Wireless charging coil type and mutual inductance identification method
By constructing a deep learning model and training a neural network using the pixel values of magnetic field cloud maps, the type of wireless charging coil and mutual inductance can be identified, solving the problem of difficult identification in existing technologies and achieving fast and accurate identification of coil type and mutual inductance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NANJING UNIV OF SCI & TECH
- Filing Date
- 2022-09-21
- Publication Date
- 2026-07-07
AI Technical Summary
Existing technologies struggle to quickly and accurately identify coil types and mutual inductance values in wireless charging systems, requiring extensive calculations of system parameters.
By constructing a deep learning model and training a neural network using the pixel values of magnetic field cloud maps, the model identifies the type of wireless charging coil and mutual inductance, including a coil type identification model and a mutual inductance identification model. Feature extraction is performed using convolutional layers, pooling layers, and fully connected layers, and the model is trained using cross-entropy loss and mean squared error function.
It enables rapid and accurate identification of wireless charging coil type and mutual inductance value, avoiding a large amount of system parameter calculation and improving the accuracy and efficiency of identification.
Smart Images
Figure CN115620269B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of wireless power transmission, specifically relating to a wireless charging coil type and a method for mutual inductance identification. Background Technology
[0002] Deep learning has seen rapid development in computer vision in recent years. Deep learning methods have been widely applied in classification, recognition, and object segmentation. Compared to traditional image processing methods, deep learning methods do not require users to define the captured image features; they are more accurate and faster. They can extract features from images through the self-learning of convolutional and pooling layers in the network. This provides a new approach to solving the nonlinearity problem of relevant parameters in magnetic field cloud image recognition systems in the field of wireless transmission. Summary of the Invention
[0003] The purpose of this invention is to provide a wireless charging coil type and mutual inductance identification method.
[0004] The technical solution for achieving the objective of this invention is a method for identifying the type and mutual inductance of a wireless charging coil, comprising the following steps:
[0005] Step 1, Dataset Creation: Establish models of circular transmitting coil and circular receiving coil, square transmitting coil and square receiving coil, circular transmitting coil and square receiving coil, square transmitting coil and circular receiving coil; obtain magnetic field cloud maps and mutual inductance values of the receiving coil and transmitting coil; construct training dataset.
[0006] Step 2, Coil Type Recognition Model Construction: Using the pixel values of the coil magnetic field cloud map as input and the coil type as output, construct a coil type recognition model;
[0007] Step 3, Construction of coil mutual inductance identification model: Using the pixel values of the coil magnetic field cloud map as input and the transmitting coil and receiving coil as output, construct the coil mutual inductance identification model;
[0008] Step 4, Model Training: Set the loss function and train the coil type recognition model and coil mutual inductance recognition model to identify the type and mutual inductance of wireless charging coils.
[0009] Further, in step 1, the dataset is created, and the specific method is as follows:
[0010] In the finite element simulation software, models of circular transmitting coil and circular receiving coil, square transmitting coil and square receiving coil, circular transmitting coil and square receiving coil, and square transmitting coil and circular receiving coil are established.
[0011] A first square plane with the same size as the transmitting coil is selected at a certain height below the transmitting coil, and a second square plane with the same size as the receiving coil is selected at a certain height above the receiving coil. The magnetic field cloud map of the transmitting coil is obtained on the first square plane, and the magnetic field cloud map of the receiving coil is obtained on the second square plane. At the same time, the mutual inductance value between the transmitting coil and the receiving coil is obtained, and a training dataset is constructed based on this.
[0012] Further, in step 2, the coil type identification model is constructed, specifically using the following method:
[0013] The coil type recognition model takes the magnetic field cloud image pixel values as input. The first layer contains a 3x3 convolutional kernel (128 kernels), with ReLU convolutional layers, normalized batch normalization (BN) layers, and dropout layers as activation functions. The second layer contains a 128x128 convolutional kernel (128 kernels), with ReLU convolutional layers, normalized BN layers, and max pooling layers as activation functions. The third layer contains a 64x64 convolutional kernel (256 kernels), with ReLU convolutional layers, normalized BN layers, and max pooling layers as activation functions. The fourth layer... The first layer contains one 32x32 convolutional kernel with 256 kernels, and uses ReLU convolutional layers, normalized BN layers, and max pooling layers as activation functions. The second layer contains one 16x16 convolutional kernel with 512 kernels, and uses ReLU convolutional layers, normalized BN layers, max pooling layers, and flattening layers as activation functions. The third layer has one output feature with 512 kernels, and uses ReLU fully connected layers, normalized BN layers, and dropout layers as activation functions. The fourth layer outputs two parameters and uses Softmax fully connected layers as activation functions.
