let α be the analysis error of the b-th decision tree on the training samples, and let α be the error decay coefficient. Using the generated integrated model data as input for tripping analysis, specific tripping fault types can be located, and live-line operation analysis and tripping cause analysis can be performed.

[0107] In a further implementation of this embodiment, to improve the accuracy and reliability of trip analysis, this invention proposes a method for determining the energized operating status of equipment by combining the open / closed status and current data of the corresponding incoming-side switch in the SCADA system before the feeder switch trips. The key technology of this method is a topology analysis algorithm. This algorithm, based on the closed status of adjacent switching equipment within the traction substation, uses the principle of energization derivation to analyze the current flow direction between the switching equipment and adjacent equipment, thereby determining the energized state of the traction substation equipment. According to the substation main wiring diagram and current flow sequence, the energized state of the incoming-side switch can deduce the energized state of the feeder switch before tripping, thus distinguishing between the energized operating state and standby state of the switching equipment and eliminating dirty trip data generated by the protection device when the traction substation equipment is not energized.

[0108] Procedure 1: Obtain the feeder voltage from the fault report at the time of feeder switch tripping, determine if the feeder voltage meets the undervoltage condition, and check if the feeder voltage in the SCADA system also meets the undervoltage condition. If both meet the undervoltage condition, a short circuit fault, traction substation fault, or voltage / current secondary circuit fault is initially suspected. If the voltage is lower than normal but the undervoltage condition is not triggered, it may be due to locomotive current drawover overload or substation line fault.

[0109] Step 2: Obtain the feeder current collected in the fault report at the moment the feeder switch trips, and determine whether the difference between the feeder current of the T-line and F-line and the overcurrent value exceeds the threshold. If the current is higher than the normal value, but the difference does not exceed the threshold, it is initially suspected that there is a locomotive current drawover overload, short circuit fault, or voltage and current secondary circuit fault on the feeder side.

[0110] Step 3: Obtain locomotive operation data at the moment the feeder switch trips, including the number of trains and their speed. If the number of trains exceeds the limit and the speed is too high, and the feeder current exceeds the limit, it is suspected that there is a fault of locomotive current drawover, causing the feeder switch to trip.

[0111] Step 4: Obtain the action flag, impedance angle, and fault distance from the fault report at the time of the feeder switch trip. Based on the protection signal (action flag) generated by the protection action, the feeder switch trip type can be preliminarily determined. If the feeder switch trip type is 'current increment' protection, a short circuit or line fault is initially suspected. If the protection type generated by the fault location device is "distance protection," a short circuit or secondary circuit fault may exist. If the impedance angle is less than... This indicates an abnormal phase difference between current and voltage, suggesting possible locomotive overload or a substation line fault. When the cause of the trip is unclear, the fault distance measured by fault location can be used to determine whether the trip is caused by a problem with equipment or secondary circuits within the traction substation, or by a line fault.

[0112] The fault report shows that the feeder voltage and equipment operation data meet the undervoltage condition, which may indicate a load change or a fault within the substation system. The significant increase in feeder current in the fault report may be related to train operation or a short circuit within the substation system. Since the load current increment is small in a short time (ms-level), while the short-circuit current increment is large in a short time (ms-level), and the protection action when the feeder switch trips is not "current increment" protection, a short circuit-related trip can be ruled out. The number and speed of locomotives operating beyond the normal range may cause undervoltage, increased current, and overcurrent protection tripping. Based on the above criteria, the cause of the feeder switch tripping can be diagnosed as an overload fault in the high-speed train's current draw.

[0113] When a feeder switch trips, the protection device's activation indicator is "current increment protection," indicating a significant increase in current within a short period, which may be due to a short circuit in the substation. A significant increase in feeder current reported in the fault report may be related to train operation or an internal short circuit in the substation system. If the feeder voltage reported as not reaching an undervoltage state and the impedance angle is less than the specified value, it may be due to the addition of capacitive equipment in the system or a fault in the substation equipment. The actual number and speed of trains operating on the traction network are within normal ranges, and the fault distance measured by the fault location device is within the specified range. Based on the above criteria, the cause of the feeder switch tripping can be diagnosed as a substation fault.

[0114] The fault report indicates that the feeder voltage meets the undervoltage condition, which may be due to load changes or a fault within the substation system; the feeder load current is within the normal range, ruling out tripping caused by load changes; when the feeder switch trips, the protection device's action flag is "Distance I," reflecting the distance between the fault point and the protection installation point; based on the above criteria, the cause of the feeder switch tripping can be diagnosed as a secondary circuit fault.

[0115] The system will push the trip cause conclusion, which includes information such as trip time, tripped line, tripped station, tripped equipment, trip cause, and handling suggestions, to the relevant system.

[0116] In another embodiment of the present invention: an alarm analysis system for power supply equipment fault tripping includes:

[0117] The data acquisition module acquires the operating data of the power supply and transformation equipment, including historical operating data and real-time operating data;

[0118] The preprocessing module preprocesses the operating data of power supply and transformation equipment;

[0119] The feature fusion module extracts features from historical operating data of power supply and transformation equipment based on kernel functions. It also combines the extracted feature vectors of power supply and transformation equipment with random forest algorithm to fuse the data, and builds a fusion regression model of power supply and transformation equipment tripping type based on the combined averaging strategy.

