Intelligent residential area group classification method fused with Transform model
Through the intelligent classification method of residential groups integrating the Transformer model, the problem that existing technology is difficult to adapt to the characteristics of residential areas in different regions is solved, efficient feature dimensionality reduction and accurate classification are achieved, and the accuracy and calculation efficiency of the analysis results are improved.
Patent Information
- Application Number
- CN202510094681.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-21
- Publication Date
- 2025-06-17
AI Technical Summary
The existing residential clustering and classification methods are difficult to fully adapt and accurately analyze residential characteristics of different geographical regions, cultural backgrounds and development levels, and feature extraction and model training face challenges when processing high-dimensional residential data.
The intelligent classification method of residential group with fusion Transformer model is adopted. By acquiring and preprocessing residential vector data, clustering and feature extraction are performed, and dimensionality reduction and classification are performed by combining multiple clustering algorithms and feature expression systems. Finally, the improved Transformer model is used for training and classification.
It improves the accuracy and reliability of the analysis results, can better adapt to the characteristics of residential areas in different regions, achieve efficient feature dimensionality reduction and accurate classification, and improves computing efficiency and model performance.
Smart Images
Figure CN120162689A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical fields of geographic information systems and deep learning, and particularly relates to an intelligent classification method for residential area groups integrating a Transformer model. Background Art
[0002] Most of the existing residential area clustering and classification methods are limited within the application framework of a single algorithm. This limitation makes it difficult for them to fully adapt to and accurately analyze the characteristics of residential areas in different geographical regions, different cultural backgrounds, and different development levels. Due to the differences in social and economic structures, natural environments, and historical cultures in each region, a single clustering algorithm often fails to capture these complex and variable residential area characteristics, thus severely affecting the accuracy and reliability of the analysis results. In addition, the existing feature extraction and classification models also face many challenges when processing residential area data. On the one hand, as a complex geographical entity, a residential area group contains rich spatial structures and attribute information, and the effective extraction and expression of this information are crucial for accurate classification. However, the existing feature extraction methods often fail to fully capture these geographical characteristics, resulting in biases in the classification model when identifying residential area types. On the other hand, with the advent of the big data era, the dimensions and scales of residential area data are constantly increasing, which brings greater difficulties to feature extraction and model training. How to effectively perform feature dimensionality reduction in the high-dimensional data space to improve the computational efficiency and classification performance of the model is also an urgent problem to be solved currently. Summary of the Invention
[0003] To solve the above technical problems, the present invention proposes an intelligent classification method for residential area groups integrating a Transformer model, which improves the accuracy and reliability of the analysis results and can better adapt to the residential area characteristics of different regions.
[0004] To achieve the above object, the present invention provides an intelligent classification method for residential area groups integrating a Transformer model, including:
[0005] Obtain a residential area vector data set, preprocess the residential area vector data set to obtain preprocessed residential area vector data;
[0006] Cluster the preprocessed residential area vector data into groups to obtain clustered residential area vector data;
[0007] Extract features from the clustered residential area vector data according to preset features, and reduce the dimensions of the extracted features to obtain an optimized feature vector matrix;
[0008] Classify the optimized feature vector matrix to obtain a classification result;
[0009] Input the optimized feature vector matrix and the classification result into the Transformer model for training to obtain a trained Transformer model;
[0010] Obtain the vector data of residential areas in the area to be measured, input it into the trained Transformer model, and obtain the intelligent classification result of residential area groups.
[0011] Optionally, preprocess the vector dataset of residential areas. The preprocessed vector data of residential areas obtained includes:
[0012] Based on the vector dataset of residential areas, check the projection coordinate system type of the vector dataset of residential areas, and identify and remove the outliers in the vector dataset of residential areas;
[0013] Fill in the missing parts in the data to obtain complete vector data of residential areas;
[0014] Perform standardization processing on the complete vector data of residential areas to obtain preprocessed vector data of residential areas.
[0015] Optionally, the vector data of residential areas clustered into groups includes:
[0016] Based on the preprocessed vector data of residential areas, select different clustering methods for different types of vector data of residential areas, and cluster the preprocessed vector data of residential areas into groups to obtain vector data of residential areas clustered into groups.
