Classification method and device for equipment problems and storage medium

Abstract

Embodiments of the invention provide an equipment problem classification method and apparatus, and a storage medium. The method comprises the steps of obtaining data features of target equipment; performing feature selection algorithm processing and feature dimension reduction processing on the data features to obtain key features; constructing a first classifier according to the key features and the classification requirements; and training the first classifier through training set data to obtain a second classifier, and classifying the equipment problems of the target equipment by using the second classifier. According to the method, the equipment supervision efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power supervision, and in particular to a classification method, a device and a storage medium for equipment problems. Background Art

[0002] As power equipment becomes increasingly complex, accurate classification and rapid processing of equipment technical problems have become increasingly important. In recent years, with the rapid development of technologies such as artificial intelligence and machine learning, equipment technical problem classification algorithms have been significantly improved. Existing technologies usually classify equipment technical problems based on manual experience, and the classification accuracy is easily affected by human factors; moreover, manual classification requires a lot of time and effort, resulting in low processing efficiency and difficulty in meeting the real-time requirements of power technical supervision. The above unfavorable factors will affect the classification effect. Summary of the invention

[0003] The purpose of the embodiments of the present invention is to provide a method, device and storage medium for classifying equipment problems, which improves the efficiency of equipment supervision.

[0004] In order to achieve the above object, an embodiment of the present invention provides a method for classifying device problems, the method comprising:

[0005] Obtain data characteristics of the target device;

[0006] Performing feature selection algorithm processing and feature dimension reduction processing on the data features to obtain key features;

[0007] Constructing a first classifier according to the key features and classification requirements;

[0008] The first classifier is trained using training set data to obtain a second classifier, and the second classifier is used to classify device problems of the target device.

[0009] Optionally, obtaining data features of the target device includes:

[0010] Acquire a data source of the target device, the data source including operation data, fault records and maintenance logs;

[0011] Extracting data features of the data source by using a convolutional neural network and / or a recurrent neural network;

[0012] The data features include equipment type, fault phenomenon, fault code and operating parameters.

[0013] Optionally, the feature selection algorithm is at least one of a chi-square test, mutual information and recursive feature elimination;

[0014] The feature dimension reduction process includes principal component analysis and linear discriminant analysis;

[0015] The principal component analysis is to project the original high-dimensional data into a low-dimensional space through linear transformation to reduce the dimension of the data features;

[0016] The linear discriminant analysis is to determine the best projection direction according to the inter-class discreteness and intra-class discreteness of the data features.

[0017] Optionally, constructing a first classifier according to the key features and classification requirements includes: determining a classification method according to the classification requirements;

[0018] Constructing a first classifier for the key features according to the classification method;

[0019] The classification method is at least one of a support vector machine, a random forest and a neural network.

[0020] Optionally, the constructing a first classifier according to the key features and classification requirements further includes:

[0021] Determine a machine learning algorithm according to the classification requirements;

[0022] Constructing a first classifier for the key features according to the machine learning algorithm;

[0023] The machine learning algorithm is at least one of a gradient boosting decision tree, extreme gradient boosting, and a lightweight gradient boosting machine.

[0024] Optionally, the first classifier is a hierarchical classifier, including an upper classifier and a lower classifier;

[0025] The upper classifier is used to classify faults into major categories, and the lower classifier is used to classify faults into minor categories.

[0026] Optionally, the method further includes:

[0027] According to the electrical connection and operation coupling relationship of the multiple classifiers, the outputs of the multiple classifiers are weightedly fused to classify the device problems of the target device.

[0028] Optionally, the method further includes:

[0029] Preprocessing the data features;

[0030] The preprocessing includes data cleaning, denoising, normalization, data enhancement and data dimension reduction;

[0031] The data cleaning includes removing duplicate data, processing missing values, and identifying and correcting erroneous data;

[0032] The denoising includes filtering and wavelet denoising;

[0033] The normalization includes maximum and minimum normalization and standardization.

[0034] Optionally, the method further includes:

[0035] The parameters of the second classifier are tuned by using at least one of cross validation, grid search, random search, Bayesian optimization and genetic algorithm.

