Magnetic levitation controller device and filtering method against electromagnetic interference

Abstract

The application belongs to the technical field of electric machines, and discloses a magnetic suspension controller device and a filtering method for resisting electromagnetic interference; the method comprises the following steps: preparing a quantum limited magnetic guide unit, collecting the internal magnetic flux and temperature data of the magnetic core in real time through an embedded sensor array, and building a corresponding magnetic permeability regulation model; obtaining the output of the magnetic permeability regulation model, and analyzing to obtain real-time inductance; based on the real-time inductance, combining a preset target inductance, and compensating the real-time inductance through a gate voltage regulation and inductance compensation collaborative strategy; filtering the compensated real-time inductance by using a PID controller; under the coupling environment of a strong magnetic field and high temperature, the application can effectively reduce the nonlinear decline amplitude of the magnetic permeability, reduce the inductance value drift, filter cutoff frequency offset and current ripple, improve the rotor suspension precision, and significantly improve the filtering stability and control reliability of the magnetic suspension system in extreme environments.

Description

Technical Field

[0001] This invention relates to the field of motor technology, and more specifically, to a magnetic levitation controller device and filtering method for resisting electromagnetic interference. Background Technology

[0002] Magnetic levitation systems achieve contactless levitation through magnetic field force. Their core control unit needs to process weak feedback signals to maintain stable rotor levitation, making it extremely sensitive to electromagnetic interference. The controller's filtering module is crucial for suppressing EMI—using a filter composed of specific inductors and capacitors, high-frequency interference signals can be attenuated, ensuring the signal-to-noise ratio of the control loop.

[0003] However, the core material of the filter inductor has significant drawbacks in the strong magnetic field and high-temperature coupling environment of a magnetic levitation system: on the one hand, the strong magnetic field generated during system operation causes disordered magnetic domain arrangement in the core, leading to a nonlinear decrease in permeability; on the other hand, the high temperature (e.g., -40℃ to 125℃) during long-term operation of the equipment exacerbates the thermomagnetic loss of the core, further deteriorating the stability of permeability. Under the coupling effect of these two factors, the inductance value fluctuates significantly with the permeability drift, directly affecting the filter cutoff frequency.

[0004] Chinese patent application CN116317717A discloses an anti-interference power supply device, a motor control device, and a control method thereof. The method involves setting a filter module body on the output side of the power supply and a control module body between the output side of the filter module body and the power supply side of the load. The system determines whether the load needs power to start and operate. If the load needs power, the control module body is turned on to connect the power supply path between the power supply and the load. If the load does not need power, the control module body is turned off to disconnect the power supply path between the power supply and the load. This invention, by setting modular filter modules and switch control modules between the power supply and the motor, allows users to flexibly select and discard filter modules and switch control modules according to actual needs, greatly improving the applicability of the filter modules and switch control modules between the power supply and the motor.

[0005] While the above methods can meet the needs of most scenarios, research and practical application of these methods and existing technologies have revealed at least the following shortcomings:

[0006] The above methods focus on suppressing external electromagnetic interference, but cannot solve the problem that the magnetic permeability of the filter inductor core decreases nonlinearly under the coupling effect of strong magnetic field and high temperature, causing the inductance value to drift, which in turn causes the filter cutoff frequency to shift and the resonant point to drift. Interference signals can then enter the magnetic levitation controller through the control loop, resulting in increased current ripple and decreased rotor levitation accuracy.

[0007] In view of this, the present invention proposes a magnetic levitation controller device and filtering method to resist electromagnetic interference in order to solve the above problems. Summary of the Invention

[0008] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a magnetic levitation controller filtering method for electromagnetic interference resistance, comprising the following steps:

[0009] Quantum confined magnetic permeability units were fabricated, and magnetic flux and temperature data inside the magnetic core were collected in real time through an embedded sensor array to construct a corresponding permeability control model.

[0010] The output of the permeability control model is obtained, and the real-time inductance is analyzed.

[0011] Based on the real-time inductance and combined with the preset target inductance, the real-time inductance is compensated through a coordinated strategy of gate voltage control and inductance compensation.

[0012] Based on the compensated real-time inductance, a PID controller is used for filtering.

[0013] Furthermore, the method for fabricating a quantum-confined magnetic permeability unit containing an embedded sensor array includes:

[0014] A superlattice composed of a preset material is selected as the main body of the magnetic core;

[0015] Pretreated SiO2 is used as a substrate for a superlattice. The superlattice is alternately grown on the SiO2 substrate through a preset growth process to form a superlattice with a preset period. During the superlattice growth process, a mask is used to block the superlattice and a preset size of interlayer gap is reserved at the interface of the superlattice with a preset period. A preset chip is embedded in the interlayer gap and connected by a conductive polymer to form an embedded sensor array.

[0016] An Au layer is deposited by evaporation on the superlattice end face through electron beam lithography, and then gold wire gates with a preset linewidth are obtained by etching, forming a substrate-superlattice-gold electrode structure to obtain a quantum confined magnetic permeability unit.

[0017] Furthermore, methods for constructing permeability modulation models include:

[0018] Calculate the fundamental quantum tunneling probability;

[0019] The vertical electric field generated is calculated based on the gate voltage and the single-layer barrier thickness. The single-layer barrier thickness is then corrected based on the vertical electric field and the electron charge to obtain the corrected barrier thickness.

[0020] The quantum tunneling probability is calculated based on the basic quantum tunneling probability, the single-layer barrier thickness, the modified barrier thickness, the tunneling probability attenuation factor correction value, and the temperature correction term; the tunneling probability attenuation factor correction value is obtained by correcting the tunneling probability attenuation factor based on the magnetic flux density inside the magnetic core.

[0021] The permeability control model is constructed by obtaining the zero-magnetic-field reference permeability, quantum tunneling probability, coupling coefficient, modified barrier thickness, gate voltage sensitivity coefficient, gate voltage, magnetic field attenuation coefficient, and magnetic flux density inside the magnetic core.

[0022] Furthermore, methods for obtaining the fundamental quantum tunneling probability include:

[0023] The electron wave vector within the potential barrier is calculated based on electron energy, barrier thickness, Dirac constant, and effective electron mass; the electron wave vector within the graphene layer is calculated based on electron energy, Dirac constant, and effective electron mass; and the fundamental quantum tunneling probability is calculated based on the electron wave vector within the potential barrier, the electron wave vector within the graphene layer, and the single-layer barrier thickness.

[0024] Furthermore, methods for obtaining the gate voltage sensitivity coefficient include:

[0025] With a fixed magnetic field and temperature, the gate voltage is changed according to a preset step size by a gallium nitride high electron mobility transistor, and the corresponding permeability is measured. The difference between the corresponding permeability and the zero magnetic field reference permeability is calculated. Combined with the modified barrier thickness and gate voltage, the coupling relationship between the difference and the gate voltage sensitivity coefficient is fitted. The gate voltage sensitivity coefficient is obtained by fitting using the least squares method.

[0026] Furthermore, methods for compensating for real-time inductance include:

[0027] Set up K groups of hollow coils, which are switched using GaN switches;

[0028] When the difference between the real-time inductance and the reference value is not greater than the preset drift threshold, the flux error change rate within the preset time period is calculated, and a state vector is defined based on the real-time flux density, real-time temperature, flux error, and error change rate. The PID parameters used to adjust the gate voltage are used as the action vector. A reward function consisting of a quantum tunneling probability term, a multi-physics coupling penalty term, and a gate voltage energy consumption penalty is constructed. The optimal action is obtained through iterative training. The gate voltage is adjusted based on the PID parameters in the optimal action.

[0029] When the difference between the real-time inductance and the reference value is greater than the preset drift threshold, the corresponding coil is activated according to the difference, and the real-time inductance is adjusted until the difference is lower than the preset safety threshold.

[0030] Furthermore, methods for obtaining the reward function include:

[0031] The quantum tunneling probability term is calculated based on the quantum tunneling probability, the target magnetic flux density, and the magnetic flux error.

[0032] The multiphysics coupling penalty term is calculated based on the current temperature, current magnetic flux density, temperature reference value, and magnetic flux density reference value.

[0033] The gate voltage energy consumption penalty is calculated based on the gate voltage and the maximum gate voltage.

[0034] Furthermore, methods for obtaining the optimal action include:

[0035] The agent randomly draws samples from the experience pool and outputs actions through an actor network built from W layers of fully connected neural networks; the experience replay pool consists of Y sets of sample data; each set of samples includes the current state vector, action, reward, and the state vector for the next time step;

[0036] Substitute the PID parameters corresponding to the action into the gate voltage control formula and apply it to the superlattice;

[0037] Collect the state and reward of the next time step, and update the Q value of the critic network. The Q value is calculated based on the reward function, the maximum Q value of the corresponding action selected in the next state, and a discount factor, which is obtained through experimental optimization.

[0038] The actor network parameters are optimized using gradient descent with the goal of maximizing the Q-value.

[0039] The target network is updated at preset step intervals until the reward converges to a preset reward interval. The action corresponding to the maximum Q value when the reward converges to the preset reward interval is taken as the optimal action.

[0040] Furthermore, methods for obtaining the inductor drift quantization model include:

[0041] The initial inductance, real-time permeability, real-time magnetic flux density, real-time temperature, corrected barrier thickness, total core length, and quantum tunneling probability are obtained from the output of the permeability control model. An inductance drift quantization model is established to calculate the real-time inductance.

[0042] Furthermore, methods for filtering using a PID controller include:

[0043] By combining the fixed capacitance value with the real-time inductance after compensation, the filter cutoff frequency is calculated.

[0044] A PID controller is used to make the PWM carrier frequency track the filter cutoff frequency for filtering.

[0045] An electromagnetic interference-resistant magnetic levitation controller device, used to implement the electromagnetic interference-resistant magnetic levitation controller filtering method, includes:

[0046] Acquisition and processing module: fabricates quantum-confined magnetic permeability units, and collects magnetic flux and temperature data inside the magnetic core in real time through an embedded sensor array to construct a corresponding permeability control model;

[0047] Inductance analysis module: acquires the output of the permeability control model and analyzes it to obtain the real-time inductance;

[0048] Inductance compensation module: Based on real-time inductance and combined with preset target inductance, it compensates for real-time inductance through a cooperative strategy;

[0049] Compensation and filtering module: Based on the real-time inductance after compensation, a PID controller is used for filtering.

[0050] The technical effects and advantages of the electromagnetic interference-resistant magnetic levitation controller device and filtering method of this invention are as follows:

[0051] This invention selects a pre-defined superlattice material as the core body, leveraging the quantum confinement effect to fundamentally enhance the core's resistance to strong magnetic fields and high temperatures. Combined with a pre-defined growth process and real-time magnetic flux and temperature acquisition via an embedded sensor array, it provides dual inputs of microscopic quantum states and macroscopic environmental parameters for permeability control. Through an inductance drift quantization model integrating parameters such as quantum tunneling probability, modified barrier thickness, and real-time permeability, it accurately captures the nonlinear decrease in permeability, achieving full-dimensional tracing from microscopic quantum state fluctuations to macroscopic inductance drift. Furthermore, it employs rapid switching between K sets of hollow coils and GaN switches to handle large amplitude fluctuations. Inductor drift is addressed by combining a reinforcement learning PID strategy based on quantum state stability terms to achieve fine-tuning of the gate voltage. Finally, the PID controller drives the PWM carrier frequency to track the cutoff frequency of the compensated inductor in real time, effectively preventing the filter cutoff frequency from deviating from the resonant point and blocking the path of interference signals into the magnetic bearing controller. Under strong magnetic field and high temperature coupling environments, this invention can effectively reduce the nonlinear decrease in permeability, reduce inductor drift, filter cutoff frequency deviation, and current ripple, improve rotor levitation accuracy, and significantly enhance the filtering stability and control reliability of the magnetic levitation system under extreme environments. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the electromagnetic interference-resistant magnetic levitation controller filtering method of the present invention;

[0053] Figure 2 This is a schematic diagram of the data flow in this invention;

[0054] Figure 3 This is a schematic diagram of the method for constructing a permeability control model according to the present invention;

[0055] Figure 4 This is a schematic diagram of the electromagnetic interference-resistant magnetic levitation controller device of the present invention. Detailed Implementation

[0056] 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.

[0057] Example 1:

[0058] Please see Figure 1 , Figure 2 As shown, this embodiment provides a filtering method for a magnetic levitation controller to resist electromagnetic interference, including the following steps:

[0059] A quantum confined magnetic permeability unit containing an embedded sensor array was fabricated to collect magnetic flux and temperature data inside the magnetic core in real time, and a corresponding permeability control model was built.

[0060] Methods for fabricating quantum-confined magnetic permeability units containing embedded sensor arrays include:

[0061] Choose a superlattice composed of a preset material as the core body, such as a superlattice composed of graphene / hexagonal boron nitride;

[0062] Pretreated SiO2 is used as the substrate for the superlattice. Through a predetermined growth process, such as low-pressure chemical vapor deposition, graphene precursors, such as CH4, and hexagonal boron nitride precursors, such as NH2, are used as gas sources. At 800℃ and 50Pa pressure, graphene layers and hexagonal boron nitride layers are grown alternately. The superlattice is controlled to grow alternately on the SiO2 substrate to form a predetermined period, such as a 5-period superlattice. During the superlattice growth process, a mask is used to block the growth. A predetermined interlayer gap of a predetermined size, such as a 1μm×1μm interlayer gap, is reserved at the superlattice interface of a predetermined reserved period, such as the 3rd period. Focused ion beam deposition is used to embed a predetermined chip within the interlayer gap. Specifically, a miniaturized magnetic flux sensor chip and a temperature sensor chip are embedded within the interlayer gap. The chip is only for monitoring the magnetic flux density and temperature inside the magnetic core and does not need to integrate other redundant functions to avoid occupying the interlayer gap of the superlattice. Based on the computational requirements of the permeability control model, the chip's measurement accuracy is pre-set, such as magnetic flux density accuracy at the μT level and temperature accuracy at ±0.5℃, to ensure that the acquired data can reflect subtle changes in the microscopic state of the magnetic core. Conductive polymer connections are used to form an embedded sensor array.

[0063] A 50nm thick Au layer is deposited by evaporation on the superlattice end face using electron beam lithography, and then gold wire gates with a preset linewidth, such as 200nm, are obtained by etching using electron beam lithography, forming a substrate-superlattice-gold electrode structure to obtain a quantum confined magnetic permeability unit.

[0064] The above method uses a graphene / hexagonal boron nitride superlattice as the main body of the magnetic core, which is alternately grown into a multi-period layered structure. This quantum confinement property enhances the magnetic core's resistance to interference from strong magnetic fields and high temperatures, and suppresses the nonlinear decreasing trend of permeability μ at the material level. At the same time, interlayer gaps are reserved during the superlattice growth process, and a pre-set chip is embedded by focused ion beam deposition and connected with conductive polymer to form an embedded sensor array. This can accurately collect magnetic flux and temperature data inside the magnetic core in real time, providing reliable input for the subsequent permeability control model and realizing real-time monitoring of inductance value L drift. In addition, gold wire gates are prepared on the end face of the superlattice by electron beam lithography. The constructed substrate-superlattice-gold electrode structure can efficiently respond to gate voltage control and dynamically adjust the permeability in combination with real-time monitoring data, thereby stabilizing the inductance value, avoiding filter cutoff frequency and resonant point shift, blocking interference signals from entering the magnetic bearing controller through the control loop, and ultimately reducing current ripple and improving rotor levitation accuracy. This effectively solves the filtering and control stability problems of magnetic levitation systems under strong magnetic field and high temperature coupling environments.

[0065] Reference Figure 3 Methods for constructing permeability modulation models include:

[0066] Calculate the fundamental quantum tunneling probability;

[0067] The vertical electric field is calculated based on the gate voltage and the monolayer barrier thickness. The monolayer barrier thickness is then corrected based on the vertical electric field, the monolayer barrier thickness, and the electron charge to obtain the corrected barrier thickness. Here, the monolayer barrier thickness specifically refers to the thickness of the single material layer constituting the barrier. If a graphene / hexagonal boron nitride superlattice is used, it corresponds to the thickness of a single layer of hexagonal boron nitride. For example, the vertical electric field... Correcting the barrier thickness ,in, Gate voltage; The thickness of a single-layer barrier; For electron charge; The dielectric constant of the barrier material;

[0068] The quantum tunneling probability is calculated based on the fundamental quantum tunneling probability, the single-layer barrier thickness, the modified barrier thickness, the tunneling probability decay factor correction value, and the temperature correction term. The tunneling probability decay factor correction value is obtained by correcting the tunneling probability decay factor based on the magnetic flux density inside the magnetic core, and the tunneling probability decay factor is obtained through experimental fitting. (Temperature correction term is also mentioned.) Quantum tunneling probability ,in, The height of the barrier; Boltzmann's constant; Absolute temperature; For reference temperature; The basic quantum tunneling probability; This is the correction value for the tunneling probability attenuation factor. , The flux correction factor can be obtained through experimental fitting. The magnetic flux density inside the magnetic core; This is the tunneling probability attenuation factor;

[0069] The permeability control model is constructed by obtaining the zero-magnetic-field reference permeability, quantum tunneling probability, coupling coefficient, modified barrier thickness, gate voltage sensitivity coefficient, gate voltage, magnetic field attenuation coefficient, and magnetic flux density inside the magnetic core. The zero-magnetic-field reference permeability is the initial permeability of the quantum confined magnetic permeability unit at zero external magnetic field and preset reference temperature, such as 25℃, which can be obtained by averaging multiple tests.

[0070] Methods for obtaining the fundamental quantum tunneling probability include:

[0071] The electron wave vector within the potential barrier is calculated based on electron energy, barrier thickness, Dirac constant, and effective electron mass; the electron wave vector within the graphene layer is also calculated based on electron energy, Dirac constant, and effective electron mass; the fundamental quantum tunneling probability is calculated based on the electron wave vector within the barrier, the electron wave vector within the graphene layer, and the monolayer barrier thickness. (Example: Electron wave vector within the potential barrier) Electron wave vector within graphene layers Fundamental quantum tunneling probability ,in, The effective mass of electrons in the barrier region; For electron energy; It is the Dirac constant; The characteristic velocity of graphene electrons;

[0072] Methods for obtaining the gate voltage sensitivity coefficient include:

[0073] With a fixed magnetic field and temperature, the gate voltage is changed according to a preset step size by a gallium nitride high electron mobility transistor, and the corresponding permeability is measured. The difference between the corresponding permeability and the zero magnetic field reference permeability is calculated. Combined with the modified barrier thickness and gate voltage, the coupling relationship between the difference and the gate voltage sensitivity coefficient is fitted. The gate voltage sensitivity coefficient is obtained by fitting using the least squares method.

[0074] The method for obtaining the magnetic field attenuation coefficient is similar to that for obtaining the grid voltage sensitivity coefficient. The magnetic permeability is collected by changing the magnetic field while keeping the grid voltage and temperature fixed, and then the magnetic field attenuation coefficient is obtained by fitting the zero magnetic field reference magnetic permeability.

[0075] The method for constructing a permeability control model integrates the quantitative relationship between quantum tunneling effect and multi-physics coupling to build a dynamic model that accurately reflects the nonlinear changes in permeability under strong magnetic fields and high temperatures, fundamentally solving a series of problems caused by inductor drift. This method overcomes the limitations of traditional models that ignore quantum effects and multi-field coupling, accurately capturing the nonlinear decrease in permeability μ under strong magnetic fields and high temperatures. It quantifies the influence of microscopic quantum states on permeability through quantum tunneling probability and modified barrier thickness, and reflects the effects of macroscopic control and environmental factors using gate voltage sensitivity coefficient and magnetic field attenuation coefficient, thus providing a reliable permeability input for accurate inductance calculation. Based on the permeability control model, the system can predict permeability change trends in advance, providing a precise basis for subsequent compensation strategies, thereby stabilizing the inductance value, avoiding filter cutoff frequency and resonant point shifts, blocking interference signals from entering the magnetic bearing controller, ultimately reducing current ripple, improving rotor levitation accuracy, and ensuring stable operation of the magnetic levitation system in extreme environments.

[0076] The output of the permeability control model is obtained, and the real-time inductance is analyzed.

[0077] Methods for obtaining inductor drift quantization models include:

[0078] The initial inductance, real-time permeability, real-time magnetic flux density, real-time temperature, corrected barrier thickness, total core length, and quantum tunneling probability are obtained from the output of the permeability control model. An inductance drift quantization model is then established to calculate the real-time inductance. (Example: Real-time inductance...) , The initial inductance; The real-time permeability is the real-time permeability output by the permeability control model. The initial permeability; Real-time magnetic flux density; The reference magnetic flux density; This is an inductor temperature correction term. , The temperature coefficient can be obtained through experimental fitting. For real-time temperature, For reference temperature; The flux sensitivity coefficient can be obtained through experimental fitting. This is the barrier thickness correction value; The reference barrier thickness.

[0079] The method for obtaining the inductance drift quantization model integrates key parameters such as the initial inductance, real-time permeability, real-time magnetic flux density, real-time temperature, corrected barrier thickness, total core length, and quantum tunneling probability output by the permeability modulation model. This constructs a quantization model that accurately reflects the coupling of multiple physical fields and quantum effects, thereby enabling dynamic calculation of real-time inductance. It overcomes the limitations of traditional inductance calculations that ignore quantum characteristics and nonlinear coupling of multiple physical fields. It can accurately capture the inductance drift law caused by the nonlinear decrease in permeability under the influence of strong magnetic fields and high temperatures. It reflects the influence of the environment on the core through macroscopic parameters such as real-time permeability, magnetic flux density, and temperature, and quantifies the modulation effect of microscopic quantum state changes on the inductance by quantum parameters such as corrected barrier thickness and quantum tunneling probability. This achieves full-dimensional source tracing of inductance drift from macroscopic to microscopic. The real-time inductance calculated based on the inductance drift quantization model can provide accurate input for subsequent compensation strategies, enabling the system to predict inductance drift trends in advance and make targeted adjustments. This suppresses filter cutoff frequency shift and resonant point drift, reduces the risk of interference signals intruding into the magnetic bearing controller through the control loop, and ultimately reduces current ripple, improves rotor levitation accuracy, and effectively solves the stability problem of magnetic levitation systems caused by inductance drift under strong magnetic field and high temperature coupling environments.

[0080] Based on the real-time inductance and combined with the preset target inductance, the real-time inductance is compensated through a coordinated strategy of gate voltage control and inductance compensation.

[0081] Methods for compensating for real-time inductance include:

[0082] Set up K groups of hollow coils, which are switched using GaN switches;

[0083] When the difference between the real-time inductance and the reference value is not greater than the preset drift threshold, the preset drift threshold is set based on the inductance fluctuation range allowed for normal system operation. The flux error change rate within a preset time period is calculated, such as the flux error change rate... , This represents the change in magnetic flux error. For preset time periods; Let be the real-time magnetic flux density at time t; define a state vector based on the real-time magnetic flux density, real-time temperature, magnetic flux error, and error rate of change; use the PID parameters for adjusting the gate voltage as the action vector; construct a reward function consisting of a quantum tunneling probability term, a multi-physics coupling penalty term, and a gate voltage energy consumption penalty; obtain the optimal action through iterative training; and adjust the gate voltage based on the PID parameters in the optimal action.

[0084] When the difference between the real-time inductance and the reference value is greater than the preset drift threshold, the corresponding coil is activated according to the difference, and the real-time inductance is adjusted until the difference is lower than the preset safety threshold. The preset safety threshold is set based on the minimum requirements for safe operation of the system.

[0085] The method for compensating for real-time inductance involves setting up K groups of air-core coils that can be rapidly switched by GaN switches, and combining two targeted compensation strategies to form a full-range inductance stabilization mechanism covering different degrees of drift, effectively solving the inductance drift problem under strong magnetic field and high temperature coupling. When the difference between the real-time inductance and the reference value is not greater than the preset drift threshold, a state vector containing real-time magnetic flux density, temperature, magnetic flux error, and error change rate is defined. The PID parameters for adjusting the gate voltage are used as the action vector to construct a reward function that integrates quantum tunneling probability, multi-physics field coupling penalty, and gate voltage energy consumption penalty. Through iterative training, the agent learns the optimal PID parameters autonomously, achieving precise control of the gate voltage. This fine adjustment based on quantum characteristics and multi-field sensing can adapt to small nonlinear fluctuations in permeability, suppressing small-amplitude inductance drift at the quantum level. When the difference exceeds the preset threshold, the corresponding air-core coil is quickly activated by the GaN switch. The fixed inductance of the coil is used to immediately correct large-amplitude inductance drift, ensuring that the inductance quickly returns to the safe range. The combined strategy of grid voltage control and inductance compensation can address the subtle dynamic drift of the quantum-confined magnetic core through grid voltage control, and compensate for large fluctuations under extreme conditions with the help of hollow coils. This stabilizes the inductance value, prevents filter cutoff frequency shift and resonant point drift, blocks the path of interference signals into the magnetic bearing controller, and ultimately reduces current ripple, improves rotor levitation accuracy, and ensures stable operation of the magnetic levitation system in the environment of strong magnetic field and high temperature coupling.

[0086] Methods for obtaining the reward function include:

[0087] The quantum tunneling probability term is calculated based on the quantum tunneling probability, target magnetic flux density, and magnetic flux error; such as the quantum tunneling probability term. , The flux error weighting coefficient can be obtained through experimental fitting. The target magnetic flux density;

[0088] The multiphysics coupling penalty term is calculated based on the current temperature, current magnetic flux density, temperature reference value, and magnetic flux density reference value; such as the multiphysics coupling penalty term. , and The coupling penalty weights can be obtained through experimental fitting;

[0089] The gate voltage energy consumption penalty is calculated based on the gate voltage and its maximum value. (Example: Gate voltage energy consumption penalty) , For real-time gate voltage; This represents the maximum gate voltage. This is the energy consumption penalty coefficient;

[0090] The method for obtaining the reward function constructs a comprehensive evaluation mechanism that integrates quantum tunneling probability, multi-physics coupling penalty, and gate voltage energy consumption penalty. This provides a precise optimization target for reinforcement learning, thereby specifically addressing the inductor drift problem under the coupling of strong magnetic fields and high temperatures. Specifically, the quantum tunneling probability is directly related to the microscopic quantum state stability of the quantum-confined magnetic core; by quantifying the deviation between the tunneling probability and the magnetic flux error, it ensures that the control strategy can accurately respond to permeability fluctuations at the quantum level, fundamentally suppressing the minute inductor drift caused by quantum effects. The multi-physics coupling penalty combines the deviation of the current temperature, magnetic flux density, and the baseline value, specifically targeting the cross-coupling effect of strong magnetic fields and high temperatures, avoiding adjustment misalignment caused by ignoring this coupling, and ensuring that the control strategy adapts to the nonlinear characteristics under complex environments. The gate voltage energy consumption penalty is calculated based on the ratio of the gate voltage to the maximum value, limiting gate voltage energy consumption while ensuring the adjustment effect, preventing excessive control from causing additional electromagnetic interference or energy loss, thus balancing stability and economy. The reward function formed by the collaboration of these three factors can guide reinforcement learning to autonomously learn an optimal control strategy that is compatible with quantum state fluctuations, multi-field coupling, and energy consumption. This will drive the inductance value to stabilize within the target range, thereby avoiding filter cutoff frequency shift and resonant point drift, blocking the path of interference signals into the magnetic bearing controller, and ultimately reducing current ripple, improving rotor levitation accuracy, and achieving stable operation of the magnetic levitation system in extreme environments.

[0091] Methods for obtaining the optimal action include:

[0092] The agent randomly draws samples from the experience pool and outputs actions through an actor network built from W layers of fully connected neural networks; the experience replay pool consists of Y sets of sample data; each set of samples includes the current state vector, action, reward, and the state vector for the next time step;

[0093] Substitute the PID parameters corresponding to the action into the gate voltage control formula and apply it to the superlattice;

[0094] Collect the state and reward of the next time step, and update the Q value of the critic network. The Q value is calculated based on the reward function, the maximum Q value of the corresponding action selected in the next state, and a discount factor, which is obtained through experimental optimization.

[0095] The actor network parameters are optimized using gradient descent with the goal of maximizing the Q-value.

[0096] The target network is updated at preset step intervals until the reward converges to a preset reward interval. The action corresponding to the maximum Q value when the reward converges to the preset reward interval is taken as the optimal action.

[0097] The method for obtaining the optimal action utilizes the dynamic interaction between the agent and the quantum confined superlattice system within a reinforcement learning framework. This enables precise optimization of the PID parameters for gate voltage regulation, thereby specifically addressing the inductor drift problem under the coupling of strong magnetic fields and high temperatures. Specifically, the Y sets of samples stored in the experience replay pool cover the current and next time-to-time state vectors, actions, and rewards, ensuring the agent can learn the multi-condition characteristics under strong magnetic fields and high temperatures. The PID parameter actions output by the actor network (built with W layers of fully connected neural networks) are applied to the superlattice via the gate voltage control formula. Combined with the Q-value updated by the critic network based on the reward function, the network parameters are continuously optimized through gradient descent. The optimal action, obtained when the reward converges to a preset interval, accurately adapts to the nonlinear decreasing characteristics of permeability. By learning the effects of quantum tunneling and multi-field coupling, the PID parameters dynamically respond to changes in microscopic quantum states to stabilize permeability. Furthermore, a discount factor balances the current regulation effect with long-term inductor stability, ensuring that the action suppresses immediate inductor drift while avoiding over-regulation that could trigger new fluctuations. This self-learning optimal action, when applied to the grid voltage adjustment, can stabilize the inductance value, prevent the filter cutoff frequency from deviating from the resonant point, block the path of interference signals into the magnetic bearing controller, thereby reducing current ripple, improving rotor levitation accuracy, and ensuring the stable operation of the magnetic levitation system in extreme environments.

[0098] Based on the compensated real-time inductance, a PID controller is used for filtering;

[0099] Methods for filtering using a PID controller include:

[0100] By combining the fixed capacitance value and the compensated real-time inductance, the filter cutoff frequency is calculated; for example, the filter cutoff frequency... , The real-time inductance after compensation; For fixed capacitance values;

[0101] A PID controller is used to make the PWM carrier frequency track the filter cutoff frequency for filtering.

[0102] The method employs a PID controller for filtering based on the compensated real-time inductance. By combining a fixed capacitance value with the compensated real-time inductance, a precise filter cutoff frequency is calculated. The PID controller then drives the PWM carrier frequency to dynamically track this cutoff frequency, forming a closed-loop frequency stabilization mechanism. This method specifically addresses the filter failure problem under strong magnetic field and high-temperature coupling. The compensated real-time inductance corrects for inductance drift caused by the nonlinear decrease in permeability, ensuring the calculated cutoff frequency accurately reflects the filter's intended operating frequency band. The PID controller continuously adjusts the PWM carrier frequency to maintain consistency with this cutoff frequency, effectively offsetting the cutoff frequency shift caused by inductance drift and suppressing the resonant point drift due to inductance changes. This directly blocks interference signals from entering the magnetic bearing controller due to filter frequency band mismatch, avoiding the performance degradation caused by frequency shift in traditional filtering. Ultimately, by stabilizing the filter's cutoff frequency and resonant point, the current ripple in the control loop is effectively reduced, rotor levitation accuracy is improved, and the magnetic levitation system maintains stable control performance even under extreme environments of strong magnetic field and high-temperature coupling.

[0103] Example 2:

[0104] This embodiment provides a method for obtaining a reward function applied to Embodiment 1, including the following steps:

[0105] The standard deviation of the quantum tunneling probability within a preset time window and the corresponding mean of the quantum tunneling probability are statistically analyzed to calculate the relative volatility of the quantum tunneling probability. Combined with the preset target volatility, the quantum state stability term is calculated and replaced with the quantum tunneling probability term in the reward function of Example 1. That is, the reward function consists of the quantum state stability term, the multi-physics coupling penalty term, and the gate voltage energy consumption penalty term.

[0106] This embodiment obtains the reward function by calculating the relative volatility through the standard deviation and mean of the quantum tunneling probability within a preset time window. This relative volatility is then combined with a preset target volatility to obtain a quantum state stability term, which replaces the quantum tunneling probability term in the original reward function. This results in a reward function composed of a quantum state stability term, a multi-physics coupling penalty term, and a gate voltage energy consumption penalty. This allows the reward function to focus more directly on the dynamic stability of the quantum state. Quantum state stability, by quantifying the fluctuation of the quantum tunneling probability, accurately captures the microscopic quantum state disturbances in a quantum-confined magnetic core under strong magnetic field and high temperature coupling. These disturbances are the root cause of the nonlinear decrease in permeability. Combined with the multi-physics coupling penalty term's suppression of temperature and magnetic field cross-interference and the gate voltage energy consumption penalty's limitation of over-tuning, the new reward function guides reinforcement learning to autonomously learn control strategies that better suit the quantum state stability requirements. This suppresses permeability drift caused by quantum state fluctuations at the microscopic level, thereby reducing inductance fluctuations. Ultimately, by stabilizing the quantum state indirectly, the permeability and inductance are stabilized, thus avoiding filter cutoff frequency shift and resonant point drift. This blocks the path of interference signals into the magnetic levitation controller, effectively reduces current ripple and improves rotor levitation accuracy, and specifically solves a series of stability problems caused by quantum state instability in magnetic levitation systems under strong magnetic field and high temperature coupling environments.

[0107] Example 3:

[0108] Please see Figure 4 As shown, this embodiment provides a magnetic levitation controller device resistant to electromagnetic interference, including:

[0109] Acquisition and processing module: fabricates quantum-confined magnetic permeability units, collects magnetic flux and temperature data inside the magnetic core in real time through an embedded sensor array, and builds a corresponding permeability control model;

[0110] Inductance analysis module: acquires the output of the permeability control model and analyzes it to obtain the real-time inductance;

[0111] Inductance compensation module: Based on real-time inductance and combined with preset target inductance, it compensates for real-time inductance through a cooperative strategy;

[0112] Compensation and filtering module: Based on the real-time inductance after compensation, a PID controller is used for filtering.

[0113] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0114] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A filtering method for a magnetic levitation controller to resist electromagnetic interference, characterized in that, include: Quantum confined magnetic permeability units were fabricated, and magnetic flux and temperature data inside the magnetic core were collected in real time through an embedded sensor array to construct a corresponding permeability control model. Methods for preparing quantum-confined magnetic permeability units include: A superlattice composed of a preset material is selected as the main body of the magnetic core; the superlattice is composed of graphene / hexagonal boron nitride; Pretreated SiO2 is used as a substrate for a superlattice. The superlattice is alternately grown on the SiO2 substrate through a preset growth process to form a superlattice with a preset period. During the superlattice growth process, a mask is used to block the superlattice and a preset size of interlayer gap is reserved at the interface of the superlattice with a preset period. A preset chip is embedded in the interlayer gap and connected by a conductive polymer to form an embedded sensor array. An Au layer is deposited by evaporation on the superlattice end face through electron beam lithography, and then gold wire gates with a preset linewidth are obtained by etching to form a substrate-superlattice-gold electrode structure, thereby obtaining a quantum confined magnetic permeability unit. The output of the permeability control model is obtained, and the real-time inductance is analyzed. Methods for constructing permeability tuning models include: Calculate the fundamental quantum tunneling probability; The vertical electric field generated is calculated based on the gate voltage and the single-layer barrier thickness. The single-layer barrier thickness is then corrected based on the vertical electric field and the electron charge to obtain the corrected barrier thickness. The quantum tunneling probability is calculated based on the basic quantum tunneling probability, the single-layer barrier thickness, the modified barrier thickness, the tunneling probability attenuation factor correction value, and the temperature correction term; the tunneling probability attenuation factor correction value is obtained by correcting the tunneling probability attenuation factor based on the magnetic flux density inside the magnetic core. The zero-magnetic-field reference permeability, quantum tunneling probability, coupling coefficient, modified barrier thickness, gate voltage sensitivity coefficient, gate voltage and magnetic field attenuation coefficient, and magnetic flux density inside the magnetic core are obtained to construct a permeability control model. Methods for obtaining real-time inductance include: The real-time permeability, real-time magnetic flux density, real-time temperature, corrected barrier thickness, total core length, and quantum tunneling probability output by the initial inductance and permeability control model are obtained. An inductance drift quantization model is established to calculate the real-time inductance. Based on the real-time inductance and combined with the preset target inductance, the real-time inductance is compensated through a coordinated strategy of gate voltage control and inductance compensation. Based on the compensated real-time inductance, a PID controller is used for filtering.

2. The electromagnetic interference-resistant magnetic levitation controller filtering method according to claim 1, characterized in that, Methods for obtaining the fundamental quantum tunneling probability include: The electron wave vector within the potential barrier is calculated based on electron energy, barrier thickness, Dirac constant, and effective electron mass; the electron wave vector within the graphene layer is calculated based on electron energy, Dirac constant, and effective electron mass; and the fundamental quantum tunneling probability is calculated based on the electron wave vector within the potential barrier, the electron wave vector within the graphene layer, and the single-layer barrier thickness.

3. The electromagnetic interference-resistant magnetic levitation controller filtering method according to claim 1, characterized in that, Methods for obtaining the gate voltage sensitivity coefficient include: With a fixed magnetic field and temperature, the gate voltage is changed according to a preset step size by a gallium nitride high electron mobility transistor, and the corresponding permeability is measured. The difference between the corresponding permeability and the zero magnetic field reference permeability is calculated. Combined with the modified barrier thickness and gate voltage, the coupling relationship between the difference and the gate voltage sensitivity coefficient is fitted. The gate voltage sensitivity coefficient is obtained by fitting using the least squares method.

4. The electromagnetic interference-resistant magnetic levitation controller filtering method according to claim 1, characterized in that, Methods for compensating for real-time inductance include: Set up K groups of hollow coils, which are switched using GaN switches; When the difference between the real-time inductance and the reference value is not greater than the preset drift threshold, the flux error change rate within the preset time period is calculated, and a state vector is defined based on the real-time flux density, real-time temperature, flux error, and error change rate. The PID parameters used to adjust the gate voltage are used as the action vector. A reward function consisting of a quantum tunneling probability term, a multi-physics coupling penalty term, and a gate voltage energy consumption penalty is constructed. The optimal action is obtained through iterative training. The gate voltage is adjusted based on the PID parameters in the optimal action. When the difference between the real-time inductance and the reference value is greater than the preset drift threshold, the corresponding coil is activated according to the difference, and the real-time inductance is adjusted until the difference is lower than the preset safety threshold.

5. The electromagnetic interference-resistant magnetic levitation controller filtering method according to claim 4, characterized in that, Methods for obtaining reward functions include: The quantum tunneling probability term is calculated based on the quantum tunneling probability, the target magnetic flux density, and the magnetic flux error. The multiphysics coupling penalty term is calculated based on the current temperature, current magnetic flux density, temperature reference value, and magnetic flux density reference value. The gate voltage energy consumption penalty is calculated based on the gate voltage and the maximum gate voltage.

6. The electromagnetic interference-resistant magnetic levitation controller filtering method according to claim 4, characterized in that, Methods for obtaining the optimal action include: The agent randomly draws samples from the experience pool and outputs actions through an actor network built from W layers of fully connected neural networks; the experience replay pool consists of Y sets of sample data; each set of samples includes the current state vector, action, reward, and the state vector for the next time step; Substitute the PID parameters corresponding to the action into the gate voltage control formula and apply it to the superlattice; Collect the state and reward of the next time step, and update the Q value of the critic network. The Q value is calculated based on the reward function, the maximum Q value of the corresponding action selected in the next state, and a discount factor, which is obtained through experimental optimization. The actor network parameters are optimized using gradient descent with the goal of maximizing the Q-value. The target network is updated at preset step intervals until the reward converges to a preset reward interval. The action corresponding to the maximum Q value when the reward converges to the preset reward interval is taken as the optimal action.

7. The electromagnetic interference-resistant magnetic levitation controller filtering method according to claim 1, characterized in that, Methods for filtering using a PID controller include: By combining the fixed capacitance value with the real-time inductance after compensation, the filter cutoff frequency is calculated. A PID controller is used to make the PWM carrier frequency track the filter cutoff frequency for filtering.

8. An electromagnetic interference-resistant magnetic levitation controller device, used to implement the electromagnetic interference-resistant magnetic levitation controller filtering method according to any one of claims 1-7, characterized in that, include: Acquisition and processing module: fabricates quantum-confined magnetic permeability units, and collects magnetic flux and temperature data inside the magnetic core in real time through an embedded sensor array to construct a corresponding permeability control model; Inductance analysis module: acquires the output of the permeability control model and analyzes it to obtain the real-time inductance; Inductance compensation module: Based on real-time inductance and combined with preset target inductance, it compensates for real-time inductance through a cooperative strategy; Compensation and filtering module: Based on the real-time inductance after compensation, a PID controller is used for filtering.

Images