CN117076871A - Battery fault classification method based on unbalanced semi-supervised countermeasure training framework - Google Patents
Battery fault classification method based on unbalanced semi-supervised countermeasure training framework Download PDFInfo
- Publication number
- CN117076871A CN117076871A CN202311333654.1A CN202311333654A CN117076871A CN 117076871 A CN117076871 A CN 117076871A CN 202311333654 A CN202311333654 A CN 202311333654A CN 117076871 A CN117076871 A CN 117076871A
- Authority
- CN
- China
- Prior art keywords
- sample
- semi
- supervised
- training
- class
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000012549 training Methods 0.000 title claims abstract description 90
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 238000009826 distribution Methods 0.000 claims description 31
- 230000004913 activation Effects 0.000 claims description 12
- 238000012795 verification Methods 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 7
- 238000002372 labelling Methods 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 102100029469 WD repeat and HMG-box DNA-binding protein 1 Human genes 0.000 claims description 2
- 101710097421 WD repeat and HMG-box DNA-binding protein 1 Proteins 0.000 claims description 2
- 238000002790 cross-validation Methods 0.000 claims description 2
- 238000010200 validation analysis Methods 0.000 claims description 2
- 238000011156 evaluation Methods 0.000 abstract description 2
- 230000005484 gravity Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 25
- 238000004422 calculation algorithm Methods 0.000 description 10
- 238000012952 Resampling Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000002411 adverse Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000007635 classification algorithm Methods 0.000 description 2
- 238000004140 cleaning Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000036541 health Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000003042 antagnostic effect Effects 0.000 description 1
- 230000008485 antagonism Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011158 quantitative evaluation Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0475—Generative networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/0895—Weakly supervised learning, e.g. semi-supervised or self-supervised learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/094—Adversarial learning
Abstract
The invention belongs to the technical field of power battery fault identification, and discloses a battery fault classification method based on an unbalanced semi-supervised countermeasure training framework, which comprises the steps of firstly constructing and preprocessing a battery fault data set, and constructing a countermeasure training model comprising a generator and a plurality of classes of semi-supervised classifiers; and carrying out data classification evaluation by using the trained countermeasure training model, and judging whether the classification performance meets the requirement. The invention converts the training gravity center of the battery fault data into the discriminant, and the trained discriminant is directly used for classifying the battery fault data; the original data set does not need to be expanded, noise is avoided being introduced, an extra training classifier is not needed, and the training cost and time are reduced.
Description
Technical Field
The invention belongs to the technical field of power battery fault identification, and particularly relates to a battery fault classification method based on an unbalanced semi-supervised countermeasure training framework.
Background
The deep learning method is developed in the field of power battery fault identification, but most applications do not consider the unbalanced distribution condition of samples of different battery health states, and the unbalanced distribution phenomenon among classes is widely existed in power battery fault data. In general, the battery is almost in a normal working state, so that data collected by the sensor is subjected to long tail distribution, namely, samples of normal operation are abundant, and various fault samples are rare. The unbalanced classification refers to the majority class, i.e., the normally operating data class, as the negative class, and the minority class, i.e., the fault class, as the positive class. The small positive class sample size results in that the classifier is easy to deviate to the negative class sample in the process of training the classifier; however, the value of the fault sample is far greater than that of the normal sample, the overall accuracy of the basic classifier trained directly by using the original unbalanced data set is high, but the recall rate is low, namely the fault sample is mostly misclassified. Therefore, how to furthest discover the hidden information of the unbalanced data set of the power battery, and alleviate the adverse effect of the inter-class distribution imbalance problem on the classifier, so that the problem of battery fault identification under data driving is an important problem.
