Line loss anomaly identification method and device based on semi-supervised learning and storage medium

Abstract

The application discloses a line loss anomaly identification method and device based on semi-supervised learning and a storage medium, wherein XGBoost is used to take the labeled sample as an initial training set to obtain an initial prediction model, the initial prediction model is used to add pseudo labels to the unlabeled sample, then the samples with high possibility of correct labeling are selected by calculating Mahalanobis distance and added to the labeled sample to form a new training set, and iterative training is performed until the model converges. The application also discloses a device capable of realizing the method, including the device and the storage medium. From the perspective of machine learning, the application identifies the distribution network data according to whether there is line loss anomaly and the type of line loss anomaly. Based on the use of the application, the number of samples manually labeled by operation and inspection personnel can be greatly reduced, the application has practicability, and the application range is wide.

Description

Technical Field

[0001] This invention belongs to the field of line loss anomaly identification technology, and in particular relates to a line loss anomaly identification method, device and storage medium based on semi-supervised learning. Background Technology

[0002] With economic and social development, the scale of my country's power distribution network has become increasingly large, and the number of operating power equipment has grown rapidly. Power losses in the operation of the distribution network can be divided into two main categories: technical line losses and management line losses. Technical line losses consist of iron and copper losses from electrical equipment in the power grid, as well as losses generated by distribution lines. Management line losses refer to losses caused by abnormal power consumption behavior at the distribution side of the power user end. The distribution network line loss rate is an important performance indicator for power companies, reflecting the planning, design, operation, and management level of my country's power grid. Therefore, conducting abnormal line loss detection is a crucial step for power companies in reducing losses.

[0003] In recent years, reducing power loss has become a research hotspot both domestically and internationally, with comprehensive and accurate line loss diagnosis being crucial. Firstly, it provides theoretical and practical basis for power sectors to analyze line loss composition, formulate loss reduction measures, and promote power supply companies to tap their development potential and improve economic efficiency. Secondly, it optimizes power grid planning and design schemes, promoting the construction and transformation of distribution networks. Thirdly, it promotes stronger operation management, ensuring the distribution network operates economically. Fourthly, it helps establish distribution network line loss indicators, saving energy, improving power supply capacity, and alleviating power shortages. Existing technologies mostly employ supervised deep learning models, but fully supervised models require labeling all samples before inputting them into the training set, significantly increasing data processing workload. Therefore, an accurate and objective identification algorithm with lower requirements for labeled samples is needed for line loss identification. Summary of the Invention

[0004] The purpose of this invention is to provide a semi-supervised learning-based method for identifying line loss anomalies, so as to identify massive unlabeled distribution network line loss anomalies based on a small portion of already labeled distribution network data, and improve the accuracy and objectivity of existing methods.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is a method for identifying line loss anomalies based on semi-supervised learning, comprising the following steps:

[0006] S1. Based on whether the original distribution network line loss data has a label for judging line loss anomalies, the data is divided into labeled sample datasets and unlabeled sample datasets.

[0007] S2. Using the labeled samples in the dataset as the training set, train the XGBoost model to obtain the initial evaluation model;

[0008] S3. Use the initial evaluation model obtained from training to evaluate the unlabeled samples, and use the evaluation results as pseudo-labels for the samples.

[0009] S4. Based on the pseudo-label sample selection strategy, select pseudo-label samples based on Mahalanobis distance and add them to the training set. Use the new training set to retrain the XGBoost model to obtain a new evaluation model.

[0010] S5. Repeat the process from S2 to S3 until the model converges, then stop iterative training of the model and obtain the final evaluation model.

[0011] S6. Input the unlabeled transformer substation line loss related data into the trained self-trained semi-supervised learning model to diagnose and identify the distribution network line loss samples.

[0012] Furthermore, the data in the dataset includes distribution network topology parameters, power equipment parameters, transformer three-phase voltage, current, power, and load data; wherein, the sampling interval for each data is 15 minutes.