[0017] Optionally, obtaining the optimized feature vector matrix includes:
[0018] Based on the vector data of residential areas clustered into groups, extract the geometric and spatial features of residential area groups to obtain a set of feature vectors;
[0019] Reduce the dimension of the set of feature vectors to obtain an optimized feature vector matrix.
[0020] Optionally, the preset features include: convex hull area, building density, ratio of building area to convex hull area, coefficient of variation of Thiessen polygon, distance between residential buildings, number of sides of residential buildings, convex hull direction, PCA principal axis direction, area of minimum bounding rectangle, ratio of building area to area of minimum bounding rectangle, angle between major axis of circumscribed ellipse and due north direction, perimeter-area ratio.
[0021] Optionally, classifying the optimized feature vector matrix to obtain the classification result includes:
[0022] Classify according to the distribution characteristics of the data in the optimized feature vector matrix;
[0023] Divide the classified data into equal - area regions to obtain the initial classification result;
[0024] Align the initial classification result with the residential area vector dataset to obtain the classification result.
[0025] Optionally, the classification result includes small - scale evacuation class, small - scale compact class, medium - scale evacuation class, medium - scale compact class, large - scale evacuation class, and large - scale compact class.
[0026] Optionally, obtaining the trained Transformer model includes:
[0027] Divide the optimized feature vector matrix and the classification result into a training set and a test set;
[0028] Input the training set into the Transformer model for training to obtain an untested Transformer model;
[0029] Use the test set to test the untested Transformer model, evaluate the performance of the Transformer model, and adjust the Transformer model to obtain the trained Transformer model.
[0030] Technical effects of the present invention:
[0031] (1) By integrating multiple algorithms such as K - means, DBSCAN, CDC local center clustering, and MSGC multi - scale grid clustering, the present invention can intelligently select the optimal clustering algorithm according to the distribution characteristics of different residential areas. When facing residential areas with diverse shapes and uneven density distributions, the present invention can flexibly respond and achieve accurate division. This flexibility not only improves the accuracy of clustering but also lays a solid foundation for subsequent analysis work.
[0032] (2) The present invention designs a feature expression system including 12 feature vectors such as convex hull area, building density, and the ratio of building area to convex hull area. These feature vectors comprehensively cover the geometric shape and spatial characteristics of residential areas and can accurately reflect key information such as the compactness of building groups and the distribution law of residential area groups. This high - precision feature extraction system provides rich and high - quality data support for subsequent dimensionality reduction and classification, ensuring the accuracy and reliability of the analysis results.
[0033] (3) When dealing with high-dimensional data, dimensionality reduction is a key step in improving computational efficiency and optimizing model performance in the present invention. The present invention uses methods such as principal component analysis (PCA) and correlation calculation to select 6 eigenvectors from 12 features that can best represent the geographical features of residential area groups. This process not only retains the maximum variance of the feature information, improves the computational efficiency, but also reduces redundant information and optimizes the performance of the classification model. Through efficient dimensionality reduction, while ensuring the classification accuracy and generalization ability of the model, the present invention reduces the computational complexity and improves the overall performance.
[0034] (4) The present invention adopts an improved Transformer model, adds an average pooling module after the residual connection and layer normalization of the Transformer model, and combines the natural break method and the equal area division method to accurately classify residential area groups into six categories. This innovative classification model can not only complete the classification task quickly and efficiently, but also improve the refinement degree of the classification results through the feature optimization algorithm. This innovative classification method not only improves the prediction accuracy, but also provides strong support for applications such as urban and rural planning and land use.
[0035] (5) The present invention proposes a complete process from clustering analysis to feature extraction, dimensionality reduction, classification and detection, with comprehensive and strong versatility. This system not only is compatible with different types of residential area distribution characteristics, but also has good portability and scalability. Therefore, it can be widely applied to multiple fields such as geographic information system (GIS), urban and rural planning, resource optimization allocation, and disaster monitoring. This comprehensiveness and versatility make the present invention have broad practical value and social significance.
