Control method and system of power factor correction circuit

Abstract

The invention discloses a control method and system of a power factor correction circuit, and the method comprises the following steps: selecting a double-closed-loop control mode of a voltage outer loop and a current inner loop, introducing a BP neural network into the current inner loop, and designing the control frequency of the BP neural network; under the low control frequency of 50Hz, correcting parameters of a PI controller by using the BP neural network; and under the high control frequency of 30kHz, based on the output of the voltage outer loop, the optimized PI controller of the current inner loop is used for controlling the power factor correction circuit. According to the invention, forward and reverse propagation of the neural network is designed, and proper PI control frequency and BP neural network control frequency are selected, so that input current harmonics can be reduced, and the power factor and adaptive performance of the circuit are improved.

Description

Technical Field

[0001] This invention belongs to the field of power electronics technology, specifically relating to a control method and system for a power factor correction circuit. Background Technology

[0002] Power factor correction (PFC) circuits, serving as interface circuits between AC power supplies and DC buses, are widely used in electric vehicle charging stations, switching power supplies, and industrial drives. A common application of PFC is in rectifier circuits, used as active power factor correction modules to enable efficient power transfer between the AC grid and DC loads, while simultaneously improving the power factor and reducing input harmonic distortion.

[0003] To improve energy conversion efficiency, PFC circuits typically employ active power factor correction (PFC) techniques, such as boost PFC circuits. These circuits control the switching transistors to make the input current follow the waveform changes of the input voltage, thereby achieving unity power factor. Building upon this, improved topologies, such as interleaved parallel PFC or resonant PFC, can further improve efficiency and reduce electromagnetic interference (EMI).

[0004] In practical applications, a key issue to be addressed is the design of the control strategy for PFC circuits. Meanwhile, the widespread application of digital control technology has made it possible for PFC circuits to achieve higher dynamic response and stability, meeting more stringent power quality requirements. Therefore, we need to introduce neural networks into the design and optimization of digital control technology to meet performance targets. Summary of the Invention

[0005] This invention aims to address the shortcomings of existing technologies and provides the following solutions:

[0006] A control method for a power factor correction circuit includes the following steps:

[0007] A dual closed-loop control mode consisting of an outer voltage loop and an inner current loop is selected. A BP neural network is introduced into the inner current loop, and the control frequency of the BP neural network is designed.

[0008] At a low control frequency of 50Hz, the parameters of the PI controller are corrected using the BP neural network.

[0009] At a high control frequency of 30kHz, the power factor correction circuit is controlled by a PI controller with an optimized current inner loop, based on the output of the voltage outer loop.

[0010] Preferably, the BP neural network consists of one input layer, one hidden layer, and one output layer;

[0011] The input layer has 3 neuron nodes, and the maximum current error, root mean square error and current effective command are selected as parameters. The hyperbolic tangent function is selected as the activation function, and the input data is normalized.

[0012] The hidden layer has four neuron nodes;

[0013] The output layer has two neuron nodes, and the two neuron nodes of the output layer correspond to the parameter correction amount ΔK of the PI controller. p and ΔK i .

[0014] Preferably, the network weights updated by backpropagation are obtained using the relationship between the hidden layer and the input layer, and between the hidden layer and the output layer of the BP neural network during forward propagation, thereby completing the training of the network.

[0015] Preferably, during forward propagation, the relationship between the hidden layer and the input layer of the BP neural network is as follows:

[0016]

[0017] Among them, net j w represents the net input to the j-th neuron in the hidden layer. ij This represents the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer, O i This represents the output of the i-th neuron in the input layer, O j This represents the output of the j-th neuron in the hidden layer, where i represents the i-th neuron in the input layer and j represents the j-th neuron in the hidden layer.

[0018] The relationship between the hidden layer and the output layer is as follows:

[0019]

[0020] Among them, net k w represents the net input to the k-th neuron in the output layer. jk This represents the connection weight from the j-th neuron in the hidden layer to the k-th neuron in the output layer, O k This represents the output of the k-th neuron in the output layer, where k represents the k-th neuron in the output layer.

[0021] The activation function is:

[0022]

[0023] Where f(x) represents the activation function, and x represents the independent variable of the activation function;

[0024] The derivative of the activation function is:

[0025] f′(x)=1-f 2 (x)

[0026] Here, f'(x) represents the derivative of the activation function.

