User payment data classification analysis method based on MSBOA-FCM-BIRCH algorithm
By using the MSBOA-FCM-BIRCH algorithm to perform multi-strategy fusion optimization and clustering of user payment data, the problem of lack of effective analysis in traditional user payment data management is solved. This enables accurate classification and efficient clustering of user payment habits, supporting grassroots power supply stations in formulating targeted service strategies.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ANSHAN POWER SUPPLY COMPANY OF STATE GRID LIAONING ELECTRIC POWER COMPANY
- Filing Date
- 2025-12-12
- Publication Date
- 2026-04-21
AI Technical Summary
Traditional user payment data management lacks effective data summarization and analysis methods, making it impossible to extract user payment cycles, payment method preferences, and arrears patterns from complex payment data, resulting in insufficient understanding of user payment habits by grassroots power supply stations.
A user payment data classification and analysis method based on the MSBOA-FCM-BIRCH algorithm is adopted. The initial cluster centers and fuzzy factors of the fuzzy C-means algorithm are globally optimized by multi-strategy fusion butterfly optimization algorithm. Combined with the construction of a clustering feature tree CF-Tree with dynamic threshold adjustment and global clustering, a secondary clustering of user payment behavior is achieved.
It enables accurate classification of user payment data, improves the precision and efficiency of payment data processing, and provides technical support for grassroots power supply stations to understand user payment habits and formulate targeted service strategies.
Smart Images
Figure 0BU8SYEJIXJGRNGDXAMUUC1PPTJWH6TPSOXETBKG 
Figure 0KVHTWJ48ZG4MJ4R7PJO00GEG7JSEZCVT9IDQC16 
Figure 0MCZTRMHTUHC5WYMWPOVTK4T1WXXZW0KWTJERAHF
Abstract
Description
Technical Field
[0001] This invention relates to the field of power system user data analysis technology, and in particular to a method for classifying and analyzing user payment data based on the MSBOA-FCM-BIRCH algorithm. Background Technology
[0002] In the management of user payment at grassroots power supply stations, traditional payment data processing and user analysis models have significant limitations. Currently, user payment data management mainly relies on traditional methods such as scattered recording and manual statistics, which can only retain basic payment information and has not formed a systematic data analysis system. Given the complexity of user payment data, there is a lack of effective summarization and analysis methods, making it impossible to extract core characteristics such as user payment cycles, payment method preferences, and arrears patterns from the complex payment data. This results in insufficient understanding of users' basic payment habits at grassroots power supply stations. Therefore, it is necessary to study data mining algorithms based on user payment data and utilize power business data, information fusion, big data analysis, and other information technologies to establish a user payment data analysis model. Summary of the Invention
[0003] This invention provides a user payment data classification and analysis method based on the MSBOA-FCM-BIRCH algorithm, which solves the problem of the lack of effective data summarization and analysis methods in the current user payment management system.
[0004] To achieve the above objectives, the present invention employs the following technical solution: A method for classifying and analyzing user payment data based on the MSBOA-FCM-BIRCH algorithm includes the following steps: S1. Collect user payment and related data, including electricity bill value, number of reminders, reminder time, and payment time, and combine them with accumulated historical data to form a complete dataset; S2. Standardize and preprocess the dataset; S3. The multi-strategy fusion butterfly optimization algorithm is used to globally optimize the initial cluster centers and fuzzy factors of the fuzzy C-means algorithm, generating the optimal parameter combination that adapts to the distribution of payment data. S4. Perform fuzzy C-means soft clustering based on the optimized parameters to calculate the membership matrix of each user's payment data point to each cluster; S5. Transform the membership matrix into the input features of comprehensive hierarchical clustering. Construct a clustering feature tree (CF-Tree) with dynamic threshold adjustment and perform global clustering to complete the secondary clustering of user payment behavior and output user clusters with clear behavioral feature labels. S6. When adding new data each month, the model is iterated by combining the incremental optimization of the MSBOA algorithm with the dynamic adjustment of the BIRCH algorithm (a comprehensive hierarchical clustering algorithm) to ensure that the classification results are consistent with users' payment habits in the long term.
