Phase modifier operation data analysis fault diagnosis method, device and equipment

Abstract

The invention relates to the field of artificial intelligence, and discloses a phase modifier operation data analysis fault diagnosis method, device and equipment, and the method comprises the steps: obtaining a historical operation data sample of a phase modifier, constructing the multi-dimensional operation characteristics of the phase modifier based on a non-coding ratio method, and carrying out the dimension reduction of the multi-dimensional operation characteristics through a kernel principal component method, obtaining a first sample data set; generating a new fault feature sample based on the auxiliary confrontation generative network to balance the sample data set and obtain a second sample data set; and training a phase modifier fault diagnosis model based on a least square support vector machine by using the second sample data set, inputting to-be-diagnosed operation data of the phase modifier, extracting multi-dimensional operation features, performing kernel principal component analysis dimension reduction processing, and inputting the trained phase modifier fault diagnosis model to perform fault diagnosis. According to the method, the problems of few fault samples, unbalanced category distribution, high feature dimension and redundancy of the phase modifier are effectively solved, and the accuracy, generalization and real-time performance of fault diagnosis are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method, device and equipment for analyzing faults in phase shifter operation data. Background Art

[0002] Phase regulator is a reactive power compensation device, which is a synchronous motor that provides or absorbs reactive power to the power system. Phase regulator is a synchronous phase regulator, which is a synchronous motor in a special operating state. When applied to the power system, it can automatically increase reactive output when the grid voltage drops according to the needs of the system. The synchronous motor runs in the motor state, without mechanical load or prime mover, and only provides or absorbs reactive power to the power system. It is also called a synchronous compensator. It is used to improve the power factor of the power grid and maintain the voltage level of the power grid. Some faults of the phase regulator may have some adverse effects on the unit, such as mechanical failure, commutation failure or DC blocking. Strong reactive power fluctuations will occur in the power grid system, resulting in a large change in the grid voltage. Timely and accurate diagnosis of potential faults of the phase regulator is crucial to avoid accident risks and formulate maintenance strategies.

[0003] Traditional phase-converter fault diagnosis methods mainly rely on manual detection and empirical judgment. This method is not only inefficient, but also easily affected by human factors, resulting in low accuracy and reliability of the diagnostic results. With the expansion of the scale of power systems and the development of technology, higher requirements are placed on phase-converter fault diagnosis, requiring more accurate, fast and automated diagnostic technologies. In recent years, with the development of artificial intelligence and machine learning technologies, data-driven fault diagnosis methods have gradually become a hot topic of research. These methods use algorithm models to identify and predict potential faults by analyzing a large amount of data generated during the operation of phase-converters, such as vibration, temperature, pressure and other parameters. However, in practical applications, due to the relatively low probability of phase-converter failures, the number of fault samples that can be used to train models is limited, and the distribution of these samples in various fault types is often uneven. This problem of small number of samples and uneven distribution brings challenges to the training and optimization of fault diagnosis models, which is prone to misjudgment in model prediction and reduces the accuracy of diagnosis.

[0004] In order to solve this problem, a variety of methods have been proposed in the prior art, such as balancing sample distribution through resampling technology and improving the generalization ability of the model by using ensemble learning methods. However, the methods of the prior art still have certain limitations when dealing with unbalanced data sets. For example, resampling may cause data distortion, and ensemble learning may increase computational complexity. Therefore, how to effectively deal with the sample imbalance problem in phase-shifting fault diagnosis and improve the accuracy and efficiency of diagnosis is still a technical problem that needs to be solved urgently. Summary of the invention

[0005] In view of this, the present application provides a phase regulator operation data analysis fault diagnosis method to solve the technical problem in the prior art that the imbalance of training data samples leads to low accuracy and low efficiency in fault diagnosis of phase regulator units.

[0006] According to a first aspect of the present application, a method for analyzing faults in phase-shifting operation data is provided, comprising:

[0007] Acquire historical operation data samples of the phase regulator, wherein the operation data samples include normal state operation data and fault state operation data;

[0008] Constructing the multi-dimensional operation characteristics of the condenser based on the uncoded ratio method;

[0009] The kernel principal component analysis method is used to reduce the dimensionality of the multi-dimensional operation characteristics to obtain the first sample data set;

[0010] Based on the auxiliary adversarial generation network, new fault feature samples are generated to balance the sample data set to obtain a second sample data set;

[0011] Using the second sample data set, a phase-shifting fault diagnosis model based on the least squares support vector machine (LSSVM) is trained;

[0012] Input the condenser operation data to be diagnosed, extract the multi-dimensional operation features and perform KPCA dimensionality reduction processing, and input the trained condenser fault diagnosis model for fault diagnosis.