[0036] On the other hand, the present invention also provides a device for classifying equipment problems, the device comprising:

[0037] An acquisition module, used to acquire data characteristics of a target device;

[0038] A first processing module is used to perform feature selection algorithm processing and feature dimension reduction processing on the data features to obtain key features;

[0039] The second processing module is used to construct a first classifier according to the key features and classification requirements, and train the first classifier through training set data to obtain a second classifier, and use the second classifier to classify the device problems of the target device.

[0040] On the other hand, the present invention further proposes a machine-readable storage medium, on which instructions are stored, and the instructions are used to enable a machine to execute the above-mentioned device problem classification method.

[0041] A classification method for equipment problems of the present invention comprises: obtaining data features of a target device; performing feature selection algorithm processing and feature dimension reduction processing on the data features to obtain key features; constructing a first classifier according to the key features and classification requirements; training the first classifier with training set data to obtain a second classifier, and using the second classifier to classify equipment problems of the target device. The method automatically learns and optimizes classification based on a machine learning classification model, thereby improving the classification accuracy of equipment technical problems.

[0042] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0044] Figure 1 It is a flow chart of a method for classifying equipment problems of the present invention;

[0045] Figure 2 It is a schematic diagram of a classification device for equipment problems of the present invention.

[0046] Description of Reference Numerals

[0047] 200-Classification device for equipment problems;

[0048] 201-Get module;

[0049] 202-first processing module;

[0050] 203-second processing module. DETAILED DESCRIPTION

[0051] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0052] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application are in compliance with the relevant provisions of national laws and regulations. In the embodiments of this application, some existing solutions in the industry such as certain software, components, and models may be mentioned, which should be considered as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of this application, but it does not mean that the applicant has or will necessarily use the solution.

[0053] Figure 1 It is a flow chart of a method for classifying equipment problems of the present invention, such as Figure 1 As shown, a classification method of equipment problems of the present invention comprises the following steps:

[0054] Step S101 is to obtain data characteristics of the target device.

[0055] Specifically, the data features of the target device are obtained, including: obtaining the data source of the target device, the data source including operation data, fault records and maintenance logs; extracting the data features of the data source through a convolutional neural network and / or a recurrent neural network; the data features include device type, fault phenomenon, fault code and operation parameters. The target device may be an electric power technical supervision device.

[0056] According to a specific implementation method, multiple data features are extracted from multiple data sources (such as operating data, fault records, maintenance logs, etc.) of power technical supervision equipment, and these data features cover equipment type, fault phenomenon, fault code, operating parameters, etc. For example, the real-time operating parameters of the equipment such as voltage, current, temperature, etc. are obtained from the operating data; the time, frequency, severity, etc. of the fault occurrence are extracted from the fault records; and the relevant features such as maintenance measures and replacement parts are extracted from the maintenance log.

[0057] The method for obtaining data features of the target device includes statistical feature extraction and frequency domain feature extraction. The statistical feature extraction includes calculating the mean, the standard deviation, the variance, the square root amplitude, the waveform index, the margin index, the skewness and kurtosis, etc. The frequency domain feature extraction includes spectrum analysis, spectrum energy calculation, spectrum mean and spectrum peak.

[0058] Specifically, the calculated mean is the value of the equipment operating parameters (such as voltage, current, temperature, etc.) within a period of time, and the average value is calculated. For example, the average value of the oil temperature of a power transformer in the past hour is calculated. The average value can reflect the average level of equipment operation and help determine whether the equipment is in normal operation.

[0059] The calculated standard deviation is used to measure the degree of dispersion of the equipment operating parameter data. Taking the current in the power line as an example, a large standard deviation may indicate abnormal current fluctuations, indicating potential problems with the equipment. The calculated variance is similar to the standard deviation and is also a statistic that reflects the degree of dispersion of the data. The larger the variance, the more drastic the fluctuation of the data, which is of great significance for judging the stability of equipment operation.