[0005] Furthermore, the complete dataset is the original user payment dataset, specifically: ; in, Let i be the feature vector of the i-th user's payment data point. This is the electricity cost. The number of times a payment reminder is issued. In order to expedite the process, For payment time, The timestamp corresponding to the data. This represents the total number of samples for payment data.
[0006] Furthermore, step S3 specifically includes the following process: 1) Initial cluster centers: ; in, This represents the initial cluster center matrix for the fuzzy C-means algorithm. For the first The initial cluster centers of each cluster. The preset number of clusters; This represents the matrix transpose operation; 2) Butterfly population initialization: ; in, For the first The position vector of the butterfly. For butterfly population size, For the first The butterfly corresponds to the first Initial centers of each cluster For the first The fuzzy factor corresponding to only one butterfly; The initial population is generated using a Logistic chaotic mapping, with the chaotic initialization formula as follows: ; in, For the first The value of the next chaotic iteration. For chaos control parameters, These are the initial cluster center components for the generated FCM; 3) Fitness function calculation; With blur Maximizing the fuzzy partitioning coefficient (FPC) of the mean is the objective, and the fitness function is: ; in, For the first The fuzzy partitioning coefficient (FPC) corresponding to each individual butterfly; ; in, The total number of samples for payment data. The preset number of clusters, For the first The payment data point for the first The membership degree of each cluster; 4) Dynamic search and location updates; The global or local search is dynamically switched based on the density of payment data clusters. The dynamic search switching formula is as follows: ; in, This represents the percentage of global searches. Given the current cluster density, The average density of all clusters; Combining pheromone-guided updates to butterfly position, the position update formula is: ; in, Let i be the position of the i-th butterfly in the t-th iteration. The pheromone guiding coefficient. Let be the pheromone concentration of the i-th butterfly. Let be the optimal butterfly position in the t-th iteration. For random search coefficients, A random number in the range [0,1]. , The positions of two randomly selected butterflies; After each iteration, the time series weights are adjusted once. The formula for the time series weights is: ; in, The weight of the payment data at time t, As the initial weights, The attenuation coefficient is... This represents the total duration of historical data. This refers to the time point corresponding to the current data. 5) Convergence condition judgment; Set dual convergence conditions: (1) Relative error condition: ; in, The convergence threshold, For the first The optimal fitness value in the next iteration. For the first The optimal fitness value in the next iteration; (2) Maximum number of iterations condition: ; in, Adjustments based on the scale of payment data; When any convergence condition is met, the optimal fuzzy C-mean parameter is output: ; in, To be the optimal initial cluster centers, It is the optimal fuzzy factor.
[0007] Furthermore, step S4 specifically includes the following process: 1) Membership initialization; The initial membership matrix is calculated based on V*, using the following formula: ; in, For the first The payment data point for the first The initial membership degree of each cluster; 2) Iterative updates; (1) Update the cluster centers according to the cluster center update formula. The formula is: ; in, For the first The user payment data point for the first The m-th power of the membership degree of a cluster As a fuzzy factor, For the first Feature vector of each user's payment data point; (2) Update the membership matrix according to the membership calculation formula. The formula is: ; in, For the first The user payment data point for the first The membership degree of each cluster, For the first Feature vectors of individual user payment data points As a fuzzy factor, For the first Cluster centers of each cluster For the first Cluster centers of each cluster; (3) Calculate the objective function The objective function formula is: ; in, The objective function value of FCM is the weighted distance sum. For the first The user payment data point to the first The squared Euclidean distance between the centers of each cluster; (4) When The iteration terminates, and the final membership matrix is output: ; in, This is the final membership matrix output after the fuzzy C-means algorithm terminates its iterations; For the first The payment data point for the first The final affiliation of each cluster.