[0013] According to a second aspect of the present application, a phase regulator operation data analysis and fault diagnosis device is provided, comprising:

[0014] An acquisition module is used to acquire historical operation data samples of the phase regulator, wherein the operation data samples include normal state operation data and fault state operation data;

[0015] A construction module for constructing the multi-dimensional operation characteristics of the condenser based on the uncoded ratio method;

[0016] A dimensionality reduction module is used to perform dimensionality reduction processing on the multi-dimensional operation characteristics by using a kernel principal component analysis (KPCA) method to obtain a first sample data set;

[0017] A generation module, used to generate new fault feature samples based on the auxiliary adversarial generation network to balance the sample data set and obtain a second sample data set;

[0018] A training module, used for training a phase-shifting fault diagnosis model based on a least squares support vector machine (LSSVM) using a second sample data set;

[0019] The diagnosis module is used to input the condenser operation data to be diagnosed, extract multi-dimensional operation features and perform dimensionality reduction processing by kernel principal component analysis, and input the trained condenser fault diagnosis model for fault diagnosis.

[0020] According to a third aspect of the present application, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned phase regulator operation data analysis fault diagnosis method when executing the computer program.

[0021] According to a fourth aspect of the present application, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the above-mentioned phase-shifting operation data analysis fault diagnosis method are implemented.

[0022] Through the above technical scheme, the present application provides a phase shifter operation data analysis fault diagnosis method, device, equipment and medium, which obtains the historical operation data samples of the phase shifter; constructs the multi-dimensional operation characteristics of the phase shifter based on the uncoded ratio method; uses the kernel principal component analysis method to reduce the dimension of the multi-dimensional operation characteristics to obtain a first sample data set; based on the auxiliary adversarial generation network, generates new fault feature samples to balance the sample data set to obtain a second sample data set; uses the second sample data set to train a phase shifter fault diagnosis model based on a least squares support vector machine (LSSVM), inputs the phase shifter operation data to be diagnosed, extracts the multi-dimensional operation characteristics and performs kernel principal component analysis. The dimensionality reduction processing is performed by sub-analysis, and the trained LSSVM phase-shifting fault diagnosis model is input for fault diagnosis. The present invention introduces an auxiliary adversarial generative network to balance the expansion of the original fault sample data, and constructs a non-coding ratio feature that integrates multi-source information, and then combines the kernel principal component analysis to perform feature dimensionality reduction extraction. Finally, the Banshee Kingfisher optimization algorithm is used to optimize the LSSVM diagnosis model parameters, forming a fault diagnosis strategy that integrates sample enhancement, feature optimization, intelligent modeling and optimization. The present invention effectively solves the problems of few phase-shifting fault samples, uneven category distribution, high and redundant feature dimensions, and improves the accuracy, generalization and real-time performance of fault diagnosis.

[0023] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application and to implement it in accordance with the contents of the specification, and to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0025] Figure 1 A schematic diagram of an application scenario of a phase regulator operation data analysis fault diagnosis method provided in an embodiment of the present application is shown;

[0026] Figure 2 A schematic diagram of a process flow of a phase regulator operation data analysis fault diagnosis method provided in the first embodiment of the present application is shown;

[0027] Figure 3 A schematic diagram showing the process of training a phase-shifting fault diagnosis model using the Kingfisher PKO algorithm to optimize the phase-shifting fault diagnosis model provided in the second embodiment of the present application is shown;

[0028] Figure 4 A schematic diagram of data generation of the base-assisted adversarial network ACGAN provided in the fourth embodiment of the present application is shown;

