CN111987779A - MC-WPT system load and mutual inductance identification model, method and system based on TensorFlow - Google Patents

Abstract

The invention relates to the technical field of MC-WPT, and specifically discloses a TensorFlow-based MC-WPT system load and mutual inductance identification model, method and system. The model is based on the TensorFlow deep learning framework and adopts a neural network model, so that the MC-WPT system is integrated into the MC-WPT system. The load and mutual inductance identification problem is equivalent to a nonlinear equation solving problem, which is then transformed into a deep learning nonlinear fitting problem, and the training set is used to train the model for tens of thousands of times, and finally the MC‑ WPT system load and mutual inductance identification model. On the whole, the present invention can realize online simultaneous identification of load and mutual inductance by training the model off-line and importing the trained model into the microcontroller. The identification speed is fast and the accuracy is high, which is beneficial to the real-time control of the system, and the cost is low and easy to use. Realization is conducive to engineering popularization and application.

Description

Load and mutual inductance identification model, method and method of MC-WPT system based on TensorFlow system

technical field

The invention relates to the technical field of MC-WPT (magnetic field coupled wireless power transmission), in particular to a TensorFlow-based MC-WPT system load and mutual inductance identification model, method and system.

Background technique

With the development of economy and society, the use of traditional wire power supply for mobile electrical equipment, such as rail trains, mobile hoisting equipment, household appliances, rotating machinery and other equipment, will affect its flexibility, and will increase in some special environments. The hidden danger of electricity safety has brought challenges to the practical application of the project. The emergence of wireless power transfer (WPT, Wireless PowerTransfer) technology provides a safe, environmentally friendly, convenient and easy-to-maintain power supply method, which has attracted the attention and research of many scholars at home and abroad, and jointly promote the continuous development of this new power supply method. Among them, Magnetic Coupling Wireless Power Transfer (MC-WPT, Magnetic Coupling Wireless Power Transfer) technology is one of the most concerned technologies.

In some practical applications of the MC-WPT system, such as the wireless charging/power supply system for electric vehicles, the change of the relative position between the energy transmitter and the energy receiver of the system will bring about changes in the mutual inductance value, and different energy receiver devices also The system load will change, and these factors will cause the load and mutual inductance to change when the electric vehicle performs wireless charging/power supply, thereby affecting the energy transmission efficiency and power transmission capability of the system. Therefore, in the MC-WPT system, when the load and mutual inductance change, the system needs to adjust the control mode of the energy transmitter according to the current situation to achieve the optimal efficiency tracking or constant voltage output of the system, and the identification of the load and mutual inductance parameters is exactly the key issues.

The current load and mutual inductance identification is mainly carried out by establishing an identification model based on the steady state characteristics of the system or adding additional circuits for auxiliary identification, and using genetic algorithms and other methods to identify the load and mutual inductance. Among the existing load and mutual inductance identification methods, some methods can only perform single-parameter identification of the load or mutual inductance; some methods have the problem of slow online identification speed, which is not conducive to real-time control of the system; some methods have high implementation costs and high accuracy. lower issues.

SUMMARY OF THE INVENTION

The invention provides a TensorFlow-based MC-WPT system load and mutual inductance identification model, method and system, and the technical problem solved is that the current load and mutual inductance identification method of the MC-WPT system cannot realize dual parameter identification at the same time, the identification speed is fast, Low cost and high recognition accuracy.

In order to solve the above technical problems, the present invention proposes a MC-WPT system load and mutual inductance identification model based on TensorFlow, and its generating steps include:

S1. Build a fully connected neural network model based on the TensorFlow framework;

S2. establish the COSMOL and Simulink simulation model of MC-WPT system, obtain the input current value of multiple groups of described MC-WPT system, the simulation data of transmission distance between coils, and described simulation data is divided into training set and test set;

S3. Input the training set into the fully connected neural network model for model training, and continuously optimize the parameters in the fully connected neural network model according to the training error value;

S4. When the training error rate of the fully connected neural network model is as low as the preset error rate, the training is ended, and the trained MC-WPT system load and mutual inductance identification model is obtained.

