Method and apparatus for determining hysteresis of transformer, and transformer hysteresis detection system

Abstract

The present application provides a method, an apparatus, and a transformer hysteresis detection system for determining the hysteresis of a transformer. The method includes: obtaining relevant parameters of the transformer, where the relevant parameters include one or more of magnetic induction intensity, magnetic field intensity, temperature, operating frequency, and DC bias; constructing a hysteresis analysis model, where the hysteresis analysis model is trained using multiple sets of training data through a convolution algorithm and a vertical and horizontal cross algorithm, and each set of training data in the multiple sets of training data includes historical relevant parameters obtained within a historical time period and historical hysteresis analysis results corresponding to the historical relevant parameters; and inputting the relevant parameters into the hysteresis analysis model to obtain a hysteresis analysis result corresponding to the relevant parameters. This solution solves the problem in the prior art that when using a BP neural network to study the hysteresis phenomenon, the accuracy of the obtained results is poor.

Description

Technical Field

[0001] The present application relates to the technical field of hysteresis characteristic analysis, and in particular, to a method and device for determining the hysteresis of a transformer, a computer program product, and a transformer hysteresis detection system. Background Art

[0002] Electromagnetic components are affected by complex operating conditions and thus exhibit different operating characteristics. In addition to the different shapes of the excitation waveforms, other operating conditions, such as different frequencies, temperatures, DC bias levels, etc., will result in different shapes of the B-H loop. Therefore, studying the hysteresis and loss characteristics of materials under different conditions is crucial for the efficient operation of electrical equipment.

[0003] Currently, the BP neural network is often used to study the hysteresis phenomenon, but the accuracy of the final result is poor. Summary of the Invention

[0004] The main object of the present application is to provide a method and device for determining the hysteresis of a transformer, a computer program product, and a transformer hysteresis detection system, so as to at least solve the problem that the accuracy of the result obtained by using the BP neural network to study the hysteresis phenomenon in the prior art is poor.

[0005] To achieve the above object, according to one aspect of the present application, a method for determining the hysteresis of a transformer is provided, including: obtaining relevant parameters of the transformer, where the relevant parameters include one or more of magnetic induction intensity, magnetic field intensity, temperature, operating frequency, and DC bias; constructing a hysteresis analysis model, where the hysteresis analysis model is trained by using a convolutional algorithm and a crosswise and longitudinal cross algorithm with multiple sets of training data, and each set of training data in the multiple sets of training data includes historical relevant parameters obtained within a historical time period and historical hysteresis analysis results corresponding to the historical relevant parameters; inputting the relevant parameters into the hysteresis analysis model to obtain the hysteresis analysis results corresponding to the relevant parameters.

[0006] Optionally, obtaining relevant parameters of the transformer includes: obtaining initial relevant parameters, where the initial relevant parameters are the parameters of the transformer obtained at an initial moment; preprocessing the initial relevant parameters to obtain the relevant parameters, where the preprocessing method includes at least one or more of interpolation processing, normalization processing, denoising processing, and outlier removal processing.

[0007] Optionally, before constructing the hysteresis analysis model, the method further includes: constructing a convolutional layer, where the convolutional layer is used to perform a convolutional operation on the input data; constructing a pooling layer, where the pooling layer is used to remove redundant information from the data output by the convolutional layer and perform a pooling operation; constructing a fusion layer, where the fusion layer is used to perform feature fusion on the data output by the pooling layer; constructing an output layer, where the output layer is used to output a preliminary analysis result and update the weights and biases of the data using the backpropagation algorithm.

[0008] Optionally, before constructing the hysteresis analysis model, the method further includes: a first step of performing horizontal crossover and vertical crossover on the population, where horizontal crossover represents arithmetic crossover between different particles of the population in the same dimension, and vertical crossover represents arithmetic crossover between the same particle of the population in different dimensions; a second step of using a greedy algorithm to select the optimal particle after crossover; a third step of decaying the probabilities of horizontal and vertical crossovers and performing iteration until a preset number of iterations is reached and the iteration stops; a fourth step of outputting a preliminary analysis result when a preset condition is satisfied, and re-executing the first step, the second step, and the third step when the preset condition is not satisfied, where the preset condition at least includes that the number of iterations is greater than the preset number of iterations.

[0009] Optionally, the first step includes: randomly pairing the particles in the population and performing horizontal crossover on any two paired particles using a horizontal crossover formula, where the horizontal crossover formula is:

[0010]

[0011] represents a particle after horizontal crossover, r 1 represents a random number within a first range, represents a particle before horizontal crossover, represents another particle after horizontal crossover, c 1 represents a random number within a second range, where the first range is smaller than the second range, represents another particle after horizontal crossover, r 2 represents the random number within the first range, c 2 Updating any two dimensions of a particle in the population and performing vertical crossover on the particle using a vertical crossover formula, where the vertical crossover formula is:

[0012]

[0013] Denote the offspring particles after vertical crossover, r represents the random number within the first range, SM ij1 Denote the parent particles in one dimension, SM ij2 Denote the parent particles in another dimension.

[0014] Optionally, the second step includes: using a fitness function to calculate the fitness of the particles at their respective positions; comparing the magnitude relationships among multiple fitness values; using a greedy algorithm to extract the particle with the highest fitness to obtain the optimal particle.

[0015] Optionally, the third step includes: obtaining an initial vertical crossover probability and an initial horizontal crossover probability; using a horizontal decay formula to decay the initial horizontal crossover probability and performing iterations until the preset number of iterations is reached and the iteration stops. The horizontal decay formula is:

[0016]

[0017] pc t Denote the horizontal crossover probability at the t-th iteration, pc min The minimum horizontal crossover probability, pc 0 Denote the initial horizontal crossover probability, T represents the preset number of iterations; using a vertical decay formula to decay the initial vertical crossover probability and performing iterations until the preset number of iterations is reached and the iteration stops. The vertical decay formula is:

[0018]

[0019] hc t Denote the vertical crossover probability at the t-th iteration, hc min Denote the minimum vertical crossover probability, hc 0 Denote the initial vertical crossover probability.

[0020] According to another aspect of the present application, there is provided a device for determining the hysteresis of a transformer, including: an acquisition unit for acquiring relevant parameters of the transformer, where the relevant parameters include one or more of magnetic induction intensity, magnetic field intensity, temperature, operating frequency, and DC bias; a first construction unit for constructing a hysteresis analysis model, where the hysteresis analysis model is trained using multiple sets of training data through a convolution algorithm and a vertical and horizontal crossover algorithm, and each set of training data in the multiple sets of training data includes historical relevant parameters acquired within a historical time period and the historical hysteresis analysis results corresponding to the historical relevant parameters; a first processing unit for inputting the relevant parameters into the hysteresis analysis model to obtain the hysteresis analysis results corresponding to the relevant parameters.

[0021] According to another aspect of the present application, there is provided a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the steps of the method for determining the hysteresis of any one of the transformers.

[0022] According to yet another aspect of the present application, there is provided a transformer hysteresis detection system, including: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include those for executing the method for determining the hysteresis of any one of the transformers.

[0023] By applying the technical solution of the present application, a plurality of relevant parameters of the transformer are obtained. The method of combining multiple factors can make up for the shortcoming that the existing hysteresis analysis model cannot fit the hysteresis characteristics. And the convolutional neural network is used for feature training. The convolutional neural network has stronger feature extraction ability and data integration ability compared with the BP neural network. At the same time, the cross and longitudinal intersection algorithm is used for optimization, and the cross and longitudinal intersection algorithm has better optimization efficiency. Therefore, the present solution uses the combination of convolution and cross and longitudinal intersection to construct the hysteresis analysis model, so that the accuracy of the finally obtained result is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

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

[0025] Figure 1 It shows a hardware structure block diagram of a mobile terminal for executing the method for determining the hysteresis of a transformer provided in the embodiment of the present application;

[0026] Figure 2 It shows a schematic flowchart of a method for determining the hysteresis of a transformer provided in the embodiment of the present application;

[0027] Figure 3 It shows a schematic diagram of obtaining a hysteresis curve;

[0028] Figure 4 It shows a schematic flowchart of a convolutional neural network and a cross and longitudinal intersection algorithm;

[0029] Figure 5 It shows a schematic flowchart of the cross and longitudinal intersection algorithm;

[0030] Figure 6 It shows a structure block diagram of a device for determining the hysteresis of a transformer provided in the embodiment of the present application.

