CN112583139A - WPT (wavelet packet transform) system based on fuzzy RBF (radial basis function) neural network and frequency tracking method - Google Patents

Abstract

The invention discloses a WPT system and a frequency tracking method based on a fuzzy RBF neural network. The system includes a main circuit of an MCR-WPT system and a frequency tracking control circuit. The frequency tracking control circuit is used for detecting and collecting the transmission in the main circuit of the MCR-WPT system. The phase difference and phase difference change rate of the resonant voltage and resonant current at the terminal, the PID controller uses the RBF neural network to adjust the control amount of the frequency of the H-bridge inverter in real time, and realizes the adaptive tracking of the resonant operating frequency of the transmitter, so that the resonance of the transmitter can be achieved. The voltage and the resonant current keep the same frequency and phase at all times, that is, the phase difference between the two is maintained at 0 degrees, so as to ensure that the resonant operating frequency of the system is consistent with the natural resonant frequency of the system, providing a guarantee for the wireless power transmission system to achieve high transmission efficiency. .

Description

WPT system and frequency tracking method based on fuzzy RBF neural network

technical field

The invention relates to the technical field of wireless charging systems, in particular to a WPT system and a frequency tracking method based on a fuzzy RBF neural network.

Background technique

At present, the Magnetic Coupled Resonance Wireless Power Transfer (MCR-WPT) system has the advantages of long transmission distance and high transmission efficiency in the near-field WPT technology. It has been widely used in the field of wireless charging and has become a research hotspot in the field of wireless charging. However, factors such as ambient temperature, operating conditions, coil size, and surface effects may cause flux and current changes in the resonant coil. These changes may lead to the change of the actual working resonant frequency, which makes the transmission efficiency drop rapidly. Therefore, keeping the MCR-WPT system working at the resonant frequency is one of the key technologies to improve the transmission efficiency.

In order to ensure that the MCR-WPT system works at the resonant frequency point, there are mainly three control methods: coil topology optimization, dynamic compensation tuning and frequency tracking. Compared with the other two methods, frequency tracking control is widely used in the system because of its easy implementation and fast response.

However, the current automatic frequency tracking methods of WPT systems mostly realize the linear adjustment of the operating frequency by detecting the phase difference between the resonant voltage and the current. However, the MCR-WPT system often has a steep change process at the moment of frequency detuning or resonant frequency drift, and there is no specific model for this nonlinear change.

SUMMARY OF THE INVENTION

In view of the above-mentioned technical deficiencies, the present invention establishes a frequency tracking closed-loop system model based on fuzzy RBF neural network by analyzing the influence of control parameters on system performance, and real-time nonlinear frequency tracking, on the basis of improving the transmission efficiency of the MCR-WPT system, The frequency tracking control with fast response and high precision is realized. The analysis results confirm that the frequency control algorithm enhances the adaptive tracking ability of the working resonant frequency of the WPT system, greatly improves the transmission efficiency of the system, and provides important information for the optimal design of the WPT system efficiency. refer to.

In order to solve the above-mentioned technical problems, the present invention adopts the following technical solutions:

The present invention provides a WPT system based on a fuzzy RBF neural network. The system includes a MCR-WPT system main circuit and a frequency tracking control circuit, wherein the frequency tracking control circuit includes a voltage and current acquisition circuit, a zero-crossing detection circuit, and a digital phase detection circuit. controller, fuzzy controller, PWM generator, PID controller and H-bridge inverter drive circuit;

Among them, the voltage and current acquisition circuit completes the detection of the transmitter voltage u ₁ and the resonant current i ₁ of the main circuit of the MCR-WPT system;

The zero-crossing detection circuit converts the main circuit voltage u ₁ and resonant current i ₁ of the MCR-WPT system into square wave signals u _u (θ) and u _i (θ) with the same frequency and phase;

The digital phase detector compares the phases of the signals u _u (θ) and u _i (θ), and the resulting phase difference Δθ phase difference rate of change

的变化率，实时调整PID参数，将相位差维持在0度；The fuzzy controller adjusts the parameters of the PID controller based on the fuzzy RBF neural network. When the PID controller works, the fuzzy RBF neural network adjusts the parameters according to Δθ and

The rate of change is adjusted in real time, and the PID parameters are adjusted in real time to maintain the phase difference at 0 degrees;

The PWM generator adjusts the frequency according to the PID parameters, generates the same frequency and in-phase H bridge drive logic signal as _ui (θ), and realizes the switching control of the MOSFET in the H bridge inverter drive circuit through the H bridge inverter drive circuit.