Preferably, the fully connected neural network model includes an input layer, an output layer, and the 1st to Nth hidden layers that are sequentially fully connected between the input layer and the output layer, N≥1; the 1st to Nth hidden layers The hidden layer has k nodes, k≥2.

Preferably, the nonlinear activation functions of the first to N hidden layers use the Sigmoid activation function in the TensorFlow framework.

Preferably, the parameters optimized in step S3 include a first weight matrix and a first bias matrix acting on the first hidden layer, and a second weight matrix and a second bias acting on the second hidden layer matrices up to the Nth weight matrix and the Nth bias matrix acting on the Nth hidden layer, and the N+1th weight matrix and the N+1th bias matrix acting on the output layer.

Preferably, N=3, k=10.

Preferably, the preset error rate is set not to exceed 2%.

Based on the above-mentioned MC-WPT system load and mutual inductance identification model, the present invention also provides a MC-WPT system load and mutual inductance identification method, comprising the steps of:

X1. Detect the input current value of the current MC-WPT system and the transmission distance between coils;

X2. Input the current input current value and the transmission distance between coils into the MC-WPT system load and mutual inductance identification model, and calculate the corresponding load value and mutual inductance value.

Further, in the step X2, the formula for calculating the MC-WPT system load and mutual inductance identification model is:

和

分别表示所述第1权重矩阵和所述第1偏置矩阵，L₁表示对l₁中每个元素代入激活函数

进行运算后得到的隐藏层输出矩阵；Among them, l ₁ represents the intermediate variable of the first hidden layer, [h I _m ] represents the matrix composed of the transmission distance between the coils of the MC-WPT system and the input current value,

and

respectively represent the first weight matrix and the first bias matrix, and L ₁ represents the activation function for each element in l ₁

The hidden layer output matrix obtained after the operation;

和

分别表示所述第2权重矩阵和所述第2偏置矩阵，L₂表示对l₂中每个元素代入激活函数

进行运算后得到的隐藏层输出矩阵；l ₂ represents the intermediate variable of the second hidden layer,

and

respectively represent the second weight matrix and the second bias matrix, and L ₂ represents that each element in l ₂ is substituted into the activation function

The hidden layer output matrix obtained after the operation;

...;

和

分别表示所述第N权重矩阵和所述第N偏置矩阵，L_N表示对l_N中每个元素代入激活函数

进行运算后得到的隐藏层输出矩阵；l _N represents the intermediate variable of the Nth hidden layer,

and

respectively represent the Nth weight matrix and the Nth bias matrix, and L _N represents that each element in l _N is substituted into the activation function

The hidden layer output matrix obtained after the operation;

和

分别表示所述第N+1 权重矩阵和所述第N+1偏置矩阵，M、R_eq分别表示所述输出层输出的互感值和负载值。l _N+1 represents the intermediate variable of the output layer,

and

represent the N+1th weight matrix and the N+1th bias matrix, respectively, and M and _Req represent the mutual inductance value and load value output by the output layer, respectively.

The present invention also provides a TensorFlow-based MC-WPT system load and mutual inductance identification system, including a controller, a current detection module and a ranging module connected to the controller; the current detection module is used to detect the MC-WPT system The input current value of the LCC circuit topology of the transmitting end is sent to the controller; the ranging module is used to detect the transmission distance between the transmitting coil and the receiving coil in the MC-WPT system and send it to the controller; the The controller is used to install the above-mentioned MC-WPT system load and mutual inductance identification model, and according to the above MC-WPT system load and mutual inductance identification method, calculate the corresponding input current value, load value and mutual inductance value under the transmission distance.

Preferably, the current detection module is a Hall sensor, and the ranging module is an infrared ranging sensor.

