A Smart Tuning Method and System for LC Filters

Abstract

This invention discloses an intelligent debugging method and system for LC filters, relating to the field of automation control technology. It generates an initialization parameter set suitable for the LC filter debugging task based on a meta-learning framework; constructs a quantum particle swarm optimization debugging model; and builds a parameter change trajectory database by combining the concept of psychological fields and knowledge distillation techniques. The database is then used to perform time-series modeling of the parameter change trajectory database through an LSTM network to predict the optimal initial guess. The initial debugging of the LC filter does not require extensive reliance on historical initial data, and the complex procedures of manual debugging are eliminated, significantly reducing the time for initializing the LC filter. It solves the dilemma of local solutions during intelligent debugging, with fast convergence speed and shorter debugging time. During debugging, it quickly and efficiently obtains result data based on a small amount of historical data, and can continuously optimize the automated intelligent debugging process of the LC filter.

Description

Technical Field

[0001] This invention relates to the field of LC filter tuning technology, and in particular to an intelligent tuning method and system for LC filters. Background Technology

[0002] With the development of artificial intelligence and automation control technologies, intelligent tuning methods will be more widely used in the field of LC filter tuning. For example, by combining more advanced technologies such as deep learning and reinforcement learning, intelligent tuning systems will be able to adaptively adjust in more complex environments, further improving filter performance and application range.

[0003] Currently, Chinese invention patent application number CN201810086329.2 discloses a precise tuning method and system for notch filters. By performing mathematical simulation or frequency sweep experiments on the target filter, a precise mathematical model of the second-order notch filter is established, allowing for tuning. During the tuning process, if the frequency characteristics of the target filter still deviate from the required specifications, the notch depth and bandwidth can be adjusted appropriately to gradually meet the specifications. The tuning process can also guide subsequent tuning based on previous tuning results, significantly improving tuning efficiency while achieving precise tuning. Existing technologies still suffer from over-reliance on historical initial data in the initial tuning of LC filters, and the complex procedures of manual tuning require significant time for initialization. They also face the challenge of finding local solutions during intelligent tuning, exhibiting slow convergence speeds and excessively long tuning times. Furthermore, they cannot quickly and efficiently obtain result data based on limited historical data, and they cannot continuously optimize the automated intelligent tuning process of LC filters through continuous learning and accumulation. Summary of the Invention

[0004] The technical problem solved by this invention is that the existing technology still relies too much on historical initial data in the initial adjustment of LC filter, and there is still a complicated manual debugging procedure. It requires a lot of time to initialize the LC filter. In the intelligent debugging process, there is a dilemma of local solution, and the convergence speed is slow and the debugging time is too long. In the debugging process, it is impossible to obtain the result data quickly and efficiently based on a small amount of historical data. Furthermore, it is impossible to continuously optimize the automated intelligent debugging process of LC through continuous learning and accumulation.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an intelligent tuning method for an LC filter, comprising the following steps:

[0006] Step S1: Based on the meta-learning framework, sample historical LC filter debugging task data from the multi-task distribution, learn the distribution characteristics of parameters through a pre-trained machine learning model, and generate an initialization parameter set suitable for the LC filter debugging task;

[0007] Step S2: Encode the inductance, capacitance, and quality factor of the LC filter into quantum states to construct a quantum particle swarm optimization and tuning model;

[0008] Step S3: Combining the concept of psychological field and knowledge distillation technique, the high-dimensional parameter space is mapped to the low-dimensional latent space to construct a parameter change trajectory database. The parameter change trajectory database is then used to perform time series modeling on the database to predict the optimal initial guess.

[0009] Preferably, the meta-learning framework adopts a two-layer optimization structure. The outer layer optimization updates the initial parameter distribution through gradient descent, and the inner layer optimization simulates the gradient update process, including minimizing the cutoff frequency error and total harmonic distortion index of the current task, and outputting an initial parameter distribution adapted to the condition of few samples.