[0036] (6) The present invention also performs excellently in terms of automation degree. By automatically selecting the optimal clustering algorithm, automatically reducing dimensions, and an automatic classification method based on the improved Transformer, the present invention greatly reduces manual intervention and improves the processing efficiency. This highly automated processing method not only improves the simplicity of operation, but also ensures the consistency of the results. This advantage makes the present invention an efficient, stable and easy-to-use residential area analysis tool, providing great convenience for users. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] The drawings forming a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation of this application. In the drawings:
[0038] Figure 1 It is a schematic flowchart of an intelligent classification method for residential area groups integrating a Transformer model according to an embodiment of the present invention;
[0039] Figure 2Schematic diagram of the vector data set in the embodiment of the present invention;
[0040] Figure 3 Schematic diagram of four clustering results in the embodiment of the present invention;
[0041] Figure 4 Schematic diagram of the correlation matrix of the feature vectors after dimensionality reduction in the embodiment of the present invention;
[0042] Figure 5 Schematic diagram of the classification result of residential area groups in the embodiment of the present invention;
[0043] Figure 6 Schematic diagram of the confusion matrix for evaluating the classification accuracy of the Transformer model in the embodiment of the present invention. Detailed implementation manners
[0044] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0045] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.
[0046] As Figure 1 - Figure 2 shown, in this embodiment, an intelligent classification method for residential area groups integrating a Transformer model is provided, including:
[0047] Obtain a residential area vector data set, preprocess the residential area vector data set to obtain preprocessed residential area vector data;
[0048] Cluster the preprocessed residential area vector data into groups to obtain clustered residential area vector data;
[0049] Extract features from the clustered residential area vector data according to preset features, and perform dimensionality reduction on the extracted features to obtain an optimized feature vector matrix;
[0050] Classify the optimized feature vector matrix to obtain a classification result;
[0051] Input the optimized feature vector matrix and the classification result into a Transformer model for training to obtain a trained Transformer model;
[0052] Obtain the residential area vector data of the area to be measured, input it into the trained Transformer model, and obtain the intelligent classification result of the residential area groups.
[0053] Further, preprocess the vector dataset of populated areas to obtain the preprocessed vector data of populated areas, including:
[0054] Based on the vector dataset of populated areas, check the projection coordinate system type of the vector dataset of populated areas, and identify and remove outliers in the vector dataset of populated areas;
[0055] Fill in the missing parts in the data to obtain complete vector data of populated areas;
[0056] Standardize the complete vector data of populated areas to obtain the preprocessed vector data of populated areas.
[0057] Specifically, to ensure data accuracy, check its projection coordinate system type and use statistical methods and data visualization tools to identify and remove outliers in the dataset. These outliers may be caused by measurement errors, data entry errors, etc. For the missing parts in the data, interpolation methods (such as linear interpolation, polynomial interpolation, etc.) or filling methods based on adjacent data are used for processing to ensure data integrity. Since different features may have different dimensions and value ranges, standardization methods (such as Z-score standardization, min-max standardization, etc.) are used to convert the data into a standard vector dataset D' with a unified dimension and value range.
[0058] Further, the obtained vector data of populated areas clustered into groups includes:
[0059] Based on the preprocessed vector data of populated areas, select different clustering methods for different types of vector data of populated areas, and cluster the preprocessed vector data of populated areas into groups to obtain the vector data of populated areas clustered into groups.
[0060] Specifically, step S1: Select the clustering algorithm with the best effect from the K-means algorithm, DBSCAN algorithm, CDC algorithm, and MSGC algorithm for the preprocessed standard vector dataset D';
[0061] Step S1.1: When the distribution of populated area groups is regular and the shapes are similar, select the K-means algorithm for clustering. This algorithm iteratively optimizes the objective function to divide the dataset into K clusters, making the data points within the same cluster as similar as possible and the data points between different clusters as different as possible. In this embodiment, the value of K is 10.
[0062] Step S1.2: For populated area groups with uneven density and irregular shapes, use the DBSCAN algorithm. In this embodiment, the capture radius is 100m.
[0063] Step S1.3: When it is necessary to capture the complex structure or hierarchical structure of residential areas, select the CDC algorithm. This algorithm can cluster based on the hierarchical structure of the data and is suitable for complex data sets. In this case, the K value, i.e. the number of neighbors, is set to 10 and the threshold ratio is set to 0.8.