[0013] Furthermore, in S2, the training of the XGBoost model is specifically performed as follows:

[0014] The dataset includes an unlabeled sample set S = {x1', x2', ..., x} of the power distribution network. m '} and the labeled sample set X={X Tr ,X Te}, where X Tr and X Te These are labeled training sample sets and test sample sets, respectively.

[0015]

[0016]

[0017] in, For the sample input vector, For sample labels, Given a labeled test sample input vector, Labels for labeled test samples;

[0018] The labeled training sample set X Tr The XGBoost model is input for training, and the learner parameters are adjusted through grid search to obtain the initial evaluation model XGB1.

[0019] Furthermore, S3 specifically includes:

[0020] Using the initial evaluation model XGB1 on the unlabeled sample set S={x1',x2'…x mThe unlabeled sample set is classified to obtain a pseudo-label set T = {y1', y2', ..., y}. m The pseudo-labeled sample set S' = {(x1',y1'),(x2',y2')...(x} is formed by '}. m ',y m ')}, where y m ' is an unlabeled sample, x m ' represents the corresponding pseudo-tag.

[0021] Furthermore, the specific process of S4 is as follows:

[0022] For labeled training sample sets Based on the sample labels, for labeled sample x of category j ij Where i represents the i-th sample in the j-th class, and its sample mean is calculated. The covariance matrix C of the labeled sample matrix j :

[0023]

[0024]

[0025] n j The number of labeled samples.

[0026] For the pseudo-labeled sample set S'={(x1',y1'),(x2',y2')…(x m ',y m Each pseudo-label sample (x) in ')} m ',y m '), according to y m 'Determine the category described in the pseudo-label sample and calculate x' m 'The average value x of the labeled training samples of category j' j Mahalanobis distance between

[0027]

[0028] Choose Mahalanobis distance Pseudo-labeled samples x that are less than the threshold θ ji Add labeled samples to form a new training set. Using the new training set The XGBoost model was retrained to obtain a new evaluation model, XGB2.

[0029] Furthermore, in S5, a t-test is used at each iteration to determine whether the classification probability has changed. Before convergence, the probability should change with each iteration; if the t-test shows no change after three consecutive iterations, the model is considered converged, and the final evaluation model XGB is obtained. p .

[0030] Furthermore, the specific steps of the t-test are as follows:

[0031] S501. Calculate the sample mean and standard deviation: Calculate the classifier XGB respectively. p and XGB p+1 The classification probability sample mean and sample standard deviation;

[0032] S502. Calculate the t value:

[0033]

[0034] Where, mean p and mean p+1 The classifiers are XGB and GGB respectively. p and XGB p+1 The classification probability sample mean, std p and std p+1 The classifiers are XGB and GGB respectively. p and XGB p+1 The standard deviation of the classification probability samples, n p and n p+1 The classifiers are XGB and GGB respectively. p and XGB p+1 The number of samples;

[0035] S503. Find the critical value: Based on the set significance level, find the t-distribution table or use statistical software to calculate the t-critical value.

[0036] S504. Determine significance: Compare the calculated t-value with the critical value. If the t-value is greater than the critical value, reject the null hypothesis, indicating a change between the classification probabilities. If the t-value is less than or equal to the critical value, accept the null hypothesis, indicating no change between the classification probabilities.

[0037] Another objective of this invention is to provide a storage medium for identifying and storing line loss anomalies based on semi-supervised learning, comprising:

[0038] One or more processors;

[0039] Memory, used to store one or more programs;

[0040] When one or more of the programs are executed by one or more of the processors, the one or more of the processors implement a semi-supervised learning-based method for identifying line loss anomalies as described in any one of claims 1-7;

[0041] The storage medium embeds computer-executable instructions, which, when executed by a computer processor, are used to perform a semi-supervised learning-based line loss anomaly identification method as described in any one of claims 1-7.