[0014] After the output data of five convolutional layers is flattened, it is passed through a fully connected layer with 512 neurons, followed by another fully connected layer with 2 neurons. The probability of the output coil being a circular or square coil is then obtained through the Softmax activation function.
[0015] Further, in step 2, the coil mutual inductance identification model is constructed, and the specific method is as follows:
[0016] The input to the coil mutual inductance identification model is the pixel value of the magnetic field cloud map. The first layer contains a 3*3 convolutional kernel with 128 kernels, and the activation functions are ReLU convolutional layer, normalized BN layer, and Dropout layer. The second layer contains a 128*128 convolutional kernel with 128 kernels, and the activation functions are ReLU convolutional layer, normalized BN layer, and max pooling layer. The third layer contains a 64*64 convolutional kernel with 256 kernels, and the activation functions are ReLU convolutional layer, normalized BN layer, and max pooling layer. The fourth layer contains a 32*32 convolutional kernel with 256 kernels, and the activation functions are ReLU convolutional layer, normalized BN layer, and max pooling layer. The fifth layer contains a 16*16 convolutional kernel with 512 kernels, and the activation functions are ReLU convolutional layer, normalized BN layer, and max pooling layer. The layers are: 1. Flatten layer; 2. Fully connected layer with 1024 output features, 3. Normalized BN layer, 4. Dropout layer; 5. Fully connected layer with 512 output features, 6. Fully connected layer with 128 output features, 7. Fully connected layer with 64 output features, 8. Fully connected layer with 64 output features, 9. Fully connected layer with 1024 output features, 10. Normalized BN layer, 10. Dropout layer; 11. Fully connected layer with Softmax activation function; 2. Fully connected layer with 1 output parameter.
[0017] The output data from five convolutional layers goes through an unfolded layer, then through fully connected layers with 1024, 512, 128, and 64 nodes respectively, followed by a fully connected layer with 1 neural node. The mutual inductance value between the transmitting and receiving coils is then output through the Softmax activation function.
[0018] Further, in step 4, model training, the specific method is as follows:
[0019] During model training, the batch size was set to 10, the training duration was fixed at 30 epochs, the initial learning rate was set to 0.001, and the learning rate was dynamically adjusted. The loss function for the coil recognition model was set to cross-entropy loss, and the loss function for the mutual inductance recognition model was set to mean squared error. Tensorboard was used to monitor the network's loss value and accuracy.
[0020] A wireless charging coil type and mutual inductance identification system is provided, which, based on the aforementioned wireless charging coil type and mutual inductance identification method, enables the identification of wireless charging coil type and mutual inductance.
[0021] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it identifies the type of wireless charging coil and the mutual inductance based on the described wire charging coil type and mutual inductance identification method.
[0022] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it identifies the type of wireless charging coil and the mutual inductance based on the aforementioned wire charging coil type and mutual inductance identification method.
[0023] Compared with the prior art, the significant advantage of this invention is that after obtaining the magnetic field cloud map, relevant parameters in the wireless power transmission system can be obtained by training a neural network, such as coil type, mutual inductance between the transmitting coil and the receiving coil, avoiding a large number of system parameter calculations and enabling rapid and accurate identification of coil type and mutual inductance value. Attached Figure Description
[0024] Figure 1 This is a flowchart illustrating the identification method for wireless charging coil types and mutual inductance.
[0025] Figure 2 This is a schematic diagram of the network structure of the coil type identification model.
[0026] Figure 3 This is a schematic diagram of the network structure of the coil mutual inductance identification model.
[0027] Figure 4 This is a curve showing the training accuracy of circle type recognition.
[0028] Figure 5 This is a graph showing the training loss value for coil type recognition.
[0029] Figure 6 This is a graph showing the accuracy of coil type identification tests.
[0030] Figure 7 This is a graph showing the loss value of the coil type identification test.
[0031] Figure 8 This is a graph showing the accuracy curve of coil mutual inductance recognition training.
[0032] Figure 9 This is a graph showing the training loss value of coil mutual inductance recognition.
[0033] Figure 10 This is a graph showing the accuracy of coil mutual inductance identification tests.