[0029] Figure 5 A schematic diagram of the structure of a phase regulator operation data analysis and fault diagnosis device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0030] The specific implementation of the present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0031] A method for analyzing faults in phase-shifting operation data provided by an embodiment of the present invention can be applied to Figure 1 As a reactive power compensation device, the synchronous condenser is essentially a synchronous motor running at no load. It does not provide active power during normal operation. It adjusts its reactive power output by changing the magnitude of the excitation current to provide or absorb reactive power to the system. Since it has no prime mover and no mechanical input, an external starting device is required. The structure of the synchronous phase modulation system is as follows: Figure 1As shown in the figure, the monitoring system monitors and controls the excitation, sequential control, starting system, lubricating oil system air cooling system, protection and automatic devices, station area auxiliary systems and electrical equipment of the phase condenser. Usually, an independent distributed control system (DCS) can be set up, and the DCS centrally controls multiple phase condensers and auxiliary systems. In the application scenario in the figure, the synchronous phase condenser is a synchronous motor without mechanical load, which is composed of a static frequency converter (SFC) system, a starting switching system, an excitation and motor system, a cooling system, and a signal transmission system. The phase condenser is used as a reactive power backup device. When reactive power cannot meet the commutation demand of the converter station, the bus voltage of the converter station drops. When the voltage drops to the minimum commutation voltage of the converter station, voltage collapse and active power transmission failure will occur. At this time, if there is a large-capacity phase-shifting device as a reactive power backup device, the phase-shifting device can be started through SFC to provide the required reactive power according to system requirements, which can stabilize the bus voltage and active power of the converter station. The control scope of the phase-shifting monitoring system mainly includes: phase-shifting body temperature, stator cooling water, rotor cooling water, lubricating oil, top shaft oil, open cooling water, closed cooling water (rotor cooling water) dosing, and phase-shifting start-up, excitation and other electrical systems. The control scope of the auxiliary system mainly includes: circulating water pump, mechanical ventilation cooling tower, industrial service water, chemical water treatment, circulating water dosing and corresponding electrical systems. The phase-shifting station does not have a separate management information system (MIS). The MIS is considered uniformly by the converter station. The DCS of the phase-shifting station reserves a communication interface with the MIS of the converter station.

[0032] The operating state of a synchronous condenser is complex, and the collected signals usually contain multiple natural oscillation modes and information unrelated to the fault type, including multiple coupling effects of multiple moving parts, operating background noise, etc., such as stator radial vibration characteristics, rotor radial vibration characteristics, stator circulating current characteristics, winding temperature characteristics, vibration and excitation relationship, and the process parameters of the condenser include the load, excitation current, speed, and shaft vibration of the condenser. The operating state quantities of the synchronous condenser are selected: online operating electrical parameters, vibration parameters, temperature parameters, and partial discharge parameters. The operating state information of the condenser is collected through the Internet of Things and stored in a database, and a condenser operating data analysis fault diagnosis method provided in an embodiment of the present invention is performed on the historical data, by obtaining the historical operating data samples of the condenser; constructing the multidimensional operating characteristics of the condenser based on the uncoded ratio method; using the kernel principal component analysis method to reduce the dimensionality of the multidimensional operating characteristics to obtain a first sample data set; based on the auxiliary adversarial generation network, a new fault feature sample is generated to balance the sample data set to obtain a second sample data set; using the second sample data set, a least squares support vector machine (LSSVM) is trained. The present invention proposes a phase shifter fault diagnosis model based on LSSVM (Low-Speed ​​Vector Machine), inputs the phase shifter operating data to be diagnosed, extracts multi-dimensional operating features and performs kernel principal component analysis dimensionality reduction processing, and inputs the trained LSSVM phase shifter fault diagnosis model for fault diagnosis. The present invention introduces an auxiliary generative adversarial network to balance the expansion of the original fault sample data, and constructs a non-coding ratio feature that integrates multi-source information, and then combines the kernel principal component analysis to perform feature dimensionality reduction extraction. Finally, the Kingfisher optimization algorithm is used to optimize the phase shifter fault diagnosis model parameters, forming a set of fault diagnosis strategies integrating sample enhancement, feature optimization, intelligent modeling and optimization. The present invention effectively solves the problems of few phase shifter fault samples, uneven category distribution, high and redundant feature dimensions, and improves the accuracy, generalization and real-time performance of fault diagnosis.

[0033] Embodiment 1:

[0034] like Figure 2 As shown, a phase regulator operation data analysis fault diagnosis method provided in an embodiment of the present invention includes:

[0035] Step 201, obtaining a sample of historical operation data of a phase regulator;

[0036] The operation data samples include normal state operation data and fault state operation data;

[0037] Step 202, constructing the multi-dimensional operation characteristics of the phase regulator based on the non-coding ratio method;

[0038] Step 203: Use a kernel principal component analysis method to perform dimensionality reduction processing on the multi-dimensional operation features to obtain a first sample data set;

[0039] Step 204: Generate new fault feature samples based on the auxiliary adversarial generation network to balance the sample data set and obtain a second sample data set;

[0040] Step 205: using the second sample data set, training a phase-shifting fault diagnosis model based on a least squares support vector machine (LSSVM);

[0041] Step 206: Input the condenser operation data to be diagnosed, extract the multi-dimensional operation features and perform KPCA dimensionality reduction processing, and input the trained condenser fault diagnosis model to perform fault diagnosis.

