Deep evolutionary multi-objective feature selection method for airport energy efficiency optimization
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANTONG UNIV
- Filing Date
- 2026-03-24
- Publication Date
- 2026-08-07
AI Technical Summary
[0003]在过去的几年中，学者们多依赖于互信息、皮尔逊相关系数等浅层统计指标来衡量特征重要性，然而，这些指标主要捕捉变量间的线性相关性，而机场运行系统是一个复杂的非线性动态系统，现有方法忽略了这种深层的非线性交互信息
Smart Images
Figure CN122527652A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of information intelligent processing technology, specifically to a deep evolutionary multi-objective feature selection method for airport energy efficiency optimization. Technical Background
[0002] With the continuous advancement of smart airport construction, the amount of data generated by airport operation systems is growing explosively, covering multiple dimensions such as flight dynamics, meteorological information, air traffic control instructions, passenger flow, and ground support. This data typically exhibits significant characteristics such as high dimensionality, strong noise, multi-source heterogeneity, and nonlinear coupling.
[0003] In the past few years, scholars have relied on shallow statistical indicators such as mutual information and Pearson correlation coefficient to measure feature importance. However, these indicators mainly capture the linear correlation between variables, while the airport operation system is a complex nonlinear dynamic system, and existing methods ignore this deep nonlinear interaction information.
[0004] The emerging deep learning technology in recent years has brought new solutions to the construction of smart civil aviation. Automatic encoders can automatically mine deep potential features from high-dimensional airport operational data, showing promising prospects in the field of airport energy efficiency optimization. Summary of the Invention
[0005] To address the aforementioned issues, this invention provides a deep evolutionary multi-objective feature selection method for airport energy efficiency optimization. In the data processing stage, it applies mutual information and K-Means clustering to mine the connections between data, assigns different weights to the data for preprocessing, trains a deep autoencoder model, and uses a deep guidance matrix to reduce irrelevant and redundant features, lowers the data dimensionality to reduce computational requirements, and improves the efficiency of feature selection. This method has strong application value for the construction of smart civil aviation.
[0006] The inventive concept of this invention is as follows: First, hierarchical weighted initialization data is constructed using mutual information and K-Means clustering; then, a deep autoencoder model is trained, and the importance of deep features reflecting the nonlinear relationship between features is extracted by analyzing the weights of the model's input layer; next, the importance of the deep features is used as prior knowledge to construct a deep guidance matrix to guide population evolution; finally, an evolutionary search is performed based on the guidance matrix, and an adaptive mechanism is used to dynamically update the guidance matrix until the termination condition is met and the optimal solution set is output.
[0007] This invention is achieved through the following measures: a deep evolutionary multi-objective feature selection method for airport energy efficiency optimization, comprising the following steps:
[0008] S10: Construct hierarchical weighted initialization data using mutual information and K-Means clustering;
[0009] S20: Train a deep autoencoder model and extract the importance of deep features that reflect the nonlinear relationship between features by analyzing the weights of the model's input layer;
[0010] S30: Using the importance of the deep features as prior knowledge, construct a deep guidance matrix to guide population evolution;
[0011] S40: Perform evolutionary search based on the guidance matrix and dynamically update the guidance matrix using an adaptive mechanism until the termination condition is met and the optimal solution set is output.
[0012] Further, step S10 includes the following steps:
[0013] S11: Collect raw samples of airport operations, apply data processing and analysis to denoise the data, and assign labels to samples based on whether there is a delay. Indicates size is The data sample set, with sample labels as The feature vector is ;
[0014] S12: First, mutual information is used to quantify the features. With category labels Degree of dependence between them:
[0015]
[0016] in and They are and The marginal probability mass function, It is their joint probability mass function.
[0017] S13: The feature set is divided using the K-means clustering algorithm based on mutual information values. Here, the number of clusters is set to... The aim is to automatically cluster features into three levels and select probabilities. Defined as the normalized centroid value:
[0018]
[0019] in, These represent the centroid values of strongly correlated, weakly correlated, and unrelated feature clusters, respectively.
