Fault diagnosis method and device based on semi-supervised support vector machine, terminal equipment and computer readable storage medium
Through iterative training of the semi-supervised support vector machine model, the problem of lack of label data in circuit breaker fault diagnosis is solved, and the accuracy and efficiency of fault diagnosis is improved, and it is suitable for complex power system environments.
Patent Information
- Application Number
- CN202411112934.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-14
- Publication Date
- 2024-11-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional circuit breaker fault diagnosis methods rely on manual inspection and empirical judgment, which are inefficient and insufficiently accurate. Especially in the absence of label data, it is difficult to effectively identify fault characteristics, affecting the fault diagnosis accuracy of the power system.
Adopt the fault diagnosis method based on semi-supervised support vector machine, by obtaining unlabeled and labeled data, performing sample overlap and time-domain feature extraction, and using semi-supervised support vector machine model for iterative training, gradually enriching the labeled data, and improving model accuracy .
It reduces the workload and cost of labeling, improves the accuracy and efficiency of fault diagnosis, makes up for the lack of label data in traditional supervised learning, and enhances the accuracy of the model.
Smart Images

Figure CN119024150A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of high-voltage circuit breaker fault diagnosis, and in particular to a fault diagnosis method based on a semi-supervised support vector machine. Background Art
[0002] With the continuous progress and development of science and technology, the power industry has widely applied technologies such as artificial intelligence and virtual reality in the trend of digitalization, showing new characteristics of information perception, data fusion and real-time decision-making. Against this background, countries around the world have increased their investment in the field of intelligent manufacturing. The "Digital Twin White Paper (2019)" released by the Ministry of Industry and Information Technology in 2019 clearly pointed out that the world has entered the era of intelligent manufacturing. Among them, digital twins, as an important direction, have become one of the key development directions in the field of industrial manufacturing in recent years. Gartner believes that digital twin technology has huge development potential in the future. The future world will be divided into the physical world and the virtual world. Digital twin technology can reflect the status of the physical world in real time. Changes in external conditions can be presented in real time in the twin model and data analysis and processing can be performed, so as to monitor and analyze the current situation and ensure the efficient operation of all parts in the physical world.
[0003] In this context, fault diagnosis of power equipment has also ushered in new development opportunities. Traditional circuit breaker fault diagnosis methods often rely on manual inspection and experience judgment, which has the problems of low efficiency and insufficient accuracy. With the help of digital twin technology, a virtual model of the circuit breaker can be created, and rapid fault diagnosis and predictive maintenance can be achieved through real-time monitoring and data analysis of its operating status. Digital twin technology can simulate the behavior of circuit breakers under various working conditions, timely discover potential problems, and provide optimization suggestions, thereby improving the reliability and safety of power systems. In the fault diagnosis process, there are still some significant problems, the most important of which is that most fault collection data lacks labels. The lack of labeled data makes it difficult to apply traditional supervised learning methods, because these methods usually rely on a large amount of labeled data to train models to ensure the accuracy and reliability of fault diagnosis. The existence of unlabeled data makes it difficult to clearly identify fault features, which in turn affects the accuracy of fault diagnosis. This situation is particularly common in power systems, because their complex operating environment and variable fault types make it very difficult to fully and accurately label each fault data. Summary of the invention
[0004] The embodiments of the present invention provide a fault diagnosis method, apparatus, terminal device and computer-readable storage medium based on a semi-supervised support vector machine, which can use a small amount of labeled data and a large amount of labeled data to train the model, thereby reducing the workload and cost of labeling while still obtaining high-quality training data.
[0005] An embodiment of the present invention provides a fault diagnosis method based on a semi-supervised support vector machine, comprising:
[0006] Acquiring a first vibration signal of the circuit breaker to be diagnosed;
[0007] Inputting the first vibration signal into the trained semi-supervised support vector machine model, so that the semi-supervised support vector machine model performs fault diagnosis on the circuit breaker to be diagnosed according to the first vibration signal;
[0008] The training of the semi-supervised support vector machine model includes the following steps:
[0009] Acquire second vibration signals of a plurality of circuit breakers with unknown working states, and perform sample overlap and time domain feature extraction on each second vibration signal to generate first unlabeled data;
[0010] Acquire third vibration signals of several circuit breakers with known working states, perform sample overlap and time domain feature extraction on each third vibration signal, and generate first label data according to the time domain feature and the corresponding working state;
[0011] The first unlabeled data and the first labeled data are normalized and recursively feature eliminated to obtain second unlabeled data and second labeled data;
[0012] The second unlabeled data and the second labeled data are input into the semi-supervised support vector machine model to be trained for iterative training until all the unlabeled data are allocated, thereby obtaining a trained semi-supervised support vector machine model; wherein, in each training, the current second labeled data is used to construct the current support vector machine sub-model; one of the current second unlabeled data is obtained as the data to be allocated; the probability that the data to be allocated belongs to each type of working state is calculated by the support vector machine sub-model, and the highest probability among each type of probability is used as the probability to be compared; the probability to be compared is compared with a preset threshold value, and if the probability to be compared is greater than the preset threshold value, the data to be allocated is added to the second labeled data, otherwise, the data to be allocated is added to other fault data; and the second labeled data and the second unlabeled data are updated.