[0042] A method for fault diagnosis of a condenser operation data analysis provided in the first embodiment of the present invention obtains a sample of the condenser operation data; constructs a multi-dimensional operation feature of the condenser based on the uncoded ratio method; performs dimensionality reduction processing on the multi-dimensional operation feature using the kernel principal component analysis method to obtain a first sample data set; generates a new fault feature sample based on an auxiliary adversarial generative network to balance the sample data set to obtain a second sample data set; and uses the second sample data set to train a least squares support vector machine (LSSVM) based on the least squares support vector machine (LSSVM). A phase tunner fault diagnosis model based on LSSVM (Long-Term Steering Frame Machine, LSSVM) is proposed. The phase tunner operation data to be diagnosed is input, multi-dimensional operation features are extracted and dimensionality reduction processing is performed through kernel principal component analysis. The phase tunner fault diagnosis model of the trained LSSVM is input for fault diagnosis. The present invention introduces an auxiliary generative adversarial network to balance the expansion of the original fault sample data, and constructs a non-coding ratio feature that integrates multi-source information. The feature dimensionality reduction and extraction is then performed in combination with kernel principal component analysis. Finally, the Kingfisher optimization algorithm is used to optimize the phase tunner fault diagnosis model parameters, forming a set of fault diagnosis strategies integrating sample enhancement, feature optimization, intelligent modeling and optimization. The present invention effectively solves the problems of few phase tunner fault samples, uneven category distribution, high and redundant feature dimensions, and improves the accuracy, generalization and real-time performance of fault diagnosis.

[0043] Embodiment 2:

[0044] In order to improve the training effect of the LSSVM phase-shifting fault diagnosis model, Figure 3 As shown, step 205 specifically includes:

[0045] Step 301, setting the initial parameters of LSSVM;

[0046] Among them, the initial parameters include the selection of kernel function, initialization weight and bias term;

[0047] Step 302, randomly generate a group of kingfishers, and encode the position of each kingfisher as a parameter vector of LSSVM;

[0048] Step 303: training an LSSVM model based on the parameter vector of each kingfisher representative;

[0049] Step 304: According to the fitness function, update and select the best kingfisher as the leader bird;

[0050] Among them, the fitness function is obtained based on classification accuracy, mean square error or other relevant indicators. The Pied Kingfisher Optimizer (PKO) introduced in this step is a new type of meta-heuristic intelligent optimization algorithm. The PKO algorithm is built around three different stages: perching / hovering predation (exploration / diversification), diving predation (development / enhancement) and promoting symbiotic relationships. These behavioral characteristics are converted into mathematical models that can effectively solve a wide range of optimization challenges in different search spaces. The algorithm achieves optimized search by simulating the behavior of the kingfisher when fishing. Compared with other optimization algorithms in the prior art, the PKO optimization process only involves simple vector operations, has low computational complexity, is easy to program, and does not introduce complex control parameters, making parameter debugging simpler and more convenient.

[0051] Step 305: update the positions of all kingfishers according to the rules of the kingfisher PKO algorithm, simulate the behavior of the kingfisher when preying, and update the search position by tracking and selecting the optimal solution;

[0052] Repeat steps 303 to 305 until the maximum number of iterations is met or the change in the fitness function is less than a preset threshold;

[0053] The parameter combination represented by the Banyan Kingfisher with the best fitness value was selected as the optimal parameters of LSSVM, and the LSSVM model was retrained based on the optimal parameters.

[0054] In the second embodiment of the present invention, the least squares support vector machine (LSSVM) is used to establish a phase-shifting fault diagnosis model, and the Kingfisher Optimization Algorithm (PKO) is innovatively introduced to intelligently optimize the model parameters, which significantly improves the accuracy and efficiency of fault diagnosis. Compared with the classic SVM, LSSVM converts inequality constraints into equality constraints, avoids quadratic programming problems, reduces computational complexity, and ensures high classification accuracy, which is more suitable for engineering practice. At the same time, in view of the problem that LSSVM is sensitive to parameters, the present invention innovatively proposes a PKO optimization strategy. PKO simulates the fishing behavior of Kingfisher, introduces mechanisms such as adaptive weights, multi-domain search, and sinusoidal perturbations, and can quickly search for the optimal solution in the parameter space, effectively overcoming the defect that traditional optimization algorithms are prone to fall into local optimality. In the fault diagnosis of a phase-shifting unit, the classification accuracy of the LSSVM model optimized by PKO on the test set is as high as 98.7%, which is 5.2% higher than that of the unoptimized model, and the convergence speed is increased by 1 times, which maximizes the performance potential of LSSVM. Therefore, the LSSVM diagnostic model based on PKO optimization has the advantages of high accuracy, fast speed, and strong generalization. It provides a powerful tool for the rapid diagnosis and accurate positioning of phase-shifting machine faults, promotes the transformation from condition-based maintenance to intelligent predictive maintenance mode, and has broad engineering application prospects.

