A shap-based naked mole rat algorithm filtering antenna design method

Abstract

The application discloses a filter antenna design method based on a naked mole rat algorithm of SHAP, and relates to the cross field of a radio frequency front end of a wireless communication system and a computer algorithm. The application quantifies the influence of antenna structure parameters on the performance of the antenna by using SHAP values, dynamically adjusts the search strategy in the NMRA, and enhances the exploration ability of the algorithm in a high-dimensional space. First, the SHAP values of each antenna parameter are initialized, and then the NMRA algorithm stage is entered. In the working stage, SHAP weighted differential evolution is performed, in the breeding stage, all breeders update their own states with a certain probability, and breeders with poor performance are eliminated; then adaptive mutation is performed with a probability pm to prevent falling into a local optimum; finally, the new population is evaluated, the current optimal solution is updated, and after the maximum number of iterations is reached, the optimal solution is output. The application is mainly applied in the fields of short-wave satellite communication, radar and the like, overcomes the limitation that a uniform search strategy is adopted for all features in a high-dimensional optimization problem in a traditional NMRA algorithm, and guarantees that the filter antenna has good performance.

Description

TECHNICAL FIELD

[0001] The present application relates to the cross field of radio frequency front end of wireless communication system and computer algorithm, and particularly to a filter antenna design method based on SHAP naked mole rat algorithm. BACKGROUND

[0002] In recent years, meta-heuristic algorithms have shown significant advantages in the field of antenna optimization. For example, genetic algorithm (GA), particle swarm optimization (PSO), differential evolution (DE) and grey wolf optimization (GWO) algorithms can perform global exploration in complex search space by simulating natural phenomena or swarm intelligence, and have been successfully applied to the design of various different antennas. Among them, the naked mole rat algorithm (NMRA) is a new meta-heuristic algorithm that simulates the social behavior and reproductive mechanism of naked mole rats. With its strong global exploration ability and local development ability, it has shown great potential in the field of antenna optimization. Studies have shown that NMRA performs well in handling high-dimensional, multi-modal optimization problems, and can effectively balance the relationship between exploration and development, avoiding premature convergence.

[0003] At the same time, the application of explainable machine learning technology in antenna design has also gradually attracted attention. SHAP (SHapley Additive exPlanations) as an explanation framework based on game theory, can quantify the contribution of each feature to the model output, providing intuitive and reliable explanations for black box models. Although SHAP was originally used for explainability analysis of machine learning models, it has great potential in feature importance evaluation and decision guidance in optimization problems.

[0004] Current antenna optimization methods can be mainly divided into three categories: traditional mathematical optimization methods, intelligent optimization algorithms and hybrid optimization frameworks. Traditional mathematical optimization methods such as gradient descent method, conjugate gradient method, etc. rely on the convexity and derivative information of the problem, and it is often difficult to find the global optimal solution for complex antenna optimization problems that are non-convex and non-differentiable. Intelligent optimization algorithms do not rely on gradient information and have strong global search ability, but also have some inherent defects. For example, genetic algorithm is prone to premature convergence; particle swarm optimization algorithm is prone to local optimum; while naked mole rat algorithm has advantages in exploration ability, but it converges slowly when dealing with high-dimensional problems. Therefore, to provide an efficient and reliable new method for antenna performance optimization and solve the problem of easy falling into local optimum in high-dimensional problems in antenna optimization is still a difficult problem to be solved. SUMMARY

[0005] Therefore, the application provides a filter antenna design method based on a naked mole rat algorithm based on SHAP.

[0006] In order to achieve the above-mentioned purpose, the technical scheme adopted by the application is as follows:

[0007] A filter antenna design method based on a naked mole rat algorithm based on SHAP, comprising the following steps:

[0008] Step 1, initialize the breeding probability pb, the mutation probability pm, and initialize the naked mole rat population, wherein the number of breeding mice B=n / 5, the number of working mice W=n-B; wherein n is the naked mole rat population size;

[0009] Step 2, associate the electromagnetic simulation software with the antenna model;

[0010] Step 3, set the maximum number of iterations of the naked mole rat algorithm max-iteration, and execute iteration based on SHAP guidance; complete the determination of the filter antenna parameters.

[0011] Further, the formula for initializing the naked mole rat population in step 1 is as follows:

[0012]

[0013] wherein , , is the i-th solution in the j-th dimension, D is the number of antenna structure parameters to be optimized, and are the lower bound and upper bound of the j-th antenna structure parameter to be optimized, respectively, is a uniform random number in.