[0042] The beneficial effects of this invention are:

[0043] This invention starts with the typological characteristics of line loss anomalies, uses the XGBoost classification method and calculates Mahalanobis distance to evaluate the sample classification results, which is more objective and accurate compared with existing line loss anomaly identification methods. Using Mahalanobis distance as a pseudo-label sample selection strategy can further improve the rationality and accuracy of pseudo-label sample selection. The semi-supervised self-learning strategy for classification greatly reduces the number of labeled samples required compared with traditional classification techniques, significantly reducing the workload of manually labeling anomaly samples when identifying line loss anomalies, making this method adaptable to line loss identification techniques in more scenarios. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart of the method for identifying abnormal line loss;

[0046] Figure 2 These are load curves under six abnormal modes. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] like Figure 1This invention provides a method, device, and storage medium for identifying line loss anomalies based on semi-supervised learning. The method of this invention is based on the labeling of the original distribution network line loss data, and divides the data into labeled sample datasets and unlabeled sample datasets. The steps of building a self-trained semi-supervised learning model include initial evaluation of model building, evaluation of unlabeled samples, selection of pseudo-labeled samples, and model retraining.

[0049] S1. Using the labeled samples in the dataset as the training set, train the XGBoost model to obtain the initial evaluation model.

[0050] The dataset includes distribution network topology parameters, power equipment parameters, transformer three-phase voltage, current, power, and load data (one sample covers one day's load data, with a sampling interval of 15 minutes). Based on records from power grid operators, each sample is identified and flagged to determine if it represents an anomaly in line loss.

[0051] The dataset includes an unlabeled sample set S = {x1', x2', ..., x...} m '} and the labeled sample set X={X Tr ,X Te}, where X Tr and X Te These are labeled training sample sets and test sample sets, respectively.

[0052]

[0053]

[0054] in, For the sample input vector, For sample labels, Given a labeled test sample input vector, Labels for labeled test samples.

[0055] The labeled training sample set X Tr The XGBoost model is input for training, and the learner parameters are adjusted through grid search to obtain the initial evaluation model XGB1.

[0056] Grid search process: The learning rate and tree depth are uniformly selected within their respective ranges. All possible values ​​of these parameters are arranged and combined to form a parameter grid. For each parameter combination in the grid, the model is trained and its performance is evaluated. The parameter combination with the best performance is selected as the final parameter values ​​for the learner.

[0057] S2. Use the initial evaluation model XGB1 obtained from training to evaluate the unlabeled samples, and use the evaluation results as pseudo-labels for the samples.

[0058] Using the initial evaluation model XGB1 on the unlabeled sample set S={x1',x2'…x m The unlabeled sample set is classified to obtain a pseudo-label set T = {y1', y2', ..., y}. m The pseudo-labeled sample set S' = {(x1',y1'),(x2',y2')...(x} is formed by '}. m ',y m ')}, where y m ' is an unlabeled sample, x m ' is the corresponding pseudo-tag;

[0059] S3. Based on the pseudo-label sample selection strategy, select samples from the unlabeled samples that have a higher probability of being correctly labeled and add them to the labeled samples to form a new training set. Utilize this new training set... The XGBoost model was retrained to obtain a new evaluation model, XGB2.

[0060] The pseudo-labeled sample selection strategy based on Mahalanobis distance is as follows:

[0061] For labeled training sample sets Based on the sample labels, for labeled sample x of category j ij Where i represents the i-th sample in the j-th class of samples, and its sample mean is calculated. The covariance matrix C of the labeled sample matrix j .

[0062]

[0063]

[0064] Where, n j This represents the number of labeled samples.

[0065] For the pseudo-labeled sample set S'={(x1',y1'),(x2',y2')…(x m ',y m Each pseudo-label sample (x) in ')} m ',y m '), according to y m 'Determine the category described in the pseudo-label sample and calculate x' m 'The average value x of the labeled training samples of category j' j Mahalanobis distance between

[0066]

[0067] Choose Mahalanobis distance Pseudo-labeled samples x that are less than the threshold θ ji 'Add labeled samples to form a new training set X' Tr new ={(x1,y1),…,(x k ,y k ),(x1',y1'),…,(x t ',y t ')}.