[0010] Preferably, constructing a quantum particle swarm optimization and debugging model includes:

[0011] The quantum particle swarm optimization tuning model is used to describe the relationship between the performance of the LC filter and the tuning parameters.

[0012] Generating a quantum state based on the initial parameter distribution includes: mapping the inductance value, capacitance value, and quality factor to the amplitude and phase of the quantum bit, respectively. The inductance value adopts a normalized logarithmic encoding form, the capacitance value adopts a piecewise linear encoding, and the quality factor is compressed to the [0,1] interval by the Sigmoid function.

[0013] The mathematical expression for quantum state encoding of the inductance, capacitance, and quality factor is as follows:

[0014] ;

[0015] in, Represents a superposition state of quantum states. The qubit amplitude corresponds to the inductance value, representing the relative contribution of the inductance to the quantum state. The qubit amplitude corresponds to the capacitance value, representing the relative contribution of capacitance to the quantum state. The amplitude of the qubit corresponding to the quality factor represents the relative contribution of the quality factor to the quantum state. The quantum state of inductance value, The quantum state of capacitance. The quantum state representing the quality factor;

[0016] The quantized parameter space is obtained, and the particle positions in the quantized parameter space are initialized by the particle swarm optimization algorithm. The initialization includes: adjusting the search direction of the particle positions through a quantum rotation gate based on a preset number of iterations and fitness change rate, and dynamically adjusting the inertial weight of the particle swarm. The inertial weight is used to control the speed and direction of the particle search.

[0017] The quantum rotation angle of the quantum rotating gate is dynamically calculated by the gradient difference between the current parameters and the globally optimal parameters. The mathematical expression for the rotation angle is:

[0018] ;

[0019] in, For rotation angle, Let be a function that maps the two input gradients to rotation angles. The gradient of the current parameters. The gradient of the global parameters.

[0020] Preferably, using a chaotic perturbation strategy to break local optima includes:

[0021] During the convergence stagnation phase of the particle swarm optimization algorithm, chaotic perturbation is injected to break the local optimum. After breaking the local optimum, a chaotic sequence is generated using Logistic mapping. The perturbation amplitude is inversely proportional to the fitness value of the current optimal solution.

[0022] When in an iterative state, the large parameter space is optimized through quantum variational optimization or quantum simulated annealing. When the preset number of iterations is reached, the final optimization result of the quantized particle swarm hybrid optimization and debugging model is output.

[0023] Preferably, constructing a parameter change trajectory database includes:

[0024] Using the total psychological needs of a person as the output prediction parameter of the LC filter, and considering certain physiological, environmental, and social needs as quantum states of inductance, capacitance, and quality factor, respectively, the LC filter identifies the missing conditions, insufficient data, desired conditions, and needs of the subject. Iterative transfer calculations are performed based on knowledge distillation to establish the objective function of the LC filter. The mathematical expression of the objective function is as follows:

[0025] ;

[0026]

[0027] Where B is the output prediction parameter of the LC filter. Let be the representative function, representing the relationship between the performance of the LC filter output by the quantized particle swarm optimization tuning model and the tuning parameters. For the total loss function, This indicates the learning of debugging task data. This represents the difference between two iterations of calculation. This represents the output prediction parameters of the nth iteration. This represents the output prediction parameters for the (n+1)th iteration. This represents the Softmax function, used to convert the model's output into a probability distribution. The weight hyperparameter is determined by the magnitude of the backpropagation gradient of the LC filter;

[0028] All output prediction parameters from the iterative calculations are stored in a parameter change trajectory database.

[0029] Preferably, the LSTM network uses a gated recurrent unit variant, with the input being the initialization parameter set of the historical debugging task, and the output layer predicting the optimal parameter direction for the next iteration stage through the Softmax function;

[0030] Based on the learning results of historical data from the LSTM network, a similarity matching algorithm is used to quickly match similar designs and parameter changes, providing initial guesses for new debugging tasks in iterative calculations.

