Gray wolf optimization feature selection method, device and equipment based on adaptive mechanism

Abstract

The present invention provides a method, device and equipment for optimizing feature selection of gray wolves based on an adaptive mechanism, and relates to the technical field of data processing. The present invention obtains a data set to be feature selected, initializes a gray wolf optimization algorithm; takes obtaining the current optimal solution as the goal, calculates the fitness value of each gray wolf, and obtains the current optimal three wolves; then uses the current optimal three wolves as search guides, combines a nonlinear parameter control strategy, and adopts an improved gray wolf step length update calculation method to adaptively adjust the variation intensity during the search process, update and calculate the relevant parameters, position and fitness value of each gray wolf, and continuously converge the distance between the prey and the gray wolf during the search process; by calculating the score of each gray wolf, select the solution candidate with the highest current score as the optimal solution in the current population until the maximum number of iterations is reached, and obtain the global optimal solution. The present invention is used to process feature selection in high-dimensional and complex classification, with fast convergence speed and high accuracy.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method, device and equipment for selecting gray wolf optimization features based on an adaptive mechanism. Background Art

[0002] Feature selection (FS) is a key step in machine learning and data mining. Its goal is to extract meaningful features for model prediction from the data set to reduce the dimension and redundancy of the data and improve the performance of the classifier. The Grey Wolf Optimization Algorithm (GWO) is a classic metaheuristic algorithm based on swarm intelligence. Due to its fast convergence and low parameter requirements, it is widely used in various optimization problems, including engineering optimization problems, neural networks, scheduling, power engineering problems, control engineering problems, and robotics problems. For example, Duan et al. proposed a hybrid optimization algorithm based on GWO and SCA, cHGWOSCA (collaboration-based Hybrid GWO-SCA optimizer, based on Grey Wolf and Sin-Cosine Optimization Algorithm), and applied it to photovoltaic model parameter extraction and three-constraint engineering problems. Mohammed et al. improved the GWO algorithm by combining the advantages of the Whale Optimization Algorithm (WOA) and proposed a hybrid Whale Optimization Algorithm and Grey Wolf Optimization (WOAGWO) to solve engineering problems such as pressure vessel design. In addition, some researchers have improved the performance of the GWO algorithm by introducing new learning strategies, improving search mechanisms, and adjusting control parameters. However, when solving high-dimensional classification problems, the existing algorithms have limited global search capabilities, are prone to falling into local optimal solutions, and show certain limitations in convergence speed and accuracy.

[0003] In view of this, the applicant filed this application after studying the existing technology. Summary of the invention

[0004] The present invention aims to provide a method, device and equipment for feature selection of grey wolf optimization based on an adaptive mechanism, so as to solve the problem that the existing methods are prone to fall into local optimal solutions when dealing with high-dimensional and complex optimization problems, and show certain limitations in terms of convergence speed and accuracy.

[0005] In order to solve the above technical problems, the present invention is implemented through the following technical solutions:

[0006] A gray wolf optimization feature selection method based on an adaptive mechanism is applied to the classification tasks of high-dimensional complex data in the process of machine learning, including:

[0007] Acquire a data set to be feature selected; wherein the types of the data set include text data, image data, voice data and electrical signal data;

[0008] Initialize the Grey Wolf Optimization Algorithm;

[0009] According to the data set and the gray wolf optimization algorithm, with the goal of obtaining the current optimal solution, the fitness value of each gray wolf is calculated to evaluate the pros and cons of each gray wolf, that is, to evaluate the degree of proximity between each candidate solution and the optimal solution; wherein each gray wolf represents one candidate solution in the population;

[0010] According to the fitness value, the current optimal solution is obtained Wolf, the second-best solution β wolf and the third-best solution δ wolf, that is, the current best three wolves, thus obtaining the current optimal solution;

[0011] Taking the position information of the current best three wolves as the search guide, each gray wolf, when searching for prey, combines the nonlinear parameter control strategy, adopts the improved gray wolf step length update calculation method, iteratively updates and calculates the relevant parameters of each gray wolf, and adaptively adjusts the mutation intensity during the search process, updates and calculates the position and fitness value of each gray wolf, and continuously converges the distance between the prey and the gray wolf during the search process;

[0012] Combined with the fitness value of each gray wolf and the distance between each gray wolf and the current optimal solution, the score of each gray wolf is calculated, and the solution candidate with the highest current score is selected as the optimal solution in the current population and used for subsequent search operations until the maximum number of iterations is reached to obtain the global optimal solution, which is the final optimal feature selection solution.

[0013] Preferably, the initialization of the gray wolf optimization algorithm includes: initializing the population size, the maximum number of iterations, and the position of each gray wolf individual.

[0014] Preferably, the fitness value of each gray wolf is calculated according to the classification result of the current candidate solution, and the specific expression is:

[0015] ;

[0016] in, is the fitness value; Indicates the total number of classifications of the data of the current candidate solution; It represents the classification error rate when the data of the current candidate solution is classified for the i-th time, which is obtained by calculating the ratio of the number of samples with classification errors to the total number of samples.

[0017] Preferably, the improved grey wolf step length update calculation formula is:

[0018] ; ; ;

[0019] ;

[0020] in, , , Respectively represent the distance between the gray wolf and the current three best wolves; , , Respectively represent the current population Wolf, β Wolf and δ Coefficient vector of wolves; , , Respectively represent the current population Wolf, β Wolf and δ The wolf's position vector; The position vector representing the improved path chosen by the gray wolf; represents the position vector of the gray wolf in the current iteration; t represents the number of current iterations; Represents the position vector of the gray wolf individual with the highest score at present; represents the remainder function, Indicates that the current iteration number is an even number.

[0021] Preferably, the mutation intensity in the search process is adaptively adjusted, and the position of each gray wolf is updated and calculated specifically as follows: by calculating the neighborhood search position of the current best three wolves after adding the mutation intensity, the new position of the gray wolf in the next iteration is obtained and updated, and the formula is:

[0022]

[0023] ;

[0024] ;

[0025] ; ; ;

[0026] ; ; ;

[0027] ;

[0028] in, , , Respectively Wolf, β Wolf and δ The wolf's neighborhood search location vector; represents the improved coefficient vector; represents the nonlinear parameter; t represents the current number of iterations; T represents the maximum number of iterations; , , Respectively represent the Wolf, β Wolf and δ The wolf's neighborhood search location vector; , , Represents a random number from 0 to 1; represents the best solution among the current optimal solutions; represents a random number from a standard normal distribution; Represents the new position vector of the gray wolf in the next iteration.