Currently, the main stream of unbalanced data classification solving methods are generally divided into two categories: data resampling and adjustment of the classification algorithm. The data resampling method utilizes an oversampling technology to generate a positive class sample or undersampling to reduce a negative class sample so as to balance an original training set between classes; the adjustment of the classification algorithm is to directly change the training process or the loss function of the classifier by considering the difference of the cost of different misclassification conditions, so that the classifier is more focused on the characteristics of positive classes. The SMOTE is a classical data oversampling method, the principle is simple and easy to realize, representative sample points are selected through a K nearest neighbor algorithm, new sample data are generated through random linear interpolation, the situation of overfitting of the positive sample caused by random copying of the positive sample can be effectively avoided, and the generalization capability of the classifier is improved. When high-dimensional data is processed, potential overall distribution characteristics of positive samples cannot be effectively captured through random linear interpolation, so that an improved oversampling method for generating a countermeasure network (GAN) is introduced, the overall distribution characteristics of the original data are learned through countermeasure training, and meanwhile, a generator and a discriminator network can be replaced by a proper network structure according to actual application scenes. There are many SMOTE-based improved algorithms and algorithms that introduce depth generation models into oversampling, but both types of approaches to deal with unbalanced distributions have some problems: noise is introduced into the data resampling method, the distribution of original data is changed, training time is easy to increase, important sample information is easy to lose, the existing resampling method based on GAN is mainly focused on improving the generation precision and stability of GAN, the generation capacity of a generator is improved, and the noise problem is not fundamentally solved; the algorithm-level method needs to define cost weights or cost matrixes in advance, and is difficult to accurately set. Furthermore, these classical algorithms require a lot of manual effort in data preprocessing or feature selection, which is not suitable for large capacity data sets.
As patent application CN113128589a discloses a novel oversampling method and system for solving the problem of unbalanced classification, which can effectively identify new noise introduced by SMOTE algorithm, and most sample noise and few sample noise existing in original training set, but the SMOTE algorithm cannot learn potential overall distribution characteristics of data, and the generated data mostly presents the problem of gathering around a few sample points, and cannot accurately represent real distribution of original data; noise samples are filtered through the combination of a plurality of experimental data, but training complexity of the whole process is improved, calculation resources are consumed, and the training time is too long, so that efficiency is low.
Patent application CN114943300a discloses an unbalanced data classification method based on cyclic consistency generation antagonism network, which can reduce the sensitivity of classification model to noise data to a certain extent, but the method needs to additionally construct twin data pairs, the required data amount is large, and the model is difficult to train; meanwhile, taking GAN as an oversampling means, balancing an original data set by generating a few types of samples, possibly introducing noise samples, and changing the distribution characteristics of the original data; and requires additional training of the classifier.
Disclosure of Invention
In order to solve the technical problems, the invention provides a battery fault classification method based on an unbalanced semi-supervised countermeasure training framework, which is characterized in that the training center of gravity of battery fault data is converted from a generator of GAN in the traditional method to a discriminator, the generator and the discriminator are trained simultaneously by using an original data set through countermeasure training by letting the discriminator output the class of samples, but the sample generation function of the generator is not used after training, and the trained discriminator is directly used for battery fault data classification; the original data set does not need to be expanded, noise is avoided being introduced, an extra training classifier is not needed, and the training cost and time are reduced.
The invention discloses a battery fault classification method based on an unbalanced semi-supervised countermeasure training framework, which comprises the following steps:
s1, constructing and preprocessing a power battery fault data set: collecting normal operation and negative fault samples of a vehicle-end power battery, randomly selecting a small part of data of different categories, classifying and labeling, sorting the collected data into a marked positive sample set, a marked negative sample set and an unmarked sample set, and carrying out normalization treatment;
s2, dividing a training set and a verification set from the marked power battery fault data set, and adding all unmarked data into the training set;
s3, constructing an countermeasure training model comprising a generator and a plurality of classes of semi-supervised classifiers;
s4, inputting the training set obtained in the S2 and the samples generated by the generator of the countermeasure training model into a multi-class semi-supervised classifier of the countermeasure training model to obtain loss values, and alternately training the generator and the multi-class semi-supervised classifier to obtain a trained countermeasure training model;
s5, using the verification set obtained in the S2 for evaluating a trained multi-class semi-supervised classifier model, and judging whether classification performance meets requirements; and if the parameters do not meet the requirements, adjusting the super parameters of the countermeasure training model, and repeating the step S4 until the classification is completed.
Further, in S1, cleaning the sample data of the running process of the power battery to remove missing values, abnormal values and noise samples; and normalizing the processed data to normalize the sample characteristics to be between-1 and-1.