[0060] For periodic signals such as vibration signals, the root mean square amplitude can be used to describe the strength of the signal. By analyzing the changes in the root mean square amplitude, it is possible to detect whether the mechanical parts of the equipment have problems such as wear and looseness. The waveform index is the ratio of the root mean square value of the signal to the average value of the absolute value. It is used to evaluate the waveform characteristics of the signal. Different equipment failures may cause changes in the waveform index, thereby providing a basis for fault diagnosis. The margin index reflects the relationship between the peak value and the root mean square value of the signal, and is more sensitive to abnormal conditions such as shock and overload during the operation of the detection equipment. Skewness is used to measure the asymmetry of the data distribution, and kurtosis describes the sharpness of the data distribution. These statistical features can be used to identify abnormal patterns in equipment operation data, such as signal mutations caused by failures.

[0061] Perform Fourier transform on the equipment operation signal (such as vibration signal, current signal, etc.), convert the time domain signal into frequency domain signal, and obtain the signal spectrum. By analyzing the frequency components, amplitude distribution and other information in the spectrum, the operating status of the equipment can be judged. For example, when a motor fails, changes in specific frequency components may appear in the spectrum of its vibration signal. Calculate the energy of the signal in different frequency bands, and the change in energy can reflect the working status of the equipment. For example, in transformer fault diagnosis, by calculating the energy in different frequency bands, changes in energy distribution caused by faults such as loose core and winding short circuit can be found. The spectrum mean represents the average energy level of the spectrum, and the spectrum peak is the frequency point with the highest energy in the spectrum. Their changes can also be used as one of the characteristics of equipment failure.

[0062] Step S102 is to perform feature selection algorithm processing and feature dimension reduction processing on the data features to obtain key features.

[0063] According to a specific implementation manner, the feature selection algorithm is processed as at least one of a chi-square test, mutual information and recursive feature elimination; the feature dimension reduction processing includes principal component analysis and linear discriminant analysis; the principal component analysis is to project the original high-dimensional data into a low-dimensional space through a linear transformation to reduce the dimension of the data features; the linear discriminant analysis is to determine the optimal projection direction based on the inter-class discreteness and intra-class discreteness of the data features.

[0064] Specifically, feature selection algorithms (such as chi-square test, mutual information, recursive feature elimination, etc.) are used to evaluate the correlation between each data feature and the classification target, and to screen out key features that have a significant impact on the classification results. For example, the chi-square test is used to calculate the chi-square statistic between the feature and the classification variable, and features with larger statistics are selected, redundant or irrelevant features are removed, and the data dimension is reduced while retaining important information to improve classification efficiency.

[0065] Among them, the chi-square test is the degree of deviation between the actual observed value of the statistical sample and the theoretical inference value. The degree of deviation between the actual observed value and the theoretical inference value determines the size of the chi-square value. If the chi-square value is larger, the deviation between the two is greater; conversely, the deviation between the two is smaller; if the two values ​​are completely equal, the chi-square value is 0, indicating that the theoretical value is completely consistent. Mutual information is a concept in probability theory and information theory that is used to measure the degree of mutual dependence between two random variables. It is not limited to real-valued variables, and expresses the similarity between the joint distribution and the marginal distribution. Recursive feature elimination is a feature selection method based on model prediction performance. It evaluates the importance of each feature by recursively removing features and retraining the model. In each iteration, it removes the least important features and retrains the model. This process continues until the specified number of features is reached or no more features can be removed.

[0066] The principal component analysis is to transform a set of variables that may be correlated into a set of linearly uncorrelated variables through orthogonal transformation. The transformed set of variables is called principal component. The linear discriminant analysis is a dimensionality reduction technology for supervised learning. Each sample in its data set has a categorical output. Its main idea is to maximize the mean between classes and minimize the variance within classes.

[0067] Feature dimensionality reduction (such as principal component analysis (PCA) and linear discriminant analysis (LDA)) is used to further reduce the number of features. Principal component analysis (PCA) projects the original high-dimensional data into a low-dimensional space through linear transformation, reducing the dimension while retaining the data variance as much as possible. Linear discriminant analysis (LDA) finds the best projection direction based on the consideration of inter-class and intra-class discreteness, making data of different categories more separable in low-dimensional space, reducing computational complexity and the risk of model overfitting.