[0031] Preferably, a dynamic constraint processing mechanism is adopted in the process of performing time-series modeling of the parameter change trajectory database using the LSTM network:

[0032] Define the elastic boundary of the feasible solution. The feasible interval for the inductance value is [L_min, L_max], and the feasible interval for the capacitance value is [C_min, C_max]. When the parameters exceed the boundary, the out-of-bounds parameters are projected to the neighboring feasible points of the feasible interval using the mirror mapping method.

[0033] An intelligent tuning system for an LC filter, comprising an initialization parameter module, a quantum particle swarm optimization tuning module, an optimal prediction module, and an embedded hardware platform, for executing an intelligent tuning method for an LC filter.

[0034] The initialization parameter module is used to sample historical LC filter debugging task data from a multi-task distribution based on a meta-learning framework, learn the distribution characteristics of parameters through a pre-trained machine learning model, and generate an initialization parameter set suitable for the LC filter debugging task.

[0035] The quantum particle swarm hybrid optimization and debugging module is used to encode the inductance, capacitance and quality factor of the LC filter into quantum states and construct a quantum particle swarm hybrid optimization and debugging model.

[0036] The optimal prediction module is used to combine the concept of psychological field and knowledge distillation technology to map the high-dimensional parameter space to the low-dimensional latent space, construct a parameter change trajectory database, and perform time series modeling on the parameter change trajectory database through LSTM network to predict the optimal initial guess.

[0037] The embedded hardware platform is used to run the meta-learning initialization module and the quantum heuristic optimization core algorithm, the dynamic parameter update interface, and temperature compensation.

[0038] Preferably, the embedded hardware platform includes:

[0039] It supports FPGA parallel computing when running the meta-learning initialization module and the quantum-heuristic optimization core algorithm;

[0040] The dynamic parameter update interface includes the ability to adjust debugging parameters online through the LabVIEW graphical programming environment, and to visualize the frequency domain response curve and optimization process.

[0041] Temperature compensation includes an integrated NTC thermistor network that monitors the temperature drift of inductors and capacitors in real time and corrects parameter values ​​using a linear regression model.

[0042] Preferably, the debugged LC filter undergoes performance verification and evaluation:

[0043] The metrics include frequency domain performance, time domain performance, energy efficiency, and stability. The frequency domain performance includes cutoff frequency error, stopband attenuation, and passband ripple. The time domain performance includes step response rise time and overshoot rate. The energy efficiency metrics include insertion loss and power loss. The stability metrics include temperature drift coefficient and component parameter tolerance sensitivity.

[0044] The beneficial effects of this invention are: the LC filter does not need to rely heavily on historical initial data at the beginning of the adjustment, and the complex procedures of manual debugging are eliminated, which greatly reduces the time for initializing the LC filter. It solves the dilemma of local solutions in the intelligent debugging process, and has a fast convergence speed and shorter debugging time. During the debugging process, it can quickly and efficiently obtain result data based on a small amount of historical data, and can continuously optimize the automated intelligent debugging process of LC through continuous learning and accumulation. Attached Figure Description

[0045] Figure 1 This is a basic flowchart illustrating an intelligent tuning method for an LC filter according to an embodiment of the present invention. Detailed Implementation

[0046] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0047] Reference Figure 1 As an embodiment of the present invention, an intelligent tuning method for an LC filter is provided, comprising the following steps:

[0048] Step S1: Based on the meta-learning framework, sample historical LC filter debugging task data from the multi-task distribution, learn the distribution characteristics of parameters through a pre-trained machine learning model, and generate an initialization parameter set suitable for LC filter debugging tasks;

[0049] Step S2: Encode the inductance, capacitance, and quality factor of the LC filter into quantum states to construct a quantum particle swarm optimization and tuning model;

[0050] Step S3: Combining the concept of psychological field and knowledge distillation technique, the high-dimensional parameter space is mapped to the low-dimensional latent space to construct a parameter change trajectory database. The parameter change trajectory database is then used to perform time series modeling on the database to predict the optimal initial guess.