[0008] Furthermore, the process of transforming the membership matrix into input features for comprehensive hierarchical clustering, constructing a CF-Tree clustering feature tree with dynamically adjusted thresholds, and performing global clustering to complete the secondary clustering of user payment behavior specifically includes the following steps: 1) CF-Tree construction; (1) For each payment data point ,according to Calculate the corresponding CF component. The formula for calculating the clustering feature CF is: ; in, For the first Clustering characteristics of individual clusters For weighted linear sums, For weighted sum of squares; The formula for the weighted sample size is: ; ; ; in, For the first The squared Euclidean norm of the feature vector of each user's payment data point; (2) The formula for calculating the cluster weighted centroid is: ; in, For the first The weighted centroids of each cluster; (3) Insert the CF component into the CF-Tree when the distance between the current node's CF and the new CF is less than the dynamic threshold. The dynamic merging threshold formula is: ; in, For the first BIRCH merging threshold at time 10:00 As the initial threshold, For the first The average cluster density at time t, The initial average cluster density is used; if the condition is met, nodes are merged; otherwise, new nodes are created. (4) Pruning redundant nodes: When the number of nodes exceeds the branching factor, redundant nodes are pruned to ensure that CF-Tree compresses data efficiently. 2) Global clustering; Perform hierarchical clustering on the leaf nodes of the CF-Tree and calculate the weighted distance between nodes: ; in, For the first , The distance between leaf nodes, For the first Weighted centroids of leaf nodes For the first Weighted centroids of leaf nodes For the first Weighted radius of leaf nodes For the first The weighted radius of each leaf node.
[0009] 3) Output the clustering results; The user payment data classification model uses a comprehensive hierarchical clustering secondary clustering operation to merge data with a distance less than [missing value]. The node, the final output Individual user payment behavior clusters.
[0010] Furthermore, each user cluster corresponds to a specific payment behavior feature label, including the average electricity bill, average number of reminders, typical payment time, and payment habit stability of users within the cluster.
[0011] Compared with the prior art, the beneficial effects of the present invention are: This invention employs a three-tiered collaborative architecture that integrates the Butterfly Optimization Algorithm (MSBOA), Fuzzy C-means (FCM), and Comprehensive Hierarchical Clustering (BIRCH) to accurately extract user payment data features and achieve efficient clustering. This solves the problems of insufficient accuracy and excessive resource consumption of traditional algorithms when processing multi-dimensional, noisy, and behaviorally ambiguous payment data. It provides technical support for grassroots power supply stations to understand user payment habits and formulate targeted service strategies. Detailed Implementation
[0012] The specific embodiments of the present invention will be further described below: This invention presents a user payment data classification and analysis method based on the MSBOA-FCM-BIRCH algorithm. It employs a hybrid clustering model combining multi-strategy fusion butterfly optimization algorithm, fuzzy C-means algorithm, and comprehensive hierarchical clustering algorithm. First, user payment and related data are collected through a data acquisition system, including multi-dimensional information such as electricity bill value, number of reminders, reminder time, and payment time, forming a complete dataset by combining it with accumulated massive historical data. Then, the dataset undergoes standardization preprocessing to eliminate dimensional differences between feature variables, laying the foundation for subsequent feature extraction and clustering. Next, a multi-strategy fusion butterfly optimization algorithm is used... The butterfly optimization algorithm globally optimizes the initial cluster centers and fuzzy factors of the fuzzy C-means algorithm to generate the optimal parameter combination that fits the distribution of payment data. Subsequently, based on the optimized parameters, fuzzy C-means soft clustering is performed to calculate the membership matrix of each payment data point to each cluster, thereby achieving accurate characterization of users with fuzzy behavioral boundaries. Finally, the membership matrix is transformed into the input features of comprehensive hierarchical clustering. Through the construction of a CF-Tree with dynamic threshold adjustment and global clustering, a secondary clustering of user payment behavior is completed, outputting user clusters with clear behavioral feature labels, ensuring the accuracy and practicality of the classification results.