Further, in S2, the marked dataset is divided into five parts by using 5-fold cross-validation, one part at a time is taken as a validation set, the remaining four parts are taken as training sets, and the unmarked sample is added into the training sets.
Further, in S3, the countermeasure training model includes a generator and a plurality of classes of semi-supervised classifiers;
the generator comprises a random variable input layer, a hidden layer and a sample output layer, wherein the number of network nodes of the random variable input layer is the dimension number of a random variable Z, the number of hidden layers and the number of nodes are set according to the situation, a ReLU function is selected by an activation function, and the number of nodes and the sample of the sample output layer are generatedThe feature dimensions are consistent, the output is limited between-1 and 1 by using a Tanh activation function, and the range of feature normalization of the data set is matched;
the multi-class semi-supervised classifier comprises a sample input layer, a feature extraction hidden layer and a discrimination output full-connection layer; the number of network nodes of the sample input layer is the sampleThe number of hidden layers and the number of nodes are set according to the situation, the activation function is a LeakyReLU function, and the number of nodes outputting the full connection layer is judged to be the class number of the data set sample +.>Activation of functions using SoftmaxAnd (3) outputting the prediction probability of the multi-class semi-supervised classifier on the input sample class.
Furthermore, the judging and outputting full-connection layer of the multi-class semi-supervised classifier not only needs to output the prediction probability of the class corresponding to the sample, but also outputs the true or false judgment of the sample for the countermeasure training of the generator; the normal output node should be set toOutput vector +.>Sample->Belonging to->Prediction probability of class->The following are provided:
，
wherein,representing the category to which the sample belongs,/->The +.f. representing the output of a multi-class semi-supervised classifier>A class-corresponding value; the output of the multi-class semi-supervised classifier is +.>I.e.)>The weight of the node is set to zero, sample +.>Belongs to the category->The prediction probabilities of (2) are as follows:
，
after adjustment，/>So that the output nodes of the multi-class semi-supervised classifier are controlled at +.>Output redundancy from the countermeasure training framework is avoided.
Further, the training process of the unbalanced semi-supervised countermeasure training framework in S4 specifically includes the steps of:
s4.1, marking dataInputting into multi-class semi-supervised classifier, outputting predictive result +.>，Is->Is a true category of (2); calculating a loss function->：
，
Wherein,representation model predictive markers sample->From category->Is a function of the probability of (1),representation->、/>Data distribution from training set, +.>Representing the distribution of +.>Category->Sample->Calculating the function expectations +.>Prediction sample representing corresponding multi-class semi-supervised classifier +.>Belongs to category->Probability of->The closer to the true probability +.>The closer to 0, the lower the impact of the loss;
s4.2, sampling random vector Z from normal distribution, inputting into generator to generate generated sampleThe method comprises the steps of carrying out a first treatment on the surface of the Multi-class semi-supervised classifier judgment +.>Belongs to the category of calculating the loss function>：
，
Wherein,representation model predictive generation sample->From->Category->Is a function of the probability of (1),representation->Data distribution from the generated sample set +.>，/>Representing the distribution of +.>Sample->Calculating the function expectations +.>Prediction sample representing multi-class semi-supervised classifier>Belongs to the category->Probability of (2);
s4.3, no tag data will beInput to a multiclass semi-supervised classifier, output the prediction +.>Calculating a loss function->：
，
Wherein,representation of model predictive unlabeled samples->From->Category(s)Probability of->Data distribution representing unlabeled exemplars, +.>Prediction samples representing multi-class semi-supervised classifierBelongs to category->Is a probability of (2).
Further, in step S5, the evaluation criteria are selected from the balance accuracy (Balanced Accuracy Score), the G-mean index (Geometric Mean Score) and the Macro average absolute error (Macro-Averaged Mean Absolute Error), and the higher the balance accuracy is, the higher the G-mean index is, the smaller the Macro average error is, which indicates that the better the method is for classifying the unbalanced semi-supervised power battery fault.