[0029] Preferably, the score of each gray wolf is calculated by combining the fitness value of each gray wolf and the distance between each gray wolf and the current optimal solution, and the score is calculated by the following formula:

[0030] ;

[0031] in, represents the score of the Gray Wolves; Represents the normalized fitness value; Represents the normalized value of the distance between the gray wolf and the current optimal solution.

[0032] Preferably, the distance DP between the gray wolf and the current optimal solution is calculated using the Euclidean distance, and the formula is:

[0033] ;

[0034] in, represents the location coordinates of the gray wolf, that is, the location coordinates of the candidate; represents the position coordinates of the optimal solution in the current optimal solution; n represents the number of populations, that is, the population size.

[0035] Preferably, it also includes: using a KNN classifier to classify the current candidate solution to obtain a classification result; and using the standard deviation To measure the stability and robustness of the algorithm; where the standard deviation The expression is:

[0036] ;

[0037] in, represents the optimal solution obtained after running H classifications, It represents the average value of classification error rate / total number of samples after running H classifications.

[0038] The present invention also provides a gray wolf optimization feature selection device based on an adaptive mechanism, which is applied to high-dimensional complex data classification tasks, including:

[0039] An acquisition unit, used for acquiring a data set to be feature selected; wherein the types of the data set include text data, image data, voice data and electrical signal data;

[0040] Initialization unit, used to initialize the gray wolf optimization algorithm;

[0041] A fitness value calculation unit is used to calculate the fitness value of each gray wolf according to the data set and the gray wolf optimization algorithm, with the goal of obtaining the current optimal solution, so as to evaluate the quality of each gray wolf, that is, to evaluate the degree of proximity between each candidate solution and the optimal solution; wherein each gray wolf represents one candidate solution in the population;

[0042] The current optimal three wolf units are used to find the current optimal solution according to the fitness value. Wolf, suboptimal solution β Wolf and the Third Optimal Solution δ Wolves, that is, the current three best wolves, thus obtaining the current optimal solution;

[0043] The iterative update unit is used to use the position information of the current best three wolves as a search guide. When each gray wolf is looking for prey, it combines the nonlinear parameter control strategy and adopts the improved gray wolf step length update calculation method to iteratively update and calculate the relevant parameters of each gray wolf, and adaptively adjust the mutation intensity during the search process, update and calculate the position and fitness value of each gray wolf, and continuously converge the distance between the prey and the gray wolf during the search process;

[0044] The global optimal solution unit is used to combine the fitness value of each gray wolf and the distance between each gray wolf and the current optimal solution, calculate the score of each gray wolf, select the solution candidate with the highest current score as the optimal solution in the current population, and use it for subsequent search operations until the maximum number of iterations is reached to obtain the global optimal solution, which is the final optimal feature selection solution.

[0045] The present invention also provides a gray wolf optimization feature selection device based on an adaptive mechanism, comprising a processor and a memory, wherein a computer program is stored in the memory, and the computer program can be executed by the processor to implement a gray wolf optimization feature selection method based on an adaptive mechanism as described above.

[0046] The present invention also provides a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor of a device where the computer-readable storage medium is located, the above-mentioned gray wolf optimization feature selection method based on an adaptive mechanism is implemented.

[0047] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0048] The present invention adopts a gray wolf optimization algorithm based on an adaptive mechanism and applies it to the process of machine learning. It is used to process feature selection in high-dimensional and complex classifications and performs well in terms of convergence speed and accuracy. First, a novel nonlinear parameter control strategy is introduced to effectively balance the relationship between the algorithm's neighborhood exploration and the use of the optimal solution, preventing the algorithm from falling into a local optimal solution. Secondly, an adaptive distance balance mechanism is proposed to effectively prevent the premature convergence of the GWO gray wolf optimization algorithm during the search process, and enhance the search efficiency by selecting high-potential solutions. Finally, an adaptive neighborhood variation mechanism is designed, and a variation factor is introduced to fully consider the The information exchange between the wolf, β wolf, δ wolf and the current global optimal solution is intended to adaptively adjust the mutation intensity during the search process, so that the algorithm of the present invention can more effectively find the global optimal solution and select the best feature set in the data set. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 This is a flow chart of an original grey wolf optimization algorithm provided in Example 1.

[0051] Figure 2 This is a diagram of the gray wolf hierarchy structure provided in Example 1.

[0052] Figure 3 This is a structural diagram of the gray wolf optimization feature selection method based on the adaptive mechanism provided in Example 1.

[0053] Figure 4 This is a comparison curve diagram of the linear and nonlinear convergence factors provided in Example 1.

[0054] Figure 5This is a comparison chart of the convergence curves of the method of the present invention and other algorithms provided in Example 1 on 15 data sets.

[0055] Figure 6 This is a comparison chart of the average running time of the method of the present invention provided in Example 1 and other algorithms on 15 data sets.

[0056] Figure 7 A schematic diagram of a gray wolf optimization feature selection device based on an adaptive mechanism provided in Example 2.

[0057] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0059] Embodiment 1

[0060] Embodiment 1 of the present invention provides a gray wolf optimization feature selection method based on an adaptive mechanism, which can be implemented by a gray wolf optimization feature selection device based on an adaptive mechanism (hereinafter referred to as a feature selection device), and in particular, executed by one or more processors in the feature selection device.

[0061] In this embodiment, the feature selection device may be an electronic device equipped with a processor, the processor having a computer program of the gray wolf optimization feature selection method based on the adaptive mechanism and the computer program can be executed, such as a computer, a smart phone, a smart tablet, a workstation, etc., which is not limited here.

[0062] In this embodiment, feature selection (FS) is a key step in data preprocessing, which aims to reduce the dimension and redundancy of data by extracting features that are meaningful to model prediction, thereby improving the performance and interpretability of the model. In the field of machine learning and data mining, feature selection FS has been widely used in various practical problems, such as extracting fetal electrocardiogram signals, detecting gender from voice data, biological data analysis, and intelligent facial emotion recognition. Effective FS can significantly improve the learning efficiency and prediction accuracy of the model.