[0013] Further, time domain features, including: standard deviation, root mean square and skewness value;
[0014] Performing time domain feature extraction on each second vibration signal, including:
[0015] The standard deviation of the second vibration signal is calculated by the following formula:
[0016]
[0017] The root mean square of the second vibration signal is calculated by the following formula:
[0018]
[0019] The skew value of the second vibration signal is calculated by the following formula:
[0020]
[0021] Among them, S represents the standard deviation, X RMS represents the root mean square, SKE represents the skewness value, N represents the number of data points of the second vibration signal, X i represents the value of the i-th data point of the second vibration signal, represents the average value of the second vibration signal.
[0022] Further, the working state includes: fault state and normal state;
[0023] Fault conditions include: loose bolts on the operating mechanism housing, spring fatigue, reduced operating voltage, and iron core jamming.
[0024] Further, the first unlabeled data and the first labeled data are normalized and recursively feature eliminated to obtain second unlabeled data and second labeled data, including:
[0025] Performing empirical mode decomposition on the first unlabeled data and the first labeled data to obtain mode function components;
[0026] Perform Hilbert transform on the modal function components to obtain the marginal spectrum energy;
[0027] The marginal spectrum energy and time domain features are summarized to obtain a data set;
[0028] The data set is normalized, and a recursive feature elimination method is used to obtain second unlabeled data and second labeled data.
[0029] Furthermore, the modal function components are Hilbert transformed to obtain the marginal spectrum energy, including:
[0030] Perform Hilbert transform on the modal function components to obtain the instantaneous frequency and instantaneous amplitude of each modal function component;
[0031] The instantaneous frequency and instantaneous amplitude of each modal function component of the complete time are obtained by continuous sampling, and the Hilbert spectrum is constructed;
[0032] The marginal spectral energy is obtained by time-integrating the Hilbert spectrum.
[0033] Furthermore, the data set is normalized, including:
[0034] Perform a linear transformation on the data set and map the data values in the data set to between [0,1] using the following formula:
[0035]
[0036] Among them, X ′ represents the data after normalization, and X represents the data before normalization.
[0037] Furthermore, the fault diagnosis method based on the semi-supervised support vector machine also includes:
[0038] After the iterative training of the semi-supervised support vector machine model is completed, the updated second labeled data is used as the test set data;
[0039] Through the trained semi-supervised support vector machine model, each data in the test set data is classified according to different continuous time windows, and according to the working status mark of each data in the test set data, the fault classification results of each data according to different continuous time windows are obtained;
[0040] According to the fault classification results of each data in different continuous time windows, the accuracy of the fault classification results of different continuous time windows is calculated, and the time window with the highest accuracy is taken as the optimal judgment time window;
[0041] The optimal judgment time window is used to adjust the trained semi-supervised support vector machine model.
[0042] Based on the above method embodiment, the present invention provides a corresponding device embodiment, including: a first signal acquisition module, a fault diagnosis module and a model training module;
[0043] A first signal acquisition module, used for acquiring a first vibration signal of the circuit breaker to be diagnosed;
[0044] A fault diagnosis module, used for inputting the first vibration signal into a trained semi-supervised support vector machine model, so that the semi-supervised support vector machine model performs fault diagnosis on the circuit breaker to be diagnosed according to the first vibration signal;
[0045] A model training module, comprising: a second signal acquisition submodule, a third signal acquisition submodule, a training data processing submodule, and an iterative training submodule;
[0046] A second signal acquisition submodule, used for acquiring second vibration signals of a plurality of circuit breakers with unknown working states, and performing sample overlap and time domain feature extraction on each second vibration signal to generate first unlabeled data;
[0047] A third signal acquisition submodule, used to acquire third vibration signals of a plurality of circuit breakers with known working states, perform sample overlap and time domain feature extraction on each third vibration signal, and generate first label data according to the time domain feature and the corresponding working state;
[0048] A training data processing submodule, configured to obtain second unlabeled data and second labeled data by normalizing and recursively eliminating features of the first unlabeled data and the first labeled data;
[0049] An iterative training submodule is used to input the second unlabeled data and the second labeled data into the semi-supervised support vector machine model to be trained for iterative training until all the unlabeled data are allocated, thereby obtaining a trained semi-supervised support vector machine model; wherein, during each training, the current second labeled data is used to construct the current support vector machine submodel; one of the current second unlabeled data is obtained as the data to be allocated; the probability that the data to be allocated belongs to each type of working state is calculated by the support vector machine submodel, and the highest probability among each type of probability is used as the probability to be compared; the probability to be compared is compared with a preset threshold, and if the probability to be compared is greater than the preset threshold, the data to be allocated is added to the second labeled data, otherwise, the data to be allocated is added to other fault data; and the second labeled data and the second unlabeled data are updated.