[0120] The tripping type early warning module inputs real-time operating data of power supply and transformation equipment into the tripping type fusion regression model of power supply and transformation equipment for calculation, and outputs the tripping fault analysis results of power supply and transformation equipment.

[0121] The above are merely preferred embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An alarm analysis method for power supply equipment fault tripping, characterized in that, The steps of this method include: Acquire operational data of power supply and transformation equipment, including historical and real-time operational data; Preprocessing of power supply and transformation equipment operation data; Feature extraction is performed on historical operating data of power supply and transformation equipment based on kernel function. At the same time, the extracted feature vectors of power supply and transformation equipment are fused by combining random forest algorithm. Finally, a fusion regression model of tripping type of power supply and transformation equipment is constructed based on the combined averaging strategy. The real-time operating data of power supply and transformation equipment is input into the tripping type fusion regression model of power supply and transformation equipment for calculation, and the tripping fault analysis results of power supply and transformation equipment are output.

2. The alarm analysis method for power supply and transformation equipment fault tripping according to claim 1, characterized in that, The preprocessing of power supply and transformation equipment operation data includes: data cleaning of power supply and transformation equipment operation data, and data transformation of power supply and transformation equipment operation data.

3. The alarm analysis method for power supply and transformation equipment fault tripping according to claim 2, characterized in that, The feature extraction based on kernel functions of historical operating data of power supply and transformation equipment is specifically as follows: A feature matrix is ​​constructed based on the preprocessed historical operating data of power supply and transformation equipment; Calculate the mean of each feature in the feature matrix, and then decenter each feature to obtain the decentralized feature matrix. The eigenvectors are mapped to the first-dimensional space using a Gaussian kernel function, the kernel matrix is ​​centered, and its covariance matrix is ​​calculated. Solve the covariance matrix, calculate the variance contribution rate of each eigenvalue, calculate the cumulative contribution rate based on the variance contribution rate of each eigenvalue, and select multiple principal components based on the cumulative contribution rate; The selected principal components are mapped to the second-dimensional space, and the power supply and transformation equipment feature vectors in the historical operation data of the power supply and transformation equipment are extracted as the final output.

4. The alarm analysis method for failure trip of a power transformation device according to claim 3, characterized by, The process of combining the extracted feature vectors of power supply and transformation equipment with the random forest algorithm for data fusion is as follows: The data in the feature vector of the power supply and transformation equipment is used as the original dataset, and the Bootstrap sampling method is used to randomly sample the original dataset with replacement according to a set ratio, and this process is repeated multiple times to generate multiple training sample sets. For each training sample set, construct a CART decision tree: Iterate through each feature and its segmentation point in the training sample set; The training sample set is divided into a left sub-region and a right sub-region according to the segmentation criteria; Calculate the optimal regression values ​​for the left and right sub-regions respectively, and calculate the segmentation loss function based on the optimal regression values; Select the feature and split point that minimizes the loss function as the optimal splitting scheme for the current node; Repeat the above segmentation process for the left and right sub-regions until the maximum number of iterations is reached to generate a complete decision tree. Test each decision tree with an unsampled set, combine each decision tree into a random forest model, and complete the training process of the random forest model.

5. The alarm analysis method for failure trip of a power transformation device according to claim 4, characterized by, The fusion regression model for tripping types of power supply and transformation equipment, based on the combined average strategy, is as follows: For each trip type: Collect the analysis results of each decision tree for the same input; Combine and average the results of each analysis; The averaged result is used as the final analysis value for this trip type; Summarize all tripping types and complete the construction of a fusion regression model for tripping types of power supply and transformation equipment.

6. The alarm analysis method for power supply and transformation equipment fault tripping according to claim 5, characterized in that, The loss function is specifically calculated using the following formula: Where Y is the loss function value, R1 and R2 are the number of samples in the left and right sub-regions, respectively, j is the feature index, s is the value of the split point, and x is the value of the split point. i Let and yi be the feature vector and the true value of the i-th sample, respectively.

7. The alarm analysis method for power supply and transformation equipment fault tripping according to claim 6, characterized in that, The calculation formula for the fusion regression model of tripping types of power supply and transformation equipment is as follows: Where G(X) is the final analysis value of the fusion model for input X, N is the total number of decision trees, and f b (X) is the analysis value of the b-th decision tree for the input X, w b Let be the weight coefficients of the b-th decision tree, ∈ b Let be the analysis error of the b-th decision tree on the training samples, and α be the error decay coefficient.

8. An alarm analysis system for power supply equipment fault tripping, characterized in that, include: The data acquisition module acquires the operating data of the power supply and transformation equipment, including historical operating data and real-time operating data; The preprocessing module preprocesses the operating data of power supply and transformation equipment; The feature fusion module extracts features from historical operating data of power supply and transformation equipment based on kernel functions. It also combines the extracted feature vectors of power supply and transformation equipment with random forest algorithm to fuse the data, and builds a fusion regression model of power supply and transformation equipment tripping type based on the combined averaging strategy. The tripping type early warning module inputs real-time operating data of power supply and transformation equipment into the tripping type fusion regression model of power supply and transformation equipment for calculation, and outputs the tripping fault analysis results of power supply and transformation equipment.

Images