The beneficial effects of the invention are as follows: according to the method, the generator is not used as an oversampling means to generate the battery fault samples at the end of training, and the generator is only used for antagonizing with a plurality of types of semi-supervised classifiers in the training process, so that the generated fault samples do not need to be added into the original battery data set, and the noise samples are prevented from being introduced to change the distribution of the original battery operation data; compared with the existing oversampling technology based on the generation countermeasure network, the method does not need to expand the original data set after the generator is trained, additionally trains a new classifier, directly trains a plurality of classes of semi-supervised classifiers on the basis of the original data set through countermeasure training, and saves training time and cost; the invention designs the class cost sensitive loss function and the semi-supervised classification training algorithm, and the introduction of the sensitive loss function relieves the adverse effect of unbalanced data distribution of different battery health states on the classifier, so that the performance of the multi-class semi-supervised classifier can be effectively improved; the semi-supervised training algorithm can effectively utilize unlabeled battery samples, save the labeling cost and improve the utilization rate of the running data of the power battery.
Drawings
FIG. 1 is a flow chart of a battery fault classification method based on an unbalanced semi-supervised countermeasure training framework of the present invention;
fig. 2 is a schematic diagram of a generator and a multi-class semi-supervised classifier according to an embodiment of the present invention.
Detailed Description
In order that the invention may be more readily understood, a more particular description of the invention will be rendered by reference to specific embodiments that are illustrated in the appended drawings.
As shown in fig. 1, a battery fault classification method based on an unbalanced semi-supervised countermeasure training framework includes the following steps:
s1, constructing and preprocessing a power battery fault data set: collecting normal operation and negative fault samples of a vehicle-end power battery, randomly selecting a small part of data of different categories, classifying and labeling, sorting the collected data into a marked positive sample set, a marked negative sample set and an unmarked sample set, and carrying out normalization treatment;
s2, dividing a training set and a verification set from the marked power battery fault data set, and adding all unmarked data into the training set;
s3, constructing an countermeasure training model comprising a generator and a plurality of classes of semi-supervised classifiers;
s4, inputting the training set obtained in the S2 and the samples generated by the generator of the countermeasure training model into a multi-class semi-supervised classifier of the countermeasure training model to obtain loss values, and alternately training the generator and the multi-class semi-supervised classifier to obtain a trained countermeasure training model;
s5, using the verification set obtained in the S2 for evaluating a trained multi-class semi-supervised classifier model, and judging whether classification performance meets requirements; and if the parameters do not meet the requirements, adjusting the super parameters of the countermeasure training model, and repeating the step S4 until the classification is completed.
In the following examples, semi-supervised imbalance data sets employ processed MNIST and KMNIST. The original dataset has 10 types of samples, 6000 instances each, with a data dimension of 784. In the experiment, 50 label examples are respectively arranged in most classes through random sampling, the rest 5950 is label-free data, 5 label examples are selected in few classes, and 110 label-free examples are selected in the few classes. MNIST selects number 4 as the minority class and KMNIST selects the "ma" character as the minority class.