[0068] Step S103 is to construct a first classifier according to the key features and classification requirements.

[0069] According to a specific implementation, constructing a first classifier based on the key features and classification requirements includes: determining a classification method based on the classification requirements; constructing a first classifier for the key features based on the classification method; the classification method is at least one of a support vector machine, a random forest, and a neural network.

[0070] Specifically, a suitable classification algorithm (i.e., classification method, such as support vector machine, random forest, neural network, etc.) is selected based on the extracted key features to build a classifier. The classifier is trained with training set data so that it can accurately identify and classify different technical problems.

[0071] Select a suitable classification algorithm according to data characteristics and classification requirements. Common classification algorithms include support vector machine (SVM), random forest (RF), neural network (NN), gradient boosting decision tree (GBDT), extreme gradient boosting (XGBoost), lightweight gradient boosting machine (LightGBM), etc.

[0072] Specifically, the principle of the support vector machine (SVM) is: by finding an optimal hyperplane, the data of different categories are separated as much as possible, so that the sum of the distances of the two categories of data to the hyperplane is maximized. This distance is called the interval. For linearly separable data, SVM can find a linear hyperplane for classification; for nonlinearly separable data, the data is mapped to a high-dimensional space by using a kernel function (such as a linear kernel, a polynomial kernel, a radial basis function kernel, etc.), and the optimal classification hyperplane is found in the high-dimensional space. Its applicable scenarios include: it has advantages when processing small samples, high-dimensional data, and there may be nonlinear relationships between features and classifications. It is suitable for the classification of features with clear classification boundaries such as fault codes, and can also be used to classify other equipment technical problems after appropriate feature conversion.

[0073] The principle of the random forest (RF) is as follows: it is composed of multiple decision trees. When constructing each decision tree, a feature subset is randomly selected from the training data (the random selection of features increases the diversity of the model and reduces the risk of overfitting). The prediction results of each decision tree are voted or averaged to obtain the final classification result. The construction of the decision tree is based on the segmentation of features, and the best features and segmentation points are selected so that the samples in each child node belong to the same category as much as possible. Its applicable scenarios include: being able to process high-dimensional data and having good robustness, being suitable for situations where the amount of data such as equipment operating parameters is large and there is a certain interactive relationship between features, having a certain tolerance for noise and outliers in the data, and can be used for classification prediction of equipment operating status.

[0074] The principle of the neural network (NN) is as follows: it contains multiple neurons and simulates the neuron structure of the human brain by constructing different layers (input layer, hidden layer, output layer). Neurons are connected by weights, data is transmitted between layers, and output is generated through the action of activation functions. During the training process, the connection weights between neurons are adjusted using the back propagation algorithm according to the error between the predicted result and the actual label to minimize the error function, so that the neural network learns the characteristic patterns and classification rules in the data. Its applicable scenarios include: it has strong nonlinear mapping capabilities and is suitable for processing complex nonlinear relationships, such as failures caused by the interaction of multiple factors during the operation of the equipment. It is especially suitable for classifying continuous equipment operating parameters, but it requires more training data and computing resources, and there is a risk of overfitting, which needs to be controlled by regularization and other methods.

[0075] The principle of the gradient boosting decision tree (GBDT) and its variants (XGBoost, LightGBM) is: by combining multiple weak classifiers (decision trees) into a strong classifier. In each round of iteration, the weight of the training sample is adjusted according to the error of the previous round of classifiers, focusing on the samples that were misclassified in the previous round, so that the subsequent decision trees can pay more attention to these difficult-to-classify samples and gradually improve the classification performance. GBDT uses the idea of ​​gradient descent to optimize the loss function during the training process. XGBoost performs a series of optimizations on the basis of GBDT, such as regularization terms to control model complexity, processing of missing values, parallel computing acceleration, etc.; LightGBM uses histogram algorithms to further improve training efficiency and memory utilization. Its applicable scenarios include: suitable for processing mixed data types (including continuous and discrete data), can automatically learn the complex relationship between different features in multi-source data, and in the classification of technical problems of power technical supervision equipment, it performs well in the case of comprehensively considering the fusion features of multi-source data such as equipment operation data, fault records, and maintenance logs, which can effectively improve the accuracy and robustness of classification.

