Filter optimization method based on artificial surface plasmon

Abstract

The invention discloses a filter optimization method based on artificial surface plasmon, and relates to the field of signal processing, and the method comprises the steps: S1, monitoring and collecting the real-time operation parameters of each filter; s2, presetting a reference amplitude range of the working signal; s3, characteristic parameters of artificial surface plasmons are preset for each filter, and the part, exceeding the reference amplitude range, of the frequency amplitude of the working signal is set as an abnormal frequency value; and S4, constructing a filter waveguide model, and setting an iteration output value as a correction operation parameter of a corresponding filter. Compared with the prior art, the abnormal frequency of the filter can be quickly identified and marked by monitoring the operation parameters of the filter in real time, the possible coupling effect can be quickly positioned by identifying the abnormal frequency in time, the debugging and correction time is shortened, and the debugging efficiency is improved. The method has the advantages that abnormal operation of the filter can be found and corrected as soon as possible, and loss and failure rate of equipment are reduced.

Description

Technical Field

[0001] The present invention relates to the field of signal processing, and in particular to a filter optimization method based on artificial surface plasmons. Background Art

[0002] With the rapid development of wireless communication, sensor technology and microwave devices, the design of low-loss devices based on artificial surface plasmons has received widespread attention. Such devices have excellent electromagnetic properties and can achieve efficient signal transmission and processing. However, in practical applications, the problem of electromagnetic crosstalk has become an important factor affecting its performance. Electromagnetic crosstalk refers to the interference between multiple signal channels caused by the interaction of electromagnetic fields, which may cause problems such as signal distortion and reduced transmission efficiency. In the control of electromagnetic crosstalk, effective decoupling technology aims to reduce or eliminate the mutual interference between different signal channels, thereby ensuring the integrity and transmission quality of the signal. The existing artificial surface plasmon filter design often leads to an increase in local electric field strength, which may lead to increased coupling between adjacent signal channels and form a waveguide structure. The signal may interfere with the signal of other channels during propagation, thereby causing crosstalk, and the filter has the possibility of additional electromagnetic crosstalk. Since the influence of the waveguide effect is present in the entire circuit device, the efficiency of troubleshooting the filter in the circuit is affected. Summary of the invention

[0003] The present invention provides a filter optimization method based on artificial surface plasmons, which solves the problem in the prior art that the filter operation status cannot be checked in time when the artificial surface plasmons generate a waveguide effect leading to an additional coupling effect in the filter.

[0004] The present invention is achieved through the following technical solutions: A filter optimization method based on artificial surface plasmons, the method comprising: Step S1: setting rated operating parameters for each filter in the circuit and marking corresponding working blocks in the circuit, operating the filter after the plasmon processing, and monitoring and collecting the real-time operating parameters of each filter; Step S2: after the working signal is input into the circuit, the operating parameters of the target filter in each working block are collected, and the working signal is placed in the frequency domain to obtain a spectrum diagram of the working signal, and a reference amplitude range of the working signal is preset in the spectrum diagram; Step S3: preset characteristic parameters of artificial surface plasmon for each filter, set the part of the working signal frequency amplitude that exceeds the reference amplitude range as an abnormal frequency value, and mark the working block with the abnormal frequency value as an abnormal block; Step S4: construct a filter waveguide model, take the real-time operating parameters of the abnormal block as input values ​​and input the characteristic parameters as constraint values ​​into the filter waveguide model for iteration, and set the iterative output values ​​as the corrected operating parameters of the corresponding filter; Among them, the variable types of the rated operating parameters, the real-time operating parameters and the corrected operating parameters are the same.

[0005] In the control of electromagnetic crosstalk, effective decoupling technology aims to reduce or eliminate mutual interference between different signal channels, thereby ensuring signal integrity and transmission quality. The existing artificial surface plasmon filter design often leads to an increase in local electric field strength, which may lead to an increase in the coupling between adjacent signal channels, forming a waveguide structure. The signal may interfere with the signal of other channels during propagation, thereby causing crosstalk, and there is a possibility of additional electromagnetic crosstalk to the filter. Since the influence of the waveguide effect is present in the entire circuit device, the efficiency of troubleshooting the filter in the circuit is affected. Based on this, the present invention provides a filter optimization method based on artificial surface plasmons to solve the problem of untimely troubleshooting of the filter operation when the artificial surface plasmon generates a waveguide effect in the prior art, resulting in an additional coupling effect in the filter.