The frequency tracking method of WPT system based on fuzzy RBF neural network includes the following steps:

S1: Build a wireless power transmission system model based on RBF neural network, form a resonant network, and perform real-time adaptive tracking of the resonant operating frequency of the transmitter through the fuzzy RBF neural network;

宽度

并计算出模糊化层内的规则适用度w_j和模糊隶属度

初始值；S2: Initialize the fuzzy RBF neural network, randomly obtain a sample in the training set, and select the center of the membership function in the fuzzy layer of the RBF neural network

width

And calculate the rule applicability w _j and fuzzy membership degree in the fuzzy layer

initial value;

Among them, sample X=[x ¹ , x ² ] ^T , T represents the transpose of X;

表示样本中第i个特征对于第i个特征中第j个模糊子集的模糊隶属度；Among them, the fuzzification layer contains h classes, forming h nodes, each node corresponds to a fuzzy rule, j=1, 2,...h;

Indicates the fuzzy membership of the ith feature in the sample to the jth fuzzy subset in the ith feature;

其中，x¹＝Δθ，

S3: Obtain the resonant current value u _i (k) at the transmitter end of the system and the resonance voltage value u _u (k) at the transmitter end by sampling, and calculate the system phase difference Δθ(k)=u _i (k)-u _u (k) and the phase difference rate of change

where x ¹ =Δθ,

S4: Determine the performance index function of the fuzzy RBF neural network

宽度

为预先设定的正数，相应模糊规则中的权值均初始化为0；S5: Compare the fuzzy rule applicability w _j in the sample with a preset threshold δ, if argmax(w _j )>δ, j=1, 2, ... h, then in the fuzzy layer of the RBF neural network A node corresponding to the h+1 fuzzy rule is added to , and the feature component of the corresponding dimension of the sample is taken as the center of the membership function

width

is a preset positive number, and the weights in the corresponding fuzzy rules are initialized to 0;

If not satisfied, go to the next step;

S6: If there are rule j and rule k that satisfy formula (20), then merge the two fuzzy rules, otherwise go to the next step of parameter learning;

Among them, ψ, σ are preset values;

宽度

和网络权值系数

S7: The RBF neural network performs parameter learning and calculates the center of the membership function in the RBF neural network

width

and network weight coefficients

S8: According to formula (15), calculate the three parameters K _p , K _i and K _d of the PID controller output by the output layer of the RBF neural network;

In the formula: y _n1 is the input of the output layer of the RBF neural network, σ(l, j) is the connection weight matrix between y _n1 and the output layer, l=1, 2, 3 correspond to three parameters K _p , K _i and K respectively _d ;

Then the fuzzy RBF neural network PID controller calculates the frequency modulation signal increment ΔZ(k) as:

S9: Obtain the control signal Z(k) that can keep the resonant operating frequency in the resonant network consistent with the natural resonant frequency of the system according to S8, wherein:

Z(k)=Z(k-1)+ΔZ(k) (23);

Add Z(k) to the resonant network formed by building the system model to ensure that the H-bridge inverter in the main circuit of the MCR-WPT system works at the resonance point and maintains the phase difference Δθ(k) to zero at all times;

S10: Let k=k+1, and then return to step S2 for the next cycle of Δθ(k+1) detection and frequency control signal Z(k+1) adjustment to form real-time detection and adjustment.

Preferably, in step S2, k-means clustering algorithm is used to discretize the fuzzing layer into h classes between [-5, 5], and the h classes form h nodes, each node has two Gaussians Membership function.

Preferably, in step S6, the merging mode of the two rules is carried out according to the following formula:

where i=1,2.