The invention provides a MC-WPT system load and mutual inductance identification model based on TensorFlow, which is based on the TensorFlow deep learning framework and adopts a neural network model, so that the load and mutual inductance identification problem of the MC-WPT system is equivalent to the solution of nonlinear equations The problem is transformed into a deep learning nonlinear fitting problem, and the training set is used to train the model for tens of thousands of times, and finally the MC-WPT system load and mutual inductance recognition model with fast recognition speed and high accuracy is obtained;

The invention also provides a TensorFlow-based MC-WPT system load and mutual inductance identification method and system. During specific identification, the model is imported into the microcontroller in advance, and the system transmitter circuit is collected through the current detection module and the ranging module respectively. The current value and transmission distance, and then the controller calls the model for operation to identify the load and mutual inductance, and obtain the corresponding load value and mutual inductance value.

On the whole, the present invention can realize online simultaneous identification of load and mutual inductance by training the model off-line and importing the trained model into the microcontroller. The identification speed is fast and the accuracy is high, which is beneficial to the real-time control of the system, and the cost is low and easy to use. Realization is conducive to engineering popularization and application.

Description of drawings

1 is a main circuit topology diagram of a dual-LCC type MC-WPT system provided by an embodiment of the present invention;

2 is a first equivalent circuit diagram of a dual-LCC-type MC-WPT system provided by an embodiment of the present invention;

3 is a schematic diagram of a coupling mechanism in a dual-LCC-type MC-WPT system provided by an embodiment of the present invention;

4 is an equivalent circuit diagram of a receiving end of a dual-LCC type MC-WPT system provided by an embodiment of the present invention;

5 is a second equivalent circuit diagram of a dual-LCC-type MC-WPT system provided by an embodiment of the present invention;

6 is a network structure diagram of the MC-WPT system load and mutual inductance identification model provided in Embodiment 1 of the present invention;

7 is a simulation model diagram of the coupling mechanism of the dual LCC type MC-WPT system provided in Embodiment 1 of the present invention;

8 is a three-dimensional diagram of the relationship between d, h and M provided in Embodiment 1 of the present invention;

Fig. 9 is the Simulink simulation model of the dual LCC type MC-WPT system provided by Embodiment 1 of the present invention;

10 is a graph showing the variation of training error with the number of training times provided by Embodiment 1 of the present invention;

FIG. 11 is a simulation diagram of a load and mutual inductance identification model of a dual-LCC type MC-WPT system based on the TensorFlow framework provided by Embodiment 3 of the present invention.

Detailed ways

The embodiments of the present invention will be explained in detail below in conjunction with the accompanying drawings. The examples are given only for the purpose of illustration and should not be construed as a limitation of the present invention. The accompanying drawings are only used for reference and description, and do not constitute a limitation on the protection scope of the patent of the present invention. limitation, since many changes may be made in the present invention without departing from the spirit and scope of the invention.

This embodiment is aimed at identifying the mutual load and mutual inductance of the MC-WPT system. First, it is necessary to understand what the MC-WPT system is and why the mutual load and mutual inductance need to be identified.

1. System overview

This embodiment takes a dual-LCC-type MC-WPT system as an example for description, but is not limited to an LCC-type MC-WPT system.

The dual LCC type MC-WPT system has the characteristics of symmetrical structure of the transmitter and receiver, easy matching of the compensation topology network, and the output current has nothing to do with the load. It has good offset resistance and is widely used in wireless charging systems such as electric vehicles.

Figure 1 shows the main circuit topology of the dual LCC type MC-WPT system. At the transmitter end of the system, the system input power E _dc can be directly provided by the DC power supply or obtained after being rectified and filtered by the AC power grid. The switching tubes S1 _- S4 form a high _- frequency inverter circuit, which outputs high-frequency alternating current. After passing through the LCC type resonance compensation network, the transmitting coil transmits the energy to the receiving end. At the receiving end of the system, after the receiving coil picks up the electric energy, it is rectified and filtered by the LCC resonance compensation network to supply power to the load _RL .