[0006] Furthermore, the filter waveguide model uses a random forest network to perform electromagnetic simulation prediction iterations on the real-time operating parameters of each filter, including: The real-time operating parameters and characteristic parameters of each filter are collected and preprocessed, and the hyperparameters of the random forest network are constructed before iterative training, and the hyperparameters of the random forest network are mapped to the characteristic parameters of the plasmon; a training data set used as the initial data for iteration is constructed based on the input real-time operating parameters, and the number of iterations is preset; the hyperparameters are used as iterative constraint parameters of the training data set for directional constraints, the rated operating parameters are used as the reference threshold of the iterative output of the real-time operating parameters, and a floating interval is set for each variable; when the number of iterations is exhausted or all variables of the real-time operating parameters of a certain iteration are within the floating interval of the reference threshold, the iteration is stopped and the output value is set to the corrected operating parameter.

[0007] Furthermore, the hyperparameters of the random forest network include the number of trees, the maximum tree depth and the minimum number of tree splits, and the characteristic parameters include structural characteristic values, dielectric conductivity and electromagnetic resonance frequency, wherein the structural characteristic value is obtained by weighted summation of the line width and depth of the plasmon; the structural characteristic value is parameter mapped to the number of trees, the dielectric conductivity is parameter mapped to the maximum tree depth, and the electromagnetic resonance frequency is parameter mapped to the minimum number of tree splits.

[0008] Furthermore, a convergence function is set for the hyperparameter, and the convergence function is used to minimize the loss function in the random forest network along the directional constraints of the hyperparameter. The setting form of the convergence function includes: Define the loss function and express it as L(θ), where θ represents the hyperparameter in the random forest network; set the total number and ordinal number of filters and real-time operation parameters as N and i respectively, and set the true value and predicted value of the real-time operation parameter input sample as p and q respectively, then the calculation formula of the loss function is expressed as: ; Suppose that the penalty term representing the degree of non-compliance of the quantized hyperparameter with the constraint is denoted as R(θ), the penalty coefficient used to control the weight between the objective function and the constraint penalty is set as λ, and the convergence function is set as C(θ), then the calculation formula of the convergence function C(θ) is expressed as: C(θ)= L(θ)+ λ∙ R(θ); The calculation formula of the merged convergence function C(θ) is expressed as: .

[0009] Furthermore, the setting content of the penalty term includes: let the ordinal number of the hyperparameter be represented by j, let the constraint value be represented by μ, wherein the constraint value μ includes the maximum constraint value μ max and the minimum constraint value μ min , The penalty term is calculated as: .

[0010] Furthermore, the penalty term is increased when the hyperparameters do not meet the constraints, which promotes the optimization of adjusting the hyperparameters to meet the constraints.

[0011] Furthermore, the difference between the loss function values ​​in two consecutive iterations is set as the loss change to represent the loss change amplitude of the random forest network iteration process; a dynamic threshold is set for the loss change, and the numerical value of the dynamic threshold of the convergence function is dynamically adjusted according to the current change in the loss function value.

[0012] Furthermore, when the loss change is greater than the dynamic threshold, the dynamic threshold is increased to increase the search range of the random forest network; when the loss change is less than the dynamic threshold, the dynamic threshold is increased to converge the search range of the random forest network.

[0013] Furthermore, the setting process of the floating interval includes: setting the upper boundary of the floating interval at a position where the reference threshold increases by 20% along the increasing direction, and setting the lower boundary of the floating interval at a position where the reference threshold decreases by 20% along the decreasing direction.

[0014] Furthermore, a multi-dimensional floating interval is set in the random forest network to maintain the mapping state between the hyperparameters and the feature parameters.