[0031] Among them, the above-mentioned drawings include the following reference numerals:

[0032] 102, processor; 104, memory; 106, transmission device; 108, input / output device. Detailed implementation manners

[0033] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other. The following will describe the present application in detail with reference to the drawings and in combination with the embodiments.

[0034] In order to enable those skilled in the art to better understand the solution of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0035] It should be noted that the terms "first", "second", etc. in the description and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that such used data may be interchanged under appropriate circumstances for the embodiments of the present application described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0036] The J-A model derives a set of ordinary differential equations with five parameters based on the domain wall motion and magnetic domain rotation mechanism of ferromagnetic materials, and obtains the hysteresis loop conforming to the characteristics of magnetic materials by solving the differential equations. However, the classical J-A model is a static model and cannot be applied to simulate dynamic magnetism. The Preisach model regards the magnetic material as composed of many small magnetic domains, and each small magnetic domain can change with the change of the applied magnetic field, thereby fitting the nonlinear behavior of the overall material, but it cannot accurately describe the reversible magnetization process of vector hysteresis. These traditional models are all established based on empirical simplification or physical approximation, which limits their ability to capture complex waveforms, temperature, and DC bias.

[0037] The neural network has the advantage of being able to unify many intertwined influencing factors (such as temperature and DC bias) into a framework, which makes it a good choice for hysteresis model modeling. Although the commonly used BP neural network today has powerful learning ability in dealing with some complex problems, when the input data volume is too large and the number of weights is too many, problems such as slow convergence or inability to calculate will occur. It is necessary to preprocess the data manually, which may cause the loss of the original feature information, resulting in the overfitting phenomenon, that is, larger errors will occur in the data that has not been trained.

[0038] As introduced in the background art, in the prior art, the BP neural network is used to study the hysteresis phenomenon, and the accuracy of the obtained results is poor. To solve the above problems, the embodiments of the present application provide a method, a device, a computer program product and a transformer hysteresis detection system for determining the hysteresis of a transformer.

[0039] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention.

[0040] The method embodiments provided in the embodiments of the present application can be executed on a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 is a hardware structure block diagram of a mobile terminal for a method of determining the hysteresis of a transformer according to an embodiment of the present invention. As Figure 1 shown, the mobile terminal may include one or more ( Figure 1 only one is shown in the figure) processors 102 (the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data. Among them, the above mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those of ordinary skill in the art can understand that Figure 1 the structure shown is only schematic and does not limit the structure of the above mobile terminal. For example, the mobile terminal may further include more or fewer components than Figure 1 shown in the figure, or have a different configuration from Figure 1 shown in the figure.

[0041] The memory 104 can be used to store computer programs, such as software programs and modules of application software, such as the computer program corresponding to the display method of device information in the embodiments of the present invention. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, that is, the above-mentioned method is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory 104 may further include a memory remotely disposed relative to the processor 102, and these remote memories may be connected to the mobile terminal through a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof. The transmission device 106 is used to receive or send data via a network. A specific example of the above-mentioned network may include a wireless network provided by a communication provider of the mobile terminal. In one instance, the transmission device 106 includes a network adapter (Network Interface Controller, abbreviated as NIC), which can be connected to other network devices through a base station and thus can communicate with the Internet. In one instance, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0042] In this embodiment, a method for determining the hysteresis of a transformer operating on a mobile terminal, a computer terminal, or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0043] Figure 2 It is a flowchart diagram of a method for determining the hysteresis of a transformer according to an embodiment of the present application. As Figure 2 shown, the method includes the following steps:

[0044] Step S201, obtain relevant parameters of the transformer, where the above-mentioned relevant parameters include one or more of magnetic induction intensity, magnetic field intensity, temperature, operating frequency, and DC bias;

[0045] Specifically, relevant parameters of the transformer can be obtained. In this solution, multi-factor parameters are obtained, and the obtained parameters include one or more of magnetic induction intensity, magnetic field intensity, temperature, operating frequency, and DC bias. By using multi-factor parameters, more efficient hysteresis analysis can be ensured.

[0046] Step S202: Construct a hysteresis analysis model. The hysteresis analysis model is trained using a convolutional algorithm and a cross and crisscross algorithm with multiple sets of training data. Each set of training data in the multiple sets of training data includes historical relevant parameters obtained within a historical time period and the corresponding historical hysteresis analysis results of the historical relevant parameters.

[0047] Specifically, the convolutional algorithm and the cross and crisscross algorithm can be combined. First, use the convolutional algorithm to construct a neural network model, and then use the cross and crisscross algorithm to optimize the neural network for machine learning training, thereby obtaining the hysteresis analysis model.

[0048] Step S203: Input the relevant parameters into the hysteresis analysis model to obtain the hysteresis analysis results corresponding to the relevant parameters.

[0049] Specifically, since the hysteresis analysis model can predict the hysteresis analysis results of a transformer, the relevant parameters of the transformer can be input into the hysteresis analysis model to obtain the output of the hysteresis analysis model, that is, the hysteresis analysis results corresponding to the relevant parameters of the transformer.

[0050] Through this embodiment, multiple relevant parameters of the transformer are obtained. The multi-factor combination method can make up for the shortcoming of the existing hysteresis analysis model that cannot fit the hysteresis characteristics. And the convolutional neural network method is used for feature training. The convolutional neural network has stronger feature extraction ability and data integration ability compared with the BP neural network. At the same time, the cross and crisscross algorithm is used for optimization, and the cross and crisscross algorithm has good optimization efficiency. Therefore, this solution uses the combination of convolution and cross and crisscross to construct the hysteresis analysis model, greatly improving the accuracy of the final result.

[0051] Aiming at the problem that the current traditional models have poor prediction performance under different working conditions, such as different amplitudes, frequencies, temperatures, and DC bias levels, and the problem that the neural network converges too slowly and has insufficient feature extraction when the data volume is too large. This solution uses the convolutional operation in the neural network to construct a non-linear mapping relationship, taking advantage of the powerful self-extracting feature of the convolutional neural network to achieve an accurate simulation of the complex magnetic characteristics of the transformer. The hysteresis analysis model has the advantages of short training time, strong learning and generalization ability, strong robustness, and being more suitable for engineering practice. At the same time, the cross and crisscross algorithm is used to optimize the hyperparameters of the convolutional neural network model, effectively solving the problem that it is difficult to determine the hyperparameters of the neural network model and improving the accuracy.

[0052] Specifically, the above solution is as Figure 3As shown, it can be roughly divided into the following four steps: ① Separate the magnetic flux density, magnetic induction intensity B, and magnetic field intensity H from the measured current and voltage data, and compare the obtained magnetic induction intensity B, magnetic field intensity H with the related factors temperature T, frequency f, DC bias H dc After normalization, the data is input into the convolutional neural network model. The processed data is divided into training set, validation set and prediction set according to the moving window; ② A convolutional neural network with dual channels is built (convolutional neural network prediction model). The main channel inputs the field strength H, and the auxiliary channel inputs various factors to realize the model training through the back propagation algorithm, and autonomously extracts and learns the magnetic induction intensity B and field strength H, temperature T, frequency f, and magnetic bias H dc The relationship between the magnetic characteristics of the two networks (the data is input into the model, the convolutional neural network extracts the relationship between the learning features and the output through the convolutional layer, the pooling layer realizes the feature dimensionality reduction, and finally the fully connected layer outputs the results); ③ The vertical and horizontal cross optimization algorithm (CSO) with the cross probability decay strategy is used to optimize the hyperparameters in the convolutional neural network, including the number of iterations N, the number of convolution kernels K, and the learning rate α, to complete the training of the prediction model (the convolutional neural network model uses the vertical and horizontal cross optimization algorithm to optimize the number of convolution kernels, the number of training iterations, and the learning path to complete the training of the model); ④ Save the trained hysteresis analysis model to simulate the complex magnetic characteristics of the transformer, and use the trained model input to obtain the corresponding transformer hysteresis curve.

