A motor corona detection method and system based on a particle swarm-convolutional neural network, an electronic device, and a non-transitory computer-readable storage medium

Abstract

The application discloses a motor corona detection method and system based on a particle swarm-convolutional neural network, electronic equipment and a non-transitory computer readable storage medium, and belongs to the technical field of motor state monitoring and fault diagnosis. The method comprises the following steps: acquiring multi-dimensional data of motor corona discharge and operation state thereof; pre-processing the multi-dimensional data and dividing the multi-dimensional data into a training set and a verification set; inputting the training set into a particle swarm-convolutional neural network and training the training set, wherein a particle swarm algorithm takes a learning rate and a random rejection rate as search dimensions, the iteration update of particle speed and position is guided by a verification set performance function, and the best parameter combination is obtained through optimization; and inputting the verification set into a trained judgment model to obtain a judgment result of whether the target motor has corona discharge. The application globally optimizes key parameters of the convolutional neural network through the particle swarm optimization algorithm, overcomes the defect that the traditional gradient descent method is prone to local optimization, improves the generalization ability and detection precision of the model in a strong noise background, and realizes early and accurate early warning of corona discharge.

Description

Technical Field

[0001] This invention belongs to the field of motor condition monitoring and fault diagnosis technology, and particularly relates to a method, system, electronic device and non-transitory computer-readable storage medium for motor corona detection based on particle swarm optimization-convolutional neural network. Background Technology

[0002] High-voltage motors, as core power equipment, are widely used in energy, manufacturing, transportation, and other fields. During long-term operation, the internal insulation materials of the motor gradually deteriorate due to factors such as electrothermal aging and mechanical vibration, leading to partial discharge phenomena. Corona discharge is the earliest and most insidious precursor to insulation faults. If early corona discharge is not detected in a timely and effective manner, continuous discharge activity will cause irreversible damage to the insulation system, ultimately leading to downtime or even safety accidents, resulting in huge economic losses.

[0003] When a motor experiences corona discharge, it generates a high-frequency pulse current signal with an extremely short duration. This signal has a very rich spectral composition, typically ranging from several megahertz to hundreds of megahertz. To capture these weak signals, engineering methods primarily employ high-frequency current detection, capacitive coupling, and ultra-high-frequency electromagnetic wave detection. However, in the complex electromagnetic environments of actual industrial sites, these detection methods face significant technical challenges:

[0004] First, strong electromagnetic interference is prevalent in motor systems. Broadband noise generated by power electronic equipment such as frequency converters, switching power supplies, and relays can couple into the detection system through the grounding loop, making it difficult for traditional filtering methods to effectively separate useful signals from background noise, which may lead to missed detections of corona discharge.

[0005] Second, pattern recognition of discharge signals is highly dependent on the quality of feature extraction. Traditional detection methods typically require manual design of statistical features based on parameters such as pulse amplitude, phase distribution, and skewness. This method is not only time-consuming and labor-intensive, but also struggles to fully characterize the nonlinear and time-varying characteristics of discharge signals. When the motor's operating conditions change, the preset feature set often becomes inapplicable.

[0006] Third, and more importantly, corona discharge signals are often submerged in strong background noise, and relying solely on hardware filtering or simple signal processing algorithms can no longer meet the requirements for high-precision detection.

[0007] Accurately detecting the characteristics of corona discharge against a background of strong noise and achieving intelligent identification of discharge types has become a key issue restricting the development of motor condition monitoring technology. Therefore, it is urgent to introduce advanced artificial intelligence algorithms and build intelligent detection models with strong adaptive learning capabilities to overcome the limitations of traditional methods in signal processing and pattern recognition.

[0008] In recent years, Convolutional Neural Networks (CNNs) have demonstrated powerful automatic feature extraction capabilities in fields such as image recognition and speech processing. However, applying CNNs to motor corona detection still faces the following technical challenges:

[0009] (1) Difficulty in parameter optimization: Hyperparameters such as learning rate and random discard rate of CNN have a significant impact on model performance. Traditional gradient descent method is prone to getting trapped in local optima, and parameter selection is highly dependent on experience.

[0010] (2) Insufficient generalization ability: Under strong noise background, the CNN model with fixed parameters is difficult to adapt to the changes in different motor operating conditions and is prone to overfitting.

[0011] (3) Low training efficiency: The manual parameter adjustment process is time-consuming and cannot meet the needs of real-time monitoring in industrial sites.