[0014] Further, the specific process of step 3 is as follows:

[0015] Step 301: For each individual in the current naked mole rat population, calculate its corresponding SHAP value;

[0016] Step 302: Calculate the target function value corresponding to the current naked mole rat population, update and record the global optimal solution;

[0017] Step 303: Sort the antenna bandwidth corresponding to each individual in the current naked mole rat population, and record the B individuals with the largest bandwidth as the breeding mice in the current naked mole rat population, and the remaining W individuals as the working mice in the current naked mole rat population;

[0018] Step 304: Enter the worker phase, determine the weight according to the SHAP value, and update the worker mouse;

[0019] Step 305: Enter the breeding phase, judge whether to move to the optimal place according to the breeding probability pb, and update the breeding mouse;

[0020] Step 306: Enter the mutation phase, perform SHAP adaptive mutation operation on the updated worker mouse and breeding mouse;

[0021] Step 307: Determine whether the maximum number of iterations max-iteration is reached, if reached, calculate the objective function value corresponding to the current naked mole rat population, and output the global optimal solution after updating and recording the global optimal solution, if not reached, repeat steps 301 to 307.

[0022] Further, the specific process of step 301 is as follows:

[0023]

[0024]

[0025]

[0026] Wherein represents a feature set composed of the values of D dimensions of the i th individual in the current naked mole rat population, represents the set obtained by deleting from the feature set , is any feature subset of , including the empty set, represents the number of elements in the set, and! represents factorial; is the marginal contribution of ; is the prediction contribution value of the feature subset , which is given by the random forest model; and represent the maximum and minimum values of the D dimensions of the i th individual in the current naked mole rat population. That is, the SHAP value of the j th dimension solution of the i th individual in the current naked mole rat population.

[0027] Further, the specific process of step 302 is as follows:

[0028] Step 3021: For the D-dimensional antenna structure parameters contained in each individual in the current naked mole rat population, simulate the corresponding antenna model according to the electromagnetic simulation software, and obtain the corresponding flatness and bandwidth BW:​

[0029]

[0030] in, and These are the sets of local maximum and local minimum values ​​of the S11 curve within the target frequency band;

[0031] Step 3022: If the current naked mole-rat population contains For individuals whose bandwidth B is not 0, the maximum bandwidth value is selected from those individuals and recorded as the objective function value of the current naked mole-rat population; otherwise, the objective function value of the current naked mole-rat population is recorded as... This refers to the penalty mechanism; the globally optimal solution is the solution obtained by excluding all iterations. The maximum objective function value other than [the objective function value].

[0032] Furthermore, the specific process of step 304 is as follows:

[0033]

[0034]

[0035] in, In the t-th iteration, the first... The SHAP value corresponding to the j-th dimension solution of the working mouse. In the t-th iteration, the first... The j-th dimension solution of a working mouse. This indicates its corresponding updated value. and This represents the j-th dimension solution of two random working mice in the t-th iteration. It is a scaling factor based on LEVY flight, ranging from... Control the update step size. .

[0036] Furthermore, the specific process of step 305 is as follows:

[0037]

[0038]

[0039] in, In the t-th iteration, the first... The SHAP value corresponding to the j-th dimension solution of a breeding mouse. In the t-th iteration, the first... The j-th dimension solution of a breeding mouse. This indicates its corresponding updated value. the jth dimension solution of the individual corresponding to the current global optimal solution, the jth dimension solution of the individual generated by the random probability for the tth iteration, the jth dimension solution of the individual generated by the random probability for the tth iteration, is a random variable subject to uniform distribution, ranging from [-1, 1], is a scaling factor based on LEVY flight, ranging from , controls the update step size, .

[0040] Further, the SHAP adaptive mutation probability in the step 306 is as follows:

[0041] .

[0042] Thanks to the above technical solutions, the present application has the following advantages over the prior art:

[0043] 1. The present application introduces the SHAP (SHapley Additive exPlanations) explainability analysis method, which quantifies the contribution of each feature parameter to the optimization objective and dynamically adjusts the search strategy: for high SHAP feature values, a local fine search is adopted to deeply mine potential high-quality solutions; for low SHAP feature values, the global exploration ability is enhanced to avoid falling into local optimum. This mechanism effectively overcomes the limitations of traditional NMRA algorithms in uniformly searching all features in high-dimensional optimization problems.