[0053] As mentioned above, this scheme proposes for the first time a multi-factor related transformer magnetic characteristic modeling method based on convolutional neural network. The model can obtain the corresponding magnetic field strength by directly inputting the time series data of magnetic flux density and other factors. Compared with the previous BP neural network hysteresis analysis model, it eliminates the process of manually extracting the characteristic parameters of magnetic flux density and considers the influence of multiple factors at the same time, realizing the autonomous extraction and learning of transformer hysteresis characteristics under complex conditions. In view of the problem that the hyperparameters of the neural network model are difficult to determine, the article introduces a vertical and horizontal cross algorithm to optimize the hyperparameters and improve the performance of the neural network hysteresis analysis model.

[0054] In this solution, the convolution algorithm and the cross-link algorithm are combined to improve the performance of the model in image recognition, natural language processing and other fields. In convolutional neural networks, convolution layers are usually used to extract local features, while the cross-link algorithm can be used to perform global cross-linking on these local features. The specific steps are as follows:

[0055] First, the convolutional layer is used to extract features from the input data to obtain a local feature map;

[0056] Then, the local feature map is input into the vertical and horizontal cross-learning algorithm, and the features at different positions are combined through cross-learning to obtain the global feature representation;

[0057] Finally, input the global feature representation into the fully connected layer for classification or other tasks.

[0058] Suppose an image needs to be classified. A convolutional neural network can be used to extract the local features of the image, and then the cross - intersection algorithm can be used to combine these local features to obtain the global features. Take a handwritten digit image as an example:

[0059] First, extract the local features of the image, such as edges, textures, etc., through the convolutional layer.

[0060] Then, input the local features into the cross - intersection algorithm to combine the features at different positions and obtain the global feature representation, such as the overall shape of the digit, the thickness of the strokes, etc.

[0061] Finally, input the global feature representation into the fully connected layer for digit classification to recognize the digit on the image.

[0062] By combining the convolutional algorithm and the cross - intersection algorithm, local and global information can be fully utilized to improve the performance and accuracy of the model in tasks such as image classification.

[0063] In the specific implementation process, obtaining the relevant parameters of the transformer can be achieved through the following steps: obtaining the initial relevant parameters, where the above - mentioned initial relevant parameters are the parameters of the above - mentioned transformer obtained at the initial moment; pre - processing the above - mentioned initial relevant parameters to obtain the above - mentioned relevant parameters, where the pre - processing methods include at least one or more of interpolation processing, normalization processing, denoising processing, and outlier removal processing.

[0064] In this scheme, the initial relevant parameters can be pre - processed. Interpolation processing of the parameters can fill in the missing parts of the parameters to make the parameters more complete and avoid inaccurate analysis results due to missing parameters. Normalization processing can unify parameters of different dimensions to the same scale, avoid the influence of parameter differences between different dimensions on the analysis results, and improve the comparability and interpretability of the parameters. Denoising processing can remove the noise in the parameters to make the parameters cleaner and more stable, improving the quality and credibility of the parameters. Outlier removal processing can remove the outliers in the parameters to avoid the influence of outliers on the analysis results and improve the accuracy and reliability of the parameters.

[0065] Specifically, the following introduces data pre - processing. The waveforms of B(t) and H(t) are calculated directly from the original measurements of voltage and current signals. By measuring the voltage and current under sinusoidal excitation, the parameters of the proposed model are determined, where t represents the time.

[0066] According to Ampere's law, the magnetic field strength H can be expressed as: H(t) = (N 1 ·I) / l m , where N1 is the number of turns of the primary coil. l m is the effective magnetic path length, and I is the excitation current.

[0067] The magnetic flux density B(t) is calculated as follows:

[0068]

[0069] where N 2 is the number of turns of the secondary coil, u 2 (t) is the induced voltage of the secondary coil, and S is the cross-sectional area of the sample.

[0070] These waveforms consist of multiple cycles captured in a periodic steady state. Each waveform is a 1×16750 time series, and these waveforms are multiple waveform cycles captured in a steady state. However, these waveforms are captured in a periodic steady state, and the repeated cycles do not provide much valuable information for training the network. Utilizing such a large number of data points will increase the computational cost of network training and inference. Moreover, the number of sampling points in each cycle of waveforms with different frequencies is different, which will lead to different numbers of points on the B-H plane. The network trained with these waveforms is prone to amplifying the noise in the low-frequency waveforms while ignoring the characteristics of the high-frequency waveforms.

[0071] To solve these problems, a single-cycle interpolation algorithm is applied to all B(t) and H(t) waveforms. Given the sampling rate fs and the fundamental frequency f, the total number of cycles contained in each waveform can be calculated as N = 16750×(f / fs). First, the waveform of 16750 samples is interpolated into N×256 samples using the spline algorithm. Then, the interpolated waveform is evenly divided into multiple complete single-cycle waveforms, with a total of 256 sampling points. Finally, all the parts are averaged into a single-cycle waveform.

[0072] To ensure better convergence of the model, the measured magnetic induction intensity B, magnetic field intensity H, and the related factors temperature T, frequency f, and DC bias H dc are normalized to the range of [-1, 1] using min-max normalization to obtain the processed list of N×256 sample time series [B(t), H(t), f(t), T(t), H dc (t)].

[0073] In the specific implementation process, before constructing the hysteresis analysis model, the above method further includes the following steps: constructing a convolutional layer, where the convolutional layer is used to perform a convolutional operation on the input data; constructing a pooling layer, where the pooling layer is used to process the data output by the convolutional layer to remove redundant information and perform a pooling operation; constructing a fusion layer, where the fusion layer is used to perform feature fusion on the data output by the pooling layer; constructing an output layer, where the output layer is used to output a preliminary analysis result and update the weights and biases of the data using the backpropagation algorithm.

[0074] In this solution, the core of the convolutional neural network is to continuously use convolutional operations to extract the features of the input data. The main structure of the convolutional neural network in this solution includes four layers: a convolutional layer, a pooling layer, a fusion layer, and an output layer. Of course, the convolutional neural network can also be improved on the basis of this solution by adding other structures. The convolutional layer can effectively capture spatial local information by extracting data features through convolutional operations, which helps the model better understand the structure of the input data. The pooling layer can reduce the computational amount of the model, lower the complexity of the model, and improve the computational efficiency of the model. The fusion layer can perform feature fusion on feature information at different levels, which helps improve the model's ability to understand the input data. The output layer can output a specific structure to implement the classification or regression function of the model.

[0075] Specifically, Figure 4 as shown Figure 4 introduces the combination process of the structure of the hysteresis analysis model and the CSO algorithm.

[0076] In the convolutional layer, sparse connection and weight sharing are achieved through convolutional kernels. This solution selects a two-channel convolutional network, so there are two different convolutional kernels, and convolutional operations without changing the weights are performed on the main channel and the auxiliary channel respectively; generating output data for two channels for the next layer, and the output results of the two convolutional kernels are placed in different channels respectively. The relevant formulas are as follows: y main = a main (W main *H + b main ), y aux = a aux (W aux *V + b aux ), where * represents the convolution operator; y main and y aux represent the main channel matrix and the auxiliary channel after the convolution operation; W represents the convolutional kernel; b represents the bias coefficient, a is a learnable parameter used to control the ratio of the features of the two paths; H represents the magnetic induction intensity as the main input channel, and V represents the factor features including temperature T, frequency f, and bias magnetic field Hdc, that is, V(t) = {f(t), T(t), Hdc(t)} as the auxiliary input channel.

