Medium-frequency induction heating power supply control method based on fuzzy neural network BP-PI control

Abstract

The invention provides a medium-frequency induction heating power supply control method based on fuzzy neural network BP-PI control, and belongs to the technical field of induction heating power supplies, and the method comprises the steps: collecting current and voltage signals in real time, and determining the real-time power; according to the difference value between the real-time power and the target power, a bridge arm of an LLC load resonance circuit is controlled through a PWM driving signal, and power adjustment is carried out; determining a resonant frequency, and obtaining a frequency error between the output frequency and the resonant frequency of the current full-bridge IGBT inverter circuit end and a frequency error change rate; and according to the frequency error and the frequency error change rate, outputting adjustment amounts of a proportionality coefficient and an integral coefficient of a PI controller through a fuzzy BP neural network, dynamically adjusting the proportionality coefficient and the integral coefficient, and determining a frequency adjustment amount of the LLC load resonance circuit for frequency tracking. According to the control method, the LLC load resonance circuit and the full-bridge IGBT inverter circuit are combined to realize medium-frequency heating, the problem of resonant frequency drift caused by workpiece heating and load change is solved, and the power supply efficiency and the power stability are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of induction heating power supply, in particular to a medium-frequency induction heating power supply control method based on fuzzy neural network BP-PI control. BACKGROUND

[0002] Induction heating technology is an advanced heating method based on electromagnetic induction principle and Joule heating effect, which can efficiently convert electrical energy into heat energy. The core process is: when an alternating current passes through the excitation coil, a high-density alternating magnetic field of the same frequency is generated around it; when metal or other conductive materials are placed in the magnetic field, a closed eddy current is induced inside the material, which follows the Joule-Lenz law and generates a large amount of heat to achieve rapid heating from the inside out. In addition, for ferromagnetic materials, the magnetic hysteresis loss caused by the continuous friction and rearrangement of the internal magnetic domains in the high-frequency magnetic field further contributes to the generation of heat. This non-contact heating method has a series of outstanding advantages such as high energy density, fast heating speed, excellent thermal efficiency, and easy to realize automatic control, so it is widely used in key industrial fields such as metal smelting, forging heat penetration, surface quenching, welding, and crystal growth.

[0003] However, the induction heating system, especially the high-efficiency power supply using LLC resonant topology, faces a key challenge in actual operation: the problem of resonant frequency dynamic drift caused by workpiece heating and load changes. During the heating process, the rapid rise in workpiece temperature will cause significant changes in its electromagnetic parameters (such as permeability and resistivity), and changes in the material, shape, number of heated workpieces, and relative position with the coil will directly manifest as time-varying characteristics of the equivalent inductance and resistance of the load. This dynamic uncertainty of the load makes the inherent resonant frequency of the resonant circuit a time-varying quantity. The existing fixed frequency or conventional PID control method is difficult to quickly and accurately track and compensate for this, resulting in system detuning. SUMMARY

[0004] To solve the above problems, the present application provides a medium-frequency induction heating power supply control method and system based on fuzzy neural network BP-PI control, which combines LLC load resonant circuit and full-bridge IGBT inverter circuit to realize medium-frequency heating and solve the problem of resonant frequency drift caused by workpiece heating and load changes, improving power efficiency and power stability.

[0005] To achieve the above purpose, the present application provides the following technical solutions.

[0006] A medium-frequency induction heating power supply control method based on fuzzy neural network BP-PI control, comprising the following steps: The current and voltage signals on the load side of the medium frequency induction heating power supply are collected in real time to determine the real-time power; the medium frequency induction heating power supply is connected to the full-bridge IGBT inverter circuit through the LLC load resonant circuit for medium frequency heating; The difference between the real-time power and the target power is used to control the bridge arm of the LLC load resonant circuit through the PWM drive signal to perform power regulation. The resonant frequency of the LLC load resonant circuit is determined, and the frequency error and the frequency error change rate between the current output frequency of the full-bridge IGBT inverter circuit and the resonant frequency are obtained. The adjustment amount of the proportional coefficient and the integral coefficient of the PI controller is output by the fuzzy BP neural network according to the frequency error and the frequency error change rate, and the proportional coefficient and the integral coefficient are dynamically adjusted to determine the frequency adjustment amount of the LLC load resonant circuit for frequency tracking.