[0012] Therefore, it is necessary to propose a corona detection method that deeply integrates Particle Swarm Optimization (PSO) with convolutional neural networks. By automatically determining the optimal parameter combination of the CNN through a global optimization mechanism, the detection accuracy, anti-interference ability and generalization performance of the model can be improved. Summary of the Invention

[0013] The purpose of this invention is to provide a method, system, electronic device, and non-transitory computer-readable storage medium for detecting motor corona discharge based on particle swarm optimization-convolutional neural networks, in order to solve the technical problems in the prior art, such as difficulty in signal extraction under strong electromagnetic interference environment, reliance on manual design for feature extraction, and easy getting trapped in local optima in model parameter optimization.

[0014] To achieve the above objectives, the present invention adopts the following technical solution:

[0015] On the one hand, a method for detecting motor corona based on particle swarm optimization-convolutional neural networks is characterized by the following steps:

[0016] Step S1: Obtain multidimensional data on motor corona discharge and its operating status;

[0017] Step S2: Preprocess the multidimensional data and construct a dataset, dividing the dataset into a training set and a validation set;

[0018] Step S3: Particle Swarm Optimization-Convolutional Neural Network Model Construction and Training

[0019] Step S3.1: Particle swarm initialization;

[0020] The adjustable parameters of the convolutional neural network, including the learning rate. Random discard rate Defined as the search dimension of the particle swarm, each particle corresponds to a parameter configuration scheme. Initialize the particle swarm, randomly generate the initial position and initial velocity of each particle, set the maximum number of iterations, inertia weight, individual learning factor, and global learning factor, and initialize the individual optimal position of each particle to its own position, and initialize the global optimal position to the random particle position.

[0021] Step S3.2: Convolutional Neural Network Construction

[0022] For each particle, a convolutional neural network model is constructed using the parameters corresponding to its current position; the convolutional neural network structure includes an input layer, at least one convolutional layer, at least one pooling layer, a fully connected layer, and a regression layer;

[0023] Step S3.3: Model Training and Performance Evaluation

[0024] The convolutional neural network is trained using the training set, and a performance function is computed on the validation set; the performance function is defined as the reciprocal of the error between the predicted value and the true value on the validation set.

[0025] Step S3.4: Particle Swarm Iterative Update

[0026] Update the individual optimal value and the global optimal value of the particles according to the performance function, and update the velocity and position of each particle;

[0027] Step S3.5: Iteration Termination and Optimal Parameter Output

[0028] Verify whether the maximum number of iterations has been reached. If not, return to step S3.2 and iterate again. If the maximum number of iterations has been reached, output the optimal parameter combination.

[0029] Step S3.6: Model Performance Evaluation

[0030] The predictive performance evaluation index of the model is calculated using the true values ​​in the training set and the predicted values ​​output by the prediction model.

[0031] Step S4: Corona discharge assessment

[0032] The data to be detected is input into the judgment model constructed by the optimal parameter combination to obtain the judgment result of whether the target motor has corona discharge.

[0033] Preferably, in step S1, the multidimensional data includes: pulse current time-domain waveform data and spectral component data detected from the motor grounding circuit during normal operation and when corona discharge occurs; waveform data of high-frequency voltage signals detected from the winding ends; operating voltage, load current, speed and power factor data during normal operation and when corona discharge occurs; and motor operation duration data.

[0034] Preferably, in step S1, the multidimensional data includes data obtained by detecting the same motor running at different time points within a preset time period; or data obtained by detecting the same motor running in different environments at the same time; each detection forms a multidimensional data matrix.

[0035] Preferably, in step S2, the preprocessing includes:

[0036] The pulse current time-domain waveform data, spectral component data, and high-frequency voltage signal waveform data are preprocessed to extract the sine wave amplitude of the pulse current, the amplitude and number of high-frequency current pulses superimposed on the sine wave, and the amplitude and number of high-frequency voltage signals. The extracted feature data, along with 10 data points including the operating voltage, load current, speed, power factor, and motor running time, are used as feature parameters. A linear transformation is performed using a standardization function to transform the size of each element to the [0,1] interval.

[0037] Preferably, in step S2, constructing the dataset includes:

[0038] The preprocessed data is divided into two parts: the first part comes from motors that are operating normally and have not experienced corona discharge, and the corresponding data matrix is ​​marked as 0; the second part comes from motors that have experienced corona discharge, and the corresponding data matrix is ​​marked as 1; each sample is an 11×1 data matrix; the data is divided into training set and validation set according to a preset ratio.