[0077] In the pooling layer, in order to extract the most representative features, a pooling layer is generally added after the convolutional layer. The pooling layer can further eliminate redundant information and improve the robustness and generalization ability of the overall network. The pooling layer in the convolutional neural network of this solution uses the max-pooling method, applying max-pooling once, with a pooling window size of 1×2 and a stride of 2 each time. Its mathematical description is as follows:

[0078]

[0079] z i is the feature matrix after max-pooling; z i-1 is the feature matrix output by the previous convolutional layer; W represents the window size of the pooling operation; S represents the stride of the pooling operation, n represents the nth pooling window, and L represents the length of the input sequence.

[0080] In the fusion layer, after the convolutional kernel pooling operation, the features of the main channel and the auxiliary channel need to be fused. The operation of the fusion layer is as follows: F fused = Concat(z main , z aux ), where F fused represents the fused features, z main and z aux represent the output results of the pooling layers of the main channel and the auxiliary channel, and Concat represents the concatenation operation in the feature dimension.

[0081] In the output layer (which can also be called the fully connected layer), the fully connected layer is generally one of the last few layers in the hidden layer of the convolutional neural network. The matrix after convolution and pooling is flattened into a one-dimensional feature vector as the input of the fully connected layer. The neurons in the layer are fully connected to all neurons in the previous layer, and its representation is as follows:

[0082] B = f(W fc ·F fused + b fc ), where W fc is the weight matrix of the fully connected layer, b fc is the bias of the output layer, and f is the activation function.

[0083] The activation function is the key to introducing non-linear transformation in the neural network. In this solution, the ReLU activation function is used in the output layer, and its expression is as follows: ReLU = max(0, x).

[0084] Finally, since the output magnetic induction intensity B is related to the previous state and the current state, each input sample includes V(t) at the current moment and V(t-1) at the previous moment. The corresponding input-output relationship can be simply expressed as the following formula: B(t) = f(Pool(W * [V(t), V(t-1)] + b)), where * represents the convolution operation, which can reduce the dimension of the input data, Pool represents the pooling operation, and f is the ReLU activation function.

[0085] In this solution, the performance index used to quantify the error between the quantization model and the experimental measurement is the following normalized root mean square error. The weights and biases in the network are updated by backpropagation through the Adam optimizer. The specific formula is as follows:

[0086]

[0087] In the formula, y represents the sample detection value, represents the sample prediction value, and N represents the total number of samples.

[0088] As described above, based on the efficient feature extraction effect of the convolutional neural network, this solution proposes a modeling method for the transformer hysteresis analysis model that includes multi-factor correlations. This method divides the input features into main features and auxiliary features, and uses two convolutional kernels in the convolutional network for parallel feature extraction and fusion. This method can comprehensively consider the effects of temperature, frequency, and DC bias on the hysteresis curve, which is of great help to the modeling of the magnetic characteristics of transformers.

[0089] In the specific implementation process, before constructing the hysteresis analysis model, the above method further includes the following steps: The first step is to perform horizontal crossover and vertical crossover on the population. Among them, horizontal crossover means the arithmetic crossover between different particles of the population in the same dimension, and vertical crossover means the arithmetic crossover of the same particle of the population in different dimensions; The second step is to select the optimal particle after crossover using the greedy algorithm; The third step is to decay the probabilities of horizontal and vertical crossovers and perform iterations until the preset number of iterations is reached and the iteration stops; The fourth step is to output the preliminary analysis result when the preset conditions are met, and when the above preset conditions are not met, re-execute the above first step, the above second step, and the above third step. Among them, the above preset conditions at least include that the number of iterations is greater than the above preset number of iterations.

[0090] In this solution, for the neural network models commonly used today, each hyperparameter is selected by experience or grid search, which easily leads to local optimality of the model, resulting in overfitting of the prediction model. Therefore, this solution improves the cross-swarm optimization (CSO) algorithm and optimizes the model parameters based on the improved CSO algorithm. Compared with the traditional CSO algorithm, a longitudinal and transverse cross probability decay strategy is introduced, enabling individuals to have a greater information search ability in the early stage of iteration and to converge quickly in the later stage.

[0091] Specifically, the objective function of the above embodiment of this solution is:

[0092]

[0093] The optimization objective is to minimize the overall error between the hysteresis loop predicted by the model and the measured value. Among them, N is the number of sampling points, and B i are the predicted value obtained by the i-th particle from the convolutional network and its corresponding theoretical value, respectively.

[0094] Execute the cross-swarm optimization algorithm, as Figure 4 and Figure 5 shown, including the following steps:

[0095] ① Execute the longitudinal cross strategy and the transverse cross strategy;

[0096] ② Calculate the fitness of the individuals after crossing and use the greedy algorithm for selection;

[0097] ③ Decay the longitudinal and transverse cross probabilities;

[0098] ④ Determine whether the constraints are satisfied. If so, output the identification result; if not, return to ①.

[0099] During each generation of iteration of the cross-swarm optimization algorithm, two cross methods, namely transverse cross and longitudinal cross, are performed, so that some dimensions of the population that fall into local optimality have the opportunity to jump out of the iteration. The solution obtained after each cross of the cross-swarm optimization algorithm is called the moderate solution. By introducing a competition operator, these two cross methods are organically combined. After each cross operation, it enters the competition operator and competes with the parent generation. Only the examples that are better than the parent generation will be retained for the next iteration, and the obtained solution is the dominant solution.

[0100] As described above, based on the good performance of the cross-swarm optimization algorithm in optimization, this solution proposes to use the cross-swarm optimization algorithm with the longitudinal and transverse cross probability decay strategy to optimize the number of iterations, the size of the convolutional kernel, and the learning rate in the convolutional neural network. During the optimization process, the cross probability is gradually reduced according to a certain rule, which can find the optimal model parameters faster and better, and further improve the fitting accuracy of the model.

[0101] In some embodiments, the above-mentioned first step can be specifically implemented through the following steps: randomly pair the particles in the population, and use the horizontal crossover formula to perform horizontal crossover on any two paired particles. Among them, the above-mentioned horizontal crossover formula is:

[0102]

[0103] represents a particle after horizontal crossover, r 1 represents a random number within the first range, represents a particle before horizontal crossover, represents another particle after horizontal crossover, c 1 represents a random number within the second range, and the above-mentioned first range is smaller than the second range, represents another particle after horizontal crossover, r 2 represents the random number within the above-mentioned first range, c 2 represents the random number within the above-mentioned second range; update any two dimensions of a particle in the population, and use the vertical crossover formula to perform vertical crossover on the particle. Among them, the above-mentioned vertical crossover formula is:

[0104]

[0105] represents the offspring particle after vertical crossover, r represents the random number within the above-mentioned first range, SM ij1 represents the parent particle of one dimension, SM ij2 represents the parent particle of another dimension.

[0106] In this solution, the characteristic information of different particles can be combined through horizontal crossover and vertical crossover to generate new particles, thereby increasing the diversity and adaptability of the population. Through horizontal crossover and vertical crossover, new genetic information can be introduced into the population, thereby promoting the evolution of the population and the improvement of adaptability.

[0107] Specifically, for the horizontal crossover strategy, first randomly pair the individuals in the population (i.e., the above-mentioned particles), and perform horizontal crossover on the two paired individuals with a certain probability. The crossover formula is the above-mentioned horizontal crossover formula. is composed of the d-th dimension individual generated after horizontal crossover, is composed of the d-th dimension individual generated after horizontal crossover, r 1 is a random number within (0, 1), r 2 is a random number within (0, 1), c 1is a random number within the range of (-1, 1), c 2 is a random number within the range of (-1, 1).