[0007] Preferably, the power regulation by controlling the bridge arm of the LLC load resonant circuit through the PWM drive signal according to the difference between the real-time power and the target power comprises the following steps: The PWM drive signals of the bridge arms of the LLC load resonant circuit are determined according to the power difference; The two reference bridge arms of the LLC load resonant circuit are controlled to work in the 180° electric angle complementary driving state through the PWM drive signal, and the pulse of the phase-shifted bridge arm is delayed relative to the reference signal to embed a zero voltage window in the output voltage waveform to adjust the output voltage amplitude for power regulation.

[0008] Preferably, the network of the fuzzy BP neural network comprises an input feature preprocessing layer, a membership function calculation layer, a fuzzy rule reasoning layer, a consequent synthesis and parameter aggregation layer, and a de-fuzzification output layer connected in sequence. The input feature preprocessing layer comprises two parallel channels for inputting the frequency error and the frequency error change rate as two input variables for signal normalization and fuzzification; the membership function calculation layer is used to construct a Gaussian membership function according to the linguistic conversion of the two input variables after fuzzification; the fuzzy rule reasoning layer is used to construct a rule activation matrix through tensor product operation according to the Gaussian membership function to form a 25-dimensional rule strength vector; the consequent synthesis and parameter aggregation layer adopts Mamdani-type inference control parameter fuzzification to obtain an aggregated output fuzzy set according to the 25-dimensional rule strength vector; and the de-fuzzification output layer is used to convert the output fuzzy set distribution into accurate control parameters through an improved gravity method.

[0009] Preferably, the input feature preprocessing layer comprises two parallel channels for inputting the frequency error and the frequency error change rate as two input variables for signal normalization and fuzzification, and comprises the following steps: Two variables are defined and fuzzified for parallel input: ; ; ; wherein, represents the net input of the i-th neuron in the first layer, represents the output of the neuron, x1 is the frequency error of the input, and x2 is the frequency error rate, E(t) is the current error, E(t+1) is the error at the next moment, and T is the sampling period; , are the quantization factors of the frequency error and the frequency error rate, respectively.

[0010] Preferably, the membership function calculation layer is used to construct a Gaussian membership function according to the linguistic conversion of the two input variables after the fuzzification, including the following steps: Divide x 1, x 2 into five fuzzy subsets {NB, NS, ZO, PS, PB}, and the membership degree calculation expression is: ; ; wherein, is the net input of the i-th neuron in the second layer, is the output of the neuron; is the output of the neuron in the upper layer; and are the center parameter and the width parameter of the i-th fuzzy set of the input xi variable, respectively, forming an adjustable parameter matrix .

[0011] Preferably, the fuzzy rule reasoning layer is used to construct a rule activation matrix through a tensor product operation according to the Gaussian membership function, forming a 25-dimensional rule strength vector, including the following steps: The third layer constructs a rule activation matrix through a tensor product operation, forming a 25-dimensional rule strength vector, representing the premise satisfaction degree of each fuzzy rule, and the rule antecedent matching expression is: ; ; wherein, represents the net input of the i-th neuron in the first layer, i , , is the output of the neuron in the second layer.

[0012] Preferably, the consequent synthesis and parameter aggregation layer uses a Mamdani-type inference control parameter fuzzification output based on a 25-dimensional rule intensity vector to obtain an aggregated output fuzzy set, including the following steps: The control parameters are fuzzy output using Mamdani-type inference, and the consequent membership degree synthesis expression is as follows: ; ; In the formula, Indicates the first i Rule number 1 j The contribution weights of each output variable constitute the adjustable parameter matrix. ; Input to the 4th layer neurons, This is the output of the third layer of neurons. This is the output of the 4th layer neuron.

[0013] Preferably, the defuzzified output layer is used to transform the output fuzzy set distribution into precise control parameters using an improved centroid method, including the following steps: Accurate output is achieved through the improved centroid method. The parameter defuzzification calculation expression is as follows: ; ; In the formula, This is the quantization factor for the output membership function.

[0014] The present invention also provides a medium-frequency induction heating power supply control system based on fuzzy neural network BP-PI control, the system comprising: The signal acquisition module and the effective value detection circuit module are used to acquire the current and voltage signals on the load side of the medium frequency induction heating power supply in real time, as well as the current output frequency of the full-bridge IGBT inverter circuit. The signal processing module is used to obtain real-time power based on the current and voltage signals of the load side of the intermediate frequency induction heating power supply; it is also used to determine the frequency error between the current output frequency and the resonant frequency of the full-bridge IGBT inverter circuit and the rate of change of the frequency error. The PWM generation module is used to generate a PWM drive signal based on the difference between the real-time power and the target power. The frequency tracking module is used to control the bridge arm of the LLC load resonant circuit according to the PWM drive signal. The control power module is used to dynamically adjust the proportional and integral coefficients of the PI controller based on the frequency error and the rate of change of the frequency error through a fuzzy BP neural network, thereby determining the frequency regulation amount of the LLC load resonant circuit for frequency tracking.