[0063] The FS problem is an NP-hard problem. As the dimension increases, the search space grows exponentially, making exhaustive search impractical. In order to improve the search efficiency of the FS algorithm, many scholars have proposed various methods. FS methods are mainly divided into three categories: filtering methods, packaging methods, and embedding methods. Among them, packaging methods are widely used due to their superior classification ability. Packaging methods mainly consist of three parts: classifier, feature subset evaluation, and search technology. Among them, effective search technology is crucial to the performance of FS algorithms, such as metaheuristic methods (MH), particle swarm optimization (PSO), differential evolution (DE), genetic algorithm (GA), artificial bee colony algorithm (ABC), Harris Hawk optimization (HHO), whale optimization algorithm (WOA), moth fire optimization (MFO), snake optimization algorithm (SO), etc., which have been widely used in the FS field.

[0064] The present invention adopts the meta-heuristic method MH for search, which has simplicity, flexibility, lack of derivation mechanism and the ability to avoid local optimal solutions. The search process of the MH algorithm includes two stages: the first stage is exploration and the second stage is exploitation. In the exploration stage, the algorithm comprehensively explores the search space to find diversified solutions to the problem. In the exploitation stage, the algorithm uses local information to generate better solutions, usually in the vicinity of the current solution. Excessive exploration will reduce the convergence rate of the algorithm, while excessive exploitation increases the risk of falling into the local optimal solution. One of the main goals of the algorithm design of the present invention is to balance exploration and exploitation to achieve optimal performance.

[0065] In this embodiment, the Grey Wolf Optimizer (GWO) is a heuristic swarm intelligence algorithm proposed by Mirjalili et al., which is inspired by the social hierarchy and hunting behavior of grey wolves.

[0066] like Figure 1 As shown in the GWO algorithm flow chart, the main steps of the Grey Wolf Optimization Algorithm include:

[0067] Population initialization: set the number of gray wolf individuals (population size), variable dimension and maximum number of iterations;

[0068] Random initialization of gray wolf positions: Randomly initialize the positions of gray wolf individuals in the search space.

[0069] Fitness value calculation: Calculate the fitness value of each individual gray wolf to evaluate its quality.

[0070] Position update: Update the position of individual gray wolves according to their fitness values ​​to simulate the hunting behavior of gray wolf groups.

[0071] Parameter update: Update relevant parameters in the algorithm, such as the number of iterations.

[0072] Iteratively calculate fitness value and update optimal position: recalculate the fitness value of the updated gray wolf individual and update the optimal position until the maximum number of iterations is reached.

[0073] like Figure 2 As shown, there is a strict hierarchy in the gray wolf group, including , β, δ and ω. The wolf is the leader of the pack and has a decisive influence on key areas such as hunting and habitat selection. The beta wolf plays a secondary role, with Wolves work closely together and play a key supporting role in decision making, planning group movements and hunting strategies. Delta wolves follow closely behind, guarding the boundaries of the territory, alerting the group when threatened, and following and the leadership of Beta Wolf. Under the leadership of , β and δ wolves, although ω wolf seems to have a lower status, it plays an indispensable role in maintaining the internal balance of the group. When constructing the GWO mathematical model, the optimal solution of the problem is analogous to wolf, the second best solution corresponds to β wolf, the third best solution corresponds to δ wolf, and other candidate solutions are considered as ω wolf. Then, use The position information of , β and δ wolves is used as a search guide. By simulating the cooperation mechanism between them, the entire search process is led to the optimal solution, thereby effectively solving complex problems.

[0074] By simulating the social hierarchy and hunting strategy of gray wolves, GWO can achieve global search and local exploitation. Specifically, the behavior of gray wolves in searching for prey is abstractly represented as:

[0075] ;

[0076] ;

[0077] ;

[0078] ;

[0079] Where D represents the distance between the prey and the wolf during the search process; C and A represent the coefficient vectors respectively; The vector representing the current position of the prey; Represents the position vector of the gray wolf in the current iteration; represents the new position vector of the gray wolf in the next iteration; a represents the linear coefficient that decreases with the number of iterations, and its value decreases from 2 to 0; , Represents a random number between 0 and 1.

[0080] In the original GWO algorithm, the mathematical model of the entire hunting behavior of the gray wolf is as follows:

[0081] ;

[0082] ;

[0083] ;

[0084] in, , , Respectively represent the distance between the gray wolf and the current three best wolves; , , Respectively represent the current population Wolf, β Wolf and δ Coefficient vector of wolves; , , Respectively represent the current population Wolf, β Wolf and δ The wolf's position vector; Represents the position vector of the path chosen by the gray wolf in the current iteration; represents the position vector of the gray wolf in the current iteration; t represents the number of current iterations; , , Respectively Wolf, β Wolf and δ The wolf's neighborhood search position vector; Represents the new position vector of the gray wolf in the next iteration; Represents the coefficient vector when the current iteration number is t.

[0085] When parameter A is greater than 1, the wolf will move away from the prey, which helps the algorithm to explore the search space more extensively. On the other hand, if A is less than 1, the wolf will tightly surround the prey, which promotes the algorithm to converge quickly to the global optimal solution. Obviously, the dynamic adjustment of the wolf's position is determined by wolf, β wolf and δ wolf guidance.

[0086] However, when dealing with high-dimensional and complex optimization problems, the original GWO algorithm is prone to falling into local optimal solutions and shows certain limitations in terms of convergence speed and accuracy. To address the deficiencies of GWO in feature selection, the present invention proposes a new variant of GWO, namely a grey wolf optimization algorithm based on an adaptive mechanism.

[0087] Such as Figure 3 shown, a grey wolf optimization feature selection method based on an adaptive mechanism, which is applied to the classification task of high-dimensional complex data in the machine learning process, includes steps S1 to S6.

[0088] S1. Obtain a data set to be feature-selected; wherein, the types of the data set include text data, image data, voice data, and electrical signal data.

[0089] In this embodiment, given a data set composed of D features and n samples, this data set can be electrocardiogram signal data, voice data, biological data (such as fingerprint recognition data, face data), human organ movement data, power signal data, etc. The types of the data set include text data, image data, voice data, electrical signal data, and biological signal data, etc.