[0108] In general crosswise intersections, the probability of horizontal intersections depends on the number of horizontal intersection points and the total number of intersection points. Suppose there is a 10x10 grid, and the total number of crosswise intersection points is (10 - 1) x (10 - 1) = 81. Among them, the number of horizontal intersection points is 9 x 10 = 90 (there is one horizontal intersection point in each row), so the probability of horizontal intersection is 90 / 81 ≈ 1.11.

[0109] For example, if there is a 15x15 grid, the total number of crosswise intersection points is (15 - 1) x (15 - 1) = 196. Among them, the number of horizontal intersection points is 14 x 15 = 210 (there is one horizontal intersection point in each row), so the probability of horizontal intersection is 210 / 196 ≈ 1.07.

[0110] Specifically, for vertical crossover, each pair of dimensions of each individual is updated with a certain probability, and other dimensions remain unchanged. Since the parameter meanings represented by different dimensions are different, normalization is required to unify the dimensions before crossover. Crossover is performed according to the vertical crossover formula, is SM ij1 and SM ij2 For the offspring generated by vertical crossover, r is a random number within the range of (0, 1). The generated offspring are compared with the parent individuals, and the better individuals are retained.

[0111] Vertical crossover is one of the commonly used crossover operations in genetic algorithms, and its update probability usually takes a value around 0.5, that is, the crossover operation is performed with a 50% probability.

[0112] For example: Suppose there are two parent individuals A = [1, 2, 3, 4, 5] and B = [6, 7, 8, 9, 10]. When performing vertical crossover, a crossover point can be randomly selected. For example, select the 3rd position, and then exchange the genes of the two parent individuals after this position to obtain new offspring individuals:

[0113] A = [1, 2, 3, 9, 10],

[0114] B = [6, 7, 8, 4, 5],

[0115] In this way, one vertical crossover operation is completed. In practical applications, the selection of the crossover point and the exchange operation can be adjusted according to specific problems and algorithms, but the update probability generally takes a value around 0.5.

[0116] Horizontal crossover refers to the operation of exchanging gene segments at the same positions of two individuals. For example, there are two individuals A and B, and their chromosomes are A = {1, 2, 3, 4, 5} and B = {6, 7, 8, 9, 10}. By performing horizontal crossover at positions 2 and 3, two new individuals A' = {1, 7, 8, 4, 5} and B' = {6, 2, 3, 9, 10} can be obtained.

[0117] Vertical gene exchange refers to the operation of exchanging genes at different positions of two individuals. For example, there are two individuals A and B, and their chromosomes are A = {1, 2, 3, 4, 5} and B = {6, 7, 8, 9, 10}. By performing vertical gene exchange at positions 2 and 4, two new individuals A' = {1, 7, 3, 9, 5} and B' = {6, 2, 8, 4, 10} can be obtained.

[0118] In some embodiments, the above-mentioned second step can be specifically implemented through the following steps: using a fitness function to calculate the fitness of the particle at its position; comparing the magnitude relationships among multiple fitness values; using a greedy algorithm to extract the particle with the highest fitness among the above-mentioned fitness values to obtain the optimal particle.

[0119] In this solution, according to the fitness function, the fitness of each example at its position is calculated, that is, the fitness value of the objective function is calculated. The fitness of the parent generation after crossover and the fitness of the offspring are compared, and the particle with the higher fitness is retained, so that the particle always remains at the optimal value, thereby increasing the convergence speed and search effect of the algorithm.

[0120] Using a greedy algorithm to find the optimal particle after crossover in horizontal and vertical crossover means that when performing the crossover operation of the genetic algorithm, the particle with the optimal fitness is selected as the result after crossover. The greedy algorithm is an optimization algorithm that always selects the current optimal solution without considering better solutions that may appear in the future.

[0121] In the genetic algorithm, crossover is an important operation. It generates new offspring individuals by exchanging parts of the chromosomes of two parent individuals. When performing the crossover operation, selecting the particle with the optimal fitness as the result can ensure that the offspring individuals after crossover have better fitness, thereby increasing the convergence speed and search effect of the genetic algorithm.

[0122] For example, suppose there are two parent individuals A and B, and their fitness values are 10 and 8 respectively. When performing the crossover operation, if A is selected as the result after crossover, then the newly generated offspring individuals will have higher fitness, which may be 10 or higher. If B is selected as the result, the fitness of the offspring individuals will be lower than 10, resulting in poor search effect.

[0123] Therefore, by using the greedy algorithm in the vertical and horizontal crossover to find the optimal particle after crossover, the search effect and convergence speed of the genetic algorithm can be improved, enabling the algorithm to find the global optimal solution faster.

[0124] In some embodiments, the above-mentioned third step can be specifically implemented through the following steps: obtaining the initial vertical crossover probability and the initial horizontal crossover probability; using the horizontal attenuation formula to attenuate the above-mentioned initial horizontal crossover probability and performing iteration until the above-mentioned preset number of iterations is reached and the iteration stops. The above-mentioned horizontal attenuation formula is:

[0125]

[0126] pc t represents the horizontal crossover probability at the t-th iteration, pc min the minimum horizontal crossover probability, pc 0 represents the above-mentioned initial horizontal crossover probability, T represents the above-mentioned preset number of iterations; using the vertical attenuation formula to attenuate the above-mentioned initial vertical crossover probability and performing iteration until the above-mentioned preset number of iterations is reached and the iteration stops. The above-mentioned vertical attenuation formula is:

[0127]

[0128] hc t represents the vertical crossover probability at the t-th iteration, hc min represents the minimum vertical crossover probability, hc 0 represents the above-mentioned initial vertical crossover probability.

[0129] In this solution, during the vertical and horizontal crossover operation, the probabilities of vertical and horizontal crossover in each iteration are gradually decreased to reduce the influence of the crossover operation gradually and increase the chance of mutation operation, so as to better search for the optimal solution.

[0130] Specifically, in this solution, the initial horizontal crossover probability pc 0 and the initial vertical crossover probability hc 0 can be preset first. The specific values are pc0 = 0.8 and hc0 = 0.2 in this solution. After each iteration, the probabilities of vertical and horizontal crossover are attenuated according to a certain rule, so that the algorithm can perform a large-scale search when the crossover probability is large in the early stage and can converge quickly when the crossover probability is small in the later stage. According to the set number of iterations, repeat the steps and stop the iteration when the preset number of iterations is reached. The preset number of iterations can be any feasible value such as 5, 10, 15, 20, etc.

[0131] By gradually reducing the probability of crossover, the algorithm can be prevented from falling into local optimal solutions, enhancing its global search ability and preventing premature convergence. By gradually decreasing the crossover probability, the algorithm can converge to the vicinity of the optimal solution more quickly, accelerating its convergence speed. Reducing the crossover probability can increase the diversity of the population, facilitating better information exchange and gene combination among individuals in the population and improving the search efficiency of the algorithm.

[0132] For example, assume that in a genetic algorithm, the initial probability of crossover is set to 0.8, and then the probability is gradually decreased by 0.1 in each iteration until it reaches 0.2. This allows for more crossover operations at the beginning of the algorithm, facilitating the rapid convergence of the population. As the number of iterations increases, reducing the crossover probability can increase the diversity of the population, which is beneficial for better searching for the optimal solution.

[0133] In addition, in the solution of this application, the measured values of magnetic induction intensity, magnetic field intensity, temperature, frequency, and DC bias experiment can be obtained at each moment. After preliminary processing, they are input into a convolutional neural network, and then the hyperparameters in the convolutional neural network model are optimized through an improved crossover algorithm to complete the training of the prediction model; finally, the trained prediction model is used to predict the magnetic flux density.