[0015] The beneficial effects of this invention are: This invention proposes a control method for a medium-frequency induction heating power supply based on fuzzy neural network BP-PI control. This method combines inverter-side PWM power regulation with zero-voltage switching (ZVS) technology to achieve direct and precise control of the output power, avoiding detuning caused by large frequency shifts. ZVS technology effectively reduces switching losses and interference, laying the foundation for high-efficiency power conversion. This invention also proposes a fuzzy neural network BP-PI composite controller. This controller learns the dynamic characteristics of the load online through a neural network, optimizes the fuzzy rule base in real time, and adaptively tunes the PI controller parameters based on the optimized fuzzy rule base and the obtained frequency error and rate of change of frequency error. This gives the system the robustness of fuzzy logic, the self-learning ability of neural networks, and the steady-state accuracy of PI control.

[0016] This invention employs a fuzzy neural network BP-PI composite controller to identify and compensate for load disturbances in real time, outputting optimal control commands; the inverter-side PWM power regulation method executes the commands quickly and accurately, achieving precise power output and extremely low losses. This collaborative mechanism enables the system to dynamically follow changes in load impedance, automatically maintaining a resonant state and stable output power, significantly improving the system's adaptability, efficiency, and overall robustness under complex operating conditions. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a circuit structure diagram of the induction heating power supply according to an embodiment of the present invention; Figure 3 This is a structural diagram of the LLC resonant load main circuit according to an embodiment of the present invention; Figure 4 This is a steady-state operating waveform diagram of the system according to an embodiment of the present invention; Figure 5 This is the working stage of the LLC-type resonant load in this embodiment of the invention; Figure 6 This is a waveform diagram of the trigger pulse during PWM control according to an embodiment of the present invention; Figure 7 This is a closed-loop block diagram of a fuzzy BP-PI control system according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the fuzzy neural network structure according to an embodiment of the present invention; Figure 9 This is the inverter drive signal with a duty cycle of 30% in this embodiment of the invention, where (a) is the drive waveform and (b) is the load voltage and current waveform. Figure 10This is the inverter drive signal when the duty cycle is 60% in this embodiment of the invention, where (a) is the drive waveform and (b) is the load voltage and current waveform. Figure 11 This is a frequency tracking diagram under steady-state operating conditions according to an embodiment of the present invention; Figure 12 These are voltage and current waveforms at different times according to an embodiment of the present invention, wherein (a) is the load voltage and current waveform at 0.05s, and (b) is the load voltage and current waveform after 0.2s. Figure 13 This is a frequency tracking diagram of inductance changes under dynamic operating conditions according to an embodiment of the present invention; Figure 14 This is a power control output diagram of an embodiment of the present invention, wherein (a) is a 10kW power output waveform and (b) is a 20kW power output waveform. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] Example 1 This invention proposes a control method for a medium-frequency induction heating power supply based on fuzzy neural network BP-PI control, the specific steps of which are as follows: Figure 1 As shown, it includes: S1: Real-time acquisition of current and voltage signals on the load side of the medium-frequency induction heating power supply to determine the real-time power; the medium-frequency induction heating power supply performs medium-frequency heating through an LLC load resonant circuit and a full-bridge IGBT inverter circuit.

[0020] S2: Based on the difference between the real-time power and the target power, the bridge arm of the LLC load resonant circuit is controlled by the PWM drive signal to perform power regulation.

[0021] S3: Determine the resonant frequency of the LLC load resonant circuit, and obtain the frequency error and frequency error change rate between the current full-bridge IGBT inverter circuit output frequency and the resonant frequency.

[0022] S4: Based on the frequency error and the rate of change of the frequency error, the fuzzy BP neural network outputs the adjustment amount of the proportional coefficient and integral coefficient of the PI controller, dynamically adjusts the proportional coefficient and integral coefficient, and determines the frequency regulation amount of the LLC load resonant circuit for frequency tracking.

