Intelligent electric meter fault classification method based on meta learning

Abstract

The embodiment of the invention provides an intelligent electric meter fault classification method based on meta-learning, and the method comprises the steps: taking fault historical data of an intelligent electric meter under different classes as an input data set, and carrying out the division to obtain a plurality of second-class data sets; for each second-class data set, constructing an unbalanced classification model based on a meta-learning strategy, sampling and constructing subtasks from the second-class data sets, and training a meta-classifier by using the meta-learning strategy; performing boundary enhancement on each second-class data set based on a Bayesian imbalance influence index; performing fine tuning on the element classifier based on the boundary enhanced second-class data set to obtain a final second-class classifier; according to a classification result under each second-class data set, obtaining a fault category through hard voting; experimental results in an actual intelligent electric meter fault data set prove that compared with other fault classification methods, the method is better in effect, the best result is obtained on most of second-class data sets, and better universality is achieved.

Description

[Technical field]

[0001] The present invention relates to a smart meter fault classification method, and in particular to a smart meter fault classification method based on meta-learning. [Background technology]

[0002] In recent years, with the reform of the power market and the adjustment of relevant national energy policies, the interaction between users and the power market has become increasingly close. At the same time, with the continuous improvement of economic and technological levels, the demand for electricity in people's lives and work has continued to increase, and the requirements for the stability and reliability of power supply have become increasingly higher. In addition, various distributed power generation equipment has emerged in an endless stream, and it is difficult for traditional power networks to coordinate and meet the processes of power generation, transmission, and power consumption. In this environment, smart grids came into being. As the intelligent terminal of smart grids, smart meters, on the basis of traditional meters measuring power consumption, also have functions such as intelligent display, remote control, and power consumption monitoring, which play an important role in the stable operation of smart grids.

[0003] With the development of smart grids, the demand for smart meters is increasing, and the number of smart meter manufacturers has increased accordingly. Different manufacturers have different design solutions, component selection and process flows. Coupled with the complex production and application environments, the fault types of smart meters are complex and diverse. In addition, the frequency of occurrence of meters with different fault types is also different, and the fault data presents a multi-mode distribution characteristic, which further increases the difficulty of smart meter prediction. Therefore, the fault classification problem of smart meters is an unbalanced multi-classification problem.

[0004] According to the research of relevant literature at home and abroad, the "one-to-many" classification framework is the most commonly used and typical method for dealing with multi-classification problems. It converts multi-classification problems into multiple unbalanced binary classification problems, thereby reducing the overall difficulty of multi-classification problems. Existing unbalanced binary classification methods are mainly divided into two categories: data level and algorithm level. The data level method alleviates the decision bias phenomenon of the basic classifier by achieving a balance between samples of different categories, but it is often difficult to avoid problems such as information loss, noise introduction, and aggravated sample overlap, which limits the improvement of its classification performance. The algorithm level method starts from the model structure or the cost of misclassification, and effectively avoids the problems caused by eliminating or introducing samples incorrectly. However, for smart meter data sets, the distribution, imbalance ratio, and degree of sample overlap of samples in different two-class data sets are quite different. For data sets with large differences in these characteristics, it is impossible to find a suitable cost matrix or a set of suitable hyperparameters, resulting in the inability of the algorithm level method to maintain excellent performance and the low universality of the algorithm. Based on the above analysis, the present invention proposes a smart meter fault classification method based on meta-learning to improve the universality of the algorithm, thereby improving the classification performance on each two-class sub-dataset and the performance of smart meter fault classification. [Summary of the invention]

[0005] In view of this, the present invention proposes a smart meter fault classification method based on meta learning (Smart Meter Fault Classification Method Based on Meta Learning, SMFC-ML) to improve the performance of smart meter fault classification.

[0006] The present invention proposes a smart meter fault classification method based on meta-learning, which includes the following steps:

[0007] (1) Taking the fault history data of different categories of smart meters as the input data set, multiple second-category data sets are obtained, specifically:

[0008] The actual fault data set of smart meters is input. The samples in this data set include 19 characteristic variables, including the number of days of use, manufacturing unit, calibrated current, active power accuracy level, specification, overload multiple, voltage, wiring method, payment terminal production batch, new and old, rotation cycle, hardware version, card meter tripping mode, load curve, power outage meter reading, reverse phase sequence judgment, current loss judgment, rate, and meter mode. The fault category labels include 7 categories: appearance fault, power supply fault, software fault, clock fault, communication unit fault, metering unit fault, and storage unit fault. Each category of samples in the fault data set is traversed, and all samples under this category are regarded as minority class sample sets, and all samples of other categories are majority class sample sets, and the original data set is converted into 7 second-category data sets. For each of the second-category data sets, the data set can be described as follows:

[0009] X=[X min ,X maj ]

[0010] Among them, X is a two-category dataset, and x is defined as any sample in the dataset, that is, x∈X; X min For the minority class sample set, define x min is any sample in the data set, that is, x min ∈X min ;X maj is the majority class sample set, define x maj is any sample in the data set, that is, x maj ∈X maj ;

[0011] (2) For each two-category dataset, an unbalanced classification model based on a meta-learning strategy is constructed, and a meta-classifier θ based on a deep neural network is constructed. N independent and identically distributed subtasks are randomly sampled from the two-category dataset, where each subtask contains a support set and a query set, and the number of majority class samples and minority class samples in the subtask is the same, and the number of samples in the query set is greater than the number of samples in the support set. The subtask division process is as follows:

[0012] sub-task={spt,qry}

[0013] Among them, sub-task is the subtask, spt is the support set in the subtask, and qry is the query set in the subtask;

[0014] For each subtask, the meta-classifier is trained using only samples in the support set. The training process is as follows:

[0015]

[0016] in, is the classification loss of the meta-classifier on the i-th subtask support set, is the gradient descent algorithm used to optimize the meta-classifier, α is the learning rate of the meta-classifier on the subtask support set, is the meta-classifier optimized on the i-th subtask support set;

[0017] Since the samples in the subtask only account for a small part of the samples in the original data set, the classifier does not need to pay attention to the overall classification boundary, but only focuses on distinguishing several samples in the subtask, so that the classifier can fully explore the classification features of the limited samples in the support set, laying the foundation for the performance optimization of the classifier on the new classification task. At the same time, the number of samples of the two categories in the support set is the same, which reduces the possibility of decision deviation in the classifier;

[0018] The number of subtasks required to optimize the classifier each time is n. After training on n subtasks, the average loss of the classifier on the corresponding subtask query set samples is obtained, and the classifier is optimized using the average loss. The training process is as follows:

[0019]

[0020] Among them, β is the learning rate of the meta-classifier on the subtask support set;

[0021] Repeat the above process until all subtasks are trained and obtain a meta-classifier trained based on the meta-learning strategy.

[0022] (3) After the meta-classifier is trained based on the meta-learning strategy, the boundary of each two-class dataset is enhanced based on the Bayesian imbalance impact index (IBI3), so as to alleviate the decision deviation phenomenon of the meta-classifier during the fine-tuning process. Specifically:

[0023] For each minority class sample x, the number of minority class samples m among its K nearest neighbor samples is obtained according to the Euclidean distance, and the probability that the sample is classified as a minority class sample is f p (x) = m / k, where k is the number of K nearest neighbor samples, and the probability of being classified as a majority class sample is f n (x) = (km) / k, assuming that the imbalance ratio of the data is r = N n / N p , where N n is the number of majority class samples in the two-class dataset, N p is the number of minority class samples in the two-class data set, then the probability that the sample is predicted as a minority class sample by the classification model that is not affected by data imbalance can be expressed as f′ p (x) = r*m / k, the calculation process of the Bayesian imbalance impact index (IBI3) of the minority class sample x is as follows:

[0024]

[0025] Use the IBI3 value as the sampling probability of the minority class sample and sample the minority class sample x i Then, use K nearest neighbors to get its neighbor samples in the training set, and randomly select a sample from the neighbor samples After performing random feature interpolation, we get a new sample x new , the calculation process is as follows:

[0026]

[0027] Among them, x i,j Represents sample x i The value of the j-th dimension feature, Representation sample The value of the j-th dimension feature of , β is a random number between 0 and 1;

[0028] Repeat the above sampling process until the number of majority class samples and minority class samples in the two-class data set is the same, and obtain a boundary-enhanced balanced two-class data set X′;

[0029] (4) Based on the boundary-enhanced two-category dataset, the meta-classifier is fine-tuned for N2 rounds to obtain the final meta-classifier θ * ,θ * is the final classifier for the two-category dataset.

[0030] In the above method, in step (2), the structure of the meta-classifier is as follows:

[0031]

[0032] Among them, Dense Layer() is the fully connected layer construction function, ReLU represents the corresponding activation function, and x_dim is the number of feature dimensions of samples in the two-category dataset.