[0076] According to another embodiment, constructing a first classifier according to the key features and classification requirements also includes: determining a machine learning algorithm according to the classification requirements; constructing a first classifier for the key features according to the machine learning algorithm; the machine learning algorithm is at least one of a gradient boosting decision tree, extreme gradient boosting, and a lightweight gradient boosting machine.

[0077] Specifically, the first classifier is a hierarchical classifier, including an upper classifier and a lower classifier; the upper classifier is used to distinguish faults into major categories, and the lower classifier is used to divide faults into minor categories.

[0078] Step S104 is to train the first classifier using training set data to obtain a second classifier, and use the second classifier to classify the device problems of the target device.

[0079] According to a specific implementation method, the selected classification algorithm is trained using labeled training set data (which may be preprocessed data features) so that the classifier learns the mapping relationship between different technical problem features and classification labels. For example, the training data containing the equipment technical problem features and the corresponding classification labels (such as fault type A, fault type B, etc.) are input into the classification algorithm, and the parameters in the algorithm are optimized (such as the kernel function parameters in the SVM, the weights and biases in the neural network, etc.), so that the classifier can achieve a higher classification accuracy on the training set.

[0080] The method further includes: performing weighted fusion on the outputs of the multiple classifiers according to the electrical connection and operation coupling relationship of the multiple classifiers, so as to classify the device problems of the target device. The method further includes: performing parameter tuning on the second classifier by using at least one of cross validation, grid search, random search, Bayesian optimization and genetic algorithm.

[0081] The present invention uses cross-validation, grid search, random search and other methods to tune the parameters of the classifier, thereby improving the classification performance and generalization ability of the model.

[0082] Specifically, the cross-validation is a k-fold cross-validation. For example, the data set is divided into k subsets of similar size, usually k is 5 or 10. Take k-1 subsets as training sets and the remaining 1 subset as test set in turn, and perform k training and testing. A model is obtained after each training, and the performance (such as accuracy, recall, etc.) is evaluated on the corresponding test set. Finally, the average of the k test results is used as the final evaluation indicator of the model. This method makes full use of the data, reduces the random influence brought by the single division of training sets and test sets, and can more accurately evaluate the generalization ability of the model on different data subsets.

[0083] The principle of the grid search is: for multiple parameters in the classification algorithm, determine their value ranges, and then generate all possible parameter combinations within these ranges by exhaustive enumeration. For example, for the SVM algorithm, different kernel function parameters (such as C value, gamma value, etc.) may be searched; for the neural network, different parameters such as the number of hidden layers and learning rate are searched. For each set of parameter combinations, the model is trained using the training set, and the performance is evaluated on the validation set (such as using accuracy as an evaluation indicator), and the parameter combination with the best performance is selected as the parameters of the final model.

[0084] The principle of random search is to randomly sample a certain number of parameter combinations in the parameter space for evaluation. Unlike grid search, it does not traverse all possible parameter combinations, but randomly selects them. In large-scale parameter spaces, random search can find better parameter combinations more quickly, especially when there are a large number of parameters and the parameters have complex interactions. Through multiple random sampling and evaluations, the parameter settings that optimize the model performance are found, further improving the performance and generalization ability of the classifier.

[0085] The method for tuning the parameters of the classifier can also be Bayesian Optimization and Genetic Algorithm. The principle of Bayesian Optimization is: based on Bayes' theorem, a probability model (called a surrogate model, such as a Gaussian process) of the objective function (usually a model evaluation index, such as accuracy) is constructed. According to the evaluated parameter combination and its corresponding objective function value, the surrogate model is used to predict the objective function value of the unevaluated parameter combination, and the next parameter combination with the greatest hope of improving the model performance is selected for evaluation by obtaining functions (such as expected improvement, upper bound of confidence interval, etc.). Bayesian optimization balances between exploring new parameter regions (to find better parameter combinations) and using the evaluated parameter combination information (to avoid repeated evaluation of regions with poor performance), so as to efficiently find the optimal or near-optimal parameter combination and reduce unnecessary parameter adjustments and model training times.