[0023] Specifically, the circuit structure diagram of the induction heating power supply of the present invention is as follows: Figure 2 As shown. The main circuit structure of the LLC resonant load is as follows. Figure 3As shown, this architecture consists of a voltage-source full-bridge inverter circuit and... Figure 2 The LLC resonant load network is constructed as described. In actual operation, the power switching transistors have turn-on / turn-off delays, and a reasonable dead time needs to be set to avoid bridge arm shoot-through faults. Figure 4 The voltage and current waveforms at key nodes under steady-state conditions are shown: S 1. S 4 and S 2. S 3. Alternating conduction, inverter output current I MN Lag voltage U MN Each work cycle can be divided into four equivalent time periods. t 0~ t The working principle of the 4-period analysis.

[0024] Figure 5 The four core operating stages of an LLC resonant load are described: (1) t 0~ t Phase 1 ( S 1. S 4 conduction, S 2. S 3. Shutdown). For example... Figure 5 (a), t The current flows through before time 0 S 1. S 4. With diode in flux, the diode voltage drop is close to the diode's forward voltage drop. Upon receiving the drive signal... S 1. S 4. Achieve zero-voltage turn-on (ZVS). Output current during this stage. I MN Phase lags behind voltage U MN The hysteresis period varies with the load characteristics.

[0025] (2) t 1~ t Phase 2 (full bridge shutdown, dead time). For example... Figure 5 (b) To prevent short circuits in the bridge arms, a dead time is set to force the entire bridge to shut down. At this time, a parallel capacitor is connected. C 1 and C 4. Enter charging state. C 2 and C 3. Discharge is performed to establish a voltage gradient for subsequent switching actions.

[0026] (3) t 2~ t 3-stage (continuation mode). For example... Figure 5 (c) After the capacitor has finished charging and discharging, the current flows through...D 2 and D 3. Continued flow, for S 2 and S 3. Create ZVS conditions. During this stage, the magnetizing inductor maintains residual current, forming the energy buffer required for soft switching.

[0027] (4) t 3~ t 4 stages ( S 1. S 4. Turn off S 2. S 3. Conductivity). For example... Figure 5 (d) The working mode of the second half of the cycle is symmetrical with that of the first half of the cycle. The current path and voltage polarity are opposite, thus completing a complete energy transfer cycle.

[0028] As described in step S2, the present invention regulates power by controlling the bridge arm of the LLC load resonant circuit through a PWM drive signal based on the difference between the real-time power and the target power. Specifically:

[0029] This control strategy achieves precise control of the equivalent output voltage amplitude by embedding an adjustable-width zero-voltage window into the output voltage waveform. The trigger pulse waveform during PWM control is as follows: Figure 6 As shown, the horizontal axis represents time. Main switch transistor. S 1 / S 2 forms the reference bridge arm, whose drive signal leading edge interval is fixed at 180° electrical angle; bridge arm S 3 / S The drive pulse of 4 is delayed by an angle α relative to the reference signal. This control architecture achieves soft-switching characteristics through zero-voltage switching (ZVS) or zero-current switching (ZCS) mechanisms, reducing energy loss by approximately 60% and electromagnetic interference intensity by more than 40dB. Not only does it cover a power regulation dynamic range of 10%-100% of rated power, but it also shortens the response time to the millisecond level and improves the power factor to above 0.98. This effectively solves the power factor limitation problem of traditional DC-side voltage regulation schemes and the high-frequency loss problem in PFM mode, making it the preferred technical solution of this invention.

[0030] As described in steps S3-S4, the present invention employs a fuzzy BP-PI frequency tracking strategy, specifically: In resonant systems such as induction heating and wireless power transmission, dynamic changes in load parameters can cause resonant frequency shifts (Δω). Traditional PID controllers with fixed parameters struggle to achieve rapid frequency tracking. The core design goals of the fuzzy BP-PI controller are: real-time detection of resonant frequency errors; dynamic adjustment of PI parameters through fuzzy inference; and self-optimization of control rules using the BP algorithm to improve tracking accuracy.

[0031] The structure of the fuzzy BP-PI control system proposed in this invention is as follows: Figure 7 As shown in the figure. E and EC represent the error of the given frequency and the output frequency, respectively, as well as the rate of change of the frequency error. The system structure mainly consists of two parts:

[0032] PI controller: Updates the proportional coefficient in real time based on the adjustment values ​​provided by the fuzzy inference system and the BP neural network optimizer. K p ), integral coefficient ( K i The updated PI parameters are used to control the resonant system, thereby reducing frequency error and improving system performance.

[0033] Fuzzy BP Neural Network: This fuzzy neural network extracts the rate of change of E and EC through feature extraction and adjusts the PI parameter based on the backpropagation algorithm. K p , K i The controller parameters are dynamically adjusted to achieve precise control.

[0034] The following section will mainly introduce the structure of fuzzy neural networks (FNNs): the BP-PI combined neural network structure is as follows... Figure 8 As shown, it is divided into five layers.

[0035] (1) Input feature preprocessing layer The first layer contains two parallel input channels to perform signal normalization and fuzzification processing. Input variable definitions: ; ; ; In the formula, the symbol Indicates the first j The first in the layer i Net input to each neuron, This represents the output of the neuron. x 1 represents the input variable, which is the frequency deviation. x 2 represents the rate of change of frequency deviation. , , E ( t ) represents the current error. E ( t +1) represents the error at the next time step. T The sampling period.

[0036] (2) Membership function calculation layer The second layer implements the linguistic transformation of input variables and constructs Gaussian membership functions. x 1,x 2 is divided into five fuzzy subsets {NB, NS, ZO, PS, PB}. The membership degree calculation expression is as follows: ; ; In the formula, and Inputs x i Variable number i The center parameters and width parameters of each fuzzy set constitute an adjustable parameter matrix. .

[0037] (3) Fuzzy rule reasoning layer The third layer constructs a rule activation matrix through tensor product operations, forming a 25-dimensional rule strength vector that represents the precondition satisfaction degree of each fuzzy rule. The rule antecedent matching expression is:

[0038] ; ; (4) Subsequent synthesis and parametric polymerization layer The fourth layer uses Mamdani-type inference to achieve fuzzy output of control parameters, and the consequent membership degree synthesis expression is: ; ; In the formula, The weights representing the contribution of the i-th rule to the j-th output variable form the adjustable parameter matrix. .

[0039] (5) Defuzzification of the output layer The fifth layer, serving as the terminal processing module for fuzzy inference, primarily transforms the fuzzy membership distribution output from the fourth layer into precise control parameters. An improved centroid method is used to achieve precise output; the parameter defuzzification expression is as follows:

[0040] ; ; In the formula, To output the quantization factor of the membership function, it is calibrated here. .

[0041] The above is one embodiment of the medium-frequency induction heating power supply control method based on fuzzy neural network BP-PI control provided in this embodiment. Based on the same idea, this embodiment also provides a corresponding medium-frequency induction heating power supply control system based on fuzzy neural network BP-PI control, specifically including: The signal acquisition module and the RMS detection circuit module are used to acquire the current and voltage signals on the load side of the medium-frequency induction heating power supply in real time, as well as the current output frequency of the full-bridge IGBT inverter circuit.

[0042] The signal processing module is used to obtain real-time power based on the current and voltage signals of the load side of the intermediate frequency induction heating power supply; it is also used to determine the frequency error between the current output frequency and the resonant frequency of the full-bridge IGBT inverter circuit and the frequency error change rate.

[0043] The PWM generation module is used to generate a PWM drive signal based on the difference between the real-time power and the target power.

[0044] The frequency tracking module is used to control the bridge arm of the LLC load resonant circuit according to the PWM drive signal.

[0045] The control power module is used to dynamically adjust the proportional and integral coefficients of the PI controller based on the frequency error and the rate of change of the frequency error through a fuzzy BP neural network, thereby determining the frequency regulation amount of the LLC load resonant circuit for frequency tracking.

[0046] Specifically, based on the above control scheme design, this invention presents an induction heating power supply control system. It employs an LLC resonant load circuit, power regulation is achieved through a PWM modulation strategy, and frequency tracking is accomplished using a PI control algorithm optimized with an integrated fuzzy neural network. The medium-frequency induction heating power supply system consists of two main parts: a main circuit and a control circuit. The main circuit includes a rectifier and filter circuit, an inverter circuit, and an LLC resonant load circuit. The control circuit comprises a control core composed of a DSP and an FPGA, a signal processing circuit, a protection circuit, a signal acquisition and RMS detection circuit, and a human-machine interface. The human-machine interface provides two interfaces: one is a touchscreen, and the other connects to a host computer via CAN communication. These components work together to achieve precise control and monitoring of the induction heating process. The overall system block diagram is shown below. Figure 2 As shown.

[0047] The system uses a DSP as the main controller. The process begins with the DSP initialization phase after system power-on, covering clock tree configuration, peripheral register mapping, and interrupt priority setting. This is followed by a system self-test, which diagnoses hardware status by detecting IGBT drive levels, the equivalent impedance of the resonant capacitor bank, and temperature sensor feedback. If the self-test fails, a fault display is triggered and the system is locked; if successful, a timer interrupt service is initiated, simultaneously executing signal acquisition and RS485 communication. Within the interrupt subroutine, the system dynamically analyzes power deviation using a fuzzy BP-PI algorithm, generating a PWM output control signal to adjust the phase angle of the full-bridge inverter circuit. The duty cycle adjustment step size is limited by an integral separation strategy to ensure that the overshoot remains stable within a set threshold. The control loop continuously checks for the end of metal processing. If the process parameters are not met, the program returns to the signal acquisition phase for iterative optimization; if the preset termination conditions are met, the PWM output is turned off and the control loop exits.

[0048] This invention establishes a simulation model of a medium-frequency induction heating power supply system in Simulink. The model consists of two parts: a main circuit and a control circuit. The main circuit uses an equivalent model of a three-phase uncontrolled rectifier filter, an IGBT full-bridge inverter, a medium-frequency transformer, and an LLC matching network. The control circuit includes PWM generation, frequency tracking, and power control modules. Specific simulation parameters are detailed in Table 1.

[0049] Table 1. Simulation parameters of the medium-frequency induction heating power supply system PWM simulation results and analysis: To verify the duty cycle power regulation function, the intermediate frequency induction heating power supply in resonant state was tested. When the duty cycle is 30%, the inverter drive signal is as follows: Figure 9 As shown in (a), PWM1 and PWM4 are fixed bridge arms, and PWM2 and PWM3 are shifting bridge arms. The high-level duration in each cycle is approximately 15% of the total cycle. The load voltage and current waveforms are as follows: Figure 9 As shown in (b), the conduction angle of both the positive and negative half-cycles of the voltage waveform is 180°, while the current waveform exhibits both upper and lower half-wave conduction states, with an amplitude significantly lower than that of the voltage waveform, conforming to impedance characteristics. When the duty cycle increases to 50%, the drive signal is as follows: Figure 10 As shown in (a), the output voltage and current waveforms at this time are as follows: Figure 10 As shown in (b), the zero-voltage state decreases, while the output voltage and current amplitudes increase. The voltage and current waveforms work in tandem, and the system is in a resonant and efficient state, verifying the feasibility of duty cycle power regulation.

[0050] Frequency tracking simulation analysis: To verify the frequency adaptive performance of the fuzzy neural network BP-PI algorithm in a medium-frequency induction heating power supply, tests need to be conducted on both steady-state regulation accuracy and dynamic disturbance rejection capability. This invention uses simulation analysis to compare the frequency tracking characteristics of traditional PI control, fuzzy PI control, and the improved BP-PI algorithm.

[0051] (1) Frequency locking under steady-state conditions The model was simulated and analyzed based on the parameters in Table 1. The initial frequency of the system was set to 1700Hz, the theoretical resonant frequency was 1983Hz, and the total simulation time was 1 second.

[0052] Figure 11 The frequency tracking curves of three algorithms are shown. Traditional PI control experiences frequency overshoot within 0.1-0.2 seconds, with a settling time as long as 0.3 seconds; fuzzy PI control takes 0.3 seconds to achieve frequency adjustment; while the BP-PI algorithm, which integrates fuzzy inference and neural networks, exhibits a frequency curve that rises smoothly exponentially within 0-0.1 seconds, stabilizing at the target value in just 0.1 seconds with no overshoot throughout. As the figure shows, the BP-PI algorithm improves response speed by 60% compared to traditional methods and has a lower steady-state error.

[0053] Figure 12 (a) The display shows that in the initial stage t < 0.05 seconds, the phase difference between the load voltage and current reaches 45°, indicating that the system is in a detuned state; if Figure 12 As shown in (b), after adjustment by the BP-PI algorithm, the phase difference is basically 0° after 0.2 seconds, verifying that the resonance point is successfully locked.

[0054] (2) Dynamic operating condition disturbance immunity test To simulate the sudden change in inductance caused by temperature drift, an inductance compensation module is configured in the load circuit to increase the inductance L2 to 2uH. A step change in the inductance L2 value is achieved by switching, corresponding to a resonant frequency offset Δf = 85Hz, meaning the LLC type load resonant frequency is 1903Hz.