[0039] Preferably, in step S3.2, the convolutional neural network structure comprises seven interconnected layers: an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a fully connected layer, and a regression layer.

[0040] The input layer is responsible for inputting parameters into the model. The input parameters are:

[0041]

[0042] In the formula, Representing the One input parameter;

[0043] The kernel size of both the first and second convolutional layers is 3×1, and the stride of the kernel is 1. The output after convolution is represented as follows:

[0044]

[0045] In the formula, Indicates the first One output unit This is the input unit of the convolutional layer. This represents the convolution kernel of the convolutional layer. For activation functions;

[0046] The first and second pooling layers are responsible for downsampling the output features of the convolutional layer. The downsampled output is as follows:

[0047] ;

[0048] In the formula, For the pooling layer One output unit for The downsampling coefficients of the function, in addition The function uses average pooling.

[0049] The fully connected layer is connected to all nodes of the second convolutional layer, and the output expression is:

[0050]

[0051] In the formula, For the fully connected layer One output unit For the first The convolutional layer neural unit and the first The connection weights between fully connected layer neural units, The first convolutional layer One output unit Indicates the first The threshold of a neural unit;

[0052] Random dropout should also be performed after the fully connected layer, and its function expression is:

[0053]

[0054] In the formula, The probability that each neural source is preserved;

[0055] The regression layer is responsible for outputting the prediction results of the convolutional neural network.

[0056] Preferably, in step S3.3, the performance function Q is defined as:

[0057]

[0058] In the formula, For the verification set The judgment value of each sample. For the true value, The number of samples in the validation set.

[0059] Preferably, in step S3.4, updating the velocity of each particle includes:

[0060]

[0061]

[0062] In the formula, is the th During the nth iteration The particle in the first Dimensional speed; and A random number within the range [0,1]; For inertial weights, The maximum inertia coefficient of the particle. The minimum inertia coefficient of the particle. The maximum number of iterations, This represents the current iteration number;

[0063] The update of the position of each particle includes:

[0064] .

[0065] Preferably, in step S3.6, the prediction performance evaluation indicators include the coefficient of determination R², the root mean square error RMSE, and the mean absolute percentage error MAPE, and their calculation formulas are as follows:

[0066] ;

[0067] ;

[0068] ;

[0069] In the formula, For predicted values, For the true value, This is the average value. This indicates the number of samples in the training set.

[0070] On the other hand, a motor corona detection system based on particle swarm optimization-convolutional neural network is characterized by comprising:

[0071] The detection module is used to acquire environmental data of the target motor stator winding, including pulse current time-domain waveform data, spectral component data, high-frequency voltage signal waveform data, operating voltage, load current, speed, power factor and motor running time.

[0072] The data processing module is used to preprocess multidimensional data according to a preset standardization function, remove dimensions, sort the standardized data by category to obtain a dataset that conforms to the model input format, and then divide the dataset into training set and validation set according to a preset ratio.

[0073] The model training module is used to input the formed training set into the convolutional neural network, and to perform global optimization training using the particle swarm optimization algorithm with the learning rate and random discard rate as search dimensions to obtain the optimal combination of learning rate and random discard rate of the convolutional neural network, thus forming the final prediction model.

[0074] The performance evaluation module is used to calculate the model's coefficient of determination R², root mean square error RMSE, and mean absolute percentage error MAPE using the results of the training set, and to evaluate the model's predictive performance.

[0075] The discharge judgment module is used to determine whether the target motor has experienced corona discharge based on the optimized and trained judgment model, and output the judgment result.

[0076] Preferably, the data processing module includes:

[0077] The first data processing submodule is used to standardize the feature sample data according to the standardization equation and remove the dimensions.

[0078] The second data processing submodule is used to sort the standardized data by category to obtain a dataset that conforms to the model input format corresponding to the detection data and sample data.

[0079] The third data processing submodule is used to divide the sample feature dataset into training and validation sets according to a preset ratio.

[0080] Preferably, the model training module includes:

[0081] The first particle swarm submodule is used for data initialization and to assign the number of particles according to the number of training set groups and the number of model parameter groups.

[0082] The second particle swarm submodule is used to guide particle transfer based on the particle's performance function Q.

[0083] The third particle swarm submodule is used to update the velocity and position of each particle according to the formula for updating particle velocity and position;

[0084] The fourth particle swarm submodule is used to determine whether the maximum number of iterations has been reached and to output the optimal parameters.

[0085] On the other hand, an electronic device is characterized by comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements a motor corona detection method based on a particle swarm optimization-convolutional neural network.