[0044] 2. The present application sets the weight function of the naked mole rat algorithm based on the SHAP value. If the current parameter importance is high, its small change may cause a large fluctuation in the optimization objective. If a large step size is used, the optimal solution may be skipped, and the optimization objective is difficult to converge. Therefore, if the parameter importance is high, its SHAP value should be large, and the weight value should be small. In the working stage, the optimization step size is small to avoid destroying the found better solution due to excessive adjustment.

[0045] 3. The present application designs a target function for filter antenna customization based on bandwidth, which ensures that the designed filter antenna has good performance. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 is the overall flowchart of the filter antenna design method based on the SHAP naked mole rat algorithm in the embodiment of the present application.

[0047] Figure 2 is a comparison diagram of the optimization results of the SG-NMRA with the maximum bandwidth as the optimization objective and the traditional algorithm in the embodiment of the present application.

[0048] Figure 3An iterative curve comparison diagram of the SG-NMRA in the embodiment of the present application compared with a traditional algorithm. DETAILED DESCRIPTION

[0049] The present application will be further described in conjunction with the accompanying drawings and specific embodiments.

[0050] A filter antenna design method based on a naked mole rat algorithm (NMRA) of SHAP, which is used to optimize the design of the filter antenna. Since there are many antenna structure parameters, and each parameter has different effects on the overall performance of the antenna, in order to analyze the influence of each design parameter of the antenna structure on the performance of the antenna, the SHAP value is used to quantify the influence of the antenna structure parameter on the performance of the antenna, and the search strategy in the NMRA is dynamically adjusted to enhance the exploration ability of the algorithm in high-dimensional space. Then enter the NMRA algorithm stage. According to the SHAP value, the weight is determined. If the importance of the current parameter is high, a small change in it may cause a large fluctuation in the optimization target, so a smaller weight is used to avoid skipping the local area where the optimal solution is located. The SHAP weighted differential evolution is carried out in the working stage; in the breeding stage, all breeders update their own state with a certain probability, which helps to eliminate breeders with poor performance; then adaptive mutation is carried out with a probability pm to prevent falling into local optimum; finally, the new population is evaluated, and the current optimal solution is updated, and after the maximum iteration number is reached, the optimal solution is output; as shown in the formula (1), the specific steps include the following steps: Figure 1 The specific steps include the following steps:

[0051] Step 1, initialize the breeding probability pb, the mutation probability pm, and initialize the naked mole rat population, wherein the number of breeding mice B=n / 5, the number of working mice W=n-B; wherein n is the naked mole rat population size, n is a multiple of 5; pb, pm∈ .

[0052] Step 2, associate the electromagnetic simulation software with the antenna model;

[0053] Step 3, set the maximum iteration number max-iteration of the naked mole rat algorithm, and execute iteration based on SHAP; complete the determination of the filter antenna parameters.

[0054] Further, the formula for initializing the naked mole rat population in step 1 is as follows:

[0055]

[0056] wherein , , is the i-th solution in the j-th dimension, D is the number of antenna structure parameters to be optimized, and are the lower bound and upper bound of the j-th antenna structure parameter to be optimized, respectively, is Uniformly random numbers within.

[0057] Furthermore, the specific process of step 3 is as follows:

[0058] Step 301: For each individual in the current naked mole-rat population, calculate its corresponding SHAP value;

[0059] Step 302: Calculate the objective function value corresponding to the current naked mole-rat population, update and record the global optimal solution;

[0060] Step 303: Sort the antenna bandwidth of each individual in the current naked mole rat population, and record the B individuals with the largest bandwidth as breeding rats in the current naked mole rat population, and the remaining W individuals as working rats in the current naked mole rat population;

[0061] Step 304: Enter the worker stage, determine the weights based on the SHAP value, and update the worker mouse;

[0062] Step 305: Enter the breeding stage. Determine whether to move to the optimal position based on the breeding probability pb, and update the breeding mice.

[0063] Step 306: Enter the mutation phase and perform SHAP adaptive mutation operation on the updated working mice and breeding mice;

[0064] Step 307: Determine whether the maximum number of iterations (max-iteration) has been reached. If it has, calculate the objective function value corresponding to the current naked mole-rat population, and output the global optimal solution after updating and recording it. If it has not been reached, repeat steps 301 to 307.