[0015] Compared with the prior art, the present invention can quickly identify any abnormal situation through real-time monitoring of the filter operating parameters, so that the abnormal frequency of the filter affected by the waveguide effect can be quickly identified and marked. Timely identification of abnormal frequency helps to quickly locate possible coupling effects and reduce the time for debugging and correction. It has the advantages of early detection and correction of filter operating abnormalities and the beneficial effect of reducing equipment loss and failure rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings: Figure 1 It is a flowchart of the present invention. DETAILED DESCRIPTION

[0017] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention. Example

[0018] like Figure 1 As shown, this embodiment is a filter optimization method based on artificial surface plasmons, and the method includes: Step S1: setting rated operating parameters for each filter in the circuit and marking corresponding working blocks in the circuit, operating the filter after the plasmon processing, and monitoring and collecting the real-time operating parameters of each filter; Step S2: after the working signal is input into the circuit, the operating parameters of the target filter in each working block are collected, and the working signal is placed in the frequency domain to obtain a spectrum diagram of the working signal, and a reference amplitude range of the working signal is preset in the spectrum diagram; Step S3: preset characteristic parameters of artificial surface plasmon for each filter, set the part of the working signal frequency amplitude that exceeds the reference amplitude range as an abnormal frequency value, and mark the working block with the abnormal frequency value as an abnormal block; Step S4: construct a filter waveguide model, take the real-time operating parameters of the abnormal block as input values ​​and input the characteristic parameters as constraint values ​​into the filter waveguide model for iteration, and set the iterative output values ​​as the corrected operating parameters of the corresponding filter; Among them, the variable types of the rated operating parameters, the real-time operating parameters and the corrected operating parameters are the same.

[0019] The rated operating parameters refer to the key parameters that are preset during the design and manufacturing process of the filter and represent its optimal working state. They are used to describe the performance standards of the filter under normal operating conditions and provide a benchmark for the optimization and troubleshooting of the filter. In a specific implementation, the rated operating parameters may include operating parameters such as center frequency value, bandwidth value, insertion and reflection loss value, and phase characteristics, and the rated operating parameters, real-time operating parameters, and modified operating parameters have the same parameter type and can correspond one by one. The working block of the filter in the circuit refers to the area designated for a specific function or task in the circuit design. Each block is responsible for a specific signal processing or functional operation, such as filtering, amplification, modulation, etc., to facilitate the management and optimization of each function; each working block is designed in the order of the signal flow to ensure that the signal is smoothly transmitted from one block to another to maintain the integrity of the circuit. The process of operating the filter after plasmon treatment is to integrate the filter into the circuit after using artificial surface plasmon treatment to start its actual operation to verify its performance under the design parameters, and monitor its operating parameters (such as frequency, amplitude, etc.) in real time during the operation of the filter to evaluate whether it has achieved the preset operating goals and identify any possible abnormalities or coupling effects. In the spectrum diagram, compare with the preset reference amplitude range to identify the frequency components of the operating signal that are out of range. These frequency components may indicate that there is an abnormality in the signal or a system failure.

[0020] More, the working signal in the time domain is converted into a frequency domain signal to obtain a spectrum diagram, and a reference amplitude range is set on the spectrum diagram to indicate the signal strength under normal working conditions. The reference amplitude range in the specific application can be set based on previous experimental data or theoretical analysis. The characteristic parameters of the artificial surface plasmon preset by each filter represent the specific application parameter content used for the artificial surface plasmon unit in the specific implementation. By detecting whether the frequency amplitude exceeds the reference range, the abnormal blocks in the circuit are identified and marked. A filter waveguide model is established, and iterative optimization is performed through the real-time parameters and SPP characteristic parameters of the abnormal blocks, and the corrected operating parameters are output. The characteristic parameters are used as constraints to enable the effective operating parameters of the target filter in the circuit to meet the application of artificial surface plasmons. The iterative mechanism is used to adjust the abnormal frequency to achieve adaptive correction of the filter operating state, thereby optimizing the filter performance and maintaining low loss characteristics. This implementation method can quickly locate and solve problems, reduce the time and labor costs of manual troubleshooting, and reduce the overall cost of equipment maintenance.

[0021] Further, as a feasible implementation, the filter waveguide model uses a random forest network to perform electromagnetic simulation prediction iterations on the real-time operating parameters of each filter, the contents of which include: The real-time operating parameters and characteristic parameters of each filter are collected and preprocessed, and the hyperparameters of the random forest network are constructed before iterative training, and the hyperparameters of the random forest network are mapped to the characteristic parameters of the plasmon; a training data set used as the initial data for iteration is constructed based on the input real-time operating parameters, and the number of iterations is preset; the hyperparameters are used as iterative constraint parameters of the training data set for directional constraints, the rated operating parameters are used as the reference threshold of the iterative output of the real-time operating parameters, and a floating interval is set for each variable; when the number of iterations is exhausted or all variables of the real-time operating parameters of a certain iteration are within the floating interval of the reference threshold, the iteration is stopped and the output value is set to the corrected operating parameter.