[0027] Preferably, when performing backpropagation of the error and updating the weights of the neural network according to the negative gradient direction, the expression is:

[0028]

[0029] in, Δw represents the root mean square error. jk (m) represents the weight change from the j-th neuron in the hidden layer to the k-th neuron in the output layer during the m-th iteration, Δw ij (m) represents the weight change from the i-th neuron in the input layer to the j-th neuron in the hidden layer in the m-th iteration, η represents the learning rate, α represents the inertia coefficient, and m represents the number of iterations;

[0030] Calculate the derivative of the expression using the chain rule:

[0031]

[0032] Among them, O k (m) represents the output of the k-th neuron in the input layer during the m-th iteration, net k (m) represents the net input of the k-th neuron in the output layer during the m-th iteration, w jk (m) represents the connection weight from the j-th neuron in the hidden layer to the k-th neuron in the output layer in the m-th iteration. j (m) represents the output of the j-th neuron in the output layer during the m-th iteration. j (m) represents the net input of the j-th neuron in the output layer during the m-th iteration, w ij (m) represents the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer in the m-th iteration;

[0033] Using the sgn symbolic function to find approximate solutions:

[0034]

[0035] Based on the approximate solution and the derivative of the activation function, calculate the network weights updated by backpropagation:

[0036]

[0037] The present invention also provides a control system for a power factor correction circuit, wherein the system applies the above-described method and includes: a BP neural network construction module, a PI controller parameter correction module, and a circuit control module;

[0038] The BP neural network construction module adopts a dual closed-loop control mode of voltage outer loop and current inner loop, introduces the BP neural network into the current inner loop, and designs the control frequency of the BP neural network.

[0039] The PI controller parameter correction module is used to correct the parameters of the PI controller using the BP neural network at a low control frequency of 50Hz.

[0040] The circuit control module is used to control the power factor correction circuit at a high control frequency of 30kHz, based on the output of the voltage outer loop and using an optimized current inner loop PI controller.

[0041] Preferably, the BP neural network consists of one input layer, one hidden layer, and one output layer;

[0042] The input layer has 3 neuron nodes, and the maximum current error, root mean square error and current effective command are selected as parameters. The hyperbolic tangent function is selected as the activation function, and the input data is normalized.

[0043] The hidden layer has four neuron nodes;

[0044] The output layer has two neuron nodes, and the two neuron nodes of the output layer correspond to the parameter correction amount ΔK of the PI controller. p and ΔK i .

[0045] Preferably, the network weights updated by backpropagation are obtained using the relationship between the hidden layer and the input layer, and between the hidden layer and the output layer of the BP neural network during forward propagation, thereby completing the training of the network.

[0046] Preferably, during forward propagation, the relationship between the hidden layer and the input layer of the BP neural network is as follows:

[0047]

[0048] Among them, net j w represents the net input to the j-th neuron in the hidden layer. ij This represents the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer, O i This represents the output of the i-th neuron in the input layer, O jThis represents the output of the j-th neuron in the hidden layer, where i represents the i-th neuron in the input layer and j represents the j-th neuron in the hidden layer.

[0049] The relationship between the hidden layer and the output layer is as follows:

[0050]

[0051] Among them, net k w represents the net input to the k-th neuron in the output layer. jk This represents the connection weight from the j-th neuron in the hidden layer to the k-th neuron in the output layer, O k This represents the output of the k-th neuron in the output layer, where k represents the k-th neuron in the output layer.

[0052] The activation function is:

[0053]

[0054] Where f(x) represents the activation function, and x represents the independent variable of the activation function;

[0055] The derivative of the activation function is:

[0056] f′(x)=1-f 2 (x)

[0057] Here, f'(x) represents the derivative of the activation function.

[0058] Preferably, when performing backpropagation of the error and updating the weights of the neural network according to the negative gradient direction, the expression is:

[0059]

[0060] in, Δw represents the root mean square error. jk (m) represents the weight change from the j-th neuron in the hidden layer to the k-th neuron in the output layer during the m-th iteration, Δw ij (m) represents the weight change from the i-th neuron in the input layer to the j-th neuron in the hidden layer in the m-th iteration, η represents the learning rate, α represents the inertia coefficient, and m represents the number of iterations;

[0061] Calculate the derivative of the expression using the chain rule:

[0062]