[0068] S4. Repeat steps S2-S3 until the model meets the required specifications, then stop the iterative training. During each iteration, use a t-test to determine if the classification probability has changed. Before convergence, the probability should change with each iteration. Three consecutive t-tests showing no change indicate model convergence. Therefore, the iteration termination condition is model convergence or reaching the maximum number of iterations. This yields the final distribution network line loss anomaly assessment model XGB. N And using the final evaluation model XGB N For the test sample set X Te The results of the distribution network line loss assessment are obtained by classification.

[0069] The t-test is a statistical inference method used to determine whether there is a significant difference between the means of two samples. In classification problems, the t-test can be used to compare whether there is a significant difference between the classification probabilities of two classifiers on a given test set.

[0070] In this case, the purpose of the t-test is to compare the classifiers XGB. p and XGB p+1 To determine whether there is a significant difference in classification probabilities, the specific steps are as follows:

[0071] Calculate the sample mean and standard deviation: calculate the classifier XGB respectively. p and XGB p+1 The classification probability sample mean (mean) p and mean p+1 ) and sample standard deviation (std p and std p+1 ).

[0072] Calculate the t-value: Use the following formula to calculate the t-value.

[0073]

[0074] Where, mean p and mean p+1 The classifiers are XGB and GGB respectively. p and XGB p+1 The classification probability sample mean, std p and std p+1The classifiers are XGB and GGB respectively. p and XGB p+1 The standard deviation of the classification probability samples, n p and n p+1 The classifiers are XGB and GGB respectively. p and XGB p+1 The number of samples.

[0075] Finding the critical value: Based on the set significance level (usually 0.05), find the t-distribution table or use statistical software to calculate the t-critical value.

[0076] Determining significance: Compare the calculated t-value with the critical value. If the t-value is greater than the critical value, the null hypothesis can be rejected, indicating a significant difference between the classification probabilities. If the t-value is less than or equal to the critical value, the null hypothesis is accepted, indicating no significant difference between the classification probabilities.

[0077] S5. Input the unlabeled transformer substation line loss related data into the trained self-trained semi-supervised learning model to diagnose and identify the distribution network line loss samples.

[0078] This invention uses a self-trained semi-supervised learning model to diagnose and identify abnormal line loss samples in distribution networks. Through analysis of a large number of actual transformer substations' measured line loss rates over multiple months, it can be found that abnormal line loss rates can be categorized into the following situations:

[0079] (1) Line loss rate is zero: The line loss rate is zero for the whole year or for some months. According to the line loss refinement platform program settings, the line loss rate is set to zero when the power supply is zero;

[0080] (2) The line loss rate is stable and too high: the line loss rate of the transformer area usually exceeds 50% continuously, or even reaches 100%;

[0081] (3) The line loss rate is stable and negative: the line loss rate is consistently negative;

[0082] (4) The line loss rate was abnormal for a few periods of time;

[0083] (5) Fluctuating line loss rate: The line loss rate of the transformer area fluctuates, even sometimes positive and sometimes negative. This abnormal line loss rate is the result of a combination of factors.

[0084] The line load data in this embodiment of the invention originates from power load data of a certain region from January 1, 2009 to December 31, 2010 (one sampling point every 15 minutes, 96 points per day, unit: kW), including data such as line topology, load conditions at sampling points, and meteorological information. This invention selects load data from January 2009 to December 2010 and temperature data collected from nearby meteorological stations. Through data cleaning, missing and abnormal data are removed, and some load curves are modified using the six anomaly patterns shown in Table 1 to simulate abnormal data. Figure 2 The load curves for six abnormal modes are shown. In the experimental setup, the proportion of labeled data was set to 20%, and the proportion of data with abnormal line loss was set to 25%. The proportions of each type of abnormal line loss are shown in Table 1.

[0085] Table 1 Six Abnormal Line Loss Patterns

[0086]

[0087] in, This is the modified daily load data. In anomaly modes 2, 3, and 4, this paper sets the coefficient parameters α and β. T (T = 1, 2, ..., 24) is randomly generated in [0.88, 0.98], and min_T takes the value 4.