The present embodiment uses the PyTorch framework, based on the python language, with the following detailed procedures:
s1, constructing and preprocessing a data set: carrying out data cleaning on the MNIST and KMNIST data set for experiments to remove missing values, abnormal values and noise samples; screening labeled positive class samples, labeled negative class samples and unlabeled sample data, and carrying out normalization processing on the data set to ensure that sample characteristics are distributed between-1 and 1;
s2, dividing the marked data set into five parts by using 5-fold cross verification, taking one part at each time without repetition as a verification set, taking the other four parts as training sets, and dividing the training sets and the verification sets from the marked data; if unmarked data exists, the number of the label-free samples is matched with that of the label-free samples, and the label-free samples and the label samples of the training set after the copying and expansion form a sample pair to construct a semi-supervision training set;
s3, constructing an countermeasure training model comprising a generator and a plurality of classes of semi-supervised classifiers; the method comprises the following steps:
the generator includes a random variable input layer, a hidden layer, and a generated sample output layer. The number of network nodes of the input layer is the Z dimension of a random variable, the Z dimension is set to be 200, the hidden layer is set to be two layers, 512 and 1024 nodes are respectively arranged, the ReLU function is selected by the activation function, and the number of nodes of the output layer is a sampleFeature dimension 784, using the Tanh activation function, limiting the output to between-1 and 1, matching the range of feature normalization of the dataset;
the multi-class semi-supervised classifier comprises a sample input layer, a feature extraction hidden layer and a discrimination output full-connection layer. The number of network nodes of the input layer is a sampleThree layers are arranged on the hidden layer, namely 1024, 512 and 256 nodes respectively, the activation function is a LeakyReLU function, and the number of nodes of the full connection layer is the category number of the data set sample=10, outputting the prediction probability of the multi-class semi-supervised classifier for the input sample class using Softmax activation function;
the optimizer selects ADAM optimization algorithm, and the learning rate is set to be 1e-4;
s4, inputting the training set obtained in the step S2 and the generated sample of the generator into a multi-class semi-supervised classifier to obtain a loss value, and alternately training the generator and the multi-class semi-supervised classifier; the method comprises the following steps:
s4.1, marking dataInputting into multi-class semi-supervised classifier, outputting prediction result，Is thatIs a true category of (2); calculating a loss function：
，
Wherein,representation->、/>Data distribution from training set, +.>Representing the distribution of +.>Category->Sample->Calculating the function expectations +.>Prediction sample representing corresponding multi-class semi-supervised classifier +.>Belongs to category->Probability of (2);
s4.2, sampling random vector Z from normal distribution, inputting into generator to generate generated sampleThe method comprises the steps of carrying out a first treatment on the surface of the Multi-class semi-supervised classifier judgment +.>Belongs to the category of calculating the loss function>；
，
Wherein,representation->Data distribution from the generated sample set +.>，/>Representing the distribution of +.>Sample->Calculating the function expectations +.>Prediction sample representing multi-class semi-supervised classifier>Belongs to category->Probability of (2);
s4.3, no tag data will beInput to a multiclass semi-supervised classifier, output the prediction +.>Calculating a loss function->；
，
Wherein,the distribution of data representing the unlabeled exemplars,prediction samples representing multi-class semi-supervised classifierBelongs to the category ofProbability of (2);
s5, using the verification set obtained in the step S2 for evaluating a trained multi-class semi-supervised classifier model, and judging whether classification performance meets requirements or not; and if not, adjusting the super parameters, and repeating the step S4.
The present embodiment adopts the balance accuracy, the G-mean index, and the macro average absolute error as quantitative evaluation criteria. The effect comparison was performed with the basic MLP classifier, balanced integrated classifier (balanced ensemble method) and self-training (self-training) with classifier binding methods. The experimental results are shown in table 1, and the methods are all the methods of the invention, so that the balance accuracy and the G-mean index of the method are far higher than those of other methods, the macro average absolute error is greatly reduced, the classification accuracy of positive class is greatly improved, and meanwhile, negative class samples are rarely misclassified.
Watch (watch)Classification Properties of different methods in unbalanced semi-supervised MNIST and KMNIST
。
The foregoing is merely a preferred embodiment of the present invention, and is not intended to limit the present invention, and all equivalent variations using the description and drawings of the present invention are within the scope of the present invention.
Claims (6)
1. A battery fault classification method based on an unbalanced semi-supervised countermeasure training framework, comprising the steps of:
s1, constructing and preprocessing a power battery fault data set: collecting normal operation and negative fault samples of a vehicle-end power battery, randomly selecting a small part of data of different categories, classifying and labeling, sorting the collected data into a marked positive sample set, a marked negative sample set and an unmarked sample set, and carrying out normalization treatment;
s2, dividing a training set and a verification set from the marked power battery fault data set, and adding all unmarked data into the training set;
s3, constructing an countermeasure training model comprising a generator and a plurality of classes of semi-supervised classifiers;
s4, inputting the training set obtained in the S2 and the samples generated by the generator of the countermeasure training model into a multi-class semi-supervised classifier of the countermeasure training model to obtain loss values, and alternately training the generator and the multi-class semi-supervised classifier to obtain a trained countermeasure training model;
s5, using the verification set obtained in the S2 for evaluating a trained multi-class semi-supervised classifier model, and judging whether classification performance meets requirements; and if the parameters do not meet the requirements, adjusting the super parameters of the countermeasure training model, and repeating the step S4 until the classification is completed.