[0064] Step S1.4: For large-scale data sets, the MSGC algorithm is used. This algorithm can efficiently process large-scale data, and can also handle noise and outliers, improving the accuracy and efficiency of clustering. In this case, the number of grids is set to 10, and the number of clusters is set to 500.
[0065] Step S1.5: After selecting a clustering algorithm, use the algorithm to cluster the standardized vector data set D' to generate a clustering result vector data set C. The clustering result set C contains multiple clusters, each cluster representing a residential area group. Figure 3 The clustering results of the standardized vector dataset D' are four clustering pairs.
[0066] Step S1.6, in order to evaluate the quality of clustering results, calculate the silhouette coefficient and cluster purity of each clustering algorithm. The silhouette coefficient is used to measure the closeness of a data point to its cluster and the degree of separation from other clusters. The larger the value, the better the clustering effect. Cluster purity is used to measure the proportion of correctly classified data points in the clustering results. Select the CDC clustering result C_opt with the highest silhouette coefficient and the best cluster purity as the input for subsequent steps; silhouette coefficient formula (1), cluster purity formula (2);
[0067]
[0068] Where a represents the average distance from a sample point to other sample points in the class to which it belongs, reflecting the compactness within the cluster. b represents the average distance from a sample point to the sample point of the nearest class, reflecting the separation between clusters. N is the total number of samples in the data set. ci is the i-th cluster set. tj is the j-th true class. |ci∩tj| represents the number of samples in the intersection between cluster set ci and true class tj.
[0069] Further, obtaining the optimized eigenvector matrix includes:
[0070] Based on the clustered residential area vector data, extracting geometric and spatial features of the residential area group to obtain a feature vector set;
[0071] The eigenvector set is reduced in dimension to obtain an optimized eigenvector matrix.
[0072] Specifically, step S2: extracting features from the clustering result C_opt and reducing the dimension of the extracted feature matrix.
[0073] Step S2.1: Feature extraction. Extract the geometric and spatial characteristics of the residential area clusters from the clustering result C_opt, including convex hull area, building density, ratio of building area to convex hull area, etc. These features can reflect key information such as the morphology, scale, and spatial distribution of the residential area clusters. Construct a feature vector set F, where each feature vector corresponds to a residential area cluster.
[0074] Step S2.2: To reduce the dimension of the feature vectors and improve the efficiency of subsequent classification tasks, use PCA (Principal Component Analysis) to reduce the dimension of the feature vector set F. PCA transforms high-dimensional data into low-dimensional data by calculating the principal components of the data (i.e., the projections of the data in the directions of the maximum variances). Retain the main features to generate an optimized feature vector matrix F_opt for subsequent classification tasks.
[0075] Step S2.3: Conduct a correlation analysis based on the optimized feature vector matrix F_opt to obtain the correlation matrix diagram of the feature vectors after dimensionality reduction as Figure 4 shown.
[0076] Furthermore, classify the optimized feature vector matrix to obtain classification results including:
[0077] Classify according to the distribution characteristics of the data in the optimized feature vector matrix;
[0078] Divide the classified data into regions of equal area to obtain the initial classification results;
[0079] Align the initial classification results with the residential area vector data set to obtain the classification results.
[0080] Specifically, Step S3: Classify the residential area clusters for the optimized feature vector matrix F_opt.
[0081] Step S3.1: Classify the optimized feature vector matrix F_opt using the natural breaks method and the equal area division method. The natural breaks method classifies according to the distribution characteristics of the data itself, which can maintain the natural discontinuity of the data; the equal area division method divides the data into regions of equal area, which is convenient for spatial analysis and comparison.
[0082] Step S3.2: Divide the residential areas into six categories: small and scattered, small and compact, medium and scattered, medium and compact, large and scattered, large and compact. These classification results can reflect the different scales and spatial distribution characteristics of the residential area clusters. Generate a classification label set T to identify the classification results of each residential area cluster. Figure 5 The classification result diagram for the residential area clusters of the label set T.
