Meridian conditioning control system based on artificial intelligence

Abstract

The invention relates to the technical field of artificial intelligence medical treatment, in particular to an artificial intelligence-based meridian conditioning control system, which comprises a meridian pulse condition acquisition module, a meridian dialectical analysis module, a meridian treatment intervention module and a conditioning parameter optimization module, according to the scheme, optimization of feature extraction is carried out by using heuristic testing in combination with an ant optimization algorithm in the meridian differentiation analysis module, K-means clustering and mutual information are combined, the dimensionality of a feature space is reduced, the correlation between a selected feature subset and a target variable can be maximized by combining the mutual information and a greedy strategy, and the accuracy of feature extraction is improved. The mode recognition capability of the meridian differentiation model is improved; a physiological signal association graph is constructed, local feature aggregation is performed by using a graph convolutional network to mine complex association among different physiological signals, the feature distribution of similar meridian states is more compact in combination with comparative learning, and a control system which can still maintain high identification degree, high adaptability and stable diagnosis capability in a complex physiological signal environment is provided.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence medical technology, and specifically to a meridian conditioning control system based on artificial intelligence. Background Art

[0002] In recent years, artificial intelligence technology has been widely used in the field of medical health, especially in pattern recognition, intelligent recommendation and optimization algorithms. In the existing meridian conditioning process, important features are difficult to extract effectively. There is redundant information between many physiological signal data, especially heart rate and heart rate variability, which leads to local optimality and lacks global optimization capabilities, resulting in the system being unable to fully explore potential efficient feature combinations. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a meridian conditioning control system based on artificial intelligence. In view of the problem that important features are difficult to be effectively extracted in the existing meridian conditioning process, the present invention uses heuristic testing combined with ant optimization algorithm to optimize feature extraction in the meridian syndrome differentiation analysis module, combines K-means clustering and mutual information, reduces the dimension of the feature space, and enhances the global search capability. In addition, through the combination of mutual information and greedy strategy, the selected feature subset can maximize the correlation with the target variable, improve the pattern recognition ability of the meridian syndrome differentiation model, and make the final diagnosis result more accurate; by constructing a physiological signal association graph, using a graph convolutional network to perform local feature aggregation to explore the complex correlation between different physiological signals, combined with comparative learning, the feature distribution of similar meridian states is made closer, while the features of different states remain distinguished, ensuring that the system can still maintain high recognition, high adaptability and stable diagnostic capabilities in a complex physiological signal environment.

[0004] The technical solution adopted by the present invention is as follows: The meridian conditioning control system based on artificial intelligence provided by the present invention includes a meridian pulse acquisition module, a meridian syndrome differentiation analysis module, a meridian treatment intervention module and a conditioning parameter optimization module, which specifically includes the following contents:

[0005] The meridian pulse condition acquisition module acquires the user's physiological signals during the meridian adjustment process, including pulse condition, meridian potential, skin resistance, heart rate, blood oxygen, skin temperature and infrared thermal imaging data;

[0006] The meridian syndrome differentiation and analysis module establishes a meridian syndrome differentiation model, performs pattern recognition on the user's physiological signal data, and obtains a meridian diagnosis report;

[0007] The meridian treatment intervention module recommends personalized intervention plans based on the meridian diagnosis report through a neural collaborative filtering recommendation algorithm;

[0008] The conditioning parameter optimization module collects the user's physiological feedback data and uses a genetic algorithm to optimize the personalized intervention plan.

[0009] Furthermore, the meridian syndrome differentiation and analysis module specifically includes the following steps:

[0010] Step S1: signal collection, receiving the user's physiological signals collected by the meridian pulse acquisition module;

[0011] Step S2: signal preprocessing, preprocessing the user's physiological signals, including denoising, standardization and outlier detection;

[0012] Step S3: feature extraction, extracting physiological features from the pre-processed user's physiological signals, including pulse features, meridian potential features, skin resistance features and heart rate variability features, and the extracted physiological features are used as an input feature set of step S4;

[0013] Step S4: Optimizing feature selection, using heuristic testing combined with ant optimization algorithm to optimize feature extraction, and obtaining the optimal feature subset according to the input feature set;

[0014] Step S5: Establishing a meridian syndrome differentiation model, and using the optimal feature subset to train the meridian syndrome differentiation model;

[0015] Step S6: pattern recognition, using the trained meridian dialectical model to perform pattern recognition.