[0050] Based on the above-mentioned method item embodiment, the present invention provides a corresponding terminal device item embodiment, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps of the fault diagnosis method based on the semi-supervised support vector machine as described in the present invention are implemented.
[0051] Based on the above-mentioned method item embodiment, the present invention provides a corresponding computer-readable storage medium item embodiment, including: a stored computer program, which controls the device where the computer-readable storage medium is located to execute the steps of the fault diagnosis method based on the semi-supervised support vector machine as described in the present invention when the computer program is running.
[0052] Compared with the prior art, the beneficial effects of the embodiment of this solution are:
[0053] The present invention uses a trained semi-supervised support vector machine model to perform fault diagnosis. In the training stage, by obtaining second vibration signals of several circuit breakers with unknown working states and third vibration signals of several circuit breakers with known working states, first unlabeled data and first labeled data are obtained through sample overlap and time domain feature extraction; then, normalization processing and recursive feature elimination are performed through data preprocessing to convert the first unlabeled data and the first labeled data into second unlabeled data and second labeled data; next, the second unlabeled data and the second labeled data are input into the semi-supervised support vector machine model for iterative training. In each training, the support vector machine sub-model is constructed using the current second labeled data, and the probability that the current second unlabeled data belongs to each type of working state is calculated, and the highest probability among each type of probability is used as the probability to be compared, then, the probability to be compared is compared with a preset threshold, and the data with the probability to be compared greater than the preset threshold is added to the labeled data, otherwise, the data to be assigned is added to other fault data, and the second labeled data and the second unlabeled data are updated, so that the information of the unlabeled data is fully utilized, and the deficiency of lacking label data in traditional supervised learning is made up. Through such an iterative training process, the number of labeled data is gradually enriched, and the accuracy of the model is improved.
[0054] In summary, the present invention fully utilizes the information of unlabeled data through the semi-supervised learning method, solves the problem that most collected data in traditional supervised learning lack labels, and thus improves the accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] Figure 1 It is a flow chart of a fault diagnosis method based on a semi-supervised support vector machine provided by an embodiment of the present invention;
[0056] Figure 2 It is a schematic diagram of the process of training a semi-supervised support vector machine model provided by an embodiment of the present invention;
[0057] Figure 3 is a window overlapping spectrum of a circuit breaker vibration signal provided by an embodiment of the present invention;
[0058] Figure 4 This is a vibration signal diagram collected when the circuit breaker provided by an embodiment of the present invention is in a core stuck state;
[0059] Figure 5 It is a variation trend diagram of the fault diagnosis accuracy of the SVM model under different numbers of features in recursive feature elimination provided by an embodiment of the present invention;
[0060] Figure 6 is a graph showing a change trend of a continuous judgment time window and an algorithm accuracy rate provided by an embodiment of the present invention;
[0061] Figure 7 This is a comparison chart of the accuracy of the SVM algorithm and the S3VM algorithm combined with the voting mechanism provided in one embodiment of the present invention;
[0062] Figure 8 It is a structural schematic diagram of a fault diagnosis device based on a semi-supervised support vector machine provided in one embodiment of the present invention. DETAILED DESCRIPTION
[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0064] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features.
[0065] like Figure 1 As shown, an embodiment of the present invention provides a fault diagnosis method based on a semi-supervised support vector machine, the method comprising at least the following steps:
[0066] Step S1: Acquire a first vibration signal of a circuit breaker to be diagnosed;
[0067] For step S1: In the present invention, taking the SF6 circuit breaker as an example, the vibration signal of the SF6 circuit breaker to be diagnosed is sampled by a piezoelectric acceleration sensor at a sampling frequency of 250 kHz and a sampling time of t when the SF6 circuit breaker to be diagnosed is in an operating state to obtain a first vibration signal.
[0068] Step S2: inputting the first vibration signal into the trained semi-supervised support vector machine model, so that the semi-supervised support vector machine model performs fault diagnosis on the circuit breaker to be diagnosed according to the first vibration signal;
[0069] For step S2:
[0070] First, the training of the semi-supervised support vector machine model is described in detail:
[0071] like Figure 2 As shown, the training of the above semi-supervised support vector machine model includes the following steps:
[0072] Step S21: obtaining second vibration signals of a plurality of circuit breakers with unknown working states, and performing sample overlap and time domain feature extraction on each second vibration signal to generate first unlabeled data;
[0073] In a preferred embodiment, the time domain features include: standard deviation, root mean square and skewness value;
[0074] Performing time domain feature extraction on each second vibration signal, including:
[0075] The standard deviation of the second vibration signal is calculated by the following formula:
[0076]
[0077] The root mean square of the second vibration signal is calculated by the following formula:
[0078]
[0079] The skew value of the second vibration signal is calculated by the following formula:
[0080]
[0081] Among them, S represents the standard deviation, X RMS represents the root mean square, SKE represents the skewness value, N represents the number of data points of the second vibration signal, X i represents the value of the i-th data point of the second vibration signal, represents the average value of the second vibration signal.