[0022] The random forest network is used to iterate the prediction of the real-time operating parameters so that the operating state of the filter can be more consistent with the characteristics of the plasmon, thereby avoiding the waveguide effect and the resulting additional coupling effect. Random forests can handle a large number of input features and select the parameters that have the greatest impact on the prediction results by evaluating the importance of the features. This can help screen the real-time operating parameters that are most relevant to the plasmon characteristics. Collect the real-time operating parameters of each filter that reflects the current working state of the filter, and collect the characteristic parameters related to the implementation of artificial surface plasmons. Clean and standardize the collected data to ensure the consistency and availability of the data. During the model training process, the hyperparameters are used as constraints to guide the model to optimize in the expected direction. Such directional constraints help improve the learning efficiency of the model. The rated operating parameters are set as the reference threshold of the real-time operating parameters, with the goal of ensuring that the parameters of the iterative output can be within the set floating range. The floating range set for each variable is allowed to vary within a certain range to adapt to the actual operating requirements of the filter. In each iteration, the real-time operating parameters are input into the random forest model for prediction. The parameters output by the model will be used as new operating parameters and input into the model again for further prediction. This feedback mechanism can gradually adjust the parameters to optimize them in a direction that is more consistent with plasmon properties.

[0023] Furthermore, as a feasible implementation method, the hyperparameters of the random forest network include the number of trees, the maximum tree depth and the minimum number of tree splits, and the characteristic parameters include structural characteristic values, dielectric conductivity and electromagnetic resonance frequency, wherein the structural characteristic value is obtained by weighted summation of the line width and depth of the plasmon; the structural characteristic value is parameter mapped to the number of trees, the dielectric conductivity is parameter mapped to the maximum tree depth, and the electromagnetic resonance frequency is parameter mapped to the minimum number of tree splits.

[0024] The number of trees refers to the total number of decision trees constructed in the random forest; the more trees there are, the higher the stability and accuracy of the model is, but the computational cost also increases accordingly. The maximum tree depth refers to the maximum depth of a single decision tree. The greater the depth, the greater the complexity and fitting ability of the model, but too deep may lead to overfitting. The minimum number of tree splits represents the minimum number of samples required when the decision tree splits, which affects the growth and complexity of the tree; setting a reasonable minimum number of splits helps to improve the generalization ability of the model. The structural characteristic value is obtained by weighted summation of the line width and depth of the plasmon, reflecting the geometric and electromagnetic properties of the material. Among the plasmon parameters, the line width and depth directly affect the propagation and coupling characteristics of the surface plasmon. A wider line width may enhance the coupling between light and the surface plasmon, but it may also introduce greater losses; the depth affects the degree of localization of the mode. The structural characteristic value is set by weighted summation of the line width and depth of the plasmon structure to form a quantitative description of the impact on the filter performance, which can realize quantitative analysis and optimization design of the filter performance, so that it can better adapt to specific operating frequencies and environmental conditions. The dielectric conductivity describes the dielectric material's ability to conduct current, and has a direct impact on the loss and transmission characteristics of plasmons. The electromagnetic resonance frequency refers to a specific frequency at which the filter operates, and refers to the frequency at which resonance occurs when electromagnetic waves interact with matter (such as filters, antennas, plasmon structures, etc.) under specific conditions. At this frequency, the system's response will be significantly enhanced, resulting in the effective transfer or accumulation of energy, reflecting the resonance characteristics of the material at a specific frequency, affecting the selectivity and transmission performance of the signal.

[0025] Combining the structural characteristic value with the number of trees ensures that the complexity of the model matches the material properties. For example, in a specific implementation, a filter with a larger structural characteristic value may require more trees to capture complex features. Changes in the conductivity of the medium may affect the loss characteristics of the plasmon, so by mapping it to the maximum tree depth, the sensitivity of the model to the loss characteristics can be adjusted. High conductivity may require a larger depth to better capture its characteristics. Mapping the electromagnetic resonance frequency to the minimum number of tree splits can control the response of the model to specific frequency characteristics; when the frequency characteristics are complex, the minimum number of splits can be reduced to enable the tree to perform more detailed splitting and modeling. When working, collect the real-time operating parameters of the target filter, set the hyperparameters of the random forest based on the value of the characteristic parameters, realize dynamic adjustment through the above mapping relationship, use the preset hyperparameters to build a random forest model, train the model and make predictions; in the iterative process, the final output correction parameters will be applied to the actual operation of the filter to optimize its performance and ensure the optimization of signal processing effects.