[0016] Further, step S4 uses heuristic testing combined with ant optimization algorithm to optimize feature extraction, which specifically includes the following steps:

[0017] Step S41: Initialization: setting a feature search space according to the input feature set, including discrete features and continuous features, initializing the number of ants, the maximum number of iterations, the feature pheromone matrix, the pheromone volatility and the reinforcement factor, and setting the initial number of clusters of K-means clustering for the continuous features;

[0018] Step S42: Initialization of feature subsets. Each ant randomly selects an initial feature subset in the feature search space, calculates the initial fitness, and uses the CMI measure to evaluate the quality of the initial feature subset. The formula used is as follows: ;

[0019] in, represents the initial feature subset, is the initial fitness value, is the K-th cluster formed by K-means clustering, for The probability of is the Kth cluster and the target variable The mutual information between

[0020] Step S43: Heuristic measurement, set the target variable as the variable that is ultimately expected to be classified, use K-means clustering to organize continuous features into clusters, and calculate the mutual information between the clusters and the target variable. The formula used is as follows: ;

[0021] in, Features With the target variable The mutual information of yes and The joint probability distribution of and They are and The marginal probability distribution of ;

[0022] Step S44: Pheromone update, calculate the fitness of the feature subset selected by each ant, the formula used is as follows: ;

[0023] in, Features The pheromone concentration, Features The initial pheromone concentration, is the pheromone volatilization rate, is the strengthening factor, is the number of ants, It is The fitness value of each ant;

[0024] Step S45: Normalization is performed based on pheromones, and a probability selection strategy based on average pheromone concentration is adopted. The formula used is as follows: ;

[0025] in, is the normalized probability value, is the total number of features, Features pheromone concentration;

[0026] Step S46: Local optimization, using greedy strategy to adjust feature subsets and calculate the optimized fitness ,when When , update the feature subset;

[0027] Step S47: Exploratory foraging, some ants randomly explore unselected features, use adaptive K to select local perturbations, and calculate new fitness ,like , then replace it to get the current optimal feature subset;

[0028] Step S48: Global pheromone update, calculate the fitness of the current optimal feature subset, and use enhanced pheromone for update. The formula used is as follows: ;

[0029] in, For the updated pheromone, is the current optimal feature subset;

[0030] Step S49: feature subset output, repeat steps S42 to S48 until the maximum number of iterations is reached, and then output the optimal feature subset.

[0031] Further, step S5 specifically includes the following steps:

[0032] Step S51: Initialize the meridian dialectical model, construct an input feature matrix based on the optimal feature subset obtained in step S4 as the input of the meridian dialectical model, and initialize the meridian dialectical model including a graph convolution layer, an attention mechanism, and a contrastive learning module;

[0033] Step S52: feature representation learning, constructing a physiological signal association graph according to the input feature matrix, defining an adjacency matrix, and using a graph convolutional network to perform local information aggregation on each input feature in the physiological signal association graph according to the adjacency matrix, using a graph attention mechanism to adjust the weight of each feature, and combining contrastive learning to optimize feature distribution to obtain enhanced feature representation;

[0034] Step S53: constructing a meridian dialectical model, using metric learning to map the enhanced feature representation to a high-dimensional embedding space, similar meridian states are close together, and vice versa, a clustering algorithm is used to group similar meridian states to form a clustered dialectical model;

[0035] Step S54: construct a meridian diagnosis classifier, introduce a classifier based on an attention mechanism, use cross entropy loss to learn the mapping relationship between meridian states and diagnostic categories, use a dynamic loss weighting mechanism to balance the distribution of samples of different categories, and obtain a trained meridian diagnosis model;

[0036] Step S55: output the trained meridian diagnosis model for pattern recognition.