[0086] The principle of the genetic algorithm is: regard the parameter combination as an individual, and search for the optimal parameters by simulating the biological evolution process. First, randomly generate an initial population (including multiple parameter combination individuals), and then calculate the fitness value according to the model performance (such as accuracy) corresponding to each individual. The higher the fitness value, the better the individual. Then, through selection operations (such as roulette selection, tournament selection, etc.), select excellent individuals from the population, perform crossover operations (simulate gene recombination, exchange some parameters of two individuals) and mutation operations (randomly perturb certain parameters of individuals) to generate new populations. Repeat this process, after multiple generations of evolution, the individuals in the population gradually approach the optimal solution, and finally find the best performance parameter combination. Genetic algorithms can search for the global optimal solution in a complex parameter space, but the computational complexity is high, and it is suitable for model optimization problems with large and complex parameter spaces.

[0087] The method also includes: preprocessing the data features; the preprocessing includes data cleaning, denoising, normalization, data enhancement and data dimension reduction; the data cleaning includes removing duplicate data, processing missing values ​​and identifying and correcting erroneous data; the denoising includes filtering and wavelet denoising; the normalization includes maximum and minimum normalization and standardization.

[0088] According to a specific implementation method, the extracted original data features are preprocessed, including data cleaning (removing outliers and erroneous data), denoising (using filtering and other methods to eliminate interference signals), normalization (mapping the values ​​of different features to a specific interval to eliminate the impact of dimensions and make each feature comparable, such as [0, 1] or [-1, 1]), etc., to ensure the quality and consistency of the data and provide a reliable data basis for subsequent classifier training.

[0089] The data cleaning includes removing duplicate data, processing missing values, and identifying and correcting erroneous data. The deduplication includes that in data sets such as equipment operation data and fault records, there may be identical duplicate records. These duplicate data will increase the data volume but do not provide additional information. They may also affect the model training effect, so they need to be deleted.

[0090] The missing value processing method is used to process missing equipment operating parameter values ​​or part of the information in the fault record using a variety of methods. For example, the missing values ​​of continuous features can be filled using the mean, median or mode; for discrete features, the most common value can be selected for filling according to the data distribution, or the missing values ​​can be treated as a special category. In addition, model-based methods (such as using other related features to predict missing values) can also be used for filling, but care should be taken to avoid introducing too many errors.

[0091] The identification and correction of erroneous data includes identifying obviously erroneous data points by setting reasonable data ranges, logical relationships and other rules. For example, the voltage value of the power equipment that exceeds the normal operating range may be erroneous data. For such data, it can be corrected according to the context of the data, historical data or other relevant information, or directly deleted (if the erroneous data is small and cannot be accurately corrected).

[0092] The denoising includes filtering and wavelet denoising. Specifically, the filtering method includes: filtering technology can be used for noise interference in the equipment operation signal. For example, mean filtering smoothes the signal by calculating the average value of the signal in a certain window to remove high-frequency noise; median filtering selects the median in the window as the output, which has a good suppression effect on salt and pepper noise; Gaussian filtering uses Gaussian function to perform weighted averaging on the signal, which is suitable for removing noise that obeys the normal distribution. The wavelet denoising includes: decomposing the signal into wavelet coefficients of different scales, setting a threshold to process the wavelet coefficients according to the distribution characteristics of the noise in the wavelet coefficients (usually the absolute value of the noise coefficient is small), retaining important signal coefficients, removing noise coefficients, and then reconstructing the signal through inverse wavelet transform to achieve the purpose of denoising. Wavelet denoising has advantages in retaining signal detail characteristics and is suitable for processing noise in non-stationary signals.