[0086] On the other hand, a non-transitory computer-readable storage medium is characterized in that it stores computer instructions thereon, which, when executed by a processor, implement a motor corona detection method based on a particle swarm optimization-convolutional neural network.

[0087] Compared with the prior art, the present invention has the following advantages:

[0088] (1) An adaptive intelligent feature extraction mechanism was established: This invention utilizes the adaptive feature learning capability of convolutional neural networks for non-stationary signals such as high-frequency pulse current and ultra-high-frequency electromagnetic waves to directly and automatically mine the deep features of corona discharge from the original detection data. This avoids the limitations of traditional methods that rely on manually designed statistical features, significantly improves the comprehensiveness and accuracy of feature extraction, and solves the problem of corona feature extraction failure under complex working conditions.

[0089] (2) Achieved globally optimal configuration of key CNN parameters: This invention overcomes the shortcomings of traditional gradient descent methods, such as being prone to getting trapped in local optima and relying on experience for parameter selection, by introducing a particle swarm optimization algorithm to globally optimize key parameters of the convolutional neural network, such as the learning rate and random discard rate. The particle swarm algorithm, through an efficient search mechanism that simulates the foraging behavior of bird flocks, can quickly lock in the optimal parameter combination, significantly shortening the model training time and improving the model's generalization ability in strong noise backgrounds.

[0090] (3) An end-to-end intelligent detection framework was constructed: The intelligent detection model constructed in this invention integrates data preprocessing, automatic feature extraction, parameter optimization, and classification judgment, and has end-to-end signal processing capabilities. It can directly extract effective features from noisy data, breaking through the bottleneck of traditional detection methods at the signal processing level. Compared with traditional methods, this model has stronger robustness to broadband noise generated by power electronic equipment such as frequency converters and switching power supplies, and reduces the false alarm rate and false alarm rate.

[0091] (4) Early and accurate warning of corona discharge is realized: The CNN model optimized by particle swarm optimization maintains stable detection performance under different motor operating conditions, effectively avoiding overfitting and realizing early and accurate warning of corona discharge. It provides reliable technical support for real-time monitoring and preventive maintenance of motor insulation status and has the potential to effectively avoid downtime accidents and safety risks caused by insulation faults. Attached Figure Description

[0092] Figure 1 This is a flowchart of the prediction method in this invention;

[0093] Figure 2 This is a schematic diagram of the connection of the prediction system in this invention;

[0094] Figure 3 This is a schematic diagram of the connection of the electronic device in this invention. Detailed Implementation

[0095] The present invention will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0096] Example 1: A method for detecting motor corona based on particle swarm optimization and convolutional neural networks

[0097] like Figure 1 As shown, step S1: Obtain multi-dimensional data on motor corona discharge and its operating status.

[0098] A detection device is set up at the motor operating site. A high-frequency current sensor collects pulse current time-domain waveform data from the motor grounding circuit, a spectrum analyzer obtains spectral component data, and a capacitive coupling sensor collects high-frequency voltage signal waveform data from the winding ends. Simultaneously, the motor control system reads operating parameters such as operating voltage, load current, speed, and power factor, as well as the motor's cumulative operating time.

[0099] The above data collection was performed under the following two conditions:

[0100] Normal condition: The motor operates stably under rated conditions, and no corona discharge occurs in the insulation system;

[0101] Fault status: The motor insulation system has shown obvious corona discharge phenomenon, which was confirmed by a partial discharge detector.

[0102] During the data collection process, the same motor is tested multiple times at different time points, or multiple motors of the same model are tested under the same environment. Each test forms a multidimensional data sample.

[0103] Step S2: Preprocess the multidimensional data and construct the dataset.

[0104] Step S2.1: Feature Extraction

[0105] The acquired pulse current time-domain waveform data were digitally filtered and envelope detected to extract the following features:

[0106] Pulse current sinusoidal wave amplitude;

[0107] The amplitude of the high-frequency current pulse superimposed on the sine wave;

[0108] Number of high-frequency current pulses;

[0109] Perform Fast Fourier Transform (FFT) analysis on the spectral component data to extract the amplitude of the dominant frequency component;

[0110] Wavelet denoising and peak detection are performed on the waveform data of high-frequency voltage signals to extract:

[0111] High-frequency voltage signal amplitude;

[0112] Number of high-frequency voltage pulses;

[0113] The above 6 extracted features, along with 10 parameters including operating voltage, load current, speed, power factor, and running time, constitute the original feature vector.