Preferably, in step S7, the membership function parameters and network weight coefficients in the RBF neural network are learned according to the following formula:

Among them, k is the number of network iteration steps; α is the learning rate, β is the learning momentum factor, and α, β∈[0,1], i=1,2.

的线性组合：Preferably, in step S8, the input y _n1 of the output layer is y _j and the normalized fitness

A linear combination of :

位于RBF神经网络的归一化层，归一化后的

将充当RBF神经网络的模糊化层与其输出层输入的连接权值。in,

Located in the normalization layer of the RBF neural network, the normalized

The connection weights that will act as the fuzzification layer of the RBF neural network to its output layer input.

Preferably, the RBF neural network is normalized by defuzzification by the centroid method:

where i=1,2.

The beneficial effects of the present invention are:

By analyzing the influence of control parameters on system performance, the invention establishes a frequency tracking closed-loop system model based on fuzzy RBF neural network, designs a fuzzy adaptive controller for frequency tracking control, real-time nonlinear frequency tracking, and improves the performance of the MCR-WPT system. On the basis of transmission efficiency, fast response and high-precision frequency tracking control is realized, which enhances the adaptive tracking ability of the working resonant frequency of the MCR-WPT system, so that the resonant voltage and resonant current at the transmitter end are always kept at the same frequency and phase, that is, the phases of the two. The difference is maintained at 0 degrees, thereby ensuring that the resonant operating frequency of the system is consistent with the natural resonant frequency of the system, providing a guarantee for the wireless power transmission system to achieve high transmission efficiency, and greatly improving the system transmission efficiency, which is MCR-WPT System efficiency optimization design provides an important reference.

Description of drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts.

1 is a system block diagram of the present invention;

Fig. 2 is the main circuit structure diagram of the series topology of the MCR-WPT system;

Figure 3 is the frequency tracking structure diagram of the fuzzy RBF neural network.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

Example

As shown in Figure 1 and Figure 2, the present invention provides a WPT system based on fuzzy RBF neural network, the system includes the main circuit of the MCR-WPT system and the frequency tracking control circuit, specifically:

表示，i₁、i₂分别为发送端和接收端的高频谐振电流，R₁、R₂为发射电路和接收电路的寄生电阻，整流二极管D₁～D₄构成全桥整流器，C_L为整流桥滤波电容，利用其充放电作用，使输出电压U_L趋于平滑；R_L为负载侧等效电阻；With reference to Figure 2, the main circuit of the MCR-WPT system adopts a typical double-coil string-string topology main circuit structure, which is mainly composed of an H-bridge inverter, a series resonance circuit at the transmitter, a series resonance circuit at the receiver, and a full-bridge rectifier circuit; among them, U _d is the DC power supply, C _S is the power supply filter capacitor, the field effect transistors Q ₁ to Q ₄ form an H-bridge inverter, L ₁ is the inductance of the transmitting coil, L ₂ is the inductance of the receiving coil, and C ₁ and C ₂ are the transmitting ends The resonance compensation capacitor corresponding to the receiving end, M is the mutual inductance between the transmitting coil and the receiving coil, and the tightness of the magnetic coupling between the two coils is determined by the coupling coefficient

Indicates that i ₁ and i ₂ are the high-frequency resonant currents of the transmitter and receiver respectively, R ₁ and R ₂ are the parasitic resistances of the transmitter circuit and the receiver circuit, the rectifier diodes D ₁ to D ₄ constitute a full-bridge rectifier, and C _L is the rectifier The bridge filter capacitor uses its charge and discharge function to make the output voltage _{UL tend to be smooth; RL is} the equivalent resistance on the load side;

With reference to Figure 1, the frequency tracking control circuit includes a voltage and current acquisition circuit, a zero-crossing detection circuit, a digital phase detector, a fuzzy controller, a PWM generator, a PID controller and an H-bridge inverter drive circuit;