[0037] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0038] (1) In view of the problem that important features are difficult to be effectively extracted in the existing meridian conditioning process, the present invention uses heuristic testing combined with ant optimization algorithm to optimize feature extraction in the meridian syndrome differentiation analysis module, combines K-means clustering and mutual information, reduces the dimension of the feature space, and enhances the global search capability. In addition, through the combination of mutual information and greedy strategy, the selected feature subset can maximize the correlation with the target variable, improve the pattern recognition ability of the meridian syndrome differentiation model, and make the final diagnosis result more accurate;

[0039] (2) By constructing a physiological signal association graph and using a graph convolutional network to perform local feature aggregation, the complex associations between different physiological signals can be explored. Combined with comparative learning, the feature distribution of similar meridian states can be made closer, while the features of different states remain distinct, ensuring that the system can still maintain high recognition, high adaptability and stable diagnostic capabilities in a complex physiological signal environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A schematic diagram of the meridian conditioning control system based on artificial intelligence proposed by the present invention;

[0041] Figure 2 The figure is a flow chart of establishing the meridian dialectical model proposed by the present invention;

[0042] Figure 3 This is a schematic flow chart of step S4 proposed by the present invention;

[0043] Figure 4 This is a schematic flow chart of step S5 proposed by the present invention.

[0044] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0045] 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 only part of the embodiments of the present invention, rather than all 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.

[0046] Example 1, see Figure 1 The meridian conditioning control system based on artificial intelligence provided by the present invention includes a meridian pulse acquisition module, a meridian syndrome differentiation and analysis module, a meridian treatment intervention module and a conditioning parameter optimization module, which specifically includes the following contents:

[0047] The meridian pulse condition acquisition module acquires the user's physiological signals during the meridian adjustment process, including pulse condition, meridian potential, skin resistance, heart rate, blood oxygen, skin temperature and infrared thermal imaging data;

[0048] The meridian syndrome differentiation and analysis module establishes a meridian syndrome differentiation model, performs pattern recognition on the user's physiological signal data, and obtains a meridian diagnosis report;

[0049] The meridian treatment intervention module recommends personalized intervention plans based on the meridian diagnosis report through a neural collaborative filtering recommendation algorithm;

[0050] The conditioning parameter optimization module collects the user's physiological feedback data and uses a genetic algorithm to optimize the personalized intervention plan.

[0051] Example 2, see Figure 2 This embodiment is based on the above embodiment, and the meridian syndrome differentiation analysis module specifically includes the following steps:

[0052] Step S1: signal collection, receiving the user's physiological signals collected by the meridian pulse acquisition module;

[0053] Step S2: signal preprocessing, preprocessing the user's physiological signals, including denoising, standardization and outlier detection;

[0054] Step S3: feature extraction, extracting physiological features from the pre-processed user's physiological signals, including pulse features, meridian potential features, skin resistance features and heart rate variability features, and the extracted physiological features are used as an input feature set of step S4;

[0055] Step S4: Optimizing feature selection, using heuristic testing combined with ant optimization algorithm to optimize feature extraction, and obtaining the optimal feature subset according to the input feature set;

[0056] Step S5: Establishing a meridian syndrome differentiation model, and using the optimal feature subset to train the meridian syndrome differentiation model;

[0057] Step S6: pattern recognition, using the trained meridian dialectical model to perform pattern recognition.

[0058] Embodiment 3, based on Figure 3 This embodiment is based on the above embodiment. Step S4 uses a heuristic test combined with an ant optimization algorithm to optimize feature extraction, specifically including the following steps:

[0059] Step S41: Initialization: setting a feature search space according to the input feature set, including discrete features and continuous features, initializing the number of ants, the maximum number of iterations, the feature pheromone matrix, the pheromone volatility and the reinforcement factor, and setting the initial number of clusters of K-means clustering for the continuous features;

[0060] Step S42: Initialization of feature subsets. Each ant randomly selects an initial feature subset in the feature search space, calculates the initial fitness, and uses the CMI measure to evaluate the quality of the initial feature subset. The formula used is as follows: ;

[0061] in, represents the initial feature subset, is the initial fitness value, is the K-th cluster formed by K-means clustering, for The probability of is the Kth cluster and the target variable The mutual information between

[0062] Step S43: Heuristic measurement, set the target variable as the variable that is ultimately expected to be classified, use K-means clustering to organize continuous features into clusters, and calculate the mutual information between the clusters and the target variable. The formula used is as follows:

[0063] ;

[0064] in, Features With the target variable The mutual information of yes and The joint probability distribution of and They are and The marginal probability distribution of ;

[0065] Step S44: Pheromone update, calculate the fitness of the feature subset selected by each ant, the formula used is as follows: ;

[0066] in, Features The pheromone concentration, Features The initial pheromone concentration, is the pheromone volatilization rate, is the strengthening factor, is the number of ants, It is The fitness value of each ant;

[0067] Step S45: Normalization is performed based on pheromones, and a probability selection strategy based on average pheromone concentration is adopted. The formula used is as follows: ;

[0068] in, is the normalized probability value, is the total number of features, Features pheromone concentration;

[0069] Step S46: Local optimization, using greedy strategy to adjust feature subsets and calculate the optimized fitness ,when When , update the feature subset;

[0070] Step S47: Exploratory foraging, some ants randomly explore unselected features, use adaptive K to select local perturbations, and calculate new fitness ,like , then replace it to get the current optimal feature subset;

[0071] Step S48: Global pheromone update, calculate the fitness of the current optimal feature subset, and use enhanced pheromone for update. The formula used is as follows: ;

[0072] in, For the updated pheromone, is the current optimal feature subset;

[0073] Step S49: feature subset output, repeat steps S42 to S48 until the maximum number of iterations is reached, and then output the optimal feature subset.

[0074] Embodiment 4, this embodiment is based on the above embodiment. In this embodiment, the input feature set is set to include 4 features: pulse feature, meridian potential feature, skin resistance feature, heart rate variability feature, the number of ants is set to 20, the maximum number of iterations is set to 100, the initial pheromone value is set to 0.1, the pheromone volatility is set to 0.1, the reinforcement factor is set to 1, and the initial number of clusters of K-means clustering is set to 3;

[0075] Ant 1 selected pulse characteristics and heart rate variability characteristics. For the feature subset selected by Ant 1 (pulse characteristics and heart rate variability characteristics), its fitness was calculated to be 0.42;

[0076] K-means clustering is used to organize continuous features into three clusters. , the mutual information with the target variable is calculated to be 0.6;

[0077] In each iteration, the "pheromone" concentration of each feature is updated. Pheromones represent the quality of features. Better features will leave more pheromones. Suppose there are two ants, and the ants choose the feature subsets (pulse features and heart rate variability features, with a fitness of 0.42) and (meridian potential features and skin resistance features, with a fitness of 0.35). Then, the pheromone update will be performed as follows:

[0078] For pulse characteristics: pheromone is updated to: ;

[0079] For feature 2 (meridian potential feature), the pheromone is updated as follows: ;

[0080] For pheromone-based normalization, if there are four features, the total information concentration is: , , , , the total pheromone concentration is 2.23, and the normalized probability value is: ; ; ; ;

[0081] Assuming that the currently selected feature subset is (pulse feature and heart rate variability feature), the fitness is 0.42, perform greedy optimization, try to replace the heart rate variability feature with the skin resistance feature, and calculate the new fitness:

[0082] The new feature subset is (pulse features and skin resistance features);

[0083] After optimization, the new fitness is =0.48;

[0084] Therefore, the updated feature subsets are pulse features and skin resistance features;

[0085] Ant 1 has selected the pulse feature and the skin resistance feature. The ant is allowed to randomly select one of the unselected features and select the heart rate variability feature. The new feature subset is the pulse feature, the skin resistance feature, and the heart rate variability feature. After the disturbance, the new fitness is ;

[0086] The current optimal feature subset is changed to pulse feature, skin resistance feature and heart rate variability feature;

[0087] The current optimal feature subset is pulse feature, skin resistance feature and heart rate variability feature, with a fitness of 0.5. Using the enhanced pheromone update formula, the pheromone after sealing is 1.0715;

[0088] When 100 iterations are completed, the optimal feature subset is output.