[0034] Figure 11 This is a graph showing the loss value of the coil mutual inductance identification test. Detailed Implementation
[0035] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0036] like Figure 1 The present invention discloses a wireless charging coil type and mutual inductance identification method, comprising the following steps: (1) dataset creation; (2) constructing a coil type identification model; (3) constructing a mutual inductance identification model between the transmitting coil and the receiving coil; (4) training the model, as detailed below:
[0037] Step 1: Dataset Creation
[0038] In finite element simulation software, models of circular and circular transmitting and receiving coils, square transmitting and receiving coils, and circular transmitting and square receiving coils are established. The relevant parameters of the coils are set to vary linearly within a specified range, including coil turn spacing and number of turns. A first square plane with the same dimensions as the transmitting coil is selected at a certain height below the transmitting coil, and a second square plane with the same dimensions as the receiving coil is selected at a certain height above the receiving coil. The magnetic field contour maps of the transmitting and receiving coils are obtained on the first and second square planes. The mutual inductance values of the transmitting and receiving coils are also obtained. After obtaining the magnetic field contour maps of the transmitting and receiving coils, the pixel values of the coil contour maps are calculated, and these pixel values will be used as input to the neural network.
[0039] Step 2: Construct a coil type identification model
[0040] like Figure 2 As shown, the coil type recognition model contains 5 convolutional layers with a kernel size of 3x3. Each convolutional layer uses the ReLU activation function. The data passing through the convolutional layers is flattened, then passed through a fully connected layer with 512 neurons, and finally through another fully connected layer with 2 neurons. The model outputs the probability of recognizing the image as a circular or square coil using the Softmax activation function. The specific model is as follows:
[0041] First layer: The first layer is the input layer, and the input is the pixel value of the magnetic field cloud map.
[0042] The second layer contains a convolutional layer with a kernel size of 3*3 and a number of 128 kernels, an activation function of ReLU, a normalized BN layer, and a Dropout layer.
[0043] The third layer contains a convolutional layer with a kernel size of 128*128 and a number of kernels of 128, with ReLU activation function, a normalized BN layer, and a max pooling layer.
[0044] The fourth layer contains a convolutional layer with a kernel size of 64*64 and a number of 256 kernels, an activation function of ReLU, a normalized BN layer, and a max pooling layer.
[0045] Fifth layer: The fifth layer contains a convolutional layer with a kernel size of 32*32 and 256 kernels, an activation function of ReLU, a normalized BN layer, and a max pooling layer.
[0046] The sixth layer contains a convolutional layer with a kernel size of 16*16 and 512 kernels, an activation function of ReLU, a normalized BN layer, a max pooling layer, and then a flatten layer to flatten the data.
[0047] Layer 7: Layer 7 contains a fully connected layer with 512 output features and ReLU activation function, a normalized BN layer, and a Dropout layer.
[0048] The eighth layer: The eighth layer is a fully connected layer with two output parameters and the activation function is Softmax.
[0049] Ninth layer: Outputs the type of image recognized by the neural network.
[0050] Step 3: Construct a mutual inductance identification model between the transmitting and receiving coils.
[0051] Coil mutual inductance identification model such as Figure 3 As shown, adjustments are made to the network based on the coil type recognition model. The convolutional layers are not adjusted. After passing through 5 convolutional layers, the data is flattened and then input into 5 fully connected layers with the number of layers being 1024, 512, 128, 64, and 64, respectively. The coil mutual inductance value is output through the Softmax activation function.
[0052] Layer 7: Layer 7 contains a fully connected layer with 1024 output features and ReLU activation function, a normalized BN layer, and a Dropout layer.
[0053] The eighth layer contains a fully connected layer with 512 output features and ReLU activation function, a normalized BN layer, and a Dropout layer.
[0054] Layer 9: Layer 9 contains a fully connected layer with 128 output features and ReLU activation function, a normalized BN layer, and a Dropout layer.
[0055] Layer 10: Layer 10 contains a fully connected layer with 64 output features and ReLU activation function, a normalized BN layer, and a Dropout layer.
[0056] The eleventh layer contains a fully connected layer with 64 output features and ReLU activation function, a normalized BN layer, and a Dropout layer.
[0057] Twelfth layer: Output the mutual inductance value between the transmitting coil and the receiving coil.