模糊控制器基于模糊RBF神经网络对PID控制器的参数进行调整，当PID控制器工作时，模糊RBF神经网络根据Δθ和

的变化率，实时调整PID参数，将相位差维持在0度；PWM发生器根据PID参数进行频率调整，生成与u_i(θ)同频同相H桥驱动逻辑信号，经H桥逆变驱动电路，实现对H桥逆变驱动电路中MOSFET管的开关控制，使H桥逆变器工作频率在谐振点。Among them, the voltage and current acquisition circuit completes the detection of the resonant voltage u ₁ and the resonant current i ₁ of the transmitter of the main circuit of the MCR-WPT system; the zero-crossing detection circuit converts the resonant voltage u ₁ and the resonant current i ₁ of the main circuit of the MCR-WPT system are square wave signals u _u (θ) and u _i (θ) with the same frequency and phase; the digital phase detector compares the phases of the signals u _u (θ) and u _i (θ), and the resulting phase difference Δθ phase difference changes Rate

The fuzzy controller adjusts the parameters of the PID controller based on the fuzzy RBF neural network. When the PID controller works, the fuzzy RBF neural network adjusts the parameters according to Δθ and

The PID parameters are adjusted in real time, and the phase difference is maintained at 0 degrees; the PWM generator adjusts the frequency according to the PID parameters, and generates the same frequency and phase H bridge drive logic signal as u _i (θ), which is driven by the H bridge inverter drive circuit. , to realize the switch control of the MOSFET tube in the H-bridge inverter drive circuit, so that the H-bridge inverter operating frequency is at the resonance point.

R_o＝8R_L/π²，Further, with reference to Figure 2, in order to facilitate the analysis of frequency detuning characteristics, the SS topology in Figure 2 is analyzed, where

R _o =8R _L /π ² ,

According to Kirchhoff's voltage law, the equivalent model column circuit equation can be obtained:

Among them, Z ₁ and Z ₂ are the equivalent impedances of the transmitter and receiver, which satisfy:

For the convenience of analysis, the structures of the transmitter coil and the receiver coil are selected to be the same, that is, L ₁ =L ₂ =L, R ₁ =R ₂ =R, and C ₁ =C ₂ =C;

Based on equations (1) and (2), the current values on both sides can be calculated as:

where ω is the inverter angular frequency, the input power P _in and output power P _out of the MCR-WPT system can be calculated as:

U ₁ is the effective value of the input power supply voltage u ₁ .

According to the electromagnetic resonance conditions, the detuning rate is defined as:

γ=ωL-1/ωC (5)

When γ=0, ω=ω ₀ =1/LC, ω ₀ is the resonant angular frequency, the resonant network is in the resonance state, and the coil loop is purely resistive; when γ>0, ω>ω ₀ , the resonant network is in over- In the resonance state, the loop is inductive; when γ<0, ω<ω ₀ , the resonant network is in an under-resonant state, and the coil loop is capacitive;

From equations (4) and (5), the transmission efficiency η can be calculated

It can be seen from equation (6) that when the resonant network is in the resonance state, the equivalent impedance of the coil loop is the smallest, and the energy in the coil can achieve the transmission with the highest transmission efficiency; in the non-resonant state, the greater the detuning rate, the higher the transmission efficiency of the system. The efficiency will be greatly reduced; because the MCR-WPT system adopts a series resonance structure, the transmitter current is a sinusoidal signal, and the voltage is a square wave signal, and the fuzzy control method is used to perform real-time adaptive frequency tracking on the transmitter resonance current.

specific:

For the WPT system, due to the influence of factors such as transmission distance, coil offset, and load changes, the structure and parameters in the model can change at any time. If the conventional PID strategy with constant parameters is used for control, it is difficult to achieve ideal frequency tracking. Effect; the WPT system is based on the conventional PID frequency tracking control, and the structure of the fusion fuzzy RBF neural network to be adopted in the present invention is shown in Figure 3;

Fuzzy RBF neural network includes input layer, fuzzification layer (ie hidden layer), normalization layer and output layer;

Input layer:

即输入样本X＝[x¹，x²]^T，其中x¹＝Δθ，

T表示对X的转置(指数学的矩阵转置，一行转换为一列)。The number of nodes in the input layer of the fuzzy RBF neural network is 2, which are the frequency phase difference Δθ and the phase difference change rate respectively.