[0093] The normalization includes minimum and maximum normalization and standardization (Z-Score Normalization). The minimum and maximum normalization includes: mapping the data (original feature data) to a specified interval, such as [0,1] or [-1,1]. Mapping each feature to the [0,1] interval or to the [-1,1] interval. This normalization method can maintain the original distribution shape of the data, and is suitable for situations where the data range is known and the data needs to be compressed to a specific interval, which is conducive to accelerating the convergence of the model.

[0094] The standardization includes: converting the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The standardization method works better when the scale differences between features are large, and can make different features have the same weight in model training. It is suitable for algorithms based on distance calculation (such as K nearest neighbor algorithm) and algorithms using gradient descent optimization (such as neural network), avoiding unstable or inaccurate model training due to feature scale differences.

[0095] The data features are cleaned, denoised, normalized and other preprocessing operations are performed to ensure the quality and consistency of the data features. Data from different data sources are integrated to form a unified data view, which is convenient for subsequent feature extraction and classification processing.

[0096] If there is a source task dataset that is similar to the target task, transfer learning methods can be used to build a classifier. Transfer learning can leverage knowledge from the source task to accelerate the learning of the target task and potentially improve classification accuracy.

[0097] When labeled data is scarce, semi-supervised learning methods can be used to build classifiers. Semi-supervised learning can use unlabeled data to assist the learning of labeled data, thereby improving the performance of the classifier.

[0098] The present application can also construct an end-to-end classification system based on deep learning, which can directly extract features from raw data and classify them without explicit feature extraction and classifier construction steps. Such a system usually has a higher degree of automation and better classification performance.

[0099] The present invention uses programming language to implement the above algorithm (including steps such as feature extraction, classifier construction, model optimization, etc.). By constructing a test set, the classification performance of the algorithm is tested. Test indicators include but are not limited to accuracy, recall rate, etc. At the same time, comparative experiments of different classification algorithms are carried out to verify the superiority of the present invention. The present invention can integrate data from different data sources to form a unified data view, improve the comprehensiveness and accuracy of the data, and do not require manual intervention, improve the efficiency and accuracy of problem classification, and have scalability and flexibility. Through model optimization and parameter tuning, the algorithm has high classification performance and robustness, and can accurately identify and classify various technical problems.

[0100] The present invention realizes automatic learning and optimization classification based on the classification model of machine learning, which significantly improves the classification accuracy of equipment technical problems. The classification model can continuously adapt to new data and problems through adaptive learning and optimization mechanism, which improves the intelligent level of power technical supervision.

[0101] On the other hand, Figure 2As shown, the present invention also proposes a classification device 200 for equipment problems, which includes: an acquisition module 201, used to obtain data features of target equipment; a first processing module 202, used to perform feature selection algorithm processing and feature dimension reduction processing on the data features to obtain key features; a second processing module 203, used to construct a first classifier according to the key features and classification requirements, and train the first classifier through training set data to obtain a second classifier, and use the second classifier to classify the equipment problems of the target equipment. The device can accurately classify technical problems of power technical supervision equipment, and improve the classification accuracy of equipment technical problems.

[0102] A classification method for equipment problems of the present invention comprises: obtaining data features of a target device; performing feature selection algorithm processing and feature dimension reduction processing on the data features to obtain key features; constructing a first classifier according to the key features and classification requirements; training the first classifier with training set data to obtain a second classifier, and using the second classifier to classify equipment problems of the target device. The method automatically learns and optimizes classification based on a machine learning classification model, thereby improving the classification accuracy of equipment technical problems.

[0103] On the other hand, an embodiment of the present invention provides a storage medium having a program stored thereon, which implements the method for classifying device problems when executed by a processor.

[0104] An embodiment of the present invention provides a processor, which is used to run a program, wherein the method for classifying device problems is executed when the program is run.