[0082] For step S21, Figure 3 As shown, a piezoelectric acceleration sensor is used to collect several second vibration signals as training samples at a sampling frequency of 250kHz and a sampling time of 0.4s / time when the circuit breaker is running in an unknown working state, that is, the number of sample data points of each second vibration signal is 100000; then, the data enhancement technology of overlapping samples is used, the window length is set to 50000, the overlap amount is 5000, and each second vibration signal is overlappedly sampled through the window, so that each complete second vibration signal is expanded to obtain samples of 11 window length periods, and more data samples are obtained after the samples are overlapped.
[0083] In the data samples after sample overlap processing, time domain features are extracted for each data sample, and the standard deviation, root mean square and skewness value of each second vibration signal are calculated by formula, and finally the first unlabeled data are generated.
[0084] Step S22: obtaining third vibration signals of several circuit breakers with known working states, performing sample overlap and time domain feature extraction on each third vibration signal, and generating first label data according to the time domain feature and the corresponding working state;
[0085] In a preferred embodiment, the working state includes: a fault state and a normal state;
[0086] Fault conditions include: loose bolts on the operating mechanism housing, spring fatigue, reduced operating voltage, and iron core jamming.
[0087] For step S22, a plurality of groups of vibration signals are collected by the piezoelectric acceleration sensor under normal working conditions, loose bolts of the operating mechanism housing, spring fatigue, reduced operating voltage, and core jamming. In the present invention, 20 groups of vibration signal data are collected for each working state, and a total of 100 groups of vibration signal data are obtained. At the same time, the data enhancement technology of overlapping samples is used, and the window length is selected as 50,000 and the overlap amount is 5,000 to enhance the vibration signal, and finally obtain data samples of 1,100 window length periods; then, time domain feature extraction is performed on each data sample, and the standard deviation, root mean square and skewness value of each third vibration signal are calculated by formula, and the first label data is generated according to these time domain features and the corresponding working state.
[0088] It should be noted that the second vibration signal and the third vibration signal are essentially vibration signals belonging to the circuit breaker, and the difference lies in the known and unknown working states. The working state of the second vibration signal is unknown, while the working state of the third vibration signal is known. Therefore, the calculation formulas of the standard deviation, root mean square and skewness value of the second vibration signal can be used to calculate the corresponding time domain characteristics of the third vibration signal.
[0089] Step S23: normalizing the first unlabeled data and the first labeled data and performing recursive feature elimination to obtain second unlabeled data and second labeled data;
[0090] In a preferred embodiment, the first unlabeled data and the first labeled data are normalized and recursively feature eliminated to obtain the second unlabeled data and the second labeled data, including:
[0091] Performing empirical mode decomposition on the first unlabeled data and the first labeled data to obtain mode function components;
[0092] Perform Hilbert transform on the modal function components to obtain the marginal spectrum energy;
[0093] The marginal spectrum energy and time domain features are summarized to obtain a data set;
[0094] The data set is normalized, and a recursive feature elimination method is used to obtain second unlabeled data and second labeled data.
[0095] In a preferred embodiment, the modal function components are subjected to Hilbert transformation to obtain marginal spectrum energy, including:
[0096] Perform Hilbert transform on the modal function components to obtain the instantaneous frequency and instantaneous amplitude of each modal function component;
[0097] The instantaneous frequency and instantaneous amplitude of each modal function component of the complete time are obtained by continuous sampling, and the Hilbert spectrum is constructed;
[0098] The marginal spectral energy is obtained by time-integrating the Hilbert spectrum.
[0099] In a preferred embodiment, normalizing the data set includes:
[0100] Perform a linear transformation on the data set and map the data values in the data set to between [0,1] using the following formula:
[0101]
[0102] Among them, X ′ represents the data after normalization, and X represents the data before normalization.
[0103] like Figure 4 The figure shows the vibration signal collected when the circuit breaker is in the core stuck state. It is impossible to directly judge whether the equipment is faulty based on the signal amplitude change state. Therefore, it is necessary to extract the intrinsic eigenvalues through empirical mode decomposition and use the energy distribution, frequency distribution and time domain characteristics of each modal component to judge the fault type.