That is, the input sample X=[x ¹ , x ² ] ^T , where x ¹ =Δθ,

T represents the transpose of X (referred to as the mathematical matrix transpose, converting a row to a column).

Hidden layer:

作为隐含层各高斯隶属度函数的初始中心参数；The hidden layer is also called the fuzzification layer. The k-means clustering algorithm is used to discretize them into h classes between [-5, 5]. These h classes form h nodes, and each node corresponds to a fuzzy rule. , each node has two Gaussian membership functions; the cluster centers of these j classes (j=1, 2, ... h)

As the initial center parameter of each Gaussian membership function of the hidden layer;

表示样本中第i个特征对于第i个特征中第j个模糊子集的模糊隶属度，

分别代表高斯隶属度函数的中心和宽度；in the formula

represents the fuzzy membership of the ith feature in the sample to the jth fuzzy subset in the ith feature,

represent the center and width of the Gaussian membership function, respectively;

The output value of the jth node of the hidden layer, that is, the applicability of the jth rule w _j , adopts a fuzzy rule applicability calculation method based on the Mahalanobis distance instead of the traditional Euclidean distance:

The method of adaptive modification of different membership function width parameters makes the fitting effect of the model better. Equation (8) can be expressed as:

w _j =exp[md ² (j)] (9)

where md(j) represents the Mahalanobis distance between the input sample and the jth node of the hidden layer,

∑ _j ^-1 is expressed as:

代表第i个特征分量中第j个模糊子集的隶属度函数宽度。In formula (11)

represents the membership function width of the jth fuzzy subset in the ith feature component.

Normalization layer:

Although the traditional standard RBF fuzzy neural network performs well in training, its generalization ability in testing is not high, and the normalized RBF fuzzy neural network can effectively improve the generalization ability of the model. After the invention integrates the T-S fuzzy model, the method of defuzzification by the center of gravity method is adopted to normalize the network;

将充当RBF神经网络的模糊化层与其输出层输入的连接权值。normalized

The connection weights that will act as the fuzzification layer of the RBF neural network to its output layer input.

output layer:

The output layer contains h fuzzy rules and the h hidden layer nodes in the normalization layer are in one-to-one correspondence, and the output y _j generated by the jth fuzzy rule is calculated by the following rule inference:

代表了第i个特征分量中的第j个模糊子集，

为实数，j＝1，2，...h；i＝1，2；in,

represents the jth fuzzy subset in the ith feature component,

is a real number, j=1, 2, ... h; i=1, 2;

的线性组合：The input y _n1 of the output layer is y _j and the normalized fitness

A linear combination of :

The output layer mainly outputs the three parameters K _p , K _i and K _d of the PID controller, and the selected activation function is:

In the formula: σ(l, j) is the connection weight matrix between y _n1 and the output layer, and l=1, 2, and 3 correspond to three parameters K _p , K _i and K _d respectively.

1 to 3, the frequency tracking method for the system is as follows:

S1: Build a wireless power transmission system model based on RBF neural network, form a resonant network, and perform real-time adaptive tracking of the resonant operating frequency of the transmitter through the fuzzy RBF neural network to ensure that the resonant current of the system and the resonant voltage are kept at the same time. frequency in the same direction;

宽度

并计算出模糊化层内的规则适用度w_j和模糊隶属度

初始值；S2: Initialize the fuzzy RBF neural network, randomly obtain a sample in the training set, and select the center of the membership function in the fuzzy layer of the RBF neural network

width

And calculate the rule applicability w _j and fuzzy membership degree in the fuzzy layer

initial value;

其中，x¹＝Δθ，

S3: Obtain the resonant current value u _i (k) at the transmitter end of the system and the resonance voltage value u _u (k) at the transmitter end by sampling, and calculate the system phase difference Δθ(k)=u _i (k)-u _u (k) and the phase difference rate of change

where x ¹ =Δθ,

即确定了模糊RBF神经网络的性能指标函数；S4: In order to find the optimal weight and make the output of the fuzzy RBF neural network the closest to the expected output value, that is, to achieve the minimum error, the cost function is defined as