[0065] Furthermore, the specific process of step 301 is as follows:

[0066]

[0067]

[0068]

[0069] in Let represent the complete set of features consisting of the D dimensions of the i-th individual in the current naked mole-rat population. Indicates the total set of features Delete from middle The resulting set for Any feature subset, including the empty set, The symbol '!' indicates the number of elements in the set, and '!' indicates factorial. for marginal contribution For feature subset The predicted contribution value is given by the random forest model. Specifically, in this embodiment, a TreeExplainer package is called to implement it. and This represents the D dimensions of the i-th individual in the current naked mole-rat population. The maximum and minimum values ​​among the values; This is the SHAP value of the j-th dimension solution for the i-th individual in the current naked mole-rat population.

[0070] Furthermore, the specific process of step 302 is as follows:

[0071] Step 3021: For each individual in the current naked mole-rat population, based on the D-dimensional antenna structure parameters, simulate the corresponding antenna model using electromagnetic simulation software to obtain its corresponding smoothness. And bandwidth (BW):

[0072]

[0073] in, and These are the sets of local maximum and local minimum values ​​of the S11 curve within the target frequency band;

[0074] Step 3022: If the current naked mole-rat population contains For individuals whose bandwidth B is not 0, the maximum bandwidth value is selected from those individuals and recorded as the objective function value of the current naked mole-rat population; otherwise, the objective function value of the current naked mole-rat population is recorded as... This refers to the penalty mechanism; the globally optimal solution is the solution obtained by excluding all iterations. The maximum objective function value other than [the objective function value].

[0075] Furthermore, the specific process of step 304 is as follows:

[0076]

[0077]

[0078] in, In the t-th iteration, the first... The SHAP value corresponding to the j-th dimension solution of the working mouse. In the t-th iteration, the first... The j-th dimension solution of a working mouse. This indicates its corresponding updated value. and This represents the j-th dimension solution of two random working mice in the t-th iteration. is the scaling factor based on LEVY flight, ranging from , which controls the update step size, . is the parameter weight obtained from the SHAP value, which adjusts the step size adaptively.

[0079] Further, the specific process of step 305 is as follows:

[0080]

[0081]

[0082] wherein, represents the SHAP value corresponding to the j-dimensional solution of the t-th iteration of the i-th breeding mouse, represents the j-dimensional solution of the t-th iteration of the i-th breeding mouse, represents the j-dimensional solution of the t-th iteration of the i-th breeding mouse, represents the j-dimensional solution of the individual corresponding to the current global optimal solution, represents the j-dimensional solution of the individual corresponding to the current global optimal solution, is a random probability generated for the j-dimensional solution of the t-th iteration of the i-th breeding mouse, is a random variable subject to uniform distribution, ranging from [-1, 1], is the scaling factor based on LEVY flight, ranging from , which controls the update step size, . Further, the SHAP adaptive mutation probability in step 306 is as follows:

[0083] .

[0084] The SHAP adaptive mutation operation is the mutation operation in the naked mole rat algorithm, and the object is the updated working mouse and breeding mouse, and the mutation probability is .

[0085]

[0086] To comprehensively evaluate the algorithm performance, this embodiment optimizes three structural parameters of the filtering antenna, selects six standard algorithms, and compares the proposed SG-NMRA algorithm with the traditional naked mole rat algorithm (NMRA), grey wolf optimization algorithm (GWO), artificial rabbit algorithm (ARO, 2022), Coati optimization algorithm (CoatiOA, 2023), egret swarm optimization algorithm (ESOA, 2022), and golden jackal optimization algorithm (GJO, 2022). Figure 2 The optimization results of each algorithm when the maximum bandwidth is the optimization objective are shown, and it can be seen that the SG-NMRA can obtain a wider bandwidth compared with the other algorithms.​​Figure 3 It is shown that the SG-NMRA can converge faster than the other six algorithms in optimizing bandwidth and will not fall into local optimal solution.