[0058] Step 4: Train the model
[0059] During model training, the batch size was set to 10, and training was performed for a fixed 30 epochs. The initial learning rate was set to 0.001, and the learning rate was dynamically adjusted. The cross-entropy loss function was used for the coil recognition model training, and the mean squared error function was used for the mutual inductance recognition model training. Tensorboard was used to monitor the network's loss value and accuracy.
[0060] The present invention also proposes a wireless charging coil type and mutual inductance identification system, which realizes the identification of wireless charging coil type and mutual inductance based on the aforementioned wireless charging coil type and mutual inductance identification method.
[0061] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it identifies the type of wireless charging coil and the mutual inductance based on the described wire charging coil type and mutual inductance identification method.
[0062] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it identifies the type of wireless charging coil and the mutual inductance based on the aforementioned wire charging coil type and mutual inductance identification method.
[0063] Example
[0064] To verify the effectiveness of the present invention, the following experiment was conducted.
[0065] Step 1: Dataset Creation
[0066] The magnetic field cloud maps and mutual inductance values of the transmitting and receiving coils of the wireless charging system were collected. The parameters of the transmitting and receiving coils are as follows: Transmitting coil parameters include: circular coil, radius 100mm, wire diameter 1mm, fixed number of turns 10, fixed turn spacing 4mm, self-inductance 8.06μH, and equivalent resistance 72.8mΩ. Receiving coil parameters include: circular or square coil, radius 100mm, wire diameter 1mm, number of turns ranging from 5 to 20, parameterization step size of 1 turn, turn spacing of 2 to 5mm, and parameterization step size of 0.5mm.
[0067] The distance between the receiving coil and the transmitting coil was fixed at 50mm. 111 effective combinations of circular and square coils were made. Magnetic field cloud maps were collected at 44 locations at distances of 2mm, 5mm, 8mm, 10mm, 12mm, 15mm, 18mm, 20mm, 22mm, 25mm, 28mm, 30mm, 32mm, 35mm, 38mm, 40mm, 42mm, 45mm, 48mm, and 50mm from the receiving and transmitting coils. A total of 4440 effective magnetic field cloud maps were collected for the circular coils and 4440 for the square coils, for a total of 8880 maps.
[0068] The dataset was randomly divided into training and test sets in an 8:2 ratio, and then the images were labeled. The dataset was divided as follows: circular coils: 3552 training images, 880 test images; square coils: 3552 training images, 880 test images.
[0069] The dataset for the mutual inductance recognition model is as follows:
[0070] The receiving coil was set to a radius of 100mm, a wire diameter of 1mm, and a number of turns ranging from 5 to 20. The parametric simulation step size was 1, the turn spacing was 2 to 5mm, and the parametric simulation step size was 0.5mm. The distance between the receiving coil and the transmitting coil was fixed at 50mm. Magnetic field cloud maps were collected at 20 locations at distances of 2mm, 5mm, 8mm, 10mm, 12mm, 15mm, 18mm, 20mm, 22mm, 25mm, 28mm, 30mm, 32mm, 35mm, 38mm, 40mm, 42mm, 45mm, 48mm, and 50mm from the receiving coil.
[0071] Because the information contained in a single magnetic field cloud image may be similar, leading to difficulties in network convergence, to ensure reasonable distance, four cloud images were randomly selected from 20 locations for combination. This resulted in 4845 combined magnetic field cloud images collected under the same combination. A total of 111 effective coil size combinations were identified, yielding 2220 effective magnetic field cloud images and 537795 combined magnetic field cloud images. Considering the excessive data volume, only 36 coil combinations (7992 images) were selected under the same coil size combination. Simultaneously, the output mutual inductance value was set to uH to prevent network convergence difficulties due to excessively large or small mutual inductance values.
[0072] Step 2: Construct a convolutional neural network model
[0073] Coil type identification model such as Figure 2As shown, the method involves setting the convolutional kernel size to 3x3, using the ReLU activation function for each convolutional layer, followed by a normalized BN layer, and then a Dropout layer after the convolutional layers. The data from the convolutional layers is flattened, passed through a fully connected layer with 512 neurons, and finally connected to another fully connected layer with 2 neurons. The Softmax activation function is then used to output the probability of recognizing the image as a circular or square coil. The specific method is as follows:
[0074] First layer: The first layer is the input layer, and the input is the pixel value of the magnetic field cloud map.