That is, the performance index function of the fuzzy RBF neural network is determined;

宽度

为预先设定的正数，相应模糊规则中的权值均初始化为0；S5: Compare the fuzzy rule applicability w _j in the sample with a preset threshold δ, if argmax(w _j )>δ, j=1, 2, ... h, then in the fuzzy layer of the RBF neural network A node corresponding to the h+1 fuzzy rule is added to , and the feature component of the corresponding dimension of the sample is taken as the center of the membership function

width

is a preset positive number, and the weights in the corresponding fuzzy rules are initialized to 0;

Among them, the function argmax() indicates whether the number in the given point set reaches the given maximum value, if not, it will go to the next step;

S6: If there are rule j and rule k that satisfy formula (20), then merge the two fuzzy rules, otherwise go to the next step of parameter learning;

Among them, ψ, σ are preset values;

If two rules are combined, follow the formula below:

Among them, i=1, 2;

宽度

和网络权值系数

并按如下公式进行学习：S7: The RBF neural network performs parameter learning and calculates the center of the membership function in the RBF neural network

width

and network weight coefficients

And learn according to the following formula:

Among them, k is the number of network iteration steps; α is the learning rate, β is the learning momentum factor, and α, β∈[0,1], i=1,2.

S8: According to formula (15), calculate the three parameters K _p , K _i and K _d of the PID controller output by the output layer of the RBF neural network;

Then the fuzzy RBF neural network PID controller calculates the frequency modulation signal increment ΔZ(k) as:

S9: Obtain the control signal Z(k) that can keep the resonant operating frequency in the resonant network consistent with the natural resonant frequency of the system according to S8, wherein:

Z(k)=Z(k-1)+ΔZ(k) (23);

Add Z(k) to the PWM generator to drive the H-bridge inverter to ensure that the H-bridge inverter in the main circuit of the MCR-WPT system works at the resonance point and maintains the phase difference Δθ(k) to zero at all times;

S10: Let k=k+1, and then return to step S2 for the next cycle of Δθ(k+1) detection and frequency control signal Z(k+1) adjustment to form real-time detection and adjustment.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit and scope of the invention. Thus, provided that these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims (7)

1. The WPT system based on the fuzzy RBF neural network is characterized by comprising an MCR-WPT system main circuit and a frequency tracking control circuit, wherein the frequency tracking control circuit comprises a voltage current acquisition circuit, a zero-crossing detection circuit, a digital phase discriminator, a fuzzy controller, a PWM (pulse-width modulation) generator, a PID (proportion integration differentiation) controller and an H-bridge inverter driving circuit;

wherein, the voltage and current acquisition circuit finishes the voltage u at the emission end of the main circuit of the MCR-WPT system₁And a resonant current i₁Detecting;

the zero-crossing detection circuit enables the resonant voltage u of the main circuit transmitting end of the MCR-WPT system to be higher than the resonant voltage u₁And a resonant current i₁Converted into square wave signal u with same frequency and phase_u(θ) and u_i(θ)；

Digital phase discriminator_u(θ) and u_i(theta) comparing the phases to generate a phase difference Delta theta and a phase difference change rate

The fuzzy controller adjusts the parameters of the PID controller based on the fuzzy RBF neural network, and when the PID controller works, the fuzzy RBF neural network adjusts the parameters of the PID controller according to the sum of delta theta and delta theta

Adjusting PID parameters in real time, and maintaining the phase difference at 0 degree;

the PWM generator adjusts the frequency according to the PID parameter to generate the frequency and the frequency u_iAnd (theta) driving a logic signal by using a same-frequency and same-phase H bridge, and realizing the switch control of an MOSFET tube in an H bridge inverter in the MCR-WPT system main circuit through an H bridge inverter driving circuit.