[0114] Step S2.2: Standardization Processing

[0115] The min-max normalization method is used to perform a linear transformation on each feature parameter, so that the size of each element is mapped to the interval [0,1].

[0116] The standardized feature vector is:

[0117] Step S2.3: Data Labeling and Dataset Partitioning

[0118] The standardized data is divided into two categories:

[0119] Normal state sample: Add a marker value of 0 to the end of the feature vector to form an 11×1 data matrix;

[0120] Fault state sample: Add a marker value of 1 to the end of the feature vector to form an 11×1 data matrix;

[0121] In this embodiment, a total of 400 samples were collected, including 200 samples in normal condition and 200 samples in fault condition. These samples were randomly divided into two groups at a ratio of 3:1.

[0122] Training set: 300 groups (150 normal groups and 150 faulty groups);

[0123] Validation set: 100 groups (50 normal groups and 50 fault groups).

[0124] Step S3: Particle Swarm Optimization-Convolutional Neural Network Model Construction and Training

[0125] Step S3.1: Particle Swarm Initialization

[0126] The adjustable parameters of the convolutional neural network, including the learning rate. Random discard rate Defined as the search dimension of the particle swarm, each particle corresponds to a parameter configuration scheme. Initialize the particle swarm, randomly generate the initial position and initial velocity of each particle, set the maximum number of iterations, inertia weight, individual learning factor, and global learning factor, and initialize the individual optimal position of each particle to its own position, and initialize the global optimal position to the random particle position.

[0127] Step S3.2: Convolutional Neural Network Construction

[0128] For each particle, a convolutional neural network model is constructed using the parameters corresponding to its current position; the convolutional neural network structure includes an input layer, at least one convolutional layer, at least one pooling layer, a fully connected layer, and a regression layer;

[0129] The convolutional neural network structure comprises seven interconnected layers: an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a fully connected layer, and a regression layer.

[0130] The input layer is responsible for inputting parameters into the model. The input parameters are:

[0131]

[0132] In the formula, Representing the One input parameter;

[0133] The kernel size of both the first and second convolutional layers is 3×1, and the stride of the kernel is 1. The output after convolution is represented as follows:

[0134]

[0135] In the formula, Indicates the first One output unit This is the input unit of the convolutional layer. This represents the convolution kernel of the convolutional layer. For activation functions;

[0136] The first and second pooling layers are responsible for downsampling the output features of the convolutional layer. The downsampled output is as follows:

[0137] ;

[0138] In the formula, For the pooling layer One output unit for The downsampling coefficients of the function, in addition The function uses average pooling.

[0139] The fully connected layer is connected to all nodes of the second convolutional layer, and the output expression is:

[0140]

[0141] In the formula, For the fully connected layer One output unit For the first The convolutional layer neural unit and the first The connection weights between fully connected layer neural units, The first convolutional layer One output unit Indicates the first The threshold of a neural unit;

[0142] Random dropout should also be performed after the fully connected layer, and its function expression is:

[0143]

[0144] In the formula, The probability that each neural source is preserved;

[0145] The regression layer is responsible for outputting the prediction results of the convolutional neural network.

[0146] Step S3.3: Model Training and Performance Evaluation

[0147] The convolutional neural network is trained using the training set, and a performance function is computed on the validation set; the performance function is defined as the reciprocal of the error between the predicted value and the true value on the validation set.

[0148] The performance function Q is defined as follows:

[0149]

[0150] In the formula, For the verification set The judgment value of each sample. For the true value, The number of samples in the validation set.

[0151] Step S3.4: Particle Swarm Iterative Update

[0152] Update the individual optimal value and the global optimal value of the particles according to the performance function, and update the velocity and position of each particle;

[0153] The update of the velocity of each particle includes:

[0154]

[0155]

[0156] In the formula, is the th During the nth iteration The particle in the first Dimensional speed; and A random number within the range [0,1]; For inertial weights, The maximum inertia coefficient of the particle. The minimum inertia coefficient of the particle. The maximum number of iterations, This represents the current iteration number;

[0157] The update of the position of each particle includes:

[0158] .

[0159] Step S3.5: Iteration Termination and Optimal Parameter Output

[0160] Verify whether the maximum number of iterations has been reached. If not, return to step S3.2 and iterate again. If the maximum number of iterations has been reached, output the optimal parameter combination.

[0161] Repeat steps S3.2 to S3.4 until the maximum number of iterations T is reached;

[0162] After the iteration terminates, the globally optimal position is output as the best combination of parameters.