[0020] S14: Using the weight matrix We weight the original data to obtain a weighted matrix:
[0021]
[0022] Further, step S20 includes the following steps:
[0023] S21: From the training dataset Get the current batch of data Using an automatic encoder Batch Mapping to the latent space yields a latent representation of the encoder network. ;
[0024] S22: Using a decoder Receive potential representation And map it back to the original space to obtain the reconstructed original data. , where the decoder The definition is as follows:
[0025]
[0026] in, The activation function for the hidden layer. and These represent the weight matrices of the first hidden layer and the output layer, respectively. and It is a bias term;
[0027] S23: Calculate the initial feature map matrix using the mean squared error combined with the regularization loss function. and the data reconstructed by the decoder The differences between them are expressed as follows:
[0028]
[0029] in, It is the Frobenius norm, which represents the sum of the squared differences of all elements in the matrix and is used to measure the reconstruction error;
[0030] S24: After the model training is complete, extract the weight vector from the one-to-one feature selection layer, normalize the weight vector, and the resulting value is the importance of the deep features. .
[0031] Further, step S30 includes the following steps:
[0032] S31: In the current population In this study, non-dominated ranking was used to evaluate each individual. The quality of. Defining individual importance. The number of other individuals that this individual dominates in the current population:
[0033]
[0034] in Represents an individual Dominant Individual , The larger the value, the higher the quality of the feature subset represented by that individual;
[0035] S32: Utilize the normalized deep feature importance vector obtained in step S24 Combined with the individual importance obtained in step S31 Construct a depth-guided matrix SM, the matrix of which is the first... Line 1 Column elements Indicates the first The feature in the first The guiding probability for each individual is calculated using the following formula:
[0036]
[0037] in, It is a normalization function that maps numerical values to the interval [0, 1].
[0038] Further, step S40 includes the following steps:
[0039] S41: Perform dimensionality reduction on non-elite individuals in the population. The Given a feature, calculate the probability that it flips to the unselected state 0. :
[0040]
[0041] in, Let the current iteration algebra be... To control the coefficient of decay rate, For the first The mean of all characteristic guide values for an individual;
[0042] S42: Use elite individuals to repair the population. Randomly select an elite individual from the non-dominated solution set. ,like If this trait is selected among elite individuals, then the individual will be forcibly... The feature is restored to 1; if Then, based on probability Learning from elite individuals, probability Defined as:
[0043]
[0044] S43: Merge the population after operations S41 and S42 with the parent population, perform non-dominated sorting and crowding distance calculation, and select the next generation population. ;
[0045] S44: Dynamically adjust the SM matrix based on feedback from population evolution. Compare the average classification error rate of the offspring elite set with that of the parent elite set. If the offspring error rate decreases, update the guiding matrix using the following formula:
[0046]
[0047] in This is the adjustment step size determined based on changes in the characteristic state. If the offspring error rate does not decrease, it indicates that the current guiding matrix has failed. In this case, the guiding matrix is reinitialized using method S30, and finally, a termination judgment is performed to end the process.
[0048] Furthermore, this invention also provides a deep evolutionary multi-objective feature selection system for airport energy efficiency optimization. The system, applying the method described in this invention, includes the following steps:
[0049] The hierarchical weighted initialization data construction module is configured to perform the following process: construct hierarchical weighted initialization data using mutual information and K-Means clustering;
[0050] The autoencoder module is configured to perform the following processes: training a deep autoencoder model and extracting the importance of deep features that reflect the nonlinear relationships between features by analyzing the weights of the model's input layer;
[0051] The deep guidance matrix module is configured to perform the following process: using the importance of the deep features as prior knowledge, constructing a deep guidance matrix to guide the evolution of the population;
[0052] The adaptive mechanism module is configured to perform the following process: perform evolutionary search based on the guidance matrix, and dynamically update the guidance matrix using an adaptive mechanism until the termination condition is met and the optimal solution set is output.
[0053] Meanwhile, the present invention proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed, it implements the steps of the method described in the present invention.
[0054] Furthermore, the present invention proposes a computer-readable storage medium having a computer program stored thereon, the computer program being configured to implement the steps of the method described in the present invention when invoked by a processor.
[0055] Finally, the present invention provides a computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the method described in the present invention.