[0013] The above steps constitute a single complete user payment data classification and analysis. To adapt to the long-term need for new data, when new data is added each month, the model is iterated in real time through incremental optimization of the MSBOA algorithm and dynamic adjustment of the BIRCH algorithm CF-Tree, ensuring that the classification results are consistent with users' payment habits in the long term.
[0014] Fuzzy C-means soft clustering was optimized using the MSBOA optimization algorithm, and BIRCH clustering was used for secondary clustering to achieve accurate classification of user payment data. The specific steps are as follows: S1. Data Acquisition and Preprocessing; Let the original user payment dataset be: ; in, Let i be the feature vector of the i-th user's payment data point. This is the electricity cost. The number of times a payment reminder is issued. In order to expedite the process, For payment time, The timestamp corresponding to the data. This represents the total number of samples for payment data.
[0015] S2. Data standardization processing: ; in, This represents the original value of the j-th feature of the i-th sample. Let be the mean of the j-th feature. Let be the standard deviation of the j-th feature. These are the standardized feature values.
[0016] S3. Multi-strategy fusion butterfly optimization algorithm optimizes fuzzy C-mean parameters; The key parameters of the fuzzy C-means are used as optimization variables in the multi-strategy fusion butterfly optimization algorithm, and the optimal parameter combination is searched iteratively. 1) Initial cluster centers: ; in, This represents the initial cluster center matrix for the fuzzy C-means algorithm. Let c be the initial cluster center of the c-th cluster, where c is the preset number of clusters; This represents the matrix transpose operation; 2) Butterfly population initialization: ; in, For the first The position vector of the butterfly. For butterfly population size, For the first The butterfly corresponds to the first Initial centers of each cluster For the first The fuzzy factor corresponding to only one butterfly; The initial population is generated using a Logistic chaotic mapping, with the chaotic initialization formula as follows: ; in, For the first The value of the next chaotic iteration. For chaos control parameters, For the initial cluster center components generated by FCM, ensure that the initial parameters uniformly cover all behavior clusters such as on-time payment and late payment; 3) Fitness function calculation; With the objective of maximizing the fuzzy partitioning coefficient FPC of the fuzzy C-mean, the fitness function is: ; in, For the first The fuzzy partitioning coefficient (FPC) corresponding to each individual butterfly reflects the compactness and separation of the fuzzy C-means clustering results; ; in, This represents the total number of samples for payment data; The number of clusters is the preset number; where, For the first The payment data point for the first The membership degree of a cluster reflects the strength of the association between a data point and the cluster; 4) Dynamic search and location updates; The global or local search is dynamically switched based on the density of payment data clusters. The dynamic search switching formula is as follows: ; in, This represents the percentage of global searches. Given the current cluster density, The average density of all clusters; Combining pheromone-guided updates to butterfly position, the position update formula is: ; in, Let i be the position of the i-th butterfly in the t-th iteration. The pheromone guiding coefficient. Let be the pheromone concentration of the i-th butterfly. Let be the optimal butterfly position in the t-th iteration. For random search coefficients, A random number in the range [0,1]. , The positions of two randomly selected butterflies; After each iteration, the time series weights are adjusted once. The formula for the time series weights is: ; in, The weight of the payment data at time t, As the initial weights, The attenuation coefficient is... This represents the total duration of historical data. For the current time point corresponding to the data, the impact of recent payment data on the optimization results is emphasized; 5) Convergence condition judgment; Set dual convergence conditions: (1) Relative error condition: ; in, The convergence threshold, For the first The optimal fitness value in the next iteration. For the first The optimal fitness value in the next iteration; (2) Maximum number of iterations condition: ; in, Adjustments based on the scale of payment data; When any convergence condition is met, the optimal fuzzy C-mean parameter is output: ; in, To be the optimal initial cluster centers, It is the optimal fuzzy factor.