[0134] In addition, in the solution of this application, an appropriate optimization algorithm can be selected to adjust the parameters of the model. Commonly used optimization algorithms include gradient descent method, genetic algorithm, particle swarm optimization, etc. These algorithms can find the parameter combination that best fits the experimental data within a given error range. Secondly, the objective function to be optimized needs to be defined. The objective function is usually the difference between the actual observed value and the model prediction value, and mean square error, correlation coefficient, etc. can be selected as evaluation indicators. Then, the iterative process of parameter optimization is carried out. By continuously adjusting the parameters of the model, the objective function is continuously reduced until the preset stopping condition is met.

[0135] Suppose there is a hysteresis analysis model, and its parameters include saturation magnetic induction intensity, remanent magnetic induction intensity, coercivity, etc. By optimizing these parameters, the prediction results of the model can be made to fit the experimental data as closely as possible.

[0136] First, the gradient descent method can be used to adjust the parameters. Define the objective function as the mean square error, that is, the square difference between the actual observed value and the model prediction value. Then, through the iterative process, the parameters are continuously adjusted to minimize the objective function.

[0137] For example, the experimental data can be input into the hysteresis analysis model to calculate the prediction value of the model. Then, calculate the difference between the actual observed value and the prediction value, and use the gradient descent method to update the parameters until the objective function converges.

[0138] Through the above steps, the parameters of the hysteresis analysis model can be optimized to better fit the experimental data and improve the prediction accuracy of the model.

[0139] In addition, based on the obtained relevant parameters of the transformer, the hysteresis curve can be constructed. By analyzing the hysteresis curve, relevant information such as the hysteresis loop diagram and hysteresis loss of the sample can be obtained. Hysteresis loss refers to the energy loss caused by the magnetic hysteresis phenomenon in the magnetic core of the transformer. On the hysteresis curve of the transformer, the hysteresis loss can be obtained by calculating the area of the hysteresis loop. The area of the hysteresis loop represents the area of the magnetic core where the magnetic flux density changes from the positive maximum value to the negative maximum value and then back to the positive maximum value in one cycle, that is, the hysteresis loss.

[0140] The method for calculating the hysteresis loss is as follows: divide the hysteresis curve into several small rectangles, calculate the area of each small rectangle; add up the areas of all small rectangles to obtain the area of the entire hysteresis loop; multiply the obtained area by the volume and magnetic field frequency of the magnetic core to obtain the hysteresis loss.

[0141] In summary, the purpose of this solution is to overcome the defects of the existing hysteresis analysis model. Based on the accurate simulation of the magnetic characteristics of the transformer under complex conditions, a method for modeling the hysteresis analysis model using a convolutional neural network is proposed. By training a well-trained neural network, the magnetic properties under specific working conditions can be predicted, or the changes in the hysteresis loop and core loss under different conditions can be tracked. Compared with the existing technical solutions, this solution has the following advantages:

[0142] (1) For the first time, a method for modeling the transformer hysteresis analysis model based on a dual-channel convolutional neural network is proposed. The main channel is used to extract the relationship between the magnetic induction intensity B and the magnetic field intensity H, and the auxiliary channel is used to extract the relationship between other external factors and the magnetic induction intensity B. It has stronger feature extraction ability and data integration ability compared with the BP neural network and can effectively fit the B-H curve.

[0143] (2) A multi-factor correlation model considering the input waveform, temperature, and DC bias is proposed for the different operating characteristics of the transformer under complex operating conditions, effectively making up for the problem that the existing hysteresis analysis model cannot fit the hysteresis characteristics under complex conditions.

[0144] (3) Based on the convolutional neural network, an improved cross-in-cross algorithm is used to optimize the relevant network parameters, and it has better optimization efficiency compared with the traditional swarm intelligence optimization algorithm.

[0145] The embodiment of the present application also provides a device for determining the hysteresis of a transformer. It should be noted that the device for determining the hysteresis of the transformer in the embodiment of the present application can be used to execute the method for determining the hysteresis of the transformer provided by the embodiment of the present application. The device is used to implement the above-mentioned embodiments and preferred implementation manners, and those that have been described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that realizes a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0146] The following introduces the device for determining the hysteresis of the transformer provided by the embodiment of the present application.

[0147] Figure 6 It is a structural block diagram of a device for determining the hysteresis of a transformer according to an embodiment of the present application. As Figure 6 shown, the device includes:

[0148] An acquisition unit 10, configured to acquire relevant parameters of the transformer, where the relevant parameters include one or more of magnetic induction intensity, magnetic field intensity, temperature, operating frequency, and DC bias;

[0149] A first construction unit 20, configured to construct a hysteresis analysis model, where the hysteresis analysis model is trained by using a convolutional algorithm and a vertical and horizontal cross algorithm with multiple sets of training data, and each set of training data in the multiple sets of training data includes historical relevant parameters acquired within a historical time period and historical hysteresis analysis results corresponding to the historical relevant parameters;

[0150] A first processing unit 30, configured to input the relevant parameters into the hysteresis analysis model to obtain a hysteresis analysis result corresponding to the relevant parameters.

[0151] Through this embodiment, multiple relevant parameters of the transformer are obtained. The multi-factor combination method can make up for the shortcoming that the existing hysteresis analysis model cannot fit the hysteresis characteristics, and a convolutional neural network method is used for feature training. The convolutional neural network has stronger feature extraction ability and data integration ability than the BP neural network. At the same time, a vertical and horizontal cross algorithm is used for optimization, and the vertical and horizontal cross algorithm has better optimization efficiency. Therefore, this solution uses a combination of convolution and vertical and horizontal cross to construct a hysteresis analysis model, greatly improving the accuracy of the final result.

[0152] In the specific implementation process, the acquisition unit includes a first acquisition module and a preprocessing module. The first acquisition module is used to acquire initial relevant parameters, where the above-mentioned initial relevant parameters are the parameters of the above-mentioned transformer acquired at the initial moment; the preprocessing module is used to preprocess the above-mentioned initial relevant parameters to obtain the above-mentioned relevant parameters, and the preprocessing methods include at least one or more of interpolation processing, normalization processing, denoising processing, and outlier removal processing.

[0153] In this solution, the initial relevant parameters can be preprocessed. Interpolation processing of the parameters can fill in the missing parts of the parameters, making the parameters more complete and avoiding inaccurate analysis results caused by missing parameters. Normalization processing can unify parameters of different dimensions to the same scale, avoiding the influence of parameter differences between different dimensions on the analysis results and improving the comparability and interpretability of the parameters. Denoising processing can remove the noise in the parameters, making the parameters cleaner and more stable, and improving the quality and credibility of the parameters. Outlier removal processing can remove the outliers in the parameters, avoiding the influence of outliers on the analysis results and improving the accuracy and reliability of the parameters.

[0154] In the specific implementation process, the above-mentioned device further includes a second construction unit, a third construction unit, a fourth construction unit, and a fifth construction unit. The second construction unit is used to construct a convolutional layer before constructing the hysteresis analysis model, where the above-mentioned convolutional layer is used to perform a convolutional operation on the input data; the third construction unit is used to construct a pooling layer, where the above-mentioned pooling layer is used to perform redundant information removal processing on the data output by the above-mentioned convolutional layer and perform a pooling operation; the fourth construction unit is used to construct a fusion layer, where the above-mentioned fusion layer is used to perform feature fusion on the data output by the above-mentioned pooling layer; the fifth construction unit is used to construct an output layer, where the above-mentioned output layer is used to output a preliminary analysis result and update the weights and biases of the data using the backpropagation algorithm.

[0155] In this solution, the core of the convolutional neural network is to continuously use convolutional operations to extract the features of the input data. The main structure of the convolutional neural network in this solution includes four layers: a convolutional layer, a pooling layer, a fusion layer, and an output layer. Of course, the convolutional neural network can also be improved on the basis of this solution by adding other structures. The convolutional layer can effectively capture spatial local information by extracting data features through convolutional operations, which helps the model better understand the structure of the input data. The pooling layer can reduce the computational amount of the model, reduce the complexity of the model, and improve the computational efficiency of the model. The fusion layer can perform feature fusion on feature information at different levels, which helps improve the model's ability to understand the input data. The output layer can output specific structures to implement the classification or regression function of the model.