[0090] The feature selection problem includes selecting d best features (d < D) from all features. For example, from colon cancer CT image data, select the optimal feature subset of colon cancer; from electrocardiogram signal data, select the optimal feature subset of fetal electrocardiogram signals to accurately classify high-dimensional electrocardiogram signal data; extract gender voice features from voice data; identify facial features of specific emotions from face data; identify features during normal movement of human organs from human organ movement data, and so on.

[0091] For the convenience of processing, the solution X to the feature selection problem is usually encoded as a binary string, where each bit represents whether a certain feature is selected for model construction. For example:

[0092] ;

[0093] = 1 indicates that the i-th feature is selected and included in the feature set X, otherwise, it indicates that it is not selected.

[0094] In the feature selection problem, it usually refers to the classification accuracy or error rate. If Represents the classification error rate, then the feature selection problem can be expressed by a mathematical formula:

[0095] ;

[0096] in, Indicates subject to constraints.

[0097] S2, initialize the gray wolf optimization algorithm.

[0098] Furthermore, the initialization gray wolf optimization algorithm includes: initializing the population size, the maximum number of iterations, and the position of each gray wolf individual.

[0099] S3, according to the data set and the gray wolf optimization algorithm, with the goal of obtaining the current optimal solution, calculate the fitness value of each gray wolf to evaluate the pros and cons of each gray wolf, that is, evaluate the degree of closeness between each candidate solution and the optimal solution; wherein each gray wolf represents one candidate solution in the population.

[0100] In this embodiment, a wrapper technique is used to run each feature selection algorithm. The core advantage of the wrapper method is that it directly uses the performance of the classifier as the criterion for selecting features, making the feature selection process both efficient and accurate. In the wrapper feature selection method, the fitness function directly reflects the classification performance.

[0101] Furthermore, the fitness value of each gray wolf is calculated according to the classification result of the current candidate solution, and the expression of the fitness function is:

[0102] ;

[0103] in, is the fitness value; Indicates the total number of classifications of the data of the current candidate solution; It represents the classification error rate when the data of the current candidate solution is classified for the i-th time, which is obtained by calculating the ratio of the number of samples with classification errors to the total number of samples.

[0104] S4, according to the fitness value, obtain the current optimal solution Wolf, suboptimal solution β Wolf and the Third Optimal Solution δ , that is, the current optimal three wolves, thus obtaining the current optimal solution.

[0105] In this embodiment, the optimal solution to the problem is analogized as wolf, the second best solution corresponds to β wolf, the third best solution corresponds to δ wolf, and other candidate solutions are considered as ω wolf. Then, use The position information of , β and δ wolves is used as a search guide. By simulating the cooperation mechanism between them, the entire search process is led to the optimal solution, thereby effectively solving complex problems.

[0106] S5, taking the position information of the current best three wolves as the search guide, each gray wolf, when searching for prey, combines the nonlinear parameter control strategy, adopts the improved gray wolf step length update calculation method, iteratively updates and calculates the relevant parameters of each gray wolf, and adaptively adjusts the mutation intensity during the search process, updates and calculates the position and fitness value of each gray wolf, and continuously converges the distance between the prey and the gray wolf during the search process.

[0107] S6, combining the fitness value of each gray wolf and the distance between each gray wolf and the current optimal solution, calculate the score of each gray wolf, select the solution candidate with the highest current score as the optimal solution in the current population, and use it for subsequent search operations until the maximum number of iterations is reached and the global optimal solution is obtained, which is the final optimal feature selection solution.

[0108] Specifically, the gray wolf optimization algorithm with adaptive mechanism proposed in the present invention mainly includes three improvements:

[0109] First, the adaptive parameter control mechanism (APCGWO). The present invention improves the existing linear parameter a and adopts nonlinear parameter The search is controlled to explore the entire search space efficiently. In the original Grey Wolf Optimization (GWO) algorithm, the convergence speed is slow or the convergence accuracy is low, partly because the convergence factor of GWO decreases linearly. Figure 4 As shown, the convergence factor of the present invention is It decreases nonlinearly from 2 to 0. In the early stage of iteration, the curve is flat, which shows that the algorithm of the present invention can effectively explore the entire search space. In the later iterations, the curve drops rapidly, indicating that the algorithm converges quickly.

[0110] Nonlinear parameters The expression is as follows:

[0111] ;

[0112] in, represents the nonlinear parameter; t represents the current number of iterations; T represents the maximum number of iterations.

[0113] Second, Adaptive Neighborhood Mutation Mechanism (ADVGWO). In order to further improve the global search capability of the Grey Wolf Optimization Algorithm (GWO) and avoid falling into the local optimal solution in the later iterations, the present invention designs an adaptive neighborhood mutation mechanism, which fully considers Information exchange between the wolf, β wolf, δ wolf and the current global optimal solution. This mechanism aims to adaptively adjust the mutation intensity during the search process, allowing the algorithm to explore the solution space more effectively and escape from the local optimal solution.

[0114] Specifically: First, define a variation intensity factor , to quantify the difference between different levels of wolves and the global optimal solution. Then, the calculated mutation intensity is combined with the random number of the standard normal distribution , the neighborhood of each level of wolves is mutated. This step aims to dynamically adjust the search direction and step size of the wolves according to their relative position to the global optimal solution, thus achieving adaptive domain change. This mechanism not only enhances the algorithm's exploration ability, but also retains the efficiency of using known information, providing a new method for solving complex optimization problems.

[0115] Third, Adaptive Fitness Distance Balancing Mechanism (AFDBGWO).

[0116] The core of the Fitness Distance Balancing (FDB) method is to balance the fitness of solution candidates in the search space with their distance from the current optimal solution. This method aims to avoid the early convergence problem of metaheuristic search algorithms and improve search efficiency by selecting solution candidates with high potential. Fitness usually indicates the quality of the solution; for minimization problems, solutions with higher fitness have lower objective function values. The distance metric (DP) represents the distance between the candidate and the optimal solution in the current population.

[0117] By calculating an FDB score for each solution candidate, namely , which combines fitness and distance factors. Based on the calculated score, the solution candidate with the highest score is selected for the selection process of subsequent search operations. This process helps the algorithm maintain a certain degree of exploration (through the distance value) while utilizing known good solutions (through the fitness value). The FDB score calculation method effectively improves the performance of the metaheuristic algorithm by comprehensively considering the fitness and diversity of the solution candidates, especially in complex problems with multiple local optimal solutions.