[0055] Embodiment three:

[0056] In order to improve the effectiveness of feature engineering and efficiently extract the operating features of the phase regulator, step 202 specifically includes:

[0057] Step 202-1, selecting characteristic parameters that have a significant impact on fault diagnosis;

[0058] Among them, the characteristic parameters include vibration peak value, maximum temperature, and average oil pressure;

[0059] Step 202-2, calculating the ratio between the selected characteristic parameters;

[0060] Step 202-3, combining all calculated ratios into a multi-dimensional operation feature;

[0061] Step 202-4: normalize the synthesized multi-dimensional operation features.

[0062] In the third embodiment of the present invention, the coding-free ratio method is used to construct the multi-dimensional operation characteristics of the phase regulator, which comprehensively considers multi-source monitoring quantities such as temperature, vibration, and current, and mines the early signs of different types of faults by calculating the ratio indicators between different physical quantities. Compared with the traditional three-ratio method, Rogers ratio method, etc., this method does not rely on expert experience to set the threshold value, avoids information loss in the encoding process, can comprehensively characterize the health status of the phase regulator, sensitively capture fault characteristics, and reduce the misdiagnosis rate. In addition, the method of the third embodiment of the present invention makes full use of the massive historical operation data of the phase regulator and extracts global indicators that reflect the internal mechanism.

[0063] Embodiment 4:

[0064] In order to reduce the dimension of the data sample, step 203 specifically includes:

[0065] Step 203-1, calculating the covariance matrix of the high-dimensional feature matrix;

[0066] Step 203-2, calculating the eigenvalues ​​and eigenvectors of the covariance matrix based on the orthogonal triangular QR algorithm;

[0067] Step 203-3, the eigenvalue represents the variation in each direction, and the eigenvector represents the direction itself;

[0068] Step 203-4: define a kernel function to map the multi-dimensional operation features to a high-dimensional space, wherein the kernel function includes a Gaussian kernel, a polynomial kernel, and a linear kernel;

[0069] Step 203-5, calculating the cumulative contribution rate of the eigenvalues, and extracting the principal components whose cumulative contribution rate is greater than a preset threshold as the eigenvectors;

[0070] Step 203-6: Perform a projection transformation on the multi-dimensional operation feature according to the feature vector and the kernel matrix to reduce the dimension of the multi-dimensional operation feature.

[0071] In the fourth embodiment of the present invention, the kernel principal component analysis (KPCA) method is used to reduce the dimension of the multi-dimensional operation characteristics of the phase regulator, which effectively solves the limitation that the traditional principal component analysis (PCA) is difficult to characterize the nonlinear data characteristics. By introducing the kernel function, KPCA maps the original data to a high-dimensional feature space and extracts the principal components in the feature space, thereby revealing the inherent nonlinear structure of the data and realizing the accurate characterization of the phase regulator state under complex working conditions. This method not only eliminates the dimensional influence between different physical quantities, but also retains the characteristic information of the original data to the maximum extent, and significantly reduces the data dimension, reducing the training burden of subsequent models. In the phase regulator fault diagnosis of a certain unit, the 180-dimensional time-frequency domain features are reduced to 5 dimensions by KPCA, and the cumulative contribution rate reaches 92.5%, which greatly compresses the size of the feature set. The SVM diagnostic model is trained by the reduced-dimensional features, and its fault classification accuracy is improved by 12% compared with the direct use of the original features, and the model training time is shortened by 70%, reflecting a significant dimensionality reduction gain. Therefore, the fourth embodiment of the present invention realizes the dimensionality reduction optimization of the diagnosis model on the basis of fully exploring the state characteristics of the phase regulator, and provides a high-quality sample set for subsequent accurate fault identification.

[0072] Embodiment five:

[0073] In order to generate balanced fault samples and solve the problem of insufficient fault samples, before step 204, the following steps are included:

[0074] Construct an auxiliary adversarial generation network, the auxiliary adversarial generation network is as follows Figure 4 As shown, the auxiliary adversarial generation network includes a generator, a discriminator, an auxiliary classifier and a category information module. The generator contains a fully connected layer or a convolutional layer, which is used to generate virtual data samples from random noise vectors; the discriminator is used to distinguish the generated virtual data samples from the real data samples, the auxiliary classifier is used to determine the category information of the virtual data samples, the category information module is used to obtain the category information of the real data samples, and the discriminator and the auxiliary classifier are used to receive the virtual data samples and category information generated by the generator.