[0056] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0057] 1. This invention presents a deep evolutionary multi-objective feature selection method for airport energy efficiency optimization, introducing a hierarchical weighted mechanism based on mutual information and K-Means clustering to effectively suppress high-dimensional noise and reduce the ineffective search space. Faced with high-dimensional, multi-source data containing strong noise, traditional feature selection methods often lack specificity in the preprocessing stage, making subsequent algorithms highly susceptible to interference from redundant features. This invention innovatively introduces a hierarchical weighted preprocessing structure based on a combination of mutual information and K-Means clustering. Mutual information is used to quantify the dependence between each feature and the target category. The K-Means algorithm automatically clusters features and divides them into three levels: strongly correlated, weakly correlated, and unrelated. Different weight matrices are then assigned to features at different levels, reducing the blind search space of the evolutionary algorithm and enhancing the robustness of feature subsets in complex data environments.
[0058] 2. This invention introduces a deep autoencoder into a multi-objective feature selection framework, mapping high-dimensional batch data to a latent space and reconstructing it through the autoencoder. After model training, the weight vectors of the input layer are extracted and normalized, directly transforming them into deep feature importance. This mechanism can uncover hidden deep semantic information and non-linear dependency interactions in the data, improving the classification accuracy and data representation quality of the final selected feature subset.
[0059] 3. This invention constructs a deep guidance matrix that combines the prior importance of features extracted by the autoencoder with the posterior evolutionary quality of the population based on non-dominated ranking. This matrix provides precise directional search navigation for dimensionality reduction operators for non-elite solutions and repair operators for elite solutions, driving the algorithm to optimize towards high-potential feature subspaces. While approaching the Pareto optimal front for multi-objective solutions, it effectively avoids local optimum traps in high-dimensional solution spaces, accelerates the global convergence process of the algorithm, and reduces the time complexity of the model in optimization scenarios. Attached Figure Description
[0060] The technical solution and its beneficial effects of the present invention will become apparent from the following detailed description of specific embodiments in conjunction with the accompanying drawings.
[0061] Figure 1 The diagram shown is a flowchart of a deep evolutionary multi-objective feature selection method for airport energy efficiency optimization in an embodiment of this application.
[0062] Figure 2 The diagram shown is a block diagram of a deep evolutionary multi-objective feature selection method for airport energy efficiency optimization in an embodiment of this application.
[0063] Figure 3 shows a model diagram of a deep evolutionary multi-objective feature selection method for airport energy efficiency optimization in an embodiment of this application. Detailed Implementation
[0064] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0065] Example 1: This example provides a deep evolutionary multi-objective feature selection method for airport energy efficiency optimization, such as... Figure 1-3 As shown, it includes the following steps:
[0066] S1: Construct hierarchical weighted initialization data using mutual information and K-Means clustering;
[0067] S2: Train a deep autoencoder model and extract the importance of deep features that reflect the nonlinear relationship between features by analyzing the weights of the model's input layer;
[0068] S3: Using the importance of the deep features as prior knowledge, construct a deep guidance matrix to guide population evolution;
[0069] S4: Perform evolutionary search based on the guidance matrix and dynamically update the guidance matrix using an adaptive mechanism until the termination condition is met and the optimal solution set is output.
[0070] Specifically, step S1 includes the following steps:
[0071] S11: Collect raw samples of airport operations, apply data processing and analysis to denoise the data, and assign labels to samples based on whether there is a delay. Indicates size is The data sample set, with sample labels as The feature vector is The processed data is shown in Table 1 below:
[0072] Table 1
[0073]
[0074] S12: First, mutual information is used to quantify the features. With category labels Degree of dependence between them:
[0075]
[0076] in and They are and The marginal probability mass function, It is their joint probability mass function.
[0077] S13: The feature set is divided using the K-means clustering algorithm based on mutual information values. Here, the number of clusters is set to... The aim is to automatically cluster features into three levels and select probabilities. Defined as the normalized centroid value:
[0078]
[0079] in, These represent the centroid values of strongly correlated, weakly correlated, and unrelated feature clusters, respectively.