[0063] Among them, O k (m) represents the output of the k-th neuron in the input layer during the m-th iteration, net k (m) represents the net input of the k-th neuron in the output layer during the m-th iteration, wjk (m) represents the connection weight from the j-th neuron in the hidden layer to the k-th neuron in the output layer in the m-th iteration. j (m) represents the output of the j-th neuron in the output layer during the m-th iteration. j (m) represents the net input of the j-th neuron in the output layer during the m-th iteration, w ij (m) represents the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer in the m-th iteration;

[0064] Using the sgn symbolic function to find approximate solutions:

[0065]

[0066] Based on the approximate solution and the derivative of the activation function, calculate the network weights updated by backpropagation:

[0067]

[0068] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0069] This invention, by designing the forward and backward propagation of the neural network and selecting appropriate PI control frequency and BP neural network control frequency, can reduce input current harmonics, improve the circuit power factor and adaptive performance, and obtain wide-range voltage regulation capability. It solves the control problem of the totem pole power factor correction circuit and enables it to have wide-range voltage regulation capability. Attached Figure Description

[0070] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0071] Figure 1 This is a schematic diagram of the control method flow according to an embodiment of the present invention;

[0072] Figure 2 This is a block diagram of the control method according to an embodiment of the present invention;

[0073] Figure 3 This is a logic diagram of the control method according to an embodiment of the present invention;

[0074] Figure 4 This is a diagram of the BP neural network model according to an embodiment of the present invention;

[0075] Figure 5 This is a topology diagram of the totem pole power factor correction circuit according to an embodiment of the present invention;

[0076] Figure 6 This is a block diagram of the voltage and current dual-loop control of the totem pole power factor correction circuit according to an embodiment of the present invention. Detailed Implementation

[0077] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0078] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0079] First, explain the totem pole power factor correction circuit that needs to be controlled, such as... Figure 5 As shown, it includes a high-frequency side switch, a power frequency side switch, and a filter inductor. The high-frequency side switch and the power frequency side switch are each a phase bridge arm and are electrically connected to the AC input terminal and the DC output terminal at the same time. The filter inductor is used to smooth the current waveform and suppress high-frequency harmonics.

[0080] Example 1

[0081] In this embodiment, as Figure 1 , Figure 2 , Figure 3 As shown, a control method for a power factor correction circuit includes the following steps:

[0082] S1. A dual closed-loop control mode with an outer voltage loop and an inner current loop is selected. A BP neural network is introduced into the inner current loop, and the control frequency of the BP neural network is designed.

[0083] BP neural network, such as Figure 4 As shown, it consists of one input layer, one hidden layer, and one output layer; the input layer has three neurons, and the maximum current error e(n) is selected. max Root mean square error and current valid command i * As parameters, the hyperbolic tangent function was chosen as the activation function, and the input data was normalized. The hidden layer had four neurons, and the output layer had two neurons, which corresponded to the parameter correction ΔK of the PI controller. p and ΔK i .

[0084] By utilizing the relationships between the hidden layer and the input layer, and between the hidden layer and the output layer during forward propagation of a BP neural network, the network weights updated during backpropagation are obtained, and the network training is completed.

[0085] During forward propagation, the relationship between the hidden layer and the input layer of a BP neural network is as follows:

[0086]

[0087] Among them, net j w represents the net input to the j-th neuron in the hidden layer. ij This represents the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer, O i This represents the output of the i-th neuron in the input layer, O j Let represent the output of the j-th neuron in the hidden layer, where i represents the i-th neuron in the input layer, and j represents the j-th neuron in the hidden layer. The relationship between the hidden layer and the output layer is as follows:

[0088]

[0089] Among them, net k w represents the net input to the k-th neuron in the output layer. jk This represents the connection weight from the j-th neuron in the hidden layer to the k-th neuron in the output layer, O k This represents the output of the k-th neuron in the output layer, where k represents the k-th neuron in the output layer; the activation function is:

[0090]

[0091] Where f(x) represents the activation function, and x represents the independent variable of the activation function; the derivative of the activation function is:

[0092] f′(x)=1-f 2 (x)

[0093] Here, f'(x) represents the derivative of the activation function.