[0033] In the above method, in step (2), the values ​​of α and β are 1e-2 and 1e-3 respectively; the value of N is 5000, and the value of n is 4.

[0034] In the above method, in step (3), the value of k is 5.

[0035] In the above method, in step (4), the value of N2 is 10.

[0036] The meta-learning-based smart meter fault classification method improves the accuracy and recall rate of smart meter fault classification.

[0037] It can be seen from the above technical solutions that the present invention has the following beneficial effects:

[0038] In the technical solution implemented by the present invention, the classification performance of the meta-classifier when facing new tasks is enhanced by the classification loss of multiple query sets in the created meta-task, the universality of the algorithm is improved, and a boundary enhancement strategy based on the Bayesian imbalance influence index is designed to alleviate the influence of decision deviation caused by imbalanced category data during the fine-tuning process of the meta-learner. This method can improve the robustness of the model, thereby improving the accuracy and recall rate of smart meter fault classification.

Brief Description of the Drawings

[0039] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0040] Figure 1 It is a schematic diagram of the framework flow of the intelligent meter fault classification method based on meta-learning proposed in the present invention;

[0041] Figure 2 It is a flowchart for solving smart meter fault classification;

[0042] Figure 3 Detailed schematic diagram of the algorithm of the present invention. [Specific implementation method]

[0043] In order to better understand the technical solution of the present invention, the present invention is described in detail below with reference to the accompanying drawings.

[0044] It should be clear that the described embodiments of the invention are only part of the embodiments of the invention, not all of the embodiments. Based on the embodiments of the invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the invention.

[0045] The present invention provides a smart meter fault classification method based on meta-learning. In order to meet the fault classification of smart meters, the present invention uses the fault history data of smart meters under different categories as input data sets, and divides them into multiple two-category data sets; for each two-category data set, an unbalanced classification model based on a meta-learning strategy is constructed, and subtasks are sampled and constructed from the two-category data set and a meta-classifier is trained using a meta-learning strategy; the boundary of each two-category data set is enhanced based on the Bayesian imbalance influence index; the meta-classifier is fine-tuned based on the boundary-enhanced two-category data set to obtain the final two-category classifier; according to the classification results under each two-category data set, its fault category is obtained through hard voting.

[0046] Figure 1 The figure is a schematic diagram of the framework flow of the intelligent meter fault classification method based on meta-learning proposed in the present invention. The method comprises the following steps:

[0047] Step 101: taking fault history data of different categories of smart meters as input data sets, and dividing them into multiple second-category data sets, specifically:

[0048] The actual fault data set of smart meters is input. The samples in this data set include 19 characteristic variables, including the number of days of use, manufacturing unit, calibrated current, active power accuracy level, specification, overload multiple, voltage, wiring method, payment terminal production batch, new and old, rotation cycle, hardware version, card meter tripping mode, load curve, power outage meter reading, reverse phase sequence judgment, current loss judgment, rate, and meter mode. The fault category labels include 7 categories: appearance fault, power supply fault, software fault, clock fault, communication unit fault, metering unit fault, and storage unit fault. Each category of samples in the fault data set is traversed, and all samples under this category are regarded as minority class sample sets, and all samples of other categories are majority class sample sets, and the original data set is converted into 7 second-category data sets. For each of the second-category data sets, the data set can be described as follows:

[0049] X=[X min ,X maj ]

[0050] Among them, X is a two-category dataset, and x is defined as any sample in the dataset, that is, x∈X; X minFor the minority class sample set, define x min is any sample in the data set, that is, x min ∈X min ;X maj is the majority class sample set, define x maj is any sample in the data set, that is, x maj ∈X maj .

[0051] Step 102, for each two-category data set, for each two-category data set, construct an unbalanced classification model based on a meta-learning strategy, sample and construct subtasks from the two-category data set and train a meta-classifier using a meta-learning strategy, specifically:

[0052] Based on the unbalanced binary classification data set X=[X min ,X maj ], build and train an imbalanced classification model based on a meta-learning strategy,

[0053] Construct a meta-classifier θ based on a deep neural network, and construct N independent and identically distributed subtasks by randomly sampling from the two-category data set. Each subtask contains a support set and a query set, and the number of majority class samples and minority class samples in the subtask is the same. The number of samples in the query set is greater than the number of samples in the support set. The subtask division process is as follows:

[0054] sub-task={spt,qry}

[0055] Among them, sub-task is the subtask, spt is the support set in the subtask, and qry is the query set in the subtask;