[0080] S14: Using the weight matrix We weight the original data to obtain a weighted matrix:
[0081]
[0082] Specifically, step S2 includes the following steps:
[0083] S21: From the training dataset Get the current batch of data Using an automatic encoder Batch Mapping to the latent space yields a latent representation of the encoder network. ;
[0084] S22: Using a decoder Receive potential representation And map it back to the original space to obtain the reconstructed original data. , where the decoder The definition is as follows:
[0085]
[0086] in, The activation function for the hidden layer. and These represent the weight matrices of the first hidden layer and the output layer, respectively. and It is a bias term;
[0087] S23: Calculate the initial feature map matrix using the mean squared error combined with the regularization loss function. and the data reconstructed by the decoder The differences between them are expressed as follows:
[0088]
[0089] in, It is the Frobenius norm, which represents the sum of the squared differences of all elements in the matrix and is used to measure the reconstruction error;
[0090] S24: After the model training is complete, extract the weight vector from the one-to-one feature selection layer, normalize the weight vector, and the resulting value is the importance of the deep features. As shown in Table 2:
[0091] Table 2
[0092]
[0093] Specifically, step S3 includes the following steps:
[0094] S31: In the current population In this study, non-dominated ranking was used to evaluate each individual. The quality of. Defining individual importance. The number of other individuals that this individual dominates in the current population:
[0095]
[0096] in Represents an individual Dominant Individual , The larger the value, the higher the quality of the feature subset represented by that individual;
[0097] S32: Utilize the normalized deep feature importance vector obtained in step S24 Combined with the individual importance obtained in step S31 Construct a depth-guided matrix SM, the matrix of which is the first... Line 1 Column elements Indicates the first The feature in the first The guiding probability for each individual is calculated using the following formula:
[0098]
[0099] in, It is a normalization function that maps numerical values to the interval [0, 1].
[0100] Specifically, step S4 includes the following steps:
[0101] S41: Perform dimensionality reduction on non-elite individuals in the population. The Given a feature, calculate the probability that it flips to the unselected state 0. :
[0102]
[0103] in, Let the current iteration algebra be... To control the coefficient of decay rate, For the first The mean of all characteristic guide values for an individual;
[0104] S42: Use elite individuals to repair the population. Randomly select an elite individual from the non-dominated solution set. ,like If this trait is selected among elite individuals, then the individual will be forcibly... The feature is restored to 1; if Then, based on probability Learning from elite individuals, probability Defined as:
[0105]
[0106] S43: Merge the population after operations S41 and S42 with the parent population, perform non-dominated sorting and crowding distance calculation, and select the next generation population. ;
[0107] S44: Dynamically adjust the SM matrix based on feedback from population evolution. Compare the average classification error rate of the offspring elite set with that of the parent elite set. If the offspring error rate decreases, update the guiding matrix using the following formula:
[0108]
[0109] in This is the adjustment step size determined based on changes in the characteristic state. If the offspring error rate does not decrease, it indicates that the current guiding matrix has failed. In this case, the guiding matrix is reinitialized using method S30, and finally, a termination judgment is performed to end the process.
[0110] Example 2: Referring to Example 1, this example uses the parameters and results calculated in Example 1 to compare with the traditional method, demonstrating the superiority of this example. In the specific comparison, we used different models and compared different indicators. The final results show that this example is preferred over the traditional method.
[0111] 1. Traditional Model
[0112] The LMOCSO model: This model is based on a competitive swarm optimization architecture and is specifically designed to solve large-scale multi-objective feature selection problems. It innovatively proposes a two-stage position update strategy: the first stage utilizes a pairwise competition mechanism among particles to quickly identify and eliminate inferior feature combinations; the second stage uses a refined local search strategy to guide the population to approach the Pareto optimal front while preserving key features, thus effectively balancing convergence speed and population diversity in the high-dimensional search space.
[0113] The SparseEA2 model leverages the prior knowledge that "optimal feature subsets are usually sparse" by introducing a sparse initialization strategy and a specially designed sparse genetic operator. Unlike traditional algorithms that roam the entire search space, SparseEA2 strictly restricts the population to evolve within a low-sparseness subspace, ensuring that newly generated offspring always maintain an extremely low feature selection rate. This strategy significantly reduces the ineffective evaluation of redundant features, substantially improving the computational efficiency and accuracy of the algorithm in finding the minimum feature subset in high-dimensional data environments.