[0094] When performing backpropagation of the error and updating the weights of the neural network according to the negative gradient direction, the expression is:

[0095]

[0096] in, Δw represents the root mean square error. jk (m) represents the weight change from the j-th neuron in the hidden layer to the k-th neuron in the output layer during the m-th iteration, Δw ij(m) represents the weight change from the i-th neuron in the input layer to the j-th neuron in the hidden layer in the m-th iteration; η represents the learning rate, which is the step size of the neural network's backpropagation. Choosing an appropriate learning rate can make the network have both high convergence speed and small oscillations; α represents the inertia coefficient. Choosing an appropriate inertia coefficient can make the network escape some local optima and obtain better results; m represents the number of iterations; the derivative of the expression is calculated using the chain rule:

[0097]

[0098] Among them, O k (m) represents the output of the k-th neuron in the input layer during the m-th iteration, net k (m) represents the net input of the k-th neuron in the output layer during the m-th iteration, w jk (m) represents the connection weight from the j-th neuron in the hidden layer to the k-th neuron in the output layer in the m-th iteration. j (m) represents the output of the j-th neuron in the output layer during the m-th iteration. j (m) represents the net input of the j-th neuron in the output layer during the m-th iteration, w ij (m) represents the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer in the m-th iteration; the approximate solution is obtained using the sgn sign function:

[0099]

[0100] Based on the approximate solution and the derivative of the activation function, calculate the network weights updated by backpropagation:

[0101]

[0102] S2. At a low control frequency of 50Hz, the parameters of the PI controller are corrected using a BP neural network.

[0103] In this embodiment, the BP neural network operates at a control frequency of 50Hz, and the output results of the two neurons in its output layer are used as the changes in the control parameters Kp and Ki of the PI controller, namely ΔKp and ΔKi, respectively, thereby achieving the purpose of correcting the PI controller parameters through the BP neural network.

[0104] S3. At a high control frequency of 30kHz, based on the output of the voltage outer loop, the optimized current inner loop PI controller is used to control the high-frequency side switch and the power frequency side switch S1-S4 of the power factor correction circuit.

[0105] In this embodiment, as Figure 6As shown, the voltage and current dual-loop control circuit operates normally at a high control frequency of 30kHz to control the power factor correction circuit. First, the sampling circuit collects AC voltage, DC voltage, and current and sends them to the controller. The controller first compares the DC voltage with the target value, and the error between them is sent to the PI controller. The output of the PI controller is used as the target value of the current. The purpose of this process is to stabilize the DC voltage near the target value, and its output does not directly control the circuit, so it is called the voltage outer loop. After that, the controller compares the current magnitude with the target value of the current, and the error between them is sent to the PI controller. The output of the PI controller is used as the duty cycle of the PWM signal of the high-frequency side switch to directly control the circuit, so this process is called the current inner loop. The PI parameters of the current inner loop are continuously corrected by the BP neural network according to the circuit state. No matter how the circuit operating state changes, it always operates with the PI parameters that best suit the circuit, ensuring that the current inner loop always has the best dynamic response performance and control effect.

[0106] Example 2

[0107] In this embodiment, a control system for a power factor correction circuit includes: a BP neural network construction module, a PI controller parameter correction module, and a circuit control module.

[0108] The BP neural network construction module adopts a dual closed-loop control mode of voltage outer loop and current inner loop, introduces the BP neural network into the current inner loop, and designs the control frequency of the BP neural network.

[0109] The BP neural network consists of one input layer, one hidden layer, and one output layer. The input layer has three neurons, using the maximum current error, root mean square error, and current effective command as parameters, and the hyperbolic tangent function as the activation function. The input data is also normalized. The hidden layer has four neurons, and the output layer has two neurons, which correspond to the parameter correction ΔK of the PI controller. p and ΔK i .

[0110] By utilizing the relationships between the hidden layer and the input layer, and between the hidden layer and the output layer during forward propagation of a BP neural network, the network weights updated during backpropagation are obtained, and the network training is completed.

[0111] During forward propagation, the relationship between the hidden layer and the input layer of a BP neural network is as follows:

[0112]

[0113] Among them, net j w represents the net input to the j-th neuron in the hidden layer. ijThis represents the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer, O i This represents the output of the i-th neuron in the input layer, O j Let represent the output of the j-th neuron in the hidden layer, where i represents the i-th neuron in the input layer, and j represents the j-th neuron in the hidden layer. The relationship between the hidden layer and the output layer is as follows:

[0114]

[0115] Among them, net k w represents the net input to the k-th neuron in the output layer. jk This represents the connection weight from the j-th neuron in the hidden layer to the k-th neuron in the output layer, O k This represents the output of the k-th neuron in the output layer, where k represents the k-th neuron in the output layer; the activation function is:

[0116]

[0117] Where f(x) represents the activation function, and x represents the independent variable of the activation function; the derivative of the activation function is:

[0118] f′(x)=1-f 2 (x)

[0119] Here, f'(x) represents the derivative of the activation function.