[0089] Embodiment 5, based on Figure 4 This embodiment is based on the above embodiment, and step S5 specifically includes the following steps:

[0090] Step S51: Initialize the meridian dialectical model, construct an input feature matrix based on the optimal feature subset obtained in step S4 as the input of the meridian dialectical model, and initialize the meridian dialectical model including a graph convolution layer, an attention mechanism, and a contrastive learning module;

[0091] Step S52: feature representation learning, constructing a physiological signal association graph according to the input feature matrix, defining an adjacency matrix, and using a graph convolutional network to perform local information aggregation on each input feature in the physiological signal association graph according to the adjacency matrix, using a graph attention mechanism to adjust the weight of each feature, and combining contrastive learning to optimize feature distribution to obtain enhanced feature representation;

[0092] Step S53: constructing a meridian dialectical model, using metric learning to map the enhanced feature representation to a high-dimensional embedding space, similar meridian states are close together, and vice versa, a clustering algorithm is used to group similar meridian states to form a clustered dialectical model;

[0093] Step S54: construct a meridian diagnosis classifier, introduce a classifier based on an attention mechanism, use cross entropy loss to learn the mapping relationship between meridian states and diagnostic categories, use a dynamic loss weighting mechanism to balance the distribution of samples of different categories, and obtain a trained meridian diagnosis model;

[0094] Step S55: output the trained meridian diagnosis model for pattern recognition.

[0095] In this embodiment, the codes used are as follows: import torch import torch.nn as nn import torch.optim as optim import torch.nn.functional as F import numpy as np from sklearn.cluster import KMeans from torch_geometric.nn import GCNConv from sklearn.model_selection import train_test_split from sklearn.metrics import accuracy_score # --------------- Step S51: Initialize the meridian dialectical model--------------- class MeridianDiagnosisModel(nn.Module): def __init__(self, in_features, hidden_dim, out_features): super(MeridianDiagnosisModel, self).__init__() self.gcn1 = GCNConv(in_features, hidden_dim) self.gcn2 = GCNConv(hidden_dim, hidden_dim) self.attention = nn.Linear(hidden_dim, 1) # Attention mechanism self.contrastive_loss = nn.CosineEmbeddingLoss() # Contrastive learning loss self.fc = nn.Linear(hidden_dim, out_features) # classification layer def forward(self, x, edge_index): x = F.relu(self.gcn1(x, edge_index)) x = self.gcn2(x, edge_index) attention_weights = torch.sigmoid(self.attention(x)) x = x * attention_weights # weighting return x # --------------- Step S52: Feature Representation Learning --------------- def feature_representation_learning(features, adjacency_matrix): edge_index = adjacency_matrix.nonzero(as_tuple=True) model = MeridianDiagnosisModel(in_features=features.shape[1],hidden_dim=64, out_features=32) optimizer = optim.Adam(model.parameters(), lr=0.01) for epoch in range(100): optimizer.zero_grad() enhanced_features = model(features, edge_index) loss = model.contrastive_loss(enhanced_features, features,torch.ones(features.shape[0])) loss.backward() optimizer.step() return enhanced_features.detach().numpy() # --------------- Step S53: Build Meridian Dialectical Modes --------------- def build_meridian_modes(enhanced_features, num_clusters=4): kmeans = KMeans(n_clusters=num_clusters, random_state=42) clusters = kmeans.fit_predict(enhanced_features) return clusters # --------------- Step S54: Construct the Meridian Diagnosis Classifier --------------- class DiagnosisClassifier(nn.Module): def __init__(self, input_dim, num_classes): super(DiagnosisClassifier, self).__init__() self.fc1 = nn.Linear(input_dim, 64) self.attention = nn.Linear(64, 1) # Attention mechanism self.fc2 = nn.Linear(64, num_classes) def forward(self, x): x = F.relu(self.fc1(x)) attention_weights = torch.sigmoid(self.attention(x)) x = x * attention_weights return F.log_softmax(self.fc2(x), dim=1) def train_diagnosis_classifier(features, labels, num_classes): X_train, X_test, y_train, y_test = train_test_split(features,labels, test_size=0.2, random_state=42) X_train, X_test = torch.tensor(X_train, dtype=torch.float32),torch.tensor(X_test, dtype=torch.float32) y_train, y_test = torch.tensor(y_train, dtype=torch.long), torch.tensor(y_test, dtype=torch.long) model = DiagnosisClassifier(input_dim=features.shape[1], num_classes=num_classes) optimizer = optim.Adam(model.parameters(), lr=0.01) loss_fn = nn.CrossEntropyLoss() for epoch in range(100): optimizer.zero_grad() output = model(X_train) loss = loss_fn(output, y_train) loss.backward() optimizer.step() y_pred = model(X_test).argmax(dim=1).numpy() accuracy = accuracy_score(y_test, y_pred) return model, accuracy # --------------- Step S55: Output the trained model --------------- def main(): num_samples, num_features = 1000, 10 raw_features = np.random.rand(num_samples, num_features) # Simulate physiological signals adjacency_matrix = np.random.randint(0, 2, (num_samples, num_samples)) # Random adjacency matrix enhanced_features = feature_representation_learning(torch.tensor(raw_features, dtype=torch.float32), adjacency_matrix) clusters = build_meridian_modes(enhanced_features) model, accuracy = train_diagnosis_classifier(enhanced_features,clusters, num_classes=4) print(f"Classification model training completed, test accuracy: {accuracy:.2f}") if __name__ == "__main__": main().