[0163] Step S3.6: Model Performance Evaluation

[0164] The predictive performance evaluation index of the model is calculated using the true values ​​in the training set and the predicted values ​​output by the prediction model.

[0165] Reconstruct the CNN model using the optimal parameter combination, perform final training on the training set, and calculate the following prediction performance metrics:

[0166] The predictive performance evaluation indicators include the coefficient of determination R², root mean square error RMSE, and mean absolute percentage error MAPE, and their calculation formulas are as follows:

[0167] ;

[0168] ;

[0169] ;

[0170] In the formula, For predicted values, For the true value, This is the average value. This indicates the number of samples in the training set.

[0171] Step S4: Corona discharge assessment

[0172] After the multidimensional data of the motor to be tested is preprocessed in step S2, it is input into the judgment model constructed by the optimal parameter combination, and the model outputs the predicted probability.

[0173] Set the judgment threshold to 0.5;

[0174] If the output prediction probability is ≥0.5, it is determined that the target motor has corona discharge, and the judgment result is "1" is output.

[0175] If the output prediction probability is less than 0.5, the target motor is judged to be operating normally, and the judgment result is "0".

[0176] Example 2: Motor Corona Detection System Based on Particle Swarm Optimization-Convolutional Neural Network

[0177] like Figure 2 As shown, the system described in this embodiment includes the following modules:

[0178] The detection module is used to acquire environmental data of the target motor stator winding, including pulse current time-domain waveform data, spectral component data, high-frequency voltage signal waveform data, operating voltage, load current, speed, power factor and motor running time.

[0179] The data processing module is used to preprocess multidimensional data according to a preset standardization function, remove dimensions, sort the standardized data by category to obtain a dataset that conforms to the model input format, and then divide the dataset into training set and validation set according to a preset ratio.

[0180] The model training module is used to input the formed training set into the convolutional neural network, and to perform global optimization training using the particle swarm optimization algorithm with the learning rate and random discard rate as search dimensions to obtain the optimal combination of learning rate and random discard rate of the convolutional neural network, thus forming the final prediction model.

[0181] The performance evaluation module is used to calculate the model's coefficient of determination R², root mean square error RMSE, and mean absolute percentage error MAPE using the results of the training set, and to evaluate the model's predictive performance.

[0182] The discharge judgment module is used to determine whether the target motor has experienced corona discharge based on the optimized and trained judgment model, and output the judgment result.

[0183] The data processing module specifically includes: the first data processing submodule: performing feature extraction and standardization. It performs digital filtering, envelope detection, FFT analysis, and wavelet denoising on the original waveform data, extracting 10-dimensional feature parameters, and mapping them to the [0,1] interval using a minimum-maximum normalization method.

[0184] The second data processing submodule: data format conversion and sorting. The standardized feature vectors are sorted by timestamp to construct a structured dataset, with each sample being an 11×1 data matrix (10-dimensional features + 1-dimensional label).

[0185] The third data processing submodule: Dataset partitioning. The dataset is randomly partitioned according to a preset ratio (75% training set and 25% validation set) to ensure a balanced distribution of the two types of samples in the training and validation sets.

[0186] The model training module specifically includes: the first particle swarm submodule: initializing particle swarm parameters. Based on the training set size and model complexity, the particle swarm size is set to m=3, the search space dimension D=2 (learning rate and random discard rate), and the initial particle positions and velocities are randomly generated.

[0187] The second particle swarm optimization submodule focuses on performance evaluation and particle transfer guidance. A CNN model is constructed for each particle's parameter combination. After training on the training set, the performance function Q is evaluated on the validation set. The Q value guides the particles to transfer in the optimal direction.

[0188] The third particle swarm submodule: velocity and position update. Following the formula described in claim 11, the velocity and position of each particle are updated based on individual optimal and global optimal information, and boundary constraint processing is performed.

[0189] The fourth particle swarm submodule: Iterative control and optimal parameter output. It determines whether the current iteration count has reached a preset maximum value; if so, it outputs the globally optimal parameter combination; otherwise, it triggers the next iteration loop.

[0190] Example 3: Electronic device and non-transitory computer-readable storage medium

[0191] like Figure 3As shown, the electronic device includes a non-volatile memory, which is used for data caching and intermediate calculation result storage during operation; it also includes a central processing unit, which, when executing a computer program stored in the non-volatile memory, implements the motor corona detection method based on particle swarm optimization-convolutional neural network as described in Embodiment 1.