[0120] When performing backpropagation of the error and updating the weights of the neural network according to the negative gradient direction, the expression is:

[0121]

[0122] in, The root mean square error Δw jk (m) represents the weight change from the j-th neuron in the hidden layer to the k-th neuron in the output layer during the m-th iteration, Δw ij (m) represents the weight change from the i-th neuron in the input layer to the j-th neuron in the hidden layer during the m-th iteration, η represents the learning rate, α represents the inertia coefficient, and m represents the number of iterations; the derivative of the expression is calculated using the chain rule:

[0123]

[0124] Among them, O k (m) represents the output of the k-th neuron in the input layer during the m-th iteration, net k (m) represents the net input of the k-th neuron in the output layer during the m-th iteration, w jk(m) represents the connection weight from the j-th neuron in the hidden layer to the k-th neuron in the output layer in the m-th iteration. j (m) represents the output of the j-th neuron in the output layer during the m-th iteration. j (m) represents the net input of the j-th neuron in the output layer during the m-th iteration, w ij (m) represents the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer in the m-th iteration; the approximate solution is obtained using the sgn sign function:

[0125]

[0126] Based on the approximate solution and the derivative of the activation function, calculate the network weights updated by backpropagation:

[0127]

[0128] The PI controller parameter correction module is used to correct the parameters of the PI controller using a BP neural network at a low control frequency of 50Hz.

[0129] The circuit control module is used to control the power factor correction circuit at a high control frequency of 30kHz, based on the output of the voltage outer loop and using an optimized current inner loop PI controller.

[0130] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A control method for a power factor correction circuit, characterized in that, Includes the following steps: A dual closed-loop control mode consisting of an outer voltage loop and an inner current loop is selected. A BP neural network is introduced into the inner current loop, and the control frequency of the BP neural network is designed. At a low control frequency of 50Hz, the parameters of the PI controller are corrected using the BP neural network. At a high control frequency of 30kHz, the power factor correction circuit is controlled by a PI controller with an optimized current inner loop, based on the output of the voltage outer loop.

2. The control method for a power factor correction circuit according to claim 1, characterized in that, The BP neural network consists of one input layer, one hidden layer, and one output layer; The input layer is configured with 3 neuron nodes, and the maximum current error, root mean square error and current effective command are selected as parameters. The hyperbolic tangent function is selected as the activation function, and the input data is normalized. The hidden layer has four neuron nodes; The output layer has two neuron nodes, and the two neuron nodes of the output layer correspond to the parameter correction amount ΔK of the PI controller. p and ΔK i .

3. The control method for a power factor correction circuit according to claim 2, characterized in that, The network weights updated during backpropagation are obtained by utilizing the relationships between the hidden layer and the input layer, and between the hidden layer and the output layer of the BP neural network during forward propagation, thus completing the training of the network.

4. The control method for a power factor correction circuit according to claim 2, characterized in that, During forward propagation, the relationship between the hidden layer and the input layer of the BP neural network is as follows: Among them, net j w represents the net input to the j-th neuron in the hidden layer. ij This represents the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer, O i This represents the output of the i-th neuron in the input layer, O j This represents the output of the j-th neuron in the hidden layer, where i represents the i-th neuron in the input layer and j represents the j-th neuron in the hidden layer. The relationship between the hidden layer and the output layer is as follows: Among them, net k w represents the net input to the k-th neuron in the output layer. jk This represents the connection weight from the j-th neuron in the hidden layer to the k-th neuron in the output layer, O k This represents the output of the k-th neuron in the output layer, where k represents the k-th neuron in the output layer. The activation function is: Where f(x) represents the activation function, and x represents the independent variable of the activation function; The derivative of the activation function is: f′(x)=1-f 2 (x) Here, f'(x) represents the derivative of the activation function.