[0017] S4. Perform fuzzy C-means soft clustering based on the optimized parameters to calculate the membership matrix of each payment data point to each cluster; Parameters optimized based on multi-strategy fusion butterfly optimization algorithm Perform fuzzy C-means clustering: 1) Membership initialization; The initial membership matrix is calculated based on V*, using the following formula: ; in, For the first The payment data point for the first The initial membership degree of each cluster, The preset number of clusters; 2) Iterative updates; (1) Update the cluster centers according to the cluster center update formula. The formula is: ; in, For the first The user payment data point for the first The m-th power of the membership degree of a cluster As a fuzzy factor, For the first Feature vector of each user's payment data point; (2) Update the membership matrix according to the membership calculation formula. The formula is: ; in, For the first The user payment data point for the first The membership degree of each cluster, For the first Feature vectors of individual user payment data points As a fuzzy factor, For the first Cluster centers of each cluster For the first Cluster centers of each cluster; (3) Calculate the objective function The objective function formula is: ; in, The objective function value of FCM is the weighted distance sum. For the first The user payment data point to the first The squared Euclidean distance between the centers of each cluster; (4) When The iteration terminates, and the final membership matrix is output: ; in, This is the final membership matrix output after the fuzzy C-means algorithm terminates its iterations; For the first The payment data point for the first The final membership degree of each cluster.
[0018] S5. Transform the membership matrix into the input features of comprehensive hierarchical clustering. Through the construction of CF-Tree with dynamic threshold adjustment and global clustering, complete the secondary clustering of user payment behavior and output user clusters with clear behavioral feature labels to ensure the accuracy and practicality of the classification results. The fuzzy C-mean membership matrix The input features are transformed into comprehensive hierarchical clustering features, CF-Tree is constructed, and secondary clustering is completed; 1) CF-Tree construction; (1) For each payment data point ,according to Calculate the corresponding CF component. The formula for calculating the clustering feature CF is: ; in, For the first Clustering characteristics of individual clusters For weighted linear sums, For weighted sum of squares; The formula for the weighted sample size is: ; in, The total number of samples for payment data. For the first The payment data point for the first The membership degree of each cluster; For the weighted linear sum, the formula is: ; in, For the first Feature vector of each payment data point; The formula for the weighted sum of squares is: ; in, For the first The squared Euclidean norm of the feature vector of each user's payment data point; (2) The formula for calculating the cluster weighted centroid is: ; in, For the first The weighted centroids of each cluster; (3) Insert the CF component into the CF-Tree when the distance between the current node's CF and the new CF is less than the dynamic threshold. The dynamic merging threshold formula is: ; in, For the first BIRCH merging threshold at time 10:00 As the initial threshold, For the first The average cluster density at time t, The initial average cluster density is used; if the condition is met, nodes are merged; otherwise, new nodes are created. (4) Pruning redundant nodes: When the number of nodes exceeds the branching factor, redundant nodes are pruned to ensure that CF-Tree compresses data efficiently; 2) Global clustering; Perform hierarchical clustering on the leaf nodes of the CF-Tree and calculate the weighted distance between nodes (based on the centroid and radius of the CF): ; in, For the first , The distance between leaf nodes, For the first Weighted centroids of leaf nodes For the first Weighted centroids of leaf nodes For the first Weighted radius of leaf nodes For the first The weighted radius of each leaf node.
[0019] 3) Output the clustering results; The user payment data classification model uses a comprehensive hierarchical clustering secondary clustering operation to merge data with a distance less than [missing value]. The node, the final output Each user payment behavior cluster corresponds to a specific payment behavior feature label. The cluster includes core information such as the average electricity bill, average number of reminders, typical payment time, and stability of payment habits of users within the cluster. This provides direct data support for grassroots power supply stations to understand users' payment habits and formulate targeted service strategies.
[0020] S6. Model Iteration Update: To adapt to the long-term need for new data, when new data is added each month, the model can be iterated through incremental optimization of the MSBOA algorithm and dynamic adjustment of the BIRCH algorithm CF-Tree to ensure that the classification results are consistent with users' payment habits in the long term. When new user payment data is added each month, incremental optimization is performed using a multi-strategy fusion butterfly optimization algorithm, combined with dynamic adjustment using a comprehensive hierarchical clustering CF-Tree, to achieve real-time model iteration. This ensures that the classification results align with users' latest payment habits and meet the long-term needs of processing massive amounts of accumulated historical data. The above embodiments are implemented based on the technical solution of the present invention, providing detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the above embodiments. Unless otherwise specified, the methods used in the above embodiments are conventional methods.