[0088] This invention provides a computer device comprising: one or more processors and a memory for storing one or more computer programs; the programs include program instructions, and the processor executes the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, used to implement one or more instructions, specifically for loading and executing one or more instructions stored in a computer storage medium to implement the aforementioned method.

[0089] This invention provides a computer storage medium storing a computer program that, when executed by a processor, performs the methods described above. The storage medium can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0090] The device of the present invention includes the following modules that can be mounted on the aforementioned computer and storage medium:

[0091] The data acquisition module includes distribution network topology parameters, power equipment parameters, transformer three-phase voltage, current and power, load data (one sample covers one day's load data, with a sampling interval of 15 minutes) and a flag indicating whether line loss is abnormal.

[0092] A line loss anomaly identification device based on self-trained semi-supervised learning, comprising:

[0093] The first building module is used to train the XGBoost model. By inputting different labeled sample sets X of the distribution network, it obtains the mapping relationship between the corresponding input features and the judgment result of whether the output label is an anomaly in line loss.

[0094] The second building module is used to obtain pseudo-labels. The trained XGBoost model is trained on the test set data and obtains the corresponding pseudo-labels.

[0095] The third construction module is used to filter samples among unlabeled samples that are more likely to be correctly labeled, based on a pseudo-label selection strategy based on Mahalanobis distance.

[0096] The fourth building module is used to construct a new training set. It adds samples from the unlabeled samples that are more likely to be correctly labeled to the training set to form a new training set, and uses this to repeatedly train the XGBoost model.

[0097] Example

[0098] The accuracy of line loss identification in the method of this invention includes the accuracy of anomaly identification and the accuracy of anomaly type classification;

[0099] Anomaly detection accuracy; Assuming the actual label of the i-th sample is f(i), f(i) = 0 indicates that the sample has no line loss anomaly, and samples with line loss anomalies are labeled according to the anomaly type f(i) = j (j = 1, 2, 3, 4, 5, 6), the classification result based on the line loss type of the classifier is g(i) = q (q = 0, 1, 2, 3, 4, 5, 6), and S(i) = 1 indicates that a normal line loss sample is identified as normal or an abnormal line loss sample is identified as abnormal. Then the line loss anomaly detection rate K1 is calculated using the following formula:

[0100]

[0101]

[0102] Anomaly type classification accuracy: R(i) represents the correctness of classification based on whether the sample has abnormal line loss. R(i) = 1 indicates that the anomaly type of the sample is correctly classified. Therefore, the anomaly type classification accuracy K2 is shown in the following formula:

[0103]

[0104]

[0105] Table 2. Accuracy of the line loss identification method based on self-trained semi-supervised learning

[0106] index <![CDATA[Abnormal line loss recognition accuracy rate K1]]> <![CDATA[Abnormal line loss type classification accuracy rate K2]]> numerical values 0.84551 0.80478

[0107] Table 3 Comparison of Sample Fault Type Classification Results with Actual Situations

[0108] Sample number Actual Category Prediction Category Sample number Actual Category Prediction Category Sample number Actual Category Prediction Category 1 3 3 11 3 3 21 3 3 2 3 3 12 4 5 22 3 3 3 5 5 13 1 3 23 5 5 4 1 1 14 3 3 24 1 0 5 3 3 15 3 3 25 3 3 6 3 3 16 1 3 26 4 4 7 4 4 17 1 0 27 3 3 8 3 3 18 5 3 28 1 1 9 3 3 19 3 3 29 3 3 10 1 1 20 6 6 30 3 3

[0109] Table 2 shows the accuracy of line loss anomaly identification using the self-trained semi-supervised learning-based method. Table 2 clearly shows that the prediction accuracy of the self-trained semi-supervised learning-based line loss anomaly identification strategy is high. Table 3 shows a comparison between the identified anomaly types and the actual anomaly types for samples with anomalies. Based on the results shown in the tables, this method also has high accuracy in identifying fault types.