[0075] The second layer contains a convolutional layer with a kernel size of 3*3 and a number of 128 kernels, an activation function of ReLU, a normalized BN layer, and a Dropout layer.
[0076] The third layer contains a convolutional layer with a kernel size of 128*128 and a number of kernels of 128, with ReLU activation function, a normalized BN layer, and a max pooling layer.
[0077] The fourth layer contains a convolutional layer with a kernel size of 64*64 and a number of 256 kernels, an activation function of ReLU, a normalized BN layer, and a max pooling layer.
[0078] Fifth layer: The fifth layer contains a convolutional layer with a kernel size of 32*32 and 256 kernels, an activation function of ReLU, a normalized BN layer, and a max pooling layer.
[0079] The sixth layer contains a convolutional layer with a kernel size of 16*16 and 512 kernels, using ReLU activation, a normalized BN layer, and a max pooling layer. A flattening layer is then added to flatten the data.
[0080] Layer 7: Layer 7 contains a fully connected layer with 512 output features and ReLU activation function, a normalized BN layer, and a Dropout layer.
[0081] The eighth layer: The eighth layer is a fully connected layer with two output parameters and the activation function is Softmax.
[0082] Ninth layer: Outputs the type of image recognized by the neural network.
[0083] Step 3: Construct a mutual inductance identification model between the transmitting and receiving coils.
[0084] Coil mutual inductance identification model such as Figure 3As shown, based on the aforementioned network with recognition capabilities, the convolutional layers remain unchanged. After five convolutional layers, there is one unfolded layer and five fully connected layers. The coil mutual inductance value is then output through the Softmax activation function. Layers seven through eleven of the mutual inductance recognition network are configured as follows:
[0085] Layer 7: Layer 7 contains a fully connected layer with 1024 output features and ReLU activation function, a normalized BN layer, and a Dropout layer.
[0086] The eighth layer contains a fully connected layer with 512 output features and ReLU activation function, a normalized BN layer, and a Dropout layer.
[0087] Layer 9: Layer 9 contains a fully connected layer with 128 output features and ReLU activation function, a normalized BN layer, and a Dropout layer.
[0088] Layer 10: Layer 10 contains a fully connected layer with 64 output features and ReLU activation function, a normalized BN layer, and a Dropout layer.
[0089] The eleventh layer contains a fully connected layer with 64 output features and ReLU activation function, a normalized BN layer, and a Dropout layer.
[0090] Step 4: Train the model
[0091] During model training, the batch size was set to 10, and training was performed for a fixed 30 epochs. The initial learning rate was set to 0.001, and the learning rate was dynamically adjusted. The cross-entropy loss function was used for the coil recognition model training, and the mean squared error function was used for the mutual inductance recognition model training. Tensorboard was used to monitor the network's loss value and accuracy.
[0092] After 20 training iterations, the network had 9,690,370 parameters, achieving an accuracy of over 98% and a loss value below 0.05. The output results for coil type recognition are as follows: Figure 4 , Figure 5 , Figure 6 and Figure 7 As shown, the output results of identifying the mutual inductance of the coils are respectively as follows: Figure 8 , Figure 9 , Figure 10 and Figure 11 As shown. In summary, this invention improves the generalization ability of identifying coil types and mutual inductance, providing new ideas for the research and engineering application of wireless power transfer methods.