[0118] The GWO algorithm is redesigned by combining three adaptive mechanism strategies. According to the number of iterations, the update step size of the gray wolf is selected by adaptive or FDB score-based individual selection (i.e. ), and update each gray wolf, which can not only maintain the diversity of the population, but also speed up the convergence speed and accuracy of the algorithm.

[0119] Specifically, the improved grey wolf step length update calculation formula is:

[0120] ; ; ;

[0121] ;

[0122] in, , , Respectively represent the distance between the gray wolf and the current three best wolves; , , Respectively represent the current population Wolf, β Wolf and δ Coefficient vector of wolves; , , Respectively represent the current population Wolf, β Wolf and δ The wolf's position vector; The position vector representing the improved path chosen by the gray wolf; represents the position vector of the gray wolf in the current iteration; t represents the number of current iterations; Represents the position vector of the gray wolf with the highest score at present, by calculating the score of each gray wolf get; represents the remainder function, Indicates that the current iteration number is an even number.

[0123] Specifically, the adaptive neighborhood mutation mechanism, that is, adaptively adjusting the mutation intensity during the search process, updates and calculates the position of each gray wolf as follows:

[0124] By calculating the neighborhood search positions of the current best three wolves after adding the mutation intensity, the new position of the gray wolf in the next iteration is obtained and updated. The formula is:

[0125]

[0126] ;

[0127] ;

[0128] ; ; ;

[0129] ; ; ;

[0130] ;

[0131] in, , , Respectively Wolf, β Wolf and δ The wolf's neighborhood search position vector; represents the improved coefficient vector; represents the nonlinear parameter; t represents the current number of iterations; T represents the maximum number of iterations; , , Respectively represent the Wolf, β Wolf and δ The wolf's neighborhood search position vector; , , Represents a random number from 0 to 1; represents the best solution among the current optimal solutions; represents a random number from a standard normal distribution; Represents the new position vector of the gray wolf in the next iteration.

[0132] Specifically, the FDB score of each wolf is calculated by combining the fitness value of each wolf and the distance between each wolf and the current optimal solution. , the FDB score Calculated by the following formula:

[0133] ;

[0134] in, Represents the normalized fitness value; Represents the normalized value of the distance between the gray wolf and the current optimal solution.

[0135] The distance DP between the gray wolf and the current optimal solution is calculated using the Euclidean distance in this embodiment, and the formula is:

[0136] ;

[0137] in, represents the location coordinates of the gray wolf, that is, the location coordinates of the candidate; represents the location coordinates of the optimal solution in the current optimal solution; n represents the number of populations.

[0138] Of course, other distance metrics may be used according to actual conditions, such as Manhattan distance, Chebyshev distance, cosine similarity, etc., which are not limited here.

[0139] According to the calculated Score, select the solution candidate with the highest score , used for the selection process in subsequent search operations.

[0140] In another preferred embodiment, the method further includes: using a KNN classifier to classify the current candidate solution to obtain a classification result for subsequent test evaluation. Of course, other classifiers may also be used for classification according to actual conditions, such as logistic regression, random forest, support vector machine SVM, decision tree, etc.

[0141] In another preferred embodiment, we use the O method as an evaluation tool to measure the time complexity of the algorithm, and conduct a comparative analysis between the algorithm of the present invention (i.e., the grey wolf optimization algorithm combining the adaptive parameter control mechanism, the adaptive neighborhood mutation mechanism and the adaptive fitness distance balance mechanism, referred to as the AMGWO algorithm) and the original GWO algorithm.

[0142] The specific application steps of the O-order method are briefly described as follows:

[0143] Simplify constant terms: First, to facilitate subsequent analysis, all additive constants in the algorithm running time are simplified to a constant of 1.

[0144] Keep the highest order terms: In the adjusted runtime function, only the highest order terms that have the greatest impact on the complexity are kept, and other lower order terms are ignored.

[0145] Removing non-essential constants: If there is a highest order term and its coefficient is not 1, then further remove the coefficient and only keep its order as the O-order representation.

[0146] According to the above steps, the time complexity analysis result of the original GWO algorithm is O(n×d×T), where n, d, and T represent the population size (i.e., the number of populations), dimension, and number of iterations, respectively. It is worth noting that in the calculation process of the algorithm AMGWO of the present invention, no new loop structure is introduced, and no fundamental change is made to the original loop order. Therefore, its time complexity analysis result is consistent with GWO, which is also O(n×d×T).

[0147] The traditional Grey Wolf Optimizer (GWO) performs well in solving problems in continuous search spaces. However, challenges arise when dealing with essentially binary optimization problems such as feature selection, where the solutions are strictly limited to binary values ​​{0,1}. To overcome this problem, researchers introduced the concept of a transformation function with the goal of adapting the original capabilities of the Grey Wolf algorithm in continuous search spaces to the binary domain. Among the many transformation functions, the Sigmoid function is the best choice due to its unique properties. This function achieves the transformation by mapping continuous input values ​​to the (0, 1) range and converting them to binary values ​​using a threshold (usually 0.5 is used as the threshold). It is worth mentioning that this approach has been verified in multiple GWO variants, confirming its effectiveness. In this study, the Sigmoid function is also used as a transformation tool. The transformation function is defined as follows:

[0148] ;

[0149] in, Represents the continuous value of the i-th search space in dimension j; is a random number between 0 and 1.

[0150] is considered as a specific example of the Sigmoid function, The mathematical expression is as follows:

[0151] ;

[0152] in, Represents continuous values ​​of input.

[0153] In another preferred embodiment, in order to verify the performance of the proposed AMGWO algorithm, 15 benchmark datasets are selected for experiments. Table 1 summarizes the basic parameter information of these datasets, such as the number of features, the number of samples, and the categories. These datasets are characterized by high dimensionality and small sample size, which is common in feature selection literature. These datasets are rich and diverse, including microarray gene expression data, image (face) detection data, email text data, etc., and have been preprocessed by their respective providers.

[0154] Table 1. Experimental results of the benchmark dataset using the feature selection algorithm of the present invention

[0155]

[0156] The above datasets are all from existing public databases, such as the Yale dataset is a face recognition dataset established by Yale University; the ORL dataset is a face recognition dataset created by the ORL Laboratory of the University of Cambridge; the Colon dataset is a colon tumor CT image set; Prostate is a prostate cancer dataset; Colon is a colon cancer dataset and ALLAML is an acute myeloid leukemia dataset.

[0157] To ensure the reliability of the feature selection results, we conducted 10 independent runs on the dataset to test the feature selection methods, focusing on classification accuracy and feature subset dimensionality. It is worth noting that in our experiments, each feature selection method generates a separate feature subset for each dataset in each run, i.e., the current optimal solution. In order to verify the effectiveness of our proposed method, a series of quantitative indicators are used for evaluation.

[0158] Best: The minimum value of classification error rate / feature subset size among all solutions obtained after the algorithm is run 10 times.

[0159] Worst: The maximum value of the classification error rate / feature subset size among all solutions obtained after the algorithm is run 10 times.

[0160] Mean: The average of the classification error rate / feature subset size of all solutions obtained after the algorithm is run 10 times.

[0161] Standard Deviation (Std): The standard deviation is calculated based on the set of all solutions obtained after running the algorithm 10 times. It is an important indicator to measure the stability and robustness of the optimization algorithm. Its mathematical expression is as follows:

[0162] ;

[0163] in, represents the optimal solution obtained after running H classifications, Represents the average of the classification error rate / feature subset size (i.e., total number of samples) after running H classifications. Typically, this involves recording the results of each run in multiple independent runs and selecting the best one from these results as the final solution. This approach helps evaluate the performance stability and reliability of the algorithm in different iterations.

[0164] In order to verify the performance of the proposed AMGWO algorithm, we compared AMGWO with other optimization algorithms including the original GWO: GWO (original grey wolf optimization algorithm), GNHGWO, BABCGWO, SOGWO, EGWO and AGWO. Table 2 lists the parameter settings of these optimization algorithms in detail. The maximum number of iterations and population size were set to 100 and 30, respectively. For a fair comparison, each algorithm was run 30 times independently. Then the standard deviation (Std), best (Best), worst (Worst) and mean (Mean) were calculated and recorded, and the best result was displayed in bold.

[0165] Table 2. Parameter settings of the optimization algorithm

[0166]

[0167] Among them, GNHGWO: Greedy Non-Hierarchical Grey Wolf Optimizer, is an improved algorithm of the traditional grey wolf optimization algorithm. It introduces a greedy strategy in the search process, ignores the social hierarchy of grey wolves, and adopts a non-hierarchical update mechanism to update the position of grey wolves.

[0168] BABCGWO: Binary Artificial Bee Colony with Grey Wolf Optimizer, which combines the advantages of Binary Artificial Bee Colony (BABC) and Grey Wolf Optimizer (GWO), is often used for feature selection of high-dimensional data.

[0169] SOGWO: Selective Opposition based Grey Wolf Optimization, a grey wolf optimization algorithm based on selective oppositional learning. The algorithm aims to enhance exploration behavior and improve convergence efficiency by implementing reverse learning on specific dimensions. SOGWO uses the Spearman correlation coefficient to identify the ω wolf with the lowest social status and uses it as the object of oppositional learning, thereby reducing invalid exploration and accelerating algorithm convergence. This strategy effectively maintains the probability of finding the optimal solution during the optimization process and optimizes the overall performance of the algorithm.

[0170] EGWO: Enhanced Grey Wolf Optimizer, enhanced grey wolf optimization algorithm, enhances the performance of GWO algorithm by introducing multiple improvement strategies.

[0171] AGWO: Adaptive Grey Wolf Optimization Algorithm, which dynamically adjusts the search intensity and scope to better approach the optimal solution of the problem.

[0172] AMGWO: The algorithm of the present invention is a grey wolf optimization algorithm that combines an adaptive parameter control mechanism, an adaptive neighborhood mutation mechanism, and an adaptive fitness distance balancing mechanism.

[0173] None: means none.

[0174] The proposed AMGWO algorithm was compared with six other competing algorithms in terms of classification error rate, feature subset size and running time.

[0175] Table 3 shows the results of AMGWO and six other competing algorithms in terms of classification error rate. The best results are shown in bold. In terms of classification error rate, AMGWO achieves better results than other competing algorithms on most datasets. The standard deviation of the classification error rate is smaller, indicating that the algorithm is more stable. The complexity of the problem is positively correlated with the dimensionality. As the dimension of the problem increases, the error rate of AMGWO is statistically the smallest, which shows that the classification performance of AMGWO is least affected by the increase in dimensionality.

[0176] Table 3. Results of AMGWO and other six competing algorithms in terms of classification error rate

[0177]

[0178] In addition, in another preferred embodiment, AMGWO and six other competing algorithms are experimentally tested in terms of feature subset size. The experimental test results show that in terms of feature subset size, AMGWO achieves the smallest feature subset on 11 data sets. On 13 data sets, the average size of the feature subset of AMGWO is optimal. For most data sets of different dimensions, the standard deviation corresponding to AMGWO is smaller.

[0179] like Figure 5 As shown, the convergence curves of seven algorithms on 15 datasets are shown. Each curve represents the average results of 10 runs in each iteration. These curves show that AMGWO has faster convergence speed on most datasets and produces higher quality solutions compared with other competing algorithms. Although the quality of solutions obtained by AMGWO on the Prostate (prostate cancer), Colon (colon cancer), and ALLAML (acute myeloid leukemia) datasets is not as good as that of some competing algorithms, the convergence speed of these algorithms on these datasets is not as fast as AMGWO. However, on most datasets, AMGWO performs better. Overall, AMGWO outperforms the other six competing algorithms in terms of convergence speed and solution quality.

[0180] like Figure 6 As shown in Figure 2, the running time of the AMGWO algorithm is lower than that of other competitors on most datasets, which benefits from the fact that each feature subset is evaluated only once in each iteration in the AMGWO algorithm. Therefore, AMGWO outperforms other algorithms in terms of computation time.

[0181] This efficiency improvement is particularly important for processing large-scale or high-dimensional data sets, because it can significantly reduce the overall computing time, allowing the algorithm to provide results to users or systems faster. The AMGWO algorithm optimizes the calculation process by reducing unnecessary repeated evaluations in each round of iteration, thereby improving the operating efficiency while maintaining the performance of the algorithm. This optimization not only saves time, but also may reduce resource consumption, making the AMGWO algorithm more attractive in practical applications.

[0182] In another preferred embodiment, in order to verify the strategies we proposed, namely the adaptive parameter control mechanism (APCGWO), the adaptive fitness distance balance mechanism (AFDBGWO), and the adaptive neighborhood mutation mechanism (ADVGWO), the original GWO, and the AMGWO that integrates these three strategies, are tested on accuracy analysis and feature quantity analysis. The experimental results are shown in Tables 4 and 5, respectively. In Table 4, it can be seen that ADVGWO ranks second in the accuracy analysis results, with more bold numbers, while AMGWO that integrates the three strategies ranks first. It can also be seen from the last row of the table that AMGWO has the smallest average Friedman ranking, indicating that it performs best. In Table 5, ADVGWO ranks second in the feature quantity analysis results, with more bold numbers, while AMGWO that integrates the three strategies still ranks first. In the last row of the table, the average Friedman ranking of AMGWO is 2.2, which is lower than other competing algorithms, so it ranks first.

[0183] In summary, although a single improvement strategy can achieve good results on some datasets, the comprehensive performance of AMGWO that combines the three strategies is better.

[0184] Table 4. Analysis results of accuracy of a single improved strategy and AMGWO that combines three strategies

[0185]

[0186] Table 5. Analysis results of the number of features of a single improved strategy and AMGWO that combines three strategies

[0187]

[0188] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0189] This paper proposes an effective feature selection method for the Grey Wolf Optimization Algorithm AMGWO based on an adaptive mechanism for classification tasks, which is applied to industries such as biomedicine, engineering optimization problems, neural networks, scheduling, power engineering problems, control engineering problems and robotics. The comparative experimental results with the original GWO and five other advanced variants on 15 high-dimensional data sets show that the proposed AMGWO algorithm has advantages in accuracy, convergence speed and feature subset size. This is mainly attributed to the following three aspects: (1) A nonlinear parameter control strategy is introduced to effectively balance exploration and utilization; (2) An adaptive degree distance balance mechanism is proposed to avoid premature convergence in the search process and select high-potential solutions to improve search efficiency; (3) An adaptive neighborhood mutation mechanism is designed to fully consider Wolf, β Wolf and δ The information exchange between the wolf and the current global optimal solution enables the algorithm to find the global optimal solution more efficiently.

[0190] Embodiment 2

[0191] like Figure 7 As shown, the second embodiment of the present invention further provides a gray wolf optimization feature selection device based on an adaptive mechanism, comprising:

[0192] An acquisition unit, used for acquiring a data set to be feature selected; wherein the types of the data set include text data, image data, voice data and electrical signal data;

[0193] Initialization unit, used to initialize the gray wolf optimization algorithm;

[0194] A fitness value calculation unit is used to calculate the fitness value of each gray wolf according to the data set and the gray wolf optimization algorithm, with the goal of obtaining the current optimal solution, so as to evaluate the quality of each gray wolf, that is, to evaluate the degree of proximity between each candidate solution and the optimal solution; wherein each gray wolf represents one candidate solution in the population;

[0195] The current optimal three wolf units are used to obtain the current optimal solution according to the fitness value. Wolf, the second-best solution β wolf and the third-best solution δ wolf, that is, the current best three wolves, the current best solution;

[0196] The iterative update unit is used to use the position information of the current best three wolves as a search guide. When each gray wolf is looking for prey, it combines the nonlinear parameter control strategy and adopts the improved gray wolf step length update calculation method to iteratively update and calculate the relevant parameters of each gray wolf, and adaptively adjust the mutation intensity during the search process, update and calculate the position and fitness value of each gray wolf, continuously converge the distance between the prey and the gray wolf during the search process, and calculate the current optimal solution in each iteration process;

[0197] The global optimal solution unit is used to combine the fitness value of each gray wolf and the distance between each gray wolf and the current optimal solution, calculate the score of each gray wolf, select the solution candidate with the highest current score as the optimal solution in the current population, and use it for subsequent search operations until the maximum number of iterations is reached to obtain the global optimal solution, which is the final optimal feature selection solution.

[0198] Embodiment 3

[0199] The third embodiment of the present invention also provides a gray wolf optimization feature selection device based on an adaptive mechanism, which includes a memory and a processor, wherein a computer program is stored in the memory, and the computer program can be executed by the processor to implement the gray wolf optimization feature selection method based on an adaptive mechanism as described above.

[0200] Embodiment 4

[0201] The fourth embodiment of the present invention further provides a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor of a device where the computer-readable storage medium is located, the above-mentioned gray wolf optimization feature selection method based on the adaptive mechanism is implemented.

[0202] In several embodiments provided in the embodiments of the present invention, it should be understood that the disclosed devices and methods may also be implemented in other ways.

[0203] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A gray wolf optimization feature selection method based on an adaptive mechanism is applied to the classification task of high-dimensional complex data in the process of machine learning, characterized in that: include: Acquire a data set to be feature selected; wherein the types of the data set include text data, image data, voice data and electrical signal data; Initialize the Grey Wolf Optimization Algorithm; According to the data set and the gray wolf optimization algorithm, with the goal of obtaining the current optimal solution, the fitness value of each gray wolf is calculated to evaluate the pros and cons of each gray wolf, that is, to evaluate the degree of proximity between each candidate solution and the optimal solution; wherein each gray wolf represents one candidate solution in the population; According to the fitness value, the current optimal solution is obtained Wolf, suboptimal solution β Wolf and the Third Optimal Solution δ Wolves, that is, the current three best wolves, thus obtaining the current optimal solution; Taking the position information of the current best three wolves as the search guide, each gray wolf, when searching for prey, combines the nonlinear parameter control strategy, adopts the improved gray wolf step length update calculation method, iteratively updates and calculates the relevant parameters of each gray wolf, and adaptively adjusts the mutation intensity during the search process, updates and calculates the position and fitness value of each gray wolf, and continuously converges the distance between the prey and the gray wolf during the search process; wherein, adaptively adjusting the mutation intensity during the search process is specifically as follows: first, defining a mutation intensity factor to quantify the difference between wolves of different levels and the global optimal solution; then using the calculated mutation intensity factor combined with the random number of the standard normal distribution to perform mutation operations on the neighborhood of each level of wolf; The improved grey wolf step length update calculation formula is: ； ； ； ； in, , , Respectively represent the distance between the gray wolf and the current three best wolves; , , Respectively represent the current population Wolf, β Wolf and δ Coefficient vector of wolves; , , Respectively represent the current population Wolf, β Wolf and δ The wolf's position vector; The position vector representing the improved path chosen by the gray wolf; represents the position vector of the gray wolf in the current iteration; t represents the number of current iterations; Represents the position vector of the gray wolf individual with the highest score at present; represents the remainder function, Indicates that the current iteration number is an even number; Combine the fitness value of each gray wolf and the distance between each gray wolf and the current optimal solution to calculate the score of each gray wolf, select the solution candidate with the highest current score as the optimal solution in the current population, and use it for subsequent search operations until the maximum number of iterations is reached to obtain the global optimal solution, which is the final optimal feature selection solution; wherein the score is calculated by the following formula: ； in, represents the score of the Gray Wolves; Represents the normalized fitness value; Represents the normalized value of the distance between the gray wolf and the current optimal solution.

2. The method for selecting gray wolf optimization features based on an adaptive mechanism according to claim 1 is characterized in that ,The initialization gray wolf optimization algorithm includes: initializing the population size, the maximum number of iterations and the position of each gray wolf individual.

3. The gray wolf optimization feature selection method based on the adaptive mechanism according to claim 1 is characterized in that ,The fitness value of each gray wolf is calculated according to the classification result of the current candidate solution. The specific expression is: ； in, is the fitness value; Indicates the total number of classifications of the data of the current candidate solution; It represents the classification error rate when the data of the current candidate solution is classified for the i-th time, which is obtained by calculating the ratio of the number of samples with classification errors to the total number of samples.

4. The method for selecting gray wolf optimization features based on an adaptive mechanism according to claim 1 is characterized in that , adaptively adjust the mutation intensity during the search process, and update the position of each gray wolf. Specifically, by calculating the neighborhood search position of the current best three wolves after adding the mutation intensity, the new position of the gray wolf in the next iteration is obtained and updated. The formula is: ； ； ； ； ； ； ； ； ； ； in, , , Respectively Wolf, β Wolf and δ The wolf's neighborhood search location vector; represents the improved coefficient vector; represents the nonlinear parameter; t represents the current number of iterations; T represents the maximum number of iterations; , , Respectively represent the Wolf, β Wolf and δ The wolf's neighborhood search location vector; Represents a random number from 0 to 1; represents the best solution among the current optimal solutions; represents a random number from a standard normal distribution; Represents the new position vector of the gray wolf in the next iteration.

5. The method for selecting gray wolf optimization features based on an adaptive mechanism according to claim 1 is characterized in that ,The distance DP between the gray wolf and the current optimal solution is calculated using the Euclidean distance, and the formula is: ； in, represents the location coordinates of the gray wolf, that is, the location coordinates of the candidate; represents the position coordinates of the optimal solution in the current optimal solution; n represents the number of populations, that is, the population size.

6. The method for selecting gray wolf optimization features based on an adaptive mechanism according to claim 1 is characterized in that , also includes: using KNN classifier to classify the current candidate solution to obtain the classification result; and using standard deviation To measure the stability and robustness of the algorithm; where the standard deviation The expression is: ； in, represents the optimal solution obtained after running H classifications, It represents the average value of classification error rate / total number of samples after running H classifications.

7. A gray wolf optimization feature selection device based on an adaptive mechanism, characterized in that: include: An acquisition unit, used for acquiring a data set to be feature selected; wherein the types of the data set include text data, image data, voice data and electrical signal data; Initialization unit, used to initialize the gray wolf optimization algorithm; A fitness value calculation unit is used to calculate the fitness value of each gray wolf according to the data set and the gray wolf optimization algorithm, with the goal of obtaining the current optimal solution, so as to evaluate the quality of each gray wolf, that is, to evaluate the degree of proximity between each candidate solution and the optimal solution; wherein each gray wolf represents one candidate solution in the population; The current optimal three wolf units are used to find the current optimal solution according to the fitness value. Wolf, the second-best solution β wolf and the third-best solution δ wolf, that is, the current best three wolves, thus obtaining the current optimal solution; The iterative update unit is used to use the position information of the current best three wolves as a search guide. When each gray wolf is looking for prey, it combines the nonlinear parameter control strategy and adopts the improved gray wolf step length update calculation method to iteratively update and calculate the relevant parameters of each gray wolf, and adaptively adjust the mutation intensity during the search process, update and calculate the position and fitness value of each gray wolf, and continuously converge the distance between the prey and the gray wolf during the search process; wherein, the adaptive adjustment of the mutation intensity during the search process is specifically as follows: first, a mutation intensity factor is defined to quantify the difference between wolves of different levels and the global optimal solution; then, the calculated mutation intensity factor is combined with a random number of a standard normal distribution to perform a mutation operation on the neighborhood of each level of wolf; The improved grey wolf step length update calculation formula is: ； ； ； ； in, , , Respectively represent the distance between the gray wolf and the current three best wolves; , , Respectively represent the current population Wolf, β Wolf and δ Coefficient vector of wolves; , , Respectively represent the current population Wolf, β Wolf and δ The wolf's position vector; The position vector representing the improved path chosen by the gray wolf; represents the position vector of the gray wolf in the current iteration; t represents the number of current iterations; Represents the position vector of the gray wolf individual with the highest score at present; represents the remainder function, Indicates that the current iteration number is an even number; The global optimal solution unit is used to combine the fitness value of each gray wolf and the distance between each gray wolf and the current optimal solution, calculate the score of each gray wolf, select the solution candidate with the highest current score as the optimal solution in the current population, and use it for subsequent search operations until the maximum number of iterations is reached to obtain the global optimal solution, which is the final optimal feature selection solution; wherein the score is calculated by the following formula: ； in, represents the score of the Gray Wolves; Represents the normalized fitness value; Represents the normalized value of the distance between the gray wolf and the current optimal solution.

8. A gray wolf optimization feature selection device based on an adaptive mechanism, characterized in that: It includes a processor and a memory, wherein a computer program is stored in the memory, and the computer program can be executed by the processor to implement a gray wolf optimization feature selection method based on an adaptive mechanism as described in any one of claims 1 to 6.

Images