[0056] For each subtask, the meta-classifier is trained using only samples in the support set. The training process is as follows:

[0057]

[0058] in, is the classification loss of the meta-classifier on the i-th subtask support set, is the gradient descent algorithm used to optimize the meta-classifier, α is the learning rate of the meta-classifier on the subtask support set, is the meta-classifier optimized on the i-th subtask support set;

[0059] Since the samples in the subtask only account for a small part of the samples in the original data set, the classifier does not need to pay attention to the overall classification boundary, but only focuses on distinguishing several samples in the subtask, so that the classifier can fully explore the classification features of the limited samples in the support set, laying the foundation for the performance optimization of the classifier on the new classification task. At the same time, the number of samples of the two categories in the support set is the same, which reduces the possibility of decision deviation in the classifier;

[0060] The number of subtasks required to optimize the classifier each time is n, that is, the number of proxy subtasks is n. After training on n subtasks, the average loss of the classifier on the corresponding subtask query set samples is obtained, and the classifier is optimized using the average loss. The training process is as follows:

[0061]

[0062] Among them, β is the learning rate of the meta-classifier on the subtask support set;

[0063] Repeat the above process until all subtasks are trained and obtain a meta-classifier trained based on the meta-learning strategy.

[0064] Step 103, based on the Bayesian imbalance impact index, the boundary of each second-class data set is enhanced, specifically:

[0065] For each minority class sample x, the number of minority class samples m among its K nearest neighbor samples is obtained according to the Euclidean distance, and the probability that the sample is classified as a minority class sample is f p (x) = m / k, where k is the number of K nearest neighbor samples, and the probability of being classified as a majority class sample is f n (x) = (km) / k, assuming that the imbalance ratio of the data is r = N n / N p , where N n is the number of majority class samples in the two-class dataset, N p is the number of minority class samples in the two-class data set, then the probability that the sample is predicted as a minority class sample by the classification model that is not affected by data imbalance can be expressed as f′ p (x) = r*m / k, the calculation process of the Bayesian imbalance impact index (IBI3) of the minority class sample x is as follows:

[0066]

[0067] Use the IBI3 value as the sampling probability of the minority class sample and sample the minority class sample x i Then, use K nearest neighbors to get its neighbor samples in the training set, and randomly select a sample from the neighbor samples After performing random feature interpolation, we get a new sample xnew , the calculation process is as follows:

[0068]

[0069] Among them, x i,j Represents sample x i The value of the j-th dimension feature, Representation sample The value of the j-th dimension feature of , β is a random number between 0 and 1;

[0070] The above sampling process is repeated until the number of majority class samples and minority class samples in the two-class data set is the same, and a boundary-enhanced balanced two-class data set X′ is obtained.

[0071] Step 104, fine-tuning the meta-classifier based on the boundary-enhanced two-class dataset to obtain the final two-class classifier, specifically:

[0072] Based on the boundary-enhanced two-class dataset, the meta-classifier is fine-tuned for N2 rounds to obtain the final meta-classifier θ * ,θ * is the final classifier for the two-category dataset.

[0073] Step 105, according to the classification results of each second-class data set, the fault category is obtained by hard voting, specifically:

[0074] Based on the cyclic training model from step 101 to step 104, the classifier corresponding to each two-class data set can be obtained. Where j∈[1,7], for the sample x to be tested test , its predicted label The calculation of is as follows:

[0075]

[0076] When the value is j, it means x test The predicted fault category is the jth fault.

[0077] Figure 2 It is a flow chart of the present invention for solving the classification of smart meter faults. First, the fault history data sets under different categories of smart meters are divided into multiple second-category data sets, and the proposed invention is applied to construct unbalanced classification models based on meta-learning strategy. For the samples to be tested, the classifiers obtained under each second-category data set can be used for discrimination, and the final prediction category can be obtained through the hard voting method.

[0078] Figure 3It is a detailed schematic diagram of the algorithm of the present invention. The algorithm of the present invention is based on the idea of ​​meta-learning. It enhances the classification performance of the meta-classifier when facing new tasks by creating classification losses of multiple query sets in the meta-task, thereby improving the universality of the algorithm. The boundary enhancement strategy based on the Bayesian imbalance influence index alleviates the decision deviation caused by imbalanced category data during the fine-tuning process of the meta-learner.