[0087] Those skilled in the art will appreciate that the embodiments described are for the purpose of helping the reader understand the principles of the present application and should be construed as not limiting the scope of the present application. The present application can be variously changed and modified by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for designing a filtered antenna based on the SHAP naked mole-mouse algorithm, characterized in that, Includes the following steps: Step 1: Initialize the reproduction probability pb, the mutation probability pm, and the naked mole-rat population, where the number of breeding rats B = n / 5 and the number of working rats W = nB; where n is the size of the naked mole-rat population. Step 2: Connect the electromagnetic simulation software with the antenna model; Step 3: Set the maximum number of iterations (max-iteration) for the naked mole-mole algorithm and guide the iteration based on SHAP; complete the determination of the filter antenna parameters, specifically: Step 301: For each individual in the current naked mole-rat population, calculate its corresponding SHAP value; Step 302: Calculate the objective function value corresponding to the current naked mole-rat population, update and record the global optimal solution; Step 303: Sort the antenna bandwidth of each individual in the current naked mole rat population, and record the B individuals with the largest bandwidth as breeding rats in the current naked mole rat population, and the remaining W individuals as working rats in the current naked mole rat population; Step 304: Enter the worker stage, determine the weights based on the SHAP value, and update the worker mouse; Step 305: Enter the breeding stage. Determine whether to move to the optimal position based on the breeding probability pb, and update the breeding mice. Step 306: Enter the mutation phase and perform SHAP adaptive mutation operation on the updated working mice and breeding mice; Step 307: Determine whether the maximum number of iterations (max-iteration) has been reached. If it has, calculate the objective function value corresponding to the current naked mole-rat population, and output the global optimal solution after updating and recording it. If it has not been reached, repeat steps 301 to 307.

2. The method for designing a filtered antenna based on the SHAP naked mole-mouse algorithm according to claim 1, characterized in that, The formula for initializing the naked mole-rat population in step 1 is as follows: ； in , , It is the i-th solution in the j-th dimension, where D is the number of antenna structure parameters to be optimized. and These are the lower and upper bounds of the j-th antenna structure parameter to be optimized, respectively. yes Uniformly random numbers within.

3. The method for designing a filtered antenna based on the SHAP naked mole-mouse algorithm according to claim 2, characterized in that, The specific process of step 301 is as follows: ； ； ； in Let represent the complete set of features consisting of the D dimensions of the i-th individual in the current naked mole-rat population. Indicates the total set of features Delete from middle The resulting set for Any feature subset, including the empty set, The symbol '!' indicates the number of elements in the set, and '!' indicates factorial. for marginal contribution For feature subset The predicted contribution value is given by the random forest model; and This represents the D dimensions of the i-th individual in the current naked mole-rat population. The maximum and minimum values ​​among the values; This is the SHAP value of the j-th dimension solution for the i-th individual in the current naked mole-rat population.

4. The method for designing a filtered antenna based on the SHAP naked mole-mouse algorithm according to claim 2, characterized in that, The specific process of step 302 is as follows: Step 3021: For each individual in the current naked mole-rat population, based on the D-dimensional antenna structure parameters, simulate the corresponding antenna model using electromagnetic simulation software to obtain its corresponding smoothness. And bandwidth (BW): ； in, and These are the sets of local maximum and local minimum values ​​of the S11 curve within the target frequency band; Step 3022: If the current naked mole-rat population contains For individuals whose bandwidth B is not 0, the maximum bandwidth value is selected from those individuals and recorded as the objective function value of the current naked mole-rat population; otherwise, the objective function value of the current naked mole-rat population is recorded as... This refers to the penalty mechanism; the globally optimal solution is the solution obtained by excluding all iterations. The maximum objective function value other than [the objective function value].

5. The method for designing a filtered antenna based on the SHAP naked mole-mouse algorithm according to claim 3, characterized in that, The specific process of step 304 is as follows: ； ； in, In the t-th iteration, the first... The SHAP value corresponding to the j-th dimension solution of the working mouse. In the t-th iteration, the first... The j-th dimension solution of a working mouse. This indicates its corresponding updated value. and This represents the j-th dimension solution of two random working mice in the t-th iteration. It is a scaling factor based on LEVY flight, ranging from... Control the update step size. .

6. The method for designing a filtered antenna based on the SHAP naked mole-mouse algorithm according to claim 3, characterized in that, The specific process of step 305 is as follows: ； ； in, In the t-th iteration, the first... The SHAP value corresponding to the j-th dimension solution of a breeding mouse. In the t-th iteration, the first... The j-th dimension solution of a breeding mouse. This indicates its corresponding updated value. Let be the j-th dimension solution of the individual corresponding to the current global optimal solution. For the t-th iteration The random probability of generating the j-th dimension solution of a breeding mouse. Let be a random variable that follows a uniform distribution and has a range of [-1, 1]. It is a scaling factor based on LEVY flight, ranging from... Control the update step size. .

7. The method for designing a filtered antenna based on the SHAP naked mole-mouse algorithm according to claim 3, characterized in that, The SHAP adaptive mutation probability in step 306 is as follows: 。

Images