2. The battery fault classification method based on the unbalanced semi-supervised countermeasure training framework of claim 1, wherein in S1, the power battery operation process sample data is cleaned to remove missing values, abnormal values and noise samples; and normalizing the processed data to normalize the sample characteristics to be between-1 and-1.
3. The method of claim 1, wherein in S2, the marked dataset is divided into five parts by 5-fold cross-validation, one part at a time is not repeated as a validation set, the remaining four parts are used as training sets, and unlabeled samples are added to the training sets.
4. A method of battery fault classification based on an unbalanced semi-supervised challenge training framework as claimed in claim 1, wherein in S3 the challenge training model comprises a generator and a plurality of classes of semi-supervised classifiers;
the generator comprises a random variable input layer, a hidden layer and a sample output layer, wherein the number of network nodes of the random variable input layer is the dimension number of a random variable Z, the number of hidden layers and the number of nodes are set according to the situation, a ReLU function is selected by an activation function, and the number of nodes and the sample of the sample output layer are generatedThe feature dimensions are consistent, the output is limited between-1 and 1 by using a Tanh activation function, and the range of feature normalization of the data set is matched;
the multi-class semi-supervised classifier comprises a sample input layer, a feature extraction hidden layer and a discrimination output full-connection layer; the number of network nodes of the sample input layer is the sampleThe number of hidden layers and the number of nodes are set according to the situation, the activation function is a LeakyReLU function, and the number of nodes outputting the full connection layer is judged to be the class number of the data set sample +.>The Softmax activation function is used to output the predicted probabilities of the multiple classes of semi-supervised classifiers for the input sample classes.
5. The battery fault classification method based on the unbalanced semi-supervised countermeasure training framework of claim 4, wherein the discrimination output full-connection layer of the multi-class semi-supervised classifier not only needs to output the prediction probability of the class corresponding to the sample, but also outputs the true or false judgment of the sample for the countermeasure training of the generator; the normal output node should be set toOutput vector +.>Sample->Belonging to->Prediction probability of class->The following are provided:
，
wherein,representing the category to which the sample belongs,/->The +.f. representing the output of a multi-class semi-supervised classifier>A class-corresponding value; the output of the multi-class semi-supervised classifier is +.>I.e.)>The weight of the node is set to zero, sample +.>Belongs to the category->The prediction probabilities of (2) are as follows:
，
after adjustment，/>So that the output nodes of the multi-class semi-supervised classifier are controlled at +.>Output redundancy from the countermeasure training framework is avoided.
6. The method for classifying battery faults based on an unbalanced semi-supervised countermeasure training framework as claimed in claim 1, wherein the training process of the unbalanced semi-supervised countermeasure training framework in S4 specifically comprises the steps of:
s4.1, marking dataInputting into multi-class semi-supervised classifier, outputting predictive result +.>，/>Is thatIs a true category of (2); calculating a loss function->：
，
Wherein,representation model predictive markers sample->From category->Is a function of the probability of (1),representation->、/>Data distribution from training set, +.>Representing the distribution of +.>Category->Sample->Calculating the function expectations +.>Prediction sample representing corresponding multi-class semi-supervised classifier +.>Belongs to category->Probability of->The closer to the true probability +.>The closer to 0, the lower the impact of the loss;
s4.2, sampling random vector Z from normal distribution, inputting into generator to generate generated sampleThe method comprises the steps of carrying out a first treatment on the surface of the Multi-class semi-supervised classifier judgment +.>Belongs to the category of calculating the loss function>：