[0026] Further, as a feasible implementation, a convergence function is set for the hyperparameter, and the convergence function is used to minimize the loss function in the random forest network along the directional constraints of the hyperparameter. The setting form of the convergence function includes: Define the loss function and express it as L(θ), where θ represents the hyperparameter in the random forest network; set the total number and ordinal number of filters and real-time operation parameters as N and i respectively, and set the true value and predicted value of the real-time operation parameter input sample as p and q respectively, then the calculation formula of the loss function is expressed as: ; Suppose that the penalty term representing the degree of non-compliance of the quantized hyperparameter with the constraint is denoted as R(θ), the penalty coefficient used to control the weight between the objective function and the constraint penalty is set as λ, and the convergence function is set as C(θ), then the calculation formula of the convergence function C(θ) is expressed as: C(θ)= L(θ)+ λ∙ R(θ); The calculation formula of the merged convergence function C(θ) is expressed as: .

[0027] In each iteration, the loss between the true value and the predicted value is calculated based on the current model parameters. The loss function L(θ) represents the prediction error of the model during training. The goal is to minimize this value through the optimization algorithm to improve the prediction accuracy of the model. The total number of real-time operating parameters N represents the total number of training instances, and the number is the same as the total number of filters in the circuit. i represents the true value of the i-th sample, and represents the target value actually observed. For example, in the application of filters, this may be the performance indicator that the filter should achieve under specific conditions. Similarly, q i represents the predicted value of the i-th sample, which represents the output calculated by the model based on the current parameters, and is the result of predicting the input sample through the random forest network. (p i -q i ) 2Represents the difference between the true value and the predicted value. The role of the square is to ensure that the error is positive, while amplifying larger errors so that more attention is paid to those predictions with larger deviations during the optimization process. The penalty term represents a term used to quantify the degree to which the hyperparameter does not meet the constraints; the penalty term can include various constraints, such as the hyperparameter must be kept within a specific range or the relationship with the feature parameter must meet certain conditions. The penalty coefficient λ represents the weight between the control loss function and the penalty term, which determines the balance between the degree of attention paid to the loss and the degree of attention paid to the hyperparameter constraints during the optimization process. In the specific setting process, a stable λ value needs to be selected. When the λ value is large, the emphasis on hyperparameter constraints will be increased, which may lead to a less lossy but less flexible model; conversely, it may lead to overfitting. During the convergence process, the final output model parameters and hyperparameters will be applied to the actual operation of the filter to ensure that the filter performance is optimized and meets the design requirements.

[0028] Further, as a feasible implementation, the setting content of the penalty term includes: let the ordinal number of the hyperparameter be represented by j, let the constraint value be represented by μ, where the constraint value μ includes the maximum constraint value μ max and the minimum constraint value μ min , The penalty term is calculated as: ; Among them, the penalty term is increased when the hyperparameters do not meet the constraints, which promotes the optimization and adjustment of the hyperparameters to meet the constraints; the difference between the loss function values ​​in two consecutive iterations is set as the loss change amount, which is used to represent the loss change amplitude of the random forest network iteration process; a dynamic threshold is set for the loss change amount, and the numerical value of the dynamic threshold of the convergence function is dynamically adjusted according to the current change in the loss function value.