[0114] 2. Comparison Indicators
[0115] During the evolutionary iteration process, to calculate the classification error rate of each individual in the population, a leave-one-out cross-validation method based on the training set DTrain, combined with a 1NN classifier, is employed. Specifically, each sample in DTrain is used sequentially as a test sample, and the remaining samples form a training subset to train the 1NN classifier, thereby obtaining the classification error rate of the corresponding feature subset. This design effectively avoids overfitting bias in the feature selection process. To reduce statistical bias caused by experimental randomness, five-fold cross-validation is performed independently 30 times, and the experimental results of all comparative algorithms are reported as the mean and standard deviation of 30 independent runs.
[0116] 3. Comparison Results
[0117] Experimental results show that the proposed method outperforms the other comparative algorithms on average across almost all test cases. Specifically, the proposed method achieves the lowest classification error rate across all three datasets; and in terms of feature subset size, the proposed method selects fewer features than other comparative algorithms. This strongly demonstrates that the proposed method surpasses existing multi-objective algorithms in both accuracy and dimensionality reduction.
[0118] Table 3
[0119]
[0120] Example 3: This example proposes a deep evolutionary multi-objective feature selection system for airport energy efficiency optimization, applying the steps of the method described in this invention. The system includes:
[0121] The hierarchical weighted initialization data construction module is configured to perform the following process: construct hierarchical weighted initialization data using mutual information and K-Means clustering;
[0122] The autoencoder module is configured to perform the following processes: training a deep autoencoder model and extracting the importance of deep features that reflect the nonlinear relationships between features by analyzing the weights of the model's input layer;
[0123] The deep guidance matrix module is configured to perform the following process: using the importance of the deep features as prior knowledge, constructing a deep guidance matrix to guide the evolution of the population;
[0124] The adaptive mechanism module is configured to perform the following process: perform evolutionary search based on the guidance matrix, and dynamically update the guidance matrix using an adaptive mechanism until the termination condition is met and the optimal solution set is output.
[0125] Example 4: This example proposes an electronic system, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method steps of the present invention.
[0126] Example 5: This example proposes a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the steps of the method described in this invention, which will not be repeated here.
[0127] Example 6: This example proposes a computer program product, including a computer program / instructions. When the computer program / instructions are executed by a processor, they implement the steps of the method described in this invention, which will not be repeated here.
[0128] It should be noted that the processing flow of embodiments 3-6 corresponds to the specific steps of the method provided in embodiment 1 of the present invention, and has the corresponding functional modules and beneficial effects of the method. Technical details not described in detail in this embodiment can be found in the method provided in embodiment 1 of the present invention.
[0129] The program code used to implement the methods of this application may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.
[0130] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A deep evolutionary multi-objective feature selection method for airport energy efficiency optimization, characterized in that, Includes the following steps: S1: Construct hierarchical weighted initialization data using mutual information and K-Means clustering; S2: Train a deep autoencoder model and extract the importance of deep features that reflect the nonlinear relationship between features by analyzing the weights of the model's input layer; S3: Using the importance of the deep features as prior knowledge, construct a deep guidance matrix to guide population evolution; S4: Perform evolutionary search based on the guidance matrix and dynamically update the guidance matrix using an adaptive mechanism until the termination condition is met and the optimal solution set is output.