[0156] In the specific implementation process, the above-mentioned device further includes a second processing unit, a third processing unit, a fourth processing unit, and a fifth processing unit. The second processing unit is used to perform the first step before constructing the hysteresis analysis model, performing horizontal crossover and vertical crossover on the population. Among them, horizontal crossover represents the arithmetic crossover between different particles of the population in the same dimension, and vertical crossover represents the arithmetic crossover between the same particle of the population in different dimensions. The third processing unit is used to perform the second step, selecting the optimal particle after crossover using the greedy algorithm. The fourth processing unit is used to perform the third step, attenuating the probability of horizontal and vertical crossover and performing iteration until the preset number of iterations is reached and the iteration stops. The fifth processing unit is used to perform the fourth step, outputting the preliminary analysis result when the preset conditions are met, and re-executing the above-mentioned first step, the above-mentioned second step, and the above-mentioned third step when the above-mentioned preset conditions are not met. Among them, the above-mentioned preset conditions at least include that the number of iterations is greater than the above-mentioned preset number of iterations.

[0157] In this solution, for the currently widely used neural network model, each hyperparameter is selected by experience or grid search, which easily causes the model to have a local optimum, resulting in overfitting of the prediction model. Therefore, this solution improves the horizontal and vertical crossover algorithm, and optimizes the model parameters based on the improved horizontal and vertical crossover algorithm (CSO). Compared with the traditional horizontal and vertical crossover algorithm, a horizontal and vertical crossover probability attenuation strategy is introduced, enabling individuals to have a greater information search ability in the early stage of iteration and being able to converge as soon as possible in the later stage.

[0158] In some embodiments, the second processing unit includes a first processing module and a second processing module. The first processing module is used to randomly pair the particles in the population and perform horizontal crossover on any two paired particles using the horizontal crossover formula. Among them, the above-mentioned horizontal crossover formula is:

[0159]

[0160]

[0161] represents a particle after horizontal crossover, r 1 represents a random number within the first range, represents a particle before horizontal crossover, represents another particle after horizontal crossover, c 1 represents a random number within the second range. The above-mentioned first range is smaller than the above-mentioned second range, represents another particle after horizontal crossover, r 2 represents the above-mentioned random number within the first range, c 2Represents a random number within the above-mentioned second range; the second processing module is used to update any two dimensions of a particle in the population, and perform vertical crossover on the particle using the vertical crossover formula, where the above-mentioned vertical crossover formula is:

[0162]

[0163] Represents the offspring particle after vertical crossover, r represents a random number within the above-mentioned first range, SM ij1 Represents the parent particle of one dimension, SM ij2 Represents the parent particle of another dimension.

[0164] In this solution, the characteristic information of different particles can be combined through horizontal crossover and vertical crossover to generate new particles, thereby increasing the diversity and adaptability of the population. Through horizontal crossover and vertical crossover, new genetic information can be introduced into the population, thereby promoting the evolution of the population and the improvement of adaptability.

[0165] In some embodiments, the third processing unit includes a calculation module, a comparison module, and an extraction module. The calculation module is used to calculate the fitness of the particle at its location using the fitness function; the comparison module is used to compare the magnitude relationship between multiple above-mentioned fitnesses; the extraction module is used to extract the particle with the highest above-mentioned fitness using the greedy algorithm to obtain the optimal particle.

[0166] In this solution, according to the fitness function, the fitness of each example at its location is calculated, that is, the fitness value of the objective function is calculated. The fitness of the parent after crossover and the fitness of the offspring are compared, and the particle with the higher fitness is retained, so that the particle always remains at the optimal value, thereby increasing the convergence speed and search effect of the algorithm.

[0167] In some embodiments, the fourth processing unit includes a second acquisition module, a third processing module, and a fourth processing module. The second acquisition module is used to acquire the initial vertical crossover probability and the initial horizontal crossover probability; the third processing module is used to decay the above-mentioned initial horizontal crossover probability using the horizontal decay formula and perform iteration until the above-mentioned preset iteration times are reached and the iteration stops. The above-mentioned horizontal decay formula is:

[0168]

[0169] pc t Represents the horizontal crossover probability at the t-th iteration, pc min The minimum horizontal crossover probability, pc 0P represents the above-mentioned initial horizontal crossover probability, and T represents the above-mentioned preset number of iterations; the fourth processing module is used to attenuate the above-mentioned initial vertical crossover probability by using the vertical attenuation formula and perform iterations until the above-mentioned preset number of iterations is reached and the iteration stops. The above-mentioned vertical attenuation formula is:

[0170]

[0171] hc t represents the vertical crossover probability at the t-th iteration, hc min represents the minimum vertical crossover probability, hc 0 represents the above-mentioned initial vertical crossover probability.

[0172] In this solution, during the horizontal and vertical crossover operations, the probability of horizontal and vertical crossover in each iteration is gradually reduced to gradually reduce the influence of the crossover operation and increase the chance of the mutation operation, so as to better search for the optimal solution.

[0173] The above-mentioned device for determining the hysteresis of the transformer includes a processor and a memory. The above-mentioned acquisition unit, the first construction unit, the first processing unit, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to implement corresponding functions. The above-mentioned modules are all located in the same processor; or, the above-mentioned each module is located in different processors in any combination form.

[0174] The processor contains a kernel, and the kernel retrieves the corresponding program unit from the memory. One or more kernels can be set, and by adjusting the kernel parameters, the problem that the accuracy of the results obtained by using the BP neural network to study the hysteresis phenomenon in the prior art is poor can be solved.

[0175] The memory may include non-permanent memory in the computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of, for example, read-only memory (ROM) or flash memory (flash RAM), and the memory includes at least one storage chip.

[0176] An embodiment of the present invention provides a computer-readable storage medium. The above-mentioned computer-readable storage medium includes a stored program. When the above-mentioned program runs, it controls the device where the above-mentioned computer-readable storage medium is located to execute the above-mentioned method for determining the hysteresis of the transformer.

[0177] An embodiment of the present invention provides a processor. The above-mentioned processor is used to run a program. When the above-mentioned program runs, it executes the above-mentioned method for determining the hysteresis of the transformer.

[0178] An embodiment of the present invention provides a device, which includes a processor, a memory, and a program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps of a method for determining at least the hysteresis of a transformer.

[0179] The device herein can be a server, a PC, a PAD, a mobile phone, etc.

[0180] A computer program product includes a non-volatile computer-readable storage medium. The non-volatile computer-readable storage medium stores a computer program. When the computer program is executed by a processor, it implements the steps of the method for determining the hysteresis of the transformer in each embodiment of the present application.

[0181] The present application also provides a transformer hysteresis detection system, which includes one or more processors, a memory, and one or more programs. Among them, the one or more programs are stored in the memory and are configured to be executed by the one or more processors. The one or more programs include those for executing any one of the methods for determining the hysteresis of the transformer.

[0182] Obviously, those skilled in the art should understand that the various modules or steps of the above-mentioned present invention can be implemented by a general-purpose computing device. They can be concentrated on a single computing device or distributed on a network composed of multiple computing devices. They can be implemented by program codes executable by the computing device. Thus, they can be stored in a storage device and executed by the computing device. And in some cases, the steps shown or described can be executed in a different order from here, or they can be separately made into individual integrated circuit modules, or multiple modules or steps among them can be made into a single integrated circuit module to implement. In this way, the present invention is not limited to any specific combination of hardware and software.

[0183] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program codes.

[0184] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and combinations of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general purpose computers, special purpose computers, embedded processors, or other programmable data processing devices to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices generate means for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0185] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means that implement the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0186] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are performed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0187] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0188] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of computer-readable media.