[0083] Step S3.3: To align the classification results with the original geographic data, a spatial matching method (such as spatial indexing, spatial join) is used to align the classification result T with the original geographic data, such as the residential area vector dataset. This can form the final classification result R, which is convenient for subsequent spatial analysis and decision support.
[0084] Furthermore, obtaining the trained Transformer model includes:
[0085] Dividing the optimized feature vector matrix and the classification result into a training set and a test set;
[0086] Inputting the training set into the Transformer model for training to obtain an untrained Transformer model;
[0087] Using the test set to test the untrained Transformer model, evaluating the performance of the Transformer model, and adjusting the Transformer model to obtain the trained Transformer model.
[0088] Specifically, step S4: Using the Transformer model to perform classification detection on the residential area group vector dataset. During the classification detection process of the Transformer model, the following steps are followed:
[0089] Step S4.1: Divide the optimized feature vector matrix F_opt and the final classification result R into a training set and a test set. The training set is used to train the Transformer model, and the test set is used to evaluate the performance of the model. Input the training set into the Transformer model for training, and adjust the model parameters to minimize the classification error. During the training process, a cross-validation method is used to evaluate the stability and generalization ability of the model.
[0090] Step S4.2: To evaluate the performance of the trained Transformer model, use the test set for testing. Calculate metrics such as the classification accuracy and recall rate of the model to evaluate the performance of the model. The classification accuracy represents the proportion of data points correctly classified by the model; the recall rate represents the proportion of positive class data points that the model can correctly identify. In addition, a confusion matrix diagram is also used to visually display the classification accuracy and misclassification situation of the model. Figure 6 Confusion matrix diagram for evaluating the classification accuracy of the transformer model for this case.
[0091] Step S4.3. After the model training and evaluation are completed, the new data D_new can be input into the trained Transformer model for prediction. The model will generate a prediction result R_new based on the feature vectors of the input data. The prediction result R_new contains the classification labels and relevant information of the new data, which can be used to guide decision-making in fields such as urban planning and land use.
[0092] By integrating multiple algorithms such as K-means, DBSCAN, CDC local center clustering, and MSGC multi-scale grid clustering, the present invention can intelligently select the optimal clustering algorithm according to the distribution characteristics of different settlements, enabling the present invention to flexibly respond and achieve accurate division when facing settlements with diverse forms and uneven density distributions. This flexibility not only improves the accuracy of clustering but also lays a solid foundation for subsequent analysis work.
[0093] The present invention designs a feature expression system including 12 feature vectors such as convex hull area, building density, ratio of building area to convex hull area, coefficient of variation of Thiessen polygon, distance between buildings in the settlement, number of sides of buildings in the settlement, convex hull direction, PCA principal axis direction, area of minimum circumscribed rectangle, ratio of building area to area of minimum circumscribed rectangle, angle between the major axis of the circumscribed ellipse and the due north direction, and perimeter-area ratio. These feature vectors comprehensively cover the geometric shape and spatial characteristics of the settlement, and can accurately reflect key information such as the compactness of the building group and the distribution law of the settlement group. This high-precision feature extraction system provides rich and high-quality data support for subsequent dimensionality reduction and classification, ensuring the accuracy and reliability of the analysis results.
[0094] When dealing with high-dimensional data, dimensionality reduction is a key step to improve the calculation efficiency and optimize the model performance. The present invention uses methods such as principal component analysis (PCA) and correlation calculation to select 6 feature vectors that can best represent the geographical characteristics of the settlement group from 12 features. This process not only retains the maximum variance of the feature information, improves the calculation efficiency, but also reduces redundant information and optimizes the performance of the classification model. Through efficient dimensionality reduction, the present invention reduces the calculation complexity and improves the overall performance while ensuring the classification accuracy and generalization ability of the model.
[0095] The present invention adopts an improved Transformer model and combines the natural break method and the equal area division method to accurately classify the settlement group into six categories. This innovative classification model can not only complete the classification task quickly and efficiently but also improve the refinement degree of the classification result through the feature optimization algorithm. This innovative classification method not only improves the prediction accuracy but also provides strong support for applications such as urban-rural planning and land use.