[0051] The meta-learning framework adopts a two-layer optimization structure. The outer layer optimization updates the initial parameter distribution through gradient descent, while the inner layer optimization simulates the gradient update process, including minimizing the cutoff frequency error and total harmonic distortion index of the current task, and outputting an initial parameter distribution that adapts to the condition of few samples.

[0052] With a small amount of sample data, meta-learning can quickly learn the tuning rules of LC filters, avoiding long trial and error from scratch and reducing the time and complexity of manual tuning.

[0053] The construction of a quantum particle swarm optimization and debugging model includes:

[0054] A quantum particle swarm optimization tuning model is used to describe the relationship between the performance of LC filters and tuning parameters.

[0055] Generating quantum states based on the initial parameter distribution includes: mapping the inductance value, capacitance value, and quality factor to the amplitude and phase of the quantum bit, respectively. The inductance value adopts a normalized logarithmic encoding form, the capacitance value adopts a piecewise linear encoding, and the quality factor is compressed to the [0,1] interval through the Sigmoid function.

[0056] The mathematical expression for quantum state encoding of inductance, capacitance, and quality factor is:

[0057] ;

[0058] in, Represents a superposition state of quantum states. The qubit amplitude corresponds to the inductance value, representing the relative contribution of the inductance to the quantum state. The qubit amplitude corresponds to the capacitance value, representing the relative contribution of capacitance to the quantum state. The amplitude of the qubit corresponding to the quality factor represents the relative contribution of the quality factor to the quantum state. The quantum state of inductance value, The quantum state of capacitance. The quantum state representing the quality factor;

[0059] The quantized parameter space is obtained, and the particle positions in the quantized parameter space are initialized by the particle swarm optimization algorithm. The initialization includes: adjusting the search direction of the particle positions through the quantum rotation gate based on the preset number of iterations and fitness change rate, and dynamically adjusting the inertia weight of the particle swarm. The inertia weight is used to control the speed and direction of the particle search.

[0060] The quantum rotation angle of the quantum rotating gate is dynamically calculated based on the gradient difference between the current parameters and the global optimal parameters, ensuring that the search direction adjusts as the optimal solution changes. The mathematical expression for the rotation angle is:

[0061] ;

[0062] in, For rotation angle, Let be a function that maps the two input gradients to rotation angles. The gradient of the current parameters. The gradient of the global parameters.

[0063] The gradient difference reflects the gap between the current parameters and the global optimum, while the change in the rotation angle helps the particle to perform a more accurate search in the neighborhood of the optimum.

[0064] Breaking local optima using chaotic perturbation strategies includes:

[0065] During the convergence stagnation phase of the particle swarm optimization algorithm, chaotic perturbation is injected to break the local optimum. After breaking the local optimum, a chaotic sequence is generated using Logistic mapping. The perturbation amplitude is inversely proportional to the fitness value of the current optimal solution, which helps to break the local optimum and explore new search regions.

[0066] When in the iterative state, the search for the global optimal solution is accelerated by optimizing the large parameter space through quantum variational optimization or quantum simulated annealing. During multiple iterations, with the dynamic adjustment of inertial weights, the introduction of chaotic perturbations and the application of quantum rotation gates, the algorithm gradually approaches the optimal solution. When the preset number of iterations is reached, the final optimization result of the quantum particle swarm hybrid optimization debugging model is output.

[0067] Quantum computing heuristic optimization helps find the global optimum or near-optimal solution in a large parameter space. Simulated annealing algorithm avoids local optima by introducing random perturbations and eventually converges to the global optimum. Quantum simulated annealing algorithm utilizes quantum superposition and interference effects in quantum computing, enabling the system to explore multiple solution spaces in parallel, thereby greatly improving the convergence speed.

[0068] This method not only utilizes the efficient optimization capabilities of quantum computing in high-dimensional parameter spaces, but also combines the collective intelligence of particle swarms and the global search capabilities of chaotic perturbation mechanisms, effectively avoiding local optima and accelerating the search for global optima.