2. The WPT system frequency tracking method based on the fuzzy RBF neural network is characterized by comprising the following steps of:

s1: constructing a wireless electric energy transmission system model based on an RBF neural network to form a resonant network, and carrying out real-time self-adaptive tracking on the resonant working frequency of a transmitting end through a fuzzy RBF neural network;

s2: initializing fuzzy RBF neural network, randomly obtaining a sample in training set, and selecting center of membership function in fuzzification layer of RBF neural network

Width of

And calculating the rule suitability w in the fuzzified layer_jAnd fuzzy degree of membership

An initial value;

wherein, sample X ═ X¹，x²]^TT represents a transposition of X;

the fuzzy layer comprises h classes, h nodes are formed, each node corresponds to a fuzzy rule, and j is 1, 2 and … h;

representing fuzzy membership of the ith feature in the sample to the jth fuzzy subset in the ith feature;

s3: obtaining the resonant current value u of the system transmitting end by sampling_i(k) And the resonant voltage value u of the transmitting terminal_u(k) Calculating the system phase difference Δ θ (k) u_i(k)-u_u(k) Rate of change of sum phase difference

wherein ,

s4: determining performance indicator function of fuzzy RBF neural network

S5: pasting rule applicability w in sample_jComparing with a preset threshold value delta if argmax (w)_j) If the distance is more than delta, j is 1, 2, a

Width of

The weights in the corresponding fuzzy rules are initialized to be 0 for a preset positive number;

if not, entering the next step;

s6: if the rule j and the rule k meet the formula (20), merging the two fuzzy rules, otherwise, entering the parameter learning of the next step;

wherein psi and sigma are preset values;

s7: the RBF neural network carries out parameter learning and calculates the center of the membership function in the RBF neural network

Width of

And network weight coefficient

S8: calculating three parameters K of the PID controller output by the RBF neural network output layer according to a formula (15)_p，K_i and K_d；

in the formula ：y_n1Is input to the output layer of the RBF neural network, and has a value of y_n1The connection weight matrix with the output layer, l 1, 2, 3 correspond to 3 parameters K respectively_p，K_i and K_d；

The fuzzy RBF neural network PID controller calculates the frequency modulation signal increment delta Z (k) as:

s9: obtaining a control signal Z (k) capable of keeping the resonant operating frequency in the resonant network consistent with the natural resonant frequency of the system according to S8, wherein:

Z(k)＝Z(k-1)+ΔZ(k) (23)；

adding Z (k) into a resonant network formed by constructing a system model to ensure that an H bridge inverter in a main circuit of the MCR-WPT system works at a resonant point and constantly maintains the phase difference delta theta (k) to be zero;

s10: let k be k +1, then return to step S2 to perform Δ θ (k +1) detection and frequency control signal Z (k +1) adjustment in the next cycle, resulting in real-time detection and adjustment.

3. The method for tracking the frequency of the WPT system based on the fuzzy RBF neural network as claimed in claim 2, wherein in step S2, the fuzzy layer is discretized into h classes between [ -5, 5] respectively using a k-means clustering algorithm, the h classes form h nodes, and each node has two gaussian membership functions.

4. The frequency tracking method of the WPT system based on the fuzzy RBF neural network as set forth in claim 2, wherein in step S6, the combination of the two rules is performed according to the following formula:

wherein i is 1, 2.

5. The method for tracking the frequency of the WPT system based on the fuzzy RBF neural network as claimed in claim 2, wherein in step S7, the membership function parameter and the network weight coefficient in the RBF neural network are learned according to the following formula:

wherein k is the number of network iteration steps; α is a learning rate, β is a learning momentum factor, and α, β ∈ [0, 1], i ═ 1, 2.

6. The WPT system for fuzzy RBF neural network as claimed in claim 2, wherein the input y of the output layer is outputted at step S8_n1Is y_jAnd normalized degree of suitability

Linear combination of (a):

wherein ,

normalized in the normalization layer of RBF neural network

And connecting the fuzzification layer serving as the RBF neural network with the input of the output layer of the RBF neural network.

7. The method for tracking the frequency of the WPT system based on the fuzzy RBF neural network as claimed in claim 6, wherein the RBF neural network is normalized by means of center-of-gravity defuzzification:

wherein i is 1, 2.

Images