Claims
1. A method for classifying and analyzing user payment data based on the MSBOA-FCM-BIRCH algorithm, characterized in that, Includes the following steps: S1. Collect user payment and related data, including electricity bill value, number of reminders, reminder time, and payment time, and combine them with accumulated historical data to form a complete dataset; S2. Standardize and preprocess the dataset; S3. The multi-strategy fusion butterfly optimization algorithm is used to globally optimize the initial cluster centers and fuzzy factors of the fuzzy C-means algorithm, generating the optimal parameter combination that adapts to the distribution of payment data. S4. Perform fuzzy C-means soft clustering based on the optimized parameters to calculate the membership matrix of each user's payment data point to each cluster; S5. Transform the membership matrix into the input features of comprehensive hierarchical clustering. Construct a clustering feature tree (CF-Tree) with dynamic threshold adjustment and perform global clustering to complete the secondary clustering of user payment behavior and output user clusters with clear behavioral feature labels. S6. When adding new data each month, the model is iterated by combining the incremental optimization of the MSBOA algorithm with the dynamic adjustment of the BIRCH algorithm (a comprehensive hierarchical clustering algorithm) to ensure that the classification results are consistent with users' payment habits in the long term.
2. The user payment data classification and analysis method based on the MSBOA-FCM-BIRCH algorithm according to claim 1, characterized in that, The complete dataset is the original user payment dataset, specifically: ； in, Let i be the feature vector of the i-th user's payment data point. This is the electricity cost. The number of times a payment reminder is issued. To expedite the payment process, For payment time, The timestamp corresponding to the data. This represents the total number of samples for payment data.
3. The user payment data classification and analysis method based on the MSBOA-FCM-BIRCH algorithm according to claim 1, characterized in that, Step S3 specifically includes the following process: 1) Initial cluster centers: ; in, This represents the initial cluster center matrix for the fuzzy C-means algorithm. For the first The initial cluster centers of each cluster. The preset number of clusters; This represents the matrix transpose operation; 2) Butterfly population initialization: ； in, For the first The position vector of the butterfly. For butterfly population size, For the first The butterfly corresponds to the first Initial centers of each cluster For the first The fuzzy factor corresponding to only one butterfly; The initial population is generated using a Logistic chaotic mapping, with the chaotic initialization formula as follows: ； in, For the first The value of the next chaotic iteration For chaos control parameters, These are the initial cluster center components for the generated FCM; 3) Fitness function calculation; With blur Maximizing the fuzzy partitioning coefficient (FPC) of the mean is the objective, and the fitness function is: ； in, For the first The fuzzy partitioning coefficient (FPC) corresponding to each individual butterfly; ； in, The total number of samples for payment data. The preset number of clusters, For the first The payment data point for the first The membership degree of each cluster; 4) Dynamic search and location updates; The global or local search is dynamically switched based on the density of payment data clusters. The dynamic search switching formula is as follows: ； in, This represents the percentage of global searches. Given the current cluster density, The average density of all clusters; Combining pheromone-guided updates to butterfly position, the position update formula is: ； in, Let i be the position of the i-th butterfly in the t-th iteration. The pheromone guiding coefficient. Let be the pheromone concentration of the i-th butterfly. Let be the optimal butterfly position in the t-th iteration. For random search coefficients, A random number in the range [0,1]. , The positions of two randomly selected butterflies; After each iteration, the time series weights are adjusted once. The formula for the time series weights is: ； in, Let t be the weight of the payment data at time t. As the initial weights, The attenuation coefficient is... This represents the total duration of historical data. This refers to the time point corresponding to the current data. 5) Convergence condition judgment; Set dual convergence conditions: (1) Relative error condition: ; in, The convergence threshold, For the first The optimal fitness value in the next iteration. For the first The optimal fitness value in the next iteration; (2) Maximum number of iterations condition: ; in, Adjustments based on the scale of payment data; When any convergence condition is met, the optimal fuzzy C-mean parameter is output: ; in, To be the optimal initial cluster centers, It is the optimal fuzzy factor.