[0110] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0111] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A line loss anomaly identification method based on semi-supervised learning, characterized in that, The method comprises the following steps: S1, dividing the data into a labeled sample data set and an unlabeled sample data set based on whether the original power distribution network line loss data has a line loss anomaly judgment mark; S2, using the labeled samples in the data set as a training set to train the XGBoost model to obtain an initial evaluation model; S3, using the initial evaluation model obtained by training to evaluate the unlabeled samples, and the evaluation result is used as the pseudo label of the samples; S4, according to the pseudo label sample selection strategy, the pseudo label samples are selected based on the Mahalanobis distance to join the training set, and the XGBoost model is retrained using the new training set to obtain a new evaluation model; S5, repeating the process of S2-S3 until the model converges, stopping the model iteration training, and obtaining the final evaluation model; S6, inputting the unmarked area line loss related data into the trained self-training semi-supervised learning model to diagnose and identify the power distribution network line loss samples; In S2, the XGBoost model is trained as follows: The data set includes an unlabelled sample set of a power distribution network and a labelled sample set of the power distribution network wherein, and are respectively a labelled training sample set and a test sample set; ； ； wherein, is a sample input vector, is a sample label, is a labeled test sample input vector, is a labeled test sample label; a labeled training sample set The XGBoost model is trained and the parameters of the learner are adjusted by grid search to obtain an initial evaluation model XGB1. The specific process of S3 is as follows: Using the initial evaluation model XGB1 to classify the unlabeled sample set to obtain a pseudo-label set of the unlabeled sample set , and the pseudo-label sample set is constituted , wherein is an unlabeled sample, is a corresponding pseudo-label​ The specific process of S4 is as follows: a set of labeled training set samples , according to the sample label, the class of labeled samples where represents the th sample in the th class of samples, the sample mean value thereof is calculated and the covariance matrix of the labeled sample matrix : ； ； the number of labeled samples, For each pseudo-label sample in the pseudo-label sample set , according to the category of the pseudo-label sample , calculate the Mahalanobis distance between the pseudo-label sample and the average value of the labeled training samples with the category ;​​ ； selecting mahalanobis distance less than a threshold pseudo-labeled samples adding the labeled samples to form a new training set ; retraining the XGBoost model using the new training set to obtain a new evaluation model XGB2.

2. The line loss anomaly identification method based on semi-supervised learning according to claim 1, characterized in that, The data in the data set includes power distribution network topology structure parameters, power equipment parameters, transformer three-phase voltage and current power, and load data; wherein, the sampling interval of each data is 15 minutes.

3. The semi-supervised learning based line loss anomaly identification method according to claim 1, characterized in that, In the S5, with each iteration, use Check if the classification probability changes. Before convergence, the probability should change with each iteration. No change for 3 consecutive times The test determines that the model converges, and the final evaluation model XGB is obtained p .

4. The line loss anomaly identification method based on semi-supervised learning according to claim 3, characterized in that, The The specific steps of the test are as follows: S501、Calculate sample mean and standard deviation: Calculate the sample mean and sample standard deviation of the classification probability of the classifier XGB p and XGB p+1 respectively; S502, compute Value: ； wherein, and are the sample mean of the classification probabilities of the classifier XGB p and XGB p+1 , and are the sample standard deviation of the classification probabilities of the classifier XGB p and XGB p+1 , and are the sample size of the classifier XGB p and XGB p+1 , respectively. S503, finding a critical value: according to the set significance level, finding a distribution table or using statistical software to calculate a critical value; S504, judging significance: comparing the calculated t value with a critical value, when the value is greater than the critical value, the null hypothesis is rejected, indicating that there is a change between the classification probabilities, when the value is less than or equal to the critical value, the null hypothesis is accepted, indicating that there is no change between the classification probabilities.

5. A line loss anomaly identification based on semi-supervised learning and storage medium, characterized in that, It comprises: One or more processors; Memory for storing one or more programs; When one or more of the programs are executed by one or more of the processors, one or more of the processors implement a line loss anomaly identification method based on semi-supervised learning according to any one of claims 1-4; The storage medium is embedded with computer executable instructions for executing a line loss anomaly identification method based on semi-supervised learning according to any one of claims 1-4 when executed by a computer processor.

Images