5. The control method for a power factor correction circuit according to claim 4, characterized in that, When performing backpropagation of the error and updating the weights of the neural network according to the negative gradient direction, the expression is: in, Δw represents the root mean square error. jk (m) represents the weight change from the j-th neuron in the hidden layer to the k-th neuron in the output layer during the m-th iteration, Δw ij (m) represents the weight change from the i-th neuron in the input layer to the j-th neuron in the hidden layer in the m-th iteration, η represents the learning rate, α represents the inertia coefficient, and m represents the number of iterations; Calculate the derivative of the expression using the chain rule: Among them, O k (m) represents the output of the k-th neuron in the input layer during the m-th iteration, net k (m) represents the net input of the k-th neuron in the output layer during the m-th iteration, w jk (m) represents the connection weight from the j-th neuron in the hidden layer to the k-th neuron in the output layer in the m-th iteration. j (m) represents the output of the j-th neuron in the output layer during the m-th iteration. j (m) represents the net input of the j-th neuron in the output layer during the m-th iteration, w ij (m) represents the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer in the m-th iteration; Using the sgn symbolic function to find approximate solutions: Based on the approximate solution and the derivative of the activation function, calculate the network weights updated by backpropagation:

6. A control system for a power factor correction circuit, said system employing the method described in any one of claims 1-5, characterized in that, include: The module includes a BP neural network construction module, a PI controller parameter correction module, and a circuit control module. The BP neural network construction module adopts a dual closed-loop control mode of voltage outer loop and current inner loop, introduces the BP neural network into the current inner loop, and designs the control frequency of the BP neural network. The PI controller parameter correction module is used to correct the parameters of the PI controller using the BP neural network at a low control frequency of 50Hz. The circuit control module is used to control the power factor correction circuit at a high control frequency of 30kHz, based on the output of the voltage outer loop and using an optimized current inner loop PI controller.

7. The control system for a power factor correction circuit according to claim 6, characterized in that, The BP neural network consists of one input layer, one hidden layer, and one output layer; The input layer has 3 neuron nodes, and the maximum current error, root mean square error and current effective command are selected as parameters. The hyperbolic tangent function is selected as the activation function, and the input data is normalized. The hidden layer has four neuron nodes; The output layer has two neuron nodes, and the two neuron nodes of the output layer correspond to the parameter correction amount ΔK of the PI controller. p and ΔK i .

8. The control system for a power factor correction circuit according to claim 7, characterized in that, The network weights updated during backpropagation are obtained by utilizing the relationships between the hidden layer and the input layer, and between the hidden layer and the output layer of the BP neural network during forward propagation, thus completing the training of the network.

9. The control system for a power factor correction circuit according to claim 7, characterized in that, During forward propagation, the relationship between the hidden layer and the input layer of the BP neural network is as follows: Among them, net j w represents the net input to the j-th neuron in the hidden layer. ij This represents the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer, O i This represents the output of the i-th neuron in the input layer, O j This represents the output of the j-th neuron in the hidden layer, where i represents the i-th neuron in the input layer and j represents the j-th neuron in the hidden layer. The relationship between the hidden layer and the output layer is as follows: Among them, net k w represents the net input to the k-th neuron in the output layer. jk This represents the connection weight from the j-th neuron in the hidden layer to the k-th neuron in the output layer, O k This represents the output of the k-th neuron in the output layer, where k represents the k-th neuron in the output layer. The activation function is: Where f(x) represents the activation function, and x represents the independent variable of the activation function; The derivative of the activation function is: f′(x)=1-f 2 (x) Here, f'(x) represents the derivative of the activation function.

10. The control system for a power factor correction circuit according to claim 9, characterized in that, When performing backpropagation of the error and updating the weights of the neural network according to the negative gradient direction, the expression is: in, Δw represents the root mean square error. jk (m) represents the weight change from the j-th neuron in the hidden layer to the k-th neuron in the output layer during the m-th iteration, Δw ij (m) represents the weight change from the i-th neuron in the input layer to the j-th neuron in the hidden layer in the m-th iteration, η represents the learning rate, α represents the inertia coefficient, and m represents the number of iterations; Calculate the derivative of the expression using the chain rule: Among them, O k (m) represents the output of the k-th neuron in the input layer during the m-th iteration, net k (m) represents the net input of the k-th neuron in the output layer during the m-th iteration, w jk (m) represents the connection weight from the j-th neuron in the hidden layer to the k-th neuron in the output layer in the m-th iteration. j (m) represents the output of the j-th neuron in the output layer during the m-th iteration. j (m) represents the net input of the j-th neuron in the output layer during the m-th iteration, w ij (m) represents the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer in the m-th iteration; Using the sgn symbolic function to find approximate solutions: Based on the approximate solution and the derivative of the activation function, calculate the network weights updated by backpropagation:

Images