[0093] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0094] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
Claims
1. A method for identifying the type and mutual inductance of a wireless charging coil, characterized in that, Includes the following steps: Step 1, Dataset Creation: Establish models of circular transmitting coil and circular receiving coil, square transmitting coil and square receiving coil, circular transmitting coil and square receiving coil, square transmitting coil and circular receiving coil; obtain magnetic field cloud maps and mutual inductance values of the receiving coil and transmitting coil; construct training dataset. Step 2, Coil Type Recognition Model Construction: Using the pixel values of the coil magnetic field cloud map as input and the coil type as output, construct a coil type recognition model; Step 3, Construction of coil mutual inductance identification model: Using the pixel values of the coil magnetic field cloud map as input and the transmitting coil and receiving coil as output, construct the coil mutual inductance identification model; Step 4, Model Training: Set the loss function and train the coil type recognition model and coil mutual inductance recognition model to identify the type and mutual inductance of wireless charging coils; in: Step 1, Dataset Creation, the specific method is as follows: In the finite element simulation software, models of circular transmitting coil and circular receiving coil, square transmitting coil and square receiving coil, circular transmitting coil and square receiving coil, and square transmitting coil and circular receiving coil are established. A first square plane with the same size as the transmitting coil is selected at a certain height below the transmitting coil, and a second square plane with the same size as the receiving coil is selected at a certain height above the receiving coil. The magnetic field cloud map of the transmitting coil is obtained in the first square plane, and the magnetic field cloud map of the receiving coil is obtained in the second square plane. At the same time, the mutual inductance values of the transmitting coil and the receiving coil are obtained, and a training dataset is constructed accordingly. Step 2, constructing the coil type identification model, the specific method is as follows: The coil type recognition model takes the magnetic field cloud image pixel values as input. The first layer contains a 3x3 convolutional kernel (128 kernels), with ReLU convolutional layers, normalized batch normalization (BN) layers, and dropout layers as activation functions. The second layer contains a 128x128 convolutional kernel (128 kernels), with ReLU convolutional layers, normalized BN layers, and max pooling layers as activation functions. The third layer contains a 64x64 convolutional kernel (256 kernels), with ReLU convolutional layers, normalized BN layers, and max pooling layers as activation functions. The fourth layer... The first layer contains one 32x32 convolutional kernel with 256 kernels, and uses ReLU convolutional layers, normalized BN layers, and max pooling layers as activation functions. The second layer contains one 16x16 convolutional kernel with 512 kernels, and uses ReLU convolutional layers, normalized BN layers, max pooling layers, and flattening layers as activation functions. The third layer has one output feature with 512 kernels, and uses ReLU fully connected layers, normalized BN layers, and dropout layers as activation functions. The fourth layer outputs two parameters and uses Softmax fully connected layers as activation functions. After the output data of five convolutional layers is flattened, it is passed through a fully connected layer with 512 neurons, followed by another fully connected layer with 2 neurons. The Softmax activation function is then used to output the probability that the coil is a circular or square coil. The specific method for constructing the coil mutual inductance identification model is as follows: The input to the coil mutual inductance identification model is the pixel value of the magnetic field cloud map. The first layer contains a 3*3 convolutional kernel with 128 kernels, and the activation functions are ReLU convolutional layer, normalized BN layer, and Dropout layer. The second layer contains a 128*128 convolutional kernel with 128 kernels, and the activation functions are ReLU convolutional layer, normalized BN layer, and max pooling layer. The third layer contains a 64*64 convolutional kernel with 256 kernels, and the activation functions are ReLU convolutional layer, normalized BN layer, and max pooling layer. The fourth layer contains a 32*32 convolutional kernel with 256 kernels, and the activation functions are ReLU convolutional layer, normalized BN layer, and max pooling layer. The fifth layer contains a 16*16 convolutional kernel with 512 kernels, and the activation functions are ReLU convolutional layer, normalized BN layer, and max pooling layer. The layers are:
1. Flatten layer; 2. Fully connected layer with 1024 output features, 3. Normalized BN layer, 4. Dropout layer; 5. Fully connected layer with 512 output features, 6. Fully connected layer with 128 output features, 7. Fully connected layer with 64 output features, 8. Fully connected layer with 64 output features, 9. Fully connected layer with 1024 output features, 10. Normalized BN layer, 10. Dropout layer; 11. Fully connected layer with Softmax activation function; 2. Fully connected layer with 1 output parameter. The output data from five convolutional layers goes through an unfolded layer, then through fully connected layers with 1024, 512, 128, and 64 nodes respectively, and then through a fully connected layer with 1 neural node. The mutual inductance value between the transmitting and receiving coils is then output through the Softmax activation function. Step 4, model training, the specific method is as follows: During model training, the batch size was set to 10, the training duration was fixed at 30 epochs, the initial learning rate was set to 0.001, and the learning rate was dynamically adjusted. The loss function for the coil recognition model was set to cross-entropy loss, and the loss function for the mutual inductance recognition model was set to mean squared error. Tensorboard was used to monitor the network's loss value and accuracy.
2. A wireless charging coil type and mutual inductance identification system, characterized in that, Based on the wired charging coil type and mutual inductance identification method described in claim 1, the identification of wireless charging coil type and mutual inductance is realized.
3. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it identifies the type of wireless charging coil and the mutual inductance based on the wire charging coil type and mutual inductance identification method of claim 1.
4. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it identifies the type of wireless charging coil and the mutual inductance based on the wire charging coil type and mutual inductance identification method of claim 1.