[0104] In one embodiment of the present invention, the first labeled data obtained in step S22 is subjected to empirical mode decomposition, and each signal sample in the first unlabeled data and the first labeled data is decomposed into a finite number of intrinsic mode functions (IMFs). In circuit breaker fault diagnosis or signal processing, 5 to 8 modal function components can be selected for decomposition. In this embodiment, each signal sample is decomposed to obtain 8 modal function components; then, the modal function components are subjected to Hilbert transform to obtain the instantaneous frequency and instantaneous amplitude of each modal function component. By continuous sampling, the instantaneous frequency and instantaneous amplitude of each modal function component in the complete time range are obtained, thereby constructing the Hilbert spectrum E(ω):
[0105]
[0106] Among them, h(ω) represents the marginal spectrum of the signal, s represents the length of the signal, and t represents the sampling time;
[0107] Subsequently, the Hilbert spectrum is time-integrated to obtain the marginal spectrum energy of each modal function component, thereby obtaining the energy distribution of each modal function component in different frequency ranges for subsequent feature extraction or fault diagnosis analysis;
[0108] The marginal spectrum energy and the time domain features obtained in step S22 are aggregated to obtain a data set; the data set is then normalized using a linear normalization method of minimum-maximum normalization to map the data values in the data set to between [0,1], eliminating the order of magnitude differences between the data in each dimension, and avoiding excessive algorithm prediction errors caused by large order of magnitude differences in input and output data;
[0109] Recursive Feature Elimination (RFE) is used to reduce the size of the feature set to find the best features for model performance. First, a SVM model is built based on the normalized data set, and the weight of each feature is calculated. Then, based on the size of the feature weight, the most important features are selected and the least important features are removed. If the number of features after screening reaches the preset number of features, the recursive feature elimination is stopped. Otherwise, the SVM model is retrained to select the most important features for the next round and remove the least important features. This process is repeated until the specified number of features is reached. Figure 5 The figure shows the changing trend of the fault diagnosis accuracy of the SVM model under different numbers of features in recursive feature elimination. It can be seen from the figure that when the number of features is 10, the fault diagnosis accuracy of the SVM model reaches the highest 91%. Therefore, this embodiment selects 10 feature parameters, removes unimportant feature data, and forms the second unlabeled data and the second labeled data.
[0110] Step S24: input the second unlabeled data and the second labeled data into the semi-supervised support vector machine model to be trained for iterative training until all the unlabeled data are allocated, thereby obtaining a trained semi-supervised support vector machine model; wherein, in each training, the current second labeled data is used to construct the current support vector machine sub-model; obtain one of the current second unlabeled data as the data to be allocated; calculate the probability that the data to be allocated belongs to each type of working state through the support vector machine sub-model, and take the highest probability among each type of probability as the probability to be compared; compare the probability to be compared with a preset threshold, if the probability to be compared is greater than the preset threshold, add the data to be allocated to the second labeled data, otherwise, add the data to be allocated to other fault data; update the second labeled data and the second unlabeled data.
[0111] In one embodiment of the present invention, the second unlabeled data and the second labeled data obtained in step S23 are used as training data and input into a semi-supervised support vector machine (Semi-Supervised Support Vector Machine, S3VM) model to be trained for iterative training, wherein the kernel function of the support vector machine is set to the 'rbf' type and the penalty parameter C is set to 10. Firstly, a support vector machine sub-model is trained using a small amount of labeled data, namely the second labeled data. The training process of the initial model is completed using a supervised learning algorithm (SVM). Then, the unlabeled data is predicted by the support vector machine sub-model, and a sample is selected from the second unlabeled data as the data to be assigned. The current support vector machine sub-model is used to calculate the probability that the data to be assigned belongs to each type of working state, and the category with the highest probability is selected as the probability to be compared. Next, the probability to be compared is compared with a preset threshold, and the preset threshold is set to 0.9. If the probability to be compared is greater than the preset threshold, the data to be assigned is added to the second labeled data, indicating that the sample belongs to this type of working state. Otherwise, the data to be assigned is added to other fault data, the second labeled data and the second unlabeled data are updated, and the assigned data is removed from the second unlabeled data. Finally, the model is retrained using the updated labeled data set until all unlabeled data are assigned, thereby obtaining a trained semi-supervised support vector machine model.
[0112] It should be noted that the second unlabeled data may contain data distribution and features that are not covered in the second labeled data. By continuously assigning these new samples to the known labeled data, more comprehensive and diverse information can be provided, which helps to improve the accuracy of the model's classification and fault diagnosis of different categories and ensure the safe and stable operation of power equipment.
[0113] In a preferred embodiment, the fault diagnosis method based on the semi-supervised support vector machine further includes:
[0114] After the iterative training of the semi-supervised support vector machine model is completed, the updated second labeled data is used as the test set data;
[0115] Through the trained semi-supervised support vector machine model, each data in the test set data is classified according to different continuous time windows, and according to the working status mark of each data in the test set data, the fault classification results of each data according to different continuous time windows are obtained;
[0116] According to the fault classification results of each data in different continuous time windows, the accuracy of the fault classification results of different continuous time windows is calculated, and the time window with the highest accuracy is taken as the optimal judgment time window;
[0117] The optimal judgment time window is used to adjust the trained semi-supervised support vector machine model.