[0079] Algorithm 1 is the pseudo code for training the meta-classifier:

[0080]

[0081]

[0082] In the specific embodiment, the fault history data set of different categories of smart meters was used for testing. The data set collected smart meter data of 7 fault types in 25 provinces. Due to the influence of various factors such as manual statistics and external conditions, the data labels were mislabeled or missed. After data cleaning using various technologies such as cluster analysis, missing value completion, and outlier processing, a total of 1,500 fault sample data were obtained. In order to reduce the randomness of the results, the data set was randomly divided into a training set and a test set with a fixed random number seed in a ratio of 8:2.

[0083] To verify the effectiveness of the proposed algorithm, four mainstream algorithm-level methods and four mainstream data-level methods are used for comparison in the embodiment of the present invention, as shown in Table 1. The embodiment of the present invention is represented by SMFC-ML in the table.

[0084] Table 1 Algorithms compared in specific embodiments

[0085]

[0086]

[0087] The embodiment of the present invention uses macro-F1 and G-mean indicators to evaluate the classification effect of the algorithm. The macro-F1 indicator is the arithmetic mean of the F1-measure of each category, which is used to comprehensively evaluate the accuracy and recall of the model in each category. The G-mean indicator is the geometric mean of the recall rate in each category, which is used to evaluate the recall of the model in each category. Both Macro-F1 and G-mean have values ​​from 0 to 1. The larger the value, the better the classification performance of the model.

[0088] The experimental results of the embodiment of the present invention and the mainstream algorithm-level method and data-level method on the F1-measure and macro-F1 indicators under each fault category of the smart meter are shown in Table 2, and the experimental results on the recall rate and G-mean indicators of each category are shown in Table 3. It can be seen that the meta-learning-based smart meter fault classification method of the present invention has obtained F1-measure and recall rates that exceed those of other methods in most categories, and has obtained the highest macro-F1 and G-mean. Combining the results of Tables 2 and 3, it is shown that the classification effect of the smart meter fault samples of the method of the embodiment of the present invention is better than that of the existing unbalanced classification method, and can obtain higher classification accuracy and recall rate.

[0089] Table 2 Experimental results of SMFC-ML and comparison methods on F1-measure and macro-F1 indicators under various fault categories of smart meters

[0090]

[0091] Table 3 Experimental results of SMFC-ML and comparison methods on recall rate and G-mean index for each fault category of smart meters

[0092]

[0093] A large number of comparison results with mainstream algorithm-level methods and data-level methods show that the present invention takes the classification performance of new tasks as the optimization goal, and only uses the support set samples to train the meta-classifier, so that the meta-classifier can fully explore the classification information of limited samples in the support set. The meta-classifier is further optimized based on the comprehensive performance in multiple query sets, so that the meta-classifier only needs a small number of training rounds to achieve good classification performance when facing new classification tasks, effectively improving the universality of facing different data sets, and having better performance improvement on data sets with poor classification effects of existing methods, which can effectively improve the accuracy and recall rate of faulty meter classification.

[0094] In summary, the embodiments of the present invention have the following beneficial effects:

[0095] In the technical solution implemented in the present invention, fault history data of smart meters under different categories are used as input data sets, and multiple two-category data sets are obtained by division; for each two-category data set, an unbalanced classification model based on a meta-learning strategy is constructed, subtasks are sampled and constructed from the two-category data set, and a meta-classifier is trained using a meta-learning strategy; the boundary of each two-category data set is enhanced based on the Bayesian imbalance influence index; the meta-classifier is fine-tuned based on the boundary-enhanced two-category data set to obtain the final two-category classifier; and according to the classification results of each two-category data set, its fault category is obtained by hard voting.