[0096] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0097] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0098] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. The meridian conditioning control system based on artificial intelligence is characterized by: It includes meridian pulse acquisition module, meridian syndrome differentiation analysis module, meridian treatment intervention module and conditioning parameter optimization module, which specifically include the following contents: The meridian pulse acquisition module acquires the physiological signals of the user during the meridian adjustment process; The meridian syndrome differentiation and analysis module establishes a meridian syndrome differentiation model, performs pattern recognition on the user's physiological signal data, and obtains a meridian diagnosis report; The meridian treatment intervention module recommends personalized intervention plans based on the meridian diagnosis report through a neural collaborative filtering recommendation algorithm; The conditioning parameter optimization module collects the user's physiological feedback data and uses a genetic algorithm to optimize the personalized intervention plan.

2. The meridian conditioning control system based on artificial intelligence according to claim 1 is characterized in that: The meridian syndrome differentiation and analysis module specifically comprises the following steps: Step S1: signal collection, receiving the user's physiological signals collected by the meridian pulse acquisition module; Step S2: signal preprocessing, preprocessing the user's physiological signals; Step S3: feature extraction, extracting physiological features from the pre-processed physiological signals of the user, and the extracted physiological features are used as the input feature set of step S4; Step S4: Optimizing feature selection, using heuristic testing combined with ant optimization algorithm to optimize feature extraction, and obtaining the optimal feature subset according to the input feature set; Step S5: Establishing a meridian syndrome differentiation model, and using the optimal feature subset to train the meridian syndrome differentiation model; Step S6: pattern recognition, using the trained meridian dialectical model to perform pattern recognition.

3. The meridian conditioning control system based on artificial intelligence according to claim 2 is characterized in that: Step S4, using heuristic test combined with ant optimization algorithm to optimize feature extraction, specifically includes the following steps: Step S41: Initialization: setting a feature search space according to the input feature set, including discrete features and continuous features, initializing the number of ants, the maximum number of iterations, the feature pheromone matrix, the pheromone volatility and the reinforcement factor, and setting the initial number of clusters of K-means clustering for the continuous features; Step S42: feature subset initialization, each ant randomly selects an initial feature subset in the feature search space, calculates the initial fitness, and uses the CMI measure to evaluate the quality of the initial feature subset; Step S43: Heuristic measurement, setting the target variable as the variable that is ultimately desired to be classified, using K-means clustering to organize continuous features into clusters, and calculating the mutual information between the clusters and the target variable; Step S44: Pheromone update, calculating the fitness of the feature subset selected by each ant; Step S45: performing normalization based on pheromone, adopting a probability selection strategy based on average pheromone concentration, and obtaining a normalized probability value; Step S46: local optimization, using a greedy strategy to adjust the feature subset, calculating the optimized fitness, and updating the feature subset; Step S47: Exploratory foraging, some ants randomly explore unselected features, use adaptive K to select local perturbations, calculate new fitness, perform replacement, and obtain the current optimal feature subset; Step S48: global pheromone update, calculate the fitness of the current optimal feature subset, and use enhanced pheromone for update; Step S49: feature subset output, repeat steps S42 to S48 until the maximum number of iterations is reached, and then output the optimal feature subset.

Images