[0069] Constructing a database of parameter change trajectories includes:

[0070] Using the total psychological needs of individuals as the output prediction parameters of the LC filter, and individual needs related to physiology, environment, and society as quantum states of inductance, capacitance, and quality factor, respectively, the LC filter identifies the missing conditions, insufficient data, desired conditions, and needs of the subject. Iterative transfer calculations are performed based on knowledge distillation to establish the objective function of the LC filter. The mathematical expression of the objective function is as follows:

[0071] ;

[0072]

[0073] Where B is the output prediction parameter of the LC filter. Let be the representative function, representing the relationship between the performance of the LC filter output by the quantized particle swarm optimization tuning model and the tuning parameters. For the total loss function, This indicates the learning of debugging task data. This represents the difference between two iterations of calculation. This represents the output prediction parameters of the nth iteration. This represents the output prediction parameters for the (n+1)th iteration. This represents the Softmax function, used to convert the model's output into a probability distribution. The weight hyperparameter is determined by the magnitude of the backpropagation gradient of the LC filter;

[0074] The goal of combining the concept of psychological field with knowledge distillation-assisted optimization is to use the concept of psychological field to refer to a space of certain correlations or interactions between tasks and parameters. Through this mapping, we can understand and analyze the similarities and changing trends between different tasks in a high-dimensional parameter space, thereby guiding parameter adjustments. By transmitting the experience and rules for adjusting parameters through an efficient teacher model, a simplified model can be adjusted more quickly to debug LC filters.

[0075] All output prediction parameters from the iterative calculations are stored in a parameter change trajectory database.

[0076] By establishing a parameter change trajectory database, intelligent decision-making can be made based on past debugging experience, further accelerating the debugging process. Reasonable initialization parameters can be generated through a small number of samples, global optimization can be performed, and the debugging process can be optimized by combining historical data and intelligent prediction, significantly improving debugging efficiency and performance. Furthermore, it can help to quickly find similar debugging cases and parameter trajectories in the past in future debugging tasks, providing a reference for new debugging tasks. The parameter change trajectory database can continuously learn and accumulate to optimize the debugging process.

[0077] The LSTM network uses a variant of the gated recurrent unit, with the input being the initialization parameter set of the historical debugging task, and the output layer predicting the optimal parameter direction for the next iteration stage through the Softmax function;

[0078] Based on the learning results of LSTM networks on historical data, a similarity matching algorithm is used to quickly match similar designs and parameter changes, providing initial guesses for new debugging tasks in iterative calculations. The similarity matching algorithm can reduce debugging time from scratch by finding the closest historical cases.

[0079] By using multimodal mixing, multiple influencing factors can be considered simultaneously, resulting in more comprehensive and efficient debugging parameters.

[0080] A dynamic constraint processing mechanism is adopted in the process of time series modeling of parameter change trajectory database using LSTM network:

[0081] Define the elastic boundary of the feasible solution. The feasible interval for the inductance value is [L_min, L_max], and the feasible interval for the capacitance value is [C_min, C_max]. When the parameters exceed the boundary, the out-of-bounds parameters are projected to the neighboring feasible points of the feasible interval using the mirror mapping method.

[0082] An intelligent tuning system for an LC filter, comprising an initialization parameter module, a quantum particle swarm optimization tuning module, an optimal prediction module, and an embedded hardware platform, for executing an intelligent tuning method for an LC filter.

[0083] The initialization parameter module is used to sample historical LC filter debugging task data from a multi-task distribution based on a meta-learning framework, learn the distribution characteristics of parameters through a pre-trained machine learning model, and generate an initialization parameter set suitable for LC filter debugging tasks.

[0084] The quantized particle swarm optimization and debugging module is used to encode the inductance, capacitance and quality factor of the LC filter into quantum states and construct a quantized particle swarm optimization and debugging model.