[0096] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A smart meter fault classification method based on meta-learning, characterized in that: The steps include: (1) Taking the fault history data of different categories of smart meters as the input data set, multiple second-category data sets are obtained, specifically: The actual fault data set of smart meters is input. The samples in this data set contain 19 characteristic variables, including usage days, manufacturing unit, calibrated current, active power accuracy level, specification, overload multiple, voltage, wiring method, payment terminal production batch, new and old, rotation cycle, hardware version, card meter tripping mode, load curve, power outage meter reading, reverse phase sequence judgment, current loss judgment, rate, meter mode; Its fault category labels include 7 categories: appearance fault, power fault, software fault, clock fault, communication unit fault, metering unit fault and storage unit fault. Traverse each category of samples in the fault data set, and take all samples under this category as the minority class sample set, and all samples of other categories as the majority class sample set, and transform the original data set into 7 two-category data sets. For each of the two-category data sets, the data set can be described as: X=[X min ,X maj ] Among them, X is a two-category dataset, and x is defined as any sample in the dataset, that is, x∈X; X min For the minority class sample set, define x min is any sample in the data set, that is, x min ∈X min ;X maj is the majority class sample set, define x maj is any sample in the data set, that is, x maj ∈X maj ; (2) For each two-category dataset, an unbalanced classification model based on a meta-learning strategy is constructed, and a meta-classifier θ based on a deep neural network is constructed. N independent and identically distributed subtasks are randomly sampled from the two-category dataset, where each subtask contains a support set and a query set, and the number of majority class samples and minority class samples in the subtask is the same, and the number of samples in the query set is greater than the number of samples in the support set. The subtask division process is as follows: sub-task={spt,qry} Among them, sub-task is the subtask, spt is the support set in the subtask, and qry is the query set in the subtask; For each subtask, the meta-classifier is trained using only samples in the support set. The training process is as follows: in, is the classification loss of the meta-classifier on the i-th subtask support set, is the gradient descent algorithm used to optimize the meta-classifier, α is the learning rate of the meta-classifier on the subtask support set, is the meta-classifier optimized on the i-th subtask support set; Since the samples in the subtask only account for a small part of the samples in the original data set, the classifier does not need to pay attention to the overall classification boundary, but only focuses on distinguishing several samples in the subtask, so that the classifier can fully explore the classification features of the limited samples in the support set, laying the foundation for the performance optimization of the classifier on the new classification task. At the same time, the number of samples of the two categories in the support set is the same, which reduces the possibility of decision deviation in the classifier. The number of subtasks required to optimize the classifier each time is n, that is, the number of proxy subtasks is n. After training on n subtasks, the average loss of the classifier on the corresponding subtask query set samples is obtained, and the classifier is optimized using the average loss. The training process is as follows: Among them, β is the learning rate of the meta-classifier on the subtask support set; Repeat the above process until all subtasks are trained and obtain a meta-classifier trained based on the meta-learning strategy. (3) After the meta-classifier is trained based on the meta-learning strategy, the boundary of each two-class dataset is enhanced based on the Bayesian imbalance impact index (IBI3), so as to alleviate the decision deviation phenomenon of the meta-classifier during the fine-tuning process. Specifically: For each minority class sample x, the number of minority class samples m among its K nearest neighbor samples is obtained according to the Euclidean distance, and the probability that the sample is classified as a minority class sample is f p (x) = m / k, where k is the number of K nearest neighbor samples, and the probability of being classified as a majority class sample is f n (x) = (km) / k, assuming that the imbalance ratio of the data is r = N n / N p , where N n is the number of majority class samples in the two-class dataset, N p is the number of minority class samples in the two-class data set, then the probability that the sample is predicted as a minority class sample by the classification model that is not affected by data imbalance can be expressed as f p ′(x)=r*m / k, the calculation process of the Bayesian imbalance impact index (IBI3) of the minority class sample x is as follows: Use the IBI3 value as the sampling probability of the minority class sample and sample the minority class sample x i Then, use K nearest neighbors to get its neighbor samples in the training set, and randomly select a sample from the neighbor samples After performing random feature interpolation, we get a new sample x new , the calculation process is as follows: Among them, x i,j Represents sample x i The value of the j-th dimension feature, Representation sample The value of the j-th dimension feature of , β is a random number between 0 and 1; Repeat the above sampling process until the number of majority class samples and minority class samples in the two-class data set is the same, and obtain a boundary-enhanced balanced two-class data set X′; (4) Based on the boundary-enhanced two-category dataset, the meta-classifier is fine-tuned for N2 rounds to obtain the final meta-classifier θ * ,θ * is the final classifier for the two-category dataset.

2. The meta-learning-based smart meter fault classification method according to claim 1 is characterized in that: In step (2), the structure of the meta-classifier is as follows: Among them, Dense Layer() is the fully connected layer construction function, ReLU represents the corresponding activation function, and x_dim is the number of feature dimensions of samples in the two-category dataset.

3. The meta-learning-based smart meter fault classification method according to claim 1, characterized in that: In the step (2), the values ​​of α and β are 1e-2 and 1e-3 respectively; the value of N is 5000, and the value of n is 4.

4. The method for classifying smart meter faults based on meta-learning according to claim 1, characterized in that: In step (3), the value of k is 5.

5. The method for classifying smart meter faults based on meta-learning according to claim 1, characterized in that: In step (4), the value of N2 is the round when the loss during fine-tuning drops below 0.005 and the loss increases for five consecutive rounds of fine-tuning.

Images