[0096] The present invention proposes a complete process from clustering analysis to feature extraction, dimensionality reduction, classification, and detection. The method is comprehensive and highly versatile. This system not only accommodates different types of residential area distribution characteristics but also has good portability and scalability. Therefore, it can be widely applied to multiple fields such as geographic information systems (GIS), urban and rural planning, resource optimization allocation, and disaster monitoring. This comprehensiveness and versatility endow the present invention with broad practical value and social significance.
[0097] The present invention also excels in terms of automation. By automatically selecting the optimal clustering algorithm, automatically performing feature dimensionality reduction, and using an automatic classification method based on an improved Transformer, the present invention significantly reduces manual intervention and improves processing efficiency. This highly automated processing method not only enhances the simplicity of operation but also ensures the consistency of results. This advantage makes the present invention an efficient, stable, and easy-to-use residential area analysis tool, providing great convenience to users.
[0098] The above is only the preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.
Claims
1. A method for intelligent classification of residential areas integrating Transformer model, characterized in that: include: Acquire a residential area vector data set, and preprocess the residential area vector data set to obtain preprocessed residential area vector data; Clustering the preprocessed residential area vector data into groups to obtain clustered residential area vector data; Extracting features from the clustered residential area vector data according to preset features, reducing the dimension of the extracted features, and obtaining an optimized feature vector matrix; Classifying the optimized feature vector matrix to obtain a classification result; Inputting the optimized feature vector matrix and the classification result into a Transformer model for training to obtain a trained Transformer model; The residential vector data of the area to be tested is obtained, and the trained Transformer model is input to obtain the intelligent classification result of the residential group.
2. The method for intelligent classification of residential areas integrating the Transformer model as claimed in claim 1, characterized in that: Preprocessing the residential area vector data set to obtain the preprocessed residential area vector data includes: Based on the residential area vector dataset, checking the projection coordinate system type of the residential area vector dataset, identifying and removing outliers in the residential area vector dataset; Fill in the missing parts in the data to obtain complete residential area vector data; The complete residential area vector data is standardized to obtain pre-processed residential area vector data.
3. The intelligent classification method for residential areas integrating the Transformer model as claimed in claim 1, characterized in that: The vector data of residential areas after clustering include: Based on the pre-processed residential area vector data, different clustering methods are selected for different types of residential area vector data, and the pre-processed residential area vector data are clustered to obtain clustered residential area vector data.
4. The method for intelligent classification of residential areas integrating the Transformer model as claimed in claim 1, characterized in that: Obtaining the optimized eigenvector matrix includes: Based on the clustered residential area vector data, extracting geometric and spatial features of the residential area group to obtain a feature vector set; The eigenvector set is reduced in dimension to obtain an optimized eigenvector matrix.
5. The method for intelligent classification of residential areas integrating the Transformer model as claimed in claim 1, characterized in that: The preset features include: convex hull area, building density, ratio of building area to convex hull area, Thiessen polygon coefficient of variation, distance between residential buildings, number of building edges in residential areas, convex hull direction, PCA main axis direction, minimum circumscribed rectangle area, ratio of building area to minimum circumscribed rectangle area, angle between the major axis of the circumscribed ellipse and the north direction, and perimeter area ratio.
6. The method for intelligent classification of residential areas integrating the Transformer model as claimed in claim 1, characterized in that: Classifying the optimized feature vector matrix to obtain a classification result includes: Classify according to the distribution characteristics of the data in the optimized feature vector matrix; Divide the classified data into regions of equal area to obtain the initial classification results; The initial classification result is aligned with the residential area vector dataset to obtain a classification result.
7. The intelligent classification method for residential areas integrating the Transformer model as claimed in claim 1, characterized in that: The classification results include small evacuation class, small tight class, medium evacuation class, medium tight class, large evacuation class, and large tight class.
8. The method for intelligent classification of residential areas integrating the Transformer model as claimed in claim 1, characterized in that: Obtaining a trained Transformer model includes: Dividing the optimized feature vector matrix and the classification results into a training set and a test set; Inputting the training set into the Transformer model for training to obtain an untested Transformer model; The untested Transformer model is tested using the test set to evaluate the performance of the Transformer model, adjust the Transformer model, and obtain a trained Transformer model.