[0029] The θ j represents the current value of the jth hyperparameter, and the penalty term represents the degree of penalty for the hyperparameter not meeting the constraint. The goal is to increase the penalty to encourage the hyperparameter θ j During the training process, the max(0, θ j -μ max ) indicates the calculation of the current hyperparameter θ j Whether it exceeds the maximum constraint value μ max , if the hyperparameter θ j Exceeding the maximum constraint value μ max , then θ j -μ max If it is positive, return this positive value; otherwise return 0; the purpose of this part is to quantify the extent to which the hyperparameter exceeds the maximum constraint. min -θ j ) indicates the calculation of the current hyperparameter θ jIs it below the minimum constraint value μ? min , if the hyperparameter θ j Below the minimum constraint value μ min , then μ min -θ j If it is positive, return this positive value; otherwise return 0; the purpose of this part is to quantify the degree to which the hyperparameter is below the minimum constraint. The summation process represents the summation of all hyperparameters, and the penalty terms of all hyperparameters are aggregated to reflect the overall degree of non-compliance with the constraints, so as to ensure that the hyperparameter set of the entire model meets the constraints, rather than a single parameter. This embodiment quantifies the degree to which the hyperparameters do not meet the constraints to form a penalty term, aiming to guide the hyperparameters to remain within the set range during the training process. After each iteration, the loss change is calculated and used to evaluate the convergence of the model. The dynamic threshold is used to be adjusted as the model training progresses to ensure that the sensitivity of the convergence function adapts to the current training state. If the loss change is large, it may be necessary to lower the threshold to promote more frequent hyperparameter adjustments; if the change is small, the threshold can be appropriately increased to reduce the frequency of adjustment to avoid unnecessary oscillations.

[0030] Furthermore, as a feasible implementation method, when the loss change is greater than the dynamic threshold, the dynamic threshold is increased to increase the search range of the random forest network; when the loss change is less than the dynamic threshold, the dynamic threshold is increased to converge the search range of the random forest network; the setting process of the floating interval includes: setting the upper boundary of the floating interval at the position where the reference threshold is increased by 20% in the increasing direction, and setting the lower boundary of the floating interval at the position where the reference threshold is reduced by 20% in the decreasing direction; setting a multidimensional floating interval in the random forest network to maintain the mapping state of the hyperparameters and feature parameters.

[0031] The dynamic threshold is a value dynamically adjusted according to the current loss change, which is used to judge the degree of model convergence and the flexibility of hyperparameter adjustment. When the loss change is greater than the dynamic threshold, it means that the prediction error of the model has large fluctuations, and a wider hyperparameter search range may be required to explore a better hyperparameter combination; in this case, the dynamic threshold is increased to relax the search restrictions and allow a wider range of hyperparameter changes. When the loss change is less than the dynamic threshold, it means that the model has a good convergence state, the loss change is small, and the hyperparameter adjustment can be more refined; in this case, the dynamic threshold is also increased to converge the search range, reduce unnecessary hyperparameter adjustments, and prevent model overfitting. The multi-dimensional floating interval is used to maintain the mapping state between hyperparameters and feature parameters, which can ensure that during the optimization process, the adjustment of hyperparameters is not only based on the current loss function, but also considers the relationship with the feature parameters, so as to maintain the effectiveness of the model.

[0032] The floating interval is used to define the upper and lower limits of the reference threshold to allow the real-time operating parameters to fluctuate within a certain range, thereby enhancing the flexibility and adaptability of the model. The real-time operating parameters can fluctuate freely within 20% of the reference threshold value. In this way, it can better adapt to different working conditions and environmental changes, and avoid frequent adjustments caused by small fluctuations. In actual operation, the system is often affected by noise or small fluctuations. Setting a 20% floating interval can effectively reduce the model's sensitivity to these small changes, improve fault tolerance, and avoid system instability due to minor changes.

[0033] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A filter optimization method based on artificial surface plasmon, characterized in that: The method includes: Step S1: setting rated operating parameters for each filter in the circuit and marking corresponding working blocks in the circuit, operating the filter after the plasmon processing, and monitoring and collecting the real-time operating parameters of each filter; Step S2: after the working signal is input into the circuit, the operating parameters of the target filter in each working block are collected, and the working signal is placed in the frequency domain to obtain a spectrum diagram of the working signal, and a reference amplitude range of the working signal is preset in the spectrum diagram; Step S3: preset characteristic parameters of artificial surface plasmon for each filter, set the part of the working signal frequency amplitude that exceeds the reference amplitude range as an abnormal frequency value, and mark the working block with the abnormal frequency value as an abnormal block; Step S4: construct a filter waveguide model, take the real-time operating parameters of the abnormal block as input values ​​and input the characteristic parameters as constraint values ​​into the filter waveguide model for iteration, and set the iterative output values ​​as the corrected operating parameters of the corresponding filter; Among them, the variable types of the rated operating parameters, the real-time operating parameters and the corrected operating parameters are the same.