，
Wherein,representation model predictive generation sample->From->Category->Is a function of the probability of (1),representation->Data distribution from the generated sample set +.>，/>Representing the distribution of +.>Sample->Calculating the function expectations +.>Prediction sample representing multi-class semi-supervised classifier>Belongs to the category->Probability of (2);
s4.3, no tag data will beInput to a multiclass semi-supervised classifier, output the prediction +.>Calculating a loss function->：
，
Wherein,representation of model predictive unlabeled samples->From->Category->Probability of->Data distribution representing unlabeled exemplars, +.>Prediction sample representing multi-class semi-supervised classifier>Belongs to category->Is a probability of (2).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311333654.1A CN117076871B (en) | 2023-10-16 | 2023-10-16 | Battery fault classification method based on unbalanced semi-supervised countermeasure training framework |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311333654.1A CN117076871B (en) | 2023-10-16 | 2023-10-16 | Battery fault classification method based on unbalanced semi-supervised countermeasure training framework |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117076871A true CN117076871A (en) | 2023-11-17 |
| CN117076871B CN117076871B (en) | 2023-12-29 |
Family
ID=88717614
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311333654.1A Active CN117076871B (en) | 2023-10-16 | 2023-10-16 | Battery fault classification method based on unbalanced semi-supervised countermeasure training framework |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117076871B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117476125A (en) * | 2023-12-27 | 2024-01-30 | 豆黄金食品有限公司 | Dried beancurd stick raffinate recovery data processing system based on data analysis |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108805188A (en) * | 2018-05-29 | 2018-11-13 | 徐州工程学院 | A kind of feature based recalibration generates the image classification method of confrontation network |
| US20190122120A1 (en) * | 2017-10-20 | 2019-04-25 | Dalei Wu | Self-training method and system for semi-supervised learning with generative adversarial networks |
| CN110110745A (en) * | 2019-03-29 | 2019-08-09 | 上海海事大学 | Based on the semi-supervised x-ray image automatic marking for generating confrontation network |
| CN111898579A (en) * | 2020-08-12 | 2020-11-06 | 中国地质大学（武汉） | Extreme gradient lifting-based unbiased semi-supervised classification model for high-resolution remote sensing images |
| CN112990342A (en) * | 2021-04-08 | 2021-06-18 | 重庆大学 | Semi-supervised SAR target recognition method |
| US20210287071A1 (en) * | 2020-03-12 | 2021-09-16 | Morgan State University | Method and Apparatus for Augmented Data Anomaly Detection |
| CN113705709A (en) * | 2021-09-02 | 2021-11-26 | 新疆信息产业有限责任公司 | Improved semi-supervised image classification method, equipment and storage medium |
| CN113884290A (en) * | 2021-09-28 | 2022-01-04 | 江南大学 | Voltage regulator fault diagnosis method based on self-training semi-supervised generation countermeasure network |
| CN115541228A (en) * | 2022-10-10 | 2022-12-30 | 重庆邮电大学 | Fault diagnosis method for wind turbine generator gearbox |
| CN116467634A (en) * | 2023-02-07 | 2023-07-21 | 大连理工大学 | Refrigerating unit fault diagnosis method based on semi-supervised contrast learning |
-
2023
- 2023-10-16 CN CN202311333654.1A patent/CN117076871B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190122120A1 (en) * | 2017-10-20 | 2019-04-25 | Dalei Wu | Self-training method and system for semi-supervised learning with generative adversarial networks |
| CN108805188A (en) * | 2018-05-29 | 2018-11-13 | 徐州工程学院 | A kind of feature based recalibration generates the image classification method of confrontation network |
| CN110110745A (en) * | 2019-03-29 | 2019-08-09 | 上海海事大学 | Based on the semi-supervised x-ray image automatic marking for generating confrontation network |
| US20210287071A1 (en) * | 2020-03-12 | 2021-09-16 | Morgan State University | Method and Apparatus for Augmented Data Anomaly Detection |
| CN111898579A (en) * | 2020-08-12 | 2020-11-06 | 中国地质大学（武汉） | Extreme gradient lifting-based unbiased semi-supervised classification model for high-resolution remote sensing images |
| CN112990342A (en) * | 2021-04-08 | 2021-06-18 | 重庆大学 | Semi-supervised SAR target recognition method |