2. The deep evolutionary multi-objective feature selection method for airport energy efficiency optimization according to claim 1, characterized in that, S1 includes the following steps: S11: Collect raw samples of airport operations, apply data processing and analysis to denoise the data, and assign labels to samples based on whether there is a delay. Indicates size is The data sample set, with sample labels as The feature vector is ; S12: First, mutual information is used to quantify the features. With category labels Degree of dependence between them: ； in and They are and The marginal probability mass function, It is their joint probability mass function; S13: Use the K-means clustering algorithm to partition the feature set based on mutual information values, and select probability... Defined as the normalized centroid value: ； in, These represent the centroid values of strongly correlated, weakly correlated, and unrelated feature clusters, respectively. S14: Using the weight matrix We weight the original data to obtain a weighted matrix: 。 3. The deep evolutionary multi-objective feature selection method for airport energy efficiency optimization according to claim 1, characterized in that, S2 includes the following steps: S21: From the training dataset Get the current batch of data Using an automatic encoder Batch Mapping to the latent space yields a latent representation of the encoder network. ; S22: Using a decoder Receive potential representation And map it back to the original space to obtain the reconstructed original data. , where the decoder The definition is as follows: ； in, The activation function for the hidden layer. and These represent the weight matrices of the first hidden layer and the output layer, respectively. and It is a bias term; S23: Calculate the initial feature map matrix using the mean squared error combined with the regularization loss function. and the data reconstructed by the decoder The differences between them are expressed as follows: ； in, It is the Frobenius norm, which represents the sum of the squared differences of all elements in the matrix and is used to measure the reconstruction error; S24: After the model training is complete, extract the weight vector from the one-to-one feature selection layer, normalize the weight vector, and the resulting value is the importance of the deep features. .
4. The deep evolutionary multi-objective feature selection method for airport energy efficiency optimization according to claim 1, characterized in that, S3 includes the following steps: S31: In the current population In this study, non-dominated ranking was used to evaluate each individual. Quality, defining individual importance The number of other individuals that this individual dominates in the current population: ； in Represents an individual Dominant Individual , The larger the value, the higher the quality of the feature subset represented by the individual; S32: Utilize the normalized deep feature importance vector obtained in step S24 The individual importance obtained from step S31 Construct a depth-guided matrix SM, the matrix of which is the first... Line 1 Column elements Indicates the first The feature in the first The guiding probability for each individual is calculated using the following formula: ； in, It is a normalization function that maps numerical values to the interval [0, 1].
5. The deep evolutionary multi-objective feature selection method for airport energy efficiency optimization according to claim 1, characterized in that, S4 includes the following steps: S41: Perform dimensionality reduction on non-elite individuals in the population. The Given a feature, calculate the probability that it flips to the unselected state 0. : ； in, Let the current iteration algebra be... To control the coefficient of decay rate, For the first The mean of all characteristic guide values for an individual; S42: Use elite individuals to repair the population; randomly select an elite individual from the non-dominated solution set. ,like If this trait is selected among elite individuals, then the individual will be forcibly... The feature is restored to 1; if Then, based on probability Learning from elite individuals, probability Defined as: ； S43: Merge the population after operations S41 and S42 with the parent population, perform non-dominated sorting and crowding distance calculation, and select the next generation population. ; S44: Dynamically adjust the SM matrix based on feedback from population evolution, compare the average classification error rate of the offspring elite set with that of the parent elite set. If the offspring error rate decreases, update the guiding matrix using the following formula: ； in The adjustment step size is determined based on the changes in the characteristic state. If the error rate of the offspring does not decrease, it indicates that the current guiding matrix is invalid. In this case, the guiding matrix is re-initialized using the S3 method, and finally, a termination judgment is made to end the process.
6. A deep evolutionary multi-objective feature selection system for airport energy efficiency optimization, characterized in that, The system comprising the steps of applying the method according to any one of claims 1 to 5, wherein the system includes: The hierarchical weighted initialization data construction module is configured to perform the following process: construct hierarchical weighted initialization data using mutual information and K-Means clustering; The autoencoder module is configured to perform the following processes: training a deep autoencoder model and extracting the importance of deep features that reflect the nonlinear relationships between features by analyzing the weights of the model's input layer; The deep guidance matrix module is configured to perform the following process: using the importance of the deep features as prior knowledge, constructing a deep guidance matrix to guide the evolution of the population; The adaptive mechanism module is configured to perform the following process: perform evolutionary search based on the guidance matrix, and dynamically update the guidance matrix using an adaptive mechanism until the termination condition is met and the optimal solution set is output.
7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed, it implements the steps of the method as described in any one of claims 1 to 5.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program is configured to implement the steps of the method according to any one of claims 1 to 5 when invoked by a processor.
9. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 5.