其有效值U_m与E_dc关系式如下：At the transmitter end of the system, the DC power source E _dc outputs high-frequency alternating current through the high-frequency inverter circuit, which can be equivalent to a high-frequency alternating voltage source

The relationship between its effective value U _m and E _dc is as follows:

At the receiving end of the system, when the rectifier and filter links are connected in parallel with the load _RL , it can be equivalent to the load resistance R _eq , and the relationship is given as follows:

为发射端高频逆变电路输出的高频交流电压源，L_p为发射端发射线圈自感，L_s为接收端接收线圈自感，R_p为发射线圈内阻， R_s为接收线圈内阻，L_f1、C_f1与C_p构成发射端补偿网络，L_f2、C_f2与C_s构成接收端补偿网络，R_eq为负载等效电阻。M表示发射端线圈与接收端线圈之间的互感。Therefore, the main circuit topology of the dual-LCC-type MC-WPT system shown in Figure 1 can be further equivalent to the system equivalent circuit shown in Figure 2, where

is the high-frequency AC voltage source output by the high-frequency inverter circuit at the transmitting end, L _p is the self-inductance of the transmitting coil at the transmitting end, L _s is the self-inductance of the receiving coil at the receiving end, R _p is the internal resistance of the transmitting coil, and R _s is the internal resistance of the receiving coil. resistance, L _f1 , C _f1 and C _p constitute the transmitter compensation network, L _f2 , C _f2 and C _s constitute the receiver compensation network, Re _eq is the load equivalent resistance. M represents the mutual inductance between the transmitter coil and the receiver coil.

2. System modeling

On the premise that the planes where the transmitting coil and the receiving coil are located are parallel, the mutual inductance value M is only related to the transmission distance h between the coils (can be referred to as the transmission distance h) and the offset distance d after the number of turns and geometric dimensions of the coils are determined, as shown in Figure 3 shown.

The relationship between the mutual inductance value M and d and h can be described by an implicit function as:

M=g(h, d) (3)

接收端电路阻抗表达式为：The detection of the offset distance h is complicated to implement, and can be replaced by other parameters. Considering that the load needs to be identified, it is necessary to detect the current value of the transmitter circuit of the system to identify the load and mutual inductance. In the circuit system, the impedance analysis of the receiver circuit is performed first, as shown in Figure 4, the pickup voltage of the receiver is

The circuit impedance expression of the receiving end is:

Z ₂ =(R _eq +Z _Lf2 )//Z _Cf2 +Z _Cs +Z _Ls +R _s (4)

Among them, Z _Cf2 =1/jωC _f2 , Z _Lf2 =jωL _f2 , ω is the system resonant frequency, according to formula (4), it can be further obtained that the overall impedance of the receiving end circuit is:

Since the load and mutual inductance identification of the system is mainly analyzed by detecting the relevant parameters from the transmitter, the impedance analysis of the transmitter circuit is performed next. The influence of the receiver circuit on the transmitter circuit can be equivalent to impedance:

表示输入电流，I_m表示其有效值，Z_in为系统的输入阻抗。It can be seen from equation (6) that the influence of mutual inductance and load changes on the system is embodied in the change of reflected impedance reflected to the transmitting end. Therefore, in order to further study the influence of reflected impedance changes on the system, the main circuit of the dual-LCC-type MC-WPT system needs to be equivalent to the equivalent circuit model shown in Figure 4. in,

Indicates the input current, I _m represents its effective value, and Z _in is the input impedance of the system.

It can be seen from Figure 5 that the following relationship is established between the input impedance and the current rms value _Im :

The total input impedance Z _in of the system is:

可得Z_in：Bring equation (6) into equation (8) and take

Z _in can be obtained:

Putting Equation (9) into Equation (7), the relationship between the load equivalent resistance R _eq and the mutual inductance M and the input current I _m can be obtained as:

In formula (10), it can be seen that the effective value of the system input current _Im is affected by the load equivalent resistance R _eq and the mutual inductance M, and the combined formula (3) can be obtained as follows:

(M, _Req )=f( _Im ,h) (11)

Therefore, the identification of load and mutual inductance can be realized by detecting the effective value of the input current _Im and the transmission distance h. The identification problem of load and mutual inductance is further transformed into the problem of solving the f function in nonlinear equation (11). In this embodiment, a neural network model is established based on the TensorFlow deep learning framework, and the approximate solution of the f function is performed by means of nonlinear function fitting.