4. The user payment data classification and analysis method based on the MSBOA-FCM-BIRCH algorithm according to claim 1, characterized in that, Step S4 specifically includes the following process: 1) Membership initialization; The initial membership matrix is calculated based on V*, using the following formula: ; in, For the first The payment data point for the first The initial membership degree of each cluster; 2) Iterative updates; (1) Update the cluster centers according to the cluster center update formula. The formula is: ； Among them, the The user payment data point for the first The m-th power of the membership degree of a cluster As a fuzzy factor, For the first Feature vector of each user's payment data point; (2) Update the membership matrix according to the membership calculation formula. The formula is: ； in, For the first The user payment data point for the first The membership degree of each cluster, For the first Feature vectors of individual user payment data points As a fuzzy factor, For the first Cluster centers of each cluster For the first Cluster centers of each cluster; (3) Calculate the objective function The objective function formula is: ； in, The objective function value of FCM is the weighted distance sum. For the first The user payment data point to the first The squared Euclidean distance between the centers of each cluster; (4) When The iteration terminates, and the final membership matrix is output: ； in, This is the final membership matrix output after the fuzzy C-means algorithm terminates its iterations; For the first The payment data point for the first The final membership degree of each cluster.
5. The user payment data classification and analysis method based on the MSBOA-FCM-BIRCH algorithm according to claim 1, characterized in that, The process of transforming the membership matrix into input features for comprehensive hierarchical clustering, constructing a CF-Tree clustering feature tree with dynamically adjusted thresholds, and performing global clustering to complete the secondary clustering of user payment behavior includes the following steps: 1) CF-Tree construction; (1) For each payment data point ,according to Calculate the corresponding CF component. The formula for calculating the clustering feature CF is: ； in, For the first Clustering characteristics of individual clusters For weighted linear sums, For weighted sum of squares; The formula for the weighted sample size is: ； ； ； in, For the first The squared Euclidean norm of the feature vector of each user's payment data point; (2) The formula for calculating the cluster weighted centroid is: ； in, For the first The weighted centroids of each cluster; (3) Insert the CF component into the CF-Tree when the distance between the current node's CF and the new CF is less than the dynamic threshold. The dynamic merging threshold formula is: ； in, For the first BIRCH merging threshold at time 10:00 As the initial threshold, For the first The average cluster density at time t, The initial average cluster density is used; if the condition is met, nodes are merged; otherwise, new nodes are created. (4) Pruning redundant nodes: When the number of nodes exceeds the branching factor, redundant nodes are pruned to ensure that CF-Tree compresses data efficiently. 2) Global clustering; Perform hierarchical clustering on the leaf nodes of the CF-Tree and calculate the weighted distance between nodes: ； in, For the first , The distance between leaf nodes, For the first Weighted centroids of leaf nodes For the first Weighted centroids of leaf nodes For the first Weighted radius of leaf nodes For the first Weighted radius of leaf nodes; 3) Output the clustering results; The user payment data classification model uses a comprehensive hierarchical clustering secondary clustering operation to merge data with a distance less than [missing value]. The node, the final output Each user payment behavior cluster corresponds to a specific payment behavior feature label. The cluster includes the average electricity bill, average number of reminders, typical payment time, and payment habit stability of users within the cluster. This provides direct data support for grassroots power supply stations to understand users' payment habits and formulate targeted service strategies.
6. The user payment data classification and analysis method based on the MSBOA-FCM-BIRCH algorithm according to claim 1, characterized in that, Each user cluster corresponds to a specific payment behavior feature label, which includes the average electricity bill, average number of reminders, typical payment time, and payment habit stability of users within the cluster.