[0075] As important modules, the generator, discriminator and auxiliary classifier need to undergo corresponding model training. The steps are as follows:

[0076] (1) The training steps of the generator in the auxiliary adversarial generative network include:

[0077] Step 1.1, randomly initialize the network parameters of the generator;

[0078] Step 1.2, the generator receives the random noise vector and the category label output by the category information module as input, and generates a virtual data sample;

[0079] Step 1.3: Input the virtual data sample into the discriminator and the auxiliary classifier, and output the probability that the virtual data sample is misclassified as real data and the category prediction of the virtual data sample;

[0080] Step 1.4: Calculate the adversarial loss of the generator based on the probability that the virtual data sample is misclassified as real data, the category prediction of the virtual data sample, and the generator loss function;

[0081] Among them, the adversarial loss of the generator is min(G)=Ε x～D(x) [logD(x)]+E z～G(z) [log(1-D(G(z)))],Ε x～D(x) [logD(x)] is the expectation of the real data, which encourages the discriminator to correctly identify the real sample x.

[0082] E z～G(z) [log(1-D(G(z)))] is the expectation of generated data, encouraging the generator to produce samples z that are good enough to fool the discriminator.

[0083] Step 1.5: Update the network parameters of the iterative generator according to the adversarial loss of the generator;

[0084] Step 1.6: When it is determined that the adversarial loss of the generator is less than the preset adversarial loss threshold, the network parameters of the generator are output.

[0085] (2) The training steps of the discriminator and auxiliary classifier in the auxiliary adversarial generation network include:

[0086] Step 2.1, randomly initialize the network parameters of the discriminator and auxiliary classifier;

[0087] Step 2.2, the discriminator and the auxiliary classifier receive the virtual data samples generated by the generator, the category information output by the category information module, and the real data;

[0088] Step 2.3: Output the probability that the virtual data sample is misclassified as real data and the category prediction of the virtual data sample;

[0089] Step 2.4: Calculate the source classification loss corresponding to the discriminator based on the probability that the virtual data sample is misclassified as real data and the discriminator loss function;

[0090] Among them, the source classification loss function is L s =Ε x～x [logP(S=real|x)]+E z～G(z)[logP(S=fake|z)], where P(S=real|x) is the probability that the discriminator predicts that the sample is real when given a real sample x, and P(S=fake|z) is the probability that the discriminator predicts that the sample is fake when given a generated sample z.

[0091] Step 2.5, based on the category prediction of the virtual data sample and the auxiliary classifier loss function, calculate the auxiliary classification loss corresponding to the auxiliary classifier;

[0092] Among them, the auxiliary classification loss function is

[0093] L C =Ε x～x [logP(C=c|x)]+E z～G(z,c) [logP(C=c|G(z,c))], where P(C=c|x) is the probability that the discriminator predicts the correct category c of a sample given a real sample x, and P(C=c|G(z,c)) is the probability that the discriminator predicts the correct category c of a sample given a generated sample G(z,c) and category c.

[0094] Step 2.6: Update the network parameters of the iterative discriminator according to the source classification loss;

[0095] Step 2.7, update the network parameters of the iterative auxiliary classifier according to the auxiliary classification loss;

[0096] Finally, when the source classification loss of the network parameters of the discriminator is less than the preset source classification loss threshold, the network parameters of the discriminator are output; when the auxiliary classification loss of the network parameters of the auxiliary classifier is less than the preset auxiliary classification loss threshold, the network parameters of the auxiliary classifier are output.

[0097] Further, as Figure 2 In the specific implementation of the method, an apparatus for analyzing faults in phase shifting machine operation data is provided in an embodiment of the present invention, such as Figure 5 As shown, the device comprises:

[0098] An acquisition module 510 is used to acquire historical operation data samples of the phase regulator, wherein the operation data samples include normal state operation data and fault state operation data;

[0099] A construction module 520 is used to construct a multi-dimensional operation feature of the phase regulator based on the non-coding ratio method;

[0100] A dimension reduction module 530 is used to perform dimension reduction processing on the multi-dimensional operation features by using a kernel principal component analysis (KPCA) method to obtain a first sample data set;

[0101] A generating module 540 is used to generate new fault feature samples based on the auxiliary adversarial generation network to balance the sample data set and obtain a second sample data set;

[0102] A training module 550 is used to train a phase-shifting fault diagnosis model based on a least squares support vector machine (LSSVM) using a second sample data set;

[0103] The diagnosis module 560 is used to input the condenser operation data to be diagnosed, extract multi-dimensional operation features and perform kernel principal component analysis dimensionality reduction processing, and input the trained condenser fault diagnosis model to perform fault diagnosis.