Example 1

Based on the above-mentioned dual-LCC type MC-WPT system, the present embodiment provides a TensorFlow-based MC-WPT system load and mutual inductance identification model, and its generating steps include:

S1. Build a fully connected neural network model based on the TensorFlow framework;

S2. establish the COSMOL and Simulink simulation model of MC-WPT system, obtain the input current value of multiple groups of described MC-WPT system, the simulation data of transmission distance between coils, and described simulation data is divided into training set and test set;

S3. Input the training set into the fully connected neural network model for model training, and continuously optimize the parameters in the fully connected neural network model according to the training error value;

S4. When the training error rate of the fully connected neural network model is as low as the preset error rate, the training is ended, and the trained MC-WPT system load and mutual inductance identification model is obtained.

The fully connected neural network model includes an input layer, an output layer, and the 1st to Nth hidden layers that are sequentially fully connected between the input layer and the output layer, N≥1; the 1st to Nth hidden layers have k nodes, k≥2; the output layer has two fixed nodes.

Among them, the nonlinear activation function of the first to N hidden layers uses the Sigmoid activation function in the TensorFlow framework. The parameters optimized in step S3 include the first weight matrix and the first bias matrix acting on the first hidden layer, and the second weight matrix and the second bias matrix acting on the second hidden layer, until The Nth weight matrix and the Nth bias matrix are applied to the Nth hidden layer, and the N+1th weight matrix and the N+1th bias matrix are applied to the output layer.

In step S4, the preset error rate is set not to exceed 2%. In other embodiments, it can be determined according to actual needs, such as 3%.

In this embodiment, preferably, N=3 and k=10. As shown in FIG. 6 , the first to N hidden layers are represented as hidden layers 1 to 3 in FIG. 6 . In other embodiments, N and k can be determined according to actual requirements.

Hidden layer 1 can be described by the formula as follows:

和

分别表示神经网路的权重矩阵和偏置矩阵；in

and

Represent the weight matrix and bias matrix of the neural network, respectively;

Substituting l ₁ into the Simgoid activation function can get the output of the first hidden layer as:

By analogy, the expressions of l ₂ , L ₂ , l ₃ , L ₃ , and l ₄ can be obtained, and then the load and mutual inductance values can be obtained:

的值，使得神经网络模型的输出尽可能的接近实际值(步骤S3)。为确定上述参数需要一批训练数据来训练模型从而确定模型的参数，本实施例采用仿真软件建立系统仿真模型的方式获取上述模型训练所需数据(步骤S2)。After the model is established, the weight matrix and bias matrix in the model need to be determined

, so that the output of the neural network model is as close to the actual value as possible (step S3). In order to determine the above parameters, a batch of training data is needed to train the model to determine the parameters of the model. In this embodiment, simulation software is used to establish a system simulation model to obtain the data required for the above model training (step S2).

First, the simulation model of the coupling mechanism is established in the COMSOL multiphysics simulation software, as shown in Figure 7, and the simulation parameters are shown in Table 1.

Table 1. Parameters of the simulation model of the coupling mechanism of the dual LCC type MC-WPT system

Through the simulation of the coupling mechanism in the COMSOL multiphysics simulation software, the self-inductance of the transmitting coil and the receiving coil are both 540uH. After parametric scanning simulation, 30 sets of h and M data were obtained. From the simulation data, it can be seen that after the number of turns and geometric dimensions of the coil are determined, the three-dimensional diagram of the relationship between the transmission distance h and offset distance d and the mutual inductance value M is shown in the figure. 8 shown.