[0085] The optimal prediction module combines the concept of psychological field and knowledge distillation techniques to map the high-dimensional parameter space to the low-dimensional latent space, construct a parameter change trajectory database, and perform time series modeling on the parameter change trajectory database through an LSTM network to predict the optimal initial guess.

[0086] The embedded hardware platform is used to run the meta-learning initialization module and the quantum heuristic optimization core algorithm, the dynamic parameter update interface, and temperature compensation.

[0087] Deployed on embedded hardware platforms, including:

[0088] It supports FPGA parallel computing when running the meta-learning initialization module and the quantum-heuristic optimization core algorithm;

[0089] The dynamic parameter update interface includes the ability to adjust debugging parameters online through the LabVIEW graphical programming environment, and to visualize the frequency domain response curve and optimization process.

[0090] Temperature compensation includes an integrated NTC thermistor network that monitors the temperature drift of inductors and capacitors in real time and corrects parameter values ​​using a linear regression model.

[0091] The LC filter after debugging was evaluated and verified for its performance.

[0092] The metrics include frequency domain performance, time domain performance, energy efficiency, and stability. Frequency domain performance includes cutoff frequency error, stopband attenuation, and passband ripple. Time domain performance includes step response rise time and overshoot rate. Energy efficiency metrics include insertion loss and power loss. Stability includes temperature drift coefficient and component parameter tolerance sensitivity.

[0093] The intelligent debugging method provided in this invention does not require extensive reliance on historical initial data at the beginning of LC filter tuning and reduces the complex procedures of manual debugging, greatly shortening the time for initializing the LC filter. It solves the dilemma of local solutions in the intelligent debugging process, and has a fast convergence speed and shorter debugging time. During the debugging process, it can quickly and efficiently obtain result data based on a small amount of historical data, and can continuously optimize the automated intelligent debugging process of LC through continuous learning and accumulation.

[0094] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0095] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method of intelligent commissioning of an LC filter, characterized in that, The method comprises the following steps: Step S1: sampling historical LC filter debugging task data from a multi-task distribution based on a meta-learning framework, learning a distribution feature of a parameter through a pre-trained machine learning model, and generating an initial parameter set suitable for the LC filter debugging task; the meta-learning framework adopts a double-layer optimization structure, outer-layer optimization updates an initial parameter distribution through gradient descent, and inner-layer optimization simulates a gradient update process, including: minimizing a cutoff frequency error and a total harmonic distortion index of a current task, and outputting an initial parameter distribution suitable for a small sample condition; Step S2: encoding inductance, capacitance and quality factor of the LC filter into a quantum state, and constructing a quantum particle swarm hybrid optimization debugging model, including: The quantum particle swarm hybrid optimization debugging model is used to describe the relationship between the performance of the LC filter and the debugging parameters; Generating a quantum state based on the initial parameter distribution includes: mapping the inductance, capacitance and quality factor to the amplitude and phase of a quantum bit, respectively, the inductance adopts a normalized logarithmic encoding form, the capacitance adopts a piecewise linear encoding, and the quality factor is compressed to the [0, 1] interval through a Sigmoid function; The mathematical expression for encoding the inductance, capacitance and quality factor into a quantum state is: ； wherein, represents a superposition state of quantum states, is a quantum bit amplitude corresponding to an inductance value, representing the relative contribution of the inductance in the quantum state, is a quantum bit amplitude corresponding to a capacitance value, representing the relative contribution of the capacitance in the quantum state, is a quantum bit amplitude corresponding to a quality factor, representing the relative contribution of the quality factor in the quantum state, is a quantum state of an inductance value, is a quantum state of a capacitance value, is a quantum state of a quality factor; A quantum parameter space is obtained, the particle position of the quantum parameter space is initialized through a particle swarm optimization algorithm, the initialization includes: based on a preset iteration number and an adaptability change rate, adjusting the search direction of the particle position through a quantum rotation gate, and dynamically adjusting the inertia weight of the particle swarm, the inertia weight is used to control the speed and direction of the particle search; The quantum rotation angle of the quantum rotation gate is dynamically calculated from the gradient difference between the current parameter and the global optimal parameter, and the mathematical expression of the rotation angle is: ； wherein, is a rotation angle, is a function that maps the two input gradients to a rotation angle, is the gradient of the current parameter, is the gradient of the global parameter; Step S3: combining the concept of psychological field and the knowledge distillation technology, mapping the high-dimensional parameter space to the low-dimensional latent space, constructing a parameter change trajectory database, and predicting the optimal initial guess through a LSTM network for time series modeling of the parameter change trajectory database; Wherein, constructing the parameter change trajectory database includes: based on the concept of psychological field, finding the data that the subject lacks, is insufficient, and needs of the LC filter, iteratively calculating based on knowledge distillation, and establishing a target function of the LC filter, the mathematical expression of the target function is: ； wherein B is an output prediction parameter of the LC filter, is a representative function, representing a relationship between the performance of the LC filter and the tuning parameters output by the quantized particle swarm hybrid optimization tuning model, is a total loss function, represents learning on the tuning task data, represents a difference between the two iterations, represents an output prediction parameter of the n-th iteration, represents an output prediction parameter of the n+1-th iteration, represents a Softmax function, used to convert the output of the model into a probability distribution, is a weight hyperparameter, determined by the magnitude of the backpropagation gradient of the LC filter; All output prediction parameters in the iterative calculation are stored in the parameter change trajectory database.