[0118] In one embodiment of the present invention, after the iterative training of the semi-supervised support vector machine model is completed, the updated second labeled data is used as the test set data; then, the trained semi-supervised support vector machine model is used to classify the faults of each data in the test set data according to 11 time windows with different window lengths. According to the working state label of each data in the test set data, we can obtain the fault classification results of each data according to the time windows with 11 different window lengths, and then use the voting mechanism to comprehensively evaluate the prediction results of multiple time windows, and at the same time set the weight of the judgment result of each time window to the final result to be the same, so as to obtain the fault diagnosis accuracy of different continuous time windows, such as Figure 6 As shown in the figure, the algorithm accuracy from 1 to 11 consecutive time windows is displayed, and 4 consecutive time windows with the highest accuracy are selected as the optimal judgment time windows for fault diagnosis using the trained semi-supervised support vector machine model.
[0119] like Figure 7 The figure shows the accuracy comparison of the supervised SVM algorithm and the S3VM combined with voting mechanism. It can be seen that the accuracy of the SVM algorithm and the S3VM combined with voting mechanism are different. The S3VM combined with voting mechanism algorithm selects 4 time windows, and the total judgment time of the 4 time windows is 0.26s. The judgment time is short. The accuracy of the S3VM combined with voting mechanism algorithm can reach 100%, while the accuracy of the SVM algorithm is 92%. Therefore, the S3VM combined with voting mechanism algorithm can achieve better diagnostic results for actual circuit breaker fault diagnosis, improve the reliability of algorithm judgment during online diagnosis, and improve the speed of fault diagnosis, which plays an important role in improving the ability of circuit breakers to operate safely and stably.
[0120] like Figure 8 As shown, based on the above method embodiment, a corresponding device embodiment is provided;
[0121] An embodiment of the present invention provides a fault diagnosis device based on a semi-supervised support vector machine, comprising: a first signal acquisition module, a fault diagnosis module and a model training module;
[0122] A first signal acquisition module, used for acquiring a first vibration signal of the circuit breaker to be diagnosed;
[0123] A fault diagnosis module, used for inputting the first vibration signal into a trained semi-supervised support vector machine model, so that the semi-supervised support vector machine model performs fault diagnosis on the circuit breaker to be diagnosed according to the first vibration signal;
[0124] A model training module, comprising: a second signal acquisition submodule, a third signal acquisition submodule, a training data processing submodule, and an iterative training submodule;
[0125] A second signal acquisition submodule, used for acquiring second vibration signals of a plurality of circuit breakers with unknown working states, and performing sample overlap and time domain feature extraction on each second vibration signal to generate first unlabeled data;
[0126] A third signal acquisition submodule, used to acquire third vibration signals of a plurality of circuit breakers with known working states, perform sample overlap and time domain feature extraction on each third vibration signal, and generate first label data according to the time domain feature and the corresponding working state;
[0127] A training data processing submodule, configured to obtain second unlabeled data and second labeled data by normalizing and recursively eliminating features of the first unlabeled data and the first labeled data;
[0128] An iterative training submodule is used to input the second unlabeled data and the second labeled data into the semi-supervised support vector machine model to be trained for iterative training until all the unlabeled data are allocated, thereby obtaining a trained semi-supervised support vector machine model; wherein, during each training, the current second labeled data is used to construct the current support vector machine submodel; one of the current second unlabeled data is obtained as the data to be allocated; the probability that the data to be allocated belongs to each type of working state is calculated by the support vector machine submodel, and the highest probability among each type of probability is used as the probability to be compared; the probability to be compared is compared with a preset threshold, and if the probability to be compared is greater than the preset threshold, the data to be allocated is added to the second labeled data, otherwise, the data to be allocated is added to other fault data; and the second labeled data and the second unlabeled data are updated.
[0129] In a preferred embodiment, the third signal acquisition submodule includes: a standard deviation calculation unit, a root mean square calculation unit, and a skewness value calculation unit;
[0130] The standard deviation calculation unit is used to calculate the standard deviation of the second vibration signal using the following formula:
[0131]
[0132] The root mean square calculation unit is used to calculate the root mean square of the second vibration signal using the following formula:
[0133]
[0134] The skewness value calculation unit is used to calculate the skewness value of the second vibration signal by the following formula:
[0135]
[0136] Among them, S represents the standard deviation, X RMS represents the root mean square, SKE represents the skewness value, N represents the number of data points of the second vibration signal, X i represents the value of the i-th data point of the second vibration signal, represents the average value of the second vibration signal.
[0137] In a preferred embodiment, the training data processing submodule includes: an empirical mode decomposition unit, a marginal spectrum energy calculation unit, a training data aggregation unit, and a training data processing unit;
[0138] An empirical mode decomposition unit, used to perform empirical mode decomposition on the first unlabeled data and the first labeled data to obtain a mode function component;
[0139] A marginal spectrum energy calculation unit is used to perform Hilbert transformation on the modal function components to obtain marginal spectrum energy;
[0140] A training data aggregation unit, used for aggregating marginal spectrum energy and time domain features to obtain a data set;
[0141] The training data processing unit is used to normalize the data set and obtain the second unlabeled data and the second labeled data by a recursive feature elimination method.