After obtaining the simulation data of the coupling mechanism, the dual-LCC-type MC-WPT simulation model shown in Figure 9 is established in Simulink. The single simulation time is set to 0.02s, and the system has been running in a steady state. The settings of the transmission distance h and the mutual inductance value M in the model use the h and M data sets simulated by the COMSOL software, and the rest of the parameter settings in the model are shown in Table 2.

Table 2. Main parameters of dual LCC type MC-WPT system simulation

parameter Numerical value parameter Numerical value System frequency f 79KHz Transmitter compensation inductance 67uH Transmitter coil inductance L_p 534uH Receiver Compensation Inductance 67uH Receiving coil inductance L_s 534uH Load R_eq 10-100Ω Transmitter compensation capacitor C_p 8.69nF DC Power E_dc 360V Receiver compensation capacitor C_s 8.69nF

In the simulation model shown in Fig. 9, the simulation model is run, and the automatic simulation is performed by writing an M file. Every time the transmission distance h and the load equivalent resistance R _eq are changed, the system input current value _Im and mutual inductance value M are obtained. 410 sets of simulation model data.

When training the TensorFlow load and mutual inductance recognition model, 370 sets of data are randomly selected as the training set, and 40 sets of data are used as the test set. The test set is only used to test the model training effect and does not participate in the model training. By continuously feeding the training set data into the model, use the TensorFlow optimizer Adam Optimizer to optimize the parameters in the model according to the training error value, until the training error value falls to the required error value (which can also be expressed by the error rate). As shown in Figure 10, after 10,000 times of training, the model training error value has been reduced to a very small value. At this time, the recognition accuracy of the mutual inductance M in the training set is 99%, and the recognition accuracy of the mutual inductance M in the test set reaches 99.5%. The training set load Re _eq recognition The accuracy is 98%, the test set load Re _eq recognition accuracy is 99%, and the model training is complete.

Embodiments of the present invention provide a TensorFlow-based MC-WPT system load and mutual inductance identification model, which is based on the TensorFlow deep learning framework and adopts a neural network model, so that the load and mutual inductance identification problem of the MC-WPT system is equivalent to a nonlinear equation. The solution problem is transformed into a deep learning nonlinear fitting problem, and the model is trained for tens of thousands of times using the training set, and finally the MC-WPT system load and mutual inductance recognition model with fast recognition speed and high accuracy is obtained.

Example 2

Based on the MC-WPT system load and mutual inductance identification model described in Embodiment 1, an embodiment of the present invention provides a MC-WPT system load and mutual inductance identification method, including the steps:

X1. Detect the input current value of the current MC-WPT system and the transmission distance between coils;

X2. Input the current input current value and the transmission distance between coils into the MC-WPT system load and mutual inductance identification model, and calculate the corresponding load value and mutual inductance value.

Wherein, in the step X2, the formula for calculating the MC-WPT system load and mutual inductance identification model is:

和

分别表示所述第1权重矩阵和所述第1偏置矩阵，L₁表示对l₁中每个元素代入激活函数

进行运算后得到的隐藏层输出矩阵；Among them, l ₁ represents the intermediate variable of the first hidden layer, [h I _m ] represents the matrix composed of the transmission distance between the coils of the MC-WPT system and the input current value,

and

respectively represent the first weight matrix and the first bias matrix, and L ₁ represents that each element in l ₁ is substituted into the activation function

The hidden layer output matrix obtained after the operation;

和

分别表示所述第2权重矩阵和所述第2偏置矩阵，L₂表示对l₂中每个元素代入激活函数

进行运算后得到的隐藏层输出矩阵；l ₂ represents the intermediate variable of the second hidden layer,

and

respectively represent the second weight matrix and the second bias matrix, and L ₂ represents that each element in l ₂ is substituted into the activation function

The hidden layer output matrix obtained after the operation;

...;

和

分别表示所述第N权重矩阵和所述第N偏置矩阵，L_N表示对l_N中每个元素代入激活函数

进行运算后得到的隐藏层输出矩阵；l _N represents the intermediate variable of the Nth hidden layer,

and

respectively represent the Nth weight matrix and the Nth bias matrix, and L _N represents that each element in l _N is substituted into the activation function

The hidden layer output matrix obtained after the operation;

和

分别表示所述第N+1 权重矩阵和所述第N+1偏置矩阵，M、R_eq分别表示所述输出层输出的互感值和负载值。l _N+1 represents the intermediate variable of the output layer,

and

represent the N+1th weight matrix and the N+1th bias matrix, respectively, and M and _Req represent the mutual inductance value and load value output by the output layer, respectively.