2. The method of smart commissioning of an LC filter of claim 1, wherein, Breaking the local optimum by using a chaotic disturbance strategy includes: In the convergence stagnation stage of the particle swarm algorithm, the local optimum is broken through chaotic disturbance injection, and a chaotic sequence is generated by using Logistic mapping after the local optimum is broken, and the disturbance amplitude is inversely proportional to the fitness value of the current optimal solution; When in the iteration state, the large parameter space is optimized through quantum variational optimization or quantum simulated annealing optimization, and when the preset iteration number is reached, the final optimization result of the quantum particle swarm hybrid optimization debugging model is output.

3. The intelligent debugging method of the LC filter of claim 1, characterized in that: The LSTM network adopts a gated recurrent unit variant, the input is the initial parameter set of the historical debugging task, and the output layer predicts the optimal parameter direction of the next iteration stage through a Softmax function; Based on the learning results of the LSTM network on historical data, similar design and parameter changes are quickly matched through a similarity matching algorithm to provide an initial guess for the new debugging task in the iteration calculation.

4. The method of intelligent commissioning of an LC filter of claim 3, wherein, A dynamic constraint processing mechanism is adopted in the time series modeling of the parameter change trajectory database by the LSTM network: Define the feasible solution elastic boundary, the feasible interval of the inductance value is [L_min, L_max], and the feasible interval of the capacitance value is [C_min, C_max]. When the parameter exceeds the boundary, the mirror mapping method is used to project the out-of-bound parameter to the neighborhood feasible point of the feasible interval.