[0104] In an embodiment of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of implementing a method for diagnosing a phase-shifting operation data fault analysis include:

[0105] Acquire historical operation data samples of the phase regulator, wherein the operation data samples include normal state operation data and fault state operation data;

[0106] Constructing the multi-dimensional operation characteristics of the condenser based on the uncoded ratio method;

[0107] The kernel principal component analysis method is used to reduce the dimensionality of the multi-dimensional operation characteristics to obtain the first sample data set;

[0108] Based on the auxiliary adversarial generation network, new fault feature samples are generated to balance the sample data set to obtain a second sample data set;

[0109] Using the second sample data set, a phase-shifting fault diagnosis model based on the least squares support vector machine (LSSVM) is trained;

[0110] Input the condenser operation data to be diagnosed, extract the multi-dimensional operation features and perform KPCA dimensionality reduction processing, and input the trained condenser fault diagnosis model for fault diagnosis.

[0111] In an embodiment of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0112] Acquire historical operation data samples of the phase regulator, wherein the operation data samples include normal state operation data and fault state operation data;

[0113] Constructing the multi-dimensional operation characteristics of the condenser based on the uncoded ratio method;

[0114] The kernel principal component analysis method is used to reduce the dimensionality of the multi-dimensional operation characteristics to obtain the first sample data set;

[0115] Based on the auxiliary adversarial generation network, new fault feature samples are generated to balance the sample data set to obtain a second sample data set;

[0116] Using the second sample data set, a phase-shifting fault diagnosis model based on the least squares support vector machine (LSSVM) is trained;

[0117] Input the condenser operation data to be diagnosed, extract the multi-dimensional operation features and perform KPCA dimensionality reduction processing, and input the trained condenser fault diagnosis model for fault diagnosis.

[0118] It should be noted that, in the above embodiments, only the common operating conditions of the phase-shifting unit are used as examples to illustrate the principles and implementation steps of the embodiments of the present invention, and no specific limitation is made to the actual application scenarios. For example, different operating environments of the phase-shifting unit, different equipment types, different climatic conditions, etc. can all be used to distinguish different operating conditions. The technical solution of the present invention can also be applied to extended scenarios of the phase-shifting unit. Regarding the functions or steps that can be implemented by the computer-readable storage medium or the computer device, please refer to the aforementioned method embodiments. In order to avoid repetition, they will not be described one by one here.

[0119] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0120] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0121] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for analyzing faults in phase condenser operation data, characterized in that: include: Acquire historical operation data samples of the phase regulator, wherein the operation data samples include normal state operation data and fault state operation data; Constructing the multi-dimensional operation characteristics of the condenser based on the uncoded ratio method; The kernel principal component analysis method is used to reduce the dimensionality of the multi-dimensional operation characteristics to obtain the first sample data set; Based on the auxiliary adversarial generative network, new fault feature samples are generated to balance the sample data set to obtain a second sample data set; Using the second sample data set, training a phase regulator fault diagnosis model based on a least squares support vector machine (LSSVM); Input the condenser operation data to be diagnosed, extract the multi-dimensional operation features and perform KPCA dimensionality reduction processing, and input the trained condenser fault diagnosis model for fault diagnosis.

2. The method for analyzing faults in phase condenser operation data according to claim 1, characterized in that: The step of using the second sample data set to train a phase-shifting fault diagnosis model based on a least squares support vector machine (LSSVM) model comprises: Step 1: Setting the initial parameters of LSSVM, wherein the initial parameters include selecting a kernel function, initializing weights and bias items; Step 2: Randomly generate a group of kingfishers, and encode the position of each kingfisher as a parameter vector of LSSVM; Step 3: Train the LSSVM model based on the parameter vector of each kingfisher representative; Step 4: According to the fitness function, update and select the best kingfisher as the leader bird, wherein the fitness function is obtained based on classification accuracy, mean square error or other relevant indicators; Step 5: Update the positions of all kingfishers according to the rules of the Kingfisher PKO algorithm, simulate the behavior of the kingfisher when preying, and update the search position by tracking and selecting the optimal solution; Repeat steps 3 to 5 until the maximum number of iterations is met or the change in the fitness function is less than a preset threshold; The parameter combination represented by the Banyan Kingfisher with the best fitness value was selected as the optimal parameters of LSSVM, and the LSSVM model was retrained based on the optimal parameters.

3. The method for analyzing faults in phase condenser operation data according to claim 1, characterized in that: The step of constructing the multi-dimensional operation characteristics of the phase regulator based on the non-coding ratio method includes: Selecting characteristic parameters that have a significant impact on fault diagnosis, wherein the characteristic parameters include vibration peak value, maximum temperature value, and average oil pressure value; Calculate the ratio between the selected feature parameters; All calculated ratios are combined into a multidimensional operating signature; The synthetic multi-dimensional operation characteristics are normalized.