[0142] In a preferred embodiment, the marginal spectrum energy calculation unit includes: a Hilbert transform subunit, a Hilbert spectrum construction subunit, and a marginal spectrum energy calculation subunit;
[0143] A Hilbert transform subunit, used for performing Hilbert transform on the modal function components to obtain the instantaneous frequency and instantaneous amplitude of each modal function component;
[0144] A Hilbert spectrum construction subunit is used to obtain the instantaneous frequency and instantaneous amplitude of each modal function component of the complete time by continuous sampling to construct the Hilbert spectrum;
[0145] The marginal spectrum energy calculation subunit is used to perform time integration on the Hilbert spectrum to obtain the marginal spectrum energy.
[0146] In a preferred embodiment, the training data processing unit includes: a normalization processing subunit;
[0147] The normalization processing subunit is used to perform linear transformation on the data set and map the data values in the data set to between [0, 1] using the following formula:
[0148]
[0149] Among them, X ′ represents the data after normalization, and X represents the data before normalization.
[0150] In a preferred embodiment, the fault diagnosis device based on the semi-supervised support vector machine further includes: an optimal judgment time window acquisition module;
[0151] An optimal judgment time window acquisition module is used to use the updated second labeled data as test set data after the iterative training of the semi-supervised support vector machine model is completed;
[0152] Through the trained semi-supervised support vector machine model, each data in the test set data is classified according to different continuous time windows, and according to the working status mark of each data in the test set data, the fault classification results of each data according to different continuous time windows are obtained;
[0153] According to the fault classification results of each data in different continuous time windows, the accuracy of the fault classification results of different continuous time windows is calculated, and the time window with the highest accuracy is taken as the optimal judgment time window;
[0154] The optimal judgment time window is used to adjust the trained semi-supervised support vector machine model.
[0155] It can be understood that the above-mentioned device item embodiments correspond to the method item embodiments of the present invention, and can implement the fault diagnosis method based on the semi-supervised support vector machine provided by any one of the above-mentioned method item embodiments of the present invention.
[0156] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art may understand and implement the present invention without creative work.
[0157] Based on the above-mentioned embodiment of the fault diagnosis method based on a semi-supervised support vector machine, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the fault diagnosis method based on a semi-supervised support vector machine of any embodiment of the present invention is implemented.
[0158] Exemplarily, in this embodiment, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more module elements may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the terminal device.
[0159] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.
[0160] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.
[0161] Based on the above method embodiment, another embodiment is provided: another embodiment of the present invention provides a computer-readable storage medium, including a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the fault diagnosis method based on the semi-supervised support vector machine described in any one of the above method embodiments of the present invention.
[0162] Wherein, the module / unit integrated in the fault diagnosis device / terminal equipment based on the semi-supervised support vector machine, if implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.
[0163] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.
Claims
1. A fault diagnosis method based on semi-supervised support vector machine, characterized in that: include: Acquire a first vibration signal of the circuit breaker to be diagnosed; Inputting the first vibration signal into a trained semi-supervised support vector machine model, so that the semi-supervised support vector machine model performs fault diagnosis on the circuit breaker to be diagnosed according to the first vibration signal; The training of the semi-supervised support vector machine model comprises the following steps: Acquire second vibration signals of a plurality of circuit breakers with unknown working states, and perform sample overlap and time domain feature extraction on each second vibration signal to generate first unlabeled data; Acquire third vibration signals of several circuit breakers with known working states, perform sample overlap and time domain feature extraction on each third vibration signal, and generate first label data according to the time domain feature and the corresponding working state; The first unlabeled data and the first labeled data are subjected to normalization processing and recursive feature elimination to obtain second unlabeled data and second labeled data; The second unlabeled data and the second labeled data are input into the semi-supervised support vector machine model to be trained for iterative training until all the unlabeled data are allocated, thereby obtaining a trained semi-supervised support vector machine model; wherein, during each training, the current second labeled data is used to construct the current support vector machine sub-model; one of the current second unlabeled data is obtained as the data to be allocated; the probability that the data to be allocated belongs to each type of working state is calculated by the support vector machine sub-model, and the highest probability among each type of probability is used as the probability to be compared; the probability to be compared is compared with a preset threshold value, and if the probability to be compared is greater than the preset threshold value, the data to be allocated is added to the second labeled data, otherwise, the data to be allocated is added to other fault data; and the second labeled data and the second unlabeled data are updated.
2. The fault diagnosis method based on semi-supervised support vector machine according to claim 1 is characterized in that: The time domain characteristics include: standard deviation, root mean square and skewness value; Performing time domain feature extraction on each second vibration signal, including: The standard deviation of the second vibration signal is calculated by the following formula: The root mean square of the second vibration signal is calculated by the following formula: The skew value of the second vibration signal is calculated by the following formula: Among them, S represents the standard deviation, X RMS represents the root mean square, SKE represents the skewness value, N represents the number of data points of the second vibration signal, X i represents the value of the i-th data point of the second vibration signal, represents the average value of the second vibration signal.