5. An intelligent commissioning system of LC filter, the system is used for executing an intelligent commissioning method of LC filter, characterized in being, It includes an initialization parameter module, a quantum particle swarm hybrid optimization debugging module, an optimal prediction module, and an embedded hardware platform: The initialization parameter module is used to sample historical LC filter debugging task data from a multi-task distribution based on a meta-learning framework, learn the distribution characteristics of the parameters through a pre-trained machine learning model, and generate an initial parameter set suitable for the LC filter debugging task; The meta-learning framework adopts a double-layer optimization structure, the outer layer optimization updates the initial parameter distribution through gradient descent, and the inner layer optimization simulates the gradient update process, including: minimizing the cutoff frequency error and total harmonic distortion index of the current task, and outputting an initial parameter distribution suitable for the few-sample condition; The quantum particle swarm hybrid optimization debugging module is used to encode the inductance value, capacitance value and quality factor of the LC filter into a quantum state, and construct a quantum particle swarm hybrid optimization debugging model, including: The quantum particle swarm hybrid optimization debugging model is used to describe the relationship between the performance of the LC filter and the debugging parameters; Based on the initial parameter distribution, the quantum state is generated by mapping the inductance value, capacitance value and quality factor to the amplitude and phase of the quantum bit, respectively. The inductance value is encoded in a normalized logarithmic form, the capacitance value is encoded in a piecewise linear form, and the quality factor is compressed to the [0, 1] interval through a Sigmoid function; The mathematical expression for encoding the inductance value, capacitance value and quality factor into a quantum state is: ； wherein, represents a superposition state of quantum states, is a quantum bit amplitude corresponding to an inductance value, representing the relative contribution of the inductance in the quantum state, is a quantum bit amplitude corresponding to a capacitance value, representing the relative contribution of the capacitance in the quantum state, is a quantum bit amplitude corresponding to a quality factor, representing the relative contribution of the quality factor in the quantum state, is a quantum state of an inductance value, is a quantum state of a capacitance value, is a quantum state of a quality factor; The quantum parameter space is obtained, and the particle position of the quantum parameter space is initialized by the particle swarm optimization algorithm, including: based on the preset iteration number and fitness change rate, adjusting the search direction of the particle position through the quantum rotation gate, dynamically adjusting the inertia weight of the particle swarm, and the inertia weight is used to control the speed and direction of the particle search; The quantum rotation angle of the quantum rotation gate is dynamically calculated by the gradient difference between the current parameter and the global optimal parameter, and the mathematical expression of the rotation angle is: ； wherein, is a rotation angle, is a function that maps the two input gradients to a rotation angle, is the gradient of the current parameter, is the gradient of the global parameter; The optimal prediction module is used to combine the concept of psychological field and knowledge distillation technology to map the high-dimensional parameter space to the low-dimensional latent space, construct a parameter change trajectory database, and predict the optimal initial guess through time series modeling of the parameter change trajectory database by the LSTM network. Wherein, the construction parameter variation trajectory database comprises: finding the subject's missing conditions, insufficient data, subject's desired conditions and demand data corresponding to the LC filter based on the concept of mental field, iteratively transferring and calculating based on knowledge distillation, establishing the objective function of the LC filter, and the mathematical expression of the objective function is: ； wherein B is an output prediction parameter of the LC filter, is a representative function, representing a relationship between the performance of the LC filter and the tuning parameters output by the quantized particle swarm hybrid optimization tuning model, is a total loss function, represents learning on the tuning task data, represents a difference between the two iterations, represents an output prediction parameter of the n-th iteration, represents an output prediction parameter of the n+1-th iteration, represents a Softmax function, used to convert the output of the model into a probability distribution, is a weight hyperparameter, determined by the magnitude of the backpropagation gradient of the LC filter; All output prediction parameters in the iterative calculation are stored in the parameter variation trajectory database; The embedded hardware platform is used for running the meta-learning initialization module and the quantum heuristic optimization core algorithm, the dynamic parameter updating interface and the temperature compensation.

6. The intelligent commissioning system of an LC filter of claim 5, wherein, The embedded hardware platform comprises: The meta-learning initialization module and the quantum heuristic optimization core algorithm support FPGA parallel calculation when running; The dynamic parameter updating interface comprises realizing online adjustment of debugging parameters through LabVIEW graphical programming environment and visually displaying frequency domain response curve and optimization process; The temperature compensation comprises integrating NTC thermistor network, monitoring inductance and capacitance temperature drift in real time, and correcting parameter value through linear regression model.

7. The intelligent commissioning system of an LC filter of claim 6, wherein, The LC filter after debugging is subjected to index verification and evaluation: The indexes comprise frequency domain performance, time domain performance, energy efficiency index and stability, the frequency domain performance comprises cutoff frequency error, stopband attenuation and passband ripple, the time domain performance comprises step response rise time and overshoot rate, the energy efficiency index comprises insertion loss and power loss, and the stability comprises temperature drift coefficient and element parameter tolerance sensitivity.

Images