| CN113705709A (en) * | 2021-09-02 | 2021-11-26 | 新疆信息产业有限责任公司 | Improved semi-supervised image classification method, equipment and storage medium |
| CN113884290A (en) * | 2021-09-28 | 2022-01-04 | 江南大学 | Voltage regulator fault diagnosis method based on self-training semi-supervised generation countermeasure network |
| CN115541228A (en) * | 2022-10-10 | 2022-12-30 | 重庆邮电大学 | Fault diagnosis method for wind turbine generator gearbox |
| CN116467634A (en) * | 2023-02-07 | 2023-07-21 | 大连理工大学 | Refrigerating unit fault diagnosis method based on semi-supervised contrast learning |
Non-Patent Citations (4)
| Title |
|---|
| HE R等: "Generative adversarial network-based semi-supervised learning for real-time risk warning of process industries", 《EXPERT SYSTEMS WITH APPLICATIONS》, vol. 150, pages 1 - 12 * |
| HU H等: "Semi-supervised semantic segmentation via adaptive equalization learning", 《ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS》, vol. 34, pages 22106 - 22118 * |
| 刘半藤等: "基于距离度量损失框架的半监督学习方法", 《浙江大学学报(工学版)》, vol. 57, no. 4, pages 744 - 752 * |
| 刘坤等: "基于半监督生成对抗网络X光图像分类算法", 《光学学报》, vol. 39, no. 8, pages 1 - 9 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117476125A (en) * | 2023-12-27 | 2024-01-30 | 豆黄金食品有限公司 | Dried beancurd stick raffinate recovery data processing system based on data analysis |
| CN117476125B (en) * | 2023-12-27 | 2024-04-05 | 豆黄金食品有限公司 | Dried beancurd stick raffinate recovery data processing system based on data analysis |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117076871B (en) | 2023-12-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110443281B (en) | Text classification self-adaptive oversampling method based on HDBSCAN (high-density binary-coded decimal) clustering | |
| CN110213222B (en) | Network intrusion detection method based on machine learning | |
| CN101464964B (en) | Pattern recognition method capable of holding vectorial machine for equipment fault diagnosis | |
| Yin et al. | Wasserstein generative adversarial network and convolutional neural network (WG-CNN) for bearing fault diagnosis | |
| CN107103332A (en) | A kind of Method Using Relevance Vector Machine sorting technique towards large-scale dataset | |
| CN111314353B (en) | Network intrusion detection method and system based on hybrid sampling | |
| CN117076871B (en) | Battery fault classification method based on unbalanced semi-supervised countermeasure training framework | |
| Wu et al. | Graphmixup: Improving class-imbalanced node classification on graphs by self-supervised context prediction | |
| CN105320961A (en) | Handwriting numeral recognition method based on convolutional neural network and support vector machine | |
| CN112069310A (en) | Text classification method and system based on active learning strategy | |
| CN111681132B (en) | Typical power consumption mode extraction method suitable for massive class unbalanced load data | |
| CN107644057A (en) | A kind of absolute uneven file classification method based on transfer learning | |
| CN108717149A (en) | Diagnosis Method of Transformer Faults based on M-RVM fusion dynamic weightings AdaBoost | |
| CN105930792A (en) | Human action classification method based on video local feature dictionary | |
| CN111832664A (en) | Borderline SMOTE-based power transformer fault sample equalization and fault diagnosis method | |
| CN111177010B (en) | Software defect severity identification method | |
| CN115545070A (en) | Intelligent diagnosis method for unbalance-like bearing based on comprehensive balance network | |
| Wang et al. | Application of an oversampling method based on GMM and boundary optimization in imbalance-bearing fault diagnosis | |
| Xueli et al. | An improved KNN algorithm based on kernel methods and attribute reduction | |
| CN111178897B (en) | Cost-sensitive dynamic clustering method for fast feature learning on unbalanced data | |
| CN112990371A (en) | Unsupervised night image classification method based on feature amplification | |
| CN116582300A (en) | Network traffic classification method and device based on machine learning | |
| CN115545111A (en) | Network intrusion detection method and system based on clustering self-adaptive mixed sampling | |
| Zhang et al. | A fault diagnosis model of pumping unit based on bp neural network | |
| Arkok et al. | Classification of Quranic Topics Using SMOTE Technique |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |