Feature selection method based on improved snow ablation optimization algorithm

By improving the snow ablation optimization algorithm, using the Good Point Set strategy and Support Vector Machine Classifier, combined with the periodic oscillation mutation strategy, the problems of large consumption of computing resources and insufficient accuracy of traditional feature selection technology are solved, and efficient feature selection is achieved, significantly improving Classification accuracy and feature subset reduction rate are obtained.

CN119025880APending Publication Date: 2024-11-26XIJING UNIV
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411046246.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-01
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

Traditional feature selection technology consumes a lot of computing resources and complex parameter tuning when processing massive data sets. The snow ablation optimization algorithm lacks accuracy in direct feature classification and has a large number of interference features, resulting in reduced efficiency.

Method used

Using an improved snow ablation optimization algorithm, initialize populations through the Good Point Set strategy, use the Support Vector Machine Classifier to calculate the fitness value, perform population division and periodic oscillation mutation strategies, calculate snow melt factor and melt rate, update individual positions, and use test functions Evaluate the classification results and output the optimal feature subset.

Benefits of technology

The accuracy of direct classification of features in the data set is significantly improved, the average number of feature subset selection is reduced, and the reduction rate of the number of features reaches 47% and 82%, improving the feature selection efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119025880A_ABST
    Figure CN119025880A_ABST
Patent Text Reader

Abstract

The invention discloses a feature selection method based on an improved snow ablation optimization algorithm, and relates to the technical field of machine learning and data mining, and the method comprises the following steps: obtaining to-be-selected features from a data set; using a good point set strategy to initialize the positions of population individuals; calculating the fitness value of each individual of the population by using a target function; uniformly dividing the original population into a sub-population and b sub-population; calculating values of a snow melting factor DDF and a melting rate M; updating individual positions of the population a and the population b in sequence; performing periodic oscillation mutation strategy disturbance on the population a; evaluating the optimal individual fitness; judging whether the current number of iterations reaches the set maximum number of iterations or not; and outputting the optimal feature subset. According to the feature selection method based on the improved snow ablation optimization algorithm, the feature subsets as few as possible can be effectively selected, and the classification effect as good as possible is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning and data mining, and in particular to a feature selection method based on an improved snow melting optimization algorithm. Background Art

[0002] In the era of big data, with the continuous innovation of data collection technology, the complexity of data analysis has risen rapidly, and the explosive growth of data dimensions has become one of the important causes of overfitting. For example, in medical data mining, there are many redundant and irrelevant features in the data set. These features may not only affect the efficiency of data collection and processing, but also have a negative impact on the accuracy of mining results, so data mining is needed. In the field of machine learning, feature selection is a crucial preprocessing step in classification, regression and data mining in machine learning. The purpose of feature selection is to use a selection method to delete redundant and irrelevant features in the data set and find the optimal feature subset. It can not only reduce data dimensions and improve the efficiency of machine learning algorithms, but also select the most useful features for the classification performance of the classifier from the original data set, thereby improving its classification accuracy.

[0003] Traditional feature selection techniques, such as filtering, packaging, and embedded methods, often face challenges such as high computing resource consumption and complex parameter tuning when processing massive data sets. In contrast, heuristic search algorithms show unique advantages with their group iterative search mechanism. Such algorithm models are intuitive in design, easy to program, require limited adjustment parameters, and have powerful optimization search capabilities. Therefore, they show higher adaptability in dealing with feature selection problems.

[0004] The Snow Ablation Optimizer (SAO) algorithm, as an innovative meta-heuristic algorithm launched in 2023, has demonstrated extraordinary performance in dealing with complex optimization problems. The algorithm simulates the sublimation and melting of snow in nature, and cleverly incorporates a balance mechanism between exploration and development into the algorithm design, effectively preventing the algorithm from converging to a local optimal solution prematurely. When the SAO algorithm is extended to the field of feature selection, its unique search logic significantly reduces the feature dimension, while significantly enhancing the performance accuracy and efficiency of the selected feature subset in the prediction model, opening up a novel path for optimizing the data preprocessing process and improving model performance. However, the following defects still exist:

[0005] 1. The accuracy of direct classification of data features by snow melting optimization algorithm is still insufficient, and there are many interference features after the direct classification of data features;

[0006] 2. After completing the highest classification of the data set, there are more interfering average feature subsets, which requires a large amount of screening of the average feature subsets, resulting in reduced efficiency;

[0007] In view of the above problems, the present invention provides a feature selection method based on snow melting optimization algorithm to solve the above problems. Summary of the invention

[0008] To achieve the above object, the present invention provides the following technical solution: a feature selection method based on an improved snow melting optimization algorithm, comprising the following steps:

[0009] S1: Get the features to be selected in the data set;

[0010] S2: Initialize the population using the good point set strategy;

[0011] S3: Use the support vector machine classifier as a model to calculate the optimal fitness value of each individual in the population;

[0012] S4: Determine the positions of the top three individuals in the fitness value ranking and the centroid position of the elite population, and store the position information of these four individuals in the elite pool array;

[0013] S5: Divide the original population uniformly and randomly into population a and population b;

[0014] S6: Determine the centroid positions of all individuals in the population, and update the positions of individuals in the population a;

[0015] S7: implementing a periodic oscillation mutation strategy perturbation on the a population;

[0016] S8: Calculate the value of snowmelt factor DDF and melting rate M;

[0017] S9: Update the individual positions in the b population;

[0018] S10: using a test function to evaluate the classification result of the individual with the best fitness;

[0019] S11: Determine whether the current number of iterations reaches the set maximum number of iterations;

[0020] S12: Output the optimal feature subset.

[0021] Further, preferably, according to step S1, the method for obtaining the features to be selected in the data set is:

[0022] S101: Obtain a data set, the data set including x rows of data and y columns of features, randomly divide the data set into 10 subsets according to the rows, randomly select one of the 10 subsets as a test set, and combine the remaining 9 subsets as a training set;

[0023] S102: Normalizing the feature data of the training set and the test set.

[0024] Further, preferably, according to step S2, the method for initializing the population with the good point set strategy is as follows:

[0025] S201: Calculate the best point r value, r = (r-1, r-2, ..., rn);

[0026] where ri = mod(2cos(2πi- / p),m i ,1), 1≤i≤n, i represents the i-th individual;

[0027] The best point r = ones(n,1)×2×cos((2×π⊙[1:d]) / p), where d is the dimension of the solution space and p is the smallest integer greater than 2×d+3;

[0028] S202: Construct a set P of good points with a number of A n (i) = {(r1i1, r2i2, ... r n i n )},i=1,2,3,...n:

[0029] Where A is the sample size; P n (i) is the set of good points;

[0030] S203: P n Mapped to the upper and lower limits of the feasible region where the population is located:

[0031] Among them, a j Indicates the lower limit of the current dimension; b j Indicates the upper limit of the current dimension; Represents the value of the i-th individual in the j-th dimension, that is, the position of the individual in the solution space of the current dimension.

[0032] Further, preferably, according to step S3, the method for calculating the fitness value includes:

[0033] minFitness=αγ R (D)+βR / Y, where γ R (D) is the classification error obtained using the selected feature subset, R is the number of selected features, Y is the number of all features, R / Y is the reduction ratio of the number of selected feature subsets compared to the original number of features, α and β are the weight coefficients for controlling the classification error and the reduction rate of the number of features, respectively. The lower the fitness function value, the better the individual.

[0034] Further, preferably, according to step S4, the method for updating the centroid position of a random element in the elite pool array and the elite individual in the population is:

[0035] Elite(t)∈[G(t),X second (t),X third (t),X c (t)]

[0036]

[0037] Among them, Elite(t) represents a random element selected from the elite pool array, G(t) represents the position of the best individual with the smallest fitness in the tth iteration; X second (t) indicates the position of the second individual ranked from low to high in fitness; X third (t) represents the position of the individual ranked third in fitness; X c (t) represents the centroid position of the individual in the elite population; n1 represents the number of individuals in the elite population.

[0038] Further, preferably, according to step S5, the original population is evenly and randomly divided into: population a and population b.

[0039] Further, preferably, according to step S6, the centroid positions of all individuals in the population and the method for updating the positions of individuals in the population a are as follows:

[0040]

[0041]

[0042] in, Indicates the centroid position of all individuals in the population; X i (t) represents the position of the i-th individual in the t-th iteration; X i (t+1) represents the position of the i-th individual in the t+1-th iteration; Elite(t) represents the position of an individual randomly selected from the elite pool array at the t-th iteration; BM i (t) represents a vector of random numbers that obey a Gaussian distribution, which is used to represent the Brownian motion of an individual in the solution space. G(t)-X i (t) represents the process of approaching the optimal solution during the process of updating the individual position in the solution space. It indicates the interaction between different individuals;

[0043] represents term-by-term multiplication, and θ1 represents a random number between 0 and 1.

[0044] Further, preferably, according to step S7, the periodic oscillation mutation strategy, the method of disturbing the individual positions of population a is:

[0045]

[0046]

[0047] Where A is the custom amplitude value; f r For custom frequency;

[0048] Every fr -1 In the next generation, the mutation strategy is applied to the characteristics of all individuals in the population a.

[0049] Further, preferably, according to step S8, the snowmelt factor DDF and the melting rate M are calculated as follows:

[0050]

[0051]

[0052] M=DDF×T

[0053] Where t is the current iteration number of the algorithm; t max Indicates the maximum number of iterations of the algorithm.

[0054] Further, preferably, according to step S9, the method for updating the individual positions in the b population is:

[0055]

[0056] Here, θ2 represents a random number between 0 and 1.

[0057] After the individual positions in subpopulations a and b are updated, the number of individuals in population a is increased by 1 and the number of individuals in population b is decreased by 1 in each iteration.

[0058] Compared with the prior art, the present invention provides a feature selection method based on an improved snow melting optimization algorithm, which has the following beneficial effects:

[0059] 1. Compared with the snow melting optimization algorithm and SVM, the accuracy of direct classification of data set features has obvious advantages and has achieved the highest classification accuracy;

[0060] 2. While achieving the highest classification accuracy, it can achieve a relatively small average number of feature subset selections;

[0061] 3. Compared with SonarEW, Heart, Hepatits, and Musk1 algorithms, the average number of feature selections is significantly reduced in the data set concentration effect, and the reduction rate of the number of features remains between 47% and 82%. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a system flow chart of a feature selection method based on an improved snow melting optimization algorithm;

[0063] Figure 2 This is a flow chart of the SVM classifier model process for a feature selection method based on an improved snow melting optimization algorithm. DETAILED DESCRIPTION

[0064] The embodiment uses a data set in the UCI public database, and the present invention performs feature selection on the data set based on the feature selection method of the improved snow melting optimization algorithm. Feature selection aims to eliminate irrelevant and redundant features in the data set, thereby improving the performance of machine learning. The quality of the candidate feature subset is measured by a test function, and the measurement indicators are usually the classification accuracy and the number of selected feature subsets.

[0065] Among them: feature selection method based on improved snow ablation optimization algorithm (GVSAO); Snow Ablation Optimizer (SAO); Whale Optimization Algorithm (WOA); Grey Wolf Optimizer (GWO); Support Vector Machine (SVM).

[0066] Reference Figure 1-2 The present invention provides a technical solution: a feature selection method based on an improved snow ablation optimization algorithm, comprising the following steps:

[0067] S1: Obtain the features to be selected in the data set, including the following steps:

[0068] S2: Initialize the population using the good point set strategy:

[0069] S3: Calculate the fitness value of each individual in the population using the objective function;

[0070] S4: Determine the positions of the top three individuals in the fitness value ranking and the centroid position of the elites defined as those in the top 50% of the population whose fitness values ​​are determined, and the centroid position is defined as the arithmetic mean of the positions of all elite individuals whose fitness values ​​are in the top 50% from low to high, and store the position information of these four individuals in the elite pool array;

[0071] S5: uniformly and randomly divide the original population into: population a and population b;

[0072] S6: Determine the centroid positions of all individuals in the population, and update the positions of individuals in the population a;

[0073] S7: implementing a periodic oscillation mutation strategy perturbation on the a population;

[0074] S8: Calculate the value of snowmelt factor DDF and melting rate M;

[0075] S9: Update the individual positions in the b population;

[0076] S10: using a test function to evaluate the classification result of the individual with the best fitness;

[0077] S11: Determine whether the current number of iterations reaches the set maximum number of iterations;

[0078] S12: Output the optimal feature subset.

[0079] As a preferred embodiment, according to step S1, the method for obtaining the features to be selected in the data set is:

[0080] S101: Obtain a data set, the data set including x rows of data and y columns of features, randomly divide the data set into 10 subsets according to the rows, randomly select one of the 10 subsets as a test set, and combine the remaining 9 subsets as a training set;

[0081] S102: Normalizing the feature data of the training set and the test set.

[0082] As a preferred embodiment, according to step S2, the method for initializing the population with the good point set strategy is as follows:

[0083] S201: Calculate the best point r value, r = (r-1, r-2, ..., rn),

[0084] where ri = mod(2cos(2πi- / p),m i ,1), 1≤i≤n, i represents the i-th individual; the best point r=ones(n,1)×2×cos((2×π⊙[1:d]) / p), where d is the dimension of the solution space and p is the smallest integer greater than 2×d+3;

[0085] S202: Construct a set P of good points with a number of A n (i) = {(r1i1, r2i2, ... r n i n )},i=1,2,3,...n:

[0086] k=[1:n]'×ones(1,d)

[0087] P n (i) = |k⊙r|,

[0088] Among them, k is the coefficient for calculating the best point r of the best point set, which is a matrix with n rows and d columns, where each column is an arrangement from 1 to n from top to bottom, A is the number of individuals in the population, d is the dimension of the solution space, and P n (i) is the position of the i-th individual in the good point set.

[0089] S203: P n Mapped to the upper and lower limits of the feasible region where the population is located:

[0090] Among them, a j Indicates the lower limit of the current dimension; b j Indicates the upper limit of the current dimension; It represents the position of the i-th individual in the j-th dimension, that is, the position of the individual in the solution space of the current dimension.

[0091] As a preferred embodiment, according to step S3, using the support vector machine classifier as a model, the method for calculating the best individual fitness value is as follows:

[0092] S301: Create a support vector machine model: set the penalty coefficient c to 4 and the parameter gamma of the RBF kernel function to 0.8;

[0093] S302: training a support vector machine model: giving the feature subset optimized by the improved snow melting optimization algorithm in the training set to the support vector machine for training;

[0094] S303: Test data: The best feature subset obtained by the trained model is tested using the test set, and the classification error rate is obtained by comparing the test results with the predicted labels.

[0095] The function value of individual fitness is as follows:

[0096] minFitness=αγ R (D)+βR / Y, where Fitness is the fitness value of each individual in the population used to evaluate the classification accuracy and feature selection rate of the classifier. The lower the feature selection rate, the fewer features are selected in the algorithm optimization result, and the better the feature selection effect is. R(D) is the classification accuracy obtained by putting the selected feature subset into the support vector machine classifier for verification, and then the classification error is obtained by subtracting it from 1. R is the number of selected features, Y is the number of all features, R / Y is the reduction ratio of the number of selected feature subsets compared with the original number of features, α and β are the weight coefficients for controlling the classification error and the reduction rate of the number of features, respectively. The lower the fitness function value Fitness is, the better the classification performance of the selected feature subset is.

[0097] It should be noted that the support vector machine classification model established in the present invention is as follows Figure 2 shown.

[0098] As a preferred embodiment, according to step S4, a random element Elite(t) in the elite pool array and the centroid position X of the elite individual in the population are c (t) The update method is:

[0099] Elite(t)∈[G(t),X second (t),X third (t),X c (t)]

[0100]

[0101] Elite(t) represents a random element selected from the elite pool array, G(t) represents the location of the best individual with the smallest fitness at the tth iteration, and X second Indicates the position of the second individual in terms of fitness when the fitness is ranked from low to high, X third Indicates the position of the third individual in fitness ranking, X c (t) represents the centroid position of the elite individuals in the population whose fitness is in the top 50%. These four positions constitute the elite pool array as elements. n1 represents the number of individuals in the elite sub-population, which is 50% of the number of individuals n in the population.

[0102] It should be noted that the number of elements in the elite pool array is 4, and the elements are the positions of the individuals with the top three fitnesses and the centroid positions of the elite individuals with the top 50% fitnesses in the population.

[0103] As a preferred embodiment, according to step S5, the original population is uniformly and randomly divided into: population a and population b, that is, the number of individuals in population a and population b is N / 2, wherein N is 50% of the number of all individuals in the original population.

[0104] As a preferred embodiment, according to step S6, the centroid positions of all individuals in the population And the method for updating the individual positions in the a population is as follows:

[0105]

[0106]

[0107] in, represents the centroid position of all individuals in population a; X i (t) represents the position of the i-th individual in the t-th iteration; X i (t+1) represents the position of the i-th individual in the t+1-th iteration; Elite(t) represents the position of an individual randomly selected from the elite pool array at the t-th iteration; BM i (t) represents a vector of random numbers that obey a Gaussian distribution, which is used to represent the Brownian motion of an individual in the solution space. G(t)-X i (t) represents the process of approaching the optimal solution during the process of updating the individual position in the solution space, X(t)-X i (t) represents the interaction between different individuals;

[0108] represents term-by-term multiplication, and θ1 represents a random number between 0 and 1.

[0109] As a preferred embodiment, according to step S7, the periodic oscillation mutation strategy, the method of disturbing the individual positions of population a is:

[0110]

[0111]

[0112] Among them, X i,j (t+1) represents the position of the i-th individual after the mutation strategy is applied to the j-th dimension in the t-th iteration, X i,j (t) represents the position of the i-th individual before the j-th dimension perturbation in the t-th iteration, rand is a random number between 0 and 1, A is the custom amplitude value, t is the number of iterations, and f r is a custom frequency; let A=5,f r =10;

[0113] The period of applying the periodic oscillation mutation strategy in each iteration is fr -1 , every fr -1 In the next generation, the mutation strategy is applied to the characteristics of all individuals in the population a.

[0114] As a preferred embodiment, according to step S8, the snowmelt factor DDF and the melting rate M value are calculated as follows:

[0115]

[0116]

[0117] M=DDF×T

[0118] Among them, t is the current iteration number of the algorithm; t max Indicates the maximum number of iterations of the algorithm.

[0119] As a preferred embodiment, according to step S9, the method for updating the individual positions in the b population is:

[0120]

[0121] Among them, X i (t+1) represents the updated position of the i-th individual in the b population in the t-th iteration, M represents the melting rate, G(t) represents the position of the individual with the minimum fitness in the current iteration, and BM i (t) represents a vector containing random numbers that follow a Gaussian distribution, X i (t) represents the position of the i-th individual before the update, represents the centroid position of all individuals in the original population, and θ2 represents a random number between 0 and 1.

[0122] It should be noted that we check whether the updated position is in the solution space, put the individual position into the set feasible domain, use the Sigmoid function to map the individual characteristics to the [0,1] interval, and then compare with the random number to binarize the individual characteristics into 0 and 1.

[0123]

[0124]

[0125] Among them, X i,j (t) represents the position of the i-th individual in the j-th dimension in the t-th iteration, 1≤i≤n, 1≤j≤d, n is the number of individuals in the population, and d is the dimension of the solution space where the individual is located.

[0126] Check whether the position of the individual in the population is out of bounds, and put the position of the individual that is out of bounds back into the solution space, evaluate the fitness of the optimal individual through the fitness function, update the data of the elite pool array, and increase the number of individuals in population a by 1, and reduce the number of individuals in population b by 1;

[0127] S10: using a test function to evaluate the classification result of the individual with the best fitness;

[0128] S11: Determine whether the current number of iterations reaches the set maximum number of iterations;

[0129] S12: Output the optimal feature subset and draw the fitness curve.

[0130] This embodiment is tested in the Windows 11 operating system, 16G memory, and MATLAB 2022b experimental environment. The maximum number of iterations is 80, the population size is 60, and the test is performed on four data sets. Each experiment is performed independently 5 times. The evaluation indicators are the average classification accuracy and the average number of selected feature subsets. The experimental results are shown in Tables 1, 2, and 3.

[0131] Table 1 Description of original dataset:

[0132]

[0133]

[0134] Table 2 Comparison of average classification accuracy of the method of the present invention with other existing methods and direct classification by SVM classifier:

[0135]

[0136] Table 3 Comparison of the average number of feature selections between the method of the present invention and other existing methods:

[0137]

[0138] Experimental results and analysis:

[0139] From the results in Table 2, the proposed method has obvious advantages over the snow melting optimization algorithm, other algorithms and SVM in the accuracy of direct classification of data set features. It has achieved the highest classification accuracy in the SonarEW, Heart, Hepatits and Musk1 data sets.

[0140] From the results in Table 3, we can see that all four algorithms can significantly reduce the average number of feature selections in the dataset, and the reduction rate of the number of features remains between 47% and 82%.

[0141] In summary, the method of the present invention can achieve the highest classification accuracy in the SoanrEW, Heart, Hepatits, vote, and Musk1 datasets while achieving a relatively small average number of feature subset selections.

[0142] What is described above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A feature selection method based on an improved snow melting optimization algorithm, characterized in that: The following steps are involved: S1: Get the features to be selected in the data set; S2: Initialize the population using the good point set strategy; S3: Use the support vector machine classifier as a model to calculate the optimal fitness value of each individual in the population; S4: Determine the positions of the top three individuals in the fitness value ranking and the centroid position of the elite population, and store the position information of these four individuals in the elite pool array; S5: Divide the original population uniformly and randomly into population a and population b; S6: Determine the centroid positions of all individuals in the population, and update the positions of individuals in the population a; S7: implementing a periodic oscillation mutation strategy perturbation on the a population; S8: Calculate the value of snowmelt factor DDF and melting rate M; S9: Update the individual positions in the b population; S10: using a test function to evaluate the classification result of the individual with the best fitness; S11: Determine whether the current number of iterations reaches the set maximum number of iterations; S12: Output the optimal feature subset.

2. The feature selection method based on the improved snow ablation optimization algorithm according to claim 1, characterized in that: According to step S1, the method for obtaining the features to be selected in the data set is: S101: Obtain a data set, the data set including x rows of data and y columns of features, randomly divide the data set into 10 subsets according to the rows, randomly select one of the 10 subsets as a test set, and combine the remaining 9 subsets as a training set; S102: Normalizing the feature data of the training set and the test set.

3. The feature selection method based on the improved snow ablation optimization algorithm according to claim 2 is characterized in that: According to step S2, the method for initializing the population with the good point set strategy is as follows: S201: Calculate the best point r value, r = (r-1, r-2, ..., rn); where ri = mod(2cos(2πi- / p),m i ,1), 1≤i≤n, i represents the jth individual; The best point r = ones(n,1)×2×cos((2×π⊙[1:d]) / p), where d is the dimension of the solution space and p is the smallest integer greater than 2×d+3; S202: Construct a set P of good points with a number of A n (i) = {(r1i1, r2i2, ... r n i n )},i=1,2,3,...n: Where A is the sample size; P n (i) is the set of good points; S203: P n Mapped to the upper and lower limits of the feasible region where the population is located: X i j =a j +P n (i)(b j -a j ); Among them, a j Indicates the lower limit of the current dimension; b j Indicates the upper limit of the current dimension; X i j Represents the value of the i-th individual in the j-th dimension, that is, the position of the individual in the solution space of the current dimension.

4. The feature selection method based on the improved snow ablation optimization algorithm according to claim 3 is characterized in that: According to step S3, the method for calculating the fitness value includes: min Fitness = αγ R (D)+βR / Y, where γ R (D) is the classification error obtained using the selected feature subset, R is the number of selected features, Y is the number of all features, R / Y is the reduction ratio of the number of selected feature subsets compared to the original number of features, α and β are the weight coefficients for controlling the classification error and the reduction rate of the number of features, respectively. The lower the fitness function value, the better the individual.

5. The feature selection method based on the improved snow ablation optimization algorithm according to claim 4 is characterized in that: According to step S4, the method for updating the centroid position of a random element in the elite pool array and the elite individual in the population is: Elite(t)∈[G(t),X second (t),X third (t),X c (t)] Among them, Elite(t) represents a random element selected from the elite pool array, G(t) represents the position of the best individual with the smallest fitness in the tth iteration; X second (t) indicates the position of the second individual ranked from low to high in fitness; X third (t) represents the position of the individual ranked third in fitness; X c (t) represents the centroid position of the individual in the elite population; n1 represents the number of individuals in the elite population.

6. The feature selection method based on the improved snow ablation optimization algorithm according to claim 5, characterized in that: According to step S6, the centroid positions of all individuals in the population and the method for updating the positions of individuals in the population a are as follows: in, Indicates the centroid position of all individuals in the population; X i (t) represents the position of the i-th individual in the t-th iteration; X i (t+1) represents the position of the i-th individual in the t+1-th iteration; Elite(t) represents the position of an individual randomly selected from the elite pool array at the t-th iteration; BM i (t) represents a vector of random numbers that obey a Gaussian distribution, which is used to represent the Brownian motion of an individual in the solution space. i (t) represents the process of approaching the optimal solution during the process of updating the individual position in the solution space. It indicates the interaction between different individuals; represents term-by-term multiplication, and θ1 represents a random number between 0 and 1.

7. The feature selection method based on the improved snow ablation optimization algorithm according to claim 6, characterized in that: According to step S7, the periodic oscillation mutation strategy, the method of disturbing the position of individuals in population a is: Every fr -1 In the next generation, the mutation strategy is applied to the characteristics of all individuals in the population a.

8. The feature selection method based on the improved snow melting optimization algorithm according to claim 7 is characterized in that: According to step S8, the snowmelt factor DDF and the melting rate M are calculated as follows: M=DDF×T Where t is the current iteration number of the algorithm; t max Indicates the maximum number of iterations of the algorithm.

9. The feature selection method based on the improved snow ablation optimization algorithm according to claim 8, characterized in that: According to step S9, the method for updating the individual positions in the b population is: Here, θ2 represents a random number between 0 and 1.

Citation Information

Cited By

  • Water quality clustering method based on improved snow ablation algorithm

    CN121278412A