[0055] The initial resonant frequency was 1988Hz, and the dynamic inductance L1 was 21.877uH. After 0.25 seconds, the dynamic inductance switched to 23.877uH, and the theoretical resonant frequency dropped to 1903Hz. Figure 13 (a) shows that the BP-PI algorithm completes frequency relocking within 30ms through online weight correction.

[0056] During reverse switching, the dynamic inductance reaches 19.877uH in 0.25 seconds, increasing the theoretical resonant frequency to 2079Hz. The algorithm rapidly reconstructs the control rules based on historical data, reducing the adjustment time to 25ms. The frequency tracking curve is shown below. Figure 13 As shown in (b).

[0057] Depend onFigure 13 (a) and Figure 13 (b) It can be seen that before 0.05 seconds, the frequency tracking process is basically the same as the frequency tracking process under static conditions. When the parameters suddenly change, the frequency tracking algorithm controlled by the fuzzy neural network BP-PI can still quickly track the new resonant frequency point within 30ms and match the theoretical calculation value.

[0058] Power control simulation analysis: To verify the effectiveness of the proposed power control algorithm, a target power value was set to drive dynamic adjustment of the duty cycle, and a system was established as follows: Figure 14 The power point tracking experiment shown. (As shown in the image) Figure 14 As shown in (a), when the set power is 10kW, the system achieves a continuous power increase to the target value within a 150ms adjustment time, after which the power remains basically stable in the steady-state phase. Figure 14 As shown in (b), after the target power is increased to 20kW, the system response time is extended to 200ms, the output steady-state power reaches the target set value and is basically stable at 20kW.

[0059] Experiments show that the target power setpoint is nonlinearly positively correlated with the system's dynamic response time, and the steady-state fluctuation amplitude is related to the power magnitude. The experimental results fully verify the effectiveness and robustness of the proposed power control algorithm over a wide power range.

[0060] This invention simulates PWM, frequency tracking algorithm, and power regulation on the Simulink platform. The frequency tracking algorithm can track the series resonant frequency and find a new resonant point when the load changes abruptly; the power control algorithm can quickly reach and stably maintain the set power. Simulation results verify the effectiveness of the theoretical analysis. Furthermore, this invention also built a physical system and conducted actual tests on the drive signal, automatic frequency tracking, and power regulation functions. The results met the design expectations, proving the feasibility of the solution.

[0061] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A control method for a medium-frequency induction heating power supply based on fuzzy neural network BP-PI control, characterized in that, Includes the following steps: The current and voltage signals on the load side of the medium-frequency induction heating power supply are acquired in real time to determine the real-time power; the medium-frequency induction heating power supply performs medium-frequency heating through an LLC load resonant circuit and a full-bridge IGBT inverter circuit. Based on the difference between the real-time power and the target power, the bridge arm of the LLC load resonant circuit is controlled by the PWM drive signal to perform power regulation. Determine the resonant frequency of the LLC load resonant circuit, and obtain the frequency error and the rate of change of the frequency error between the current full-bridge IGBT inverter circuit output frequency and the resonant frequency. Based on the frequency error and the rate of change of the frequency error, the fuzzy BP neural network outputs the adjustment amounts of the proportional coefficient and integral coefficient of the PI controller. The proportional coefficient and integral coefficient are dynamically adjusted to determine the frequency regulation amount of the LLC load resonant circuit for frequency tracking.

2. The medium-frequency induction heating power supply control method based on fuzzy neural network BP-PI control according to claim 1, characterized in that, The method of adjusting the power by controlling the bridge arm of the LLC load resonant circuit with a PWM drive signal based on the difference between the real-time power and the target power includes the following steps: The PWM drive signal for each bridge arm of the LLC load resonant circuit is determined based on the power difference. The two reference bridge arms of the LLC load resonant circuit are controlled by the PWM drive signal to operate in a 180° electrical angle complementary drive state, so that the pulse of the phase-shifting bridge arm is delayed relative to the reference signal, so as to embed a zero voltage window in the output voltage waveform to adjust the output voltage amplitude for power regulation.

3. The medium-frequency induction heating power supply control method based on fuzzy neural network BP-PI control according to claim 1, characterized in that, The fuzzy BP neural network includes, in sequence, an input feature preprocessing layer, a membership function calculation layer, a fuzzy rule inference layer, a consequent synthesis and parameter aggregation layer, and a defuzzified output layer. The input feature preprocessing layer includes two parallel channels for input frequency error and frequency error change rate as input variables, used for signal normalization and fuzzification. The membership function calculation layer constructs Gaussian membership functions based on the linguistic transformation of the two fuzzified input variables. The fuzzy rule inference layer constructs a rule activation matrix based on the Gaussian membership functions through tensor product operations, forming a 25-dimensional rule strength vector. The consequent synthesis and parameter aggregation layer uses the 25-dimensional rule strength vector and Mamdani-type inference to fuzzify the output control parameters, obtaining the aggregated output fuzzy set. The defuzzification output layer transforms the distribution of the output fuzzy set into precise control parameters using an improved centroid method.