4. The method for analyzing faults in phase condenser operation data according to claim 1, characterized in that: The step of using the kernel principal component analysis (KPCA) method to perform dimensionality reduction processing on the multidimensional operation features to obtain the first sample data set includes: Calculate the covariance matrix of a high-dimensional feature matrix; Based on the orthogonal triangular QR algorithm, the eigenvalues ​​and eigenvectors of the covariance matrix are calculated; The eigenvalues ​​represent the amount of variation in each direction, and the eigenvectors represent the directions themselves; Define a kernel function to map the multi-dimensional operation characteristics to a high-dimensional space, wherein the kernel function includes a Gaussian kernel, a polynomial kernel, and a linear kernel; Calculate the cumulative contribution rate of the eigenvalues, and extract the principal components whose cumulative contribution rate is greater than a preset threshold as eigenvectors; According to the feature vector and the kernel matrix, a projection transformation is performed on the multi-dimensional operation feature to reduce the dimension of the multi-dimensional operation feature.

5. The method for analyzing faults in phase condenser operation data according to claim 1, characterized in that: Before the step of generating new fault feature samples based on the auxiliary adversarial generation network to balance the sample data set and obtain the second sample data set, the method includes: An auxiliary adversarial generation network is constructed, wherein the auxiliary adversarial generation network includes a generator, a discriminator, an auxiliary classifier and a category information module, the generator includes a fully connected layer or a convolutional layer, and is used to generate virtual data samples from a random noise vector; the discriminator is used to distinguish the generated virtual data samples from the real data samples, the auxiliary classifier is used to determine the category information of the virtual data samples, the category information module is used to obtain the category information of the real data samples, and the discriminator and the auxiliary classifier are used to receive the virtual data samples and category information generated by the generator.

6. The method for analyzing faults in phase condenser operation data according to claim 5, characterized in that: The training step of the generator in the auxiliary adversarial generation network includes: Randomly initialize the network parameters of the generator; The generator receives a random noise vector and a category label output by the category information module as input, and generates a virtual data sample; Input the virtual data sample into the discriminator and the auxiliary classifier, and output the probability that the virtual data sample is misclassified as real data and the category prediction of the virtual data sample; Calculating the adversarial loss of the generator according to the probability that the virtual data sample is misjudged as real data, the category prediction of the virtual data sample, and the generator loss function; According to the adversarial loss of the generator, updating and iterating the network parameters of the generator; When it is determined that the adversarial loss of the generator is less than a preset adversarial loss threshold, the network parameters of the generator are output.

7. The method for analyzing faults in phase condenser operation data according to claim 6, characterized in that: The training steps of the discriminator and the auxiliary classifier in the auxiliary adversarial generation network include: Randomly initialize the network parameters of the discriminator and the auxiliary classifier; The discriminator and the auxiliary classifier receive the virtual data samples generated by the generator, the category information output by the category information module, and the real data; Outputting the probability that the virtual data sample is misjudged as real data and the category prediction of the virtual data sample; Calculating the source classification loss corresponding to the discriminator according to the probability that the virtual data sample is misjudged as real data and the discriminator loss function; Calculating the auxiliary classification loss corresponding to the auxiliary classifier according to the category prediction of the virtual data sample and the auxiliary classifier loss function; According to the source classification loss, updating and iterating the network parameters of the discriminator; According to the auxiliary classification loss, updating and iterating the network parameters of the auxiliary classifier; When it is determined that the source classification loss of the network parameters of the discriminator is less than a preset source classification loss threshold, the network parameters of the discriminator are output; when it is determined that the auxiliary classification loss of the network parameters of the auxiliary classifier is less than a preset auxiliary classification loss threshold, the network parameters of the auxiliary classifier are output.

8. A phase regulator operation data analysis and fault diagnosis device, characterized in that: include: An acquisition module, used for acquiring historical operation data samples of the phase regulator, wherein the operation data samples include normal state operation data and fault state operation data; A construction module for constructing the multi-dimensional operation characteristics of the condenser based on the uncoded ratio method; A dimensionality reduction module is used to perform dimensionality reduction processing on the multi-dimensional operation characteristics by using a kernel principal component analysis (KPCA) method to obtain a first sample data set; A generation module, used to generate new fault feature samples based on the auxiliary adversarial generation network to balance the sample data set and obtain a second sample data set; A training module, used for training a phase-shifting fault diagnosis model based on a least squares support vector machine (LSSVM) using the second sample data set; The diagnosis module is used to input the condenser operation data to be diagnosed, extract multi-dimensional operation features and perform dimensionality reduction processing by kernel principal component analysis, and input the trained condenser fault diagnosis model for fault diagnosis.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the phase regulator operation data analysis fault diagnosis method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the phase regulator operation data analysis fault diagnosis method as described in any one of claims 1 to 7 are implemented.

Images