[0189] A computer-readable medium includes permanent and non-permanent, removable and non-removable media and can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory, or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD), or other optical storage, magnetic cassette tapes, disk storage, or other magnetic storage devices, or any other non-transitory medium that can be used to store information accessible by a computing device. As defined herein, a computer-readable medium does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0190] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or apparatus comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or apparatus. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising the element.

[0191] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:

[0192] 1), In the method for determining the hysteresis of the transformer of the present application, multiple relevant parameters of the transformer are obtained. The method of combining multiple factors can make up for the shortcoming that the existing hysteresis analysis model cannot fit the hysteresis characteristics. And the convolutional neural network is used for feature training. The convolutional neural network has stronger feature extraction ability and data integration ability compared with the BP neural network. At the same time, the vertical and horizontal crossover algorithm is used for optimization, and the vertical and horizontal crossover algorithm has better optimization efficiency. Therefore, this solution uses the combination of convolution and vertical and horizontal crossover to construct a hysteresis analysis model, so that the accuracy of the finally obtained result is greatly improved.

[0193] 2) The hysteresis determination device of the transformer in this application has obtained multiple relevant parameters of the transformer. The method of combining multiple factors can make up for the shortcoming that the existing hysteresis analysis model cannot fit the hysteresis characteristics. And the convolutional neural network is used for feature training. The convolutional neural network has stronger feature extraction ability and data integration ability compared with the BP neural network. At the same time, the vertical and horizontal crossover algorithm is used for optimization, and the vertical and horizontal crossover algorithm has better optimization efficiency. Therefore, this solution uses the combination of convolution and vertical and horizontal crossover to construct a hysteresis analysis model, greatly improving the accuracy of the final result obtained.

[0194] The above are only the preferred embodiments of this application and are not intended to limit this application. For those skilled in the art, various changes and modifications can be made to this application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included within the protection scope of this application.

Claims

1. A method for determining the hysteresis of a transformer, characterized in that: include: Obtaining relevant parameters of the transformer, wherein the relevant parameters include one or more of magnetic induction intensity, magnetic field intensity, temperature, operating frequency and DC bias; Constructing a hysteresis analysis model, wherein the hysteresis analysis model is obtained by training with a convolution algorithm and a cross-cutting algorithm using a plurality of training data sets, wherein each of the plurality of training data sets includes historical related parameters obtained within a historical time period and historical hysteresis analysis results corresponding to the historical related parameters; Inputting the relevant parameters into the hysteresis analysis model to obtain hysteresis analysis results corresponding to the relevant parameters; Before constructing the hysteresis analysis model, a convolution layer, a pooling layer, a fusion layer and an output layer are constructed. The convolution layer is used to perform convolution operations on the input data and perform convolution operations on the main channel and the auxiliary channel; the pooling layer is used to remove redundant information from the data output by the convolution layer and perform pooling operations; the fusion layer is used to perform feature fusion on the data output by the pooling layer and fuse the features of the main channel and the auxiliary channel; the output layer is used to output preliminary analysis results and update the weights and biases of the data using the back propagation algorithm through the Adam optimizer; Before constructing the hysteresis analysis model, the method also includes: first, performing horizontal and vertical crossover on the population; second, using a greedy algorithm to select the optimal particle after the crossover; The third step is to attenuate the probability of vertical and horizontal crossing and iterate until the preset number of iterations is reached and the iteration is stopped; the fourth step is to output the preliminary analysis results when the preset conditions are met, and to re-execute the first step, the second step and the third step when the preset conditions are not met, and the preset conditions at least include that the number of iterations is greater than the preset number of iterations.

2. The method according to claim 1, characterized in that Get the relevant parameters of the transformer, including: Acquire initial relevant parameters, wherein the initial relevant parameters are parameters of the transformer acquired at an initial moment; The initial correlation parameters are preprocessed to obtain the correlation parameters, wherein the preprocessing method includes at least one or more of interpolation processing, normalization processing, denoising processing and outlier removal processing.

3. The method according to claim 1, characterized in that The first step comprises: The particles in the population are randomly paired, and a horizontal crossover formula is used to perform horizontal crossover on any two paired particles, wherein the horizontal crossover formula is: represents a particle after horizontal crossing, r1 represents a random number in the first range, represents a particle before the lateral crossing, represents another particle after the horizontal crossing, c1 represents a random number within a second range, the first range is smaller than the second range, represents another particle after the lateral crossing, r2 represents a random number within the first range, and c2 represents a random number within the second range; Update any two dimensions of a particle in the population, and use the vertical crossover formula to perform vertical crossover on the particle, where the vertical crossover formula is: represents the offspring particle after vertical crossover, r represents a random number within the first range, SM ij1 Represents a parent particle of one dimension, SM ij2 Represents a parent particle in another dimension.

4. The method according to claim 1, characterized in that The second step comprises: Use the fitness function to calculate the fitness of the particle at its location; Comparing the size relationships among a plurality of the fitness values; A greedy algorithm is used to extract the particle with the highest fitness to obtain the optimal particle.

5. The method according to claim 1, characterized in that The third step comprises: Obtaining initial longitudinal crossing probability and initial transverse crossing probability; The initial horizontal crossover probability is attenuated by using a horizontal attenuation formula, and iteration is performed until the preset number of iterations is reached and the iteration is stopped. The horizontal attenuation formula is: pc t represents the horizontal crossover probability at the tth iteration, pc min Minimum horizontal crossover probability, pc0 represents the initial horizontal crossover probability, T represents the preset number of iterations; The initial longitudinal crossing probability is attenuated by using a longitudinal attenuation formula, and iteration is performed until the preset number of iterations is reached and the iteration is stopped. The longitudinal attenuation formula is: hc t represents the vertical crossover probability at the tth iteration, hc min represents the minimum vertical crossover probability, and hc0 represents the initial vertical crossover probability.

6. A device for determining the hysteresis of a transformer, characterized in that: include: An acquisition unit, used to acquire relevant parameters of the transformer, wherein the relevant parameters include one or more of magnetic induction intensity, magnetic field intensity, temperature, operating frequency and DC bias; A first construction unit is used to construct a hysteresis analysis model, wherein the hysteresis analysis model is obtained by training with a plurality of sets of training data through a convolution algorithm and a vertical and horizontal cross algorithm, wherein each set of training data in the plurality of sets of training data includes historical related parameters obtained in a historical time period and historical hysteresis analysis results corresponding to the historical related parameters; A first processing unit, configured to input the relevant parameters into the hysteresis analysis model to obtain hysteresis analysis results corresponding to the relevant parameters; The device is also used to construct a convolution layer, a pooling layer, a fusion layer and an output layer before constructing a hysteresis analysis model. The convolution layer is used to perform a convolution operation on the input data and perform a convolution operation on the main channel and the auxiliary channel; the pooling layer is used to remove redundant information from the data output by the convolution layer and perform a pooling operation; the fusion layer is used to perform feature fusion on the data output by the pooling layer and fuse the features of the main channel and the auxiliary channel; the output layer is used to output a preliminary analysis result and update the weight and bias of the data using a back propagation algorithm through an Adam optimizer; The device is also used to execute the first step, before building the hysteresis analysis model, to perform horizontal and vertical crossover on the population; execute the second step, use a greedy algorithm to select the optimal particles after the crossover; execute the third step, attenuate the probability of vertical and horizontal crossover, and iterate until the iteration is stopped when a preset number of iterations is reached; execute the fourth step, output the preliminary analysis results if the preset conditions are met, and re-execute the first step, the second step and the third step if the preset conditions are not met, and the preset conditions at least include that the number of iterations is greater than the preset number of iterations.

7. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method for determining the hysteresis of the transformer according to any one of claims 1 to 5 are implemented.

8. A transformer hysteresis detection system, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include a method for determining the hysteresis of the transformer according to any one of claims 1 to 5.

Images