[0192] The non-transitory computer-readable storage medium is a solid-state drive (SSD) or a USB flash drive, which stores computer instructions. When the computer instructions are executed by the processor, they implement the motor corona detection method based on particle swarm optimization-convolutional neural network as described in Example 1.

[0193] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0194] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

Claims

1. A method for detecting motor corona based on particle swarm optimization-convolutional neural network, characterized in that, Includes the following steps: Step S1: Obtain multidimensional data on motor corona discharge and its operating status; Step S2: Preprocess the multidimensional data and construct a dataset, dividing the dataset into a training set and a validation set; Step S3: Particle Swarm Optimization-Convolutional Neural Network Model Construction and Training Step S3.1: Particle swarm initialization; The adjustable parameters of the convolutional neural network, including the learning rate. Random discard rate Defined as the search dimension of the particle swarm, each particle corresponds to a parameter configuration scheme. Initialize the particle swarm, randomly generate the initial position and initial velocity of each particle, set the maximum number of iterations, inertia weight, individual learning factor, and global learning factor, and initialize the individual optimal position of each particle to its own position, and initialize the global optimal position to the random particle position. Step S3.2: Convolutional Neural Network Construction For each particle, a convolutional neural network model is constructed using the parameters corresponding to its current position; the convolutional neural network structure includes an input layer, at least one convolutional layer, at least one pooling layer, a fully connected layer, and a regression layer; Step S3.3: Model Training and Performance Evaluation The convolutional neural network is trained using the training set, and a performance function is computed on the validation set; the performance function is defined as the reciprocal of the error between the predicted value and the true value on the validation set. Step S3.4: Particle Swarm Iterative Update Update the individual optimal value and the global optimal value of the particles according to the performance function, and update the velocity and position of each particle; Step S3.5: Iteration Termination and Optimal Parameter Output Verify whether the maximum number of iterations has been reached. If not, return to step S3.2 and iterate again. If the maximum number of iterations is reached, output the optimal parameter combination; Step S3.6: Model Performance Evaluation The predictive performance evaluation index of the model is calculated using the true values ​​in the training set and the predicted values ​​output by the prediction model. Step S4: Corona discharge assessment The data to be detected is input into the judgment model constructed by the optimal parameter combination to obtain the judgment result of whether the target motor has corona discharge.

2. The method for detecting motor corona based on particle swarm optimization-convolutional neural network according to claim 1, characterized in that, In step S1, the multidimensional data includes: pulse current time-domain waveform data and spectral component data detected from the motor grounding circuit during normal operation and when corona discharge occurs; waveform data of high-frequency voltage signals detected from the winding ends; operating voltage, load current, speed and power factor data during normal operation and when corona discharge occurs; and motor running time data.

3. The method for detecting motor corona based on particle swarm optimization-convolutional neural network according to claim 1 or 2, characterized in that, In step S1, the multidimensional data includes data obtained by detecting the same motor running at different time points within a preset time period; or data obtained by detecting the same motor running in different environments at the same time; each detection forms a multidimensional data matrix.

4. The method for detecting motor corona based on particle swarm optimization-convolutional neural network according to claim 1, characterized in that, In step S2, the preprocessing includes: The pulse current time-domain waveform data, spectral component data, and high-frequency voltage signal waveform data are preprocessed to extract the sine wave amplitude of the pulse current, the amplitude and number of high-frequency current pulses superimposed on the sine wave, and the amplitude and number of high-frequency voltage signals. The extracted feature data, along with 10 data points including the operating voltage, load current, speed, power factor, and motor running time, are used as feature parameters. A linear transformation is performed using a standardization function to transform the size of each element to the [0,1] interval.

5. The method for detecting motor corona based on particle swarm optimization-convolutional neural network according to claim 1, characterized in that, In step S2, constructing the dataset includes: The preprocessed data is divided into two parts: the first part comes from motors that are operating normally and have not experienced corona discharge, and the corresponding data matrix is ​​marked as 0; the second part comes from motors that have experienced corona discharge, and the corresponding data matrix is ​​marked as 1; each sample is an 11×1 data matrix; the data is divided into training set and validation set according to a preset ratio.