2. The filter optimization method based on artificial surface plasmon according to claim 1, characterized in that: The filter waveguide model uses a random forest network to iterate electromagnetic simulation prediction of the real-time operating parameters of each filter, including: The real-time operating parameters and characteristic parameters of each filter are collected and preprocessed, and the hyperparameters of the random forest network are constructed before iterative training, and the hyperparameters of the random forest network are mapped to the characteristic parameters of the plasmon; a training data set used as the initial data for iteration is constructed based on the input real-time operating parameters, and the number of iterations is preset; the hyperparameters are used as iterative constraint parameters of the training data set for directional constraints, the rated operating parameters are used as the reference threshold of the iterative output of the real-time operating parameters, and a floating interval is set for each variable; when the number of iterations is exhausted or all variables of the real-time operating parameters of a certain iteration are within the floating interval of the reference threshold, the iteration is stopped and the output value is set to the corrected operating parameter.

3. The filter optimization method based on artificial surface plasmon according to claim 2, characterized in that: The hyperparameters of the random forest network include the number of trees, the maximum tree depth and the minimum number of tree splits, and the characteristic parameters include structural characteristic values, dielectric conductivity and electromagnetic resonance frequency, wherein the structural characteristic value is obtained by weighted summation of the line width and depth of the plasmon; the structural characteristic value is parameter mapped to the number of trees, the dielectric conductivity is parameter mapped to the maximum tree depth, and the electromagnetic resonance frequency is parameter mapped to the minimum number of tree splits.

4. The filter optimization method based on artificial surface plasmon according to claim 3, characterized in that: A convergence function is set for the hyperparameter, and the convergence function is used to minimize the loss function in the random forest network along the directional constraints of the hyperparameter. The setting form of the convergence function includes: Define the loss function and express it as L(θ), where θ represents the hyperparameter in the random forest network; set the total number and ordinal number of filters and real-time operation parameters as N and i respectively, and set the true value and predicted value of the real-time operation parameter input sample as p and q respectively, then the calculation formula of the loss function is expressed as: ; Suppose that the penalty term representing the degree of non-compliance of the quantized hyperparameter with the constraint is denoted as R(θ), the penalty coefficient used to control the weight between the objective function and the constraint penalty is set as λ, and the convergence function is set as C(θ), then the calculation formula of the convergence function C(θ) is expressed as: C(θ)= L(θ)+ λ∙ R(θ); The calculation formula of the merged convergence function C(θ) is expressed as: .

5. The filter optimization method based on artificial surface plasmon according to claim 4, characterized in that: The setting content of the penalty term includes: let the ordinal number of the hyperparameter be j, let the constraint value be μ, where the constraint value μ includes the maximum constraint value μ max and the minimum constraint value μ min , The penalty term is calculated as: .

6. A filter optimization method based on artificial surface plasmon according to any one of claims 4-5, characterized in that: The penalty term is increased when the hyperparameters do not meet the constraints, promoting the optimization to adjust the hyperparameters to meet the constraints.

7. A filter optimization method based on artificial surface plasmon according to any one of claims 4-5, characterized in that: The difference between the loss function values ​​in two consecutive iterations is set as the loss change to represent the loss change amplitude of the random forest network iteration process; a dynamic threshold is set for the loss change, and the numerical value of the dynamic threshold of the convergence function is dynamically adjusted according to the current change in the loss function value.

8. The filter optimization method based on artificial surface plasmon according to claim 7, characterized in that: When the loss change is greater than the dynamic threshold, the dynamic threshold is increased to increase the search range of the random forest network; when the loss change is less than the dynamic threshold, the dynamic threshold is increased to converge the search range of the random forest network.

9. The filter optimization method based on artificial surface plasmon according to claim 2, characterized in that: The setting process of the floating interval includes: setting the upper boundary of the floating interval at a position where the reference threshold value increases by 20% along the increasing direction, and setting the lower boundary of the floating interval at a position where the reference threshold value decreases by 20% along the decreasing direction.

10. The filter optimization method based on artificial surface plasmon according to claim 2, characterized in that: A multi-dimensional floating interval is set in the random forest network to maintain the mapping state between the hyperparameters and the feature parameters.

Images