[0105] The embodiment of the present invention provides a device, the device includes a processor, a memory and a program stored in the memory and can be run on the processor, and the processor implements the following steps when executing the program: obtaining entity data of a physical entity, the entity data including entity motion data and entity graphics; establishing a data group library according to the entity motion data; establishing a model group library according to the entity graphics; setting an index identifier for the physical entity according to the data group library, and setting the same index identifier for the physical entity according to the model group library, and the index identifier is used to retrieve whether the model group library contains the entity graphics of the physical entity to be tested. The device in this article can be a server, a PC, a PAD, a mobile phone, etc.

[0106] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program that initializes the following method steps: obtaining data features of a target device; performing feature selection algorithm processing and feature dimension reduction processing on the data features to obtain key features; constructing a first classifier based on the key features and classification requirements; training the first classifier with training set data to obtain a second classifier, and using the second classifier to classify device problems of the target device.

[0107] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0108] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0109] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0110] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0111] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0112] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0113] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0114] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0115] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for classifying equipment problems, characterized in that: The method includes: Obtain data characteristics of the target device; Performing feature selection algorithm processing and feature dimension reduction processing on the data features to obtain key features; Constructing a first classifier according to the key features and classification requirements; The first classifier is trained using training set data to obtain a second classifier, and the second classifier is used to classify device problems of the target device.

2. The method according to claim 1, characterized in that The step of obtaining data characteristics of the target device includes: Acquire a data source of the target device, the data source including operation data, fault records and maintenance logs; Extracting data features of the data source by using a convolutional neural network and / or a recurrent neural network; The data features include equipment type, fault phenomenon, fault code and operating parameters.

3. The method according to claim 1, characterized in that The feature selection algorithm is at least one of a chi-square test, mutual information and recursive feature elimination; The feature dimension reduction process includes principal component analysis and linear discriminant analysis; The principal component analysis is to project the original high-dimensional data into a low-dimensional space through linear transformation to reduce the dimension of the data features; The linear discriminant analysis is to determine the best projection direction according to the inter-class discreteness and intra-class discreteness of the data features.

4. The method according to claim 1, characterized in that: The constructing a first classifier according to the key features and classification requirements includes: determining a classification method according to the classification requirements; Constructing a first classifier for the key features according to the classification method; The classification method is at least one of a support vector machine, a random forest and a neural network.

5. The method according to claim 1, characterized in that: The constructing a first classifier according to the key features and classification requirements also includes: Determine a machine learning algorithm according to the classification requirements; Constructing a first classifier for the key features according to the machine learning algorithm; The machine learning algorithm is at least one of a gradient boosting decision tree, extreme gradient boosting, and a lightweight gradient boosting machine.

6. The method according to claim 1, 4 or 5, characterized in that: The first classifier is a hierarchical classifier, including an upper classifier and a lower classifier; The upper classifier is used to classify faults into major categories, and the lower classifier is used to classify faults into minor categories.

7. The method according to claim 1, characterized in that The method further includes: According to the electrical connection and operation coupling relationship of the multiple classifiers, the outputs of the multiple classifiers are weightedly fused to classify the device problems of the target device.

8. The method according to claim 1, characterized in that: The method further includes: Preprocessing the data features; The preprocessing includes data cleaning, denoising, normalization, data enhancement and data dimension reduction; The data cleaning includes removing duplicate data, processing missing values, and identifying and correcting erroneous data; The denoising includes filtering and wavelet denoising; The normalization includes maximum and minimum normalization and standardization.

9. The method according to claim 1, characterized in that: The method further includes: The parameters of the second classifier are tuned by using at least one of cross validation, grid search, random search, Bayesian optimization and genetic algorithm.

10. A device for classifying equipment problems, characterized in that: The device includes: An acquisition module, used to acquire data characteristics of a target device; A first processing module is used to perform feature selection algorithm processing and feature dimension reduction processing on the data features to obtain key features; The second processing module is used to construct a first classifier according to the key features and classification requirements, and train the first classifier through training set data to obtain a second classifier, and use the second classifier to classify the device problems of the target device.

11. A machine-readable storage medium having instructions stored thereon, characterized in that: The instruction is used to enable the machine to execute the device problem classification method described in any one of claims 1-9.

Images