4. The medium-frequency induction heating power supply control method based on fuzzy neural network BP-PI control according to claim 3, characterized in that, The input feature preprocessing layer includes two parallel channels for input frequency error and frequency error change rate as input variables, used for signal normalization and fuzzification processing, including the following steps: Define and fuzzify the two variables for parallel input: ； ； ； In the formula, Indicates the first layer. i Net input to each neuron, This represents the output of the neuron. x 1 represents the input frequency error. x 2 represents the rate of change of frequency error. , E ( t ) represents the current error. E ( t +1) represents the error at the next time step. T The sampling period; , These are the quantization factors for frequency error and the rate of change of frequency error, respectively.

5. The medium-frequency induction heating power supply control method based on fuzzy neural network BP-PI control according to claim 4, characterized in that, The membership function calculation layer is used to construct Gaussian membership functions based on the linguistic transformation of the two input variables after fuzzification, including the following steps: Will x 1, x The two sets are divided into five fuzzy subsets: {NB, NS, ZO, PS, PB}. The membership degree calculation expression is as follows: ； ； In the formula, This is the net input to the i-th neuron in the second layer. This is the output of the neuron; This is the output of the upper-layer neurons; and Inputs x i Variable number i The center parameters and width parameters of each fuzzy set constitute an adjustable parameter matrix. .

6. The medium-frequency induction heating power supply control method based on fuzzy neural network BP-PI control according to claim 5, characterized in that, The fuzzy rule inference layer is used to construct a rule activation matrix based on Gaussian membership functions through tensor product operations, forming a 25-dimensional rule intensity vector, including the following steps: The third layer of tensor product operations constructs the rule activation matrix, forming a 25-dimensional rule strength vector, which represents the precondition satisfaction degree of each fuzzy rule. The rule antecedent matching expression is: ； ； In the formula, Indicates the first layer. i Net input to each neuron, , This is the output of the second layer of neurons.

7. The medium-frequency induction heating power supply control method based on fuzzy neural network BP-PI control according to claim 6, characterized in that, The consequent synthesis and parameter aggregation layer, based on a 25-dimensional rule intensity vector, uses Mamdani-type inference control parameters to fuzzify the output, obtaining the aggregated output fuzzy set, including the following steps: The control parameters are fuzzy output using Mamdani-type inference, and the consequent membership degree synthesis expression is as follows: ； ； In the formula, Indicates the first i Rule number 1 j The contribution weights of each output variable constitute the adjustable parameter matrix. ; Input to the 4th layer neurons, This is the output of the third layer of neurons. This is the output of the 4th layer neuron.

8. The medium-frequency induction heating power supply control method based on fuzzy neural network BP-PI control according to claim 7, characterized in that, The defuzzification output layer is used to transform the output fuzzy set distribution into precise control parameters using an improved centroid method, including the following steps: Accurate output is achieved through the improved centroid method. The parameter defuzzification calculation expression is as follows: ； ； In the formula, This is the quantization factor for the output membership function.

9. A medium-frequency induction heating power supply control system based on fuzzy neural network BP-PI control, characterized in that, The system includes: The signal acquisition module and the effective value detection circuit module are used to acquire the current and voltage signals on the load side of the medium frequency induction heating power supply in real time, as well as the current output frequency of the full-bridge IGBT inverter circuit. The signal processing module is used to obtain real-time power based on the current and voltage signals of the load side of the intermediate frequency induction heating power supply; it is also used to determine the frequency error between the current output frequency and the resonant frequency of the full-bridge IGBT inverter circuit and the rate of change of the frequency error. The PWM generation module is used to generate a PWM drive signal based on the difference between the real-time power and the target power. The frequency tracking module is used to control the bridge arm of the LLC load resonant circuit according to the PWM drive signal. The control power module is used to dynamically adjust the proportional and integral coefficients of the PI controller based on the frequency error and the rate of change of the frequency error through a fuzzy BP neural network, thereby determining the frequency regulation amount of the LLC load resonant circuit for frequency tracking.

Images