6. The method for detecting motor corona based on particle swarm optimization-convolutional neural network according to claim 1, characterized in that, In step S3.2, the convolutional neural network structure comprises seven interconnected layers: an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a fully connected layer, and a regression layer. The input layer is responsible for inputting parameters into the model. The input parameters are: ； In the formula, Representing the One input parameter; The kernel size of both the first and second convolutional layers is 3×1, and the stride of the kernel is 1. The output after convolution is represented as follows: ； In the formula, Indicates the first One output unit This is the input unit of the convolutional layer. This represents the convolution kernel of the convolutional layer. For activation functions; The first and second pooling layers are responsible for downsampling the output features of the convolutional layer. The downsampled output is as follows: ； In the formula, For the pooling layer One output unit for The downsampling coefficients of the function, in addition The function uses average pooling. The fully connected layer is connected to all nodes of the second convolutional layer, and the output expression is: ； In the formula, For the fully connected layer One output unit For the first The convolutional layer neural unit and the first The connection weights between fully connected layer neural units, The first convolutional layer One output unit Indicates the first The threshold of a neural unit; Random dropout should also be performed after the fully connected layer, and its function expression is: ； In the formula, The probability that each neural source is preserved; The regression layer is responsible for outputting the prediction results of the convolutional neural network.

7. The method for detecting motor corona based on particle swarm optimization-convolutional neural network according to claim 1, characterized in that, In step S3.3, the performance function Q is defined as: ； In the formula, For the verification set The judgment value of each sample. For the true value, The number of samples in the validation set.

8. The method for detecting motor corona based on particle swarm optimization-convolutional neural network according to claim 1, characterized in that: In step S3.4, updating the velocity of each particle includes: ； ； In the formula, is the th During the nth iteration The particle in the first Dimensional speed; and A random number within the range [0,1]; For inertial weights, The maximum inertia coefficient of the particle. The minimum inertia coefficient of the particle. The maximum number of iterations, This represents the current iteration number; The update of the position of each particle includes: 。 9. The method for detecting motor corona based on particle swarm optimization-convolutional neural network according to claim 1, characterized in that, In step S3.6, the prediction performance evaluation indicators include the coefficient of determination R², the root mean square error RMSE, and the mean absolute percentage error MAPE, and their calculation formulas are as follows: ； ； ； In the formula, For predicted values, For the true value, This is the average value. This indicates the number of samples in the training set.

10. A motor corona detection system based on particle swarm optimization-convolutional neural network, characterized in that, include: The detection module is used to acquire environmental data of the target motor stator winding, including pulse current time-domain waveform data, spectral component data, high-frequency voltage signal waveform data, operating voltage, load current, speed, power factor and motor running time. The data processing module is used to preprocess multidimensional data according to a preset standardization function, remove dimensions, sort the standardized data by category to obtain a dataset that conforms to the model input format, and then divide the dataset into training set and validation set according to a preset ratio. The model training module is used to input the formed training set into the convolutional neural network, and to perform global optimization training using the particle swarm optimization algorithm with the learning rate and random discard rate as search dimensions to obtain the optimal combination of learning rate and random discard rate of the convolutional neural network, thus forming the final prediction model. The performance evaluation module is used to calculate the model's coefficient of determination R², root mean square error RMSE, and mean absolute percentage error MAPE using the results of the training set, and to evaluate the model's predictive performance. The discharge judgment module is used to determine whether the target motor has experienced corona discharge based on the optimized and trained judgment model, and output the judgment result.

11. The motor corona detection system based on particle swarm optimization-convolutional neural network according to claim 10, characterized in that, The data processing module includes: The first data processing submodule is used to standardize the feature sample data according to the standardization equation and remove the dimensions. The second data processing submodule is used to sort the standardized data by category to obtain a dataset that conforms to the model input format corresponding to the detection data and sample data. The third data processing submodule is used to divide the sample feature dataset into training and validation sets according to a preset ratio.

12. The motor corona detection system based on particle swarm optimization-convolutional neural network according to claim 10, characterized in that, The model training module includes: The first particle swarm submodule is used for data initialization and to assign the number of particles according to the number of training set groups and the number of model parameter groups. The second particle swarm submodule is used to guide particle transfer based on the particle's performance function Q. The third particle swarm submodule is used to update the velocity and position of each particle according to the formula for updating particle velocity and position; The fourth particle swarm submodule is used to determine whether the maximum number of iterations has been reached and to output the optimal parameters.

13. An electronic device, characterized in that, include: The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the motor corona detection method based on particle swarm optimization-convolutional neural network as described in any one of claims 1 to 9.

14. A non-transitory computer-readable storage medium, characterized in that, It stores computer instructions, which, when executed by a processor, implement the motor corona detection method based on particle swarm optimization-convolutional neural network as described in any one of claims 1 to 9.

Images