3. The fault diagnosis method based on semi-supervised support vector machine according to claim 1 is characterized in that: The working state includes: a fault state and a normal state; The fault conditions include: loose bolts of the operating mechanism box, spring fatigue, reduced operating voltage, and iron core jamming.
4. The fault diagnosis method based on semi-supervised support vector machine according to claim 1 is characterized in that: The first unlabeled data and the first labeled data are subjected to normalization processing and recursive feature elimination to obtain second unlabeled data and second labeled data, including: Performing empirical mode decomposition on the first unlabeled data and the first labeled data to obtain mode function components; Performing Hilbert transformation on the modal function components to obtain marginal spectrum energy; Summarizing the marginal spectrum energy and the time domain features to obtain a data set; The data set is normalized, and a recursive feature elimination method is used to obtain second unlabeled data and second labeled data.
5. The fault diagnosis method based on semi-supervised support vector machine according to claim 4 is characterized in that: The modal function components are subjected to Hilbert transformation to obtain marginal spectrum energy, including: Performing Hilbert transform on the modal function components to obtain the instantaneous frequency and instantaneous amplitude of each modal function component; The instantaneous frequency and instantaneous amplitude of each modal function component of the complete time are obtained by continuous sampling, and the Hilbert spectrum is constructed; The Hilbert spectrum is time-integrated to obtain marginal spectrum energy.
6. The fault diagnosis method based on semi-supervised support vector machine according to claim 4 is characterized in that: Normalize the data set, including: Perform a linear transformation on the data set and map the data values in the data set to between [0,1] using the following formula: Among them, X ′ represents the data after normalization, and X represents the data before normalization.
7. The fault diagnosis method based on semi-supervised support vector machine according to claim 1 is characterized in that: Also includes: After the iterative training of the semi-supervised support vector machine model is completed, the updated second labeled data is used as the test set data; Through the trained semi-supervised support vector machine model, each data in the test set data is classified according to different continuous time windows, and according to the working status mark of each data in the test set data, the fault classification results of each data according to different continuous time windows are obtained; According to the fault classification results of each data in different continuous time windows, the accuracy of the fault classification results of different continuous time windows is calculated, and the time window with the highest accuracy is taken as the optimal judgment time window; The optimal judgment time window is used to adjust the trained semi-supervised support vector machine model.
8. A fault diagnosis device based on a semi-supervised support vector machine, characterized in that: include: A first signal acquisition module, a fault diagnosis module and a model training module; The first signal acquisition module is used to acquire a first vibration signal of the circuit breaker to be diagnosed; The fault diagnosis module is used to input the first vibration signal into the trained semi-supervised support vector machine model, so that the semi-supervised support vector machine model performs fault diagnosis on the circuit breaker to be diagnosed according to the first vibration signal; The model training module includes: a second signal acquisition submodule, a third signal acquisition submodule, a training data processing submodule, and an iterative training submodule; The second signal acquisition submodule is used to acquire second vibration signals of a plurality of circuit breakers with unknown working states, and perform sample overlap and time domain feature extraction on each second vibration signal to generate first unlabeled data; The third signal acquisition submodule is used to acquire third vibration signals of a plurality of circuit breakers with known working states, perform sample overlap and time domain feature extraction on each third vibration signal, and generate first label data according to the time domain feature and the corresponding working state; The training data processing submodule is used to obtain second unlabeled data and second labeled data by normalizing and recursively eliminating features of the first unlabeled data and the first labeled data; The iterative training submodule is used to input the second unlabeled data and the second labeled data into the semi-supervised support vector machine model to be trained for iterative training until all the unlabeled data are allocated, thereby obtaining a trained semi-supervised support vector machine model; wherein, during each training, the current second labeled data is used to construct the current support vector machine submodel; one of the current second unlabeled data is obtained as the data to be allocated; the probability that the data to be allocated belongs to each type of working state is calculated by the support vector machine submodel, and the highest probability among each type of probability is used as the probability to be compared; the probability to be compared is compared with a preset threshold, and if the probability to be compared is greater than the preset threshold, the data to be allocated is added to the second labeled data, otherwise, the data to be allocated is added to other fault data; and the second labeled data and the second unlabeled data are updated.
9. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the fault diagnosis method based on a semi-supervised support vector machine as described in any one of claims 1 to 7 is implemented.
10. A computer-readable storage medium, characterized in that: include: A stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the fault diagnosis method based on a semi-supervised support vector machine as described in any one of claims 1 to 7.
Citation Information
Patent Citations
Fault diagnosis method and device of low-voltage-frame circuit breaker and computing device
CN109116224A
Transformer fault type diagnosis method based on semi-supervised SVM
CN110689068A
Circuit breaker operation state detection method, medium and system
CN114676750A
Bearing fault intelligent diagnosis method and system, medium, equipment and terminal
CN115219197A
Feature selection method for intelligent fault diagnosis of high-voltage circuit breaker
CN116166952A
Cited By
Proton exchange membrane fuel cell state monitoring method, equipment and medium
CN119315064A