的值，在线识别负载与互感时，通过检测系统输入电流值I_m和传输距离h，将上述导出的值代入式(16)进行运算，从而可实现在微型控制器中进行负载与互感识别。Likewise, in this embodiment, N=3, and k=10. After the model training of Example 1 is completed, the weight matrix and the bias matrix are derived

When identifying the load and mutual inductance online, by detecting the input current value _Im and the transmission distance h of the system, and substituting the above-derived value into the formula (16) for calculation, the load and mutual inductance identification in the microcontroller can be realized.

Example 3

An embodiment of the present invention provides a TensorFlow-based MC-WPT system load and mutual inductance identification system, including a controller, a current detection module and a ranging module connected to the controller; the current detection module is used to detect the LCC of the transmitter in the MC-WPT system The input current value I _m of the circuit topology is sent to the controller; the ranging module is used to detect the transmission distance h between the transmitting coil and the receiving coil in the MC-WPT system and send it to the controller; the controller is used to install the The load and mutual inductance identification model of the MC-WPT system is described, and the load and mutual inductance values corresponding to the input current value and transmission distance are calculated according to the MC-WPT system load and mutual inductance identification method described in Embodiment 2.

Preferably, the current detection module is a Hall sensor. The ranging module is an infrared ranging sensor, which is installed in the center of the transmitting coil of the coupling mechanism.

In order to verify the feasibility and effectiveness of the system (including the identification model and method on it), this embodiment establishes the load and mutual inductance identification model shown in Figure 11 based on the dual LCC type MC-WPT system and the above-mentioned TensorFlow-based identification model. Simulink simulation validates the model. The model is mainly composed of the following parts: ①Dual LCC type MC-WPT system main circuit; ②Inverter circuit signal generation module; ③Transmitting end LCC topology network inductance L _f1 current value _Im and transmission distance h acquisition unit, simulation In , the current value I _m is detected in real time by the detection module, and the transmission distance h is automatically changed by programming input; ④ identification model algorithm unit. The system simulation parameters are consistent with Table 2.

In the simulation verification model, the load and mutual inductance recognition model algorithm (16) based on the TensorFlow framework is encapsulated in the recognition model algorithm unit in Figure 11 by writing M language. 10 groups of load and mutual inductance values are brought into the simulation verification model to obtain the identification results and relative errors as shown in Table 3.

Table 3. Load and mutual inductance identification results of LCC type MC-WPT system

It can be seen from Table 3 that the maximum relative errors of load and mutual inductance identification are only 0.5% and 0.34%, and the identification results obtained through the TensorFlow load and mutual inductance identification model are very close to the set value.

In summary, this embodiment proposes a TensorFlow-based dual-LCC-type MC-WPT system load and mutual inductance identification model, method, and system. The model trains the model offline, and the trained model is imported into the microcontroller and can be performed online. Simultaneous identification of load and mutual inductance has fast identification speed and high precision, which is conducive to real-time control of the system, and has low cost, easy implementation, and is conducive to engineering popularization and application.

The above-mentioned embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited by the above-mentioned embodiments, and any other changes, modifications, substitutions, combinations, The simplification should be equivalent replacement manners, which are all included in the protection scope of the present invention.

Claims (8)

1. based on the MC-WPT system load of TensorFlow and mutual inductance identification model, it is characterized in that, its generation step comprises: S1. Build a fully connected neural network model based on the TensorFlow framework; S2. establish the COSMOL and Simulink simulation model of MC-WPT system, obtain the input current value of multiple groups of described MC-WPT system, the simulation data of transmission distance between coils, and described simulation data is divided into training set and test set; S3. Input the training set into the fully connected neural network model for model training, and continuously optimize the parameters in the fully connected neural network model according to the training error value; S4. When the training error rate of the fully connected neural network model is as low as the preset error rate, the training is ended, and the trained MC-WPT system load and mutual inductance identification model is obtained. 2. the MC-WPT system load and mutual inductance identification model based on TensorFlow as claimed in claim 1, it is characterized in that: described fully connected neural network model comprises input layer, output layer and sequence fully connected in described input layer and all. The 1st to Nth hidden layers between the output layers, N≥1; the 1st to Nth hidden layers have k nodes, and k≥2. 3. The MC-WPT system load and mutual inductance identification model based on TensorFlow as claimed in claim 2, wherein the nonlinear activation function of the first to N hidden layers uses the Sigmoid activation function in the TensorFlow framework. 4. the MC-WPT system load and mutual inductance identification model based on TensorFlow as claimed in claim 2, it is characterized in that: the parameter optimized in described step S3 comprises the 1st weight matrix and the 1st weight matrix acting on the 1st hidden layer bias matrix, and the second weight matrix and the second bias matrix acting on the second hidden layer until the Nth weight matrix and the Nth bias matrix acting on the Nth hidden layer, and acting on all The N+1th weight matrix and the N+1th bias matrix of the output layer. 5. MC-WPT system load and mutual inductance identification method based on TensorFlow, is characterized in that, comprises the steps: X1. Detect the input current value of the current MC-WPT system and the transmission distance between coils; X2. Input the current input current value and the transmission distance between coils into the MC-WPT system load and mutual inductance identification model described in any one of claims 1 to 4, and calculate the corresponding load value and mutual inductance value. 6. the MC-WPT system load and mutual inductance identification method based on TensorFlow as claimed in claim 5 is characterized in that, in described step X2, the MC-WPT system load described in claim 4 and mutual inductance identification model are calculated The formula is:

Among them, l ₁ represents the intermediate variable of the first hidden layer, [h I _m ] represents the matrix composed of the transmission distance between the coils of the MC-WPT system and the input current value,

and

respectively represent the first weight matrix and the first bias matrix, and L ₁ represents that each element in l ₁ is substituted into the activation function

The hidden layer output matrix obtained after the operation; l ₂ represents the intermediate variable of the second hidden layer,

and

respectively represent the second weight matrix and the second bias matrix, and L ₂ represents that each element in l ₂ is substituted into the activation function

The hidden layer output matrix obtained after the operation; ...; l _N represents the intermediate variable of the Nth hidden layer,

and

respectively represent the Nth weight matrix and the Nth bias matrix, and L _N represents that each element in l _N is substituted into the activation function

The hidden layer output matrix obtained after the operation; l _N+1 represents the intermediate variable of the output layer,

and

represent the N+1th weight matrix and the N+1th bias matrix, respectively, and M and _Req represent the mutual inductance value and the load value output by the output layer, respectively. 7. based on the MC-WPT system load of TensorFlow and mutual inductance identification system, it is characterized in that: comprise controller and the current detection module and distance measuring module that connect described controller; Described current detection module is used to detect in the MC-WPT system The input current value of the LCC circuit topology of the transmitting end is sent to the controller; the ranging module is used to detect the transmission distance between the transmitting coil and the receiving coil in the MC-WPT system and send it to the controller; the The controller is used to install the MC-WPT system load and mutual inductance identification model. According to the MC-WPT system load and mutual inductance identification method, it calculates the corresponding input current value, load value and mutual inductance value under the transmission distance. 8. The MC-WPT system load and mutual inductance identification system based on TensorFlow as claimed in claim 7, wherein the current detection module is a Hall sensor, and the ranging module is an infrared ranging sensor.

Images